Assignments

Homework policy: You may work and submit in pairs, but each student is responsible for understanding the entire solution. I will occasionally ask you to explain your solution in class; an inability to justify a solution you handed in will be taken as evidence of copying. Please cite all sources: people, papers, books, internet.

Significant thought and effort is put into crafting exercises so as to maximally facilitate understanding of the material. Solutions will be graded for clarity and correctness. You are encouraged to write them up in LaTeX – mastering this now will greatly help you later on when writing papers.

Project topics

Multiclass, multilabel classification
PCA, Johnson-Lindenstrauss and other dimensionality reduction techniques
Clustering: K-means, spectral, other kinds, Kleinberg's impossibility theorem
multiplicative updates, winnow algorithm; related: this and this
scale-sensitive dimensions; also this (a very technical topic)
learning boolean functions with Fourier analysis
model selection
locally-linear embeddings and manifold techniques
nonnegative matrix factorization
Markov Decision Processes
convex loss as a surrogate for 0-1 loss
statistical queries and learning with noise