FPGA Implementation of Bidirectional Associative Memory Via Simultaneous Perturbation Rule

Abstract

Bidirectional Associative Memory (BAM) is a typical example of recurrent neural networks. Software implementation of BAM does not obtain sufficient speed in the operation. Therefore, hardware implementation, especially, FPGA implementation of BAM is very promising. Originally, the weights of BAM are calculated by patterns to be memorized. However, we adopt a recursive learning method via the simultaneous perturbation learning rule.