Title :
Optimal gradient descent learning for bidirectional associative memories
Author_Institution :
Istituto di Elettronica, Perugia Univ., Italy
Abstract :
A learning algorithm for bidirectional associative memories (BAMs) is presented, which results in a greatly enhanced storage capacity. The design strategy is formulated as a convex optimisation problem, and then solved by a steepest-descent approach. The proposed method guarantees the storage of all the training pairs as stable states of the BAM. Computer simulation results are presented to demonstrate the performance of the proposed algorithm.<>
Keywords :
content-addressable storage; learning (artificial intelligence); memory architecture; neural nets; optimisation; bidirectional associative memories; convex optimisation problem; design strategy; learning algorithm; stable states; steepest-descent approach; storage capacity; training pairs;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:19931037
