[talks] reminder: 1-2pm talk today in room 302 on Learning Structured Bayesian Networks by Marie desJardins

[Jennifer Rexford]

Speaker: Professor Marie desJardins, University of Maryland
Title: Learning Structured Bayesian Networks: Combining Abstraction 
Hierarchies and Tree-Structured Conditional Probability Tables
Date/time: 1-2pm Wednesday May 7
Location: 302 in the CS Building

Abstract:

   In this talk, I will describe our research on incorporating
   background knowledge in the form of feature hierarchies during
   Bayesian network learning. Feature hierarchies enable the learning
   system to aggregate categorical variables in meaningful ways, thus
   enabling an appropriate "discretization" for a categorical variable.
   In addition, by choosing the appropriate level of abstraction for
   the parent of a node, we also support compact representations for
   the local probability models in the network. We combine this notion
   of selecting an appropriate abstraction with context-specific
   independence representations, which capture local ndependence
   relationships among the random variables in the Bayesian network.
   Capturing this local structure is important because it reduces the
   number of parameters required to represent the distribution. This
   can lead to more robust parameter estimation and structure
   selection, more efficient inference algorithms, and more
   interpretable models.

   I will describe our primary contribution, the Tree-Abstraction-Based
   Search (TABS) algorithm, which learns a data distribution by
   inducing the graph structure and parameters of a Bayesian network
   from training data. TABS combines tree structure and attribute-value
   hierarchies to compactly represent conditional probability tables.
   In order to construct the attribute-value hierarchies, we
   investigate two data-driven techniques: a global clustering method,
   which uses all of the training data to build the attribute-value
   hierarchies, and can be performed as a preprocessing step; and a
   local clustering method, which uses only the local network structure
   to learn attribute-value hierarchies. Empirical results in several
   benchmark domains show that (1) combining tree structure and
   attribute-value hierarchies improves the accuracy of generalization,
   while providing a significant reduction in the number of parameters
   in the learned networks, and (2) data-derived hierarchies perform as
   well or better than expert-provided hierarchies.

   BIOGRAPHY Dr. Marie desJardins is an associate professor in the
   Department of Computer Science and Electrical Engineering at the
   University of Maryland, Baltimore County. Her research is in
   artificial intelligence, focusing on the areas of machine learning,
   multi-agent systems, planning, interactive AI techniques,
   information management, reasoning with uncertainty, and decision
   theory.

   Dr. desJardins can be contacted at the Dept. of Computer Science and
   Electrical Engineering, University of Maryland Baltimore County,
   1000 Hilltop Circle, Baltimore MD 21250, mariedj at cs.umbc.edu,(410)
   455-3967.