UTCS AI Colloquia - Jude Shavlik, University of Wisconsin-Madison, "Improving Learning and Inference in Statistical Relational Learning," PAI 3.14

Contact Name: 
Ray Mooney
PAI 3.14
Nov 28, 2012 11:00am - 12:00pm

Signup Schedule: http://apps.cs.utexas.edu/talkschedules/cgi/list_events.cgi

Talk Audience: UTCS Faculty, Grads, Undergrads, Other Interested Parties

Host:  Ray Mooney

Talk Abstract: The two primary mathematical underpinnings of artificial intelligence havebeen first-order predicate logic and probability. Over the 15 or so years there has been substantial research activity on approaches that combine the two, producing various forms of probabilistic logic. Within machine learning, this work is commonly called Statistical Relational Learning (SRL).

At Wisconsin we have been investigating an approach to SRL where we learn probabilistic concepts expressed as a sequence of first-order regression trees. In such trees, nodes are expressions in first-order logic and the leaves are numbers (hence the phrase 'regression trees,' rather than the more common 'decision trees'). I will present our learning algorithms for two SRL knowledge representations, Relational Dependency Networks (RDNs) and Markov Logic Networks (MLNs), and describe their performance on a variety of 'real world' testbeds, including comparison to alternate approaches. Time permitting, I will also present our work on using a relational database management system (RDBMS) and an optimization method called 'dual decomposition' to substantially speed up inference ('question answering') in MLNs. Our approach allowed us to handle inference in an MLN testbed with 240 million facts (which lead to 64 billion 'factors' in the grounded Markov network). Joint work with Bernd Gutmann, Kristian Kersting, Tushar Khot, Sriraam Natarajan, Feng Niu, Chris Re, and Ce Zhang. Papers available at http://pages.cs.wisc.edu/~shavlik/mlrg/publications.html

Speaker Bio: Jude Shavlik is a Professor of Computer Sciences and of Biostatistics and Medical Informatics at the University of Wisconsin - Madison, and is a Fellow of the American Association for Artificial Intelligence. He has been at Wisconsin since 1988, following the receipt of his PhD from the University of Illinois for his work on Explanation-Based Learning. His current research interests include machine learning and computational biology, with an emphasis on using rich sources of training information, such as human-provided advice. He served for three years as editor-in-chief of the AI Magazine and serves on the editorial board of about a dozen journals. He chaired the 1998 International Conference on Machine Learning, co-chaired the First International Conference on Intelligent Systems for Molecular Biology in 1993, co-chaired the First International Conference on Knowledge Capture in 2001, was conference chair of the 2003 IEEE Conference on Data Mining, and co-chaired the 2007 International Conference on Inductive Logic Programming. He was a founding member of both the board of the International Machine Learning Society and the board of the International Society for Computational Biology. He co-edited, with Tom Dietterich, "Readings in Machine Learning." His research has been supported by DARPA, NSF, NIH (NLM and NCI), ONR, DOE, AT&T, IBM, and NYNEX.