Theoretical and Experimental Research Into Biological Mechanisms Underlying Learning and Memory
Abstract
We describe an extended model of backward propagation incorporating gain modification and compare the performance of the extended model with ordinary backward propagation. We also describe our work on a statistical model for feature extraction based on the BCM neural network model. The model is presented as an exploratory (PP) (Projection Pursuit) algorithm. The formulation, which is similar in nature to PP, is based on a minimization of a cost function over a set of parameters, yielding an optimal decision rule under some norm. A new projection index (cost function) was presented that favors directions possessing multi-modality, where the multi-modality is measured in terms of the separability property of the data. The synaptic modification equations, which perform the minimization of the cost function, turn out to be similar to the synaptic modification equations governing learning in BCM neurons. Keywords: Backward propagation, Statistical model.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 24, 1990
- Accession Number
- ADA223615
Entities
People
- Leon Cooper
Organizations
- Brown University