Memory dynamics in a recurrent neural network with cycle memories embedded by pseudo-inverse method

Memory dynamics in a recurrent neural network model with cycle memories consisting of similar face patterns embedded by a pseudo-inverse method are numerically investigated from the point of view of harnessing of chaos in a high-dimensional memory system for sampling and search. The model has both non-chaotic and chaotic regimes which are realized by varying a system parameter. Simulations show that there is chaos which is useful in the sense that chaotic attractors are localized near the original attractors, there is intermittency between memory patterns and there is merging of the memory patterns. It is proposed that these properties will be useful for search and synthesis functions.