Neural Networks Processing Mean Values of Random Variables

TL;DR

This paper presents neural networks derived from Bayesian belief networks that require no training, can integrate multiple evidence sources, and handle contradictory information, effectively capturing probabilistic properties.

Contribution

It introduces a new class of neural networks based on probabilistic models that do not need training and can process multiple evidence sources.

Findings

Neural networks require no training to determine weights.

Networks can pool multiple evidence sources.

They handle inconsistent or contradictory evidence effectively.

Abstract

We introduce a class of neural networks derived from probabilistic models in the form of Bayesian belief networks. By imposing additional assumptions about the nature of the probabilistic models represented in the belief networks, we derive neural networks with standard dynamics that require no training to determine the synaptic weights, that can pool multiple sources of evidence, and that deal cleanly and consistently with inconsistent or contradictory evidence. The presented neural networks capture many properties of Bayesian belief networks, providing distributed versions of probabilistic models.