Associative pattern retrieval by stochastic resonance in FitzHugh-Nagumo neural network

摘要

We report stochastic resonance (SR) in an associative memory network which is composed of a FitzHugh-Nagumo neuron model. The network is modulated by a Gaussian white noise independent among neurons and times, and by a temporally sinusoidal pattern. The modulation pattern is used from a randomly flipped learning pattern which consists of 2 values pattern. We study the pattern retrieval properties by SR when the overlap between the original learning pattern and the modulation pattern decreases. The output of the network is more similar to the learning pattern than the modulation pattern when the overlap is greater than moderate values and the noise variance is in some ranges. When the effect becomes large with large weight, the retrieval region of the learning pattern is extended in the overlap and also contracted in the noise variance. These responses show that the network retrieves the learning pattern even when the modulation pattern is slightly different from the learning pattern. This comes from a cooperative phenomenon by interactions and the retrieval function by the associative memory network.