Analysis of Chaotic Neuron Models with Information Theory.

A one-dimensional mapping of a chaotic neuron model which generates deterministic chaos is analyzed from the viewpoint of the information theory. First, the mutual information between the initial value and the value after iterations of the mappings is calculated, and it is shown that the information of the initial value decays almost exponentially. To examine the decaying behavior in detail, the dynamics of the chaotic neuron model is represented as the information flow of‘0’and‘1’in a bitwise space (register) and the behavior of the mutual information contained in each bit is examined. It is shown through this analysis that the information in the chaotic neuron model is maintained, being mixed and distributed among the hits. Moreover, the chaotic neuron model is compared with two typical chaotic dynamical systems with different information structures, i.e., the B-Z mapping and the logistic mapping to clarify the information structure. Finally, it is shown for the system, where the chaotic neuron models are unidirectionally connected as a one-dimensional network, that the information transmission is made possible by its peculiar information structure. [ABSTRACT FROM AUTHOR]