The study of an ATM switching architecture controlled by neural network

This paper points out the disadvantages of Banyan communication network switching realized by using usual bidirectional associative memories (BAM) neural network, and presents a method for realizing Banyan communication network switching by using adaptive and asymmetric BAM, i.e. AABAM neural network. The empirical results have shown that the new architecture is more adaptive to many arbitrary inputs patterns and the memory capacity than that of the old one, and the ability to prevent pseudo-memories has been improved.