November 6, 1998
School of Information Technology and Engineering
University of Ottawa
Learning in Belief Networks and its Application in Distributed Databases
In this talk we present the problem of learning in belief networks and its application to caching data with repeated read-only accesses in distributed databases. Our goal is to build a probabilistic network from the distribution of the data which adequately represents the data. We describe two classes of techniques for the induction of Bayesian networks from data, methods based on probabilistic-graph models and methods using a Bayesian learning approach. The probabilistic methods for learning Bayesian Belief Networks(BBN)s focus on finding the most likely structure, implicitly assuming that there is a true underlying structure. The Bayesian methods for learning BBN search the network structure hypothesis to yield the network which maximizes the relative a posteriori probability. Once constructed, such a network can provide insight into probabilistic dependencies that exist among the variables. We consider representations using Chow's dependence trees and Polytrees (Singly Connected Networks) as structures for inferring causal knowledge. We apply this approach to learn patterns or sequences in query accesses to distributed databases.