Associative memory with a controlled chaotic neural network

Abstract

A chaotic neural network (CNN) composed of chaotic neurons exhibits chaotic associative dynamics for some values of the parameters, which indicate that the CNN is a promising technique that can be applied to information processing such as pattern recognition and memory recalling. However, the outputs of the CNN in those parameter areas wander around all the stored patterns and cannot be stabilized to one of the stored patterns or a periodic orbit because of its chaotic behavior. Furthermore, it is also difficult to judge when the chaotic dynamics must be terminated, thereby hampering the application of the chaotic associative dynamics of CNNs to information processing. In this paper, we propose a novel chaos control scheme, i.e. a parameter modulated control method, for the CNN, which is applied particularly for associate memory. This scheme can be viewed as a type of adaptive control method in which the refractory scaling parameter decreases by the addition of a delay feedback control signal to the network. By means of this control method, the outputs of the controlled CNN converge to the periodic orbits and are dependent on the initial patterns. We observed that the controlled CNN can distinguish two initial patterns even if they have a small difference. This implies that such a controlled CNN can be feasibly applied to information processing such as pattern recognition.