Adaptive genetic algorithm for the binary perceptron problem
Heinrich Köhler · Journal of Physics A Mathematical and General · 1990
For neural networks with J couplings the perceptron problem for random unbiased patterns is considered. An algorithm that uses concepts of the continuous perceptron problem as well as ideas of biological optimization is proposed and investigated. The distribution of local stabilities and the critical storage capacity alpha c are determined. While for N less than 50 the value of alpha c is approximately 0.83, the storage capacity goes down to alpha c =0.7 for N=255.