An Application of a Genetic Algorithm to the Construction of a Minimally Admissible Training Sample for a Neural Network Decision-Making System

摘要

The problem is considered of constructing a minimally admissible training sample for a neural network. This problem consists in finding the subset of patterns from the source sample, the description of which allows us to reduce the computational cost of training and provide high-accuracy operation of the network on the source sample. A genetic algorithm for constructing such a sample is described.