An analysis of associative chaotic neurodynamics by using surrogate neurons

In the present paper, associative chaotic neurodynamics is analyzed by using a method for nonlinear time series analysis. The aim of the analysis is to finding out which statistic of the deterministic chaos of the constituent neurons is important for the chaotic associative neurodynamics. A method comparing features of the original time series with that of artificially made time series preserving some statistics of the original one is applied for the analysis as follows. Some of the constituent neurons in the chaotic neural network are replaced by their surrogate data. The retrieval frequencies of the original network and the network with three surrogate methods, that preserve the dynamic range of the original data, are compared. The results show that not only the auto-correlation in neuronal output of a neuron but also the cross-spectra among the neurons in the network play certain role for maintaining the associative chaotic neurodynamics.