The variance of covariance rules for
associative matrix memories and reinforcement learning.
Peter Dayan Terry Sejnowski
Neural Computation, 5 205-209.
Abstract
Hebbian synapses lie at the heart of most associative matrix memories
and are also biologically plausible. Their analytical and computational
tractability make these memories the best understood form of distributed
information storage. A variety of Hebbian algorithms for estimating the
covariance between input and output patterns has been proposed. The
authors point out that one class of these involves stochastic estimation
of the covariance, show that the signal-to-noise ratios of the rules are
governed by the variances of their estimates, and consider some parallels
in reinforcement learning.