A Gaussian function based chaotic neural network
Zuohan Zhou, Weifeng Shi, Yan Bao, Ming Juan Yang · 2010
In this paper we choose the non-monotonic Gaussian function as activation function of the recurrent neural network to built a Gaussian function based chaotic neural network. The discrete dynamics of this network are discussed to find the proper network parameters, such as weight, bias and input. Numerical simulations demonstrate that this network can exhibit period doubling bifurcations from stationary states to stable period-2 orbits, and even the routes to chaos over certain parameter domains. The parameterized Gaussian function as an iterated map presents abundant dynamic behavior and its application in chaotic neural network may help to improve the global searching ability of the optimization problem.