- 1 3rd Workshop on Algorithmic issues for Inference in Graphical Models - AIGM13 - September 13, 2013.
3rd Workshop on Algorithmic issues for Inference in Graphical Models - AIGM13 - September 13, 2013.
Location: Institut Henri Poincaré (IHP) Hermite amphitheater, Paris, France.
Inscription is free but each participant should register here.
Most real (e.g. biological) complex systems are formed or modelled by elementary objects that locally interact with each other. Local properties can often be measured, assessed or partially observed. On the other hand, global properties that stem from these local interactions are difficult to comprehend. It is now acknowledged that a mathematical modelling is an adequate framework to understand, to be able to control or to predict the behaviour of complex systems, such as gene regulatory networks or contact networks in epidemiology.
More precisely, graphical models (GM), which are formed by variables linked to each other by deterministic or stochastic relationships, allow researchers to model dependencies in possibly high-dimensional heterogeneous data and to capture uncertainty. Analysis, optimal control, inference or prediction about complex systems benefit from the formalisation proposed by GM. To achieve such tasks, a key factor is to be able to answer general queries: what is the probability to observe such events in this situation ? Which model best represents my data ? What is the most acceptable solution to a query of interest that satisfies a list of given constraints ? In many situations, an exact resolution cannot be achieved either because of computational limits, or because of the intractability of the problem; hence approximate methods are needed.
The aim of this workshop is to bridge the gap between Statistics and Artificial Intelligence communities where approximate inference methods for GM are developped. We are primarily interested in algorithmic aspects of probabilistic (e.g. Markov random fields, Bayesian networks, influence diagrams), deterministic (e.g. Constraint Satisfaction Problems, SAT, weighted variants, Generalized Additive Independence models) or hybrid (e.g. Markov logic networks) models.
We expect both:
- reviews that analyze similarities and differences betwen approaches developped by computer scientists and statisticians in these areas and
- original research works which propose new algorithms and show their performance on data sets as compared to state-of-the-art methods.
- Francis Bach, SIERRA, INRIA, ENS Paris, France.
- Nic Wilson, Cork Constraint Computation Centre, Ireland.
- Amos Storkey, School of Informatics, University of Edinburgh, Scotland, UK.
- 8.45 - 9.00: workshop introduction
- 9.00 - 10.00: F. Bach, Beyond stochastic gradient descent for large-scale machine learning
- 10.00 - 10.30: E. Aljohani and J. Cussens, Informative priors for learning in graphical models
- 10.30 - 11.00: break
- 11.00 - 11.30: J. Whittaker, Y. Xiang and F. Martin, Forward elements in graphical model search
- 11:30 - 12.00: P. Fernique, J.-B. Durand and Y. Guédon, Parametric modelling of multivariate count data using probabilistic graphical models
- 12.00 - 13.30: lunch
- 13.30 - 14.30: A. Storkey, Relationships Between Probabilistic Inference and Decision Processes
- 14.30 - 15.00: J. Radoszycki, N. Peyrard and R. Sabbadin, Evaluation of stochastic policies for factored Markov decision processes
- 15.00 - 15.30: A. Franc, N. Peyrard, R. Sabbadin, T. Schiex, An example of a link between artificial intelligence and statistical physics: exact computation of the partition function of a graphical model on a series-parallel graph
- 15.30 - 16.00: break
- 16.00 - 17.00: N. Wilson, Reasoning with Tradeoffs in Multi-Objective Optimisation Problems
- 17.00 - 17.30: S. de Givry, S. Prestwich and B. O'Sullivan, Dead-End Elimination for Weighted constraint satisfaction problems
Call for paper
Topics include, but are not limited to:
- answering queries in GMs (MAP/MPM/MPE, satisfaction, optimization...),
- evaluation of the normalisation constant of a Markov random field,
- solution counting in deterministic GM or enumeration of k-best solutions,
- decision variable optimisation (optimisation within deterministic or mixed deterministic/stochastic GM),
- variational methods,
- Monte-Carlo methods,
- bounds for approximate inference,
- stochastic satisfiability (SAT) and stochastic constraint programming (CP),
- bridge between probabilistic and logic formalisms
We will consider papers from 1 to 2 pages
Contributions (pdf files) can be submitted no later than the 10th of June, by sending an email to the organisation committee.
- Submission deadline: June 10, 2013
- Notification to authors: July 1, 2013
- Submission of final version: July 12, 2013
- Meeting date: September 13, 2013.
Simon de Givry, Nathalie Peyrard, Régis Sabbadin, Thomas Schiex, Matthieu Vignes (MIA-T, INRA Toulouse, France) and Stéphane Robin (AgroParisTech, Paris, France).
For paper submission or enquiries about the meeting, please contact the organisation committee.
Links to past workshops
- AIGM 2012.
- ECCS 2010 workshop on 'Graphical models for reasoning on biological systems: computational challenges'.