Roles of Fluctuations in Pulsed Neural Networks

Concerning the fluctuation which is observed in biological sensory systems and cortical neuronal networks, the roles of fluctuations in pulsed neural networks are investigated. As a model of a single neuron, the FitzHugh-Nagumo model is used, and two kinds of couplings of neurons are considered, namely, the electrical coupling which is often observed in sensory systems, and the chemical coupling which is widely seen in cortical neuronal networks. The network with electrical couplings with the periodic input shows typical properties of stochastic resonance, which is the phenomenon where a weak input signal is enhanced by its background fluctuations. It is found that the optimal fluctuation intensity which maximizes the correlation coefficient between the input and the output increases with the increase of the coupling strength of the network. Using these properties, the network in which the fluctuations in the system play significant roles in the information processing is proposed. The dependence of the correlation coefficient on the coupling strength of the network is also investigated, and it is found that the correlation coefficient takes a maximum also as a function of the coupling strength. This phenomenon called arrayenhanced stochastic resonance is analyzed theoretically. In the network with chemical couplings, the associative memory in the network is considered, and it is found that the memory retrieval is induced by the fluctuations in the system. Besides, for the network storing sparse patterns with hierarchical correlations, it is observed that the retrieved pattern is selected by controlling the fluctuation intensity. Based on the results for both couplings, it is claimed that the fluctuations in the pulsed neural network might have beneficial effects on the information processing in the neural system.