Search for cyclic patterns by a chaotic neural network

Abstract

When a chaotic neural network learns a pattern by associative memory of autocorrelative type, it can recall the acyclic pattern sequence, including the learned patterns. The authors have discussed such neural networks and have reported that a search of associative memory can be executed by utilizing presynaptic inhibition to control the extent of chaos. In this paper, an associative memory search by chaotic neural network is applied to cyclic memory content based on cross-correlation associative memory, and a successful search is demonstrated. Nara and others have reported that a chaotic situation can be generated in cyclic associative memory by adjusting the number of synaptic connections in the cyclic associative memory, and that search of the cyclic associative memory can be executed. From this perspective, the authors use the same cyclic pattern as was used by Nara and coworkers for the same search by a chaotic neural network, and the results are compared. It is shown that search by a chaotic neural network can achieve a better success rate, although the search is slower. © 1998 Scripta Technica, Syst Comp Jpn, 29(11): 1–8, 1998