Mutually Connected Neural Network Can Learn Some Patterns by Means of GA

abstract We simulated an associative memory with mutually connected neural network, and successfully made the connection matrix learn some binary patterns only by means of genetic algorithm. Although the memory capacity is about 12 % of the number of neurons, the fact that it was made without any learning algorithm like Hebbian rule is very interesting. The structure of connection matrices we obtained is quite diierent from that of Hoppeld network. Our overall goal for this research is two fold. One is to know if we can use genetic algorithm as a more eeective learning method than that proposed so far, and another is to understand this learning mechanism of genetic algorithm.