Structured statistical models: a tool for cognition and modeling

Abstract

A promising class of models, namely, probabilistic models of dependences based on acyclic directed graphs (ADG), primarily of the Bayesian networks, is reviewed. The expressive and cognitive properties of the ADG models, their ability to convey a causal relationship are described. The role and place of the Bayesian networks as a tool for analysis and deneralization of empirical data, their relation to logic and induction problem are shown in comparison with other approaches to cognition and model identification.