Implementation of Parallel Evolutionary Convolutional Neural Network for Classification in Human Activity and Image Recognition

Abstract

Pattern recognition has been evolving to include problems posed by new sceneries containing a high number of pattern components. Processing this volume of information allows a more exact classification in wider types of applications; however, some of the difficulties of this scheme is the maintenance of numerical precision and mainly the reduction of the execution time. During the last 15 years, several Machine Learning solutions have been implemented to reduce the number of pattern components to be analyzed, such as artificial neural networks. Deep learning is an appropriate tool to accomplish this task. In this paper, a convolutional neural network is implemented for recognition and classification of human activity signals and digital images. It is achieved by automatically adjusting the parameters of the neural network through genetic algorithms using a multiprocessor and GPU platform. The results obtained show the reduction of computational costs and the possibility of better understanding of the solutions provided by Deep Learning.