Construction of activation functions for wavelet neural networks

The article describes the procedure of constructing activation functions for wavelet-neural networks. This paper proposes an algorithm for synthesis of wavelets of a given shape for wavelet neural networks. A comparative analysis of different mathematical models of wavelets is conducted. Their complexity and possibility of application in the process of wavelet-neural networks design, modeling and implementation are estimated. The article demonstrates wavelet graphs obtained using these models. The main results indicate the possibility to use the wavelets derived from polynomial models to construct activation functions for these networks as well as the neural network and spline models proposed by the author. The final choice of the mathematical model is determined by the task to be solved and the available computational resource.