Training Deep Autoencoder via VLC-Genetic Algorithm

Abstract

Recently, both supervised and unsupervised deep learning techniques have accomplished notable results in various fields. However neural networks with back-propagation are liable to trapping at local minima. Genetic algorithms have been popular as a class of optimization techniques which are good at exploring a large and complex space in an intelligent way to find values close to the global optimum. In this paper, a variable length chromosome genetic algorithm assisted deep autoencoder is proposed. Firstly, the training of autoencoder is done with the help of variable length chromosome genetic algorithm. Secondly, a classifier is used for the classification of encoded data and compare the classification accuracy with other state-of-the-art methods. The experimental results show that the proposed method achieves competitive results and produce sparser networks.