The recollection characteristics of a generalized MCNN

As a dynamic auto-associative memory model, Aihara et al. has proposed a chaotic neural network (CNN) which is consisted by interconnected chaotic neurons is able to recollect stored patterns dynamically. To realize mutual association of plural time series patterns, Kuremoto et al. proposed to combine multiple CNN layers as a MCNN and applied it to a mathematical hippocampus model. However, recollection simulation of MCNN was limited in a two-layer model, and the recollection characteristics concerning with the different external inputs (stimuli) was not investigated. In this paper, we extend the MCNN to be a general form (GMCNN) with more layers and show the recollecting characteristics of different GMCNNs with 2, 3, and 4 CNN layers by computer simulation.