Storage properties of correlated perceptrons

A. Engel, D. Malzahn, I. Kanter

Research output: Contribution to journalArticlepeer-review

Abstract

Feedforward multilayer neural networks implementing random input—output mappings develop characteristic correlations between the activity of their hidden nodes which are important for the understanding of their storage and generalization performance. It is shown how these correlations can be calculated within the replica-symmetric approximation. Replacing the multilayer network by an ensemble of perceptrons displaying the same correlations the relative influence of these correlations on the storage capacity can be studied.

Original languageEnglish
Pages (from-to)1507-1513
Number of pages7
JournalPhilosophical Magazine B: Physics of Condensed Matter; Statistical Mechanics, Electronic, Optical and Magnetic Properties
Volume77
Issue number5
DOIs
StatePublished - May 1998

Fingerprint

Dive into the research topics of 'Storage properties of correlated perceptrons'. Together they form a unique fingerprint.

Cite this