Spatio-temporal summation and self-organization in chaotic neural networks

It has been considered to be difficult for an asynchronous network without time-delayed signal transmission to learn and retrieve spatio-temporal patterns. We propose a neural network model composed of chaotic neurons with a leaky integrator at the synapse and show that it is capable of both learning and retrieving spatio-temporal patterns. The leaky integrator at the synapse works to store mutual-correlation of the patterns in the sequence as well as auto-correlation. We also utilize the feature of the chaotic neural network which jumps out of stable states and does the chaotic wandering among the stored patterns.