Optimal storage capacity of neural networks at finite temperatures
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Description
Gardner's analysis of the optimal storage capacity of neural networks is extended to study finite-temperature effects. The typical volume of the space of interactions is calculated for strongly-diluted networks as a function of the storage ratio $α$, temperature $T$, and the tolerance parameter $m$, from which the optimal storage capacity $α_c$ is obtained as a function of $T$ and $m$. At zero temperature it is found that $α_c = 2$ regardless of $m$ while $α_c$ in general increases with the tolerance at finite temperatures. We show how the best performance for given $α$ and $T$ is obtained, which reveals a first-order transition from high-quality performance to low-quality one at low temperatures. An approximate criterion for recalling, which is valid near $m=1$, is also discussed.
22 pages in LaTex, 4 figures upon request, SNUTP-93-26
22 pages in LaTex, 4 figures upon request, SNUTP-93-26