A main difficulty of building intelligent systems is the ability to reason under uncertainty. One of the most successful approaches for dealing with this challenge is via use of probabilistic graphical models which provide a flexible framework for describing large, complex, heterogeneous collections of variables. Probabilistic inference of these models is important for a variety of desired applications in medical diagnosis, sensor fusion, automated help systems, credit assessment, computational biology, computer vision, data mining, and many more. Since inference across larger, more complex, problems is generally intractable, many approximation schemes have been developed relying on a vast array of methods including variational approaches, search, sampling, and learning from data.

The goal of this challenge is to explore your various approaches of approximating quantities of graphical models such as their partition function, probability of given evidence, marginal probabilities, and the MAP or marginal MAP values evaluating your algorithms on an array of problems with varying difficulties. This year, we also include a new category hosting a learning challenge for multi-label classification (MLC) through machine learning. We hope this event can fuel the continued growth of algorithms in our field as well as lead to new fruitful collaborations.

Organizers

Rina Dechter (University of California, Irvine)
Alexander Ihler (University of California, Irvine)
Vibhav Gogate (University of Texas, Dallas)
Junkyu Lee (IBM Research)
Bobak Pezeshki (University of California, Irvine)
Contact: uaicompetition at gmail dot com