Method of Selecting an Optimal Activation Function in Perceptron for Recognition of Simple Objects
Regina Latypova, Dmitrii Nikolaevich Tumakov · 2018
Artificial feedforward neural networks for simple objects recognition of different configurations are considered. The novel family of activation functions for neural networks intended for objects recognition is proposed. The method of selecting an optimal activation function from this family is presented. The results on performance evaluation of the activation functions at recognition of handwritten digits are obtained.