12:00, 30 Oct 96, WeH 7220
Automated Search for Bayesian Networks
Peter Spirtes (CMU Philosophy)
Bayesian networks are directed acyclic graphs that have both a
statistical and and a causal interpretation. They have proved useful
as expert systems, and can be used to predict the effects of
interventions upon a causal system. It would be useful to have systems
which constructed Bayesian networks from statistical data and
background knowledge. I will describe why automated search is a
difficult problem, introduce some axioms relating the probabilistic
and causal interpretations which make automated search possible,
briefly describe some algorithms for automated search, and some
simulation and empirical results of such searches.