Efficient Inference Methods for Probabilistic Logical Models

Event Speaker
Sriraam Natarajan
Department of Computer Science, University of Wisconsin, Madison
Event Type
Colloquium
Date
Event Location
KEC 1001
Event Description

Probabilistic Logical Models (PLMs) combine the powerful formalisms of probability theory and first-order logic to handle uncertainty in large, complex problems. While PLMs provide a very effective learning paradigm under the umbrella of Statistical Relational Learning (SRL) methods, tractable inference is a significant problem in these models. Earlier approaches focused on grounding the model to a propositional network to use existing inference algorithms. Other popular techniques include sampling and lifted inference, with a lot of interest in the latter recently.

In this talk, I will present three different approaches to accelerate inference in PLMs. First, a preprocessing method for Markov Logic Networks that makes exact grounded inference tractable; second, an approximate inference method called `counting belief propagation' that performs belief propagation on compressed factor graphs; and finally, an `anytime'inference algorithm that returns a bound over the marginal distribution of the query variable. I will present experimental results to demonstrate the usefulness of these three distinct, yet related, inference methodologies.

Speaker Biography

Sriraam Natarajan is currently a Post-Doctoral Research Associate at the Department of Computer Science at University of Wisconsin-Madison. He graduated with his PhD from ¾«¶«Ó°ÊÓworking with Dr. Prasad Tadepalli. His research interests lie in the field of Artificial Intelligence, with emphasis on Machine Learning, Statistical Relational Learning, Reinforcement Learning, Graphical Models and Bio-Medical Applications.