Benchmark of Some Learning Algorithms for Single-Layer and Hopfield Networks
Eddy Mayoraz
Ecole Polytechnique Federale de Lausanne, Departement de Mathematiques,
Chaire de Recherche Operationnelle, CH-1015 Lausanne Switzerland
Abstract
Many algorithms have been proposed for training a single layer or a Hopfield network with binary activations. The purpose of this work is to compare some of these algorithms experimentally and point out the advantages of each. Experiments are also reported in which the density of the synaptic connections is reduced or in which a few quantization levels are used for the synaptic weights.
