Abstract
In the present paper, we investigate roles of the randomness of memory patterns in the sensitive response of the chaotic associative memory dynamics to memory pattern fragments in the chaotic neural network model referred to as CNN hereafter. In order to realize a memory search for hierarchical memory patterns, we overcome the problem how to construct the hierarchical memory patterns, whose basin volumes and visiting measures are sufficiently large. Therefore, we investigate (i) how to construct the memory patterns which gives sufficiently large basin volumes of theirs in a recurrent neural network model referred to as RNN hereafter, and (ii) the sensitivity of the chaotic associative memory dynamics in CNN to memory pattern fragments, focusing on the randomness in the memory patterns. From computer experiments, the basin volumes of the memory patterns become much larger as the randomness increases. In addition, the sensitive and robust response to the memory pattern fragments is achieved as the randomness becomes larger. Thus, ensuring sufficient large basin volumes and visiting measures with the same frequency, and the quite sensitive and robust response to the memory pattern fragments, the randomness in memory patterns is practical, which introduces the small overlap among each inter-cycle pattern.
