Adaptation of Number of Filters in the Convolution Layer of a Convolutional Neural Network Using the Fuzzy Gravitational Search Algorithm Method and Type-1 Fuzzy Logic

Abstract

This paper presents a model of the search for adaptation of parameters and the creation of the membership functions of various fuzzy systems created using the fuzzy gravitational algorithm (FGSA). These fuzzy systems were created to find the optimal number of filters to enter a convolutional neural network (CNN) with an architecture of two convolution layers, as well as two pooling layers respectively and a classification layer, which is responsible for recognizing images. With this model, the results obtained by optimizing this CNN with the FGSA algorithm and the adaptation of parameters using this same algorithm are compared to form the membership functions of fuzzy systems. Both methods and their results are comparing with each other.