A Cerebral Cortex Model that Self-Organizes Conditional Probability Tables and Executes Belief Propagation

This paper describes a neural network model of cerebral cortex, BESOM model, that acquires conditional probability tables for a Bayesian network using self-organizing maps and estimates states of random variables with an approximate belief propagation algorithm. The approximate algorithm is derived from some assumptions. A neural network that executes the derived algorithm is in good agreement with six-layer and column structures that represent the anatomical characteristics of a cerebral cortex in many respects. This model has scalable time and space complexities and is therefore qualified to be a model of the brain, a large-scale information processor.