How Real-valued Random Synapses Evolve toward Symmetry with Diploid Chromosomes?

We apply genetic algorithms to the Hopfield's neural network model of associative memory. Previously, we observed that random synaptic weights of a network evolved to create the fixed point attractors in pattern space exactly at the locations corresponding to a set of given patterns to be stored. In that simulation, a weight configuration was represented as a sequence of genes which might be called a haploid chromosome. Then, a population of these haploid chromosomes underwent evolution. In this paper, we employ diploid chromosomes, a pair of chromosomes. We have reported elsewhere two different versions of this evolution. One is the evolution of the Hebbian synapses in which some of the synapses are adaptively pruned according to information in diploid chromosomes. The other is the evolution of random synaptic weights whose values are restricted to 61, and are encoded directly into diploid chromosomes. Here, we describe the evolution of real-valued random synaptic weights. keywords ..