A novel hysteretic neural network and its application

To improve the optimization performance of hysteretic chaotic neural networks, a hysteretic chaotic neural network with the properties of characterizing local detailed features and stochastic chaotic simulated annealing (SCSA) is proposed. A hysteretic activation function, which can provide hysteretic dynamics for neural networks, is composed of two offset sigmoid activation functions. The exponentially decaying dilation parameter in wavelet is used for chaotic simulated annealing, and the stochastically varying translation parameter in wavelet is used to construct stochastic simulated annealing. Phenomena of chaos and hysteresis make the network escape from local minima, while the properties of SCSA and the local characterizing ability of wavelet make the network contribute to improving the probability of finding the global minima. The experimental results show that it has a higher probability to obtaining a global optimization solution.