A GA-based flexible learning algorithm with error tolerance for digital binary neural networks

This paper presents a learning algorithm of digital binary neural networks for approximation of desired Boolean functions. In the learning, the genetic algorithms is used with flexible fitness that tolerates error: it is suitable to reduce the number of hidden neurons and to tolerate noise and outliers. We then apply the algorithm to design of cellular automata with rich spatio-temporal patterns and various applications. Performing basic numerical experiment, the algorithm efficiency is confirmed.