A novel chaos associative memory

Proposes a chaos neural network model that is applied to a chaotic autoassociation memory. This artificial neuron model is properly characterized in terms of a time-dependent periodic activation function that involves chaotic dynamics as well as the energy steepest-descent strategy. This neural network has a remarkable ability for dynamic memory retrieval, beyond that of the conventional models, by using a nonmonotonic activation function as well as a monotonic one (such as a sigmoidal function). This advantage is found to result from the property of the analogue periodic mapping accompanied by a chaotic behaviour of the neurons. It is also concluded that the present analogue neuron model with periodicity control has an apparently larger memory capacity as compared to the previously proposed association models.