Surrogating neurons in an associative chaotic neural network

A method of surrogate data, which is originally proposed for nonlinear time series analysis, is applied to an associative chaotic neural network in order to see which statistic of the deterministic chaos of the constituent neurons in the network for the dynamical association is important. The original associative network consists of 16 chaotic model neurons whose individuals may exhibit deterministic chaos by themselves. The associative network, whose synaptic weights are determined by a conventional auto-associative matrix of the three orthogonal patterns, shows chaotic retrievals of the stored patterns. The method of surrogation is applied to replace several neuronal sites by the surrogate data to see which statistic of the deterministic chaos of the constituent neurons is important to show the chaotic retrieval. The result shows that the auto-correlation of the time series of the output of the constituent neurons is important for maintaining the chaotic retrieval.