Improved Chaotic Multidirectional Associative Memory

In this paper, we propose an Improved Chaotic Multidirectional Associative Memory (ICMAM). The proposed model is based on the Chaotic Multidirectional Associative Memory (CMAM) which can realize one-to-many associations. In the conventional CMAM, the one-to-many associative ability is very sensitive to chaotic neuron parameters. Moreover, although the Chaotic Multidirectional Associative Memory with adaptive scaling factor of refractoriness can select appropriate scaling factor of refractoriness \(\alpha \) based on internal states of neurons automatically, their one-to-many association ability is lower than that of well-tuned Chaotic Multidirectional Associative Memory with variable scaling factor of refractoriness when the number of layers is large. In the proposed model, one-to-many association ability which does not depend on the number of layers is realized by dividing internal states of neurons by the number of layers. We carried out a series of computer experiments in order to demonstrate the effectiveness of the proposed model, and confirmed that the one-to-many association ability of this model almost equals to that of well-tuned Chaotic Multidirectional Associative Memory with variable scaling factor of refractoriness even when the number of layers is large.