Neural networks and emergent adaptive signal processing

Abstract

The relationships between neural networks and adaptive signal processing are made evident by treating the connection weights in neural networks as integrators. The integrators are often preceded and followed by multipliers, leading to a multiplier-integrator-multiplier structure for the weights. Neural networks calculate correlations in the input data and develop correlative codes, as opposed to analog-to-digital conversions. Hebbian learning adjusts weight values to minimize the expected variance of the neuron outputs. The correlation processing of neural networks may lead to the development of alternate methods for adaptive signal processing. >