Generalization ability of chaotic complex-valued multidirectional associative memory with adaptive scaling factor

Abstract

In this paper, we propose a Chaotic Complex-valued Multidirectional Associative Memory (CCMAM) with adaptive scaling factor and investigate its generalization ability for network size. The proposed model is based on the conventional CCMAM with variable scaling factor and can realize one-to-many associations of M-tuple multi-valued patterns. In the proposed model, the scaling factor of refractoriness is determined based on not only the time but also the internal states of neurons. We carried out a series of computer experiments and confirmed that the proposed model can determine the scaling factor of refractoriness automatically in various size networks. © Springer-Verlag 2013.