Neural Networks Processing Mean Values of Random Variables

Department of Anatomy and Neurobiology, Washington University School of Medicine, Saint Louis, MO 63110(Dated: February 2, 2008)We introduce a class of neural networks derived from probabilistic models in the form of Bayesianbelief networks. By imposing additional assumptions about the nature of the probabilistic modelsrepresented in the belief networks, we derive neural networks with standard dynamics that requireno training to determine the synaptic weights, that can pool multiple sources of evidence, and thatdeal cleanly and consistently with inconsistent or contradictory evidence. The presented neuralnetworks capture many properties of Bayesian belief networks, providing distributed versions ofprobabilistic models.I. INTRODUCTION