Efficient training strategy for classification using deep neural network with hybridization of quantum enthused artificial bee colony algorithm

Abstract

Deep Neural Network (DNN) is commonly used in many applications of Artificial Intelligence (AI) like robotics, speech recognition and computer vision. DNN is mainly used as it produces higher classification accuracy and its cost of computational complexity is also less when compared with other classifiers. The DNN training (DNNT) is a process by which weight set has to be optimized in order to get higher classification accuracy. Normally this training is done using Back propagation algorithm (BP) which has several drawbacks like very slow in finding local minima and computational complexity. Hence to overcome these drawbacks an evolutionary algorithm like swarm intelligence has been modified and applied in this proposed method. Normal PSO, ABC algorithm has also many drawbacks which have to be rectified using some hybrid techniques. Hence a new hybrid technique of quantum enthused artificial bee colony (QEABC) has been proposed for training DNN. The parameter setting used for this algorithm has helped a lot to get higher performance. In our proposed hybrid technique the parameter setting is performed automatically to enhance results of classification accuracy and other metrics.