We will look at search problems from different domains of Artificial Intelligence: constraint satisfaction, adversarial game playing, and planning - and show how the techniques of rational meta-reasoning could help significantly decrease the search times. We will also discuss the difficulties of applying rational meta-reasoning in each of the case studies, and how they were overcome. Informed search algorithms employ heuristics to speed-up solving of certain problem instances. However, the cost of computing an heuristic can occasionally exceed the benefit, slowing down the search. For the best results, different heuristics should be applied at different search states, and sometimes a cheaper less informative heuristic, or even blind search, results in a shorter overall search time.

A key challenge in designing graph algorithms is to compress a graph $G$ so as to preserve some of its basic properties, such as distances and cuts. Both spanners [Peleg and Schaffer, 1989] and cut sparsifiers [Benczur and Karger, 1996] fall into this category, as they reduce the number of edges in $G$ without changing any distance or cut by more than a bounded factor.

I will discuss another flavor of this challenge, which asks instead to reduce the number of vertices. Specifically, given a graph $G$ and $k$ terminal vertices, we wish to construct a small graph that is a minor of $G$, and in which all the terminal distances equal those in $G$ (exactly). Can we bound the size of $G'$ by some function of $k$? I will also talk about a similar question regarding terminal cuts, called mimicking networks by [Hagerup, Katajainen, Nishimura, and Ragde, 1998].

Joint works with Inbal Rika and with Tamar Zondiner.

NoSQL databases have been initially designed to provide extreme scalability and availability for Internet applications, often at the expense of data consistency. The recent generation of Web-scale databases fills this gap, by offering transaction support. However, transaction processing implies a performance overhead that is redundant for many applications that do not require strong consistency. The solutions offered by state-of-the-art technologies, either static separation of the data accessed by transaction-enabled and native applications, or transforming all native operations into transactions in the latter, are both inadequate. We present a novel scalable transaction processing system, Mediator, that accommodates both transactional and native operations in the same database without compromising data safety. Our work introduces a lightweight synchronization protocol that enables conflict resolution between transactions and native operations that share the same data. We evaluate Mediator’s implementation on top of the HBase NoSQL database on a large-scale distributed testbed. Our results show that despite a slight overhead to the transactional traffic, Mediator substantially outperforms the best-in-class traditional system on a vast majority of mixed workloads – in particular, on all workloads in which the fraction of native operations exceeds 50%.

I will give a tour through a sparse sample of the information theory literature - both classical and recent - on relations between information and estimation. Beyond aesthetic value, these relations underlie some of the main tools in Shannon theory, such as the Entropy Power Inequality. They also give considerable insight into and a quantitative understanding of several estimation theoretic objects, such as the costs of causality and of mismatch, as well as the performance and structure of minimax estimators. Further, they enable the transfer of analytic tools and algorithmic know-how from one domain to another. Examples will be given to illustrate these points.

In a distributed message passing model a communication network is represented by an n-vertex graph whose vertices host processors, and edges serve as communication links. One of the most fundamental goals in this setting is breaking the symmetry in the network. Specifically, the tasks of computing vertex coloring, maximal matching, and maximal independent set are of great importance. In the mid-eighties several randomized distributed algorithms for these problems were devised. [Luby 86, Alon, Babai and Itai 86, Israeli and Itai 86]. These algorithms require O(log n) rounds, and until recently they remained the best-known ones. We present significantly improved results for these problems. These results include: 1. A randomized algorithm for computing a maximal matching in O(log Delta+ (log log n)^4) rounds, where Delta is the maximum degree in the graph. This result is provably optimal for all log Delta in the range [(log log n)^4, sqrt {log n}]. 2. A randomized maximal independent set algorithm requiring O(log Delta sqrt{ log n}) rounds. 3. A randomized (Delta+1)-coloring algorithm requiring O(log Delta + exp{ sqrt {log log n}}) rounds. We also introduce a new technique for reducing symmetry breaking problems on low arboricity graphs to low degree graphs. Low arboricity graphs include planar graphs, graphs of bounded genus, graphs of bounded treewidth, graphs that exclude any fixed minor, and many other graphs. These graphs may have unbounded maximum degree. Nevertheless, for low arboricity graphs we obtain a maximal matching algorithm that runs in O(sqrt {log n}) time, and a maximal independent set algorithm that runs in O(log^{3/4} n) time. The talk is based on a joint work with Michael Elkin, Seth Pettie, and Johannes Schneider.

A {it Euclidean (1+eps)-spanner} for a point set S in the plane is a sparse subgraph H of the complete Euclidean graph corresponding to S, which preserves all pairwise distances to within a factor of 1+eps. Euclidean spanners constitute a fundamental graph structure as they can be used to approximately solve many geometric proximity problems. In many applications of Euclidean (1+eps)-spanners, in addition to being sparse, the spanners should achieve small {it weight}, {it maximum degree} and {it hop-diameter}. Understanding the inherent tradeoffs between these parameters is the foremost open question in this area. In this talk I will survey the previous state-of-the-art results, and then present our new constructions of optimal Euclidean spanners. If time permits, I will discuss extensions to {it doubling metrics} and to {it fault-tolerant spanners}, and present some open problems in this area.

The talk will be self-contained. It is based on my FOCS'08 paper (with Dinitz and Elkin), my SODA'11 paper, my STOC'13 paper (with Elkin), and my recent manuscript http://arxiv.org/abs/1304.8135

One of the most significant recent developments in machine learning has been the resurgence of ``deep learning'', usually in the form of artificial neural networks. These systems are based on a multi-layered architecture, where the input goes through several transformations, with higher-level concepts derived from lower-level ones. Thus, these systems are considered to be particularly suitable for hard AI tasks, such as computer vision and language processing. Nevertheless, despite decades of research,a major caveat of deep learning is - and always has been - its strong reliance on heuristic methods. I will describe a new, provably efficient, algorithm for learning deep neural networks.

Joint work with Roi Livni and Ohad Shamir.

When we refer to efficient algorithms, we usually mean polynomial-time algorithms. In particular this is true for graph algorithms, which are considered efficient if they run in time polynomial in the number of vertices and edges of the graph. However, in some cases we may seek even more efficient algorithms whose running time is sublinear in the size of the input graph. Such algorithms do not even read the whole input graph, but rather sample random parts of the graph and compute approximations of various parameters of interest. In this talk I will survey various such algorithms, where the parameters I will discuss are: (1) The average degree the number of small stars (2) The weight of a minimum spanning tree (3) The size of a minimum vertex cover and a maximum matching (4) The number of edges that should be added/removed in order to obtain various properties

Clustering is one of the most widely used techniques for exploratory data analysis. Across many disciplines, people try to get intuition about their data by identifying meaningful groups among the data points. In the past five decades, many clustering algorithms have been developed and applied to a wide range of practical problems. However, in spite of the abundance of clustering research published every year, we are far from having satisfactory understanding of some of the most basic general issues in clustering. In this talk I survey some recent work on developing a general theory of clustering, aiming, among other things, to provide the clustering user with some systematic guidance concerning the matching between clustering algorithms (and their parameter settings) and specific clustering tasks. This talk is based on joint work with my past students, David, Pal, Rita Ackerman and David Loker.

In this paper we study a novel model in which agents arrive sequentially one after the other and each in turn chooses one action from a fixed set of actions to maximize his expected rewards given the information he possesses at the time of arrival. The information that becomes available affects the incentives of an agent to explore and generate new information. We characterize the optimal disclosure policy of a planner whose goal is to maximizes social welfare. The planner's optimal policy is characterized and shown to be intuitive and very simple to implement. As the number of agents increases the social welfare converges to the optimal welfare of the unconstrained mechanism. One interpretation for our result is the implementation of what is known as the 'Wisdom of the crowds'. This topic has become more relevant during the last decade with the rapid adaptation of the Internet. This is a joint work with Ilan Kremer and Motty Perry.

The goal of discriminative learning is to train a system to optimize a certain desired measure of performance. In simple classification we seek a function that assigns a binary label to a single object, and tries to minimize the error rate (correct or incorrect) on unseen data. In structured prediction we are interested in the prediction a structured label, where the input is a complex object. Typically, each structured prediction task has its own measure of performance, or task loss, such as word error rate in speech recognition, the BLEU score in machine translation or the intersection-over-union score in object segmentation. Not only that those task losses are much more involved than the binary error rate, the structured prediction itself spans an exponentially large label space. In the talk, I will present two algorithms each designed to the minimize a given task loss, and applied to speech recognition. In the first part of the talk, I will present an algorithm which aims at achieving a high area under the receiver operating characteristic (ROC) curve. This measure of performance is often used to evaluate the detection of events over time. We will give some theoretical bounds that relate the performance of the algorithm with the expected area under the ROC curve, and will demonstrate its efficiency to that task of keyword spotting, i.e., that detection of all occurrences of any given word in a speech signal. In the second part of the talk, I will describe a new algorithm which aims to minimize a regularized task loss. The algorithm is derived by directly minimizing a generalization bound for structured prediction, which gives an upper-bound on the expected task loss in terms of the empirical task loss. The resulting algorithm is iterative and easy to implement, and as far as we know, the only algorithm that can handle a non-separable task loss. We will present experimental results on the task of phoneme recognition, and will show that the algorithm achieves the lowest phoneme error rate (normalized edit distance) compared to other discriminative and generative models with the same expressive power.

Detection of edges in images is a fundamental task in low level image processing. Edges are important as they mark the locations of discontinuities in depth, surface orientation, or reflectance. Their detection can facilitate a variety of tasks including image segmentation and object recognition, with many applications ranging from medical to security. In this talk we focus on accurate detection of faint, low-contrast edges in very noisy images. This challenging problem raises some fundamental statistical and computational questions, for which we shall provide some (partial) answers: What are detection limits and how do these depend on the complexity of the assumed edges ? What are computationally efficient methods to detect various families of edges ? and finally, how well can one detect edges under severe computational constraints - namely with sub-linear complexity in the number of image pixels. Joint work with Inbal Horev, Sharon Alpert, Meirav Galun, Ronen Basri (WIS) and with Ery Arias-Castro (UCSD).

Workflow Satisfiability Problem (WSP) is defined as follows. We are given sets $S

{s_1,ldots ,s_k}$ of steps and $U

{u_1,ldots ,u_n}$ of users. For every step $s_i$ we have a list $L(s_i)$ users that are allowed to do it. There are relations $rho_jsubseteq Utimes U$ and some constraints of the form $(rho_j, S', S'')$ where $S',S''subseteq S$.

We are to decide whether there is a function $pi: S rightarrow U$ satisfying the following conditions:

* $pi(s_j)in L(s_j)$;

* Each constraint $(rho_j, S', S'')$ must be satisfied, i.e., there are $s'in S', s''in S''$ such that $(pi(s'),pi(s''))in rho_j$.

Let k-WSP be WSP with parameter k (k is often small in practice).

Wang and Li (2010) showed that (i) WSP is NP-complete, (ii) k-WSP is W[1]-hard (iii) WSP(

,neq) is fixed-parameter tractable. Here $

$ and $neq$ are relations ${(s,s): sin S}$ and ${(s,s'): sneq s'in S}$, respectively.

We 'significantly' improved the FPT-algorithms of Wang and Li and proved that our algorithms cannot be 'significantly' improved any further unless the Exponential Time Hypothesis fails, extended the set of relations, and investigated some cases when WSP has polynomial kernel or not (subject to a well-known complexity theory hypothesis). Our results also improve some results of Fellows et al. (2011) on a problem related to WSP(neq).

Joint work with Jason Crampton (Information Security Group, RHUL) and Anders Yeo (Singapore Univ. Design & Technology) accepted for publication at ACM Transactions on Information and System Security; preliminary version in ACM Conference on Computer and Communications Security 2012, 857-868.

Given m copies of the same problem, does it take m times the amount of resources to solve these m problems? This is the direct sum problem, a fundamental question that has been studied in many computational models. In the first part of the talk we will focus on the advice complexity of this kind of questions, in a probabilistic Turing machine model, and discuss how many bits of "help" we need to accept such languages. In the second part we will consider several questions inspired by the direct-sum problem in (two-party) communication complexity. In all questions, there are $k$ fixed Boolean functions $f_1,dots,f_k$ and Alice and Bob have $k$ inputs $x_1,dots,x_k$ and $y_1,dots,y_k$, respectively. We consider In the {em eliminate} problem, Alice and Bob should output a vector $sigma_1,dots,sigma_k$ such that $f_i(x_i)neq sigma_i$ for at least one $i$ (i.e., their goal is to eliminate one of the $2^k$ output vectors); in {em choose}, Alice and Bob should return $(i,f_i(x_i,y_i))$ and in {em agree} they should return $f_i(x_i,y_i)$, for some $i$. The question, in each of the three cases, is whether one can do better than solving one instance.

Living organisms and intelligent agents are characterized by the flow of sensory information from a stochastic environment and the value of their decisions and actions. We first argue that the future value can be related, in Partially Observed Markov Decision Processes (POMDP), to the information needed for controlling the process. This leads to a simple relationship between sensory information, control information and reward rate, which completely characterizes metabolic information processing. Next, we argue that efficient planning and learning are related to the predictive information (mutual information between past and future) of the environment. We further argue that the sub-extensive nature of predictive information is responsible for the emergence of hierarchies and reverse hierarchies in planning and perception, the fundamental building blocks of cognition.

Proteins are the major building blocks of all living things; they mediate almost all biochemical reactions; and are the targets of almost all medicines. Proteins exert their biological function through unique three dimensional structures, encrypted in a linear genetic code. This linear coding of three dimensional structures is one of the hallmarks of life on earth. Deciphering the code would provide deep insights into fundamental questions in biology, and at the same time have major impact on medicine. This task, known as the protein structure prediction problem, has been a major challenge in computational biology over the last four decades. In my talk, I will give a brief introduction to the protein structure prediction problem and then present the way my group approaches it emphasizing software engineering and machine learning aspects. The talk is self-contained and does not require biological background.

We offer a new perspective on the nearest neighbor classifier, which yields tighter risk asymptotics than the classic Cover-Hart analysis. As a by-product, our analysis suggests a natural solution to the problem of noisy labels/outliers. Our result holds in doubling metric spaces, of which Euclidean spaces are a special case. The classifier may be learned in linear time and evaluated on new points in logarithmic time via approximate nearest neighbors. Time permitting, we'll discuss recent results on metric dimensionality reduction as well. Joint work with Lee-Ad Gottlieb and Robert Krauthgamer

The bulk of previous research on global image statistics has focused either on first order statistics such as luminance and contrast distributions, or on second order statistics such as the autocorrelation, or equivalently the power spectrum. While numerous links between global statistics of natural images and the functional architecture of biological visual systems have been established, it is widely agreed that perceptually important details such as edges and object boundaries are not captured by such low order statistics but rather by higher order ones. In this talk, I'll address the global first and second order statistics, but unlike previous studies, we consider them jointly. For this purpose we have developed an algorithm that produces random images with prescribed first and second order statistics. That is, given a natural image, the algorithm generates a new image with a similar distribution and spatial correlations of color as those of the original one, but otherwise being random. Our surprising observation is that the original images might be reconstructed only from their low order statistics. We conclude that the perceptual information content of first and second order statistics of natural images, when considered jointly, is greater than has been previously appreciated, and therefore might be utilized for efficient representations and processing of visual information.A joint work with prof. Ohad Ben-Shahar.

In many fields of application we encounter very large graphs. Because of their size and for many other practical reasons, it is completely infeasible to calculate complicated graph parameters of such graphs. Yet, and since we still want to extract some useful information from the data, what should we do? Current practices in this area are very unsatisfactory and often concentrate on easy-to-compute but fairly uninformative parameters such as the degree distribution of the vertices. Another approach which is also practiced in certain circles is take a ``local" view of the graph. Namely, fix some small integer k; sample at random k-tuples of veritces and look at the distribution of the induced k-vertex subgraphs. This leads to many interesting questions, on some of which we now have fairly satisfactory answers, while others are still very hard and challenging: 1. To what extent do local profiles characterize the big graph? 2. What local profiles are possible? 3. What global conclusions can be drawn from local information?

This talk focuses on the classic problem of Matrix Rigidity, introduced by Valiant '77, motivated by algebraic circuit lower bounds. We suggest a new route for resolving the problem by reducing it to the problem of efficiently constructing a hitting set for a class of polynomials, which we call U-polynomials. To showcase the reduction, we deduce that (large..) small-bias sets induce rigid matrices. We also show how to construct (large..) rigid sets from unbalanced expanders. Joint work with Noga Alon. No prior knowledge is assumed.

Real-time compression and deduplication for primary storage is quickly becoming widespread as data continues to grow exponentially, but adding compression and deduplication on the data path consumes scarce CPU and memory resources on the storage system. In this talk we present different approaches to efficient estimation of the potential data reduction ratio of data and how these methods can be applied in advanced storage systems. The main focus is on compression ratio evaluation where we employ two filters: The first level of filtering that we employ is at the data set level( e.g., volume or file system), where we estimate the overall compressibility of the data at rest. According to the outcome, we may choose to enable or disable compression for the entire data set, or to employ a second level of finer-grained filtering. The second filtering scheme examines data being written to the storage system in an online manner and determines its compressibility. We also discuss the challenges in achieving similar results when deduplication is involved and suggest alternatives for this scenario.

Histogram distance functions are the cornerstone of numerous computer vision and machine learning tasks (e.g. image retrieval, descriptor matching and k-nearest neighbor classification). It is common practice to use distances such as the Euclidean and Manhattan norms to compare histograms. This practice assumes that the histogram domains are aligned. However, this assumption is violated through quantization, shape deformation, light changes, etc. The Earth Mover’s Distance (EMD) is a cross-bin distance that addresses this alignment problem. We present several new Earth Mover's Distance variants that are robust to outlier noise and global deformations. Additionally, we present efficient algorithms for their computation. We show state-of-the-art results for descriptor matching and image retrieval. These tools have already been used by other groups and demonstrated state-of-the-art results for a range of tasks such as superpixel matching, descriptor matching, image retargeting, image segmentation, social graph comparisons and population density comparison. Finally, we describe several future directions including learning with general (not necessarily positive-semidefinite) similarity functions.

We describe a framework for specifying and analyzing complex system models composed of heterogeneous components using an SMT (Satisfiability Modulo Theories) based approach. We apply this method to the emerging fields of Synthetic biology and DNA Computing, and more broadly to study Biological Computation. This work highlights biological engineering as a domain that can benefit extensively from the application of software engineering and formal methods, while some of the methods and challenges addressed have the potential to influence development of novel tools for system and software engineering.

Parallelizing compilers are known to work well for scientific computations, such as matrix and vector manipulations. However, modern software has long moved on. The current landscape of applications written in high-level languages, like Java and C#, introduces new and important challenges. The compiler needs to handle multiple levels of abstraction, extensive pointer-based computations over dynamic memory, unbounded data structures, and deployment-specific program behaviors. Data dependence analysis – tracking whether statements in the sequential code depend on each other due to interfering accesses to some shared memory location – has been the foundation of parallelizing compilers. The fundamental observation, going (at least) 30 years back, is that two code blocks that have no (transitive) data dependencies can be executed in parallel, resulting in the same final state as running the codes sequentially. For all the challenges listed above, only in rare cases are parallelizing compilers able to prove this precondition: The candidate code blocks are often dependent, and even if not, the compiler's (static) dependence analysis is typically too conservative to prove independence, failing due to spurious dependencies. I will propose a new view of data dependencies, showing that they remain a useful tool for reasoning about parallelism, albeit in different ways and different form. I will show how data dependencies can be lifted to higher levels of abstraction, per the abstractions governing the structure of the application, as well as how effective use can be made of partial yet precise dependencies to guide the behavior a parallelization system while preserving its correctness (i.e. serializability guarantees). This can be done in more than one way, including (i) building specialized, client-specific conflict-detection oracles, (ii) synthesizing concurrency monitors that predict the available parallelism per input data and/or computation phase, and (iii) finding true, semantic dependencies that limit parallelism.

Driven by the recent surge of personal mobile devices we explore the means to apply the accumulated knowledge on distributed search to realistic multi agent applications. Our work focuses on interactions where the agents are cooperative (follow the distributed interaction protocol) but have personal preferences (evaluate the outcome of a joint action differently).

The talk will provide a bird’s eye view on my Ph.D. dissertation and will begin by discussing the representation of multi agent interactions where agents are cooperative but have personal preferences. Although the standard model can capture such interactions, it limits many existing algorithms – a problem remedied by introducing asymmetric constraints.

Applying the asymmetric model to the problems at hand I will then go on to describe the implications of personal preferences on the search objective. Traditionally this objective takes one of two forms: “cooperative” - the utilitarian social welfare, for example; or “competitive” - a game theoretic stable joint action (equilibrium). I will show that both objectives are attainable within the new asymmetric distributed constraint setting but also suggest some alternative concepts which attempt to leverage the inherent agents’ cooperation in an attempt to provide a partial answer to the seemingly simple philosophical question: “what is a fair solution?”.

The processor landscape has fractured into latency-optimized CPUs, throughput-oriented GPUs, and soon, custom accelerators. Future applications will need to cohesively use a variety of hardware to achieve their performance and power goals. However building efficient systems that use accelerators today is incredibly difficult.

In this talk I will argue that the root cause of this complexity lies in the lack of adequate operating system support for accelerators. While operating systems provide optimized resource management and Input/Output (I/O) services to CPU applications, they make no such services available to accelerator programs.

I propose GPUfs - an operating system layer which enables access to files directly from programs running on throughput-oriented accelerators, such as GPUs. GPUfs extends the constrained GPU-as-coprocessor programming model, turning GPUs into first-class computing devices with full file I/O support. It provides a POSIX-like API for GPU programs, exploits parallelism for efficiency, and optimizes for access locality by extending a CPU buffer cache into physical memories of all GPUs in a single machine.

Using real benchmarks I show that GPUfs simplifies the development of efficient applications by eliminating the GPU management complexity, and broadens the range of applications that can be accelerated by GPUs. For example, a simple self-contained GPU program which searches for a set of strings in the entire tree of Linux kernel source files completes in about third of the time of an 8-CPU-core run.

Joint work with Idit Keidar (Technion), Bryan Ford (Yale) and Emmett Witchel (UT Austin)

We would like to know the distances to moving objects and their velocities. The radar system is built to fulfill this task. The radar transmits a waveform S which bounds back from the objects and the echo R is received. In practice we can work in the digital model, namely S and R are sequences of N complex numbers (e.g., N=1023). THE RADAR PROBLEM IS: Design S, and an effective method of extracting, using S and R, the distances and velocities of all targets. In many applications the current sequences S which are used are pseudo-random and the algorithm they support takes O(N²logN) arithmetic operations. In the lecture we will introduce the Heisenberg sequences, and a much faster detection algorithm called the Cross Method. It solves the Radar Problem in O(NlogN+m²) operations for m objects. This is a joint work with S. Gurevich (Math, Madison), A. Sayeed (EE, Madison), K. Scheim (General Motors, Herzeliya), O. Schwartz (EECS, Berkeley).

A large number of Natural Language Processing applications, including syntactic parsing, information extraction and discourse analysis, involve the prediction of a linguistic structure. It is often times challenging for standard feature-based machine learning algorithms to perform well on these tasks due to modeling and computational reasons. Moreover, creating the large amounts of manually annotated data required to train supervised models for such applications is usually labor intensive and error prone. In this talk we describe a serious of works that integrate feature based methods with declarative task and domain knowledge in a unified framework. We address a wide variety of NLP tasks and domain knowledge: for syntactic parsing we show how to parse multiple sentences together while imposing consistency constraints, for information extraction we present a joint model that ties together a number of related tasks through task and domain constraints and for discourse analysis we present a model that exploit within and cross document regularities in a collection of documents. Our models are implemented in the Markov Random Field (MRF) framework and the resulted global hard optimization task is addressed by approximate inference techniques based on linear programming (LP) relaxations. We present improvements over state of the art models in five languages and a wide range of supervision levels - from fully unsupervised to fully supervised scenarios.

One of the main problems in Statistical Learning Theory is to estimate an unknown function by a function from a given family, using random data. Although seemingly unrelated, it turns out that this question has very strong connections to problems in Asymptotic Geometric Analysis. In this talk I will present some of these connections, and show why some natural problems in Asymptotic Geometric Analysis (e.g., embedding theorems, estimates on the singular values of random matrices, etc), are essential in the study of the basic problems in Learning Theory.

High throughput DNA sequencing has transformed the landscape of genomic data by providing an affordable means to sequence the genomes of numerous species and multiple individuals per species. There has been a particularly dramatic increase in the last five years in the availability of individual human genomes and the genomes of closely related primate species. These data provide a rich source of information about human evolution and the forces that helped shape the human genome. This talk will focus on two specific problems I explored during my postdoctoral research using these new data sets.

The first problem I will be presenting is recovery of ancient human demography and the evolutionary relationships between different human population groups. I developed a new method, called G-PhoCS (Generalized Phylogenetic Coalescent Sampler) to infer demographic parameters along a population tree, and applied this method to the complete genomes of six human individuals from major human population groups. This approach allowed us to recover very ancient trends in human demography dating back as far back as 130 thousand years ago. The second problem is that of inferring signatures of recent natural selection acting on regulatory elements in the human genome. Much of the DNA in the human genome is devoted to the regulation of gene expression, but regulatory DNA elements are typically short, dispersed and often not conserved across long evolutionary timescales. This has made it very difficult for researchers to study the selective constraints that act on regulatory DNA in the human genome. I developed a method, called INSIGHT (Inference of Natural Selection from Interspersed Genomically coHerent elemenTs), that addresses these challenges by making use of individual human genomes and the genomes of closely related primates. This method was used to perform the first comprehensive study of natural selection acting on transcription factor binding sites, which are the most well characterized regulatory elements in the human genome. Our study sheds light on the selective forces that shaped these elements, and has possible implications to the study of human disease.

This talk will highlight the methodological and algorithmic challenges in these problems, and will not require any prior biological knowledge.

Graph products are a basic combinatorial object, introduced by Alfred Whitehead and Bertrand Russell in their 1912 Principia Mathematica. They are widely studied and used in applications in different fields, for example information theory, and hardness of approximation. In the first part of the talk, motivated by a problem in communication, we present a new type of graph product, which we call the Bi-partite Graph Product, or BGP for short. We characterize the spectrum of the BGP, and derive two results: an edge discrepancy result, and a new explicit construction of a co-spectral graph family. In the second part of the talk we show how to use the well-known Tensor graph product to obtain a family of uniquely k-colorable graphs. Our main tool of analysis is Connelly's condition in Rigidity Theory combined with an explicit construction and algebraic calculations of the rigidity stress matrix. Based on joint works with Michael Langberg and Igor Park.

Pebble automata (PA) over infinite alphabets are an extension of the classical PA. They were introduced by Neven, Schwentick, and V. Vianu about ten years ago and recently found applications in XML. The notion of PA over infinite alphabets is very robust. In addition to equivalence of various models of PA, the set of languages they accept is closed under all Boolean operations. However, the emptiness problem for these languages is undecidable for three or more pebbles. Reducing Hilbert's tenth problem to the emptiness problem for 2-PA languages, we prove that the latter is also undecidable. In addition, we present an example of a 3-PA language that is not accepted by 2-PA, i,e., 2-PA automata are weaker than 3-PA. Finally, we show that 2-PA can accept a non-semi-linear language. The talk does not presume any prior knowledge of models of computation over infinite alphabets. Joint work with Tony Tan.

Given three permutations on the integers 1 through n, consider the set system consisting of each interval in each of the three permutations. In 1982, Beck conjectured that the discrepancy of this set system is O(1). In other words, the conjecture says that each integer from 1 through n can be colored either red or blue so that the number of red and blue integers in each interval of each permutation differs only by a constant. (The discrepancy of a set system based on two permutations is at most two.) Our main result is a counterexample to this conjecture: for any positive integer n = 3^k, we construct three permutations whose corresponding set system has discrepancy at least k/3.

Time permitting I will also discuss an implication of this result to the integrality gap of the Gilmore-Gomory LP relaxation for Bin Packing.

Joint with Alantha Newman and Aleksandar Nikolov

Faster algorithms can be obtained by minimizing communication. That is, reducing the amount of data sent across the memory hierarchy and between processors. The communication costs of algorithms, in terms of time or energy, are typically much higher than the arithmetic costs. We have computed lower bounds on these communication costs by analyzing geometric and expansion properties of the underlying computation DAG of algorithms. These techniques (honored by SIAG-LA prize for 2009-2011 and SPAA'11 best paper award) inspired many new algorithms, where existing ones proved not to be communication-optimal. Parallel matrix multiplication is one of the most studied fundamental problems in parallel computing. We obtain a new parallel algorithm based on Strassen's fast matrix multiplication that is communication-optimal. It exhibits perfect strong scaling, within the maximum possible range. The algorithm asymptotically outperforms all known parallel matrix multiplication algorithms, classical and Strassen-based. It also demonstrates significant speedups in practice, as benchmarked on several super-computers (Cray XT4, Cray XE6, and IBM BG/P). Our parallelization approach is simple to implement, and extends to other algorithms. Both the lower bounds and the new algorithms have an immediate impact on saving power and energy at the algorithmic level. Based on joint work with Grey Ballard, James Demmel, Olga Holtz, Ben Lipshitz

This talk will briefly introduce the streaming computational model in the context of data mining. We will focus on working with matrices that are revealed over time and are too large to store. Working with such massive matrices requires creating a concise yet accurate approximation for them. These are called matrix sketches. This talk will shortly survey new results for matrix sketching in two streaming models. In the first, the matrix is presented to the algorithm one entry at a time. Examples include recommender systems whose input is of the form "user i rated item j 3 stars". In the second, the matrix is revealed row by row, for example, "document i contains terms j_1,…,j_k". This is the case in web crawling where the crawler cannot store all the documents it visits.

We present a method for structural summarization and abstraction of complex spatial arrangements found in architectural drawings. The method is based on the well-known Gestalt rules, which summarize how forms, patterns, and semantics are perceived by humans from bits and pieces of geometric information. Although defining a computational model for each rule alone has been extensively studied, modeling a conjoint of Gestalt rules remains a challenge. In this work, we develop a computational framework which models Gestalt rules and more importantly, their complex interactions. We apply {em conjoining rules} to line drawings, to detect groups of objects and repetitions that conform to Gestalt principles. We summarize and abstract such groups in ways that maintain structural semantics by displaying only a reduced number of repeated elements, or by replacing them with simpler shapes. We show an application of our method to line drawings of architectural models of various styles, and the potential of extending the technique to other computer-generated illustrations, and three-dimensional models.

Cloud computing is transforming IT. The way IT deploys infrastructure, provisions resources, builds applications and provides access to these applications is changing dramatically, promising both greater efficiency and agility. At the same time, many businesses now see data volumes growing so large they break traditional infrastructures – this is what EMC calls Big Data.

Managing Big Data represents a huge challenge for IT, but using analytics to provide insight into Big Data promises to transform business, delivering competitive advantage and breakthrough business value. Cloud computing and Big Data will only be viable if they are underpinned by a foundation of trust – trust that the infrastructure will be available 24x7 and trust that data will not be compromised by the latest advanced threat.

Brian Gallagher will discuss these trends and the changes necessary to move to a new model – one which requires IT to transform to an internal service provider, shifting focus from infrastructure cost to business agility.

New technologies and new use paradigms are revolutionizing IT and creating challenging technical opportunities. New disk technologies, ranging from slow and inexpensive large SATA drives to very fast and very expensive solid state drives create unique opportunities to improve performance and reduce cost simultaneously. Automatic Tiered Storage is a promising new technology, but it is in its infancy, and a lot of more research is required to explore its full potential. The move to virtual data centers and Cloud, and with that the elimination of the role of the expert storage administrator, requires a leap in how storage systems are managed. Storage systems need to be wise enough to automatically achieve performance goals, be aware of their environments and improve over time. The talk will focus on the new challenges that IT vendors like EMC face today in optimizing the use of the new technologies and addressing the needs of the new users of IT. Another focus is on how academia can help us address these challenges.

MicroRNAs (miRNAs) are an abundant class of small non-coding RNAs that can affect gene expression by post-transcriptional regulation of mRNAs. miRNAs have been shown to play important roles in various cellular and pathogenic processes. Some viral organisms also encode miRNAs, a fact that contributes to the complex interactions between viruses and their hosts. miRNAs down-regulate translation of genes via imperfect binding of the miRNA to a specific site or sites on their coding transcripts. A critical step in miRNA functional studies is to identify the target genes that are directly regulated by miRNAs. The current target prediction tools have two major limitations. First, they are time consuming for large datasets. Second, they are noisy and predict an excess of targets for each miRNA. To overcome the first limitation, we developed our own method for target prediction, which extends the "threshold all-against-all" sequence alignment algorithm. To get over the second limitation, we developed methods, related to bi-clustering, that combine target prediction results with new host-viral post transcriptional regulation modes and additional information sources, and narrow the list of target genes to more reliable candidates. Our algorithm, called bi-targeting, searches for modules of miRNAs (host and viral) and their common host target genes that have a similar biological function. We applied it to the discovery of modules consisting of human and Epstein-Barr virus (EBV) miRNAs and human genes, in a cooperating mode of regulation. Later we relaxed the bi-targeting algorithm to compute quasi-modules, where many more interactions were captured and reported. We used our new relaxed bi-targeting algorithm to study the compensating regulation of miRNAs in Human cytomegalovirus (HCMV) infection, using new expression data of miRNAs in HCMV infected vs. un-infected cells. Since not much is known about the function of viral miRNAs, finding modules that link the viral miRNAs and the human miRNAs, might help in understanding the role of miRNAs in host-viral interactions. Furthermore, since the identification and validation of miRNA targets remains a hard problem, focusing on small sets of miRNAs and their effects on particular biological pathways may give a significant advantage in target identification.

The increasing availability of Electronic Health Record (EHR) data and specifically free-text patient notes presents opportunities for the extraction of phenotypes, treatment and treatment outcome on a large scale. These data can make a significant contribution to basic science in many fields that require detailed phenotypic information such as linking phenotypes to genetic variance. General purpose text processing tools perform poorly on text in the medical domain, because the medical language uses specific words, word distributions and syntactic constructs. I present three domain-adaptation techniques to help adapt existing text processing tools to the medical domain, with specific attention to Hebrew:

a) Medical text uses many technical terms to refer to anatomy, biology or diseases. Most Medical-NLP tools rely on the UMLS, a medical vocabulary with over 300K unique terms and more than 1M synonyms. We present a method for automatically creating a Hebrew-UMLS lexicon. We show that acquiring this resource reduces the error for the NLP tasks of segmentation and Part of Speech (POS) tagging. We examine the impact of this improvement on a classification task: identifying patients with Epilepsy from the notes of the Children Neurology Unit in Soroka, resulting in F1 improvement from 92% to 96%.

b) EHR text is characterized by high-level of copy-paste redundancy in medical notes of the same patients. We quantatively show that this type of redundant word distribution is highly prevalent in both Hebrew and English notes and empirically demonstrate that this characteristics of medical notes collections has a deleterious effect on classical NLP algorithms. We present a novel algorithm for Topic Modeling with Latent Dirichlet Allocations (LDA) which is immune to the redundancy noise. This algorithm also performs better than the baseline for redundant news reports clusters.

c) Syntactic dependency parsing is a useful technique for Information Extraction, widely used in the biomedical domain. However, syntactic parsers suffer a major decline in accuracy when used in a domain different from the training data. We present a method for using Selectional Preferences, the affinity of different word pair or triplets, modeled with LDA to improve dependency parsing using un-annotated data in the target domain with significant improvement.

Taken together, the techniques provide infrastructure which allows practical processing of medical text in Hebrew. We make available a first set of language resources for Hebrew medical text processing (treebank, lexicon, part of speech tagger, syntactic parser, topic modeling toolkit). This infrastructure has been applied for practical text mining of hospital patient reports.

First is the FiniteSat algorithm, an efficient detection method for finite satisfiability problems in UML class diagrams. I will sketch the main arguments for its correctness and scope. The algorithm is strengthened by a propagation method for implied missing constraints. Next, I will present an identification method which points to the causes for finite satisfiability problems. > The second contribution of my research deals with analysis of inter-association constraints in UML class diagrams. These constraints although intensively used in the UML meta-model, have obscure and contradictory semantics. The analysis reaches semantic agreements based on constraint observables that minimize meta-model changes. This analysis has yielded recommendations and guidelines for modelers. > The last part of the talk relates to the human factor in modeling. We describe a catalog of anti-patterns that characterize correctness and quality problems in class diagrams. Formalization of the anti-patterns involves a template-oriented extension of the class diagram language. The catalog role was tested in a series of experiments. The novelty of the research lies in the integration of formal and educational methods for improving model design quality in class diagrams.

Privacy Preserving Data Publishing (PPDP) is a research field that deals with the development of methods to enable publishing of data while minimizing distortion, for maintaining usability on one hand, and respecting privacy on the other hand. Sequential release is a scenario of data publishing where multiple releases of the same underlying table are published over a period of time. A violation of privacy, in this case, may emerge from any one of the releases, or as a result of joining information from different releases. Similarly to [Wang and Fung 2006], our privacy definitions limit the ability of an adversary who combines information from all releases, to link values of the quasi-identifiers to sensitive values. We extend the framework that was considered in [Wang and Fung 2006] in three ways: We allow a greater number of releases, we consider the more flexible local recoding model of ``cell generalization" (as opposed to the global recoding model of ``cut generalization" in [Wang and Fung 2006]), and we include the case where records may be added to the underlying table from time to time. Our extension of the framework requires also to modify the manner in which privacy is evaluated. We show that while [Wang and Fung 2006] based their privacy evaluation on the notion of the Match Join between the releases, it is no longer suitable for the extended framework considered here. We define more restrictive types of join between the published releases (the Full Match Join and the Kernel Match Join) that are more suitable for privacy evaluation in this context. We then present a top-down algorithm for anonymizing sequential releases in the cell generalization model, that is based on our modified privacy evaluations. Our theoretical study is followed by experimentation that demonstrates a staggering improvement in terms of utility due to the adoption of the cell generalization model, and exemplifies the correction in the privacy evaluation as offered by using the Full or Kernel Match Joins instead of the Match Join.

A client on the earth surface wants to know his/her geographical location. The Global Positioning System (GPS) was built to ful…ll this task. It works as follows: Satellites send to earth their location. For simplicity, the location of a satellite is a bit b 2 f1g: The satellite transmits to the earth a sequence of N  1000 complex numbers S[0]; S[1]; :::; S[N 􀀀1] multiplied by its location b: The client receives the sequence R of the form R[n] = b
0
e 2i N !0
n
S[n +  0] +W[n]; (1) where 0 2 C is the complex amplitude, with j0j  1; !0 2 ZN encodes the radial velocity of the satellite with respect to the client,  0 2 ZN encodes the distance between the satellite and the client, and W is a random white noise. The GPS Problem is: Problem 1 (GPS Problem) Design S, and an e¤ective method of extracting (b;  0) from S and R satisfying (1). A client can compute his/her location by knowing the locations of at least three satellites and distances to them. The current sequences S which are used are pseudo-random and the algorithm they support takes O(N2 logN) arithmetic operations: In my lecture I will explain our recent construction of sequences S that allow us to introduce a much faster algorithm called the "Flag Method". It solves the GPS Problem in O(N logN) operations. This is a joint work with A. Fish (Mathematics, Madison), R. Hadani (Math- ematics, Austin), A. Sayeed (Electrical and Computer Engineering, Madison), and O. Schwartz (Electrical Engineering and Computer Science, Berkeley).

Our visual attention is attracted by salient stimuli in our environment and affected by primitive features such as orientation, color, and motion. Perceptual saliency due to orientation contrast has been extensively demonstrated in behavioral experiments with humans and other primates and is commonly explained by the very particular functional organization of the primary visual cortex. We challenge this prevailing view by studying orientation-based visual saliency in two non-mammalian species with enormous evolutionary distance to humans. The surprising results not only imply the need to reestablish our understanding of how these processes work at the neural level, but they also suggest that orientation-based saliency has computational optimality in a wide variety of ecological contexts, and thus constitutes a universal building block for efficient visual information processing in general.

We show that an ``old dog'', namely – the classical Johnson-Lindenstrauss transform, ``performs new tricks'' – it gives a novel way of preserving differential privacy. We show that if we take two databases, D and D', such that (i) D'-D is a rank-1 matrix of bounded norm and (ii) all singular values of D and D' are sufficiently large, then multiplying either D or D' with a vector of iid normal Gaussians yields two statistically close distributions in the sense of differential privacy. Furthermore, a small, deterministic and public alteration of the input is enough to assert that all singular values of D are large. We apply the Johnson-Lindenstrauss transform to the task of approximating cut-queries: the number of edges crossing a (S,bar S)-cut in a graph. We show that the JL transform allows us to publish a sanitized graph that preserves edge differential privacy (where two graphs are neighbors if they differ on a single edge) while adding only O(|S|/epsilon) random noise to any given query (w.h.p). Comparing the additive noise of our algorithm to existing algorithms for answering cut-queries in a differentially private manner, we outperform all others on small cuts (|S| = o(n)). We also apply our technique to the task of estimating the variance of a given matrix in any given direction. The JL transform allows us to publish a sanitized covariance matrix that preserves differential privacy w.r.t bounded changes (each row in the matrix can change by at most a norm-1 vector) while adding random noise of magnitude independent of the size of the matrix (w.h.p). In contrast, existing algorithms introduce an error which depends on the matrix dimensions.

Despite their gross simplicity, random (linear) codes gained great fame with their use in Shannon's 1948 noisy-channel coding theorem. In this talk, we survey three of our applications of these codes to the field of networks. Our first work is a concrete implementation of random coding for over-the-air programming of sensor motes. Our second work relates to fair and secure bandwidth sharing of assymetric channels, where we utilize a game-theoretic framework to craft a protocol resistant to maliciously colluding parties. Our third application involves the use of extreme value theory to predict system-level error rates for wireless broadcast. Each case serves to demonstrate the astonishing power that can be harnessed from these amazingly simple codes.

Inferring drug-associated traits such as drug targets, drug indications and drug-drug interactions is essential for drug development and drug administration. I will present a novel method for the large-scale prediction of such traits able to handle both approved drugs and novel molecules. Our method is based on the observation that similar drugs tend to associate with similar traits and utilizes various similarity measures for the prediction task. Furthermore, our method is utilized for inferring causative factors of drug-drug interactions and can be extended to handle personalized medicine.

In this talk we consider weak and strong games played on the edge set of the complete graph $K_n$. Given a graph $G=(V,E)$ and a graph property P, in the weak game $P$ two players, called Maker and Breaker, alternately claim edges from $E$. Maker wins the game as soon as the graph spanned by his edges possesses the property P. If by the time all the edges have been claimed Maker does not win, then the game ends in a draw. In the strong game $P$ two players, called Red and Blue, alternately claim edges from $E$. The winner is the FIRST player whose graph possesses $P$. We consider the $k$-vertex-connectivity game, played on the edge set of $K_n$. We first study the weak version of this game and prove that, for any positive integer $k$ and sufficiently large $n$, Maker has a strategy for winning this game within $lfloor k n/2 rfloor + 1$ moves which is clearly best possible. This answers a question of Hefetz, Krivelevich, Stojakovich and Szabo. We then consider the strong $k$-vertex-connectivity game. For every positive integer $k$ and sufficiently large $n$, we describe an explicit first player's winning strategy for this game. this is a joint work with Dan Hefetz.

The disagreement coefficient of Hanneke has become a central data independent invariant in proving active learning rates. It has been shown in various ways that a concept class with low complexity together with a bound on the disagreement coefficient at an optimal solution allows active learning rates that are superior to passive learning ones. We present a different tool for pool based active learning which follows from the existence of a certain uniform version of low disagreement coefficient, but is not equivalent to it. In fact, we present two fundamental active learning problems of significant interest for which our approach allows nontrivial active learning bounds. However, any general purpose method relying on the disagreement coefficient bounds only fails to guarantee any useful bounds for these problems. The tool we use is based on the learner's ability to compute an estimator of the difference between the loss of any hypotheses and some fixed "pivotal" hypothesis to within an absolute error of at most $eps$ times the $ell_1$ distance (the disagreement measure) between the two hypotheses. We prove that such an estimator implies the existence of a learning algorithm which, at each iteration, reduces its excess risk to within a constant factor. Each iteration replaces the current pivotal hypothesis with the minimizer of the estimated loss difference function with respect to the previous pivotal hypothesis. The label complexity essentially becomes that of computing this estimator. The two applications of interest are: learning to rank from pairwise preferences, and clustering with side information (a.k.a. semi-supervised clustering). They are both fundamental, and have started receiving more attention from active learning theoreticians and practitioners. Joint work with Ron Begleiter and Esther Ezra

In 1934 H. Whitney posed the following problem: Let $f$ be a function defined on a closed subset $E$ of $R^n$. How can we tell whether $f$ extends to a $C^m$-smooth function defined on all of $R^n$? We discuss different aspects of this classical problem including its interesting connections with Convex Geometry (Helly's theorem), Lipschitz selections of set-valued functions and Analysis on Riemannian manifolds. The main part of the talk will be addressed to the "finiteness principal" which states that the Whitney extension problem can be reduced to the same kind of the problem, but for finite sets with prescribed numbers of points. We will present several constructive criteria for restrictions of $C^2$-functions and Sobolev $W^1_p$ and $W^2_p$-functions to arbitrary closed subsets of $R^2$.

The k-server problem is one of the most fundamental and extensively studied problems in online computation. Suppose there is an n-point metric space and k servers are located at some of the points of the metric space. At each time step, an online algorithm is given a request at one of the points of the metric space, and this request is served by moving a server to the requested point (if there is no server there already). The cost of serving a request is defined to be the distance traveled by the server. Given a sequence of requests, the task is to devise an online strategy minimizing the sum of the costs of serving the requests. We give the first polylogarithmic-competitive randomized online algorithm for the k-server problem on an arbitrary finite metric space. In particular, our algorithm achieves a competitive ratio of O(log^3 n log^2 k) for any metric space on n points. Our algorithm improves upon the deterministic (2k-1)-competitive algorithm of Koutsoupias and Papadimitriou whenever n is sub-exponential in k. Joint with Nikhil Bansal, Aleksander Madry and Seffi Naor

We present a novel framework for the query-performance prediction task. That is, estimating the effectiveness of a search performed by a search engine in response to a query in lack of relevance judgments. The framework is based on estimating the utility that a given document ranking provides with respect to an information need expressed by the query. To address the uncertainty in inferring the information need, we estimate the utility by the expected similarity between the given ranking and those induced by relevance language models. Specific query-performance predictors instantiated from the framework are shown to substantially outperform state-of-the-art predictors. In addition, we present an extension of the framework that results in a unified prediction model that can be used to derive and/or explain several previously proposed post-retrieval predictors which are presumably based on different principles.

Monitoring, interpretation, and analysis of large amounts of time-stamped data are tasks that are at the core of tasks such as the management of chronic patients, the detection of malware, or the integration of Intelligence data from multiple sources, and the related task of learning new knowledge from the accumulating data. In the case of the medical domain, these tasks are crucial for chronic-patient care, retrospective quality assessment, and clinical research. I will briefly describe several conceptual and computational architectures developed over the past 20 years, mostly by my research teams at Stanford and Ben Gurion universities, for knowledge-based performance of these tasks, and will highlight the complex and interesting relationships amongst them. Examples of such architectures include the IDAN and Momentum goal-directed and data-driven temporal-abstraction architectures, the KNAVE-II and VISITORS interactive-exploration frameworks for single and multiple longitudinal records, and the KarmaLego temporal data mining methodology. I will also point out the progression from individual-subject monitoring and therapy, to multiple-patient aggregate analysis and research, and finally to the learning of new knowledge. This progression, however, can also be viewed as a positive-feedback loop, in which new knowledge is brought back to bear upon both individual-patient management as well as on the learning of new and meaningful (temporal) associations.

In many real-world problems, a big scale gap can be observed between micro- and macroscopic scales of the problem because of the difference in mathematical (engineering, social, biological, physical, etc.) models and/or laws at different scales. The main objective of the multiscale algorithms is to create hierarchies of coarse problems, each representing the original problem at different scales with fewer degrees of freedom. We will discuss different strategies of creating these hierarchies and other components of multiscale methods for several large-scale applications: linear ordering for mapping problems in HPC, response to infection spread and cyber attacks, network compression, graph partitioning/clustering, etc.

Distance oracle is a data structure that provides fast answers to distance queries. Recently, the problem of designing distance oracles capable of answering restricted distance queries, that is, estimating distances on a subgraph avoiding some forbidden vertices, has attracted a lot of attention. In this talk, we will consider forbidden set distance oracles for planar graphs. I’ll present an efficient compact distance oracle that is capable of handing any number of failures. In addition, we will consider a closely related notion of fully dynamic distance oracles. In the dynamic distance oracle problem instead of getting the failures in the query phase, we rather need to handle an adversarial online sequence of update and query operations. Each query operation involves two vertices s and t whose distance needs to be estimated. Each update operation involves inserting/deleting a vertex/edge from the graph. I’ll show that our forbidden set distance oracle can be tweaked to give fully dynamic distance oracle with improved bounds compared to the previously known fully dynamic distance oracle for planar graphs. Joint work with Ittai Abraham and Cyril Gavoille

Many significant historical corpora contain leaves that are mixed up and no longer bound in their original state as multi-page documents. The reconstruction of old manuscripts from a mix of disjoint leaves can therefore be of a paramount importance to historians and literary scholars. In collaboration with the The Friedberg Genizah Project, we show that visual similarity provides meaningful pair-wise similarities between handwritten leaves and then go a step further and suggest a semi-automatic clustering tool that helps reconstruct the original documents. The proposed solution is based on a graphical model that makes inferences based on catalog information provided for each leaf as well as on the pairwise similarities of handwriting. Several novel active clustering techniques are explored, and the solution is applied to a significant part of the Cairo Genizah, where the problem of joining leaves remains unsolved even after a century of extensive study by hundreds of human scholars.

Augmented Reality is the technology and art of fusing virtual content into our reality. The technology was invented before the personal computer and runs nowadays on any smartphone. Yet, the “ultimate” Augmented Reality application that is as useful and as natural as taking a picture with a camera phone, have not been demonstrated so far.

We propose an intuitive framework that captures the challenges in creating disruptive AR experiences. We name this framework The Reality Mixing Continuum. The continuum spans between experiences that assume nothing in advance about the environment (the real world) and those that make strict assumptions about it. We discuss why assumptions about the environment are key in shaping an “ultimate” experience.

In the scope of that framework we describe our own quest, focused on Augmented Reality technological paradigms that enable natural experiences, while minimizing the assumptions about the environment. We describe three new paradigms: In-Place content creation with the premise of blending virtual and real content naturally. In-Place Sketch Interaction for manipulating augmented content. And, Out of the Cube, enabling a tangible interaction based on the Rubik’s Cube.

The presentation includes video demonstrations reviewing the latest and greatest in the field of Augmented Reality with the intention to introduce the field to the audience as well as to stimulate thinking about existing challenges.

Einstein didnt believe in quantum mechanics because "God does not play dice". We will show that one possible interpretation of Einstein's own law, governing free fall in space-time, is that god plays a considerable amount of dice. We arrived at this line of inquiry through our study of airplane boarding, which in turn started from a problem in disk scheduling which is a part of "practical" systems performance analysis, landing squarly in the realm of computer science with or without software engineering. We will try to survey the interconnections between all these topics. No prior knowledge is required, experience with boarding airplanes will be useful.

A novel approach is suggested for reducing traffic conflicts in at-grade (2D) urban networks. Intersections without primary vehicular conflicts are defined as zero traffic conflict (ZTC) designs. Complete classification of maximal ZTC designs is presented, including designs that combine driving on the right side in some streets and driving on the left side in other streets. It is proved that there are 9 four-way and 3 three-way maximal ZTC intersection designs, to within mirror, rotation, and arrow reversal symmetry. Vortices are used to design networks where all or most intersections are ZTC. Increases in average travel distance, relative to unrestricted intersecting flow, are explicitly calculated for grid-networks of sizes 10 by 10, 10 by 20 and 20 by 20 nodes with evenly distributed origins and destinations. The exact increases depend primarily on various short-range conditions, such as the access to the network. The average distance increase in most cases examined is up to four blocks. These results suggest that there is a potential for the new designs to be relevant candidates in certain circumstances, and that further study of them is worthwhile.

While many of the classical codes are cyclic, a long standing conjecture asserts that there are no 'good' cyclic codes. In recent years the intrest in symmetric codes has been promoted by Kaufamn, Sudan, Wigderson and others (where symmetric means that the acting group can be any group). Answering their main question (and in contrary to the common expectation), we show that there DO exist symmetric good codes. In fact, our codes satisfy all the "golden standards" of coding theory. Our construction is based on the Ramanujan graphs contructed by Lubotzky-Samuels-Vishne as a special case of Ramanujan complexes. The crutial point is that these graphs are edge transitive and not just vertex transitive as in prevous constructions of Ramanujan graphs. All notions will be explained. Joint work with Tali Kaufman.

We contribute a further step towards the plausible real time construction of suffix trees by presenting an on-line algorithm that spends O(log log n) time processing each input symbol and takes O(n log log n) time in total. Our results improve on a previously published algorithm that take O(log n) time per symbol and O(n log n) time in total. The improvements are achieved using a new data structure for the fringe marked ancestor problem, a special case of the nearest marked ancestor problem, which may be of independent interest. Joint work with Pino Italiano, University of Rome.

An important feature of web search interfaces are the restrictions enforced on input values – those reflecting either the semantics of the data or requirements specific to the interface. Both integrity constraints and “access restrictions” can be of great use to web exploration tools. We demonstrate here a novel technique for discovering constraints that requires no form submissions whatsoever. We work via statically analyzing the JavaScript client-side code used to enforce the constraints, when such code is available. We combine custom recognizers for JavaScript functions relevant to constraint-checking with a generic program analysis layer. Integrated with a web browser, our system shows the constraints detected on accessed web forms, and allows a user to see the corresponding JavaScript code fragment.

Making faultless complex objects from potentially faulty building blocks is a fundamental challenge in computer engineering, nanotechnology and synthetic biology. We show for the first time how recursion can be used to address this challenge and demonstrate a recursive procedure that constructs error-free DNA molecules and their libraries from error-prone oligonucleotides. Divide and Conquer (D&C), the quintessential recursive problem-solving technique, is applied in silico to divide the target DNA sequence into overlapping oligonucleotides short enough to be synthesized directly, albeit with errors; error-prone oligonucleotides are recursively combined in vitro, forming error-prone DNA molecules; error-free fragments of these molecules are then identified, extracted and used as new, typically longer and more accurate, inputs to another iteration of the recursive construction procedure; the entire process repeats until an error-free target molecule is formed. Our recursive construction procedure surpasses existing methods for de novo DNA synthesis in speed, precision, amenability to automation, ease of combining synthetic and natural DNA fragments, and ability to construct designer DNA libraries. It thus provides a novel and robust foundation for the design and construction of synthetic biological molecules and organisms.

All living cells use very large protein assemblies to carry out complex tasks. The structural understanding of these efficient nano-machines is often very limited, because conventional techniques like X-ray crystallography or cryo-EM cannot resolve them to sufficient resolution. Firsthand knowledge with cases where the low resolution resulted in limited or even wrong biological conclusions, had led us to the realization that completely unbiased methods are essential. To that end, we developed a combinatorial approach that exhaustively enumerates all the possible arrangements of the biological system and assesses them objectively against the structural data. I will present two such applications on a very suitable system: the eukaryotic chaperonin TRiC/CCT. This large complex is essential to the correct and efficient folding of many proteins in our cells. The overall structure is a spherical particle made of sixteen different subunits, whose exact arrangement was hitherto unknown. We have modeled all the 40,320 possible arrangements and compared them to two sets of structural data: (i) cross-linking and mass-spectrometry and (ii) crystallographic dataset at 3.8.. Both sets have single out the same arrangement, which to our surprise was very different than any previous model suggested for TRiC.

World Wide Web, the Internet, biology PPI networks, and social networks, are only a few examples of networks that contain characteristic patterns, termed network motifs, which occur far more often than in randomized networks with the same degree sequence. Our first set of results is related to motif local counting, namely, counting the number of motifs a vertex is part of. We present several efficient algorithms that approximate the local number of occurrences of k-length cycles and k-length cycles with a chord, where k = O(log n). We also provide efficient algorithms that approximate the local number of occurrences of all motifs of size of at most four.Our second set of results relates to general approximate motif counting. We design sublinear algorithm for approximating the number of copies of constant-size stars in a graph. We prove that our algorithm is tight up to polylogarithmic factors. Our work extends the work of Feige and Goldreich and Ron on approximating the number of edges (or average degree) in a graph. In addition, we give some (negative) results on approximating the number of triangles and on approximating the number of length-3-paths in sublinear time.

Personalized treatment approaches which optimize drugs doses according to pre-treatment and early response-to-therapy evaluation hold the promise to improve treatment success rates and reduce severe adverse side-effects due to drugs toxicity in variety of pathologies. Reliable assessment of tissue microenvironment including cell proliferation, density and size and tissue perfusion as a biomarker for disease activity is a key necessity for personalized, response-based treatment regimes. Histology-based tissue microenvironment analysis requires invasive, surgical procedure to obtain the tissue sample. Moreover, the histological analysis is limited to the obtained tissue sample which may not be sufficient in heterogeneous microenvironment. Instead, we developed the TM-MRI method utilized short-duration free-breathing diffusion-weighted MRI acquired with multiple b-values coupled with global tissue microenvironment model and reliable fitting technique that enables radiation and toxicity-free non-invasive insight into the entire three-dimensional tissue microstructure. In the lecture I’ll present the core components of this method: 1) the TM-MRI image acquisition scheme; 2) the global tissue microenvironment model, and; 3) reliable model fitting technique with intrinsic fit-quality assessment. In addition, I'll present initial clinical results of non-invasive disease activity assessment using the TM-MRI quantitative biomarkers in pediatric Crohn’s disease patients and discuss future applications of this technique.

The doubling dimension is a natural measure of the richness of a metric space. It was first considered by Assouad in the context of metric embeddings, and has since found its way into the algorithms and machine learning communities. Ultimately, the doubling dimension provides a way to generalized many results tailored for low-dimensional Euclidean space to apply to more general metric space. The Traveling Salesman Problem (TSP) is among the most famous NP-hard optimization problems. We design for this problem a randomized polynomial-time algorithm that computes a (1+eps)-approximation to the optimal tour, for any fixed eps>0, in TSP instances that form an arbitrary metric space with bounded intrinsic dimension. The celebrated results of Arora (A-98) and Mitchell (M-99) prove that the above result holds in the special case of TSP in a fixed-dimensional Euclidean space. Thus, our algorithm demonstrates that the algorithmic tractability of metric TSP depends on the dimensionality of the space and not on its specific geometry. This result resolves a problem that has been open since the quasi-polynomial time algorithm of Talwar (T-04).

Structured Prediction is a branch of Machine Learning which is concerned with prediction of complex outputs, such as sequences, trees and graphs, with applications in Natural Language Processing, Computer Vision and Computational Biology. Most structured prediction inference problems are intractable, and as a result many algorithms sacrifice model expressivity (i.e. the kinds of informations that can be taken into account when making predictions) in favor of polynomial-time exact inference. I advocate a different framework, in which exact inference is sacrificed in favor of expressive models. Instead of being trained to optimize a global objective function, the models are trained to make a sequence of greedy locally-optimal decisions, while taking easier choices before harder ones, and relaying on earlier predictions do disambiguate later ones. The resulting algorithms are very fast while remaining competitive in terms of prediction accuracy.

This talk is going to show some of the new and cool hacking & security topics. 3 main topics will be presented. It would be focus on the hacking part and would include topics as TV cable hacking, Hardware hacking, SCADA and the 4 top ways to get infected from the internet.

Heuristic state-space search is one of the major success stories of artificial intelligence. Designing successful search heuristics, however, is often considered to be some kind of black magic. In my talk, I will argue that "heuristic" does not imply "unprincipled" and that rigorous scientific discipline can be successfully and fruitfully applied to the study of search heuristics. Specifically, I will explore formal theoretical relationships between the major classes of heuristics that have been suggested in the area of automated planning and will show that the exploration of such relationships can guide us in the design of better, practically effective heuristics.

Efficient concurrent data structures are extremely important for obtaining good performance for most parallel programs. However, ensuring the correctness of concurrent data structure implementations can be very tricky because of concurrency bugs such as race conditions and deadlocks. In systems that use optimistic parallel execution such as boosted transactional memory systems and the Galois system, the implementation of concurrent data structures is even more complex because data structure implementations must also detect conflicts between concurrent activities and support the rollback of conflicting activities. At present, these types of concurrent data structures are implemented manually by expert programmers who write explicitly parallel code packaged into libraries for use by application programmers. This solution has its limitations; for example, it does not permit the customization or tuning of a data structure implementation for a particular application. In this talk, we present Autograph, which is the first concurrent data structure compiler that can synthesize concurrent relational data structure implementations for use by application programmers. The input to Autograph is a high-level declarative specification of an abstract data type (ADT); the output is a concurrent implementation of that ADT with conflict detection and rollback baked in. Our synthesizer is parameterized by a set of data structures called “tiles”, which are building blocks that the compiler composes to create the overall data structure. Application programmers can use a simple expression language to tune the composition of these tiles, thereby exercising high level, fine-grain control of data structure implementations. We have used Autograph to synthesize concurrent sparse graph data structures for a number of complex parallel graph benchmarks. Our results show that the synthesized concurrent data structures usually perform better than the handwritten ones; for some applications and thread counts, they improve performance by a factor of 2.

The ease of seeing conceals many complexities. A fundamental one is the problem of fragmentation – we are able to recognize objects although they are optically incomplete, e.g., due to occlusions. To overcome this difficulty, biological and artificial visual systems use a mechanism for contour completion, which has been studied by the many disciplines of vision science, mostly in an intra-disciplinary fashion. Recent computational, neurophysiological, and psychophysical studies suggest that completed contours emerge from activation patterns of orientation selective cells in the primary visual cortex, or V1. In this work we suggest modeling these patterns as 3D curves in the mathematical continuous space R^2 × S^1, a.k.a. the unit tangent bundle associated with the image plane R^2, that abstracts V1. Then, we propose that the completed shape may follow physical/biological principles which are conveniently abstracted and analyzed in this space. We implement our theories by numerical algorithms to show ample experimental results of visually completed curves in natural and synthetic scenes.

The field of automated, domain-independent planning seeks to build general-purpose algorithms enabling a system to synthesize a course of action that will achieve certain goals. Such algorithms perform reachability analysis in large-scale state models that are implicitly described in a concise manner via some intuitive declarative language. And though planning problems have been studied since the early days of Artificial Intelligence research, recent developments (and, in particular, recent developments in planning as heuristic search) have dramatically advanced the field, and also substantially contributed to some related fields such as software/hardware verification, control, information integration, etc. The difference between various algorithms for planning as heuristic search is mainly in the heuristic functions they define and use. Most typically, an (admissible) heuristic function for domain-independent planning is defined as the (optimal) cost of achieving the goals in an over-approximating abstraction of the planning problem in hand. Such an abstraction is obtained by relaxing certain constraints that are present in the specification of the real problem, and the desire is to obtain a provably poly-time solvable, yet informative abstract problem. The main questions are thus: What constraints should we relax to obtain such an effective over-approximating abstraction? How should we combine information provided by multiple such abstractions? In this talk we consider both these questions, and present some recent formal and empirical results that help answering these questions (sometimes even to optimality). Specifically, Considering Q1, we introduce a generalization of the pattern-database (PDB) homomorphism abstractions to what we called implicit abstractions. The basic idea is in abstracting the problem in hand into provably tractable fragments of optimal planning, alleviating by that the constraint of PDBs to use projections of only low dimensionality. We then introduce concrete instance of this framework called fork-decomposition, and show both formally and empirically that the admissible heuristics induced by the latter abstractions provide state-of-the-art worst-case informativeness guarantees on several standard domains. Considering Q2, we describe a procedure that takes a classical planning task, a forward-search state, and a set of abstraction-based admissible heuristics, and derives an optimal additive composition of these heuristics with respect to the given state. Most importantly, we show that this procedure is polynomial-time for arbitrary sets of all known to us abstraction-based heuristics such as PDBs, constrained PDBs, merge-and-shrink abstractions, fork-decomposition implicit abstractions, and implicit abstractions based on tractable constraint optimization. The talk is based on a joint work with Prof. Carmel Domshlak.

A particularly large class of macromolecular interactions consists of those mediated by a linear sequence motif in the partner molecule: peptide, DNA or RNA. Two approaches to computationally characterize such motif-mediated interactions have been put forward, using either machine learning algorithms or molecular modeling. Using the example of MHC class I proteins, I will discuss each approach's strengths and demonstrate their ability to shed light on intriguing biological phenomena.

A fundamental task in Natural Language Processing (NLP) is learning the syntax of human languages from text. The task is defined both in the sentence level ("syntactic parsing") where a syntactic tree describing the head-argument structure is to be created, and in the word level ("part-of-speech tagging") where every word is assigned a syntactic category such as noun, verb, adjective etc. This syntactic analysis is an important building block in NLP applications such as machine translation and information extraction. While supervised learning algorithms perform very well on these tasks when large collections of manually annotated text (corpora) exist, creating manually annotated corpora is costly and error prone due to the complex nature of annotation. Since most languages and text genres do not have large syntactically annotated corpora, developing algorithms that learn syntax with little human supervision is of crucial importance. The work I will describe is focused on learning better parsing and tagging models from limited amounts of manually annotated training data.Our key observation is that existing models for these tasks are defined at the sentence level, keeping inference tractable at the cost of discarding inter-sentence information. In this work we use Markov random fields to augment sentence-level models for parsing and part-of-speech tagging with inter-sentence constraints. To handle the resulting inference problem, we present a dual decomposition algorithm for efficient, exact decoding of such global objectives. We apply our model to the lightly supervised setting and show significant improvements to strong sentence-level models across six languages. Our technique is general and can be applied to other structured prediction problems in natural language processing and in other fields, to enable inference over large collections of data. Joint work with Alexander Rush, Amir globerson and Michael Collins.

A typical problem in speech and language processing has a very large number of training examples, is sequential, highly structured, and has a unique measure of performance, such as the word error rate in speech recognition, or the BLEU score in machine translation. The simple binary classification problem typically explored in machine learning is no longer adequate for the complex decision problems encountered in speech and language applications. Binary classifiers cannot handle the sequential nature of these problems, and are designed to minimize the zero-one loss, i.e., correct or incorrect, rather than the desired measure of performance. In addition, the current state-of-the-art models in speech and language processing are generative models that capture some temporal dependencies, such as Hidden Markov Models (HMMs). While such models have been immensely important in the development of accurate large-scale speech processing applications, and in speech recognition in particular, theoretical and experimental evidence have led to a wide-spread belief that such models have nearly reached a performance ceiling. In this talk, I first present a new theorem stating that a general learning update rule directly corresponds to the gradient of the desired measure of performance. I present a new algorithm for phoneme-to-speech alignment based on this update rule, which surpasses all previously reported results on a standard benchmark. I show a generalization of the theorem to training non-linear models such as HMMs, and present empirical results on phoneme recognition task which surpass results from HMMs trained with all other training techniques. I will then present the problem of automatic voice onset time (VOT) measurement, one of the most important variables measured in phonetic research and medical speech analysis. I will present a learning algorithm for VOT measurement which outperforms previous work and performs near human inter-judge reliability. I will discuss the algorithm’s implications for tele-monitoring of Parkinson’s disease, and for predicting the effectiveness of chemo-radiotherapy treatment of head and neck cancer.

Machine learning is an effective tool for predicting structured outputs, such as those arising from holistic scene understanding or 3D protein folding. As these problems grow in size, scalable and efficient methods are important for accurate prediction. In this talk I will present how to use duality theory to decompose large-scale prediction problems to many small-scale problems interdependent by messages that are sent along its correlated variables. The use of duality enabled us to handle efficiently web-scale data and achieve several improvements over existing prediction methods for indoor scene understanding, 3D depth estimation and 3D protein folding. In some cases, such as pose estimation, prediction is ambiguous. In this talk I will also present efficient methods for sampling from the set of probable predictions. This approach is based on new approximations and bounds for weighted counting which compute the weight of the heaviest configuration.

A popular approach to deal with the "curse of dimensionality" in relation with high-dimensional data analysis is to assume that points in these datasets lie on a low-dimensional manifold immersed in a high-dimensional ambient space. Kernel methods operate on this assumption and introduce the notion of local affinities between data points via the construction of a suitable kernel. Spectral analysis of this kernel provides a global, preferably low-dimensional, coordinate system that preserves the qualities of the manifold. In this presentation, the scalar relations used in this framework will be extended to matrix relations, which can encompass multidimensional similarities between local neighborhoods of points on the manifold. We utilize the diffusion maps methodology together with linear-projection operators between tangent spaces of the manifold to construct a super-kernel that represents these relations. The properties of the presented super- kernels are explored and their spectral decompositions are utilized to embed the patches of the manifold into a tensor space in which the relations between them are revealed. Two applications of the patch- to-tensor embedding framework for data clustering and classification will be presented.

For every fixed d and every n, we construct an n-point set G in R^d such that every line in R^d is contained in a halfspace that contains only 2n/(d+2) points of G (up to lower-order terms). Apparently, the point set G satisfies the following more general property: For every k, every k-flat in R^d is contained in a halfspace that contains only (k+1) n / (d+k+1) points of G (up to lower-order terms). In 2008,Bukh, Matousek, and Nivasch conjectured the following generalization of Rado's centerpoint theorem: For every n-point set S in R^d and every k, there exists a k-flat f in R^d such that every halfspace that contains f contains at least (k+1) n / (d+k+1) points of S (up to lower-order terms). (The centerpoint theorem is obtained by setting k=0.) Such a flat f would be called a "centerflat". Thus, our upper bound construction shows that the leading constant (k+1)/(k+d+1) in the above conjecture is tight (certainly for k = 1, and apparently for all k). The set G of our construction is the "stretched grid" – a point set which has been previously used by Bukh et al. for other related purposes. Joint work with Boris Bukh.

Many image restoration tasks are ill posed problems, often solved with image priors. Since image priors are only approximate, in general this yields suboptimal restoration results. Given the large body of work on image priors and on image restoration algorithms, this begs fundamental questions of what are optimal restoration procedures, what are (if any) inherent limitations imposed by the statistics of natural images, and what potential gains we may expect from additional years of research efforts.

In this talk we focus on these problems for the simplest restoration task of natural image denoising, where the goal is to estimate a clean natural image, given a noise-corrupted version of it.

We propose a statistical framework and a non-parametric computational approach to study these questions: what is optimal natural image denoising ? what are its fundamental lower bounds, and how far are current algorithms from optimality ?

As we shall see, answers to these questions involve both computational limitations, information-statistical issues and a fundamental property of natural images - scale invariance. Their combination allows us to give a ballpark estimate on the best achievable denoising, and to suggest directions for potential improvements of current algorithms.

Joint work with Anat Levin, Fredo Durand and Bill Freeman.

The Sliding Scale Conjecture was posed by Bellare, Goldwasser, Lund and Russell in 1993 and has been open since. It says that there are PCPs with constant number of queries, polynomial alphabet and polynomially small error. We show that the conjecture can be proved assuming a certain geometric conjecture about curves over finite fields. The geometric conjecture states that there are small families of low degree curves that behave, both in their distribution over points and in the intersections between pairs of curves from the family, similarly to the family of all low degree curves.

Forty years ago, Wiesner pointed out that quantum mechanics raises the striking possibility of money that cannot be counterfeited according to the laws of physics. We propose the first quantum money scheme that is (1) public-key—meaning that anyone can verify a banknote as genuine, not only the bank that printed it, and (2) cryptographically secure, under a "classical" hardness assumption that has nothing to do with quantum money.Our scheme is based on hidden subspaces, encoded as the zero-sets of random multivariate polynomials. A main technical advance is to show that the "black-box" version of our scheme, where the polynomials are replaced by classical oracles, is unconditionally secure.Previously,such a result had only been known relative to a quantum oracle (and even there, the proof was never published). Even in Wiesner's original setting—quantum money that can only be verified by the bank—we are able to use our techniques to patch a major security hole in Wiesner's scheme. We give the first private-key quantum money scheme that allows unlimited verifications and that remains unconditionally secure, even if the counterfeiter can interact adaptively with the bank. Our money scheme is simpler than previous public-key quantum money schemes, including a knot-based scheme of Farhi et al.The verifier needs to perform only two tests, one in the standard basis and one in the Hadamard basis—matching the original intuition for quantum money, based on the existence of complementary observables.Our security proofs use a new variant of Ambainis's quantum adversary method, and several other tools that might be of independent interest.

A compression algorithm for a computation problem is a polynomial-time algorithm that compresses instances of the given problem into equivalent instances. The performance of the compression is naturally measured with respect to its worst-case output size. While NP-hard problems cannot have compression algorithms with non-trivial performance guarantees in terms of the original input size (assuming NP is not in P), some NP-hard problems have surprisingly good compressions when performance is measured in terms of the input solution size. In this talk we discuss some recently developed lower bounds for the compressibility of NP-hard problems when the latter measure is used.

In algorithmic graph theory one strives to exploit graph structure to design efficient algorithms for important, practical problems. In this talk we make the case for this paradigm by discussing recent progress in algorithms for fundamental optimization problems in planar graphs, with applications to basic problems in computer vision known as early vision tasks. We first discuss nearly linear time algorithms for computing shortest paths in directed planar graphs with positive and negative length arcs. We then describe in more detail a nearly linear time algorithm for finding a maximum single commodity flow in a planar network with multiple sources and sinks. These algorithms rely on a host of structural properties of planar graphs, including a beautiful relation between circulations in a planar graph and shortest paths in the planar dual of that graph. We conclude with a brief discussion of distance oracles for planar graphs. No prior knowledge of planar graphs is assumed.

Multi-level marketing is a marketing approach that motivates its participants to promote a certain product among their friends. The popularity of this approach increases due to the accessibility of modern social networks, however, it existed in one form or the other long before the Internet age began (the infamous Pyramid scheme that dates back at least a century is in fact a special case of multi-level marketing). In this talk we lay foundations for the study of reward mechanisms in multi-level marketing within social networks. We provide a set of desired properties for such mechanisms and show that they are uniquely satisfied by geometric reward mechanisms. The resilience of mechanisms to false-name manipulations is also considered; while geometric reward mechanisms fail against such manipulations, we exhibit other mechanisms which are false-name-proof. The talk will be self-contained.

The age of the internet and the pervasiveness of networked computing enabled the creation of large computational systems in which multiple autonomous entities interact. The designers of such systems face difficult challenges: they must bring about the desired behavior of the system as a whole while accounting for the disjoint behavior and incentives of individual agents. I will present a few examples of research (with various collaborators) that deals with these challenges in the context of different systems. I will discuss recent work on incentives for information dissemination in the Bitcoin protocol (a distributed electronic currency that has received much attention recently) and in social networks, as well as work on the interactions within closed P2P communities, and on core network protocols such as TCP and BGP.

A typical learning process begins with some prior beliefs about the target concept. These beliefs are refined using evidence, typically a sample. The evidence is used to compute a “posterior belief”, which assigns a probability distribution over the hypotheses class. The Bayes optimal hypothesis is the average hypothesis, weighted by the posterior. However, this hypothesis is often prohibitively costly to compute. Therefore, there is a need for an algorithm to construct a hypothesis (either a single hypothesis or some combination), given the posterior belief. Several methods have been proposed for this problem: for example, Gibbs sampling, ensemble methods, and choosing the maximum posterior. We propose a new method: choosing the median hypothesis. This method is close to the average Gibbs classifier and Bayes optimal classifier in terms of accuracy while having the same run-time efficiency, during the generalization phase, as the maximum posterior method. In this talk, we will define a measure of depth for hypotheses, from which we derive the median hypothesis. We present generalization bounds which leverage the PAC-Bayes analysis technique. We present an algorithm to approximate the median hypotheses and we prove its correctness. Our definition of median is closely related to Tukey's median; in fact our algorithm provides a polynomial approximation to the problem of finding the Tukey median. This is a joint work with Chris J. C. Burges

We will introduce a “learning” multi-scale iterative process for data analysis. This process approximates a task related function that is defined on a given data-set by using the geometric structures of the data in different scales. The constructed multi-scale representation can be easily extended to new data points. We will provide a number of examples including classification and regression, extension of non-linear embedding coordinates, and forecasting time series.

Linear programming (LP) and semidefinite programming (SDP) are powerful convex optimization tools which have become ubiquitous in the field of approximation algorithms in the last few decades. Given a combinatorial optimization problem (e.g. Maximum Independent Set), a standard approach to obtain an approximation algorithm is as follows: formulate the problem as an integer program (IP), relax this formulation to an LP or SDP, solve the relaxation, and then "round" the solution back to an integer solution. This approach is limited by how well the convex program (LP or SDP) approximates the original IP formulation, i.e. the integrality gap. One way to circumvent this limitation is through hierarchies of convex programs, which give a systematic way of iteratively strengthening any relaxation (at the cost of increased running time to solve it), so that the integrality gap gradually decreases. While initially, most of the literature on hierarchies in the context of approximation algorithms had focused on impossibility results,there has been a surprising surge of recent positive results. I will survey this recent development, by describing a number of combinatorial optimization problems for which we have been able to achieve improved approximations using hierarchies.

Two key synchronization paradigms for the construction of scalable concurrent data-structures are software combining and elimination.Elimination-based concurrent data-structures allow operations with reverse semantics (such as push and pop stack operations) to "collide" and exchange values without having to access a central location. Software combining, on the other hand, is effective when colliding operations have identical semantics: when a pair of threads performing operations with identical semantics collide, the task of performing the combined set of operations is delegated to one of the threads and the other thread waits for its operation(s) to be performed. Applying this mechanism iteratively can reduce memory contention and increase throughput. We present DECS, a novel Dynamic Elimination-Combining Stack algorithm,that scales well for all workload types. While maintaining the simplicity and low-overhead of an elimination-based stack, DECS manages to benefit from collisions of both identical- and reverse-semantics operations. Our empirical evaluation shows that DECS scales significantly better than both blocking and non-blocking best prior stack algorithms. This is joint work with Gal Bar-Nissan and Danny Hendler.

In the past decade the importance of natural language text processing (NLP) has grown immensely. The first steps in NLP applications involve identification of topics, entities, concepts, and relations in text. Traditionally, statistical models have been successfully deployed for the aforementioned problems. However, the major trend so far has been:"scaling up by dumbing down"- that is, applying sophisticated statistical algorithms operating on very simple or low-level features of the text. In this talk I will be making an argument that it is essential to use knowledge in NLP, propose several ways of doing it, and provide case studies on several fundamental NLP problems. The first problem I will address is entifying "important" expressions in input text and cross-linking them to Wikipedia pages describing these expressions.This approach allows to enrich the input text with knowledge from Wikipedia. Then, I'll describe an approach for utilizing knowledge imported from Wikipedia for co-reference resolution, the task of understanding that in the expression "Obama addressed the nation, I need your help, he said", ""he" refers to Obama and "your" refers to the American nation.

Database transformations (queries, views, mappings) take apart, filter,and recombine source data in order to populate warehouses, materialize views,and provide inputs to analysis tools. As they do so, applications often need to track the relationship between parts and pieces of the sources and parts and pieces of the transformations' output. This relationship is what we call database provenance. This talk presents an approach to database provenance that relies on two observations. First, provenance is a kind of annotation, and we can develop a general approach to annotation propagation that also covers other applications, for example to uncertainty and access control.In fact, provenance turns out to be the most general kind of such annotation,in a precise and practically useful sense. Second, the propagation of annotation through a broad class of transformations relies on just two operations: one when annotations are jointly used and one when they are used alternatively.This leads to annotations forming a specific algebraic structure, a commutative semiring.The semiring approach works for annotating tuples, field values and attributes in standard relations, in nested relations (complex values), and for annotating nodes in (unordered) XML. It works for transformations expressed in the positive fragmentof relational algebra, nested relational calculus, unordered XQuery, as well as for Datalog, GLAV schema mappings, and tgd constraints. Finally, when properly extended to semimodules it works for queries with aggregates. Specific semirings correspond to earlier approaches to provenance, while others correspond to forms of uncertainty, trust, cost, and access control. This is joint work with Y. Amsterdamer, D. Deutch, J.N. Foster, T.J. Green,Z. Ives, and G. Karvounarakis, done in part within the frameworks of the Orchestra and pPOD projects.

In recent years the computing industry has been showing a growing interest in distributed computing frameworks as the only way to achieve the massive scale, high reliability and continuous availability that are required by enterprise data centers and cloud computing infrastructure. This talk will focus on research we did on the Virtual Synchrony framework of Birman and Jones. We created an axiomatic model for their framework that enables a great simplification of the description of failure behaviors in the cluster and very precise proofs of the correctness of their protocols.

Applying Software Engineering and Formal Verification Methods to Biological Systems Studies of biological systems are often facilitated by diagram models that summarize the current understanding of underlying mechanisms. The increasing complexity of our understanding of biology necessitates computational models that can extend these representations to include their dynamic behavior. I will present progress on foundations and tools, towards enabling biologists and modelers to construct high-level theories and models of biological systems using formal visual languages, capturing biological hypotheses, inferred mechanisms, and experimental results. I will describe some of the applications (assuming no background in biology) and outline challenges and future research directions.

In this talk I will present state of the art algorithms for solving combinatorial optimization tasks defined over graphical models (Bayesian networks, Markov networks, and constraint networks) and demonstrate their performance on a variety of benchmarks. Specifically I will present branch and bound and best-first search algorithms which explore the AND/OR search space over graphical models and will demonstrate the gain obtained by exploiting problem’s decomposition (using AND nodes), equivalence (by caching) and irrelevance (via the power of new lower bound heuristics such as mini-buckets). The impact of additional principles such as exploiting determinism via constraint propagation, the use of good initial upper bounds generated via stochastic local search and the variable orderings ideas may be discussed, as time permits. Current research is in extending the algorithms into distributed/parallel solving, anytime solutions, m-best solutions, and dvanced heuristic generations. Joint work with Radu Marinescu.

The current state of the art in the development of search methodologies is focused around the design of bespoke systems, which are specifically tailored to a particular situation or organisation.Such bespoke systems are necessarily created by human experts, and so they are relatively expensive. Some of the research at the ASAP research group, at the University of Nottingham, is concerned with how to build intelligent systems which are capable of automatically building new systems. In other words to automate some of the creative process, to make it less expensive by being less reliant on human expertise. In this talk, I will present some work we have recently published on the automatic design of heuristics, particularly for two dimensional stock cutting problems. The research shows that genetic programming can be used to evolve novel heuristics which are at least as good as human designed heuristics for this problem. Research into the automatic design of heuristics could represent a change in the role of the human expert, from designing a heuristic methodology, to designing a search space within which a good heuristic methodology is likely to exist. The computer then takes on the more tedious task of searching that space, while we can focus on the creative aspect of designing it.

The problem of multi-robot patrol has become a canonical problem in multi-robot, in which a team of mobile robots is required to jointly visit some target area in order to monitor change in state. The goal of the robots can vary from optimizing point-visit frequency, to maximizing the chances of detecting an adversary that tries to pass through the patrol path undetected. In this talk I will describe theoretical results that are used as a baseline for my work, in which strategies for the patrolling robots can be found efficiently based on, among others, a Markovian modeling of the world. I will then describe various adaptations of the theoretical results to handle real world constraints, among them description of new patrolling strategies, reevaluation of coordination restrictions, and development of new adversarial models.

Approximating an arbitrary graph by a simpler structure while preserving some properties of the original graph, is a very active research direction that has found numerous applications.In this talk we will discuss the problem of finding a spanning tree that preserves the distances of the original graph on average, up to a universal constant. Based on joint work with Ittai Abraham and Yair Bartal.

Detection and containment of unknown malware are challenging tasks. Typically the detection is performed by experts who use anomaly detection systems or Honeypots-based systems. Such a detection process is very slow and it is not suited for detection of rapidly propagating threats such as worms. In this research we propose to automate the detection process by introducing an innovative distributed framework for detection and containment of new malware. The framework consists of distributed agents that are installed in several client computers and a Centralized Decision Maker module (CDM) that interacts with the agents. The new detection process is performed in two phases. In the first phase agents detect potential malware on local machines and send their detection results to the CDM. In the second phase, the CDM builds a propagation graph for every potential malware. These propagation graphs are compared to known malware propagation characteristics in order to determine whether the potential malware is indeed a malware. All the agents are notified of a final decision in order to start the containment process. Another contribution of this study is a method for detecting new malicious executables locally. It is based on monitoring run-time system calls and comprises the following steps: (a) in an offline training phase, finding a set of (not necessary consecutive) system call sequences that are characteristic only to malicious files, when such malicious files are executed, and storing said sequences in a database; (b) in a real time detection phase, for each running executable, continuously monitoring its issued system calls and comparing them with the stored sequences of system calls within the database to determine whether there exists a match between a portion of the sequence of the run-time system calls and one or more of the database sequences, and when such a match is found, declaring said executable as malicious. In addition to the collaborative detection, the method can be used for independent (local) malware detection, replacing (or in addition to) traditional antivirus software.

Protection of personal information is one of the most challenging problems in emerging information society. The traditional approach, based on purely organizational protection seems to be insufficient – it became clear that personal data protection should be backed by technical mechanisms working in automatic way, independently of human behavior. One of the recent ideas in this area is restricted identification introduced by German authorities. New electronic personal identity documents in Germany are equipped with a cryptographic protocol that supports anonymous identification. The idea is based on the notion of independent sectors of activity. The cryptographic protocol implemented there provides unlinkable passwords created with strong asymmetric cryptography from a single secret key of the user. This approach differs from anonymous credentials in the sense that a single person may have only one pseudonym in a given sector for the lifetime of a given personal identity document. In particular, no Sybil attack is possible. We present related solutions for different anonymity scenarios, such as access to personal medical information or contacts with law enforcement authorities. Despite some differences these solutions build together a common framework for unlinkable authentication in different sectors. We also provide reduction proofs supporting unlinkability claims. joint work with: P.Kubiak, L.Krzywiecki, Jun Shao, M.Koza concern results presented at IEEE CCNC 2011 and to be presented at INTRUST 2011

We propose a 2-party UC-secure computation protocol that can compute any function securely. The protocol requires only two messages, communication that is poly-logarithmic in the size of the circuit description of the function, and the workload for one of the parties is also only poly-logarithmic in the size of the circuit. This implies, for instance, delegatable computation that requires no expensive off-line phase and remains secure even if the server learns whether the client accepts its results. To achieve this, we define a notion of extractable hash functions, propose an instantiation based on the knowledge of exponent in an RSA group, and build succinct zero-knowledge arguments in the CRS model.

Let a,b,c,… in S_n be random permutations on n elements, chosen at uniform distribution. What is the distribution of the permutation obtained by a fixed word in the letters a,b,c,…, such as ab,a^2, a^2bc^2b, or aba^(-2)b^(-1)? More concretely, do these new random permutations have uniform distribution? In general, a free word w in F_k is called uniform if for every finite group G, the word map $w: G^k to G$ induces uniform distribution on G (given uniform distribution on G^k). So which words are uniform? This question is strongly connected to the notion of primitive words in the free group $F_k$. The word w is called primitive is it belongs to some basis, i.e. a generating set of size $k$. It is an easy observation that a primitive word is uniform. It was conjectured that the converse is also true. We prove it for F_2. In a very recent joint work with O. Parzanchevski, we manage to prove the conjecture in full. Akey ingredient of the proofs is a new algorithm to detect primitive elements.

A maximal independent set (MIS) on a graph is an inclusion-maximal set of mutually non-adjacent nodes. This basic symmetry breaking structure is vital for many distributed algorithms. There is a simple and elegant randomized algorithm that computes an MIS in O(log n) rounds. However, despite much effort, there has been no improvement in the general case for decades, leaving a large gap to the known lower bound of Omega(sqrt(log n)) rounds. We present a randomized algorithm that achieves a running time of O(sqrt(log n log log n)) on forests. In contrast to previous techniques achieving sublogarithmic running times, our approach does not rely on any bound on the number of independent neighbors (possibly with regard to an orientation of the edges). We therefore hope that it might ultimately contribute to improved algorithms for the general case.

absract.pdf

In the index coding problem, introduced by Birk and Kol (INFOCOM, 1998), the goal is to transmit n bits to n receivers (one bit to each), where the receivers reside at the nodes of a graph G and have prior access to the bits corresponding to their neighbors in the graph (side information). The objective is to find a code word of minimum length which will allow each receiver to learn their own bit given access to the code word and their side information. When the encoding is linear (this is known as linear index coding), the minimum possible code word length corresponds to a graph parameter known as the minrank of G. In this talk, we will describe an algorithm which approximates the minrank of a graph in the following sense: when the minrank of the graph is a constant k, the algorithm finds a linear index code of length O(n^(f(k))). For example, for k=3 we have f(3) ~ 0.2574. This algorithm exploits a connection between minrank and a semidefinite programming (SDP) relaxation for graph coloring introduced by Karger, Motwani and Sudan. A result which arises from our analysis, and which may be of independent interest, gives an exact expression for the maximum possible value of the Lovasz theta-function of a graph, as a function of its minrank. This compares two classical upper bounds on the Shannon capacity of a graph. Based on joint work with Ishay Haviv.

In recent years, we have used software engineering tools to develop reactive models to simulate and analyze the development of organs. The modeled systems embody highly complex and dynamic processes, by which a set of precursor stem cells proliferate, differentiate and move, to form a functioning tissue. Three organs from diverse evolutionary organisms have been thus modeled: the mouse pancreas, the C. elegans gonad, and partial rodent brain development. Analysis and execution of the models provided dynamic representation of the development, anticipated known experimental results and proposed novel testable predictions. In my talk, I will l discuss challenges, goals and achievement in this direction in science.

Throughout the past decade there has been extensive research on algorithmic and data mining techniques for solving the problem of influence maximization in social networks: if one can convince a subset of individuals to influence their friends to adopt a new product or technology, which subset should be selected so that the spread of influence in the social network is maximized? Despite the progress in modeling and techniques, the incomplete information aspect of problem has been largely overlooked. While the network structure is often available, the inherent cost individuals have for influencing friends is difficult to extract. In this talk we will discuss mechanisms that extract individuals' costs in well studied models of social network influence. We follow the mechanism design framework which advocates for truthful mechanisms that use allocation and payment schemes that incentivize individuals to report their true information. Beyond their provable theoretical guarantees, the mechanisms work well in practice. To show this we will use results from experiments performed on the mechanical turk platform and social network data that provide experimental evidence of the mechanisms' effectiveness.

The ability to share and reconcile similar data on remote hosts with minimum communication is fundamental and inherent to a wide variety of networking applications, ranging from maintenance of your contacts across smartphones to synchronizing video fragments from moon rovers. In this talk we provide a selective survey of ten years of our research on this problem, together with applications to the information theory and cryptography communities.

We will begin with a concrete formalization of the problem of reconciling sets, and our initial solution based on polynomial interpolation. We will then present and analyze information-theoretic bounds, estimation techniques, and interactive solutions to this problem, together with an extension to string reconciliation. Throughout, we will describe applications tied to this problem and developed by others, such as the current standard for synchronizing PGP key databases and secure sketches for biometric authentication.

A coreset (or, core-set) for a given problem is a "compressed" representation of its input, in the sense that a solution for the problem with the (small) coreset as input would yield an approximate solution to the problem with the original (large) input. Using traditional techniques, a coreset usually implies provable linear time algorithms for the corresponding optimization problem, which can be computed in parallel, via one pass over the data, and using only polylogarithmic space (i.e, in the streaming model). During the recent years, coresets were suggested for problems such as k-means clustering, classification, facility location, linear regression, PCA, and matrix approximation. I will give an introduction to this new paradigm, including recent implementation in the context of computer vision. Based on a paper with Michael Langberg (STOC'11), and a paper Nir Sochen and Micha Feigin (SSVM'11)

This talk describes a body of work aiming for a paradigm shift in automatic multiview reconstruction and calibration. The new paradigm correlates the differential geometry of image curves and surfaces in contrast to using isolated image features such as SIFT/HOG. The current multiview geometry techniques have been developed based on the ideas from Projective Geometry and generally use isolated feature points in conjunction with RANSAC and Bundler, and result in unorganized cloud of point reconstructions. Our approach is complementary and uses the differential geometry attributes of curves and surfaces as the basic signature to correlate across views for both calibration and reconstruction. We discuss two specific efforts. First, we show that curve fragments can be correlated in multiple views and reconstructed to form a 3D Curve Sketch, an initial scaffold onto which surface patches will be reconstructed. Second, we show that the differential geometry attribute of curves can be used in registration of 3D models to 2D images. The results of other projects underway are also highlighted.

We explore dynamic environments in which computational nodes, or decision makers, follow simple and unsophisticated rules of behavior (e.g., repeatedly "best replying" to others' actions, and minimizing "regret") that have been extensively studied in game theory and economics. Our aim is to understand when convergence of the resulting dynamics to an equilibrium point is guaranteed even if nodes' interaction is not synchronized (e.g., as in Internet protocols and large-scale markets). We exhibit general positive and negative results and consider their implications across a wide variety of interesting and timely applications: routing, congestion control, game theory, social networks and circuit design. We also investigate incentives in our framework. Joint work with Aaron D. Jaggard and Rebecca N. Wright (2011) Noam Nisan, Gregory Valiant and Aviv Zohar (2011) Alex Fabrikant (in progress).

TBA

I discuss the syntactic-parsig problem: the automatic assignment of syntactic structure to sentences in natural language text, with a focus on parsing Hebrew, a language with rich morphology. Syntactic-parsing is an instance of a structured-prediction, a subfield of machine-learning concerned with learning to map complex inputs (like sentences) to complex outputs (like trees). I present the common framework for structured prediction, based on problem-decomposition and dynamic-programming, and show some cases in natural language where independence assumptions of that framework fail. I then present Easy-First parsing, a novel algorithm for syntactic parsing which is based on a different set of assumptions, and which makes it easy to incorporate a much richer feature sets than the dynamic-programming approach. The algorithm works by modeling the set of decisions made by constructing the output structure instead of directly modeling the structure itself. As a consequence, at each step the algorithm can observe partial structures and exploit rich contextual information to guide further decisions. The algorithm works in a greedy fashion by delegating the search to the training phase, making it particularly fast at inference time. I present results of experiments on Hebrew data sets indicating a parsing accuracy of up to 81.5%, a substantial improvement over other approaches. The method also works well for English, leading to state-of-the-art parsing results.

Structured prediction techniques have applications in other domains, I briefly report on application to computational biology (RNA structured prediction) where we also reach state-of-the-art results (joint work with Shay Zakov and Michal Ziv-Ukelson).

Symmetry breaking is an important task for distributed systems. Well-known problems involving symmetry breaking are the coloring and maximal independent set problems. They lie at the heart of many scheduling problems such as computing a TDMA schedule in wireless networks or for scheduling transactions in a multi-core system. In this talk, I consider both problems and look at different graph classes. A few randomized and deterministic algorithms will be presented in the message passing model.

Galperin and Wigderson proposed a succinct representation for graphs, that uses number of bits that is logarithmic in the number of nodes. They proved complexity results for various decision problems on graph properties, when the graph is given in a succinct representation. Later, Papadimitriou and Yannakakis showed, that under the same succinct encoding method, certain class of decision problems on graph properties becomes exponentially hard. We consider the complexity of the Permanent problem when the graph/matrix is given in a restrict succinct representation. We present an optical architecture that is based on the holographic concept for solving balanced inputs of the Succinct Permanent problem.

The talk will include both a theoretical results (relating establishing a computational complexity of some problems) and an optical implementation of the proposed algorithm.

Demanding historical data analysis like forecasting, formerly dependent on specialized data warehouses and temporal databases, can become available to everyday applications in off-the-shelf data stores. The challenge is to organize historical states so that they are ``always there'' when needed and "not in the way" when not. Retro approach integrates a low-level long-lived consistent snapshot system into a data store, allowing to run historical data analysis programs against the snapshots, side by side with programs running against the current state. The approach is attractive for several reasons. Applications can take snapshots at any frequency without disrupting the data store, and can garbage collect unneeded snapshots at low cost, an important feature in a long-lived system. Analysis programs can access snapshots at a cost that is independent of update workload and history length even for very old snapshots. A principled methodology allows to construct the snapshot system protocols in a modular way from the data store protocols, allowing to implement the snapshot system without extensive modifications to the data store internals, making the approach suitable in off-the-shelf data stores. The talk will describe the new techniques that underly Retro and present preliminary performance results from a prototype we built in Berkeley DB, indicating Retro is efficient, imposing moderate performance penalty on the native data store, on expected common workloads.

Bio: Liuba Shrira is a Professor in the Computer Science Department at Brandeis University, and is affiliated with the Computer Science and Artificial Intelligence Laboratory at MIT. She has been affiliated with Microsoft Research, Cambridge, UK, Microsoft Research Asia, Beijing, and is currently visiting the Computer Science Department, in the Technion, Haifa. Her research interests span aspects of design and implementation of distributed systems and especially storage systems. This includes fault-tolerance, availability and performance issues. Her recent focus is on long-lived transactional storage, time travel (in storage), software upgrades, and support for collaborative access to long-lived objects.

First-Order Probabilistic Models extend the Propositional Graphical Models (e.g. Markov Network) by introducing the concept of domain entities, along with a first-order language which depicts the properties of each entity and the various interactions which they exhibit. One way of performing inference in first-order models is to construct a propositional model of the same joint distribution, and apply one of the known inference methods. However, it is desirable to apply inference directly to the first-order model, thus avoiding an explicit extraction of the propositional model, which can be very large. Moreover, an inference task in a first-order model may require exponentially less amount of time compared with its propositional counterpart.

Bellare and Rogaway [ACM CCS ’93] introduced the famous random oracle model as a “paradigm for designing efficient protocols”. This paradigm has led to several highly efficient and widely used in practice constructions, such as the RSA Ful l Domain Hash signature scheme (RSA-FDH). Unfortunately, little is known about the security of the resulting schemes in the standard model, when the random oracle is replaced by a concrete function. In particular, it is unknown whether we can reduce their (standard model) security to any “natural” assumption. Prior work has shown several “uninstantiability” results for various abstractions of RSA- based schemes, where the RSA function was replaced by a random permutation. These ab- stractions, however, do not allow either the reduction or the hash function instantiation to use any algebraic properties of RSA function, such as the multiplicative group structure of Z*n . In this work we develop new techniques which rule out such algebraic instantiations, focusing specifically on the case of RSA-FDH. We show that it is impossible to reduce the security of RSA-FDH to any natural assumption, provided that the construction and the reduction treat the multiplicative RSA group Z*n in a black-box way. To the best of our knowledge, this restriction is satisfied by all positive results for RSA-based signatures, including standard model constructions of Gennaro et al. [EUROCRYPT ’99], Cramer and Shoup [ACM TISS ’00] and Hohenberger and Waters [CRYPTO ’09]. As our main technical (and conceptual) contribution, we show how to adapt the powerful “short description” paradigm of Gennaro and Trevisan [FOCS ’00] to the “generic group” setting. This paradigm is typically used to rule out one-way permutations based reductions, showing that their security proofs would imply a (provably impossible) way to “compress” a random permutation. In our setting, the reduction has access to a random group G isomorphic to Z*n , and can use the algebraic properties of G. Still, we show that such a reduction must “know” the factorization of n (and, hence, does not “benefit” from the signature forger), since otherwise it can be used to “compress” our group G. We demonstrate the optimality of our negative result, at least in some sense, by showing that the RSA-FDH signatures can be proven secure in the standard model, under the standard RSA assumption, provided the number of signing queries is a-priori bounded.

Significant progress has been made in the past two decades in the construction of software systems that relate directly to mathematics: computer algebra systems, theorem provers,proof assistants, mathematical knowledge bases and so forth. A variety of powerful algorithms have been implemented in general-purpose symbolic algebra systems, special purpose systems, and state of the art numerical libraries. I will discuss how computational tools and computationalthinking influence science and engineering and demonstrate several accessible examples carried out in the spiritof experimental mathematics.

A k-spanner of a given graph is a subgraph that preserves all distances within factor k. This notion is useful in several contexts, from distributed computing to property testing. From an algorithmic viewpoint, the goal is to find a k-spanner with as few edges as possible. In this talk I will examine two variants of graph spanners: directed spanners and fault-tolerant spanners. In the directed setting we design an O(n^{2/3})-approximation algorithm for the directed k-spanner problem that works for all k. This is the first approximation independent of k, as well as being the first to handle arbitrary edge lengths. In the fault-tolerant setting we give a new construction of fault-tolerant spanners of undirected graphs that is both simpler than previous work and improves the dependence on the number of faults from exponential to polynomial. For the special case of k=2, we also provide an O(log n)-approximation algorithm for the smallest fault-tolerant 2-spanner. This approximation ratio is, notably, independent of the number of faults. Our main tool in both the directed and fault-tolerant settings is a new flow-based linear programming relaxation.

Joint work with Robert Krauthgamer.

The Internet has enabled the creation of virtual worlds and communities, where user interactions imitate and, to some extent, even replace the more traditional "real-life" equivalents on a larger scale. The existence of easily accessible virtual communities makes it both possible and legitimate to communicate with total strangers. We can now anonymously interact with other virtual community members whom we do not really know in ways that break the boundaries and limitations of the real world. Trust and reputation systems are considered key enablers of virtual communities, especially communities of strangers, where users are not required to reveal their real identities and use nicknames or pseudonyms instead. These systems support the accumulation of member reputation information and leverage this information to increase the likelihood of successful member interactions and to better protect the community from fraudulent members. Reputation information is a valuable resource both for the users and for the communities. We developed the Cross-Community Reputation (CCR) model for the sharing of reputation knowledge across virtual communities. The CCR model defines the major stages required to compute the cross-community reputation of a community member based on the reputation of that member in other communities. We addressed major privacy concerns that are not present or that are less significant in single community domains. In this talk we provide a brief introduction to trust and reputation systems. We outline our major research directions, and focus on the CCR model and its related privacy concerns.

Search algorithms are usually designed to return either optimal, suboptimal or any solution. In the talk I will address a different but very realistic search task: find solution with cost smaller than or equal to a given fixed constant. This task is relevant in scenarios where a fixed budget is available and we would like to find solution under the given budget with minimum search effort. For this problem, we introduced an algorithm called Potential Search (PTS) which is a best-first search that expands nodes according to the probability that they will lead to a solution whose cost is less than the given budget. Surprisingly, it is often possible to implement PTS even without explicitly calculating these probabilities, by learning the error model of the available heuristic. In addition, PTS can be modified to an anytime search algorithm. Experimental results show that PTS outperforms other relevant algorithms in most cases, and is more robust.

In the second part of the talk, I will present a novel algorithm that can predict the cost optimal solution of a given search problem. The proposed prediction algorithm builds on the CDP formula (Zahavi et. al., JAIR 2010) used to predict the number of nodes expanded in a single iteration of IDA*. Experiments on benchmark search domains show that the prediction is extremely accurate. In addition, the predicted cost can be used to enhance existing search algorithm. Experiments show that combining the prediction algorithm with Potential Search is able to find almost optimal solutions with orders of magnitude less nodes expanded when compared to A*.

Originating in statistical physics, the Abelian Sandpile Model is a diffusion process on finite graphs with a remarkably rich theory that connects the fields of algebraic graph theory, discrete dynamical systems, stochastic processes, commutative semigroups and groups, number theory, algorithms and complexity theory, and more. After a general introduction highlighting classical result by Deepak Dhar and others, I will outline recent work, in part with my former students Evelin Toumpakari and Igor Gorodezky, on the transition from "transient" to "recurrent"states in this model. I will conclude with open algorithmic problems.

In the square jigsaw puzzle problem one is required to reconstruct the complete image from a set of non-overlapping, unordered, square puzzle parts. Here we propose a fully automatic solver for this problem, where unlike some previous work, it assumes no clues regarding parts location and requires no prior knowledge about the original image or its simplified (e.g., lower resolution) versions. To do so, we introduce a greedy solver which combines both informed piece placement and rearrangement of puzzle segments to find the final solution. Among our other contributions are new compatibility metrics which better predict the chances of two given parts to be neighbors, and a novel estimation measure which evaluates the quality of puzzle solutions without the need for ground-truth information. Incorporating these contributions, our approach facilitates solutions that surpass state-of-the-art solvers on puzzles of size larger than ever attempted before.

This work is on the seam between differential privacy and mechanism design. We study the implementation challenge in an abstract interdependent values model and an arbitrary objective function. We design a mechanism that allows for approximate optimal implementation of insensitive objective functions in ex-post Nash equilibrium. If, furthermore, values are private then the same mechanism is strategy proof. We cast our results onto two specific models: pricing and facility location. The mechanism we design is optimal up to an additive factor of the order of magnitude of one over the square root of the number of agents and involves no utility transfers.

Underlying our mechanism is a lottery between two auxiliary mechanisms: with high probability we actuate a mechanism that reduces players' influence on the choice of the social alternative, while choosing the optimal outcome with high probability. This is where the recent notion of differential privacy is employed. With the complementary probability we actuate a mechanism that is typically far from optimal but is incentive compatible. The joint mechanism inherits the desired properties from both.

Joint work with Rann Smorodinsky and Moshe Tennenholtz.

Negotiation is a process by which interested parties confer with the aim of reaching agreements. The dissemination of technologies such as the Internet has created opportunities for computer agents to negotiate with people, despite being distributed geographically and in time. The inclusion of people presents novel problems for the design of autonomous agent negotiation strategies. People do not adhere to the optimal, monolithic strategies that can be derived analytically, as is the case in settings comprising computer agents alone. Their negotiation behavior is affected by a multitude of social and psychological factors. Furthermore, culture plays an important role in their decision making and people of varying cultures differ in the way they make offers and fulfill their commitments in negotiation. In this talk I will present the following two agents that negotiate well with people by modeling several social factors: The PURB agent that can adapt successfully to people from different cultures in complete information settings, and the SIGAL agent that learns to negotiate successfully with people in games where people can choose to reveal private information. These agents were evaluated in extensive experiments including people from three countries.

(Joint work with Y. Gal, N. Peled and G. Haim)

Trees, bushes, and other plants are ubiquitous in urban environments, and realistic models of trees can add a great deal of realism to a digital urban scene. In recent years mobile laser scanning become an efficient technology for fast acquisition of large urban scenes. Such systems generate clouds of points that are often sparse, incomplete, and noisy representation of the real scenes. In this talk I present an approach that robustly reconstructs the existing trees in a scene. This is done by a series of optimizations that construct skeletal structure and foliage for a tree based on its points.

A significant benefit of our approach is its ability to reconstruct multiple overlapping trees simultaneously without segmentation. We demonstrate the effectiveness and robustness of our approach on many raw scans of different tree varieties.

This talk is composed of two sub titles (projects): 1) Automatic Reconstruction of Tree Skeletal Structures from Point Clouds 2) Texture-Lobes for Tree Modelling

RNA molecules are fundamental participants in many biological processes, where recent discoveries revile that the role of RNA is much more important than was previously known. A key property for analyzing the functionality of RNA is its structure. Due to their size, it is difficult to determine structures of RNA molecules physically, and thus computational RNA structure prediction (a.k.a "RNA folding") is an important bioinformatic application.

In this talk, we present a new approach which significantly improves the accuracy of RNA folding. The method combines rich feature representations of RNA molecules with a light-weight machine learning algorithm (the "online passive aggressive" algorithm), and obtains about 50% reduction of error rate with respect to the best previous result.

The security of various cryptosystems in common use has been completely compromised by "side channel attacks", namely, by attacks exploiting leakage of information on the underlying secret keys. Such information leakage typically emanates from physical characteristics inevitably involved in real-world implementations of cryptographic protocols (say, power consumption, timing, or electro-magnetic radiation).

In this talk I will discuss *leakage resilient cryptography* – cryptographic protocols protecting against such side channel attacks. I will focus on the _bounded memory leakage model_ (with Goldwasser and Vaikuntanathan), a model capturing a large class of side channel attacks that laid the foundations for many follow-up works on leakage resilient cryptography, and will exhibit public key encryption schemes resilient to such leakage. As time permits I will also mention extensions of these results to _unbounded memory leakage in distributed settings_ (with Goldwasser and Hazay).

The BlueGene supercomputers in production today run a small limited-function kernel called CNK (Compute-Node Kernel), which is designed to expose bare-metal performance to the application. The problem is that this comes at the expense of limiting BlueGene to only executing HPC applications, while potentially the platform can execute any type of workload. A possible solution is to replace CNK with standard Linux, but in order to get this accepted, Linux must first demonstrate performance that is comparable to CNK.

In a research conducted at the IBM T.J. Watson research center, for the first time we booted Linux on the compute-nodes of BlueGene/L and compared its performance to CNK. We identified two major obstacles to performance under Linux: the well known daemon noise problem that affects application scalability, and the high cost of TLB (Translation Lookaside Buffer) misses that affects performance at the node-level. We then leveraged unique hardware features in the system using proper software support in the Linux kernel, and demonstrated comparable performance to CNK over a wide set of HPC benchmarks.

My talk has two parts: I will begin with an introduction to supercomputing, describe the BlueGene software and hardware architecture, the structure of HPC applications, and related performance issues. I will then focus on the Linux research, present the experiments we performed, the issues we encountered, and the solutions we developed to address them.

Paper (1): "Feature based Locomotion Controllers" - abstract: This paper introduces an approach to control of physics-based characters based on high-level features of movement, such as center-of-mass, angular momentum, and end-effectors. Objective terms are used to control each feature, and are combined by a prioritization algorithm. We show how locomotion can be expressed in terms of a small number of features that control balance and end-effectors. This approach is used to build controllers for human balancing, standing jump, and walking. These controllers provide numerous benefits: human-like qualities such as arm-swing, heel-off, and hip-shoulder counter-rotation emerge automatically during walking; controllers are robust to changes in body parameters; control parameters and goals may be modified at run-time; control parameters apply to intuitive properties such as center-of-mass height; and controllers may be mapped onto entirely new bipeds with different topology and mass distribution, without modifications to the controller itself. No motion capture or off-line optimization process is used.

Paper (2): "Robust Physics based Locomotion Using Low Dimensional Planning" - abstract: This paper presents a physics-based locomotion controller based on online planning. At each time-step, a planner optimizes locomotion over multiple phases of gait. Stance dynamics are modeled using a simplified Spring-Load Inverted (SLIP) model, while flight dynamics are modeled using projectile motion equations. Full-body control at each instant is optimized to match the instantaneous plan values, while also maintaining balance. Different types of gaits, including walking, running, and jumping, emerge automatically, as do transitions between different gaits. The controllers can traverse challenging terrain and withstand large external disturbances, while following high-level user commands at interactive rates.

e-Science involves the application of advanced computational methods to other areas of science and technology. It It has attracted a good deal of support over the past 10 years, and numerous groups have developed new techniques and software prototypes. Importantly, e-Science requires advanced in both computer science and the application area, making it an ideal driver for computer science research. In this talk, I will explore whether any of this work is actually making a difference. I will discuss our own projects work at the Monash e-Science and Grid Engineering (MeSsAGE) Lab, a computer science research laboratory devoted to new software development techniques that support e-Science applications. I will show how high throughput (aka parallel) scientific workflows have not only contributed to the state of the art in computer science, but are being adopted in research labs at Monash and internationally. In particular, I will highlight case studies in the medical imaging, chemistry and cardiac science.

Milgram's landmark six-degrees-of-separation experiment led to the fascinating small world hypothesis: take any two people in a social network, and they will be connected by a short chain of acquaintances. The extent to which the hypothesis is true is still actively debated. In this talk we give new experimental and theoretical results concerning Milgram's experiment. In particular, we discuss the possibility and the importance of performing purely digital replicas of the experiment, without human subjecs taking part in it. Joint work with Silvio Lattanzi and D.Sivakumar

Given a set of n points in L_1, how many dimensions are needed to represent all pairwise distances within a specific distortion ? This dimension-distortion tradeoff question is well understood for the L_2 norm, where O((log n)/epsilon^{2}) dimensions suffice to achieve 1+epsilon distortion. In sharp contrast, there is a significant gap between upper and lower bounds for dimension reduction in L_1. In this work, we show the first near linear lower bounds for dimension reduction in L_1. In particular, we show that 1+epsilon distortion requires at least n^{1-O(1/log(1/epsilon))} dimensions. Our proofs are combinatorial, but inspired by linear programming. In fact, our techniques lead to a simple combinatorial argument that is equivalent to the LP based proof of Brinkman-Charikar for lower bounds on dimension reduction in L_1.

Joint work with Alex Andoni, Moses Charikar and Huy Nguyen

There are numerous real-world problems that translate to searching for short distances. These distances can either be explicit (think of road navigation in Google maps) or implicit (think of the distance between humans and mice as the number of changes in their genomes). In my talk, I will describe some techniques that are useful for solving various shortest paths problems efficiently. I will begin with the problem of finding a shortest path in a "grid-like" graph. This problem arrises when computing the distance (minimum number of changes) between two sequences (say two genomes), and has many applications in computational biology, speech recognition, and information retrieval. A slightly more complicated grid-like graph arises when computing the distance (minimum number of changes) between two trees (say two XML files). This has applications in computer vision, compiler optimization, and natural language processing. I will then move to layered graphs (where shortest paths can be used to decode Hidden Markov Models) and then to planar graphs (where shortest paths can be used for VLSI design and geographical routing problems). Finally, I will discuss shortest paths in general graphs, and describe how to maintain shortest paths in a (more realistic) network whose edges occasionally fail. The problem has applications in game (auction) theory and is in tight connection with classical problems such as all-pairs shortest paths. My goal is to illustrate some general techniques that are useful for many of these problems. These techniques include the efficient computation of distance products, the reduced lengths method, and the use of compression as a tool for acceleration.

We consider the vertex coloring problem in the message-passing model of distributed computing. In this model the network is represented by a graph G of maximum degree Delta, in which the vertices host processors, and the communication is performed over the edges of G. Vertex coloring has numerous applications in communication networks, and thus it has drawn a considerable attention of researchers in the last three decades. The question of how many colors an efficient algorithm (that is, an algorithm with polylogarithmic time) is required to use is a central long-standing open question. Currently, efficient randomized algorithms that employ Delta + 1 colors are known, but it is unknown whether this number of colors can be achieved deterministically and efficiently. In 1987, in a seminal FOCS paper, Linial devised an efficient deterministic algorithm that employs Delta^2 colors. In his paper Linial also asked whether it is possible to employ a significantly smaller number of ! colors while still maintaining a deterministic algorithm with polylogarithmic running time. Despite a very intensive research in this direction in the last twenty years, this question remained open prior to our work. We present a deterministic algorithm that runs in polylogarithmic time, and employs Delta^{1 + epsilon} colors, for an arbitrarily small positive constant epsilon. Thus, our algorithm settles the long-standing question of Linial. Moreover, our results give rise to significantly improved algorithm for O(Delta)-coloring, and even for o(Delta)-coloring of very wide graph families. These results are achieved using a novel graph-decomposition technique. Specifically, the vertex set of the graph is partitioned into subsets that satisfy certain helpful properties. In particular, the subsets induce sparse subgraphs. This allows coloring the subgraphs with sufficiently small number of colors while utilizing the parallelism of the network to the full extent. Our recent results show that this technique is very powerful, and is useful for additional problems such as edge coloring.

We consider the problem of finding a maximum flow in directed planar graphs with multiple sources and sinks. For general (i.e., non-planar) graphs, the multiple-source multiple-sink maximum flow problem is as difficult as the standard single-source single-sink one. The reduction, however, does not preserve planarity. No efficient planarity exploiting algorithms for this problem were known until the current line of work. We present a divide-and-conquer algorithm that solves the problem on a planar graph with n nodes in O(n^1.5 log(n)) time. This is asymptotically faster than any previously known algorithm. Joint work with Glencora Borradaile, Philip Klein, Yahav Nussbaum and Christian Wulff-Nilsen. Preprints can be found at http://arxiv.org/abs/1012.5870 and http://arxiv.org/abs/1008.5332

Parallel programming is notoriously difficult and is still considered an artisan's job. Recently, the shift towards on-chip parallelism has brought this issue to the front stage. Commonly referred to as the Programmability Wall, this problem has already motivated the development of simplified parallel programming models, and most notably task-based models. In this talk, I will present Task Superscalar Multiprocessors, a conceptual multiprocessor organization that operates by dynamically uncovering task-level parallelism in a sequential stream of tasks. Task superscalar multiprocessors target an emerging class of task-based dataflow programming models, and thus enables programmers to exploit manycore systems effectively, while simultaneously simplifying their programming model. The key component in the design is the Task Superscalar Pipeline, an abstraction of instruction-level out-of-order pipelines that operates at the task-level and can be embedded into any manycore fabric to manage cores as functional units. Like out-of-order pipelines that dynamically uncover parallelism in a sequential instruction stream and drive multiple functional units, the task superscalar pipeline uncovers task-level parallelism in a stream of tasks generated by a sequential thread. Utilizing intuitive programmer annotations of task inputs and outputs, the task superscalar pipeline dynamically detects inter-task data dependencies, identifies task-level parallelism, and executes tasks out-of-order. I will describe the design of the task superscalar pipeline, and discuss how it tackles the scalability limitations of instruction-level out-of-order pipelines. Finally, I will present simulation results that demonstrate the design can sustain a decode rate faster than 60ns per task and dynamically uncover data dependencies among as many as ~50,000 in-flight tasks, using 7MB of on-chip eDRAM storage. This configuration achieves speedups of 95-255x (average 183x) over sequential execution for nine scientific benchmarks, running on a simulated multiprocessor with 256 cores.

Over the past decades the evolutionary approach has been used in many fields of computer science research. Lately, with the growth of computation power, Genetic Programming (GP) has been showing much promise. Our work is an attempt to apply the tree based GP approach to several zero-sum deterministic full knowledge to board games. We present published results on Lose Checkers as well as yet unpublished improved results, and also results on two other board games: 10X10 Checkers, and Reversi. Our system implements strongly typed GP trees, explicitly defined introns and multi-tree individuals. We use the GP trees to evaluate possible future game states. Used together with traditional search techniques the results show much promise and imply that tree based GP may be useful in finding good players for other similar games.

In egocentric perceptual control, an agent is tasked with enforcing a particular perseption on a given sensory system through actions within its environment. As an example, consider turning a robot's camera to keep a particular colour blob centered in its field of vision. Consider now what happens if the blob size represents proximity of a dangerous object? If we would like to keep safe distance (or simply run away) we would like the perceived size of the colour blob to diminish with time. However, that means that the task is described though a dynamical concept: how does the blob size (and location in visual field) changes over time. To enable a consistent treatment of such egocentric perceptual tasks, I will introduce a control framework termed Dynamics Based Control (DBC), and its implementation for partially observed Markovian environments. I will then show how this implementation can be extended for a multi-agent scenario. In particular, I will demonstrate a domain where this method works even when agents can not directly monitor the activity of other participants.

RNA mutational analysis at the secondary structure level can be useful to a wide range of biological applications. It can be used to predict an optimal site for performing a nucleotide mutation at the single molecular level, as well as to analyze basic phenomena at the systems level. In the past several years, the program RNAmute that is structure-based and relies on RNA secondary structure prediction has been developed for assisting in RNA mutational analysis. It has been extended from single-point mutations to treat multiple-point mutations efficiently by initially calculating all suboptimal solutions, after which only the mutations that stabilize the suboptimal solutions and destabilize the optimal one are considered as candidates for being deleterious.

In the broadcast model on a tree, information is sent from the root over the edges which act like independent noisy channels, to the leaves of the tree at depth n. The reconstruction problems asks whether the information at the root can be recovered from random observations of the leaves with good probability as n becomes larger. This problem arises naturally in biology, information theory and statistical physics. The analysis involves understanding the tradeoff between the replication of information over the leaves and the increasing noise as the distance from the root increases. There is evidence that reconstruction on trees plays an important role in explaining threshold phenomena in random constraint satisfaction problems such as Random k-SAT or random colorings of a random graph as well as the efficiency of finding and sampling solutions for these problems. In this talk, I will present tight bounds on the threshold for reconstruction for independent sets and results on the colorings reconstruction problem on trees. I will also mention algorithmic implications and the connection with random constraint satisfaction problems. Based on joint works with Sly and Tetali; Maneva; and Vera, Vigoda and Weitz.

Scheduling is a major challenge in the design of operating systems, and is used to achieve quality of service guarantees. In many real-life environments, e.g., cloud computing, the service provider and users can have different, possibly conflicting, interests, and behave strategically. In this talk we take an economic mechanism design approach to scheduling, and examine scheduling from both a theoretical perspective and an experimental perspective. We consider the challenge of designing mechanisms, that is, algorithms coupled with pricing schemes, that are both manipulation resistant and guarantee good quality of service. Our contribution is twofold: (1) An inapproximability result for randomized mechanisms in a well-studied machine scheduling context. (2) A game-theoretic model for provider-consumer dynamic interaction and simulation results based on real data that establish existence of a single pure Nash equilibrium in practice.

No prior knowledge is assumed. Joint work with Lior Amar, Amnon Barak, Michael Schapira, and Sergei Shudler

A key question in cooperative game theory is that of coalitional stability, usually captured by the notion of the core- the set of outcomes such that no subgroup of players has an incentive to deviate. However, some coalitional games have empty cores, and any outcome in such a game is unstable. In this talk, we investigate the possibility of stabilizing a coalitional game by using external payments (subsidies). We consider a scenario where an external party, which is interested in having the players work together, offers a supplemental payment to the grand coalition, thereby extending the core. We define the cost of stability (CoS) as the minimal external payment that stabilizes the game. We provide general bounds on the cost of stability in several classes of games, and explore its algorithmic properties and its relations with other solution concepts. Joint work with Yoram Bachrach, Edith Elkind, Dmitrii Pasechnik, Michael Zuckerman, Joerg Rothe, and Jeffrey S. Rosenschein.

Recent advances in large-margin classification of data residing in general metric spaces (rather than Hilbert spaces) enable classification under various natural metrics, such as edit and earthmover distance. The general framework developed for this purpose by von Luxburg and Bousquet [JMLR, 2004] left open the question of computational efficiency and providing direct bounds on classification error. We design a new algorithm for classification in general metric spaces, whose runtime and accuracy depend on the doubling dimension of the data points. It thus achieves superior classification performance in many common scenarios. The algorithmic core of our approach is an approximate (rather than exact) solution to the classical problems of Lipschitz extension and of Nearest Neighbor Search. The algorithms generalization performance is established via the fat-shattering dimension of Lipschitz classifiers. This is joint work with Aryeh Kontorovich and Robert Krauthgamer.

In this talk I will give an overview of the field of parameterized complexity. This is a relatively new and rapidly developing branch in theoretical computer science that provides a framework for coping with hard computational problems. The overview will be influenced by my research on this topic in recent years. I will start with general motivation, and attempt to describe the main focus of research in the area. I will then review some of my own work, and explain how its related to the general interests of the field. The talk will be in most of its parts non-technical, and is intended for a general computer scientist audience.

Consider an NP-hard optimization problem with input size $n$ which is associated with a parameter $k$ (the parameter may be, for instance, the maximum allowed output size). We say that this problem is fixed-parameter tractable (FPT) if it can be solved by a fixed-parameter algorithm, i.e. an algorithm whose runtime is uniformly polynomial in $n$, though exponential (or even superexponential) in $k$. Examples of such runtimes are $O((2^k)*n)$, $O(3^k+n^2)$ and so on. When $k$ is small, the hope is that the respective dependence on $k$ is also small. In this case, an NP-hard optimization problem can be solved in a low polynomial time. Thus, if the considered parameter is reasonably small in practical applications, fixed-parameter algorithms can be used as a method of coping with NP-hardness, both precise (unlike approximation algorithms) and efficient (unlike ordinary brute-force algorithms). Graph separation problems comprise a large class of problems where the objective is to remove the smallest set of vertices (or edges) from the given graph so that certain structures in the graph are broken. Examples of such structures: paths between certain pairs of vertices, directed cycles, odd cycles, etc. In real-world applications of these problems it is often reasonable to assume that the separator (i.e. the set of vertices removed) is much smaller than the size of the whole graph. It is therefore natural to solve these problems by the means of fixed-parameter algorithms. However, designing good fixed-parameter algorithms for these problems is a very tricky task. In fact, despite many years of active research, for a number of separation problems it was not even clear if they are FPT. In this talk I will overview current results related to design of fixed-parameter algorithms for separation problems. To make the talk self-contained, I will start from definition of the fixed-parameter algorithm and provide a simple example of such algorithm. Then I will informally describe a fixed parameter algorithm for the multiway cut problem. Then I will show how the methodology underlying this algorithm has helped to resolve fixed-parameter tractability of a number of challenging problems that stayed open for many years despite attempts of many researchers. Finally, I will overview some challenging questions that are still open.

Mining sequential patterns is a key objective in the field of data mining due to its wide range of applications. Given a database of sequences, the challenge is to identify patterns which appear frequently in different sequences. Well known algorithms have proved to be efficient, however these algorithms do not perform well when mining databases that have long frequent sequences. We present CAMLS, Constraint-based Apriori Mining of Long Sequences, an efficient algorithm for mining long sequential patterns under constraints. CAMLS is based on the apriori property and consists of two phases, event-wise and sequence-wise, which employ an iterative process of candidate-generation followed by frequency-testing. The separation into these two phases allows us to: (i) introduce a novel candidate pruning strategy that increases the efficiency of the mining process and (ii) easily incorporate considerations of intra-event and inter-event constraints. Experiments on both synthetic and real datasets show that CAMLS outperforms previous algorithms when mining long sequences.

The standard problem of authorship attribution is that of determining which of a small closed set of candidates (for each of whom we have known writings) is the author of some anonymous text. This problem is well understood and can be regarded as solved (using standard NLP and machine learning tools). However, in the real world, we rarely encounter the standard problem, but rather much harder problems in which there are thousands of candidate authors and we have no guarantee that any of them is the actual author. We will review the state of the art regarding the harder versions and will discuss a number of very useful applications. If time permits, we will discuss some recent astonishing results on biblical literature.

The popularity of game theory can be witnessed through the widespread application of its models to different research areas. However, the application of game theoretical models is often done with limited consideration to the underlying assumptions of the models. In this talk I present two problem domains and show how such inherited assumptions limit the accuracy and usefulness of the solutions. The first domain comes form mainstream AI, game-tree search. Game-tree search algorithms are heavily based on theoretical results that assume both unbounded computational resources and rational players. In reality, for most games, players cannot search the entire tree due to computational limitations. As such, algorithms use various techniques to increase their search horizon under a common assumption that deeper search yields more accurate decisions. However, it was shown more than 30 years ago that there exist a class of games, namely Pathological games, in which this assumption is incorrect. In this research I show that game-tree pathology is a local phenomena that might exist in *all* games. I will then present an algorithm that recognizes pathological sub-trees and adapts its decision procedure accordingly. The second problem is the evolution of cooperation, an interdisciplinary problem from AI and Theoretical Biology. To study this problem researchers often use the Prisoner's dilemma game to model the interactions between players. Most of the existing works use the selfish, self-maximizing player model that was inherited from game theoretical analysis. However, theories from the social and behavioral sciences show that people explicitly consider the payoff of other players when making decisions. As such, we utilize the Social Value Orientation theory to present a new player model which provide a more accurate description of human behavior. With this new model we were able to gain new insights on the evolution of cooperative societies.

Linear classification is a fundamental problem of machine learning, in which positive and negative examples of a concept are represented in Euclidean space by their feature vectors, and we seek to find a hyperplane separating the two classes of vectors. We give the first sublinear-time algorithms for linear classification and other related problems in machine learning, including the kernelized versions of these problems. These new algorithms are based on a primal-dual approach, and use a combination of novel sampling techniques and the randomized implementation of online learning algorithms. We give lower bounds which show our running times to be nearly best possible in the unit-cost RAM model. Joint work with Ken Clarkson and David Woodruff, appeared in FOCS 2010.

Boosting is a general method in machine learning to improve the prediction accuracy of weak learners. A classic boosting algorithm that deals with the case of deciding between two possible classes is the Adaboost algorithm of Freund and Shpire (JCSS 1997). We propose a new approach to generalize AdaBoost to the multi-class setting. The basic idea is to map labels and confidence-based classifiers to a normed vector space, and to measure performance by distances in this space. The result is a meta-algorithm whose concrete implementations can address various scenarios, from the standard case where each example is assigned to a single class, to more complex settings where each example may belong to a set of classes, and where there is a structure on the label-space which can be captured by distances in a normed space. Joint work with Michal Rosen-Zvi.

Given a weighted graph, the maximum weight matching problem (MWM) is to find a set of vertex-disjoint edges with maximum weight. In the 1960s Edmonds showed that MWMs can be found in polynomial time. At present the fastest MWM algorithm, due to Gabow and Tarjan, runs in roughly $O~(msqrt{n})$ time, where $m$ and $n$ are the number of edges and vertices in the graph. Surprisingly, restricted versions of the problem, such as computing $(1-epsilon)$-approximate MWMs or finding maximum cardinality matchings, are not known to be much easier. The best algorithms for these problems also run in $O~(msqrt{n})$ time.

In this paper we present the first near-linear time algorithm for computing $(1-epsilon)$-approximate MWMs. Specifically, given an arbitrary real-weighted graph and $epsilon>0$, our algorithm computes such a matching in $O(mepsilon^{-2}log^3 n)$ time. The previous best approximate MWM algorithm with comparable running time could only guarantee a $(2/3-epsilon)$-approximate solution. In addition, we present a faster algorithm, running in $O(mlog nlogepsilon^{-1})$ time, that computes a $(3/4-epsilon)$-approximate MWM.

I will introduce confidence-weighted linear classifiers, a class of algorithms that maintain confidence information about classifier parameters. Instead of a single weight vector, learned hypotheses are given by Gaussian distributions over weight vectors, with a covariance matrix that represents uncertainty about weights and correlations between different weights. Learning in this framework updates parameters by estimating weights and increasing model confidence. I will describe few online algorithms that maintain a Gaussian distribution over weight vectors, updating the mean and variance of the model with each instance. A mistake bound analysis shows that indeed the algorithm performs better under some conditions and also relates between our model and the margin and loss analysis of previous models. Empirical evaluation on a range of NLP tasks show that our algorithm improves over other state of the art online and batch methods, learns faster in the online setting, ends itself to better classifier combination after parallel training and is suites better for active learning. Based on joint work with Mark Dredze,Alex Kulesza and Fernando Pereira.

The classical randomness notions of Schnorr and Kurtz permit gamblers to bet any amount of money within their means. In this talk, we consider a more realistic paradigm in which gamblers must place a minimum bet of one dollar. The corresponding randomness notion turns out to be incomparable with the classical notions mentioned above. Most casinos operate by exploiting the law of large numbers, however, by examining an even more restrictive model in which we ban wagers greater than a million dollars, we obtain an alternate principle through which an online casino might operate profitably. The open questions at the end of this talk should be accessible to a general audience.

I will present an approach to solving NP-Complete problems in polynomial time using an optical architecture. Specifically I will present our solution for the Hamiltonian Cycle Problem, and the Permanent problem.

The process of extracting useful knowledge from large datasets has become one of the most pressing problems in today's society. The problem spans entire sectors, from scientists to intelligence analysts and web users, all of whom are constantly struggling to keep up with the larger and larger amounts of content published every day. With this much data, it is often easy to miss the big picture.

In this paper, we investigate methods for automatically connecting the dots – providing a structured, easy way to navigate within a new topic and discover hidden connections. We focus on the news domain: given two news articles, our system automatically finds a coherent chain linking them together. For example, it can recover the chain of events starting with the decline of home prices (January 2007), and ending with the ongoing health-care debate.

We formalize the characteristics of a good chain and provide an efficient algorithm (with theoretical guarantees) to connect two fixed endpoints. We incorporate user feedback into our framework, allowing the stories to be refined and personalized. Finally, we evaluate our algorithm over real news data. Our user studies demonstrate the algorithm's effectiveness in helping users understanding the news.

Cooperative game theory is all about how selfish agents might agree to collaborate and then share their spoils. It allows answering questions such as: Can selfish behaviour jeopardize our heath treatments in hospitals? Would the political power balance change if a big party decided to split into two smaller parties? How might pirates share a hidden treasure when they need each other to find it? Cooperation can be problematic when agents collaborate to attack an economic or political system. For example, agents participating in an auction can coordinate their bids in order to pay less for obtaining their items and political parties may strategically merge or split to increase their influence. This talk examines computational aspects of such phenomena, focusing on collusion in auctions and attacks in decision making bodies. Auctions based on the VCG mechanism are excellent in achieving truthful bids and an optimal allocation when agents do not collude. However, they are very susceptible to collusion. I will demonstrate this in multi-unit auctions and path procurement auctions, showing how the colluders can compute their optimal joint bidding strategy and reasonable agreements for sharing the gains. I will then consider attacks in weighted voting games, a known model for cooperation between agents in decision-making bodies, showing how agents can compute strategies that increase their power. The analysis for both domains is based on the core and the Shapley value, prominent solution concepts from cooperative game theory.

Active contours are used extensively in vision for more than two decades, primarily for applications such as image segmentation and object detection. The vast majority of active contours models make use of closed curves and the few that employ open curves rely on either fixed boundary conditions or no boundary conditions at all. We discuss a new class of open active contours with free boundary conditions, in which the end points of the open active curve are restricted to lie on two parametric boundary curves. We discuss how this class of curves may assist and facilitate various vision applications and we demonstrate its utility in applications such as boundary detection, feature tracking, seam carving, and image stitching.

Identification of rare variants by resequencing is important both for detecting novel variations and for screening individuals for known disease alleles. New technologies enable low-cost resequencing of target regions, although it is still prohibitive to test more than a few individuals. We propose a novel pooling design that enables the recovery of novel or known rare alleles and their carriers in groups of individuals. The method is based on a Compressed Sensing (CS) approach, which is general, simple and efficient. CS allows the use of generic algorithmic tools for simultaneous identification of multiple variants and their carriers. We model the experimental procedure and show via computer simulations that it enables the recovery of rare alleles and their carriers in larger groups than were possible before. Our approach can also be combined with barcoding techniques to provide a feasible solution based on current resequencing costs. For example, when targeting a small enough genomic region (~100 bp) and using only ~10 sequencing lanes and ~10 distinct barcodes per lane, one recovers the identity of 4 rare allele carriers out of a population of over 4000 individuals. We demonstrate the performance of our approach over several publicly available experimental data sets.

Joint work with Amnon Amir from the Weizmann Institute of Science, and Or Zuk from the Broad Institute of MIT and Harvard

Hidden Markov models (HMMs) were introduced and studied in the late 1960s, and early 1970s. One of the first applications of HMMs was speech recognition, starting in the mid-1970s. In the second half of the 1980s, HMMs began to be applied to recognition of handwritten text in images, commonly known as offline handwriting recognition (OHR). In this talk, I'll present the Hidden Markov Model, and show some examples, of how it can be applied to offline handwriting recognition.

Supervised Learning problems often involve inference of complex structured labels such as image segmentations of grid shape graphs. To achieve high accuracy in these tasks, one is often interested in introducing dependencies between label parts. However this usually results in inference problems that are NP hard. A natural approach is to use tractable approximation of the inference problems. In this talk I will present our recent work on approximate inference, using duality to extend the belief propagation algorithms to convex programs. Specifically, we show how convex belief propagation algorithms solve convex relaxations of the partition function, also referred as the free energy, as well as linear programming relaxations of integer linear programs. Also, I will present how duality and local inference can be applied to approximate current learning frameworks of conditional random fields (CRFs) and structured support vector machines (SVMs), and show highly scalable message-passing algorithms for these approximations. I will also present how these approximations can be applied to learn image segmentations. Based on joint work with: Amnon Shashua, Raquel urtasun, Ross Girshik

Can we use statistical Natural Language Processing methods on redundant data? Document collections (corpora) may include large amount of redundancy due to copied texts, this phenomena is common in news articles and Electronic Health Records. Methods for detecting and handling redundancy are common in the fields of Bioinformatics for creating sequence databases as well as for plagiarism detection. We will show that redundant text may bias statistical methods for processing such corpora as well as a robust heuristic for identifying a non-biased subset a corpus.

Gene Translation (GT) is the complex process of decoding the mRNA sequences by the ribosome to produce proteins. The rapid accumulation of genomic data and large scale measurements that are related to GT (e.g. sequencing of genomes and measurements of gene expression, protein abundance, and ribosome densities) enables developing computational models and performing large scale systems biological study of this biological process.In this talk, I will survey our recent results in understanding and modeling GT.Specifically, I will describe: (1) a novel systems biological analysis of the dynamic of ribosome movement based on the different features mRNA sequences, a key to understanding GT. (2) A computationally efficient predictive model of GT that is based on the physical and dynamic nature of this process. The talk is self-contained and requires no prior knowledge in Biology.

We define a novel variation on the constrained sequence alignment problem, the Sequence Alignment with Regular Expression Path Constraint (SA-REPC) problem, in which the constraint is given in the form of a regular expression. Our definition extends and generalizes the existing definitions of alignment-path constrained sequence alignments to the expressive power of regular expressions. We give a solution for the new variation of the problem and demonstrate its application to integrate microRNA-target interaction patterns into the target prediction computation. Our approach can serve as an efficient filter for more computationally demanding target prediction filtration algorithms. We compare our implementation for the SA-REPC problem, cAlign, to other microRNA target prediction algorithms.

The notion of Generalized Oblivious Transfer (GOT) was introduced by Ishai and Kushilevitz. In a GOT protocol, Alice holds a set $U$ of messages. A decreasing monotone collection of subsets of $U$ defines the retrieval restrictions. Bob is allowed to learn any permissible subset of messages from that collection, but nothing else, while Alice must remain oblivious regarding the selection that Bob made. We propose a simple and efficient GOT protocol that employs secret sharing. We compare it to another secret sharing based solution for that problem that was recently proposed by Shankar, Srinathan and Pandu Rangan. In particular, we show that the access structures that are realized by the two solutions are related through a duality-type relation that we introduce here. We show that there are examples which favor our solution over the second one, while in other examples the contrary holds. Two applications of GOT are considered — priced oblivious transfer, and oblivious evaluation of multivariate polynomials.

Coin-tossing protocols are protocols that generate a random bit with uniform distribution. These protocols are used as a building block in many cryptographic protocols. Cleve [STOC 1986] has shown that if at least half of the parties can be malicious, then, in any r-round coin-tossing protocol, the malicious parties can cause a bias of Omega(1=r) to the bit that the honest parties output. However, for more than two decades the best known protocols had bias t pr , where t is the number of corrupted parties. Recently, in a surprising result, Moran, Naor, and Segev [TCC 2009] have shown that there is an r-round two-party coin-tossing protocol with the optimal bias of O(1=r). We extend Moran et al. results to the multiparty model when less than 2/3 of the parties are malicious. The bias of our protocol is proportional to 1=r and depends on the gap between the number of malicious parties and the number of honest parties in the protocol. Speci cally, for a constant number of parties or when the number of malicious parties is somewhat larger than half, we present an r-round m-party coin-tossing protocol with optimal bias of O(1=r).

We will discuss some management problems that a minimarket manager may encounter. We consider the simplistic situation in which there are two checkout counters, with one of them operating as an express line. As will be explained, managing this extra simple situation is usually problematic. We will then give examples of some favorable circumstances in which the management problem suddenly becomes easy. Some of these examples are surprising to say the least and involve some of the greatest achievements of human intelect. No prior minimarket work experience is assumed.

In many fields observations are performed irregularly along time, due to either measurement limitations or lack of a constant immanent rate. While discrete-time Markov models (as Dynamic Bayesian Networks) introduce either inefficient computation or an information loss to reasoning about such processes, continuous-time Markov models assume either a discrete state space (as Continuous-Time Bayesian Networks), or a flat continuous state space (as stochastic differential equations). To address these problems, we present a new modeling class called Irregular-Time Bayesian Networks (ITBNs), generalizing Dynamic Bayesian Networks, allowing substantially more compact representations, and increasing the expressively of the temporal dynamics. In addition, a globally optimal solution is guaranteed when learning temporal systems, provided that they are fully observed at the same irregularly spaced time-points, and a semiparametric subclass of ITBNs is introduced to allow further adaptation to the irregular nature of the available data.

Some of the fundamental algorithms in RNA secondary structure are dynamic programming algorithms. The goal of improving these algorithms is motivated by the understanding that RNA structure helps to determine function. While these algorithms have been around since the early 1980’s it was only recently that the well known dynamic programming speed up –Four Russians Speed Up– been applied. In this talk presented is the Four Russians Speed Up applied to the Basic RNA folding algorithm and the Sankoff alignment and folding algorithm. The speedup leads to a time reduction from O(n^3) to O(n^3/log(n) ) for RNA secondary structure and from O(n^6) to O(n^6/log(n^2)) time reduction for the Sankoff algorithm of simultaneous alignment and folding.

Document image analysis (DIA) refers to the process of converting a raster image of a document page (a matrix of pixels) to a symbolic form consisting of textual (characters, digits, punctuation, words) and graphical (lines, geometric shapes, etc.) objects. Document descriptions in terms of these higher-level objects are significantly more compact than their image counterparts. More importantly, the rich semantic content of such descriptions makes it possible to manipulate these documents to serve a variety of uses such as searching them for specific patterns or classifying and combining them according to some criteria. Most DIA systems consist of, Binarization, Page analysis and segmentation, Preprocessing, Feature extraction, Classification, and Post processing. This talk will include a summery of some results we had, including page layout segmentation, key word searching & spotting and script recognition.

The DNA in eukaryotic cells is packed into the chromatin that is composed of nucleosomes. High resolution mapping of the nucleosome core particles on the sequence is a problem of great interest because of the role nucleosomes play in different cellular processes. Nucleosome positioning is determined by multiple factors, including the action of chromatin remodelers, competition with site specific DNA-binding proteins and DNA sequence preferences of the nucleosomes themselves. More than 30 years ago scientists started to investigate the role of the sequence in nucleosome positioning. It has emerged since then that there are intrinsic signals in the DNA sequence, such as 10.4 bp periodicity, correlated with the nucleosome formation, but the question is still not fully answered. We have designed universal high resolution nucleosome mapping probe using nucleosome DNA bendability pattern extracted from large nucleosome DNA database of C. elegans. The probe can be used to predict nucleosome affinity to any DNA sequence of interest. Our implementation of high resolution nucleosome mapping algorithm is freely available for the scientific community via our FineStr web-server.

In recent studies non-coding RNA gets more attention. For these RNAs that perform various roles within the cell, structure is much more important then the sequence. This talk will encompass: Different approaches to gauge similarity/distance between RNA structures. Contraction of RNA structures from favoritble building blocks in order to manifacture active RNAs. Extended RNA inverse - design RNA that will fold to a desired structure and physical properties.

Let H be a graph. By TH we denote the graphs obtained from H by replacing the edges of H with vertex-disjoint paths that internally consist of an arbitrary number of degree two vertices (possibly zero). A member of TH is called a topological H (or a subdivision of H).

The question: "what is the structure of a graph containing no member of TH as a subgraph, for some fixed H?" is one of the main problems in graph theory today since the 1930s. Such "TH-free" graph families are found in many of the so called "pearls" of graph theory: Kuratowski's theorem, Tutte's integer-flow conjectures, Hajos' conjecture, Kelmans-Seymour conjecture and more.

Over the years, some partial results unfolding the structure of such "TH-free" graph families were discovered for a small number of configurations H. Sadly, each such result required some special handeling well-suited for the specific H being considered. Thus, no systematic approach towards answering the above question is known (or even conjectured).

In the last year or so I have been trying to make progress on the above question.

In the talk I propose I shall begin with an overview of the notion of "subdivisions in graphs". I then hope to introduce some of my results regarding "TH-free" graphs and spend some time on the notion of "semi-topological minors" (special forms of subdivisions). I will make an attempt to include some of my more recent results regarding the Kelmans-Seymour conjecture postulating that: "the 5-connected nonplanar graphs contain a TK_5".

The popularity of web-based mapping services such as Google Earth/Maps and Microsoft Virtual Earth (Bing), has led to an increasing awareness of the importance of spatial data and its incorporation into both web-based search and the databases that support it, whereas in the past attention to spatial data had been primarily limited to geographic information systems (GIS). An immediate byproduct of this awareness is the expectation of a real time response as is the experience of users of spreadsheets. Spatial data is distinguished from conventional data by having extent, which means that rather than being limited to locations, it also includes collections of locations [and, most importantly in both cases, their attributes]. Having extent is challenging in several respects. First, it is not easy to order such data which impacts the ability to retrieve it quickly. Second, the specification of the data is both vague and ambiguous by virtue of the amount of precision that is needed to make it useful. The ambiguity is very clear when one considers that a location as well as a collection of locations can be specified either or both geometrically via, for example, its centroid or its boundary, and verbally via the name that is used to refer to it. The latter is both aided and complicated by the possible need to make use of knowledge, whether implicit or explicit, of the information that is inherent in its container hierarchy. In this lecture we explore how these issues are manifested and resolved, both conceptually, and via demonstrations of real systems, thereby demonstrating how wide a net has been cast on geographic information systems by todaysapplications.

*The lecture is sponsored by the ACM Distinguished Speaker Program*

The problem of deciding which advertisements a publisher should display, given some contextual information about its users, is nicely captured by the contextual bandit setting. In this talk, I will give an overview of the contextual bandit problem (also known as the multiarmed bandit problem with expert advice) and present new algorithms for this setting. I will focus on a couple recent theoretical developments that are bringing us closer to getting similar guarantees in the bandit setting as we have in supervised learning. I will also discuss some generalizations of the contextual bandit problem that are particularly relevant to computational advertising.

Imposing constraints in the form of a finite automaton or a regular expression is an effective way to incorporate additional a priori knowledge into sequence alignment procedures. With this motivation, Arslan introduced the Regular Language Constrained Sequence Alignment problem and proposed an $O(n^2t^4)$ time and $O(n^2t^2)$ space algorithm for solving it, where $n$ is the length of the input strings and $t$ is the number of states in the non deterministic automaton which is given as input. Chung et al. proposed a faster $O(n^2t^3)$ time algorithm for the same problem.

In this talk, I present an additional speed up to the algorithms for Regular Language Constrained Sequence Alignment by reducing their worst case time complexity bound to $O(n^2t^3/log t)$. This is done by establishing an optimal bound on the size of Straight-Line Programs solving the maxima computation subproblem of the basic dynamic programming algorithm.

In addition, I present a another solution we have studied, which is based on a Steiner Tree computation. While this solution does not yield an improvement in the worst case, our simulations show that both approaches are efficient in practice, especially when the input automata are dense.

Joint work with Gregory Kucherov and Michal Ziv-Ukelson.

Let $X$ be a simple region, and let $Q$ be a set of $m$ points in $X$. A disk cover of $Q$ with respect to $X$, is a set of disks (of variable radii), such that the union of these disks covers $Q$ and is contained in $X$. In other words: (i) each disk in the cover is contained in $X$, and (ii) each point $q in Q$, lies in a disk of the cover. A minimum disk cover of $Q$ with respect to $X$ is a disk cover of $Q$ with respect to $X$ of minimum cardinality. The problem of computing a minimum disk cover of a point set $Q$ with respect to a simple region $X$ arises, e.g., in the following setting. $X$ represents a secured area, and each point of $Q$ represents a client of a radio network. One must place the smallest possible number of transmitters, such that each client is served by at least one of the transmitters (i.e., is within the transmission range of at least one of the transmitters), and any point outside the area, is outside the range of each of the transmitters. Surprisingly, this problem is solvable in polynomial time, due to very nice combinatorial structure. In a paper presented at WADS 2009 and recently accepted to Algorithmica, we presented an exact polynomial time algorithm for the above problem (joint work with Matya Katz). In a paper accepted to ESA 2010 we give an improved almost-linear time algorithm (Joint work with Haim Kaplan, Matya Katz and Micha Sharir). I'll give the highlights of the proof, which I find simple, nice and elegant.

Many real-world problems are distributed by nature. Examples include network file systems, the scheduling of meetings among multiple agents, flight scheduling, industrial control systems, mobile sensor nets, wireless routing, etc. Agents or nodes that solve such distributed problems often coordinate their moves and share information to improve the efficiency of the problem solving process. Distributed search for solving distributed constraints problems is a domain in which agentsare naturally cooperative. In problems such as the meetings scheduling, agents are assumed to cooperate faithfully in finding a globally optimal solution. The degree of information sharing by the agents during search spans a wide range. At one extreme one can have an algorithm that involves sharing complete information by the agents. At this extreme case several agents can appoint a single representative (mediator) to solve their combined search problem by itself. The mediator agent receives the complete information of the agents and uses it for the search it performs. When coordination among the agents during search is less tight, agents could keep their private information about their constraints with other agents and only respond to specific requests about constraints values. At this other extreme we propose the use of asymmetric constraints among agents.

A {it generalized} Davenport-Schinzel sequence is one over a finite alphabet, none of whose subsequences are isomorphic to some fixed {it forbidden subsequence}. Davenport-Schinzel sequences have been used in bounding the complexity of geometric arrangements and the running time of algorithms and data structures. Let Ex(s,n) be the maximum length of an s-free sequence over an n-letter alphabet. The foremost open problem in this area is to characterize the asymptotic growth of Ex(s,n), for different forbidden subsequences s. In this talk I will survey everything that is known and worth knowing about the function Ex(s,n), including some new results

In this talk I will begin with a short background for computer scientists on proteins and their structures. Then I will give a brief overview on some aspects of the challenging field of computational protein structure prediction. In the second part of the talk I will summarize the main project of my PhD thesis: A novel cooperative energy term for hydrogen bonds. (In more details: hydrogen bonds are major constituents of protein structures and thus have been subject to many works. Yet, the common treatment of hydrogen bonds by energy functions is a considerable contributor to the infamous "local minima problem", which has annoyed predictors since the earliest days of this field. Cooperative terms for hydrogen bonds attempt to ease this problem by shifting the focus from the individual hydrogen bond to patterns of hydrogen bonds. I will present the main ideas of our novel cooperative hydrogen-bond term that is both effective and computationally efficient).

We review simple numerical methods for approximating solutions to second order differential equations. The specific aim is to explain why the simple "Verlet" algorithm, which is a direct second order finite difference approximation to the second order differential, is so desirable and widely used for initial value problems with conservation properties. The application of broad interest is Molecular Dynamics, but we will review the method by interpreting the basic properties of the method applied to the harmonic oscillator. Stability, energy conservation and time step control will be addressed for the discrete time.

We define a simple mathematical problem which we call the optimal base problem: Given a multiset S of positive integers, find a numeral base B such that the size of the representation of S in B is minimal. For example if S={16,30,54,60} then in base 10 the sum of the digits is 25, while in base 2 it is 13, and in base 3 it is 12. The problem gets more interesting when we allow mixed radix bases such as B=<3,5,2,2} with respect to which the sum of the digits in S is only 9. We analyze the complexity of this problem and propose a quasi-polynomial algorithm to solve it. Our implementation improves significantly on existing techniques. Finally we present also an application.

Joint work with Michael Codish, Carsten Fuhs and Peter Schneider-Kamp

The phenomenon of visual curve completion, where the visual system completes the missing contours of an occluded object, is a major problem in computer vision and vision research. I will first present some of the state-of-the-art algorithms for curve completion, as well as relevant perceptual and biological insights. Then I will discuss our own approach for curve completion, which is based on the abstraction of the visual cortex with modern differential geometry tools. In this work we present formal theoretical analysis and numerical solution methods, and we show results on natural images.

To save on municipal budget we wish to minimize the number of policemen in a city described by a graph. The policemen are positioned at graph vertices (cross-roads) and are in control of their own vertex along with its neighbors. Whenever a criminal occurs in her vicinity, the guard sends alarm signal to the central office. The condition is that having known the set of worried policemen, the department head should be able to locate the unique vertex where the crime happens. We will prove that in Manhattan the police posts have to be placed at exactly 35% of the street intersections. We will also survey known results on the problem in alternative settings: different types of graphs, the number of thieves greater than one, and the graphical radius of control greater than one.

Decision making is the ability to decide on the best alternative among a set of candidates based on their value. In many real-world domains the value depends on events that occur dynamically, so that the decision is based on dynamically changing uncertain information. When there is a cost to waiting for more information, the question is when to make the decision. Do you stop and make the best decision you can, given the information you have so far, or do you wait until more information arrives so you can make a better decision? We propose a model that characterizes the influence of dynamic information on the utility of the decision. Based on this model, we present an optimal algorithm that guarantees the best time to stop. Unfortunately, its complexity is exponential in the number of candidates. We present an alternative framework in which the different candidates are solved separately. We formally analyze the alternative framework, and show how it leads to a range of specific heuristic algorithms. We evaluate the optimal and the simplest heuristic algorithms through experiments, and show that the heuristic algorithm is much faster than the optimal algorithm, and the utility of the winner it finds is close to the optimum.

In this paper we present a new approach for interactive view-dependent rendering of large polygonal datasets, which relies on advanced features of modern graphics hardware. Our preprocessing algorithm starts by generating a simplified representation of the input mesh. It then builds a multiresolution hierarchy for the simplified model. For each face in the hierarchy, it generates and assigns a displacement map that resembles the original surface represented by that face. At runtime, the multiresolution hierarchy is used to select a coarse viewdependent level-of-detail representation, which is sent to the graphics hardware. The GPU then refines the coarse representation by replacing each face with a planar patch, which is elevated according to the assigned displacement map. Initial results show that our implementation achieves quality images at high rates.

The celebrated Johnson-Lindenstrauss lemma says that every n points in Euclidean space can be represented with O(log n) dimensions with only a minor distortion of pairwise distances. It has been conjectured that a such-improved dimension reduction representation is achievable for many interesting data sets, by bounding the target dimension in terms of the intrinsic dimensions of the data (for example, by replacing the log n term with the doubling dimension). This question appears to be quite challenging, requiring new (nonlinear) embedding techniques. We make progress towards resolving this question by presenting two dimension reduction theorems with similar flavor to the conjecture. For some intended applications, these results can serve as an alternative to the conjectured embedding. [Joint work with Robert Krauthgamer.]

A power assignment is an assignment of transmission power to each of the nodes of a wireless network, so that the induced communication graph has some desired properties. The cost of a power assignment is the sum of the powers. The {em energy} of a transmission path from node $u$ to node $v$ is the sum of the squares of the distances between adjacent nodes along the path. For a constant $t >1$, an energy $t$-spanner is a graph $G'$, such that for any two nodes $u$ and $v$, there exists a path from $u$ to $v$ in $G'$, whose energy is at most $t$ times the energy of a minimum-energy path from $u$ to $v$ in the complete Euclidean graph.

We study the problem of finding a power assignment, such that (i) its induced communication graph is a `good' energy spanner, and (ii) its cost is `low'. We show that for any constant $t >1$, one can find a power assignment, such that its induced communication graph is an energy $t$-spanner, and its cost is bounded by some constant times the cost of an optimal power assignment (where the sole requirement is strong connectivity of the induced communication graph).

Based on joint work with Rom Aschner, Paz Carmi and Matya Catz.

Admissible heuristics are a key tool for optimizing heuristic-search planning procedures such as A* or IDA*. Most typically, such a heuristic for domain-independent planning is defined as the (optimal) cost of achieving the goals in an over-approximating abstraction of the planning problem in hand. Such an abstraction is obtained by relaxing certain constraints in the specification of the real problem, and the desire is to obtain a provably poly-time solvable, yet informative abstract problem. The main questions are thus:

1. What constraints should we relax to obtain such an effective over-approximating abstraction? 2. How should we combine information provided by multiple such abstractions?

In this talk we consider both these questions, and present some recent formal and empirical results that help answering these questions (sometimes even to optimality). Specifically,

- Considering Q1, we introduce a generalization of explicit abstractions (such as pattern databases (PDB) and merge-and-shrink) to what we call implicit absractions. The basic idea is in abstracting the problem in hand into provably tractable fragments of optimal planning, alleviating by that the constraint of PDBs to use projections of only low dimensionality. We then introduce concrete instance of this framework called fork-decomposition, and show both formally and empirically that the admissible heuristics induced by the latter abstractions provide state-of-the-art worst-case informativeness guarantees on several standard domains.

- Considering Q2, we describe a procedure that takes a classical planning task, a forward-search state, and a set of abstraction-based admissible heuristics, and derives an optimal additive composition of these heuristics with respect to the given state. Most importantly, we show that this procedure is polynomial-time for arbitrary sets of all known to us abstraction-based heuristics such as PDBs, constrained PDBs, merge-and-shrink abstractions, fork-decomposition implicit abstractions, and implicit abstractions based on tractable constraint optimization.

The talk is based on a joint work with Michael Katz (Technion).

The talk will cover the following: Introduction to Evolutionary Algorithms,Introduction to Informed / Heuristic Search, Generating Hand Crafted Heuristics and My research theories and results

It is well known that there cannot be a single ``best'' heuristic for optimal planning in general. One way of overcoming this is by combining admissible heuristics (e.g. by using their maximum), which requires computing numerous heuristic estimates at each state. However, there is a tradeoff between the time spent on computing these heuristic estimates for each state, and the time saved by reducing the number of expanded states. We present a novel method that reduces the cost of combining admissible heuristics for optimal search, while maintaining its benefits. Based on an idealized search space model, we formulate a decision rule for choosing the best heuristic to compute at each state. We then present an active online learning approach for that decision rule, and employ the learned model to decide which heuristic to compute at each state. We evaluate this technique empirically, and show that it substantially outperforms each of the individual heuristics that were used, as well as their regular maximum.

In recent years there is increasing interest in learning from both labeled and unlabeled data (a.k.a. semi-supervised learning, or SSL). The key assumption in SSL, under which an abundance of unlabeled data may help, is that there is some relation between the unknown response function to be learned and the marginal density of the predictor variables.

In the first part of this talk I'll present a statistical analysis of two popular graph based SSL algorithms: Laplacian regularization method and Laplacian eigenmaps.

In the second part I'll present a novel multiscale approach for SSL as well as supporting theory. Some intimate connections to harmonic analysis on abstract data sets will be discussed.

Joint work with Nati Srebro (TTI), Xueyuan Zhou (Chicago), Matan Gavish (WIS/Stanford) and Ronald Coifman (Yale).

We introduce pairwise cardinality networks, networks of comparators, derived from pairwise sorting networks, which express cardinality constraints. We show that pairwise cardinality networks are superior to the cardinality networks introduced in previous work which are derived from odd-even sorting networks. Our presentation identifies the precise relationship between odd-even and pairwise sorting networks. This relationship also clarifies why pairwise sorting networks have significantly better propagation properties for the application of cardinality constraints.

The reconstruction of phylogenetic trees (evolutionary trees) is central to many problems in Biology. Accurate reconstruction methods are currently limited to a maximum of few dozens of species. Therefore, in order to construct a tree over larger sets of species, a method capable of inferring accurately trees over small, overlapping sets, and subsequently merging these sets into a tree over the complete set, is required. A "quartet tree" (a non-rooted phylogenetic tree over four species) is the smallest informative piece of information, and the extensively-studied "quartet approach" is based on combining quartet trees into a big tree. However, even this case is NP-hard, and remains so even when the set of quartet trees is compatible (agree on a certain tree). The general problem of approximating quartets is known as the "maximum quartet consistency" (MQC), even for compatible inputs, is open for nearly twenty years. Despite its importance, the only rigorous results for approximating quartets are the naive $1/3$ approximation that applies to the general case and a PTAS when the input is the complete set of all ${n choose 4}$ possible quartets. Even when it is possible to determine the correct quartet induced by every four species, the time needed to generate the complete set of all quartets may be impractical. A faster approach is to sample at random just $m << {n choose 4}$ quartets, and provide this sample as an input. We present the first approximation algorithm whose guaranteed approximation is strictly better than $1/3$ when the input is any random sample of $m$ compatible quartets. The approximation ratio we obtain is close to $1/2$, and experimental results suggest that the ratio is much larger. An important ingredient in our algorithm involves solving a weighted MaxCut in a certain graph induced by the set of input quartets. We also show an extension of the PTAS algorithm to handle dense, rather than complete, inputs. Joint work with Raphy Yuster.

The distributed (Delta + 1)-coloring problem is a fundamental problem in distributed computing, which is motivated by various tasks such as symmetry breaking, channel allocation in wireless networks, and more. The network is modeled by a graph G = (V,E), where vertices represent processors, and edges represent communication links. The goal of legal vertex coloring is assigning colors to vertices, such that each two connected vertices are assigned distinct colors. The number of colors used is at most (Delta + 1), where (Delta) is the maximum degree in G. The distributed (Delta + 1)-coloring problem was intensively studied starting from mid-eighties. In most of the algorithms, a common technique is used, called “iterative color reduction”. In 1993, Szegedy and Vishwanathan came up with a heuristic argument stating that any algorithm that uses the iterative color reduction technique is unlikely to achieve smaller running time than Omega(Delta log Delta). In 2006, Kuhn and Wattenhoffer presented an algorithm with running time O(Delta log Delta + log^*n). This algorithm was the best known prior to our work.

We improve this result, and break the heuristic barrier of Szegedy and Vishwanathan by devising a (Delta + 1)-coloring algorithm with running time O(Delta) + 1/2 log^* n. Our algorithm employs a novel technique in distributed coloring that is apparently stronger than the iterative coloring reduction technique. In this technique, we efficiently construct colorings with relaxed requirement that each vertex has at most m neighbors colored by the same color. Such coloring is called an m-defective coloring. We show how to construct such coloring very efficiently for certain range of values of m. This construction can be then transformed into a legal (Delta + 1)-coloring. In addition this technique yields a tradeoff between the number of colors and the running time, resulting in O(Delta * t)-coloring with running time O(Delta/t + log^* n). Since many problems use the coloring problem as a block box, our results imply better solutions for these problems. In particular, our results yield an O(Delta) + 1/2 log^* n time algorithm for the problem of Maximal Independent Set, which improves previous results on bounded degree graphs. Appeared in STOC 2009. A joint work with Michael Elkin.

Segmentation of organs and vascular structures from clinical Computed Tomography (CT) images is a crucial task in many clinical applications including diagnosis, patient specific training simulations, and intra-operative navigation. The segmentation is a challenging task due to the unclear distinction between the required structure and its surrounding tissue, artifacts in the CT images, and the presence of pathologies. We present a shape constrained graph min-cut approach for the segmentation. Discrete energy functions are defined with respect to fixed or latent shape models and optimized using the graph min-cut algorithm to obtain an accurate segmentation. Extensive evaluation of our approach for different tasks, including carotid artery bifurcation segmentation, patient specific modeling of the entire carotid arteries system for simulation, abdominal aortic aneurysms segmentation, and kidney and liver segmentation shows that our method is accurate, robust, and practical for clinical use.

1) Introduction to the setting of private mechanisms. 2) Definition of private algorithm. 3) Example of a technique for getting private mechanisms. 4) Definition of privacy breach. 5) Known results of the field. 6) My attack on privacy preserving mechanisms. 7) Some lower bounds on attacks.(if time allows)

Since their introduction in 1987 by Haussler and Welzl, Epsilon-nets have become one of the central concepts in computational and combinatorial geometry, and have been used in a variety of applications, such as range searching, geometric partitions, and bounds on curve-point incidences. In particular, they are strongly related to geometric set-cover and hitting-set problems. A range space (or a hypergraph) (X,R) is a pair consisting of an underlying universe X of objects, and a certain collection R of subsets of X (also called ranges). Given a range space (X,R), and a parameter 0 < Epsilon < 1, an Epsilon-net for (X,R) is a subset N of X with the property that any range that captures at least Epsilon-fraction of X contains an element of N. In other words, N is a hitting set for all the ``heavy'' ranges. Of particular interest are geometric range spaces, since then they admit small-size Epsilon-nets. Specifically, the Epsilon-Net Theorem of Haussler and Welzl asserts that in this case there exists an Epsilon-net of size O(1/Epsilon log{1/Epsilon}). One of the major questions in the theory of Epsilon-nets, open since their introduction more than 20 years ago, is whether the factor log{1/Epsilon} in the upper bound on their size is really necessary, especially in typical low-dimensional geometric situations. A central motivation then arises from the technique of Bronnimann and Goodrich to obtain, in polynomial time, improved approximation factors for the geometric hitting-set and set-cover problems: The existence of an Epsilon-net of size O((1/Epsilon)f(1/Epsilon)), for some slowly-growing function f(.), implies an approximation factor of O(f(OPT)), where OPT is the size of the smallest such set. In this talk I will survey some of the fundamental results concerning small-size Epsilon-nets. I will then discuss range spaces of points and axis-parallel boxes in two and three dimensions, and show that they admit an Epsilon-net of size O(1/Epsilon loglog{1/Epsilon}). Joint Work with Boris Aronov (Polytechnic Institute of NYU),and Micha Sharir (Tel Aviv University).

Decoding of binary LDPC codes using linear-programming (LP) decoder was proposed by J. Feldman et al. The connections between LP decoding and classical message-passing decoding were established in that paper. In our work, we extend the above approach to nonbinary coded modulations, in particular to codes over rings mapped to nonbinary modulation signals. In both cases, the principal advantage of the linear-programming framework is its mathematical tractability. We define two alternative polytope representations, which offer a smaller number of variables and constraints for many classes of nonbinary codes. These polytope representations, when used with the respective nonbinary LP problems, lead to polynomial-time decoders for a wide variety of classical nonbinary codes. Finally, we present an LP decoder for nonbinary expander codes. We show that that this decoder corrects any pattern of errors of a relative weight up to approximately $1/4 delta_A delta_B$ (where $delta_A$ and $delta_B$ are the relative minimum distances of the constituent codes). Parts of this work are joint work with Mark F. Flanagan, Eimear Byrne and Marcus Greferath.

Images taken through a medium may suffer from poor visibility and loss > of contrast. Light passing through undergoes absorption and > scattering. Wavelength dependent attenuation causes changes in color > and brightness. In addition, light that is scattered back from the > medium into the camera (backscatter) veils the object, degrading > visibility and contrast. Low signal to noise ratio imposes resolution > limits, even if there is no blur. Moreover, refraction between the > medium and the camera (in air) causes geometric distortions that harm > geometric reconstruction. Nevertheless, there is a strong need to > perform vision tasks in such media. Thus, in this work, we look both > at photometric and geometrical aspects of imaging in these conditions. > > In the talk I give an overview of our contributions in this subject: > - Resolution limits imposed by noise > - Geometry limits: The non-single viewpoint nature of imaging systems > looking into water through a flat glass. > - Polarization-based removal of backscatter. > All the above is demonstrated in field experiments underwater and in haze. > > * A PhD research under the supervision of Prof. Yoav Y. Schechner > > >

Probabilistic models are ever growing in richness and diversity. Probabilistic programming languages have the potential to make representing, reasoning about, and learning models easier by allowing them to be represented using the power of programming languages, and providing general reasoning and learning algorithms. To this point, most probabilistic programming research has focused on the power of the languages and not on usability. This paper presents Figaro, a probabilistic programming language that addresses usability without sacrificing power. Figaro uses an object-functional style to achieve four goals. First, it is implemented as an extensible library that can be used by Java programs. Second, it can conveniently represent directed and undirected models with arbitrary constraints. Third, it can naturally represent models with interacting objects Fourth, it provides for declarative algorithm specification. Declarative algorithm specification means that algorithms are implemented as a library, and model class designers declare which algorithms they support and what they require of related models in order to support those algorithms. Given these specifications, the system automatically determines which algorithms can be applied to a given model.

Many businesses offer their services to customers via Web-based application interfaces. Reasoning about execution flows of such applications is extremely valuable for companies: it can be used to optimize business processes, employ targeted advertisements, reduce operational costs, and ultimately increase competitiveness. Such reasoning often operates in an environment inducing partial information and uncertainty of various flavors. First, the execution traces recorded for a Web application often contain only partial information on the activities that were performed at run-time, due to confidentiality, lack of storage space, etc. Second, even in the presence of fully detailed traces of the past executions, prediction of future executions may still operate under terms of uncertainty. This is because executions often depend on unknown external parameters, such as users behavior, interaction with other applications, servers response time, etc.

In this talk I will consider (1) models for capturing Web applications and their executions. These models are expressive enough to capture common scenarios, while restrictive enough to allow for efficient query evaluation; (2) query evaluation algorithms over applications/execution traces under these models, and (3) practical implementations, e.g. for recommending navigation flows within web applications.

In software development today there is a tension between the advantages of model-based methodologies and the need for rapid development methodologies, such as agile development methodologies (TDD, XP and Scrum, to name a few). Our work aim at synthesizing those approaches by providing lightweight tools for fleshing out a model from existing code. In this talk we present OPM/PL, a suite of modeling tools for the Object-Process Methodology (OPM), implemented in Prolog. We discuss the gap between model-based and agile software development methodologies, and show how OPM/PL bridges the gap. Finally, we show an example of using OPM/PL for round-trip modeling: Import information from code, link it to a model, generate new code from the model, and import information from the new code.

A democratic election is a classic example of a task in which multiple adversarial parties must collaborate and agree on an outcome. Traditional election systems (such as the one used in Israel) employ various means to ensure that the result will be accurate even if some of the people involved are corrupt or dishonest. However, the final tally is only as trustworthy as the people who count the votes. Even in the most secure systems these are usually fairly small committees. If an entire committee colludes, they can manufacture their own results. Even worse, depending on the exact setup, it may be feasible to stuff ballot boxes, destroy votes or perform other manipulations.

Using cryptographic techniques, it is possible to design a fair voting system whose correct operation can be verified by anyone, while still retaining ballot secrecy. This can be done even if the computers used to run the election are untrustworthy.

In the talk, I will briefly survey the techniques used to accomplish this and present in more detail examples with some unique properties, such as a practical solution for securely tallying Single Transferable Vote elections (a ranked voting system that is used in Australia, Ireland and Malta, among others). The talk will contain any necessary cryptographic background.

If time permits, I will also give a brief introduction to the Qilin project, a Java SDK for rapid prototyping of cryptographic protocols. The purpose of the Qilin project is to make it easier to build practical implementations of new cryptographic protocols, such as those for cryptographic voting. To this end, the API attempts to use the concepts and language from the theory of cryptography. The SDK is open-sourced and available on the web.

Based on joint works with Moni Naor and with Josh Benaloh, Lee Naish, Kim Ramchen and Vanessa Teague.

The Multiscale method is a class of algorithmic techniques for solving efficiently and effectively large-scale computational and optimization problems. This method was originally invented for solving elliptic partial differential equations and up to now it represents the most effective class of numerical algorithms for them. During the last two decades, there were many successful attempts to adapt the multiscale method for combinatorial optimization problems. Whereas the variety of continuous systems' multiscale algorithms turned into a separate field of applied mathematics, for combinatorial optimization problems they still have not reached an advanced stage of development, consisting in practice of a very limited number of multiscale techniques. In the first part of this talk we formulate the principles of designing multiscale algorithms for combinatorial optimization problems defined on a simple graph model. We present the results for the k-partitioning and a variety of linear ordering functionals (minimum linear arrangement/2-sum/bandwidth/workbound/wavefront sum). Since our algorithms were developed for practical purposes we compared them to many different heuristics: Spectral Sequencing, Optimally Oriented Decomposition Tree, Multilevel based, Simulated Annealing, Genetic Hillclimbing and other (including their combinations). In almost all cases we observed significant improvement of previous state-of-the-art results. We will cover a recently introduced measure of graph connectivity which can be important component of multilevel algorithms and other optimization problems on graphs.

If time permits we will present a multiscale coarsening scheme for minimizing a quadratic objective functional under planar inequality constraints. The scheme is demonstrated on a graph drawing problem in which the economical space utilization demand is evolved over the desired area rather than the widely used force-directed method, which preserves the non-overlapping property of the graph vertices. The non-overlapping property is automatically almost preserved as a result of equidensity constraints defined over the entire area. This demonstrates an ability of the algorithm to be used for solving a famous VLSI placement problem.

I will begin with presenting some problems from computational and combinatorial geometry. After that I will discuss problems of the following type: Is there a constant $f=f(k)$ such that any finite set of points in the plane can be colored with $k$ colors so that any halfplane that contains at least $f$ points, also contains a point from every color class? Similarly, one can reformulate the problem by changing halfplanes to a different family of regions. For halfplanes, Pach and G. Toth proved that $f(k)=O(k^2)$. This bound was later improved by Aloupis et al. to $f(k)=O(k)$. We will see that $f(k)=2k-1$, thus completely solving this question for the case of halfplanes. The above questions are related to problems of battery consumption in sensor networks and some other fields in computational geometry.

Whether or not parallel repetition improves security, is a fundamental question in the study of protocols. While parallel repetition improves the security of 3-message protocols and of public-coin protocols, Bellare, Impagliazzo and Naor (FOCS '97) gave an example of a protocol for which parallel repetition does not improve the security at all. We show that by slightly modifying any protocol, in a way that preserves its other properties, we get a protocol for which parallel repetition does improve the security (to any degree) .

In the second part of the talk (if time permits), I will presents our recent results on basing cryptography on minimal hardness assumptions, where we give simpler and more efficient (in some cases tight) constructions of pseudorandom generators, statistically hiding commitments and universal one-way hash functions based on one-way functions.

In this talk I will present several problems I have been working on in geometric modeling, computational geometry and computer graphics. The first problem is the construction, under the exact computation paradigm, of arrangements of Bezier curves. The second is the computation of Voronoi cells of free-form curves and the third is the visualization of solid models using the graphics processing unit (GPU). The common theme in these problems is that they contain geometric constructions, which either cannot be represented using their standard geometric representation or computing them is too expensive. Previous methods for attacking these problems typically use approximations, either of the input or of the problematic geometric constructions. Our methods, on the other hand, use procedural representations, which enable to answer a set of queries that are sufficient for solving the problem at hand. In arrangements of Bezier curves, we represent intersection vertices with references to intersecting curves, and to bounding polygons. This enables us to avoid the prohibitive running times incurred by exact algebraic arithmetic. In the computation of Voronoi cells of free-form curves, the bisector curves cannot be represented in standard (Bezier or B-spline) form. Instead we use a representation based on an implicit function in the curves parametric domain combined with a mapping to the Euclidean plane. Using this representation we can answer the queries required to compute the lower envelope of the bisector distance functions and thus compute the boundary of the Voronoi cell. When rendering solid models using the GPU, a common problem is the appearance of cracks between faces in the model visualization. These are a result of the non-exact representation of trimming curves in the model. Using a representation that stores references to intersecting surfaces we are able to avoid these cracks and render a smooth water tight model. This work is part of an ongoing project. In my talk, I will present the different problems and how applying procedural representations helps in their computation. I will also present other problems for which I believe applying such representations can be useful. The work described in this talk was done in collaboration with Gershon Elber, Ron Wein, Kirk Haller and others.

Over the past decades the evolutionary approach has been used in many fields of computer science research. Lately, with the growth of computation power, Genetic Programming (GP) has been showing much promise. We present an attempt to apply the tree based GP approach to zero-sum deterministic full knowledge board games, using Lose Checkers as a test-case. Our system implements strongly typed GP trees, explicitly defined introns and multi-tree individuals. We use the GP trees to evaluate possible future game states. Used together with traditional search techniques the results show much promise and imply that tree based GP may be useful in finding good players for other similar games.

How much computational power is required to perform basic cryptographic tasks? We will survey a number of recent results on the complexity of basic cryptographic primitives such as one-way functions, pseudorandom generators, encryption schemes and digital signatures. Specifically, we will consider the possibility of computing instances of these primitives by NC0 functions, in which each output bit depends on only a constant number of input bits. Somewhat surprisingly and unintuitively, it turns out that most cryptographic tasks can be carried out by such simple functions. We will also explore the cryptographic strength of several interesting subclasses of NC0. This includes simple forms of natural computation that can be performed by real-world dynamical systems in a constant number of steps. On the application side, we will mention new connections between NC0 cryptography and other seemingly unrelated areas such as secure distributed computation, program checking, and hardness of approximation. We will focus on a new approach for constructing public-key encryption schemes based on the intractability of random NC0 functions. Most, if not all, existing constructions of public-key encryption use hardness assumptions with significant algebraic structure. The main advantage of the new schemes is the relatively general and combinatorial nature of the new assumptions, which seem qualitatively different than previous ones. Based on joint works with Yuval Ishai and Eyal Kushilevitz and with Boaz Barak and Avi Wigderson. The talk is self-contained, and does not assume previous background in cryptography.

Digitization of health care data is creating an opportunity for a new way of studying diseases and improving medical care. Using the vast amount of patient data collected daily at hospitals can assist us in circumventing the inherent faults of current medical research of studying phenomenon using various imperfect models, as experimenting on humans is unethical and illegal. The majority of information is stored in free text written by doctors. Using that data requires adapted Natural Language Processing methods combined with domain specific knowledge. Here I will present one project that originated from challenges in Medical Hebrew Processing: In most professional domains of languages with non-Latin alphabet, proper names, named entities and adjectives are transliterated from English. We show that recognizing these words as well as the original word is important for term recognition. We developed a method for identifying said words combining unsupervised classifiers and a lexicon. The lexicon based approach produced F-Measures of 87%-92% across domains, the combined approach produced F-Measures of 93%-94% respectively. Using this classifier to improve term matching we obtained 77% more matches with precision of 92%.

The images acquired via medical imaging modalities are frequently subject to low signal-to-noise ratio, bias field and partial volume effects. These artifacts, together with the naturally low contrast between image intensities of some neighboring structures, make the extraction of regions of interest (ROIs) in clinical images a challenging problem. Probabilistic atlases, typically generated from comprehensive sets of manually labeled examples, facilitate the analysis by providing statistical priors for tissue classification and structure segmentation. However, the limited availability of training examples that are compatible with the images to be segmented, renders the atlas-based approaches impractical in many cases. In the talk I will present a generative model for joint segmentation of corresponding regions of interest in a collection of aligned images that does not require labeled training data. Instead, the evolving segmentation of the entire image set supports each of the individual segmentations. This is made possible by iteratively inferring a subset of the model parameters, called the spatial parameters, as part of the joint segmentation processes. These spatial parameters are defined in the image domain and can be viewed as a latent atlas, that is used as a spatial prior on the tissue labels. Our latent atlas formulation is based on probabilistic principles, but we solve it using partial differential equations (PDEs) and energy minimization criteria. We evaluate the method successfully for the segmentation of cortical and subcortical structures within different populations and of brain tumors in a single-subject multi-modal longitudinal experiment.

In experimental optimization the quality of candidate solutions can be evaluated only by means of an experiment in the real-world. These experiments are often time-consuming and/or expensive, and are typically limited to several dozens or hundreds of trials. High-dimensional problems (i.e., at least 80 search variables) cannot be efficiently handled by classical convex optimizers, and thus require an alternative treatment. Derandomized Evolution Strategies (DES) are powerful bio-inspired search methods, originating from Evolutionary Algorithms, that incorporate statistical learning for efficient derandomized search. This talk will focus on the theory behind state-of-the-art DES, as well as on their application to experimental optimization. Especially, it will discuss optimization efficiency, attainment of robust solutions, exploration of the actual search landscape, and the generalization into Pareto optimization of multiple objectives. Special emphasis will be put on a particular experimental platform employing DES at present times, namely Quantum Control experiments. The Quantum Control (QC) field aims at altering the course of quantum dynamics phenomena for specific target realizations, by means of closed-loop, adaptively learned laser pulses. The optimization task of QC experiments typically poses many algorithmic challenges, e.g., high-dimensionality, noise, constraints handling, and thus offers a rich domain for the development and application of specialized optimizers. Toward that end, the computational aspects of several real-world laboratory optimization case-studies will be presented. **This talk will be self-contained, and will target the general audience of CS, Engineering, and Applied Physics. It will not require any specialized background in Quantum Mechanics nor in Optimization.

Today's computers run numerous processes of different sensitivity and trustworthiness, and often the only boundary between a hostile network and sensitive data relies on flimsy confinement assumptions. The platform purports to protect processes from each other, but side channels arise from lower architectural layers, such as contention for shared hardware resources, and create inadvertent cross-talk. For example, we have shown how observing contention for the CPU cache allows an attacker to steal other users' encryption keys in a few milliseconds. Such cross-talk is especially grievous in the context of cloud computing ("infrastructure as a service"), where users acquire computational capacity in the form of virtual machines running on a service provider's shared hardware pool. The presence of multiple mutually-untrusting virtual machines on the same hardware creates the risk of information exfiltration across virtual machines and between clients, as we demonstrated on Amazon EC2. These security vulnerabilities raise the challenge of achieving trustworthy computation on leaky platforms. We discuss potential solutions, including a new work on mitigating side channels using just-in-time dynamic transformation of x86 machine code.

This talk includes joint works with Saman Amarasinghe, Dag Arne Osvik, Thomas Ristenpart, Ron Rivest, Stephan Savage, Hovav Shacham, Adi Shamir and Qin Zhao.

Regardless of the technology that surrounds us and the virtual realms we use and inhabit, we are still physical beings, living and acting in physical spaces. Various current trends in human-computer interaction attempt to enhance usability by exploring new physical interfaces attributes and mappings.

My group's research is directed at designing and exploring new interactive experiences based on physical entities and environments. Our interests include human-robot interaction, physical and tangible user interfaces, mixed reality, computer game interfaces and ubiquitous computing. In this talk I will present a brief overview of physicality in interaction, its potential and challenges. The talk includes a discussion of several current projects my students and I are pursuing, ranging from 3D sketch-based interaction and mixed reality games, to sociable robotic interfaces and robotic cartoon art expression.

Many applications call for estimating probabiities of rare, even previously unseen, events. We briefly describe the problem's theory, applications to classification and data compression, relation to works by Fisher, Shakespeare, Laplace, Good, Turing, Hardy, Ramanujan, and Shannon, and recent constructions of asymptotically optimal estimators. The talk is self contained and based on work with P. Santhanam, K. Viswanathan, J. Zhang, and others.

We will discuss complexity in worlds with a weak algebraic structure. We will start with a brief introduction to algebraic complexity, and explain Valiant's definitions of the algebraic analogs of P and NP. We will then explore these notions in weak algebraic structures which are not necessarily commutative or even associative. It turns out that even in such a world, permanent is NP-complete. We also consider hardness in these weak computational worlds: First, we show that the non-commutative complexity of permanent is related to the classical sum-of-squares problem. We also show that in a non-associative world, NP is strictly stronger than P. Joint work with Pavel Hrubes and Avi Wigderson.

The k-server problem is one of the most central and well studied problems in competitive analysis and is considered by many to be the "holy grail" problem in the field. In the k-server problem, there is a distance function d defined over an n-point metric space and k servers located at the points of the metric space. At each time step, an online algorithm is given a request at one of the points of the metric space, and it is served by moving a server to the requested point. The goal of an online algorithm is to minimize the total sum of the distances traveled by the servers so as to serve a given sequence of requests. The k-server problem captures many online scenarios, and in particular the widely studied paging problem. A twenty year old conjecture states that there exists a k-competitive online algorithm for any metric space. The randomized k-server conjecture states that there exists a randomized O(log k)-competitive algorithm for any metric space. While major progress was made in the past 20 years on the deterministic conjecture, only little is known about the randomized conjecture. We present a very promising primal-dual approach for the design and analysis of online algorithms. We survey recent progress towards settling the k-server conjecture achieved using this new framework.

In this talk I'll exhibit a simple method for analyzing dynamic data structures using various forbidden substructure theorems. The idea is to (1) transcribe the behavior of the data structure as some kind of combinatorial object, (2) show that the object does not contain some forbidden substructure, and (3) bound the size of any such object without this substructure. I'll show how to analyze path compression-based data structurses using forbidden 0/1 submatrices, then discuss some extremal problems on 0-1 matrices. This talk covers results from two papers to appear in SODA 2010.

In 1951, John Nash proved that every game has an equilibrium, i.e. a set of strategies such that no player has an incentive to change her strategy. Nash's proof is non-constructive in nature, since it relies on Brouwer's fixed point theorem. This raises the following question - given a game, what is the computational complexity of finding a Nash equilibrium?

In this talk I will give an introduction to the area of Computing Equilibria in Algorithmic Game Theory and overview a breakthrough result suggesting that the problem of finding a Nash equilibrium is hard – Daskalakis et al. prove it is complete for the complexity class PPAD. The proof is based on a reduction from fixed point problems to finding equilibria, using small “gadget games” as means of performing arithmetic computations.

No prior knowledge is assumed.

References: C. Daskalakis, P. W. Goldberg and C. H. Papadimitriou. - The Complexity of Computing a Nash Equilibrium, SIAM Journal on Computing, 1:195-259, 2009. K. Etessami and M. Yannakakis. On the Complexity of Nash Equilibria and Other Fixed Points, - In the 48th Annual IEEE Symposium on Foundations of Computer Science, FOCS 2007.

Most 3D shapes are nowadays acquired using range scanning devices.Currently, scanners are capable of capturing complex shapes, large urban scenes and lately even motion. The initial representation of the shape consists of several properly transformed depth images, resulting in a point sampling of the surface. Typically, scan data consist of missing parts, noise in point coordinates and orientation, outliers and non-uniform sampled regions. Without prior assumptions and user interventions, the reconstruction problem is ill posed; an infinite number of surfaces pass through or near the data points. One of today's principal challenges is the development of robust point processing and reconstruction techniques that deal with the inherent inconsistencies in the acquired data set. In my talk I will present recent advances in geometric modeling, processing and reconstruction of point data. I will describe a deformable model for watertight manifold reconstruction. The model yields a correct topology interpretation of the reconstructed shape and allows topology control to a certain extent. Next, I will present a topology-aware interactive reconstruction technique. Topological ambiguities in the data are automatically detected and user interaction is used to consolidate topology reconstruction. Following, I will present a space-time technique for the reconstruction of moving and deforming objects. The motion of the object is described as an incompressible flow of matter which overcomes deficiencies in the acquired data such as persistent occlusions, errors and even entirely missing frames. Motivated by recent advancements in sparse signal reconstruction, I will present a "lower-than-L2" minimization scheme for sparse reconstruction. The sparsity principle gives rise to a novel global reconstruction paradigm for sharp point set surfaces which is robust to noise.

Two of most serious and fundamental problems Natural Language Processing are data sparsity and non-local reasoning. For example, let's consider the task of identifying people, locations and organizations in the following text (taken from Reuters): "BLINKER BAN LIFTED. Dutch forward Reggie Blinker had his indefinite suspension lifted by FIFA on Friday and was set to make his Sheffield Wednesday comeback on Saturday… Blinker missed his club's last two games after FIFA slapped a worldwide ban on him for appearing to sign contracts for both Wednesday and Udinese …" Many of the words, like 'Udinese' and 'Sheffield' are rare words that are unlikely to appear in the training data. On the other hand, the words 'Blinker' and 'Wednesday' in this text refer to a player and to a soccer team, and successfully identifying them as such requires global understanding of the text. In this talk I will discuss algorithms for reducing data sparsity and making non-local decisions in NLP. the running example will be the task of named entity recognition. Bio. Lev Ratinov received his BSc and MSc from BGU, now he's a Phd student in University of Illinois at Urbana-Champaign. He has published at ACL, AAAI, and gave a tutorial at EACL

This talk attempts to glimpse into the field of Computational Biology, specifically through the topic of constrained sequence alignment and its applications. Introducing constraints into alignment processes facilitates improved speed and allows for fine-tuning of the alignment results. Constraints are usually a formulation of a priory knowledge regarding the expected alignment result and often originate from biological studies. We survey different types of constraints, the central approaches to constrained sequence alignment and discuss several algorithms and their applications.

A secret-sharing scheme is a mechanism for sharing data among a set of participants such that only pre-defined authorized subsets of participants can reconstruct the data, while any other subset has absolutely no information on the data. The talk will have 2 parts. In the first part, I will give some background on secret sharing schemes and introduce several simple schemes. In addition, I will discuss the connection between secret sharing and information theory. (This part is really simple) In the second part I will briefly show some results from my thesis. Specifically, the result from the paper Secret Sharing and Non-Shannon Information Inequalities  Amos Beimel and Ilan Orlov [TCC09]

the abstract of this paper follows The known secret-sharing schemes for most access structures are not efficient; even for a one-bit secret the length of the shares in the schemes is $2^{O(n)}$, where $n$ is the number of participants in the access structure. It is a long standing open problem to improve these schemes or prove that they cannot be improved. The best known lower bound is by Csirmaz (J. Cryptology 97), who proved that there exist access structures with $n$ participants such that the size of the share of at least one party is $n/ log n$ times the secret size. Csirmaz's proof uses Shannon information inequalities, which were the only information inequalities known when Csirmaz published his result. On the negative side, Csirmaz proved that by only using Shannon information inequalities one cannot prove a lower bound of $omega(n)$ on the share size. In the last decade, a sequence of non- Shannon information inequalities were discovered. This raises the hope that these inequalities can help in improving the lower bounds beyond $n$. However, in this paper we show that all the inequalities known to date cannot prove a lower bound of $omega(n)$ on the share size.

Consider the setting of emph{randomly weighted graphs}, namely, graphs whose edge weights are chosen independently according to probability distributions with finite support over the non-negative reals. Under this setting, weighted graph properties typically become random variables and we are interested in computing their statistical features. Unfortunately, this turns out to be computationally hard for some weighted graph properties albeit the problem of computing the properties per se in the traditional setting of algorithmic graph theory is tractable. For example, there are well known efficient algorithms that compute the emph{diameter} of a given weighted graph, yet, computing the emph{expected} diameter of a given randomly weighted graph is SharpP{}-hard even if the edge weights are identically distributed. In this work, we define a family of weighted graph properties and show that for each property in this family, the problem of computing the $k^{th}$ moment (and in particular, the expected value) of the corresponding random variable in a given randomly weighted graph $G$ admits a emph{fully polynomial time randomized approximation scheme (FPRAS)} for every fixed $k$. This family includes fundamental weighted graph properties such as the diameter of $G$, the emph{radius} of $G$ (with respect to any designated vertex) and the weight of a emph{minimum spanning tree} of $G$. Joint work with Amos Korman and Yuval Shavitt.

Phylogenetic trees represent evolutionary history, by showing the course of evolution from some extinct common ancestor to today's species. These trees can reveal connections between different species, teach us about extinct species, and help us understand the development of viruses, and other organisms with high mutation rate. The most common way of reconstructing phylogenetic trees is based on sequencing DNA from different species, and using similarities between sequences to infer the tree. This requires some model of how DNA evolves between an ancestor species and its descendants. The simplest model assumes that there are i.i.d mutations, and that each mutation is a substitution, and under this model, there are provably good reconstruction algorithm. However, recent studies show that insertions and deletions mutations are a serious cause of reconstruction errors, and can not be ignored. We present the first efficient algorithm for tree reconstruction when the mutations can be substitutions, insertions and deletions. The algorithm uses a clustering based approach, which builds small parts of the tree, and glues them together. In addition, the algorithm outputs a global alignment of the DNA sequences, which respects the evolutionary history. Based on joint works with Alex Andoni, Mark Braverman, Costis Daskalakis and Sebasiten Roch.

Modern software engineering heavily relies on clearly specified interfaces for separation of concerns among designers implementing components and programmers using those components. The interface of a component describes the functionality and constraints on the correct usage in a succinct manner. The need for interfaces is evident for assembling complex systems from components, but more so in modern control applications where a network of sensors and actuators is deployed to achieve complex control objectives. The notion of an interface for a control component must incorporate information about timing and resource needs, information that is not represented in current programming interfaces.

In the talk, I'll propose formal languages as an interface for resource sharing and timing. For example, imagine two components that share some resource (e.g. CPU), allocated in a time-triggered manner. With the proposed interface, each component specifies a formal language over the alphabet ${0,1}$ of ``safe schedules", i.e., if $w=w(1),w(2),…$ is in this languages, and the component gets the resource in every time slot t for which $w(t)=1$, then the component meets its performance requirements. The main advantage of this interface is that it allows modular analysis of each component and, then, constraints can be combined using languages intersection.

I'll present theoretical results concerning applications of the above approach to software control systems (where controllers are implemented by a software that shares resources). I'll also introduce a Real-Time-Java based tool, called RTComposer, that supports the proposed methodology. In RTComposer, scheduling specifications can be given as periodic tasks, or using temporal logic, or as omega-automata, or stability requirements. Components can be added dynamically, and non-real-time components are allowed. The benefits of the approach will be demonstrated and applications to wireless sensor/actuator networks based on the WirelessHART protocol and to distributed control systems based on the Control Area Network (CAN) bus (used, e.g., in automotive applications) will be discussed.

Distributed Constraint Satisfaction and Optimization problems provide a widely accepted framework for many multi agent tasks in AI and Operation Research. In this talk I will give a very brief introduction to Distributed Constraint Satisfaction Problems (DCSPs) and Distributed Constraint Optimization Problem (DCOPs) and mention some of the research being done in our DisCSP group. I will then move on to describe my work which involves the introduction of self interested agents into a cooperative framework. While most researchers of self interested agents apply game theoretic tools and ideas when dealing with such agents, in my work I attempt to examine the implication of selfishness on the inherently cooperative DCOP framework. My work revolves around three main aspects: the problem formulation, the notion of global optimality and the quality of a solution for each participant.

In the classical theory of social choice, a theory developed by game-theorists and theoretical economists, we consider a set of agents (voters) and a set of alternatives. Each agent ranks the alternatives, and the major aim is to find a good way to aggregate the individual preferences into a social preference. The major tool offered in this theory is the axiomatic approach: study properties (termed axioms) that characterize particular aggregation rules, and analyze whether particular desired properties can be simultaneously satisfied. In a ranking system the set of voters and the set of alternatives coincide, e.g. they are both the pages in the web; in this case the links among pages are interpreted as votes: pages that page p links to are preferable by page p to pages it does not link to; the problem of preference aggregation becomes the problem of page ranking. Trust systems are personalized ranking systems where the ranking is done for (and from the perspective of) each individual agent. Here the idea is to see how to rank agents from the perspective of a particular agent/user, based on the trust network generated by the votes. In a trust-based recommendation system the agents also express opinions about external topics, and a user who has not expressed an opinion should be recommended one based on the opinions of others and the trust network. Hence, we get a sequence of very interesting settings, extending upon classical social choice, where the axiomatic approach can be used.

On the practical side, ranking, reputation, recommendation, and trust systems have become essential ingredients of web-based multi-agent systems. These systems aggregate agents’ reviews of products and services, and of each other, into valuable information. Notable commercial examples include Amazon and E-Bay’s recommendation and reputation systems, Google’s page ranking system, and the Epinions web of trust/reputation system. Our work shows that an extremely powerful way for the study and design of such systems is the axiomatic approach, extending upon the classical theory of social choice. In this talk we discuss some representative results of our work.

Decision-making in the face of uncertainty over future outcomes is a fundamental machine learning problem, with roots in statistics and information theory, and applications to signal processing, network routing and finance. In this talk I'll describe recent algorithmic advances, both in terms of accuracy as well as computational efficiency. We describe the first efficient algorithm for the problem of online linear optimization in the limited-feedback (bandit) setting which achieves the optimal regret bound. This resolves an open question since the work of Awerbuch and Kleinberg in 2004, and is made possible via a new technique for controlling the exploration-exploitation tradeoff, inspired by convex optimization. Next we describe new online learning algorithms which attain optimal regret bounds in both worst case and stochastic scenarios. Tight performance bounds for decision making which interpolate between the worst-case and stochastic approaches were considered a fundamental open question. Based on work with Jacob Abetnethy, Satyen Kale and Alexander Rakhlin.

I will survey my lines of work in the fields of theoretic machine learning, automata learning, and statistical estimation. I will present some interesting recent results, and mention some open problems and topics I would like to look into further. The talk is intended to a relatively wide audience of CS graduate students, and not to machine learning professionals.

In numerous areas of science and engineering, we seek to extract information from linearly derived measurements in a computationally feasible manner. Linear measurement systems include medical imaging, financial prediction, seismic imaging in the oil industry, multiuser communication among cellphones, etc. Much of my discussion will focus on compressed sensing, an emerging area based on the revelation that optimization routines can reconstruct a sparse signal from a small number of linear projections of the signal - but the results are far more general. I will describe recent investigations into the ultimate performance limits that can be achieved in linear measurement systems, and describe an algorithm that achieves asymptotically optimal performance when the system is large and the measurement process can be modeled by a sparse matrix. An overview of potential future directions will be provided.

The visibility of personal computers hides the fact that most software development is done for systems in industry and business, and not for standalone software packages. Programmers and engineers in these environments are required to have a different set of skills and a different approach to software than are currently taught in most computer science (CS) curricula. Furthermore, the talk of a hi-tech, internet-driven revolution during the last decade is inaccurate from a historical perspective, and this loss of perspective has led to demands for an artifact-driven CS curriculum. A comparison of the ACM/IEEE CC2001 curriculum with the curriculum of a traditional engineering discipline points to what I believe the future of CS education should be. The talk will conclude with a survey of my work on teaching concurrent and distributed computation using model checking. I will show how this advanced technique can be presented to undergraduate and even high-school students.

Standard search techniques for detecting homology in DNA sequences start by detecting small matching parts, called seeds, between a query sequence and database sequences. Contiguous seed models (k-mers, k-tuples, etc.) have existed for many years and are used in programs like BLAST and BLAT. Newer models include spaced seeds and indel seeds. Both of these seed models have been shown to be more sensitive than contiguous seeds while maintaining similar specificity, where sensitivity measures the ability to find true homologies, and specificity measures the ability to avoid wasting computation time on false candidates for homology. The domains of application for the seed classes differ: spaced seeds are superior under alignment models which only allow matches and mismatches, indel seeds under models which also allow insertions and deletions in the alignments. For any value k, there is only one contiguous seed of length k, but there can be many, many spaced seeds and indel seeds. Optimal seed selection is a resource intensive activity because essentially all possible seed shapes must be tested. In this talk, I describe the various seed models, show how to efficiently compute optimal seeds, and discuss an application in the context of new technologies for genome sequencing, in particular, mapping of short sequencing reads to a reference genome.

We evolve heuristics to aid search algorithms. The main method applied is Genetic Programming (GP), a sub-class of evolutionary algorithms, in which the individuals undergoing artificial evolution are (simple, yet large) LISP programs. In the domain of games, where search spaces are enormous, and the number of possible heuristics is even larger, the advantages of GP come to the fore, as an alternative to manually fine-tuning static evaluation functions. First, this method is demonstrated in the domain of Chess - both for endgames and the Mate-in-N problem. We further analyze the performance of evolved players and show that they demonstrate some emergent features. Second, we move to single-player games (puzzles) and show that a slightly-modified version of this methodology (using a form of policies), produces strong players for the 6x6 and 8x8 Rush-Hour puzzle, a PSPACE-Complete problem (for the nxn case). Some broader conclusions are drawn for the interplay between search and knowledge.

UML is now widely accepted as the standard modeling language for software construction. The Class Diagram is its core view, having well formed semantics and providing the backbone for any modeling effort. Class diagrams are widely used for purposes such as software specification, database and ontology engineering, meta-modeling, and model transformation. The central role played by class diagrams emphasizes the need for strengthening UML modeling tools with features such as recognition of erroneous models and the detection of errors’ sources.

Correctness of UML class diagrams refers to the capability of a diagram to denote a finite but not empty reality. This is a natural, unquestionable requirement. Nevertheless, incorrect diagrams are often designed, due to interaction of contradicting constraints and limitations of current tools. In this talk, I will discuss correctness problems in class diagrams, the methods we have develop for tackling them, and the latest results. I will also suggest directions for future work in this field.

Agents operating in the real world need to handle both uncertainty and resource constraints. Typical problems in this framework are rational observation control (ROC) and optimal allocation of computation tasks during reasoning and search (also known as meta-reasoning). In both tasks, a crucial issue is value of information, a quantity hard to compute in general, and thus usually estimated using severe assumptions, such as myopic look-ahead, independence of information sources and separability of reward function.

Recently, a dynamic programming approach, which bypasses the myopic assumption, was introduced. An efficient method, based on this approach, constructs an optimal observation plan for a chain-shaped dependency model with exact measurements and additive reward function.

In this work we consider several extensions for the above method, that allow inexact and multiple measurements, various reward functions and more general dependency models. We developed a unifying approach to a wide class of ROC problems. To this end we extended the concept of a conditional performance profile, and developed an efficient technique for compiling a composite system beyond the input monotonicity assumption. The resulting framework can be applied to many real-world domains, such as medical diagnostics, setup optimization, mobile robot navigation and game-tree search.

We present an algorithm for curve skeleton extraction from imperfect point clouds where large portions of the data may be missing. Our construction is primarily based on a novel notion of generalized rotational symmetry axis (ROSA) of an oriented point set. Specifically, given a subset S of oriented points, we introduce a variational definition for an oriented point that is most rotationally symmetric with respect to S. Our formulation effectively utilizes normal information to compensate for the missing data and leads to robust curve skeleton computation over regions of a shape that are generally cylindrical. We present an iterative algorithm via planar cuts to compute the ROSA of a point cloud. This is complemented by special handling of non-cylindrical joint regions to obtain a centered, topologically clean, and complete 1D skeleton. We demonstrate that quality curve skeletons can be extracted from a variety of shapes captured by incomplete point clouds. Finally, we show how our algorithm assists in shape completion under these challenges by developing a skeleton-driven point cloud completion scheme. Joint work with *Andrea Tagliasacchi - School of Computing Science, Simon Fraser University *Hao Zhang - School of Computing Science, Simon Fraser University *Daniel Cohen-Or - School of Computer Science, Tel-Aviv University

The mitochondrial genome (mtDNA) is an exceptional model for bioinformatics. It is the only genome which was already completely sequenced in thousands of individuals. We developed a special method for comparative whole-genome analysis of mtDNA mutations in human evolution and in cancer, and show that distinct patterns recur in both. Some challenges of whole-genome analysis, especially with the ongoing “1000 genomes” data, will be demonstrated through this anlaysis.

Service-Oriented Architecture (SOA) is the new phase in the evolution of distributed ente rprise applications. It could enable the design and realization of flexible applications across multiple organizations. However, there are many security issues associated with SO A, e.g. trust establishment among actors in an inter-organizational context and its chan nels of communication are more vulnerable. Solutions include the production of numerous, often overlapping security standards by the industry (but there is no clear view of how t o use them), specialized security mechanisms and methodologies. We propose a methodology to design and build secure service-oriented applications. Our approach builds upon Model- Driven Engineering (MDE) and the use of security patterns. We show the methodology and so me of the related patterns we have produced.

We consider the following sequential decision problem. Given a set of items of unknown utility, we need to select one of as high a utility as possible (``the selection problem''). Measurements (possibly noisy) of item values prior to selection are allowed, at a known cost. The goal is to optimize the overall sequential decision process of measurements and selection. Value of information (VOI) is a well-known scheme for selecting measurements, but the intractability of the problem typically leads to using myopic VOI estimates. In the selection problem, myopic VOI frequently badly underestimates the value of information, leading to inferior sensing plans. We relax the strict myopic assumption into a scheme we term semi-myopic, providing a spectrum of methods that can improve the performance of sensing plans. In particular, we propose the efficiently computable method of ``blinkered'' VOI, and examine theoretical bounds for special cases. Empirical evaluation of ``blinkered'' VOI in the selection problem with normally distributed item values shows that is performs much better than pure myopic VOI.

Alternative splicing (AS) is an RNA processing mechanism creating protein diversity in higher eukaryotes. It is regulated by splicing factors that interact with their binding sites along exons and introns. One of the main challenges in the study of AS regulation is to accurately map splicing factor binding motifs on the RNA. Therefore we developed a new method for mapping binding sites of known splicing factors which considers both the genomic environment of a single binding site and the evolutionary conservation of these sequences. The method was successfully applied to map splicing factor binding sites on specific genes and also on a wide genomic scale. By applying the algorithm on a subset of splicing factors we constructed a splicing regulatory network. This network presented a hierarchical structure that correlated with the tissue specificity levels of the splicing factors. Further to study the relationship between splicing factors and transcription regulation we derived a transcription-splicing co-regulatory network, where the nodes of the network are the Splicing Factors and the Transcription Factors (genes/proteins) and the edges represent either splicing regulation or transcription regulation. The latter network demonstrated a high level regulation between proteins involved in the gene-expression pathway, involving both splicing and transcription regulation.

In 1992, X. Zha conjectured that the line graph of a 3-connected non-planar graph contains a subdivision of $K_5$. In this talk I will present recent progress on the conjecture. (This will be the same talk as given in the DCG seminar three weeks ago).

We present an automated approach to prove termination of Java Bytecode programs by automatically transforming them to term rewrite systems (TRSs). In this way, the numerous techniques and tools developed for TRS termination can also be used for imperative object-oriented programming languages that can be compiled into Java Bytecode. Compared to direct termination analysis of imperative programs, rewrite techniques have the advantage that they are very powerful for algorithms on user-defined data structures, since they can automatically generate suitable well-founded orders comparing arbitrary forms of terms. Moreover, by using term rewriting with built-in integers, rewrite techniques are also powerful for algorithms on pre-defined data structures like integers..

The talk will focus on three topics from my PhD thesis, which have applications also in other fields. The first concerns a clustering problem, in which the spatial data of objects are to be integrated with labels (e.g. features) of the objects. In the second the objects form the vertices of a graph and their labels are used to infer connections between communities of objects. The third is a new approach to statistical enrichment analysis, which takes into account degree of similarity between labels (instead of equality or non-equality). From a computational perspective the labels can be viewed (and will be presented in the talk) as colors, but the motivation for my resarch, as you all know, comes from attempts to improve the analysis of large scale data-sets of genes/gene-products.

In this work we study the list-decoding size of Reed-Muller codes. Given a received word and a distance parameter, we are interested in bounding the size of the list of Reed-Muller codewords that are within that distance from the received word. We provide asymptotic bounds for the list-decoding size of Reed-Muller codes that apply to all distances. Previous results of Gopalan, Klivans and Zuckerman apply to distances only up to the minimum distance of the code.

Additionally, we study the weight distribution of Reed-Muller codes. We provide bounds for the weight distribution of Reed-Muller codes that apply to all distances. Previous results by Azumi, Kasami and Tokura apply to distances only up to 2.5 times the minimum distance of the code. Joint work with Tali Kaufman.

The talk will begin with some intro to Evolutionary Algorithms and Genetic Programming, so no related background is required.

We peruse the idea of algorithmic design through Darwinian evolution, focusing on the problem of evolving list search algorithms. Specifically, we employ genetic programming (GP) to evolve iterative algorithms for searching for a given key in an array of integers. Our judicious design of an evolutionary language renders the evolution of linear-time search algorithms easy. We then turn to the far more difficult problem of logarithmic-time search, and show that our evolutionary system successfully handles this case. Subsequently, because our setup might be perceived as being geared towards the emergence of binary search, we generalize our genomic representation, allowing evolution to assemble its own useful functions via the mechanism of automatically defined functions (ADFs). We show that our approach routinely and repeatedly evolves general and correct efficient algorithms.

The leveraging of todays unprecedented capability to manipulate biological systems by state-of-the-art computational, mathematical and engineering techniques , may profoundly affect the way we approach the solution to pressing grand challenges such as the development of sustainable green energy, next generation healthcare, etc. The conceptual cornerstone of Synthetic Biology a field very much on its infancy- is that methodologies commonly used to design and construct non-biological artefacts (e.g. computer programs, airplanes, bridges, etc) might also be mastered to create designer living entities. Computational methods for modeling in Synthetic Biology consist of a list of instructions detailing an algorithm that can be executed and whose computation resembles the behavior of the biological system under study. This computational approach to modelling biological systems has been termed executable biology. In this talk I will describe current approaches for the automated generation and testing of executable biology models for synthetic biology.

The problem of carrying out cryptographic computations when the participating parties are rational in a game-theoretic sense has recently gained much attention. One problem that has been studied considerably is that of rational secret sharing. In this setting, the aim is to construct a mechanism (protocol) so that parties behaving rationally have incentive to cooperate and provide their shares in the reconstruction phase, even if each party prefers to be the only one to learn the secret. Although this question was only recently asked by Halpern and Teague (STOC 2004), a number of works with beautiful ideas have been presented to solve this problem. However, they all have the property that the protocols constructed need to know the actual utility values of the parties (or at least a bound on them). This assumption is very problematic because the utilities of parties are not public knowledge. We ask whether this dependence on the actual utility values is really necessary and prove that in the basic setting, rational secret sharing cannot be achieved without it. On the positive side, we show that by somewhat relaxing the standard assumptions on the utility functions, it is possible to achieve utility independence. In addition to the above, observe that the known protocols for rational secret sharing that do not assume simultaneous channels all suffer from the problem that one of the parties can cause the others to output an incorrect value. (This problem arises when a party gains higher utility by having another output an incorrect value than by learning the secret itself; we argue that such a scenario is not at all unlikely.) We show that this problem is inherent in the non-simultaneous channels model, unless the actual values of the parties' utilities for this attack is known, in which case it is possible to prevent this from happening.

This is a joint work with Yehuda Lindell.

The prevalence of large racks of interconnected processor nodes forces us to take another look at how to exploit parallelism when taking the join of large relations. Sometimes, there is a gain in total cost to be had by distributing pieces of each relation to several different nodes and computing the join of several large relations at once. The optimization problem is to pick the degree of replication of each relation, under the constraint that the total number of compute-nodes is fixed. We set up this problem as a nonlinear optimization and show that there is always a solution (which must be approximated by rounding to the nearest integers). For some of the most common types of join – star joins and chain joins – we give closed-form solutions to the optimization problem. Finally, we point out that the join algorithm we propose can be implemented using features already present in Hadoop, the open-source implementation of map-reduce

Predicting secondary structures of RNA molecules is a classic example of a biological problem which is solved via the usage of computational tools. The original algorithms for the RNA folding problem (dating back to the late 70's) solve it in O(n^3) time and O(n^2) space complexities (where n denotes the length of the input sequence, and can get up to several hundreds). While these complexities are reasonable for the processing of a single input, they are inadequate for modern analysis of genome-wide data, which may include several billions of input sequences. Recently, the running time was improved to O(nZ), where Z is a sparsity parameter that satisfies n < Z < n^2. In the talk, new results will be presented, which reduce the space complexity to O(Z), and the time complexity to O(LZ), where L is a sparsity parameter that satisfies L < n. The presented techniques extend to a family of RNA folding algorithms, which include more sophisticated problem variants with higher time and space complexities, as well as to algorithms from other fields, such as parsing sentences with a given probabilistic context free grammar.

Power consumption is a major issue in today's datacenters. I will give an overview of the general issue of power consumption in dataceneters and will focus on storage power consumption, which typically comprises a significant percentage of datacenter power. Understanding, managing, and reducing storage power consumption is an essential aspect of any efforts that address the total power consumption of datacenters. We developed a scalable power modeling method that estimates the power consumption of storage workloads. The modeling concept is based on identifying the major workload contributors to the power consumed by the disk arrays. To estimate the power consumed by a given host workload, our method translates the workload to the primitive activities induced on the disks. In addition, we identified that I/O queues have a fundamental influence on the power consumption. Our power estimation results are highly accurate, with only 2% deviation for typical random workloads with small transfer sizes (up to 8K), and a deviation of up to 8% for workloads with large transfer sizes. We successfully integrated our modeling into a power-aware capacity planning tool to predict system power requirements and integrated it into an online storage system to provide online estimation for the power consumed.

The path packing problem for graphs is defined as follows: given a supply graph G=(N,E), and a demand graph (T,S), what is the maximal number of edge-disjoint paths with the end-pair in S in G? The problem is considered under two assumptions: (1) the node degrees in NT T are even, and (2) the demand graph satisfies condition defined by A. Karzanov in his 1989 paper. For any demand graph violating the above condition, the problem is known to be NP-hard even under (1), and only a few cases satisfying (1) and (2) have been solved. The talk will address the relation between the path packing problem and an auxiliary flow problem and will contain a short survey of results achieved in my dissertation. No special background in graph theory! is required.

We study statistical properties of interacting protein interfaces and predict two strong, related effects: (i) statistically enhanced self-attraction of proteins; (ii) statistically enhanced attraction of proteins with similar structures. The effects originate in the fact that the probability to find a pattern self-match between two identical, interacting protein interfaces is always higher compared with the probability for a pattern match between two different, promiscuous protein interfaces. This theoretical finding explains statistical prevalence of homodimers in protein-protein interaction networks reported earlier. Further, our findings are confirmed by the analysis of curated database of protein complexes that showed highly statistically significant overrepresentation of dimers formed by structurally similar proteins with highly divergent sequences (“superfamily heterodimers”). We predict that significant fraction of heterodimers evolved from homodimers with the negative design evolutionary pressure applied against promiscuous homodimer formation. This is achieved through the formation of highly specific contacts formed by charged residues as demonstrated both in model and real superfamily heterodimers. In addition we introduce the notion of structural correlations of amino acid interface density. We predict that protein interfaces with enhanced structural correlations are statistically more promiscuous as compared with proteins possessing a lower degree of interface structural correlations.

References:

1. D. B. Lukatsky and E. I. Shakhnovich, Statistically Enhanced Promiscuity of Structurally Correlated Patterns, Phys. Rev. E 77, 020901(R) (2008). 2. D. B. Lukatsky, B. E. Shakhnovich, J. Mintseris, and E. I. Shakhnovich, Structural Similarity Enhances Interaction Propensity of Proteins, J. Mol. Biol. 365, 1596 (2007). 3. D. B. Lukatsky, K. B. Zeldovich, and E. I. Shakhnovich, Statistically Enhanced Self-Attraction of Random Patterns, Phys. Rev. Lett. 97, 178101 (2006). .

We prove the equivalence, up to a small polynomial approximation factor sqrt(n/ log n), of the lattice problems uSVP (unique Shortest Vector Problem), BDD (Bounded Distance Decoding) and GapSVP (the decision version of the Shortest Vector Problem). This resolves a long-standing open problem about the relationship between uSVP and the more standard GapSVP, as well the BDD problem commonly used in coding theory. The main cryptographic application of our work is the proof that the Ajtai-Dwork and the Regev cryptosystems, which were previously only known to be based on the hardness of uSVP, can be equivalently based on the hardness of the more standard worst-case GapSVP. Also, in the case of uSVP and BDD, our connection is very tight, establishing the equivalence (within a small constant approximation factor) between the two most central problems used in lattice based public key cryptography and coding theory.

This is joint work with Daniele Micciancio and will appear at CRYPTO 2009

In the setting of secure computation, a set of parties wish to securely compute some function of their inputs, in the presence of an adversary. The adversary in question may be static (meaning that it controls a predetermined subset of the parties) or adaptive (meaning that it can choose to corrupt parties during the protocol execution and based on what it sees). In this talk, we study two fundamental questions relating to the basic zero-knowledge and oblivious transfer protocol problems: Adaptive zero-knowledge proofs: We ask whether it is possible to construct adaptive zero-knowledge proofs (with unconditional soundness). Beaver (STOC 1996) showed that known zero-knowledge proofs are not adaptively secure, and in addition showed how to construct zero-knowledge arguments (with computational soundness). Adaptively secure oblivious transfer: All known protocols for adaptively secure oblivious transfer rely on seemingly stronger hardness assumptions than for the case of static adversaries. We ask whether this is inherent, and in particular, whether it is possible to construct adaptively secure oblivious transfer from enhanced trapdoor permutations alone. We provide surprising answers to the above questions, showing that achieving adaptive security is sometimes harder than achieving static security, and sometimes not. First, we show that assuming the existence of one-way functions only, there exists adaptive zero-knowledge proofs for all languages in NP. In order to prove this, we overcome the problem that all adaptive zero-knowledge protocols known until now used equivocal commitments (which would enable an all-powerful prover to cheat). Second, we prove a black-box separation between adaptively secure oblivious transfer and enhanced trapdoor permutations. As a corollary, we derive a black-box separation between adaptively and statically securely oblivious transfer. This is the first black-box separation to relate to adaptive security and thus the first evidence that it is indeed harder to achieve security in the presence of adaptive adversaries than in the presence of static adversaries. A joint work with Yehuda Lindell

Frequent patterns are patterns that appear in a data set frequently. For example, a set of items, such as milk and bread, that appear together frequently in a transaction data set of shopping carts is a frequent itemset. Finding such patterns plays an essential role in mining correlations and other interesting relationships among data. Thus this topic has become an important data mining task and a focused theme in data mining research. The task of finding frequent patterns holds many challenges, such as huge data sets to mine and combinatorial number of frequent patterns to find. The two leading algorithms will be presented in our talk together with some other advanced issues such as mining frequent closed and maximal patterns, using generalizations to get interesting patterns, and more.

We revisit the question of secure multiparty computation (MPC) with two rounds of interaction. It was previously shown by Gennaro et al. (Crypto 2002) that three or more communication rounds are necessary for general MPC protocols which tolerate $tge 2$ corrupted parties, regardless of the total number of parties, and even if {em broadcast} messages are allowed in each round. We complement this negative result by presenting matching positive results. Our main result is that if only {em one} party can be corrupted, then $nge 5$ parties can securely compute any function of their inputs using only {em two} rounds of interaction over secure point-to-point channels (without broadcast or any additional setup). Our protocol provides computational security while making only a black-box use of a pseudorandom generator, or alternatively can provide unconditional security for ``computationally simple'' functions (e.g., functions in $NC^1$). We also prove a similar result in a client-server setting, where there are $mge 2$ clients who hold inputs and should receive outputs, and $n$ additional servers with no inputs and outputs. For this setting we obtain a general MPC protocol which requires a single message from each client to each server, followed by a single message from each server to each client. The protocol is secure against a single corrupted client and against coalitions of $t<n/3$ corrupted servers.

Affective computing is a field of AI and HCI that aims to incorporate the behavioural cues of affective states (emotions, mental states, moods and attitudes) that are common in human-human communication into human-computer and human-robot interfaces, in order to improve the usability and performance of these interfaces. The talk will focus on recognition of subtle affective states from their non-verbal expressions in speech, their analysis and implementation to HCI. Temporal abstractions of paralinguistic speech events, borrowed from various disciplines such as musicology, engineering and linguistics, were extracted from speech signals and used as attributes for the recognition. The recognition was based on a novel multiclass and semi-blind multi-label inference system which was designed to infer co-occurring affective states. The system was used for analysis of expressions during sustained human-computer interaction. The results were validated through correlation to various indicators (multi-modal analysis) and comparison to human performance. Inference results of over 500 different affective states allowed analysis of the relations between these affective states, serving as a tool for verification of taxonomies. The implication is that the system and its architecture can be generalised to new affective states, speakers, languages and scenarios with no additional training.

It is well known that the string edit distance and the alignment of strings coincide, while the alignment of trees differs from the tree edit distance. In this talk, we recall various constraints on directed acyclic graphs that allow for a unique (up to isomorphism) representation, called the path multiplicity representation, and present a new method for the alignment of trees and directed acyclic graphs that exploits the path multiplicity representation to produce a meaningful optimal alignment in polynomial time.

Formal grammars are constraint systems that made up of rules that define the permissible arrangements of tokens (words) in a language. Using grammars we are able to also represent mobile genetic elements (MGE), assemblages of genes and protein interaction sites that can move between molecules in a bacterial cell and often move between organisms through conjugation facilitated by plasmids (horizontal gene transfer). MGEs are the principal cause in the evolution of strains of bacteria that are simultaneously resistant to multiple antibiotics as resistance genes accumulate in MGEs. I will present a novel method to computationally “read” MGEs in bacterial DNA using a context sensitive grammar but deterministic parser. I will also present how the grammar can be used to produce high-fidelity annotations of high level structures such as MGEs and two applications: large-scale surveys and a new method for gene discovery which is complementary to contemporary discovery methods.

Dr. Guy Tsafnat is the head of the Translational Bioinformatics group at the Centre for Health Informatics at UNSW – Australia’s largest and oldest centre for medical informatics research. The Translational Bioindformatics group researchers the role of computational biology in clinical decision support systems. Prior to academia, for 8 years he has conducted research in machine learning in the software industry both in Sydney and in Silicon Valley, California. His PhD and later research are in computational biology and translational bioinformatics, and is focused on computational discovery.

We present a novel approach for model based segmentation of gray-scale images of highly degraded historical documents. Given a training set of characters (of a certain letter), we construct a small set of shape models that cover most of the training set's shape variance. For each gray-scale image of a respective degraded character, we construct a custom made shape prior using those fragments of the shape models that best fit the character's boundary. Experiments show that our method achieves very accurate results both in segmentation of highly degraded characters and both in recognition

Consider the problem of lossless source coding for networks with side information. This model captures scenarios where the network capacity is insufficient to describe the source to its intended destinations, yet, it can still be delivered without loss provided there is sufficient capacity from a helper. The problem of source coding with side information has numerous applications, from sensor networks, where transmission occurs through nodes which may have correlated data, to multimedia networks, where intermediate nodes may have, for example, a lower-resolution version of the required information.

While several spacial cases of this problem have been addressed in the current literature (e.g., the three-node network of Ahlswede and Korner), the general problem remains unsolved. In this work, we derive inner and outer bounds on the rate region and describe sufficient conditions for the tightness of these bounds. Our approach demonstrates how strategies intended for small canonical problems, combined with network coding, can tackle complex networks, while still inheriting the desirable properties of the building blocks used. Furthermore, due to the complexity of solving large networks, it is highly desirable to identify the key parameters which dictate their rate region. This work substantially extends the network scenarios for which maxflow-mincut analysis is know to describe the rate region in full. Finally, in this work we open a new connection between networking and successive refinement of information.

Joint work with Salman Avestimehr and Michelle Effros.

Modern cameras and scanners technologies are capable of generating quality images for models. These images are usually stored as a collection of pictures or points in 3D space. Several approaches are available to process these collections into parametric 3D surfaces that geometrically represent the models. These approaches are usually automatic or semi-automatic, and are supported by algorithms and theories from many research areasin computer science (probability, graph theory, numerical analysis).

In this presentation, I will present several selected papers in computer graphics that attempt to solve the mentioned reconstruction problem. The papers were presented in top rated conferences related to computer graphics (SIGGRAPH2007, SIGGRAPH2008, EUROGRAPHICS2009). Some elementary backgroundis required in the following research areas: random algorithms, probability, numerical analysis, graphs algorithms, and linear algebra, however all the algorithms will be represented clearly (first degree background in related courses should be sufficient).

Topics and approaches that will be presented as time and interest allow: The efficient RANSAC approach, TheGRAPH-CUT as an optimization problem, The "4 is less than 3" approach for random validation and scoring algorithms, Advanced approach for cleaningnoisy datasets.

Many of us are familiar with the concept of an express line (say, 1-10 products) in the supermarket. While traveling in Argentina I saw a supermarket with two express lanes, 0ne for 1-10 items and a second for 11-20 items. Imagine that we manage all the lanes in this way, each one responsible for a different range of items. In the context of computer systems, this has first been suggested and studied by Mor Harchol-Balter and her collaborators in the last 10 years. However, basic questions remained unresolved. How to determine the best range for each lane, should the first lane handle 1-8 items, 1-12 items, etc.? What will the resulting performance be and how does it compare with other methods? How does the distribution of the number of items customers purchase change the behavior of the system? We provide some answers to these and several other issues and explore future directions of research. Joint work with Hagit Sarfati.

Collaboration of multiple intelligent agents on a shared task is a complex research issue, made particularly difficult when communication is limited or impossible. A common solution in multi-agent systems is to commit a team of collaborating agents to a joint plan. Since any deviation from the plan by an agent is hazardous, the common treatment of potential unplanned opportunities is either by ignoring them (even when ``opportunistic'' actions increase the expected utility of the team), or by ad-hoc rules determining whether to accept such opportunities. Neither of these solutions is desirable. In our framework, Abstract World for Opportunistic Local decisions ? AWOL for shortfootnote{Unlike the military term AWOL (Absent WithOut Leave), here the agent defaults only in order to {em increase} (expected) team utility.}, we attempt a disciplined treatment of opportunistic actions, in the context of an existing joint plan. The idea is to model the (stochastic) tradeoff of such opportunistic actions vs. continued commitment to the joint plan, while abstracting away as much as possible from the state of the world. The abstract model is evaluated using strict decision-theoretic criteria, with the goal of applying the optimal decision on whether to accept an opportunistic action in the original domain.

Jonathan Katz and Luca Trevisan defined the concept of Locally Decodable Codes (LDC). LDC are error correcting codes where a bit of the message can be probabilistically recovered by looking at a limited number of symbols of a corrupted encoding. After defining LDCs formally we will see some negative results. First, that local decoding is impossible when looking at only one symbol of the corrupted encoding. Then we will see a lower bound on the codewords' length.

This paper appeared in STOC00.

We are used to studing the runtime of algorithms as an increasing function of the data set size, and are happy when this increase is not so bad (e.g. when the runtime increases linearly, or even polynomiall, with the data set size). Traditional runtime analyzis of learning is also viewed this way, and studies how training runtime increases as more data is available. However, considering the true objective of training, which is to obtain a good predictor, I will argue that training runtime should actually be studied as a *decreasing* function of training set size. Focusing on training Support Vector Machines (SVMs), I will then present both theoretical and empirical results demonstrating how a simple stochastic subgradient descent approach indeed displays such monotonic decreasing behavior. I will also discuss a similar phenomena in the context of Gaussian mixture clustering, where it appears that excess data turns the problem from computationally intractable to computationally tractable.

Joint work with Shai Shalev-Shwartz, Karthik Sridharan, Yoram Singer, Greg Shakhnarovich and Sam Roweis

Verifiable random functions, introduced by Micali, Rabin and Vadhan, are a primitive aimed at coping with the absence of random oracles in the real world. While replacing the random oracle with a pseudorandom function can be a solution in some cases, its drawback is the function's seed owner's ability to adaptively change its answers and thus manipulate the protocol. In a verifiable random function, the seed owner produces a public-key at the beginning of the interaction, which constitutes a commitment to all (pseudorandom) values of the function. It can then produce, for any input, a proof that the output has been computed correctly (with respect to that public-key). Even a maliciously chosen public-key must not enable proving different outputs for the same input.

Verifiable random functions found various implementations and uses. However, no implementation based on a general assumption (such as the existence of one-way functions or even stronger assumptions like trapdoor permutations) is known.

We define a relaxation of verifiable random functions, which we call weak verifiable random functions, in the spirit of Naor and Reingold's weak pseudorandom functions. We show that this relaxed notion can be derived from various assumptions (including general assumptions); that the existence of weak verifiable random functions is essentially equivalent to the existence of non-interactive zero-knowledge proof systems for all of NP (in the common random string model); and prove that weak verifiable random functions (and thus also "standard" verifiable random functions) cannot be constructed from one-way permutations in a black-box manner, providing a first separation result for (standard) verifiable random functions from any cryptographic primitive.

No prior knowledge on black-box separations is assumed.

Joint work with Shafi Goldwasser, Guy Rothblum and Vinod Vaikuntanathan.

In the Multislope Ski Rental problem, the user needs a certain resource for some unknown period of time. To use the resource, the user must subscribe to one of several options, each of which consists of a one-time setup cost (“buying price”), and cost proportional to the duration of the usage (“rental rate”). The larger the price, the smaller the rent. The actual usage time is determined by an adversary, and the goal of an algorithm is to minimize the cost by choosing the best option at any point in time. Multislope Ski Rental is a natural generalization of the classical Ski Rental problem (where the only options are pure rent and pure buy), which is one of the fundamental problems of online computation. The Multislope Ski Rental problem is an abstraction of many problems, where online choices cannot be modeled by just two alternatives, e.g., power management in systems which can be shut down in parts. In this work we study randomized online strategies for Multislope Ski Rental. Our results include an algorithm that produces the best possible randomized online strategy for any additive instance, where the cost of switching from one alternative to another is the difference in their buying prices; and an e-competitive randomized strategy for any (non-additive) instance. We also provide a randomized strategy with a matching lower bound for the case of two slopes, where both slopes have positive rents.

Joint work with Zvi Lotker and Boaz Patt-Shamir.

The Black-and-White Coloring (BWC) problem is defined as follows. Given an undirected graph $G$ and positive integers $b, w$, determine whether there exists a partial vertex-coloring of $G$ such that $b$ vertices are colored black and $w$ vertices white (with all other vertices left uncolored), such that no black vertex and white vertex are adjacent.

The BWC problem has been introduced in general, and proved to be $NP$-complete, by Hansen, Hertz and Quinodoz, who also gave an $O(n^3)$ algorithm for trees. In this talk we introduce another algorithm, whose running time is $O(n^2 lg ^3 n)$. We also present an improvement to our algorithm, which works for almost all labelled trees in time~$n^{1+o(1)}$.

If time permits, we will present an algorithm which solves the BWC problem for chordal graphs.

The celebrated Minimum Spanning Tree protocol of Gallager, Humblet and Spira has been very influential in the network algorithms arena. Although it is very intuitive it has proven to be a very hard protocol to verify correct. Despite many attempts, only two proofs were previously known, each roughly 180 pages long. We describe a new proof of correctness that is considerably shorter, and is the first to formalize the reasoning suggested in the original paper. The proof is based on a new abstraction that facilitates modular reasoning about the protocol. The talk will be self contained and will assume no prior knowledge of the subject matter.

Based on joint work with Benny Shimony.

One goal of Artificial Intelligence is to enable the creation of robust, fully autonomous agents that can coexist with us in the real world. Such agents will need to be able to learn, both in order to correct and circumvent their inevitable imperfections, and to keep up with a dynamically changing world. They will also need to be able to interact with one another, whether they share common goals, they pursue independent goals, or their goals are in direct conflict. This talk will present current research directions in machine learning, multiagent reasoning, and robotics, and will advocate their unification within concrete application domains. Ideally, new theoretical results in each separate area will inform practical implementations while innovations from concrete multiagent applications will drive new theoretical pursuits, and together these synergistic research approaches will lead us towards the goal of fully autonomous agents.

Given a set F

f_1, …, f_n of n continuous functions from R^d to R, we consider the problem of finding a center x* that minimizes OPT

sum_{f in F} f(x*). We present an efficient bi-criteria approximation that returns a small set X such that sum_{fin F}min_{xin X}f(x) exceeds OPT by a factor of no more than (1+eps). The quality of our algorithm (namely, the running time and the size of X) depends on the combinatorial complexity of the set F. I.e., the complexity of the arrangement formed by the functions f_i when viewed as manifolds in R^{d+1}. Given a set of n data elements e_1, …, e_n, a parameter k, and a cost function d(-,-), the ``bi-criteria approximate clustering'' problem addressed the finding of k'>k centers c_1, ..,c_{k'} for which sum_{i=1}^n min_j d(e_i,c_j) is at most (1+eps)-times the optimal k clustering. Bi-criteria approximate clustering stands at the center of several approximation algorithms for clustering and has seen a significant amount of research over the past decade.

Defining the function set F appropriately, our algorithm can be applied to obtain bi-criteria approximate clustering in several standard and new clustering settings. These include variants of k-mean clustering, where both the data elements e_i and the centers c_j are points in R^d, and the projective clustering in which the centers c_j are flats (affine subspaces) instead of points. We suggest the first approximation for the generalized projective clustering, where both the data elements e_i and the centers c_j are flats. New settings for which our scheme applies include recovering missing entries in a matrix, solving k-means in the context of missing data, motion analysis, and camera pose estimation. Not only does our framework unify several previous approaches for bi-criteria approximation, in the case of projective clustering it significantly improves the state of the art. Joint work with Michael Langberg.

We introduce planning games, the study of interactions of self-motivated agents in automated planning settings. Planning games extend STRIPS-like models of single-agent planning to systems of multiple self-interested agents, providing a rich class of structured games that capture subtle forms of local interactions. We consider two basic classes of planning games and adapt game-theoretic solution concepts to these models. In both models, agents may need to cooperate in order to achieve their goals, but are assumed to do so only in order to increase their net benefit. For each model we study the computational problem of finding a stable solution and provide efficient algorithms for systems exhibiting acyclic interaction structure.

Generalization sets are a software analysis concept that describes specialization of behavior in a way less restrictive than inheritance in OOP languages. According to the UML 2.0 standard, a generalization set divides a type into disjoint sub-types. Each sub-type may override the behavior of the super-type. A type may have several generalization sets. The sets are non-disjoint. An instance of the super-type may be an instance of each sub-type in each set. For example, an instance of a Person can be both an instance of an Employed or Unemployed person, and of a Resident or a Foreigner person.

In this talk we present several possible designs for the concept of Generalization Sets. Each design is centered around a particular design pattern: Inheritance, State or Enumeration. We discuss the advantages and disadvantages of each approach, and show examples of combining instances of these patterns to implement a type that has more then one generalization set. Finally, we compare the different properties of the patterns.

This talk is based on a submission to ECOOP 2009 by Mayer Goldberg and Guy Wiener

We consider the following question: given a two-argument boolean function $f$, represented as an $Ntimes N$ binary matrix, how hard is to determine the (deterministic) communication complexity of $f$?

We address two aspects of this question. On the computational side, we prove that, under appropriate cryptographic assumptions (such as the intractability of factoring), the deterministic communication complexity of $f$ is hard to approximate to within some constant. Under stronger (yet arguably reasonable) assumptions, we obtains even stronger hardness results that match the best known approximation.

On the analytic side, we present a family of functions for which determining the communication complexity (or even obtaining non-trivial lower bounds on it) imply proving circuit lower bounds for some corresponding problems. Such connections between circuit complexity and communication complexity were known before (Karchmer-Wigderson 1988) only in the more involved context of relations (search problems) and not in the context of functions (decision problems). This result, in particular, may explain the difficulty of analyzing the communication complexity of certain functions such as the ``clique vs. independent-set'' family of functions, introduced by Yannakakis (1988).

Locally Decodable Codes (LDC) allow one to decode any particular symbol of the input message by making a constant number of queries to a codeword, even if a constant fraction of the codeword is damaged. In a recent work ~cite{Yekhanin08} Yekhanin constructs a $3$-query LDC with sub-exponential length of size $exp(exp(O(frac{log n}{loglog n})))$. However, this construction requires a conjecture that there are infinitely many Mersenne primes. In this paper we give an unconditional $3$-query LDC construction with a shorter codeword length of $exp(exp(O(sqrt{log n log log n })))$. We also give a $2^r$-query LDC with length of $exp(exp(O(sqrt[r]{log n log log^{r-1} n })))$. The main ingredient in our construction is the existence of super-polynomial size set-systems with restricted intersections by cite{Grolmusz00} which hold only over composite numbers.

The topology of metabolic networks may provide important insights not only into the metabolic capacity of species, but also into the habitats in which they evolved. In this talk I will present several analyses of metabolic networks and show how various ecological insights can be obtained from genomic-based data. I will first introduce various factors that affect the structure of metabolic networks, and specifically the environmental and genetic determinants that affect network modularity. I will then present the first large-scale computational reconstruction of metabolic growth environments, analyzing the metabolic networks of hundreds of species and using a graph-theory based algorithm to identify for each species a set of seed compounds that must be exogenously acquired. Such seed sets form ecological "interfaces" between metabolic networks and their surroundings, approximating the effective biochemical environment of each species. Phylogenetic analysis of the seed sets reveals the complex dynamics governing gain and loss of biosynthetic capacity across the phylogenetic tree. I will further present an extension of this framework, accounting for interactions between species, by introducing a pair-wise, topology-based measure of biosynthetic support, which reflects the extent to which the nutritional requirements of one species could be satisfied by the biosynthetic capacity of another. I will show that this measure is aligned with host-parasite interactions and facilitates successful prediction of such interactions on a large-scale. Finally, I will discuss the application of this approach to the analysis of microbial communities and metagenomic data of the human microbiota and outline future research directions; The "reverse ecology" approach demonstrated in these analyses lays the foundations for further studying the complex web of interactions characterizing various ecosystems and the evolutionary interplay between organisms and their habitats on a large scale.

Traditionally, text categorization has been studied as the problem of training of a classifier using labeled data. However, people can categorize documents into named categories without any explicit training because we know the meaning of category names. In this paper, we introduce {em Dataless Classification}, a learning protocol that uses world knowledge to induce classifiers without the need for any labeled data. Like humans, a dataless classifier interprets a string of words as a set of semantic concepts. We propose a model for dataless classification and show that the label name alone is often sufficient to induce classifiers. Using Wikipedia as our source of world knowledge, we get 85.29% accuracy on tasks from the 20 Newsgroup dataset and 88.62% accuracy on tasks from a Yahoo! Answers dataset without {em any labeled or unlabeled}

Short Bio: Lev Ratinov is a Phd candidate in University of Illinois at Urbana-Champaign. He has done work on Machine Learning in Natural Language Processing and Information Extraction and has published a number of papers in several international conferences including "Dataless Classification"(AAAI08), "Learning and Inference with Constraints" (AAAI08), and "Guiding Semi-Supervision with Constraint-Driven Learning" (ACL07

The study of ranking crosses many disciplines. Social choice theoreticians have been interested for centuries in finding a good way to rank a set of candidates. Econometricians have been asking for decades how people choose from (and more generally rank) alternative sets. More recently, information retrieval theoreticians and practitioners have been interested in ranking search query results. In verification, practitioners have been interested in ordering variable sets so as to reduce the time to detect software errors. These recent problems have been identified in both machine learning and combinatorial optimization communities. I will present my contribution on both fronts.

Randomized algorithms have been using measure concentration phenomena in countless cases, and in particular in dimension reduction and streaming. These tools have become a standard black box, and their computational aspects have been almost taken for granted. In recent work I have discovered an exciting and surprising new method for computing a standard dimension reduction algorithm (Johnson-Lindenstrauss) more efficiently than previously assumed. This method has been later used to speed up various standard high dimensional linear algebraic computations (such as SVD) and fueled increased collaboration between computer science and analysis. The main ingredient is the use of a Fast Fourier Transform and a type of uncertainty principle.

For most natural language processing tasks, unsupervised methods significantly underperform their supervised counterparts. In this talk, I will demonstrate that multilingual learning can narrow this gap. The key insight is that joint learning from several languages reduces uncertainty about the linguistic structure of individual languages. These methods exploit the deep structural connections between languages, connections that have driven many important discoveries in anthropology and historical linguistics.

I will present multilingual unsupervised models for morphological segmentation and part-of-speech tagging. Multilingual data is modeled as arising through a combination of language-independent and language-specific probabilistic processes. This approach allows the model to identify and learn from recurring cross-lingual patterns, ultimately to improve prediction accuracy in each language. I will also discuss ongoing work on unsupervised decoding of ancient Ugaritic tablets using data from related Semitic languages.

This is joint work with Benjamin Snyder, Tahira Naseem and Jacob Eisenstein.

Proteins are biological molecules which are involved in virtually every process and aspect in life - from the flexing of our muscles to our immune system response. It is widely accepted that proteins are dynamic molecules with well-defined three-dimensional structures and that understanding the structure and dynamics of proteins is crucial for understanding their function and the processes they mediate. Various computational methods exist for modeling and simulating protein structure and dynamics, but several traditional methods are limited due to the large amount of calculations involved. This talk will present a wide spectrum of methods for searching the conformational space of proteins and their application to specific cases. On the one end of the spectrum, Molecular Dynamics calculations are used for detailed analysis. On the other end, robotics-inspired search techniques are used to characterize the structure and dynamics of proteins by representing them using a mechanistic/geometric models subject to physics constraints. Two novel search methods will be discussed, aimed to characerize the structure, dynamics and flexibility of protein structures at various levels of details: the first is a hybrid method that combines an efficient robotics-inspired sampling algorithm with small scale, detailed simulations to characterize the local dynamics of proteins. The second is a fast, approximate search method that samples large scale structural changes in proteins.

Getting the deterministic complexity closer to the best known randomized complexity is an important goal in algorithms and communication protocols. In this work, we investigate the case where instead of one input, the algorithmprotocol is given multiple inputs sampled independently from an arbitrary unknown distribution. We show that in this case a strong and generic derandomization result can be obtained by a simple argument.

Our method relies on extracting randomness from ``same-source'' product distributions, which are distributions generated from multiple independent samples from the same source. The extraction process succeeds even for arbitrarily low min-entropy, and is based on the order of the values and not on the values themselves (This may be seen as a generalization of the classical method of Von-Neumann extended by Elias for extracting randomness from a biased coin).

The tools developed in the paper are generic, and can be used elsewhere. We present applications to streaming algorithms, and to implicit probe search cite{FiatNaor93}. We also refine our method to handle product distributions, where the i'th sample comes from one of several arbitrary unknown distributions. This requires creating a new set of tools, which may also be of independent interest.

Joint work with Ariel Gabizon

Alon and Tarsi have introduced an algebraic technique for proving upper bounds on the choice number of graphs; the upper bound on the choice number of $G$ obtained via their method, was later coined the emph{Alon-Tarsi number of $G$} and was denoted by $AT(G)$. In the talk I will relate this parameter to a certain weighted sum of the proper colorings of $G$. Other than the appealing notion of obtaining upper bounds on the choice number of a graph via its proper colorings (in some sense), this result has many applications. Some of them are known; for these we give unified, and often also simpler and shorter proofs; and some are new.

In the first part of the talk I will introduce chromatic, choice, and Alon-Tarsi numbers of graphs. In the second part I will state the main result and some of its applications.

We consider private data analysis in the setting in which a trusted and trustworthy curator, having obtained a large data set containing private information, releases to the public a ``sanitization'' of the data set that simultaneously protects the privacy of the individual contributors of data and offers utility to the data analyst. We focus on the case where the process is non-interactive; once the sanitization has been released the original data and the curator play no further role.

Blum et al. [STOC '08] showed a remarkable result: for any collection of counting predicate queries, the exponential mechanism of McSherry and Talwar [FOCS '07] can be used to (inefficiently) generate a synthetic data set that maintains approximately correct fractional counts for all of the queries, while ensuring a strong privacy guarantee, known as differential privacy.

In this work we investigate the computational complexity of such non-interactive privacy mechanisms, mapping the boundary between feasibility and infeasibility. We show:

1. For any polynomial-size query set C and polynomial-size data universe there is an efficient algorithm for generating a privacy-preserving synthetic data-set that maintains approximate fractional counts, even when the size of the original dataset is sub-polynomial in |C|.

2. When either the query set or the data universe are super-polynomial, assuming one-way functions exist, there is no efficient general method for releasing a privacy-preserving synthetic data-set that maintains approximate fractional counts. In particular, this is a case where the exponential mechanism cannot, in general, be implemented efficiently.

3. Turning to the potentially easier problem of privately generating an arbitrary data structure (not necessarily synthetic data) from which it is possible to approximate counts for a given concept, there is a tight connection between hardness of sanitization and the existence of traitor tracing schemes, methods of content distribution in which keys are assigned to subscribers in a way that given any useful ``pirate'' decoding box, constructed by a coalition of malicious subscribers, it is possible to identify at least one of them.

Joint work with Cynthia Dwork, Moni naor, Omer Reingold and Salil Vadhan.

We will describe some recent trends in the storage industry such as, thin provisioning, de-duplication, MAID, SSD drives and others. We will show how these may be used to provide less power consuming storage while maintaining or improving price/performance.

If time permits we will also discuss power issues in the context of queueing theory.

The filesystem API of contemporary systems exposes programs to TOCTTOU (time of check to time of use) race-condition vulnerabilities, which occur between pairs of check/use system calls that involve a name of a file. Existing solutions either help programmers to detect such races (by pinpointing their location) or prevent them altogether (by altering the operating system). But the latter alternative is not prevalent, and the former is just the first step: programmers must still address TOCTTOU flaws within the limits of the existing API with which several important tasks cannot be safely accomplished in a portable straightforward manner. The recent "filesystem maze" attack further worsens the problem by allowing adversaries to deterministically win races and thus refuting the common perception that the risk is small. In the face of this threat, we develop a new algorithm that allows programmers to effectively aggregate a vulnerable pair of distinct system calls into a single operation that is executed "atomically". This is achieved by emulating one kernel functionality in user mode: the filepath resolution. The surprisingly simple resulting algorithm constitutes a portable solution to a large class of TOCTTOU vulnerabilities, without requiring modifications to the underlying operating system.

---

Joint work with Tomer Hertz (Microsoft Research), David Wagner (Berkeley), and Dilma Da Silva (IBM T.J. Watson Research Center). Based on http://www.usenix.org/events/fast08/tech/tsafrir.html USENIX File & Storage Technologies (FAST'08). Awarded best paper.

We survey lower bound techniques in communication complexity. Lower bounds in communication complexity can be roughly divided into two groups. One group uses information theoretic principles and the other uses geometric tools. We focus on the later family of lower bounds. A partial list of the geometric tools that are used is: Fourier analysis, singular values, operator norms, factorization norms and approximation theory. But perhaps the strongest tool, which in a way ties everything together, is duality. Duality and its role in proving lower bounds in communication complexity will be a main subject of our talk. Although we discuss applications to communication complexity, the techniques we present can be used in a much more general settings. We will mention connections to learning theory and circuit complexity.

The talk is self-contained, no prior knowledge is assumed.

I will discuss a notion of privacy called *differential privacy* in conjunction with *learning problems*, and give a complexity-theoretic classification of learning problems that can be efficiently solved in a differentially private manner.

Roughly speaking, the challenge in constructing private learning algorithms is that the outcome should depend very weakly on each of the specific examples. Interestingly, it happens that two of the main models considered in computational learning theory – PAC and SQ – exactly correspond to Imposing restrictions on the way private information is distributed, and on interaction. I will discuss this correspondence, and also issues of efficiency of private learning, and a few of the open problems related to private learning.

The talk with be self-contained.

Joint work with Shiva Kasiviswanathan, Homin Lee, Sofya Raskhodnikova, and Adam Smith, FOCS 2008.

The collaborative filtering approach to recommender systems predicts user preferences for products or services by learning past user-item relationships. Their significant economic implications made collaborative filtering techniques play an important role at known e-tailers such as Amazon and Netflix. This field enjoyed a surge of interest since October 2006, when the Netflix Prize competition was commenced. Netflix released a dataset containing 100 million anonymous movie ratings and challenged the research community to develop algorithms that could beat the accuracy of its recommendation system, Cinematch. In this talk I will survey the competition together with some of the principles and algorithms, which have led us to winning the Progress Prizes in the competition.

Bio: Yehuda Koren completed his PhD in CS at The Weizmann Institute on 2003. He was with AT&T Research during 2003-2008, and recently joined Yahoo! Research (Haifa)

We consider scenarios where a sequence of tests is to be applied to an object, where the result of a test may be that a decision (such as classification of the object) can be made without running additional tests. Thus, one seeks an ordering of the tests that is optimal in some sense, such as minimum expected resource consumption. Such sequences of tests are commonly used in computer vision cite{Viola2001} and other applications. We examine conditions under which one can efficiently find an optimal ordering of the tests. Two types of dependencies between tests are examined: ordering constraints, and statistical dependencies. We show that with dependencies the optimization problem is NP-hard in the general case, and provide low-order polynomial time algorithms for special cases with non-trivial constraint structures. (Patent pending)

Fixed-parameter algorithms provide an alternative methodology of coping with NP-hardness that allows to solve hard optimization problems exactly and in a low-polynomial time.The necessary condition for design of a fixed- parameter algorithm is the presence of a parameter associated with the given problem. The time complexity of a fixed-parameter algorithm can be represented in a form $O(f(k)*n^c)$, where $k$ is the parameter, $f(k)$ is an exponential function on the parameter, $c$ is a constant usually not greater than 3. Thus a fixed-parameter algorithm is exponential in the parameter but polynomial in the input size. The low-polynomial time is achieved for small values of $k$.

The design of fixed-parameter algorithms was inspired by observation that in areas like bioinformatics and network design many hard problems are associated with natural parameters that in practice are very small.

Problems that admit fixed-parameter algorithms are called fixed-parameter tractable (FPT) and the area of Theoretical Computer Science studying various aspects of fixed-parameter tractability and intractability is called Parameterized Complexity. One of the key questions investigated in the area is classification of problems i.e. telling whether a particular problem is FPT. The community has identified a number of the most challenging open questions related to classification of problems. One of such questions is understanding whether the following problem is FPT: given a 2-CNF formula, is it possible to remove at most $k$ clauses so that the resulting formula becomes satisfiable? This problem is called 2-CNF deletion. It attracted attention of researchers due to a large number of potential applications of this problem. Despite a number of attempts to solve the problem, the status of fixed-parameter tractability of the problem remained unresolved for more than 10 years. The fixed-parameter tractability of the problem has been confirmed in Razgon, O'Sullivan "Almost 2-SAT is fixed-parameter tractable", ICALP 2008.

In this talk I will briefly introduce the area of Parameterized Complexity and overview main ideas of the fixed-parameter algorithm for the 2-CNF deletion problem. The talk is designed for the audience completely unfamiliar with the area of Parameterized Complexity.

Poker is a challenging problem for artificial intelligence research: multiple opponents (up to 10), stochastic element (cards being dealt), imperfect information (don't know the opponent's cards), deception (bluffing), user modeling (identifying player patterns), and risk management (betting decisions). Unlike the classic AI game, chess, poker is more relevant to real-world situations including negotiations, military strategy, and e-commerce. For over a decade, the University of Alberta Computer Poker Group has been working on building a high-performance poker program. This work has led us through multiple distinct phases of program design, each new idea "promising" to be the breakthrough to world-class play. Finally, we appear to be close. In a recent Man Versus Machine Match, University of Alberta programs narrowly defeated a team of world-class players. In this talk we will motivate the research, compare the different program designs, and discuss what it will take to raise the stakes in man-machine poker competitions.

Minesweeper is a popular single player game. It has been shown that the Minesweeper consistency problem is NP-complete and the Minesweeper counting problem is #P-complete. We present a polynomial algorithm for solving these problems for minesweeper graphs with bounded tree width. There is a close connection between the minesweeper problems and general CSP problems.

Modern software engineering heavily relies on clearly specified interfaces for separation of concerns among designers implementing components and programmers using those components. The interface of a component describes the functionality and constraints on the correct usage in a succinct manner. The need for interfaces is evident for assembling complex systems from components, but more so in modern control applications where a network of sensors and actuators is deployed to achieve complex control objectives. The notion of an interface for a control component must incorporate information about timing and resource needs, information that is nor represented in current programing interfaces.

In the talk, I'll propose formal languages as an interface for resource sharing and timing. For example, imagine two components that share some resource (e.g. CPU), allocated in a time-triggered manner. With the proposed interface, each component specifies a formal language over the alphabet {0,1} of "safe schedules", i.e., if w=w(1),w(2),… is in this languages, and the component gets the resource in every time slot t for which w(t)=1, then the component meets its performance requirements. The main advantage of this interface is that it allows modular analysis of each component and, then, constraints can be combined using languages intersection.

I'll present theoretical results concerning applications of the above approach to software control systems (where controllers are implemented by a software that shares resources). I'll also introduce a Real-Time-Java based tool, called RTComposer, that supports the proposed methodology. In RTComposer, scheduling specifications can be given as periodic tasks, or using temporal logic, or as omega-automata, or stability requirements. Components can be added dynamically, and non-real-time components are allowed. The benefits of the approach will be demonstrated and applications to wireless sensor/actuator networks based on the WirelessHART protocol and to distributed control systems based on the Control Area Network (CAN) bus (used, e.g., in automotive applications) will be discussed.

The topology of metabolic networks may provide important insights not only into the metabolic capacity of species, but also into the habitats in which they evolved. In this talk I will present several analyses of metabolic networks and show how various ecological insights can be obtained from genomic-based data. I will first introduce various factors that affect the structure of metabolic networks, and specifically the environmental and genetic determinants that affect network modularity. I will then present the first large-scale computational reconstruction of metabolic growth environments, analyzing the metabolic networks of hundreds of species and using a graph-theory based algorithm to identify for each species a set of seed compounds that must be exogenously acquired. Such seed sets form ecological "interfaces" between metabolic networks and their surroundings, approximating the effective biochemical environment of each species. The seed sets' composition significantly correlates with several properties characterizing the species' environments and agrees with biological observations concerning major adaptations (e.g. obligate parasitism). Phylogenetic analysis of the seed sets reveals the complex dynamics governing gain and loss of biosynthetic capacity across the phylogenetic tree. I will further present an extension of this framework, accounting for interactions between species, by introducing a pair-wise, topology-based measure of biosynthetic support, which reflects the extent to which the nutritional requirements of one species could be satisfied by the biosynthetic capacity of another. I will show that this measure is aligned with host-parasite interactions and facilitates successful prediction of such interactions on a large-scale. Integrating this method with phylogenetic analysis and calculating the biosynthetic support of ancestral Firmicutes species further reveal a large-scale evolutionary trend of biosynthetic capacity loss in parasites. Finally, I will discuss the application of this approach to the analysis of microbial communities and metagenomic data of the human microbiota and outline future research directions; The "reverse ecology" approach demonstrated in these analyses lays the foundations for further studying the complex web of interactions characterizing various ecosystems and the evolutionary interplay between organisms and their habitats on a large scale.

The heritage of a community defines the character and uniqueness of that community. However, because of the high degree of degradation suffered by very old documents, analyzing them is a difficult task. In this presentation, a degradation model is provided which is based on the physical understanding of degradation. This model can also be used to generate ground truth datasets. Based on this degradation model, various enhancement and restoration methods are proposed in different paradigms such as variational framework, PDE-based methods, and statistical methods. For improving the performance of these methods, several classifiers on different levels ranging from local level to global level are introduced. Based on the latter concept, all preprocessing methods can be modified and improved. The methods are tested on several datasets with promising results.

Short bio:

Mohamed Cheriet was born in Algiers (Algeria) in 1960. He received his B.Eng. from USTHB University (Algiers) in 1984 and his M.Sc. and Ph.D. degrees in Computer Science from the University of Pierre et Marie Curie (Paris VI) in 1985 and 1988 respectively. Since 1992, he has been a professor in the Automation Engineering department at the École de Technologie Supérieure (University of Quebec), Montreal, and was appointed full professor there in 1998. He co-founded the Laboratory for Imagery, Vision and Artificial Intelligence (LIVIA) at the University of Quebec, and was its director from 2000 to 2006. He also founded the SYNCHROMEDIA Consortium (Multimedia Communication in Telepresence) there, and has been its director since 1998. His interests include document image analysis, OCR, mathematical models for image processing, pattern classification models and learning algorithms, as well as perception in computer vision. Dr. Cheriet has published more than 200 technical papers in the field, and has served as chair or co-chair of the following international conferences: VI’1998, VI’2000, IWFHR’2002, and ICFHR’2008. He currently serves on the editorial board and is associate editor of several international journals: IJPRAI, IJDAR, and Pattern Recognition. He co-authored a book entitled,”Character Recognition Systems: A guide for Students and Practitioners,” John Wiley and Sons, 2007. Dr. Cheriet is a senior member of the IEEE and the chapter chair of IEEE Montreal Computational Intelligent Systems (CIS).

The initial sequencing and analysis of the human genome, completed in 2003, was a major biological breakthrough, leading to a better understanding of the evolution and function of many human genes. The key to translating this newly acquired knowledge into medical advances relies on the availability of the genomes of many individuals, and in the study of correlation between genomes and diseases. Because the initial human genome was sequenced over 8 years and at the cost of $3 billion, another technological leap was necessary in order to allow for the economical sequencing of the genomes of many humans. Today this leap has been accomplished: Next-Generation Sequencing (NGS) technologies are able to sequence a human genome in a few weeks, at a cost of $10,000 to $100,000. Using these technologies, scientists are hoping to sequence thousands of human genomes in the next few years, and eventually allow each individual to know his or her personal genome.

Some of the biggest remaining challenges on the path to the personal genome are algorithmic. The NGS technologies are only able to read many small fragments of a genomic sequence, and reconstructing the source genome from these fragments, as well as the analysis of the differences between the sets of fragments from various individuals are difficult computational problems. Furthermore, the challenges of using the NGS datasets are exacerbated by the errors and biases in the underlying sequencing technologies. In this talk I will give an overview of genome sequencing and NGS technologies, and discuss some of the computational methods used to address the challenges posed by NGS datasets.

A new aspect for erasure coding is considered, namely, the possibility that some portion of the arriving packets are corrupted in an undetectable fashion. In practice, the corrupted packets may be attributed to a portion of the communication paths that are leading to the receiver and are controlled by an adversary. Alternatively, in case packets are collected from several sources, the corruption may be attributed to a portion of the sources that are malicious.

Corruption resistant fountain codes are presented; the codes resemble and extend the LT and Raptor codes. To overcome the corrupted packets received, our codes use information theoretic techniques, rather than cryptographic primitives such as homomorphic one-way-(hash) functions. Our schemes overcome adversaries by means of using slightly more packets than the minimal number required for revealing the encoded message, and using a majority over the possible decoded values. We also present a more efficient randomized decoding scheme.

Beyond the obvious use, as a rateless corruption resilient erasure code, our code has several important applications in the realm of distributed computing.

Shape inference is a crucial part of image understanding and indeed many studies have addressed this problem in different ways over the years. However, when the object is specular, existing studies usually allow only spare reconstruction and have strong restrictions such as known or calibrated environment. We suggest a new approach for specular shape reconstruction when the environment is neither calibrated nor known. Furthermore unlike previous work, our approach considers general surfaces and allows dense reconstruction. We consider far-field illumination, where the object-environment distance is relatively large, and we examine the dense specular flow that is induced on the image plane through relative object-environment motion. We show that, under these very practical conditions, the observed specular flow can be related to surface shape through a pair of coupled nonlinear partial-differential equation, which we call the '3D shape from specular flow' equation. We show that by solving this equation one can recover the surface. Importantly, this relationship depends only on the environment's relative motion and not on its content. We present two novel reconstruction algorithms. The first one presents an analytic method for recovery of the shape under special environment rotation. In the second algorithm a specular surface is recovered in the presence of arbitrary environment motion. We show that by a linear combination of the observed specular flows we reduce the problem to the same special case that the first algorithm can solve. Finally, we discuss some numerical issues related to the suggested reconstruction algorithms and validate our results in several experiments.

Given a geometric structure of a building (B) and the location of transmitter (T) and a receiver (R) inside B. We would like to calculate all bounded length Radio Paths between T and R. This work presents a new Radio Paths computation frame-work specially design for complex indoor prediction. The suggested new algorithm (IDRP) uses a geometric visibility graph (of B) to traverse all possible bounded paths. Such paths are needed in order to compute the signal strength of T as received at R. We have implemented a prototype version of the IDRP algorithm and performed preliminary experiment testing the radio paths over complex buildings. The main conclusion is that the new IDRP is time efficient and can compute all relevant radio paths even on relatively complex building in a fraction of a second. The suggested work is currently being implemented as part of the Israeli Short Range Consortium (ISRC), at the last part of the talk several open questions will be posted. This talk requires NO background in Radio Frequency.

Keywords: Indoor wireless communication, 3D Geometric paths, Approximating Signal Strength, Visibility Graph.

Boaz is a faculty member in the department of computer science at Ariel University Center. He received the B.Sc., M.Sc. and Ph.D. degrees in Computer Science from Ben-Gurion University, Israel. During 2004-2005 he was a post doctoral fellow at Simon Fraser University - Vancouver Canada. His main area of research is Computational Geometry and GIS algorithms, his research includes terrain simplification, layout design and simulation of wireless networks, visibility graphs, navigation, and vehicle routing problems

The following conjecture made by Gyarfas in 1973 is of interest.

For any integer k, there exists an integer g(k) such that every graph with chi(G) geq g(k) contains either a K_k or an odd hole of length at least 5.

In 1996, in a rather short and elegant proof, Scott provided a partial answer to the above conjecture. In addition, the result of Scott also provides a partial answer to another conjecture put forth by Sumner in 1981. Recently, Chudnovsky, Robertson, Seymour, and Thomas have obtained additional advancement regarding the conjecture of Gyarfas.

A discussion as to the result of Scott is offered.

Adviser: Luis Gravano and Panagiotis (Panos) Ipeirotis

Abstract: The vast majority of Internet users utilize search functionality to navigate the text and text-annotated collections of a variety of web sites. Users of sites such as the New York Times archive, YouTube, and others often face long lists of results for their queries due to the large size of the collections. Processing numerous items is also a hurdle for "exploratory" users who have no specific query in mind, such as a new shopper in an online store or a researcher accessing a news archive. In this work, we attempt to address this problem. We investigate faceted searching and browsing to provide users with access methods that are useful for discovering the content and the structure of long search results or large collections. Hierarchies that organize items based on their topics are common for browsing a large set of items. For example, Yahoo! uses a topic-based hierarchy to guide users to their web pages of interest. Google News and Newsblaster enable news readers to quickly navigate the daily news based on a hierarchy of topics and related events. We first present summarization-aware topic faceted searching and browsing, which integrates clustering and summarization, so that users can browse a list of summarized clusters in the query results instead of individual documents. We have built a fully functional summarization-aware system for daily news. In addition to the topic facet, time can be used as an alternative facet for browsing search results. We explore time as an important dimension and suggest a general framework for time-based language models to consider time in the retrieval task. In fact, many facets, other than topic and time, can be useful for faceted searching and browsing. As a result, we propose supervised and unsupervised methods to identify and extract multiple relevant facets from collections. Yet incorporating such facets in searching or browsing is not an easy task. A typical approach to utilize facets in searching and browsing is to build individual hierarchies for each facet. Unfortunately, these hierarchies are currently manually or semi-manually constructed and populated. This prevents deploying such hierarchies for large collections due to the cost of manually annotating each item in the collections. To solve this problem, we propose a system to automate the construction of hierarchies for the extracted facets, and corresponding human studies to verify the effectiveness of our methods. We apply the faceted hierarchies to a range of large data sets, including collections of annotated images, television programming schedules, and web pages.

Computing the Fourier transform is a basic building block used in numerous applications. For data intensive applications, even the $O(Nlog N)$ running time of the Fast Fourier Transform (FFT) algorithm may be too slow, and sub-linear running time is necessary. Clearly, outputting the entire Fourier transform in sub-linear time is infeasible; nevertheless, in many applications it suffices to find only the significant Fourier coefficients, that is, the Fourier coefficients whose magnitude is at least a $tau$-fraction (say, 1%) of the $ell_2$-norm of the entire Fourier transform (ie, the sum of squared Fourier coefficients). In this paper we present an algorithm that finds the significant Fourier coefficients of functions $f$ over any finite abelian group $G$, which is: {bf Local.} The running time and number of function entries read by the algorithm is polynomial in $log N$, $1/tau$ and $L_1(f)$ (for $N=card G$ and $L_1(f)$ denoting the sum of absolute values of the Fourier coefficients of $f$). {bf Input-oblivious.} The set of entries read by the algorithm depends only on the domain $G$ and on an upper bound on the sum of Fourier coefficients $L_1(f)$, and not on the input function $f$ itself. That is, the {em same fixed} set of entries is read for all functions over the same domain and with the same upper bound on their sum of Fourier coefficients. {bf Robust to noise.} The algorithm finds the significant Fourier coefficients of $f$, even if the function entries it receives are corrupted by random noise. Furthermore, we present extensions of this algorithm to handle {em adversarial noise} in running time {em sub-linear} in the domain size. Our algorithm improves on the prior input-oblivious algorithms in: (1) handling functions over any finite abelian group, (2) being robust to noise, and (3) achieving {better running time in terms of $L_1(f)$.

We present applications of our algorithm to compressed sensing, to solving noisy variants of the Hidden Number Problem with advice, and to decoding polynomial rate Homomorphism codes and polynomial rate Multiplication codes.

An error correcting code is said to be locally testable if there is a test that can check whether a given string is a codeword of the code, or rather far from the code, by reading only a constant number of symbols of the string. Locally Testable Codes (LTCs) were first explicitly studied by Goldreich and Sudan (J. ACM 53(4)) and since then several constructions of LTCs were suggested.

While the best known construction of LTCs achieves very efficient parameters, it relies heavily on algebraic tools and on PCP machinery. We present a new and arguably simpler construction of LTCs that is purely combinatorial and does not rely on PCP machinery. Finally, our construction matches the parameters of the best known construction.

The rapid accumulation of genetic material (e.g. sequencing of genes and genomes) and the prediction of new proteins should help us to better understand evolution at the molecular level. This task, however, is not trivial. First, the evolution of many organisms includes operations such as horizontal transfers (i.e. events of transferring genetic material from one lineage in the evolutionary tree to a different lineage) and rearrangements of genetic material. Thus, it requires accurate modeling of complex biological phenomena. Second, as most of the problems in the field are NP-hard, it requires designing sophisticated algorithms and heuristics for accurate and fast inference of evolutionary models. In this talk I will describe our recent attempts to address these challenges. I will discuss hardness issues related to inferring phylogenetic trees and networks, and will describe a few approaches for modeling and inferring evolution in the presence of horizontal gene transfer, partial horizontal gene transfer, and rearrangements of genetic material. No background in biology will be assumed.

I will present two results due to Hartmanis and Hopcroft.

Let F be any formal system for proving theorems. We assume that F is axiomatizable, that F is consistent, and that F is of sufficient power to prove basic theorems of set theory.

Theorem 1: For every F we can effectively construct an i such that $\phi_i$ is recursive and $P^{\phi_i}=NP^{\phi_i}$ can neither be proved nor disproved in F.

Theorem 2: There exist an algorithm which can be explicitly given whose running time is $n^2$ but there is no proof in F that it runs in time $<2^n$

This work presents research in the field of machine learning and transportation studies. Urban Traffic Control System (UTCS), which rely on a network of sensors, aim to describe real-time traffic in urban areas using a set of parameters and estimating them. Though the state of the art systems focus on data analysis, little has been done to date in terms of prediction. In this work, we describe a machine learning system for traffic flow management and control to address the problem of traffic regulation. This new algorithm is obtained by combining a Random Forests algorithm with an Adaboost algorithm as a weak learner. We show that o! ur algorithm performs relatively well on real data, and enables, according to the traffic flow evaluation model, to estimate and predict whether there is congestion or not at a given time at a given road intersection. The problem of avoiding traffic congestion can be solved by updating the signal timing of the traffic lights at an intersection in advance, based on the prediction of heavy traffic.

In the problem of secure computation, two or more parties wish to compute a function of their private inputs while maintaining several security properties, such as independence of their inputs, privacy, correctness of the computation and others. General protocols for computing all of the above properties, and others, are well known. An exception is the emph{fairness} property, which guarantees that if one party receives output, then all parties receive output. Fairness is only generally achievable if a strict majority of the players are honest. Indeed, it was proven by Cleve cite{Cleve86} in 1986 that it is impossible for two parties to flip a fair coin.

In this talk, I will first present a recent surprising result (STOC 08) that demonstrates the possibility of computing certain particular functions of interest with complete fairness. I will then introduce a new (weaker) security definition called "partial fairness", and present a general feasibility results in this model.

Abstract: The speaker works in requirements engineering (RE) and has not done any work on aspect orientation, cross-cutting cocerns, or early aspects. Nevertheless, RE is certainly early. The speaker believes that in RE, one never sees any cross-cutting concerns, but there are plenty of concerns and these are all requirements. The speaker is of the view that, in the early stages, requirements are not well enough structured that any concern can be identified as cross cutting as opposed to not cross cutting. Everything is just a concern. Nevertheless, there will certainly be cross-cutting concerns later when the required system is built.

---

Bio: Daniel M. Berry got his Ph.D. in Computer Science from Brown University in 1974. He was on the faculty of the Computer Science Department at the University of California, Los Angeles, USA from 1972 until 1987. He was in the Computer Science Faculty at the Technion, Israel from 1987 until 1999. From 1990 until 1994, he worked for half of each year at the Software Engineering Institute at Carnegie Mellon University, USA, where he was part of a group that built CMU's Master of Software Engineering program. During the 1998-1999 academic year, he visited the Computer Systems Group at the University of Waterloo in Waterloo, Ontario, Canada. In 1999, Berry moved to what is now the Cheriton School of Computer Science at the University of Waterloo. Berry's current research interests are software engineering in general, and requirements engineering and electronic publishing in the specific.

Abstract: This talk describes grounded analysis (GA) as a method to discover grounded theories (GTs) to be subjected to later empirical validation. It shows that a good instance of Requirements Engineering (RE) is an instance of GA for the purpose of discovering the artifacts that RE produces. Therefore, these artifacts are also GTs.

Work with Michael W. Godfrey, Ric Holt, Cory J. Kapser, and Isabel Ramos

Bio: Daniel M. Berry got his Ph.D. in Computer Science from Brown University in 1974. He was on the faculty of the Computer Science Department at the University of California, Los Angeles, USA from 1972 until 1987. He was in the Computer Science Faculty at the Technion, Israel from 1987 until 1999. From 1990 until 1994, he worked for half of each year at the Software Engineering Institute at Carnegie Mellon University, USA, where he was part of a group that built CMU's Master of Software Engineering program. During the 1998-1999 academic year, he visited the Computer Systems Group at the University of Waterloo in Waterloo, Ontario, Canada. In 1999, Berry moved to what is now the Cheriton School of Computer Science at the University of Waterloo. Berry's current research interests are software engineering in general, and requirements engineering and electronic publishing in the specific.

This talk should (hopefully) be of interest to people working in distributed systems, distributed CSPs, and multi-agent systems. Loosely coupled multi-agent systems are perceived as easier to plan for because they require less coordination between agent sub-plans. In this paper we set out to formalize this intuition. We establish an upper bound on the complexity of multi-agent planning problems that depends exponentially on two parameters quantifying the level of agents' coupling, and on these parameters only. The first parameter is problem-independent, and it measures the inherent level of coupling within the system. The second is problem-specific and it has to do with the minmax number of action-commitments PER AGENT required to solve the problem. Most importantly, the direct dependence on the number of agents, on the overall size of the problem, and on the length of the agents' plans, is only polynomial.

This result is obtained using a new algorithmic methodology which we call ``planning as CSP+planning''. We believe this to be one of the first formal results to both quantify the notion of agents' coupling and to demonstrate a tractable planning algorithm for fixed coupling levels. Joint work with Carmel Domshlak from the Technion.

This talk will describe some of the challenges currently addressed by Web Information Retrieval systems, namely search engines. The first generation of search engines closely resembled classic IR systems in their goal to return relevant documents to a user's query. Those systems relied mostly on on-page contents. The second generation of search engines added the links between Web pages to the mix, and differentiated between several types of information needs. Today's engines tap huge amounts of user generated content, and focus on helping users to complete tasks. This involves both assisting users to break down their tasks into sub-tasks, as well as interpretation, aggregation and integration of diverse content that allows users to more readily digest complex information. The talk will explore some of the search assistance tools available today and will highlight some of the technical challenges in their development.

We develop a novel approach to the classical problem of learning regular languages from labeled samples. Rather than attempting to construct small consistent automata (long known to be a very difficult computational problem), we embed the strings in a Hilbert space and compute a maximum-margin hyperplane, which becomes our classifier for new strings. We accomplish this via a universal kernel that renders all regular languages linearly separable. Under this kernel, the image of every regular language is linearly separable from its complement in some finite-dimensional space with a strictly positive margin. Thus, we are able to efficiently (in sample size) compute a maximum-margin separating hyperplane (via SVM, for example) and use margin bounds to control the generalization error.

A brute-force computation of this universal kernel has super-exponential complexity. We conjecture that this problem is intractable (a likely candidate for $#$P-complete). However, we propose a simple randomized scheme for efficiently obtaining an $eps$-approximation to our universal kernel. We show that the approximate kernel preserves the distances and margins with low distortion, and therefore may be used as a surrogate for the original one.

To our knowledge, the framework we propose is the only one capable of inducing unrestricted regular languages from labeled samples (modulo standard cryptographic limitations). Along the way, we touch upon several fundamental questions in complexity, automata, and machine learning.

Join work with Boaz Nadler.

Secret-sharing schemes are an important tool used in many cryptographic protocols. In these schemes, a dealer holding a secret string distributes shares to the parties such that only pre-defined authorized subsets of participants can reconstruct the secret from their shares. The collection of pre-defined authorized subsets is called an access structure.

In this talk we will discuss ideal secret sharing schemes, that is, schemes in which the shares are taken from the same domain as the secrets (the domain of shares cannot be smaller). Brickell and Davenport have shown an interesting connection between ideal secret sharing schemes and matroids – a combinatorial structure that generalizes the linear independence. They give a necessary condition for an access structure to have an ideal secret sharing scheme – the access structure must be induced by a matroid. Seymour has proved that the necessary condition is not sufficient: There exists an access structure induced by the Vamos matroid that does not have an ideal scheme. We strengthen the result of Seymour by showing that the access structure induced by the Vamos matroid does not have a scheme that is close to being ideal. Our bounds are obtained by using non-Shannon inequalities for the entropy function.

The talk is based on joint works with Carles Padro and Noam Livne.

Execution traces of software programs are sequences of events corresponding to the execution of low level procedures. Understanding the process that generated these traces can provide helpful insights as to the cause of failures, or the impact of program and user behavior on a service. An hierarchical segmentation of these events can provide the information needed for the understanding of the application behavior.

If time permits, I will also discuss another project: Title: Searching Large Indexes on Tiny Devices: Extending Binary Search with Character Pinning Abstract: Motivated by the shrinking size and increasing capacity of mobile devices, we consider the problem of helping a user to find an item of interest in a large sorted list of strings using a limited user interface. We extend the standard binary search tree by allowing characters from the prefix of the current item to be selected (pinned). BIO: I am a researcher at Microsoft Research, working in the Machine Learning and Applied Statistics group. I graduated from the Ben Gurion University, the Computer Science department on 2007, studying under the supervision of Prof. Ronen Brafman and Prof. Eyal Shimony. My main research interests lie in the control of processes, especially under partial observability (POMDPs). I am also interested in Planning, Data Mining, Machine Learning and Recommender Systems.

Polygonal meshes dominate the representations of 3D graphics models due to their compactness and simplicity. Recent advances in design, modeling, and acquisition technologies have simplified the generation of 3D models, which have led to the generation of large 3D models. These models consist of millions of polygons and often exceed the rendering capabilities of advanced graphics hardware. Therefore, there is a need to reduce the complexity of these models to match the hardware's rendering capability, while maintaining their visual appearance. Numerous algorithms have been developed to reduce the complexity of graphics models. These include level-of-detail rendering with multi-resolution hierarchies, occlusion culling, and image-based rendering. View-dependent rendering approaches change the mesh structure at each frame to adapt to the appropriate level of detail. Traditional view-dependent rendering algorithms rely on the CPU to extract a level-of-detail representation. However, within the duration of a single frame, the CPU often fails to extract the frame's geometry. This limitation usually results in unacceptably low frame rates. In this presentation I will present new approaches for interactive view-dependent rendering of large polygonal datasets, which use the advanced features of modern graphics hardware, and free the CPU.

We present a new randomized algorithm for counting, rare-events simulation, uniform sampling on different regions, and approximating the solutions of quite general NP-hard combinatorial (integer) optimization problems. Similar to the existing randomized algorithms the proposed one is based on the MCMC (Gibbs) sampler equipped with importance sampler and it also uses a sequential sampling plan to decompose a ``difficult'' problem into a sequence of ``easy'' ones. The main differences between the existing and the proposed algorithm is that the latter one has a special device, called the ``cloning'' device, which makes our algorithm very fast and accurate. In particular it is well suited for solving problems associated with the Boltzmann distribution, like estimating the partition functions in an Ising model and for sampling random variables uniformly distributed on different convex bodies. We present efficient numerical results, while solving quite general integer and combinatorial optimization problems as well as counting ones, like satisfiability problem and Hamiltonian cycles. Bio: Reuven Rubinstein is a world known expert in stochastic modeling, applied probability and Monte Carlo simulation methods. He has written over 100 articles and published six books with Wiley and Springer. His citation index is in the top 5% among the scientists in OR and applied probability worldwide. Rubinstein spent most of his sabbatical years at top universities in the world including Columbia, Harvard, Michigan and Stanford University. He is a consultant to many leading firms, including IBM, Motorola and NEC. Rubinstein is the inventor of the popular score-function method in simulation analysis and the generic cross-entropy methods for combinatorial optimization and counting. For details on the cross-entropy, which includes over 100 publications, go to http://en.wikipedia.org/wiki/Cross-entropy_method.

Morphological disambiguation is the process of assigning one set of morphological features (e.g. ילד is a singular masculine noun – a child) to each individual word in a text. When the word is ambiguous (e.g. ילד can be alternatively analyzed as a verb – gave birth), a disambiguation procedure based on the word context must be applied (e.g. given the phrase ילד ירח, the noun analysis is more probable). The most common model for unsupervised learning of stochastic processes is Hidden Markov Models (HMM). In this generative model, a given sequence of observed events is considered to be the emitted output of a stochastic process, over a set of states. In order to estimate the parameters of the model, the Expectation Maximization (EM) algorithm of Baum-Welch is applied over the observed emissions. In the first part of this talk (a type of 'end of PhD. event'), I will overview the Hebrew morphological disambiguation problem, presenting a word-based text encoding, which enables EM learning of HMM model for this task. Then, we will discuss the influence of the initial conditions on the learning process, suggesting several methods for initial estimation of syntagmatic and morpho-lexical distributions.

Women who have given birth can testify that the process is not infinitesimally short. Objects are no different than babies in this respect: it takes time to mature a raw memory block into a live object, and during that time computation may occur. Consider a class /D/ that inherits from a class /B/. Then, (in most object oriented (OO) programming languages) the process of construction of a /D/-object includes an invocation of a suitable constructor of /B/, followed by an invocation of a constructor of /D/. What is the status of the object in the course of the evaluation of the constructor of /B/? On one hand, this object cannot be thought of as a mature, ordinary object of class /D/, since /D/'s constructor was not invoked yet. On the other hand, thinking of the object as an instance of class /B/, may lead to surprising results, e.g., in the case that /B/ is an abstract class. Concretely, suppose that /B/'s constructor invokes a dynamically bound member function implemented in both /B/ and /D/. The dominating /thesis/, taken by languages such as Java and C#, is that of /dynamic binding/ within constructors, i.e., /D/'s implementation is executed. The /anti-thesis/ of /static binding,/ taken in languages such as C++, dictates that /B/'s implementation is executed. This research sets its objective in understanding how such ``half-baked'' objects are used in actual programs. In a data set comprising a dozen software collection with over sixty thousand class, we found that although the potential for such a situation is non negligible i.e., there are many constructors that make calls to methods which /may /be overridden in derived classes, this potential is realized scarcely. We found this behavior in less than 1.5% of all constructors, inheriting from less than 0.5% of all constructors. Further, we find that over 80% of these incidents fall into eight "patterns", which can be relatively easily transformed into equivalent code which does not make pre-mature calls to methods. Another similarly undesirable situation occurs when a constructor exposes the self identity to external code, and this external code chooses to call methods overridden in the derived class. Our estimates on the prevalence of this exposition are less accurate due to the complexity of interprocedural dataflow analysis. The resulting estimates are high, but there are indications that it arises from a relatively small number of base constructors. Joint work with Tali Shragai

Many complex domains offer limited information about their exact state and the way actions affect them. There, autonomous agents need to make decisions at the same time that they learn action models and track the state of the domain. This combined problem can be represented within the framework of reinforcement learning in POMDPs, and is known to be computationally difficult.

In this presentation I will describe a new framework for such decision making, learning, and tracking. This framework applies results that we achieved about updating logical formulas (belief states) after deterministic actions. It includes algorithms that represent and update the set of possible action models and world states compactly and tractably. It makes a decision with this set, and updates the set after taking the chosen action. Most importantly, and somewhat surprisingly, the number of actions that our framework takes to achieve a goal is bounded polynomially by the length of an optimal plan in a fully observable, fully known domain, under lax conditions. Finally, our framework leads to a new stochastic-filtering approach that has better accuracy than previous techniques.

* Joint work with Allen Chang, Hannaneh Hajishirzi, Stuart Russell, Dafna Shahaf, and Afsaneh Shirazi (IJCAI'03,IJCAI'05,AAAI'06,ICAPS'06,IJCAI'07,AAAI'07).

Short Bio Eyal Amir is an Assistant Professor of Computer Science at the University of Illinois at Urbana-Champaign (UIUC) since January 2004. His research includes reasoning, learning, and decision making with logical and probabilistic knowledge, dynamic systems, and commonsense reasoning. Before UIUC he was a postdoctoral researcher at UC Berkeley (2001-2003), and did his Ph.D. on logical reasoning in AI. He received B.Sc. and M.Sc. degrees in mathematics and computer science from Bar-Ilan University, Israel in 1992 and 1994, respectively. Eyal a Fellow of the Center for Advanced Studies and of the Beckman Institute at UIUC (2007-2008), was chosen by IEEE as one of the "10 to watch in AI" (2006), received the NSF CAREER award (2006), and awarded the Arthur L. Samuel award for best Computer Science Ph.D. thesis (2001-2002) at Stanford University.

In this talk, we present on-line algorithms that can be used to explore an unknown polygonal environment by a point robot. Our algorithms compute visibility polygons only from a set of chosen points on the path of a robot, and the criteria for minimizing the cost for robotic exploration is to reduce the number of visibility polygons that the algorithms compute. Efficiency of these algorithms are measured using competitive ratio. We show that these exploration algorithms are also approximation algorithms for the art gallery problem with an additional visibility constraint.

The Fermat-Weber center of an object $Q$ in the plane is a point in the plane, such that the average distance from it to the points in $Q$ is minimal. For an object $Q$ and a point $y$, let $mu_Q(y)$ be the average distance between $y$ and the points in $Q$, that is, $mu_Q(y)

int_{xin{Q}}left|{xy}right|dx/Area(Q)$, where $left|xyright|$ is the Euclidean distance between $x$ and $y$. Let $FWQ$ be a point for which this average distance is minimal, that is, $mu_Q(FWQ) = min_y mu_Q(y)$, and put $mu_Q^*

mu_Q(FWQ)$. The point $FWQ$ is a Fermat-Weber center of $Q$. I will show that for any convex object $Q$ in the plane, the average distance between the Fermat-Weber center of $Q$ and the points in $Q$ is at least $4Delta(Q)/25$, and at most $2Delta(Q)/(3sqrt{3})$, where $Delta(Q)$ is the diameter of $Q$.

Speaker BIO: Dr. Swidan was a fellow in the Technion Excellence program and finished his B.A. in Mathematics at the Technion with Summa Cum Laude. He got his PhD in computer science (bioinformatics) from the Technion. Dr. Swidan was the first post doc to join the newly established Janelia Farms Research campus of the Howard Hughes Medical Institute. Currently, he is the CEO and founder of Olymons (TM), Blessing Machines with Vision (TM), and a fellow in the Edmond J. Safra bioinformatic program at Tel-Aviv University. His research work includes the following topics: image processing and computer vision, combinatorial algorithms and their applications to computational molecular biology, genome rearrangement, comparative genomics, phylogenetic inference, string algorithms, graph algorithms, and patten matching algorithms. Title: Reconstructing repeat-annotated phylogenetic trees Abstract: A new problem in phylogenetic inference is presented, based on recent biological findings indicating a strong association between reversals and repeats. These biological findings are formalized here in a new mathematical model, called repeat-annotated phylogenetic trees (RAPT). We show that, under RAPT, the evolutionary process - including both the tree-topology as well as internal node genome orders - is uniquely determined, a property of major significance both in theory and in practice. Furthermore, the repeats are employed to provide linear-time algorithms for reconstructing both the genomic orders and the phylogeny, which are NP-hard problems under the classical model of sorting by reversals (SBR). For that, a new data structure - namely set-tries - is presented, and is shown to support efficient linear-time insert and find operations. Joint work with Michal Ziv-Ukelson and Ron Pinter. The presentation is self contained.

A conflict-free coloring (CF-coloring) of hypergraph $H=(V,E)$ is a coloring of V such that for any hyperedge e in E there exists a vertex v in e with a unique color in e (i.e., no other vertex of e has the same color as v). The problem was introduced by Even et al. In 2002 and has applications in frequency assignment for cellular networks. We will briefly review the literature. We investigate deterministic online algorithms for the above problem for hypergraphs that arise in geometry (conflict-free coloring collinear points with respect to intervals). The best offline algorithm uses $O(\log n)$ colors, whereas the best online algorithm from Chen et al. uses $O(log^2 n)$ colors. We give the first deterministic online algorithm using $O(\log n)$ colors, in a non-trivial online model. We also provide a tight analysis of the worst-case performance of the greedy algorithm, solving an open problem from Chen et al.

We provide a framework for online CF-coloring any hypergraph in the oblivious adversary model. We use this framework to obtain efficient (order-optimal, i.e. using logarithmic number of colors) randomized online algorithms for CF-coloring some hypergraphs that arise in geometry and model an important version of the frequency assignment task for cellular networks. Our algorithms use fewer random bits and fewer colors than other algorithms in the literature. We also initiate the study of deterministic online CF-coloring with recoloring. The goal is to use few colors, but also resort to recoloring as little as possible. We provide algorithms using $O(\log n)$ colors and doing $O(n)$ recolorings.

Joint work with Amotz Bar-Noy, Svetlana Olonetsky, and Shakhar Smorodinsky

One of the consequences of fast computers, the Internet and inexpensive storage is the widespread collection of data from a variety of sources and of a variety of types. Sources of data include web click streams, financial transactions, and observational science data. Data types include categorical vs. numerical, static vs. dynamic, points in a metric space vs. vertices in a graph. The nagging question often posed about these data sets is: Can we find something interesting that we did not already know? The first answer to this question is often: Let's try clustering the data! Indeed, clustering is one of the most widely used tools for analyzing data sets. Some modern applications of clustering include clustering the web, clustering search results, clustering click streams, customer segmentation, and community discovery in social networks.

Because of its recent ubiquitous applicability, the field of clustering has undergone major revolution over the last few decades characterized by advances in approximation and randomized algorithms, novel formulations of the clustering problem and algorithms for clustering data streams. This mini-course will cover some of these major advances particularly in the context of modern applications.

The vertex coloring and maximal independent set (henceforth, MIS) problems are central problems in the field of distributed algorithms and are intensively studied. Currently there are no known deterministic distributed algorithms that solve these problems in polylogarithmic time on general graphs. Goldberg, Plotkin, and Shannon, STOC'87, devised a logarithmic time algorithm for these problems on planar graphs. We devise a technique based on Nash-Williams forests decomposition that enables to compute efficiently a vertex coloring that uses a small number of colors on graphs with bounded arboricity. Then this coloring is used to compute an MIS in sublogarithmic time on this family of graphs. This family includes planar graphs, graphs with bounded genus, and graphs that exclude fixed minors, and thus our result improves and generalizes the twenty-year old result of Goldberg, Plotkin, and Shannon. We also show a nearly tight lower bound for the time required for computing distributed O(a)-coloring and O(a)-forests-decomposition. Based on joint work with Michael Elkin.

Microarrays have evolved over recent years to address a range of genomic measurement tasks. The development of array based genome wide DNA copy number profiling has opened the door to a variety of other applications.

Manufacturing techniques that enable flexible design of probes and complete arrays in various sizes complete the picture to yield a versatile measurement platform.

I will describe the use of microarrays in several collaborative efforts, including studies of the time of replication in S phase (collaboration with HUJI), studies of chromosomal aberrations in cancer (collaborations with NCI and with Oslo University) and of copy number vaiations (CNVs) in normal populations (collaboration with Harvard). The talk will specifically address some of the computational aspects of flexible designs and applications.

We present a new randomized algorithm for finding a maximum matching in H-minor free graphs. For every fixed H, our algorithm runs in ${O}(n^{3\omega/(\omega+3)}) < O(n^{1.326})$ time, where n is the number of vertices of the input graph and $\omega < 2.376$ is the exponent of matrix multiplication. This improves upon the previous $O(n^{1.5})$ time bound obtained by applying the $O(m{n}^{1/2})$-time algorithm of Micali and Vazirani on this important class of graphs. For graphs with bounded genus, which are special cases of H-minor free graphs, we present a randomized algorithm for finding a maximum matching in ${O}(n^{\omega/2}) < O(n^{1.19})$ time. This extends a previous randomized algorithm of Mucha and Sankowski, having the same running time, that finds a maximum matching in planar graphs. We also present a deterministic algorithm with a running time of $O(n^{1+\omega/2}) < O(n^{2.19})$ for counting the number of perfect matchings in graphs with bounded genus. This algorithm combines the techniques used by the algorithms above with the counting technique of Kasteleyn. Using this algorithm we can also count, within the same running time, the number of T-joins in planar graphs. As special cases, we get algorithms for counting Eulerian subgraphs ($T=\phi$) and odd subgraphs ($T=V$) of planar graphs. The proof uses tools from structural graph theory, computational algebra, and linear algebra. Joint work with Uri Zwick.

Visual texture segregation is the perceptual phenomenon in which perceptually coherent regions can be discriminated solely on the basis of texture information (rather than by their color, luminance, etc…). A central notion in the study of texture segregation is the one of feature gradient (or feature contrast). Indeed, perceptual boundaries in texture stimuli occur where texture features (such as orientation) vary drastically within small spatial distances. However, recent work has shown that salient perceptual singularities occur in visual textures even in the absence of feature gradients. In particular, in smoothly varying orientation-defined textures (ODTs) these non-smooth percepts can be predicted from a differential measure involving two texture curvatures, one tangential and one normal (Ben-Shahar 2006). Based on this recent theory, in this work, we devise a biologically-plausible algorithm for detecting perceptual singularities in images of orientation-defined textures. The model uses a three-layer circuit in which both even-symmetric cells and odd-symmetric cells are used to compute all possible directional derivatives of the dominant orientation, from which the tangential and normal curvatures at each spatial position are selected using non linear shunting inhibition. We present result of this biologically plausible model on ODT images and discuss how this model may be extended to handling general textures and natural images.

Design by contract is a practical methodology for developing code together with its specification. It enhances code quality in several significant ways. It is an integral part of the Eiffel language, but is little used outside that community. A major reason for that is the lack of tools supporting the methodology in other languages. In this talk I will describe several tools that support design by contract in Java. One tool instruments Java programs with their contracts for runtime checking. Another is an Eclipse plugin that refactors programs with their contracts, modifying the contracts as well as using them to check the correctness of the proposed transformations. A third tool attempts to discover draft contracts for existing code, enabling the use of the methodology for code originally developed without contracts.

We show that for every $n$-point metric space $M$ there exists a spanning tree $T$ with unweighted diameter $O(log n)$ and weight $omega(T) = O! (log n) cdot omega(MST(M))$. Moreover, there is a designated point $rt$ such that for every point $v$, $dist_T(rt,v) le (1+epsilon) cdot dist_M(rt,v)$, for an arbitrarily small constant $epsilon > 0$. We extend this result, and provide a tradeoff between unweighted diameter and weight, and prove that this tradeoff is emph{tight up to constant factors} in the entire range of parameters. These results enable us to settle a long-standing open question in Computational Geometry. In STOC'95 Arya et al. devised a construction of Euclidean Spanners with unweighted diameter $O(log n)$ and weight $O(log n) cdot omega(MST(M))$. Ten years later in SODA'05 Agarwal et al. showed that this result is tight up to a factor of $O(log log n)$. We close this gap and show that the result of Arya et al. is tight up to constant factors. Joint work with Yefim Dinitz and Michael Elkin.

Wireless networks are particularly susceptible to malicious and malfunctioning devices. For example, a malicious device can easily jam the airwaves, disrupting all communication. In this talk, I will present new techniques for overcoming network disruptions in wireless networks, specifically in the context of multi-channel networks. In order to provide some intuition as to the challenges posed by malicious disruption, I will begin by demonstrating a lower bound for oblivious gossip algorithms. (Underlying the lower bound proof lies an interesting connection between epsilon-gossip and extremal graph theory.) I will then present an adaptive algorithm that improves on the lower bound, relying on a new combinatorial tool, the multiselector (which, as a natural generalization of a selector, we believe to be of potentially independent interest). Finally, I will present a randomized algorithm that can tolerate even more severe forms of malicious misbehavior. Joint work with Shlomi Dolev, Rachid Guerraoui, Darek Kowalski and Calvin Newport

An out-tree T in a digraph D is subgraph of D which is an orientation of a tree that has only one vertex of in-degree 0 (root). A vertex of T is a leaf if it has out-degree 0. A spanning out-tree is called an out-branching. We overview the following recent results:

• the problem of finding an out-branching with minimum number of leaves is polynomial-time solvable for acyclic digraphs (it is NP-hard for general digraphs) [Gutin, Kim, Razgon, 2008]
• the problem of finding an out-branching with at least k non-leaves is fixed-parameter tractable (FPT) [Gutin, Kim, Razgon, 2008]
• the problem of finding an out-tree with at least k leaves is FPT [Alon, Fomin, Gutin, Krivelevich, Saurabh, 2007]
• the problem of finding an out-branching with at least k leaves is FPT [Bonsma and Dorn, 2007]

The Internet is an indispensable part of our society, and yet its basic foundations remain vulnerable to simple attacks. In recent years, the networking community has considered a variety of proposals for securing Internet routing; path-quality monitoring is a crucial component of many of these proposals.

In this talk, we give a rigorous treatment of the path-quality monitoring problem. We consider protocols that allow a router to robustly raise an alarm when packet loss and delay exceeds some threshold, even when an adversary tries to bias monitoring results. Despite the strong threat model we considered, we develop secure protocols that efficient enough to be deployed in the highly constrained environment of high-speed routers. We present a simple protocol that combines sketching techniques with ideas from cryptography and requires O(log T) storage in order to monitor T packets sent on an Internet data path; e.g., monitoring billions of packets requires only 200-600 bytes of storage and a single IP packet of communication. We then show how to compose instances of this protocol to obtain a protocol that localizes faulty or malfunctioning links on a data path.

This is joint work with David Xiao, Eran Tromer, Boaz Barak, and Jennifer Rexford.

Sensors are small, low-cost resource limited processors with sensing abilities. Sensor networks are wireless ad-hoc networks of sensors collaborating on a mission. In this talk I will briefly describe the first two parts of my thesis and elaborate on the third one. The first part suggests a polygonal based model as an ad-hoc infrastructure for a implementing basic tasks in sensor networks; such as, broadcast, sense of direction and simultaneous activation. The second part suggests a unique-permutation hash function as an efficient implementation of a home location service application for tracking a mobile node.

This application is built on top of a Virtual Stationary Automata (VSA) layer. Finally, I will elaborate on my latest work "Swarm Unit - Reactive k-Secret Sharing". Motivated by the virtual automata abstraction and swarm computing, we investigate an extension of the $k$-secret sharing scheme, in which the secret shares are changed on the fly, independently and without (internal) communication, as a reaction to a global external trigger. The changes are made while maintaining the requirement that $k$ or more secret shares may reconstruct the secret and no $k-1$ or fewer can do so. The application considered is a swarm of mobile processes, each maintaining a share of the secret which may change according to common outside inputs e.g., inputs received by sensors attached to the process. The proposed schemes support addition and removal of processes from the swarm as well as corruption of a small portion of the processes in the swarm. hash function as an efficient implementation of a home location service application for tracking a mobile node.

This talk describes a research project exploring new ways for augmenting search using multiple types and sources of social information. Our goal is to allow searching for all object types such as documents, persons and tags, while also retrieving related objects of all types. To realize this goal, we implemented a social-search engine using a unified approach. In this approach, the search space is expanded to represent heterogeneous information objects that are interrelated by several relation types. Our novel solution is based on multifaceted search and it provides an efficient update mechanism for relations between objects, as well as efficient search over the heterogeneous data. We describe a social search engine positioned within a large enterprise, applied over social data gathered from several Web 2.0 applications. We conducted a large user study with over 600 people to evaluate the contribution of social data for search. Our results demonstrate the high precision of social search results and confirm the strong relationship of users to the topics they were retrieved for. Joint work with Einat Amitay, David Carmel, Nadav Golbandi, Nadav Har'El and Shila Ofek-Koifman

Let $M$ be some finite graph or a finite metric space and assume that some of the automorphisms of $M$ are stored together with the structure of $M$. Let $N$ be some substructure of $N$. A partial automorphism $p$ of $N$ is implemented by an automorphism $f$ of $M$, if the automorphism $f$ extends $p$. For example, an automorphism of $M$ which moves point $a$ in $N$ to point $b$ in $N$ (but may take other points of $N$ out of $N$) implements the partial automorphims $amapsto b$. Consider the following problem: given a structure $N$ and a list $L_0$ of partial automorphisms of $N$, can you find a larger structure $M$ which cotains (a copy of) $N$ as a substructure so that every partial automorphism $p$ in $L_0$ is implemented in $M$. For example, can you find a graph $M$ extending the graph $N$ such that every vertex of $N$ can be moved to any other vertex of $N$ by an automorphisms of $M$ (here the answer is YES, by a theorem of Babai and Sos). The new results I will present assure that in some cases (including the case of the example above) one can find, for given $N$ and $L_0$, a structure $M$ with a list $L_1$ of (full) automorphisms of $M$, such that: (1) the size of $M$ and the size of $L$ are polynomial in the the size of $N$. (2) every partial automorphism in $L_0$ is implemented by some automorphisms from $L_1$. (3) Property (2) is indestructible by partitions: whenever $M$ is partitioned into two disjoint parts, at least one of the parts contains a copy of $N$ for which (2) holds. I will discuss some open problems at the end of the talk. The results are joint with Stefan Geschke.

Parsing a natural language sentence is the task of automatically assigning a (syntactic) structure to a sentence in a non-formal language (such as English or Hebrew). Parsing a sentence is one of the first steps towards understanding it. This talk is an introduction to one (very common) parsing framework, namely parsing with PCFGs (probabilistic CFGs). I'll start by describing the overall framework, and present some early results. Then, I'll describe several extensions and refinements, ending in the current state-of-the-art in parsing technology.

k-CNF formulas with m clauses over n variables show a phase transition phenomenon in the following aspect. There exists $d=d(k,n)$ such that almost all formulas with $m/n>d$ are not satisfiable whereas most formulas with m/n<d are. While random k-CNFs below the threshold received much attention in recent years, above-threshold distributions over satisfiable k-CNFs were far less studied. One possible reason is that it is not clear how to define such distributions in a natural way, while keeping the problem approachable in some sense. In this talk we will survey recent developments in this area. We shall concentrate on three distributions: the planted k-SAT distribution, the uniform distribution over satisfiable k-CNF formulas (in the regime $m/n>d(k,n)$), and an "on-line" version ofthe uniform distribution. In all cases we are able to show that unlike the typically complicated structure of the solution space of below-threshold formulas, above threshold formulas have a simple, basically single-solution structure. We also present some algorithmic ideas that are useful for solving certain clause-density regimes of these distributions.

Based on joint works with: Amin Coja-Oghlan, Uriel Feige, Abraham Flaxman, Michael Krivelevich, and Benjamin Sudakov

Our next guest is Prof. David Harel (Weizmann Institute of Science). Harel is best known for his work on dynamic logic, computability and software engineering. In the 1980s he invented the graphical language of Statecharts, which has been adopted as part of the UML standard. He has also published expository accounts of computer science, such as his award winning 1987 book "Algorithmics: The Spirit of Computing" and has made appearances on Israeli radio and television. He currently works on many diverse topics, including visual languages, graph layout, systems biology and the communication of odors.

Abstract: The talk shows the way software and systems engineering, especially of reactive systems, can be applied beneficially to the life sciences. We will discuss the idea of comprehensive and realistic computerized modeling of biological systems. In comprehensive modeling the main purpose is to understand an entire system in detail, utilizing in the modeling effort all that is known about the system, and to use that understanding to analyze and predict behavior in silico. In realistic modeling the main issue is to model the behavior of actual elements, making possible totally interactive and modifiable realistic executions/simulations that reveal emergent properties. I will address the motivation for such modeling and the philosophy underlying the techniques for carrying it out, as well as the crucial question of when such models are to be deemed valid, or complete. The examples I will present will be from among the biological modeling efforts my group has been involved in: T cell development in the thymus, lymph node behavior, embryonic development of the pancreas, the C. elegans reproductive system, and a generic cell model. I will also discuss a long-term "grand challenge" — to model a full multi-cellular organism.

In this talk we present efficient secure protocols for a variety of tasks, including oblivious transfer, set intersection and pattern matching. Our protocols for securely computing the set intersection functionality are based on secure pseudorandom function evaluations, in contrast to previous protocols that used secure polynomial evaluation. We also use secure pseudorandom function evaluation in order to achieve secure pattern matching. In this case, we utilize specific properties of the Naor-Reingold pseudorandom function in order to achieve high efficiency. Finally, we show that using standard smartcards it is possible to construct truly practical secure protocols, and demonstrate this on the problem of set intersection. We consider a variety of adversary models and definitions of security in our results. Some of our protocols are secure in the presence of malicious adversaries with full simulation (via the ideal/real paradigm), and some provide only privacy. We also present protocols that are secure in the presence of covert adversaries. Loosely speaking, this means that a malicious adversary can cheat, but will then be caught with good probability

A new problem in phylogenetic inference is presented, based on recent biological findings indicating a strong association between reversals and repeats. These biological findings are formalized here in a new mathematical model, called repeat-annotated phylogenetic trees (RAPT). We show that, under RAPT, the evolutionary process - including both the tree-topology as well as internal node genome orders - is uniquely determined, a property of major significance both in theory and in practice. Furthermore, the repeats are employed to provide linear-time algorithms for reconstructing both the genomic orders and the phylogeny, which are NP-hard problems under the classical model of sorting by reversals (SBR). For that, a new data structure - namely set-tries - is presented, and is shown to support efficient linear-time insert and find operations. Joint work with Michal Ziv-Ukelson and Ron Pinter. The presentation is self contained.

This will be a kind of an erasure coding tutorial for the layman. I will give a (very) brief survey of erasure coding, and will then concentrate on the more interesting case of rateless codes. I will present one of the first (if not THE first) works in this field, LT codes, by Michael Luby. As I am not an expert on this subject, I'll not present the finer details and proofs, but the general idea (and the more interesting points).

This will be a kind of an erasure coding tutorial for the layman. I will give a (very) brief survey of erasure coding, and will then concentrate on the more interesting case of rateless codes. I will present one of the first (if not THE first) works in this field, LT codes, by Michael Luby. As I am not an expert on this subject, I'll not present the finer details and proofs, but the general idea (and the more interesting points).

Precise control of gene activation lies at the heart of nearly all biological processes. However, despite enormous advances in understanding this process from both experimental and theoretical perspectives, we are still missing a quantitative description of the underlying transcriptional control mechanisms, and the remaining questions, such as how DNA sequence elements 'compute' expression from the inputs they receive, are still very basic. In this talk, I will present our progress towards the ultimate goal of developing integrated quantitative models for transcription regulation. I will describe a novel thermodynamic model that computes gene activation patterns as a function of DNA sequence and show that this model accurately predicts gene activation patterns in fly from DNA sequence alone.

We present a new method for inferring complexity properties for imperative programs with bounded loops. The properties handled are: polynomial (or linear) boundedness of computed values, as a function of the input; and similarly for the running time.

We overcome the classic undecidability obstacle by defining a ``core" programming language that is Turing-incomplete, but strong enough to model real programs of interest. For this language, our method is the first to give a certain answer; in other words, our inference is both sound and complete. This improves on previous approaches that yield sound, but always incomplete, conclusions. Further, analyzing a program takes, itself, polynomial time.

This is joint work with Neil Jones (Copenhagen) and Lars Kristiansen (Oslo).

In recent years there has been a growing interest in bridging Game Theory and Cryptography. While the standard Cryptographic settings view players as either totally honest or arbitrarily malicious, Game Theory assumes that players are rational, selfish individuals. Designing protocols for such settings poses new challenges. We investigate rational versions of two classical problems in foundations of cryptography: secret sharing and function evaluation. We show that schemes for these tasks, in which players' values come from a bounded domain, cannot satisfy some of the most desirable properties. In contrast, we suggest protocols for rational secret sharing where the shares come from an unbounded domain, but have a finite expectation.

The goal of private data analysis is to release aggregate statistics about a dataset while protecting the privacy of individuals whose information is contained in the dataset. A popular technique for privacy is output perturbation where a small amount of random noise is added to the released statistics, carefully crafted to protect individual privacy.

We introduce a new framework that expands the applicability of output perturbation. Departing from previous works, the noise magnitude in out framework is determined not only by the function $F()$ we want to release, but also by the content of the dataset itself. One of the challenges is to ensure that the noise magnitude does not become a source of information leakage by itself. For that we introduce a quantity that we call smooth sensitivity. This is a measure of the variability of $F()$ in the neighborhood of the particular instance dataset $x$. To our knowledge, this is the first formal analysis of the effect of instance-based noise in the context of data privacy.

Our framework raises many interesting algorithmic questions. Namely, to apply the framework one must compute or approximate the smooth sensitivity of $F(x)$ on particular dataset instances. We show how to do this efficiently for several functions, including the median and the cost of the minimum spanning tree.

Finally, we also give a generic procedure based on sampling that allows one to release $F(x)$ accurately on many well behaved datasets x. This procedure is applicable even when no efficient algorithm for approximating smooth sensitivity of $F()$ is known or when $F()$ is given as a black box. We illustrate the procedure by applying k-means clustering and learning mixtures of Gaussians.

The talk with be self contained. Joint work with Sofya Raskhodnikova and Adam Smith. STOC 2007.

Polygonal meshes dominate the representations of 3D graphics models due to their compactness and simplicity. Recent advances in design, modeling, and acquisition technologies have simplified the generation of 3D models, which have led to the generation of large 3D models. These models consist of millions of polygons and often exceed the rendering capabilities of advanced graphics hardware. Therefore, there is a need to reduce the complexity of these models to match the hardware's rendering capability, while maintaining their visual appearance. Numerous algorithms have been developed to reduce the complexity of graphics models. These include level-of-detail rendering with multi-resolution hierarchies, occlusion culling, and image-based rendering.

View-dependent rendering approaches change the mesh structure at each frame to adapt to the appropriate level of detail. Traditional view-dependent rendering algorithms rely on the CPU to extract a level-of-detail representation. However, within the duration of a single frame, the CPU often fails to extract the frame's geometry. This limitation usually results in unacceptably low frame rates. In this presentation I will present new approaches for interactive view-dependent rendering of large polygonal datasets, which use the advanced features of modern graphics hardware, and free the CPU.

In many application areas, complex data sets are often represented by some metric space and metric embedding is used to provide a more structured representation of the data. In many of these applications much greater emphasis is put on the preserving the local structure of the original space than on maintaining its complete structure. In this paper we initiate the study of local embeddings of metric spaces and provide embeddings with distortion depending solely on the local structure of the space.

Joint work with Ittai Abraham and Yair Bartal

In this work we combine two directions in the field of privacy. While in both the goal is to privately evaluate some function $f$ on $n$ inputs, the two approaches differ in their privacy goals and requirements. The first direction is secure function evaluation (SFE), where $n$ parties, sharing a common interest in distributively and securely computing $f$ applied to their private inputs, are given a protocol to do so. An SFE protocol for evaluating $f$ is private if no subset of the parties may deduce information, other than what can be learned from the result of $f$ and the initial inputs of the members of the subset. The second direction is private data analysis where computation of $f$ is considered to be private if the privacy of each record is preserved individually. One possible way to obtain useful information on a collection of individual information, while protecting individual data, is by adding carefully chosen ``noise'', i.e., by evaluating a random approximation $widehat{f}$ of $f$. This choice of $widehat{f}$ is usually done while abstracting away implementation issues, assuming that the computation is performed by a trusted server holding the data.

A natural paradigm for implementing the computation without a trusted party is by constructing an SFE protocol for $widehat{f}$. This conceptual separation, between the decision on which $widehat{f}$ to compute and how to compute it, is valuable. However, this approach may result in non-optimal protocols, as SFE protocols need to comply with strong privacy requirements. For instance, publishing outputs of intermediate computations, as long as they do not compromise individual privacy, may be useful. However, publishing these outputs may be impossible under SFE definitions (e.g., if for some party these outputs cannot be simulated from the input of the party and the output of the function).

We initiate an examination whether there are advantages to choosing $widehat{f}$ and the protocol for computing it simultaneously. In particular, we investigate the case of sum queries and specify under which accuracy requirements, it is beneficial to adapt this paradigm.

Joint work with Amos Beimel and Kobbi Nissim.

Until recently, the study of gene regulation has focused mostly on two major mechanisms: the regulation of transcription (typically described in terms of the binding to transcription factors to specific DNA sequence motifs) and on direct protein-protein interactions. A string of high-throuhput experiments, however, has resulted in overwhelming evidence that several other layers of regulation are of comparable importance and cannot be neglected. Indeed, at least half of the diversity of the observable transcriptome is comprised by non-coding RNAs, and novel types and classes of ncRNAs keep being discovered. The oldest and best known, but as it seems by no means the only one of these RNA-mediated mechanism is post-transcriptional regulation through microRNAs. Anti-sense transcription is another one of the better-known effects.

In my presentation I will briefly review recent findings, both experimental and computational, that suggest that a substantial fraction of transcripts has regulatory function. Then I will focus on bioinformatics approaches to disentangle - at least partially - the various layers using two systems as examples: direct RNA-RNA and RNA-protein interactions. Because of their abundance (even though we understand very few specific systems in detail), we cannot ignore these multiple regulatory layers in attempts to reconstruct and eventually understand gene regulation networks.

Tools that analyze and modify binary code are crucial to many areas of computer science, including cyber forensics, program tracing, debugging, testing, performance profiling, performance modeling, and software engineering. Many tools used to support these activities have significant limitations in functionality, efficiency, accuracy, portability and availability.

Our goal is the design and implementation of a new framework that will allow for interoperability of the static and dynamic code modification, and enable the sharing and leveraging of this complex technology.

Characteristics of this framework include: multi-architecture, multi-format, and multi-operating system; library-based; open source; extensible data structures; exportable data structures; batch enabled; testable, with each separate component provided with a detailed test suite; and functions with non-contiguous and shared code.

This component-based approach requires identifying key portable and multi-platform abstractions at every level of functionality. Transitioning to this model successfully will enable us to break the "hero model" of tool development of having each group trying to produce its own end-to-end complete tool sets.

A polychromatic $k$ - coloring of a plane graph $G$ is an assignment of k colors to the vertices of $G$, such that each face of $G$, except possibly for the outer face, has all $k$ colors on its boundary. Polychromatic colorings of plane graphs are an essential tool for guarding and covering problems. In a polychromatic coloring of a plane graph $G$ every color class is present on every face of $G$. Therefore, any color class forms a set of vertex guards for the faces of $G$.

We survey recent results concerning polychromatic colorings, and prove that every 2-connected plane bipartite cubic graph admits a polychromatic $4$-coloring.

In previous work on mediation and negotiation support systems, we have focused upon using integrative bargaining. When end-users of our family-mediation and plea-bargaining systems have evaluated our systems, they commented that the systems met their personal needs, but not the interests of fairness or justice. For example, in Family Mediation, they meet the interests of the parents and not the children.

For negotiation support systems to be widely used, issues of fairness and justice must be addressed. We are currently developing measures for assessing the outcomes of online negotiation in the legal domains of sentencing, plea-bargaining and family mediation. Such methods will form the basis of a new model for evaluating fairness and consistency within online dispute resolution systems. This model will inform the construction of fairer and more consistent systems of IT based negotiation support.

We conclude by discussing a new project we are about to commence focusing upon developing negotiation support systems that promote constructive relationships following disputes.

An $(n,k)$-affine source over a finite field F is a random variable $X=(X_1,...,X_n) \in F^n$, which is uniformly distributed over an (unknown) k-dimensional affine subspace of $F^n$. We show how to (deterministically) extract practically all the randomness from affine sources, for any field of size larger than $n^c$ (where $c$ is a large enough constant).

Our main results are as follows: (For arbitrary k): For any n,k and any F of size larger than $n^{20}$, we give an explicit construction for a function $D: F^n \rightarrow F^{k-1}$, such that for any $(n,k)$-affine source X over F, the distribution of $D(X)$ is $\epsilon$-close to uniform, where $\epsilon$ is polynomially small in $|F|$. (For k=1): For any n and any F of size larger than $n^{c}$, we give an explicit construction for a function $D: F^n \rightarrow \{0,1\}^{(1-\delta) \log_2|F|}$, such that for any $(n,1)$-affine source X over F, the distribution of $D(X)$ is $\epsilon$-close to uniform, where $\epsilon$ is polynomially small in $|F|$. Here, $\delta > 0$ is an arbitrary small constant, and $c$ is a constant depending on $\delta$.

Joint work with Ran Raz

Precise control of gene activation lies at the heart of nearly all biological processes. However, despite enormous advances in understanding this process from both experimental and theoretical perspectives, we are still missing a quantitative description of the underlying transcriptional control mechanisms, and the remaining questions, such as how DNA sequence elements 'compute' expression from the inputs they receive, are still very basic. In this talk, I will present our progress towards the ultimate goal of developing integrated quantitative models for transcription regulation. I will describe a novel thermodynamic model that computes gene activation patterns as a function of DNA sequence and show that this model accurately predicts gene activation patterns in fly from DNA sequence alone.

We often like to build a model for one scenario based on data from similar or nearby cases. For example, consider the problem of building a model which predicts a sentiment about books from short reviews, using reviews and sentiment of DVDs. Another example is of learning movies preference for one viewer from ratings provided by other similar users. There is a natural tradeoff between using data from more users and using data from only similar users.

In this talk, I will discuss the problem of learning good models using data from multiple related or similar sources. I will present a theoretical approach which extends the standard probably approximately correct (PAC) learning framework, and show how it can be applied in order to determine which sources of data should be used and how. The bounds explicitly model the inherit tradeoff between building a model from many but inaccurate data sources or building it from a few accurate data sources. The theory shows that optimal combinations of sources can improve performance bounds on some tasks.

Random numbers are essential in every cryptographic protocol. The quality of a system's random number generator (RNG) is therefore vital to its security. In Microsoft Windows, the operating system provides an RNG for security purposes, through an API function named CryptGenRandom. This is the cryptographic RNG used by the operating system itself and by important applications like the Internet Explorer, and the only RNG recommended for security purposes on Windows. This is the most common security RNG in the world, yet its exact algorithm was never published until now. We provide a description of the Windows RNG, based on examining the binary code of Windows 2000. We reconstructed the RNG's algorithm, and present its exact description and an analysis of the design. Our analysis shows a number of weaknesses in the design and implementation, and we demonstrate practical attacks on the RNG. We propose our recommendations for users and implementers of RNGs on the Windows platform. In addition, we describe the reverse-engineering process which led to our findings.

Live Distributed Objects Prof. Kenneth P. Birman Abstract Although we've been building distributed systems for decades, it remains remarkably difficult to get them right. Distributed software is hard to design and the tools available to developers have lagged far behind the options for building and debugging non-distributed programs targeting desktop environments. At Cornell, we're trying to change this dynamic. The first part of this talk will describe "Live Distributed Objects", a new and remarkably easy way to create distributed applications, with little or no programming required (in fact, the talk will include a demo of how this works). Supporting these kinds of objects forced us to confront a number of scalability, security and performance questions not addressed by prior research on distributed computing platforms. The second part will look at Cornell's Quicksilver system and the approach it uses to solve these problems

This research is joint with PhD candidate Krzys Ostrowski (the "real" leader on the effort) and with Danny Dolev.

Abstract Machine learning algorithms are often used to extract rules and structure from large datasets, and have been successfully used in fields such as machine vision, natural language processing and computational biology. With the growing availability of text and images on the web, and high-throughput experiments in biology, learning algorithms are becoming a key tool for organizing, searching, and interpreting this data.

Learning algorithms are typically required to work with very large datasets of high dimensional signals. Thus scalability of algorithms is a key issue that must be addressed. In this talk I will show how the concept of convex duality can be used to design simple and scalable algorithms, and to help understand convergence of previously existing ones.

I will first address the problem of probabilistic inference in multivariate distributions, and show how convex duality results in simple convergent message passing algorithms for this problem. Specifically, I will show that approximate inference may be cast as a geometric program, where coordinate descent yields message passing updates. These results resolve a long standing problem regarding the convergence of message passing algorithms for approximate inference.

I will next address the problem of learning classifiers from labeled data, and present an exponentiated gradient algorithm that is also derived using convex duality. The algorithm can be shown to converge, and its convergence rate can be analyzed. I will conclude by showing an application of the algorithm to a large scale natural language parsing task, and demonstrate that it converges significantly faster than previous state of the art algorithms.

Abstract Worst-case complexity, which measures the performance of algorithms with respect to the most difficult instances of the problem at hand, is a standard and convenient complexity measure. On the other hand, average-case complexity, which measures the performance of algorithms with respect to typical (or simply random) instances, may be a more realistic complexity measure for instances that actually appear in practice. Furthermore, it seems to be the right measure in settings where computational hardness is beneficial (e.g. in cryptography and pseudo-randomness).

A central question in theoretical computer science is for which classes of computational problems (and against which families of algorithms) there is an equivalence between their worst-case and average-case complexities.

This question and the techniques to attack it are related to many other areas in computer science such as cryptography, pseudo-randomness, error-correcting codes, program checking and learning.

In this survey talk I will describe some recent results on worst-case/average-case equivalence. I will concentrate on two important classes of computational problems: NP and EXPTIME.

Abstract RNA secondary structure is defined along with three different ways to display them. Two distinct approaches are presented for determining secondary structure from sequence data. The comparative method requires a multiple sequence alignment of a collection of homologous RNA sequences. It uses phylogeny to determine common, conserved base pairs that are more likely to be the result of evolution than to exist by chance. On the other hand, recursive algorithms may be used on single sequences to compute minimum free energy structures, partition functions and other biophysical quantities. These algorithms ignore evolution and use empirically derived energy parameters based on physical chemistry. Examples will be given for both methods.

The latter part of the lecture will include recent work on computing the entropy of the Boltzmann distribution of foldings and how this quantity may be used to judge the overall reliability of free energy based methods for particular molecules.

In this talk we present BP-QL, a novel system for querying and monitoring business processes. The BP-QL query language is based on an intuitive model of business processes, an abstraction of the emerging BPEL (Business Process Execution Language) standard. It allows users to query business processes specifications, as well as their run time behavior, visually, in a manner very analogous to how such processes are typically specified, and can be employed in a distributed setting, where process components may be provided by distinct providers (peers).

We describe here the query language as well as its underlying formal model. We consider the properties of the various language components and explain how they influenced the language design. In particular we distinguish features that can be efficiently supported, and those that incur a prohibitively high cost, or cannot be computed at all. We also present our implementation which complies with real life standards for business process specifications, XML, and Web services, and is used in the BPQL system.

One of the surprising developments in the area of program verification is how several ideas introduced by logicians in the first part of the 20th century ended up yielding at the start of the 21st century an industrial-standard property-specification language called PSL. This development was enabled by the equally unlikely transformation of the mathematical machinery of automata on infinite words, introduced in the early 1960s for second-order arithmetics, into effective algorithms for model-checking tools. This talk attempts to trace the tangled threads of this development.

This research concerns the image deblurring and noise removal problem in a variational framework. Energy functionals in this study consist of a fidelity term and a regularizer that is based on Mumford-Shah segmentation, such that the recovered image and its discontinuities set are simultaneously extracted in the course of the deblurring process. The functionals are formulated using the Gamma-convergence approximation and are iteratively optimized via the alternate minimization method. We first consider the image deblurring problem in the presence of Gaussian/impulsive noise. The suggested approach integrates and extends the robust statistics, anisotropic diffusion and line process (half quadratic) points of view. Then, we extend the deblurring and denoising problem to vectorvalued images. Further, we present the shift-variant deblurring case following by simultaneous motion estimation and restoration of motion-blurred video.

In this talk, I will give an overview of several recent results of mine in computer vision and computational geometry. Within the realm of computer vision, I will focus on applications of partial differential equations (curve and surface flows) and combinatorial optimization to segmentation, tracking, and optical flow. Within computational geometry, I will discuss results on manifold reconstruction from unorganized points, as well as computational algebraic topology, in particular the measurement of homology. These results may be applied to the study of shape.

The field of Augmentative and Alternative Communication (AAC) is concerned with studying methods of communication that can be added to natural communication (speech and writing), especially when an individual lacks some of the skills to achieve it. Natural language processing techniques have been used for AAC applications to enhance the rate of communication and extend the range of expressions that can be generated. The key applications include message generation, abbreviation expansion, word prediction and text simplification. In my talk I will present the world of AAC, the usage of NLP within AAC, and specifically natural language generation systems including the one developed in BGU.

There is an intrinsic connection between computational geometry and networking. In this talk I will describe theoretical results in computational geometry with practical implications in networking. More specifically, I will first present a new construction of geometric spanners that allows optimal routing between points in the plane. In the second part of the talk I will describe the first construction ever of a disk graph spanner. The practical importance of the later is enormous as it can serve as a topology for ad-hoc networks whose nodes have a *variable* transmission range.

The talk is based on the following papers:

1. Improved Algorithms for Fully Dynamic Geometric Spanners Lee-Ad Gottlieb (NYU) and Liam Roditty

2. Spanners for Ad Hoc Networks with Variable Transmission Range David Peleg (Weizmann) and Liam Roditty

The most popular object-oriented programming (OOP) languages nowadays - Java, C++, C# - are statically-typed and based on a static model. They do not allow to manipulate their classes at run-time. As a result, many programmers view OOP as a methodology with limited flexibility and expressiveness. OOP languages do not have to be static. Several new scripting languages break this rule by implementing a highly dynamic and flexible version of OOP. These are dynamically-typed languages that allow to manipulate objects and classes at run-time. This ability gives a lot of flexibility to the programmer, and those languages are gaining much popularity.

One of the more impressive and useful usages of this approach is the implementation of a meta-classes system in Python. Meta-classes are classes that create other classes as their instances. Using meta-classes lets programmers implement design patterns and aspect-oriented programming by using the programming language itself, giving a higher level of expressiveness and code re-use. In this talk I will give a short introduction to those new approaches in OOP, and present meta-OOP programming in Python, with its powers and problems.

Revenue from electronic commerce and Internet-based advertising has been growing exponentially in the last decade, hand in hand with the birth of large-scale marketplaces (e.g., eBay and Amazon), information providers ( e.g., Google and Yahoo) and web-based social networks (e.g., MySpace and Facebook). Participants in such systems have access to information that is known only to them (like how much buyers are willing to pay for commodities), and economic mechanisms are designed to gather sufficient information for computing the desired outcomes. Complex communication protocols are often required in order to achieve good results, whereas practical and behavioral reasons encourage the use of simple communication exchanges.

This talk will survey several results characterizing tradeoffs between the simplicity of markets and their achievable economic properties. On the positive side, nearly optimal markets exist even when the agents have restricted expressive power. On the negative side, we show that the communication overhead of computing equilibrium-supporting prices may be significant. Finally, we show that simple pricing schemes cannot support close-to-optimal results in ascending-price multi-unit auctions.

Based on joint works with Moshe Babaioff, Michal Feldman, Moni Naor, Noam Nisan, Michael Schapira and Ilya Segal.

We study the round complexity of various cryptographic protocols. Our main result is a tight lower bound on the round complexity of any fully-black-box construction of a statistically-hiding commitment scheme from one-way permutations, and even from trapdoor permutations. This lower bound matches the round complexity of the statistically-hiding commitment scheme due to Naor, Ostrovsky, Venkatesan and Yung (CRYPTO '92). As a corollary, we derive similar tight lower bounds for several other cryptographic protocols, such as single-server private information retrieval, interactive hashing, and oblivious transfer that guarantees statistical security for one of the parties. Our techniques extend the collision-finding oracle due to Simon (EUROCRYPT '98) to the setting of interactive protocols (our extension also implies an alternative proof for the main property of the original oracle). In addition, we substantially extend the reconstruction paradigm of Gennaro and Trevisan (FOCS `00). In both cases, our extensions are quite delicate and may be found useful in proving additional black-box separation results. Joint work with Jonathan J. Hoch, Gil Segev and Omer Reingold

The Computational Geometry Algorithms Library, CGAL, is the largest software collection of algorithms and data structures in Computational Geometry available today. It started a little over a decade ago as a European research project with a small number of partners and has grown over the years to be a huge open source project. The arrangement package of CGAL, developed at Tel Aviv University, constructs, maintains, traverses, and answers queries on two-dimensional arrangements (subdivisions) of general curves. We will start with a bird's eye view of the overall project, and then briefly present the underlying design principles of the arrangement package. The talk will mostly focus on recent innovations and applications of the arrangement package, including the construction of: general 2D Voronoi diagrams, envelopes of surfaces in three-dimensional space, Boolean set operations for generalized (curved) polygons, and more.

The new components that we will review were developed by Efi Fogel, Michal Meyerovitch, Ophir Setter, Ron Wein, and Baruch Zukerman.

Motivated by battery consumption in sensor networks we study two hypergraph coloring problems: 1. For a hypergraph $H=(V,E)$ (The elements of E are subsets of V) define the parameter $p=p(k)$ to be the minimum number p for which there exists a coloring of V with k colors such that every hyperedge in E with cardinality at least p(k) contains elements from all color classes. If such a coloring does not exist then we say that $p(k) = \infty$. We show that $p(k) < 4k$ when the vertices of H is a finite set P of points in $R^2$ and E consists of all subsets of P that can be obtained by intersecting P with some closed half-plane.

This improves a previous bound of Pach and Toth of $O(k^2)$

We also study the following parameter: 2. For a hypergraph $H = (V,E)$, Let $c = c(k)$ denote the minimum number c such that there exists a coloring of H with c colors such that every hyperedge with cardinality at least k contains elements from some k distinct colors. We show for example that for some geometrically defined hypergraphs, $c(k) = O(k)$ (When H is the hypergraph defined by intersecting a set of points in $R^3$ with all half spaces, the statement $c(2) = 4$ is equivalent to the Four-Color Theorem)

Joint work with G. Aloupis, J. Cardinal, S Collette and S. langerman from Universite Libre de Bruxelles (ULB)

Effective resizing of images should not only use geometric constraints, but consider the image content as well. We present a simple image operator called seam carving that supports content-aware image resizing for both reduction and expansion. A seam is an optimal 8-connected path of pixels on a single image from top to bottom, or left to right, where optimality is defined by an image energy function. By repeatedly carving out or inserting seams in one direction we can change the aspect ratio of an image. By applying these operators in both directions we can retarget the image to a new size. The selection and order of seams protect the content of the image, as defined by the energy function. Seam carving can also be used for image content enhancement and object removal. We support various visual saliency measures for defining the energy of an image, and can also include user input to guide the process. By storing the order of seams in an image we create multi-size images, that are able to continuously change in real time to fit a given size.

Text remains the main medium to convey information and knowledge. Machine-readable text has become plentiful and accessible through search engines. Many applications are concerned with transforming text from one to another. The field that studies such transformation is called text to text generation (T2T). T2T is a sub-field of natural language generation (NLG), and stands in contrast with data to text (D2T) generation (where the input data is any non-textual representation).

In T2T, textual units (sentences, clauses, phrases) are extracted from one context and recombined into a new text. As in NLG in general, one can split the overall task of generation into several steps:

• Content selection
• Content organization
• Realization: the rendering of the selected content into fluent language.

In T2T, "content" is encoded as textual units - generally called "SCU" (Shared Content Unit). The input to a T2T system includes a collection of texts that are related in topic, for example a collection of news reports describing the same event (from different sources or published at different times). Content selection and organization in D2T applications is generally related to knowledge representation and inferencing. In the case of T2T, it is related to Information Extraction, which includes named entity recognition, coreference resolution, entity identification, relation identification and scenario identification.

Realization is the linguistic component of generation, and is organized around the following steps:

• Lexicalization (selection of the words)
• Aggregation
• Referring expression generation
• Rhetorical structuring and ordering
• Centering and salience
• Syntactic realization (ordering of the words, morphological inflection)

In the lecture, I will review sample T2T applications and the corresponding methods. I will focus on the type of knowledge required to perform T2T and how it can be acquired. I will end with a review of open issues and possible topics for further research.

For a function g from natural numbers to natural numbers, a g-regressive coloring of pairs of natural numbers is a coloring C satisfying $C(m,n) \le g(\min{m,n})$ for all $m,n$.

The g-regressive Ramsey number of k is the least N so that for every g-regressive coloring C there exists a subset H of of size at least k which is min-homogeneous for C, namely, the color $C(m,n)$ of a pair ${m,n}$ of H depends only on $\min(m,n)$.

We compute a sharp threshold on g at which g-regressive Ramsey numbers cease to be primitive recursive and become Ackermannian. We prove:

Suppose g is nondecreasing and unbounded. Then the g-regressive Ramsey number is bounded by some primitive recursive function in k if and only if for every $t>0$ there is some $M(t)$ so that for all $n \ge M(t)$, $g(n) < n^{1/t}$ and $M(t)$ is primitive recursive in t.

We also give an extremely large lower bound on the identity-regressive Ramsey number of k=82. Joint work with Menachem Kojman, Gyesik Lee and Andreas Weiermann.

This talk is about the application of Boolean SAT solvers to the problem of determining program termination. Proving termination is all about the search for suitable ranking functions. The key idea in this work is to encode the search for particular forms of ranking functions to Boolean statements which are satisfiable if and only if such ranking functions exist. In this way, proving termination can be performed using a state-of-the-art Boolean satisfaction solver.

We have applied this approach to several variants of the so-called LPO termination problem which comes up in the context of term rewrite systems. Experimental results are unequivocal, indicating orders of magnitude speedups in comparison with previous implementations for LPO termination. Our results have had a direct impact on the design of several major termination analyzers for term rewrite systems. Moreover, there are an increasing number of new results illustrating additional applications of SAT to proving termination.

A function is Turing computable if it can be specified by a Turing machine. Turing's definition [1936] of a function, computable by a Turing machine, was followed in his dissertation [1939], in which he defined the notion of a set $A$ being computable relative to a set $B$. We then say $A$ is Turing reducible to $B$ ($A\leq_T B$). We identify a degree as an equivalence class of sets of natural numbers, each reducible to the other

The least degree is the set of all recursive subsets of $\omega$, denoted $\deg{0}$. A degree $\deg{b}$ is minimal if $\deg{b}>\deg{0}$ and there is no $\deg{c}$ such that $\deg{0}<\deg{c}<\deg{b}$. That is, it is minimal in the ordering of degrees. I will present Spector's result [1956]: There is a minimal degree. It follows that the ordering $\leq$ on $\D$ (the structure of the degrees) is not dense. This result proved to be useful in many embedability results as well as other results concerning the structure of $\D$.

The parallel repetition theorem, proven by Raz in 1995, is a fundamental result that in addition to its philosophical appeal, plays a key role in complexity theory. The theorem studies the behavior of success probabilities of two prover games, when many copies of such a game are played in parallel. It shows that the success probability decreases exponentially in the number of repetitions, but the parameters given by the theorem do not seem tight. It is natural to ask what are the best parameters for which the theorem holds, and improving them would have complexity theoretic implications.

This talk describes an attempt to improve the parameters in a very special case of the parallel repetition theorem. Our attempt had only limited success, but it turns out that the reason we got stuck was that the following seemingly hard question from the geometry of "foams" was hidden in the special case that we were trying to solve: What is the least surface area of a cell that tiles $R^d$ by the lattice $Z^d$? Very little about this foam problem is known. It is interesting to see such a geometric question encoded inside the problem of parallel repetition in two prover games.

The Asynchronous Backtracking algorithm is considered the state of the art algorithm for solving Distributed Constraint Satisfaction Problems. A strong assumption of all previous studies of this algorithm is that a static order is required for its correctness and completeness when polynomial space is used. In the present study, an algorithm that performs asynchronous backtracking on distributed CSPs, with dynamic ordering of agents was presented, ABT_DO. Agents propose reorderings of lower priority agents and send these proposals whenever they send assignment messages. The dynamic ordered asynchronous backtracking algorithm uses polynomial space, similarly to standard ABT. The ABT_DO algorithm with three different ordering heuristics was compared to standard ABT on randomly generated DisCSPs. A Nogood-triggered heuristic,inspired by dynamic backtracking, was found to outperform static order ABT by a large factor in run-time and improve the network load. A further investigation of this successful heuristic shows the relation between it and the Min-Domain property. This relation was brought to an extreme in a later study in which a Retroactive ordering heuristic improved the run of both distributed and centralized algorithms. The improvement on structured realistic problems was much larger than on random problems.

Symmetry and self-similarity is the cornerstone of Nature, exhibiting itself through the shapes of natural creations and omnipresent laws of physics. Since many natural objects are symmetric, the absence of symmetry can often be an indication of some anomaly or abnormal behavior. Therefore, detection of asymmetries is important in numerous practical applications, including crystallography, medical imaging, and face recognition, to mention a few. Conversely, the assumption of underlying shape symmetry can facilitate many problems in shape reconstruction and analysis. Traditionally, symmetries are described as extrinsic geometric properties of the shape. While being adequate for rigid shapes, such a description is inappropriate for non-rigid ones. Extrinsic symmetry can be broken as a result of shape deformations, while its intrinsic symmetry is preserved. We pose the problem of finding intrinsic symmetries of non-rigid shapes and propose an efficient method for their computation.

*Research under the supervision of Prof. Ron Kimmel, in collaboration with Dr. Alex Bronstein and Dr. Michael Bronstein.

The talk will be on Geometric Spanners with Small Chromatic number. Two variants of this problem will be discussed

Variant 1: Given a complete k-partite geometric graph K whose vertex set is a set of n points in $\R^d$, compute a spanner of K that has a "small" stretch factor and "few" edges. We present two algorithms for this problem. The first algorithm computes a $(5+\epsilon)$-spanner of K with $O(n)$ edges in $O(n \log n)$ time. The second algorithm computes a $(3 + \epsilon)$-spanner of $K$ with $O(n log n)$ edges in $O(n \log n)$ time. The latter result is optimal: We show that there exist complete k-partite geometric graphs K such that every subgraph of K with a subquadratic number of edges has stretch factor at least 3.

Variant 2: Given an integer $k > 1$, we consider the problem of computing the smallest real number $t(k)$ such that for each set $P$ of points in the plane, there exists a $t(k)$-spanner for $P$ that has a chromatic number at most $k$. We prove that $t(2) = 3$, $t(3) = 2$, $t(4) = \sqrt{2}$, and give upper and lower bounds on $t(k)$ for $k>4$. We also show that for any $\epsilon >0$, there exists a $(1+\epsilon)t(k)$-spanner for $P$ that has $O(|P|)$ edges and whose chromatic number is at most $k$. We also consider an on-line variant of the problem, in which the points of $P$ are given one after another, and the color of a point must be decided at the moment the point is given.

I will give a survey talk on randomness extractors focusing on "seeded extractors". Randomness extractors can be viewed in two ways: The first is as bipartite graphs with certain "expansion properties". The second is as functions that extract pure randomness from somewhat random sources of randomness (so that this randomness can be used in probabilistic algorithms and protocols). In the talk I will give a brief introduction to the area, show some applications and some explicit constructions

This talk addresses one of the most the most fundamental tasks in Cryptography, that of constructing pseudorandom generators from one-way functions. The seminal paper of Hastad et al. (known as HILL), proved that these notions are equivalent. However, the reduction is not nearly as security preserving as one may desire. In this context, we revisit a technique that we call the Randomized Iterate, introduced by Goldreich et al. In the talk I will introduce the randomized iterate and survey its recent applications. Our results simplify and strengthen the basic technique and achieve constructions with improved security of pseudorandom generators from regular one-way functions as well as general one-way functions. I will focus on another result: a new construction of a pseudorandom generator from any exponentially hard one-way function that gains substantial improvement in both security and efficiency.

Joint work with Iftach Haitner and Omer Reingold.

Metric embeddings has been a useful tool in designing approximation algorithms in the last decade. One of the basic paradigm in which it is employed is as follows: Given a hard optimization problem over metric data $X$, first embed $f:X \rightarrow H$, and then solve the problem over $f(X)$, assuming that H is a "nice" metric space to optimize over. One such a successful approach is "probabilistic embedding into trees" due to Alon, Karp, Peleg, West, and Bartal. This metric embedding technique is useful for minimization problems over distances in which the optimized function is linear, and (relatively) easy to solve on trees.

In this talk, I will describe a strengthening of the probabilistic embedding, that enable reducing a class of NONLINEAR problems (mainly nonlinear clustering problems) to trees.

Based on a joint work with Assaf Naor

In the last decade range division scheduling policies have been proposed to overcome the large waiting times which are experienced in multi-server systems with heavy-tailed job size distributions. we analyze the effectivness of such scheduling policies, joint work with Hagit Sarfati.

mRNA molecules are folded in the cells and therefore many of their substrings may actually be inaccessible to protein and microRNA binding. The need to apply an accessability criterion to the task of genome-wide mRNA motif discovery raises the challenge of overcoming the core $O(n^3)$ factor imposed by the time complexity of the currently best known algorithms for RNA secondary structure prediction [Zuker et al., Nussinov et al.].

We speed up the dynamic programming algorithms that are standard for RNA folding prediction. Our new approach significantly reduces the computations without sacrificing the optimality of the results, yielding an expected time complexity of $O(n^2 \psi(n))$, where $\psi(n)$ is shown to be constant on average under standard polymer folding models. Benchmark analysis confirms that in practice the runtime ratio between the previous approach and the new algorithm indeed grows linearly with increasing sequence size.

The fast new RNA folding algorithm is utilized for genome-wide discovery of accessible cis-regulatory motifs in data sets of ribosomal densities and decay rates of S. cerevisiae genes and to the mining of exposed binding sites of tissue-specific microRNAs in A. Thaliana. This paper, joint work with Ydo Wexler and Chaya Zilberstein, has received the RECOMB2006 Special Mention Award.

We consider a new question regarding random walks on graphs: How long does it take for several independent random walks, starting from the same location, to cover an entire graph? We study the cover time, the expected time required to visit every node in a graph at least once, and we show that for a large collection of interesting graphs, running many random walks in parallel yields a speed-up in the cover time that is linear in the number of the parallel walks. We demonstrate that an exponential speed-up is sometimes possible, but that some natural graphs allow only a logarithmic speed-up.

We present and discuss the problem placing a sequence of disjoint rectangles on a larger rectangle. In the formal definition of the problem, the inputs are the integral dimensions of (the larger) rectangle , and a sequence of natural numbers, . The output is a sequence of rectangles, placed in a disjoint fashion on such that the area of is smaller than or equal to . Our goal is to place a maximal length prefix of . In this talk, we briefly motivate the problem, than we show the problem is hard. The main part of the talk is devoted to an approximation algorithm whose output satisfies: The area covered by all placed rectangles divided by the area covered by all rectangles in an optimal placement lies within a factor of This approximation ratio is achieved under the assumption that for each , , where . The complexity of the algorithm is linear. This is a joint work with Dror Rawitz and Oran Sharon.

This lecture summarizes the work done by the industrial partners in the MODELWARE project to measure the return on investment due to the adoption of a model driven development (MDD) strategy. The industrial partners in the project were France Telecom, Thales Air Traffic Management, Western Geco, WM-Data, and Enabler. Each of the industrial partners was teamed with one or two academic partners to assist in the creation and customization of MDD processes and tools appropriate to the industrial partner, taking into account their application domain, size of the company, software process maturity, and level of sophistication.

Each team carried out a series of experiments in order to measure the improvement in productivity. A baseline was established at the beginning of the project, and subsequently measurements were made using the MDD tools and processes. Different experiments focussed on different phases in the software lifecycle.

It was found that MDD provides little or no improvement in productivity in the initial phases of development, but significant gains of 20-60% were observed in the execution of maintenance tasks.

An instance of the sorting buffer problem consists of a metric space and a server, equipped with a finite-capacity buffer capable of holding a limited number of requests. An additional ingredient of the input is an online sequence of requests, each of which is characterized by a destination in the given metric; whenever a request arrives, it must be stored in the sorting buffer. At any point in time, a currently pending request can be served by drawing it out of the buffer and moving the server to its corresponding destination. The objective is to serve all input requests in a way that minimizes the total distance traveled by the server.

In this paper, we focus our attention on instances of the problem in which the underlying metric is either an evenly-spaced line metric or a continuous line metric. Although such restricted settings may appear to be very simple at first glance, we demonstrate that they still capture one of the most fundamental problems in the design of storage systems, known as the disk arm scheduling problem. Our main findings can be briefly summarized as follows:

1. We present a deterministic $O(\log{n})$-competitive algorithm for n-point evenly-spaced line metrics. This result improves on a randomized $O(\log^2{n})$-competitive algorithm due to Khandekar and Pandit (STACS '06). It also refutes their conjecture, stating that a deterministic strategy is unlikely to obtain a non-trivial competitive ratio.
2. We devise a deterministic $O(\log{N}\log\log{N})$-competitive algorithm for continuous line metrics, where N denotes the length of the input sequence. In this context, we introduce a novel discretization technique, which is of independent interest, as it may be applicable in other settings as well.
3. We establish the first non-trivial lower bound for the evenly-spaced case, by proving that the competitive ratio of any deterministic algorithm is at least 2.154. This result settles, to some extent, an open question due to Khandekar and Pandit (STACS '06), who posed the task of attaining lower bounds on the achievable competitive ratio as a foundational objective for future research.

Remark: For sake of social welfare, I intend to avoid delving into technicalities at all cost. Relatively involved arguments will be "established" by employing drawings and intuition, while some inaccuracies will be introduced without any hesitation. For this reason, my talk should be accessible to just about anyone.

Joint work with Iftah Gamzu (TAU).

We propose the use of tree automata as abstractions in the verification of branching time properties, and show several benefits. In this setting, soundness and completeness are trivial. It unifies the abundance of frameworks in the literature, and clarifies the role of concepts therein in terms of the well-studied field of automata theory. Moreover, using automata as models simpli fies and generalizes results on maximal model theorems.

Given a positive integer $k$, a sequence $A=(a_n)_{n=0}^{\infty}$ over a finite alphabet set $\Omega$ is said to be $k$-automatic if the $n$th term of this sequence is generated by a finite state machine with $n$ in base-$k$ as an input. Those sequences occur in many different areas (such as number theory, combinatorics, computer graphics, ergodic theory, physics and even music). We begin with some background on the theory of automatic sequences. Then we study the subword complexity of those sequences and of some more general sequences which are defined in terms of context-free languages. In particular we consider a problem of Allouche and Shallit [1] on $k$-context-free sequences.

J.-P. Allouche and J. O. Shallit, Automatic Sequences. Theory, Applications, Generalizations. Cambridge University Press, Cambridge, 2003.

The problem that is addressed in the talk is connecting a collection of agents that are randomly placed in a rectangular area in the plane or 3D space by a wireless network in a way that minimizes the sum of the powers of the agents' transmitters. Depending on a variant, the problem is proved or believed to be NP-hard. Known approximation algorithms use computing means and/or information that is usually not available in practice (e.g., knowledge of the whole network globally available, positions of agents known). Standard power allocation algorithms like unit (uniform) disk model or k-th nearest neighbor graph are provably asymptotically inefficient when applied to a random set of agents

The first result of the talk is a distributed algorithm that provably achieves almost surely a result that is optimal up to a multiplicative constant. However, while being a perfect solution of a problem of constuction of a connected network, a network built by the algorithm mentioned in the previous paragraph is not well suited for communication - it is similar to a minimum spanning tree and hence many physically close pairs of nodes are connected by a very long path in a network (high stretch factor), which causes important congestion in a network. The second part of the talk shows that connectivity of a network can be substantially improved while not increasing the total transmit power too much by inserting a certain number of bridges - direct connections of high stretch factor pairs of nodes.

Algovision is a collection of Java applets for visualization of algorithms. The applets were originally written for an undergraduate course on algorithms at Charles University by the speaker. They cover elementary stuff (tree data structures, heaps, sorting), basic graph algorithms (shortest path, spanning tree, flows), arithmetic algorithms (carry look-ahead addition, FFT), geometric algorithms (convex hull and Voronoi diagram in 2D), string matching (Rabin-Karp, Knuth-Morris-Pratt, Aho-Corasick), and geometric background of the simplex algorithm of LP

Unlike usual algorithm animation applets, Algovision aims to present visual tools to explain the underlying algorithmic idea and invariants that are basis for termination, correctness and complexity analysis; animation and visualization is often used to explain corresponding proofs. Moreover, many Algovision applets present visual tools to show other features of the algorithm in question, e.g., data structures used to implement the algorithm, applications, etc.

We study the $k$-Relaxed Tower of Hanoi Problem $BTH_k(n)$, posed by D. Wood in 1981. It differs from the classical problem by the relaxed placement rule: a larger disk can be placed above a smaller one if their size difference is less than $k$. The shortest sequence of moves from the standard state of disks $[1..n]$ on peg A to that on peg C, using peg B, is in question. For $k=1$ we get the classical problem.

In 1992, D. Poole suggested a solution for this problem, which is now known to have a fundamental gap. Beneditkis, Berend, and Safro suggested in 1998 a sequence of moves, denoted by $\AA_k(n)$, and for the case $k=2$ proved its optimality for $BTH_k(n)$.

We prove optimality of $\AA_k(n)$ for $BTH_k(n)$, for the general $k$. (Interestingly, a proof of optimality of $\AA_n$ for $BTH_n$ was suggested independently by Xiaomin Chen et al., approximately at the same time.)

We also survey some related problems: 1. Finding an optimal move sequence for $BTH_k(n)$, when allowing the sizes of the $n$ disks be any set of $n$ distinct integers. 2.a Finding the diameter of the configuration graph of $BTH_k(n)$. 2.b Finding the average distance between nodes in the configuration graph of $BTH_k(n)$. 3. Finding the family of all optimal solutions for $BTH_k(n)$ and their number.

If time permits, we would survey some related open problems.
In joint work with Yefim Dinitz.

We shall talk about three topics in which there has been recent progress. First, PageRank, the key idea behind Google, has been modified to involve a "teleport set," leading to the ability to measure importance of Web pages according to their relevance to a specific topic. The concept may also lead to methods for detecting spam pages. Second, a number of problems involving massive data are solved using a pair of "hashing" techniques together: minhashing and locality-sensitive hashing. We shall explain how these work and how they are used, for example, to find similar Web pages. Finally, we look at the classic problem of frequent itemsets in market baskets ("what to people buy together?"). The a-priori algorithm can be improved by careful attention to how main memory is managed

We consider broadcasting in radio networks, modeled as unit disk graphs (UDG). Network stations are modeled as points in the plane, where a station is connected to all stations at Euclidean distance at most 1 from it. A message transmitted by a station reaches all its neighbors, but a station hears a message (receives the message correctly) only if exactly one of its neighbors transmits at a given time step. One station of the network, called the source, has a message which has to be disseminated to all other stations. Stations are unaware of the network topology. Two broadcasting models are considered. In the conditional wake up model, the stations other than the source are initially idle and cannot transmit until they hear a message for the first time. In the spontaneous wake up model, all stations are awake (and may transmit messages) from the beginning.

It turns out that the running time of deterministic broadcasting algorithms depends on two parameters of the UDG network, namely, its diameter $D$ and its granularity $g$, which is the inverse of the minimum Euclidean distance between any two stations. We present a deterministic broadcasting algorithm which works in time $O(D g)$ under the conditional wake up model. On the negative side, we prove that any deterministic algorithm requires $\Omega(D \sqrt{g})$ time to accomplish broadcasting. For the spontaneous wake up model, we establish a lower bound of $\Omega(\min\{D + g^2, D \log{g}\})$ on deterministic broadcasting time and prove that this is tight. Thus our results yield a provable separation between the two models: for some parameter values, the lower bound in the first model is significantly larger than the upper bound in the second.

Recent research on shared-memory blocking algorithms (such as mutual exclusion and leader election algorithms) focuses on local-spin algorithms, in which processes busy-wait on locally-accessible variables. The most widely used metric for analyzing the time-complexity of such algorithms is the remote-memory-references (RMRs) metric, which counts the number of non-local accesses performed by processes.

While there is a lower bound of $\Omega(\log n / \log \log n)$ RMRs on the worst-case complexity of mutual exclusion algorithms that use only read/write operations, it was unknown whether such a lower bound applied also for the leader election problem, which may be regarded as `one-shot' mutual exclusion.

We provide a negative answer to this question by presenting an $O(1)$ RMRs leader election algorithm, using reads/writes only, improving on $\Theta(\log n)$ best prior-art algorithms. We show that our technique can be extended for deriving constant RMRs implementations of synchronization primitives that are considered stronger, such as consensus and comparison primitives (e.g., the widely used compare-and-swap).

Herlihy has shown that synchronization primitives vary widely in their power to support wait-free implementations and classified them in the wait-free hierarchy based on this computability power . For blocking implementations, this hierarchy is flat, since all primitives can be implemented in a blocking manner from reads and writes. Instead of comparing the computability power of a pair of primitives, it is therefore natural to ask whether we can base such a comparison on the RMR complexities of implementing these primitives from reads and writes. Our results establish that looking at the relative power of primitives through this lens reveals a landscape sharply different from that of Herlihy's wait-free hierarchy.

This talk is for a general audience and, to the extent possible, will be self-contained.

Joint work with Wojciech Golab, Vassos Hadzilacos, and Philipp Woelfel.

The talk summarized my Ph.d thesis and presents various distributed storage solutions. From distributed file systems to long-term large scale peer-to-peer storage networks. I will focus on how to achieve asynchronous consensus among processors in presence of crashes (using a failure detector).

Self-stabilizing algorithms can cope with transient faults. Transient faults can alter the system state to an arbitrary state and hence, cause a temporary violation of the algorithm correctness. Our algorithms can be started in an arbitrary state. Thus, can converge to their designed behavior. The talk will focus on a self-stabilizing failure detector, asynchronous consensus and replicated state-machine algorithm suite. We define new requirements for consensus that fit the on-going nature of self-stabilizing algorithms. The wait-free consensus (and the replicated state-machine) algorithm is a classic combination of a failure detector and a (memory bounded) rotating coordinator consensus that satisfy both eventual safety and eventual liveness. Several new techniques and paradigms are introduced. The bounded memory failure detector abstracts away synchronization assumptions using bounded heartbeat counters combined with a balance-unbalance mechanism. The practically infinite paradigm is introduced in the scope of self-stabilization, where an execution of, say, 2^64 sequential steps is regarded as (practically) infinite. Finally, we present the first self-stabilizing wait-free reset mechanism that ensures eventual safety and can be used in other scopes.

The technique of k-anonymization allows the releasing of databases that contain personal information while ensuring some degree of individual privacy. Given a database D that needs to be released, one produces a so-called k-anonymized version of that database where each record is indistinguishable from at least k-1 additional records. The anonymization process is usually performed by suppressing or generalizing database entries. We formally study the concept of generalization, and propose two information-theoretic measures for capturing the amount of information that is lost during the anonymization process. We call these two measures, the entropy measure and the non-uniform entropy measure.

The proposed measures are more eneral and more accurate than the measures that were proposed by Meyerson and Williams ([MW]),and Aggarwal et al. ([AFK]). We then study the problem of achieving k-anonymity with minimal loss of information. We prove that this problem is NP-hard and then study polynomial approximations for the optimal solution. Our first algorithm relies on similar ideas as the approximation algorithm that was proposed in [MW]. It gives an approximation guarantee of $O(ln k)$, for the entropy measure as well as for the previously studied measures. This mproves the best-known $O(k)$-approximation of [AFK]. While the approximation algorithms of [AFK] and [MW] relied on the so-called graph representation framework, which was shown in [AFK] to be limited to Omega(k)-approximations, our algorithm relies on a novel hypergraph representation that enables the improvement in the approximation ratio from $O(k)$ to $O(ln k)$.

As the running time of the algorithm is $O(n^(2k))$,we also show how to adapt the algorithm of [AFK] inorder to obtain a strongly polynomial approximation algorithm for our entropy measure with approximation guarantee of $O(k)$. We leave as an open problem to design an approximation algorithm, strongly polynomial or not, for the non-uniform entropy measure.

Joint work with Aristides Gionis

Graph theory and its applications is an exciting mathematical discipline which motivates the search for new algorithms, exact structures and combinatorial properties. To illustrate this point, consider the following simple question: When can you factor a logic (Boolean) formula, into a (logically equivalent) form in which each variable appears once and only once?

For example, the function $f = ab + acd + ace$ satisfies this property since it can be factored into the "read-once" expression $f = a(b+c(d+e))$. However, the function $h = ab + bc + cd$ does not satisfy the property.

In this talk, we will present the mathematical and computational aspects of this problem. We will show several classical characterizations of read-once functions which involve combinatorics, graph theory and properties of positive (monotone) Boolean functions. We also present the first polynomial time algorithm for recognizing and factoring read-once functions. The algorithm is based on a theorem of Gurvich and on algorithms for cograph recognition and a new efficient method for checking normality.

Finally, we raise a number of questions regarding the factoring certain non-read-once functions. In particular, we are able to show that if the co-occurrence graph of a positive Boolean function f is a tree, then the function is read-twice. However, no characterization is known for general read-twice functions.

Gene expression is directly regulated by protein transcription factors that bind at particular DNA or RNA sites in a sequence specific manner. A comprehensive characterization of these functional non-coding elements,or motifs, remains a formidable challenge, especially for higher eukaryotes.

I will present a rigorous computational methodology for ab-initio motif discovery from expression data, that utilizes the concept of mutual information, and have the following characteristics:

• directly applicable to any type of expression data, thus conceptually unifying existing motif discovery techniques,
• model-independence,i.e.,model-related assumptions commonly made by other methods are not required,
• simultaneously finds DNA motifs in upstream regions and RNA motifs in 3'UTRs and highlights their functional relations,
• scales well to metazoan genomes,
• yields very few false positive predictions if any,
• incorporates systematic analysis of the functional coherence of the predicted motifs, their conservation, positional and orientation biases, cooperativity, and co-localization with other motifs,
• displays predictions via a novel user-friendly graphical interface.

I will present results for a variety of data types, measured for different organisms, including yeast, worm, fly, mouse, human, and the Plasmodium parasite responsible for malaria. I will further discuss in detail surprising observations regarding gene expression regulation that were overlook by previous studies and naturally arise from our analysis. As a shorthand for our methodology we use the acronym FIRE, standing for Finding Informative Regulatory Elements.

Based on joint work with Olivier Elemento and Saeed Tavazoie.

In this talk we will describe the current efforts towards defining a Semantic Web knowledge representation language. The talk will deal with the foundations of the current standard, the Web Ontology Language (OWL), its limitations, and the efforts to extend OWL with a rule-based component. We will describe the problems facing researchers who work in this area as well as their possible resolution.

Verification of safety (e.g., invariance) and liveness (e.g., termination) properties of sequential and concurrent systems has traditionally required significant manual effort. For the case of safety, the method of predicate abstraction, suggested by Graf and Saidi, has become a popular semi-manual solution. However, it is generally too weak to deal with liveness. This talk presents a method known as Ranking Abstraction, which is based on program instrumentation, predicate abstraction, and model checking. The method applies the lessons learned from predicate abstraction to the verification of liveness properties. It is then shown that, just like predicate abstraction, the method can be embedded in an refinement process, resulting in a fully automatic method. Time permitting, I will then show how the method is applied to Shape Analysis, i.e., analysis of programs that destructively manipulate heaps. Joint work with Amir Pnueli and Lenore Zuck.

An important problem in genome rearrangements is sorting permutations by transpositions. Its complexity is still open, and two rather complicated 1.5-approximation algorithms for sorting linear permutations are known (Bafna and Pevzner, Christie). In this talk, we prove that the problem of sorting circular permutations by transpositions is equivalent to the problem of sorting linear permutations by transpositions. Hence, all algorithms for sorting linear permutations by transpositions can be used to sort circular permutations. Then, we derive our main result: A new 1.5-approximation algorithm, which is considerably simpler than the previous ones, and achieves running time which is equal to the best known. The analysis of the algorithm is significantly less involved. Joint work with Ron Shamir

If time permits we will discuss briefly two subsequent studies: "A 1.375-approximation algorithm for sorting by transpositions" (with Isaac Elias), and "Matrix Tightness: A Linear-Algebraic Framework for Sorting by Transpositions" (with Elad Verbin).

The computer industry has been able to keep an exponential improvement in performance for the last few decades. This unbelievable phenomenon is known as "Moore's law". As the computer architecture industry start reaching the "power wall", two new trends are being developed; the first trend calls to trade single thread performance with multithreaded performance and so to increase the overall performance of a processor by accommodate it with large number of small cores. The second trend calls to increase both the single thread performance and the multithreaded performance and so to divide the "transistor budget" of the processor between relatively small number of "large cores".

In this talk I will present the root cause of the "power wall", I will extend the discussion on each of the new trends in computer architectures described above and provide an analysis of what is needed for each of them to succeed. I will conclude the talk with a discussion on few open research topics which I believe are important for the computer architecture industry.

Bio: Avi Mendelson is a principal engineer in Intel’s Mobile Platform Group in Haifa, Israel, and adjunct professor in the CS and EE departments, Technion – Israel Institute of Technology. He received his B.Sc. and M.S.c degrees from the Technion, Israel Institute of Technology and his Ph.D from the University of Massachusetts at Amherst. Avi has been with Intel for 7 years. He started as senior researcher in Intel Labs, later he moved to the Microprocessor group where he served as the CMP architect of Intel Core Duo. Avi's work and research interests are in computer architecture, low power design, parallel systems, OS related issues and virtualization. His e-mail address is avi.mendelson@intel.com

Shape symmetry is an important cue for image understanding. In the absence of more detailed prior shape information, segmentation can be significantly facilitated by symmetry. However, when symmetry is distorted by perspectivity, the detection of symmetry becomes non-trivial,thus complicating symmetry-aided segmentation.

We present a novel approach for segmentation of symmetrical objects accommodating perspective distortion. The key idea is the use of the replicative form induced by the symmetry for challenging segmentation tasks. This is accomplished by dynamic extraction of the object boundaries, based on the image gradients, gray levels or colors, concurrently with registration of the image symmetrical counterpart e.g. reflection) to itself. The symmetrical counterpart of the evolving object contour supports the segmentation by resolving possible ambiguities due to noise, clutter, distortion, shadows, occlusions and assimilation with the background.

The symmetry constraint is integrated in a comprehensive level-set functional for segmentation that determines the evolution of the delineating contour. The proposed framework is exemplified on various images of skew-symmetrical objects and its superiority over state of the art variational segmentation techniques is demonstrated.

Joint work with Nahum Kiryati and Nir Sochen, Tel-Aviv University

In many organizations, private data should be revealed to some people while being concealed from others. In a hospital database system, for instance, a physician should be allowed to see the medical history of her patients; however, such medical data should not be available to the public. Thus, to support research over medical data while protecting privacy, only some of the data should be accessible to researchers. A common approach for protecting privacy is to manipulate sensitive data so that private information would not be revealed (e.g., by changing data values or transforming its structure). But, such manipulations can mislead users who are not aware of them and, thus, cause errors. In my talk, I will present a novel access-control mechanism for XML that protects privacy without misleading users. XML is a primary format for exchanging and publishing data on the Internet, in which data is presented in a hierarchical format. Our model uses the hierarchal nature of XML but also guarantees that private information will not be inferred because of the hierarchy, a challenge that is not required in the relational model. The mechanism employs rules for specifying the private data, and queries are validated with respect to these rules. Only queries that do not reveal private information are authorized and executed. I will talk about the complexity of validating queries, the privacy protection provided by our approach and how to test that a set of rules provides the desired concealment. No prior knowledge of XML or privacy is required. This is a joint work with Alberto Mendelzon, Renee Miller and Zheng Zhang.

The emergence of the Internet has initiated a radical shift of focus of our thinking about computational networked systems. While traditional system design assumes that all participants behave according to the intentions of the system designers, in reality, computer networks are built, operated and used by multiple users with diverse sets of interests. Hence, Internet protocols must be explicitly designed to work with interacting strategic individuals. Recently, there has been a growing interest in using tools from game theory and mechanism design to tackle incentive-related problems in these complex environments. In the first part of the talk, I will give an overview of the field, and demonstrate the inherent tension between individual rationality and collective welfare in computer networks through prime examples from my research. In the second part of the talk, I will concentrate on the inefficiency that is incurred due to users' selfishness in network routing, network formation and multicast transmission. In contrast to the common measure of "price of anarchy", which quantifies the loss incurred due to both selfishness and lack of coordination, we isolate the inefficiency originated from selfishness from that originated from lack of coordination. We show that coordination among selfish users can significantly improve the efficiency of the studied applications.

Topological persistence measures the "persistence" of topological features, such as connected components, tunnels and cavities, within the sublevel sets, ${x: f(x) < c}$, of a scalar field. The topological persistence of a feature is stable under small perturbations of the scalar field. Interval persistence is a generalization of topological persistence which better represents topological features in level sets, ${x: f(x) = c}$, gives a fuller matching of critical values, and is the appropriate setting for a stability theorem for critical points under small perturbations of scalar fields.

I will give an introduction to topological persistence and then discuss interval persistence, its relationship to topological persistence, and stability and matching of critical points under interval persistence. Joint work with Dr. Tamal Dey and Dr. Yusu Wang from The Ohio State University. Dr. Rephael Wenger is an associate professor in the Department of Computer Science and Engineering at The Ohio State University. He works on computational geometry, geometric algorithms and geometric modelling, particularly as they apply to graphics, visualization and biomedical image processing.

Issues on the border of Computer Science and Economics have drawn much recent research attention, mostly motivated by the Internet, where selfish agents interact in computational settings. I discuss various issues on this borderline, and in particular the problem of market design under computational constraints and in the presence of selfish agents with private preferences. To overcome strategic behavior, the predominant approach in the literature is to construct dominant strategy truthful mechanisms, but unfortunately not every approximation algorithm can be used for such a construction. In this talk I present a general deterministic technique to decouple the algorithmic allocation problem from the strategic aspects, by a procedure that converts any approximation algorithm to a dominant-strategy ascending mechanism. I also show how approximation can be achieved when agents do not have dominant strategies, using our new notion of "Algorithmic Implementation in Undominated Strategies". These results are taken from a joint research with Ron Lavi and Elan Pavlov.

During the last two years, I took part in a project at NASA Ames Research Center that attempts to improve the technology used to do activity planning for the Mars Exploration Rovers, and rovers for future NASA missions. The Mars rovers are autonomous vehicles that roam around the surface of Mars looking for interseting data. They can take pictures and perform various experiments. They have to operate autonomously for long strecthes at a time. These rovers are considered a great success story, but they are actually idle a great portion of the time. In this talk I'll explain the problems with current technology used to select their plans, and the work we did on devising planning algorithms that can come up with much better plans. This is joint work with a large group at NASA headed by Nicolaus Meuleau.

A mediator can help non-cooperative agents obtain an equilibrium that may otherwise not be possible. We give matching upper and lower bounds on the ability of rational agents to obtain the same equilibrium without a mediator, simply by engaging in non-binding pre-play communication, known as ``cheap talk''. Our upper bounds are based on k-resilient Nash equilibria for the secret sharing game, joint strategies where no member of a coalition of size up to k can do better, even if the whole coalition defects. Our results on implementing mediators with ``cheap talk'' reveal some connections between game theory, cryptography and distributed computing. Our upper bounds require both known and new tools in secure multi party computation. Our lower bounds require both known and new results in Byzantine fault tolerance and distributed computing. Joint work with Danny Dolev, Rica Gonen and Joe Halpern

The paradigm of Service Oriented Architectures provides a framework for dynamic Business Processes and for the emerging ideas of "programming-in-the-large". This talk addresses fundamental notions of services, in particular their proper termination and their composition, as well as the construction of (most liberal) partners, operating guidelines, (most abstract) substitutes, adapters, etc.

Finally, a feature complete semantics of the WS-BPEL language is glanced, together with techniques to analyze properties of WS-BPEL program