Redes Neurais Atratoras em Topologias Complexas

Abstract

In this work we present some modifications in the Hopfield model for attractor neural networks. First, in order to bring the model closer to a biological network of neurons, we introduce a controlled dilution generating complex topologies. This modification is supported by experimental evidences [1, 2]. Then, study the relation between eficiency and the topologies, in the recovery stored patterns. We study how the network topology play a role in the model, by introducing correlated patterns and studing the performance of each topology as the correlation between patterns is increased.