Logic and probability: the computational Connection

 Prof. Adnan Darwiche

 

I will discuss in this talk the subject of knowledge compilation of propositional knowledge bases and how it can form the basis for building state-of-the-art probabilistic reasoning systems. In the first part of the talk, I will focus on exact probabilistic reasoning and how it can be reduced to knowledge compilation, leading to an award-winning system for exact inference in the presence of local structure (determinism and context specific independence).

 

In the second part of the talk, I will focus on approximate inference, and how it can be formulated in terms of "relaxing" logical constraints (equalities in particular) for the purpose of aiding the process of knowledge compilation. The resulting formulation, known as "Relax, Compensate and then Recover," subsumes loopy belief propagation and some of its generalizations, and is the basis for an award-winning system at the UAI inference competition in 2010. I will also review in the talk some recent advances and ongoing work in the world of knowledge compilation and the impact it has on the state-of-the-art in probabilistic and symbolic reasoning.

Bio: Adnan Darwiche is a professor of computer science and director of the automated reasoning group at UCLA. The group focuses on both symbolic and probabilistic reasoning and is known for several award-winning and publicly available reasoning systems, including ACE, SamIam, Rsat and c2d. Professor Darwiche is a AAAI Fellow, a former chair of the computer science department at UCLA, a former editor-in-chief for JAIR, and author of the book "Modeling and Reasoning with Bayesian Networks," published in 2009 by Cambridge University Press. 

24 February, 2011

FENS L055, 13:40