A cosine-modulated Gaussian activation function for hyper-hill neural networks

Abstract

We present here a new class of activation functions for neural networks, which is called the CosGauss function. This function is a cosine-modulated Gaussian function. In contrast to the sigmoidal-, hyperbolic tangent- and Gaussian activation functions, more ridges can be obtained by the CosGauss function. It is proved that this function can be used to approximate polynomials and step functions. The CosGauss function was tested with a cascade-correlation-network on the sonar problem and results are compared with those obtained with other activation functions.