STRUCTURAL EVOLUTION OF NEURAL NETWORKS HAVING ARBITRARY CONNECTIONS BY A GENETIC METHOD

A genetic method to generate a neural network which has both structure and connection weights adequate for a given task is proposed. A neura l network having arbitrary connections is regarded as a virtual living thing which has genes representing its connections among neural units . Effectiveness of the network is estimated from its time sequential i nput and output signals. Excellent individuals, namely appropriate neu ral networks, are generated through generation iterations. The basic p rinciple of the method and its applications are described. As an examp le of evolution from randomly generated networks to feedforward networ ks, an XOR problem is dealt with, and an action control problem is use d for making networks containing feedback and mutual connections. The proposed method is available for designing a neural network whose adeq uate structure is unknown.