STSM Reports – WG3
1.
Short Term Scientific
within the ESF-COST Action on Algorithmic Decision Theory (Action number
IC0602)
March 2008
Scientific Report
The main purpose of the Short Term Scientific
Mission was the development and discussion of a sensitivity analysis framework
allowing to investigate the impact of simultaneously
varying the different preference and utility parameters on the results of a
multi-attribute utility theory (MAUT) analysis.
In
general, many fruitful discussions in the broad field of decision support,
participatory approaches and the impact of risk and uncertainty took place.
Moreover, potential future collaboration opportunities in the area of decision
making under uncertainty were discussed throughout the whole time of the visit.
A paper
entitled “Sensitivity Analysis Techniques in Multi-Attribute Utility Theory”
has been prepared during the visit (see attached document). This paper will be
submitted to OR Spectrum.
The aim in this paper is to extend some of the
existing sensitivity analysis techniques to allow a more comprehensive
investigation of the impact of the different sources of uncertainty on the
results of a MAUT analysis. In particular, in order to be able to explore the
robustness of decision processes, it is important to analyse which
uncertainties are most relevant in terms of the results. Furthermore, the
methods proposed in the paper are aimed at improving group decision support,
i.e. at fostering consensus building by facilitating the elicitation of
preference and utility parameters through sensitivity analyses. An important
and challenging area of applying multi-attribute methods is industrial risk and
emergency management. In modern industrial production networks and their
external environment complex decision situations need to be resolved with
respect to the potential impact on the society in a wide variety of
circumstances. Usually, various scientific expert groups are involved with
heterogeneous technical background knowledge in different disciplines. Know-how
from economic, ecological, engineering and natural sciences must be brought
together, taking into account political and socio-psychological factors
resulting in a typical multi-attribute decision analysis (MADA) problem.
Additionally, with the increasing demand from the media and the public for
information and justification from authorities, methods are required to assess
how decisions are taken. MADA seeks to facilitate the communication with the
public and the media and can be helpful in forming an audit trail and in
enhancing public confidence and understanding in relation to complex group
decisions. The aim in the paper, therefore, is to apply the new sensitivity
analysis approaches within a case study in the area of industrial risk
management. Special emphasis is given to appropriate graphical illustrations in
order to support the tangibility of the results.
Finally, I would like to thank the European
Science Foundation and the steering committee of the COST action on Algorithmic
Decision Theory for the funding of my visit. I had the opportunity to work with
very nice, friendly and supportive colleagues in the research group at
2. Ulle Endriss to
During
an STSM of UE to the
3. Nicolas Maudet to U. of Amsterdam.
I
benefited from the support of an STSM to visit Ulle Endriss at ILLC (
The purpose of this STSM was to discuss, establish and initiate new research directions in the field of multiagent resource allocation (MARA). It was related to both WG2 and WG3. More specifically, the topics investigated include: (1) the dynamics of resource allocation (with a special focus on fairness-oriented measures); (2) the generation of problem instances for MARA problems (with the objective of setting up the specifications of a generic problem instance generator); and more prospectively (3) experimental theory of justice (consisting in interpreting different facets of theories of justice in terms of MARA concepts) The proposal was intended to reinforce an existing collaboration between LAMSADE and ILLC, and involved also several PhD students (eg. Joel Uckelman, Cedric Degremont) at the host institution.
4. STSM -- Hélène Fargier (IRIT) visiting B. O'Sullivan and N.
Wilson (and also Greg Provan and T. Hadzic) at the Cork Constraint
The main topics of discussion have been :
- Approximate compilation (with G. Provan and B. O'Sullivan). In particular, discussion of Greg Provan's new results about incremental algorithms for approximate compilation.
- Configuration problems (with T. Hadzic).
- The
handling of partially ordered preferences in combinatorial problems (with
5. Manfred Jaeger visited Ronen
Brafman at Ben-Gurion University, Israel, March 9-14, 2008. The purpose of this
visit was twofold: the first goal was to explore the connections between probabilistic
relational models (Jaeger's field of expertise) and Brafman's
recent work on relational preference rules. This connection also establishes a
link between working group 1 (uncertainty models, Jaeger) and working group 3
(preferences, Brafman). In the
investigations during the stsm
a possible common computational framework for reasoning with probabilities and
reasoning with preferences in relational contexts was identified. Based on
these insights, further collaboration with possible joint publications is
anticipated.
A second goal of this stsm was educational: Jaeger gave an extended tutorial (in
3 parts over 2 hours each) on probabilistic logic models at the department of
computer science at Ben-Gurion University.
This tutorial was attended by 10-15 graduate students and faculty from
different groups at the department.