REDUCING ERROR RATE OF DEEP LEARNING USING AUTO ENCODER AND GENETIC ALGORITHMS

Deep Learning (DL) techniques are considered as one of machine learning classes that model hierarchical abstractions in data input with the assistance of multiple layers. DL techniques have accomplished high performance in computer vision, natural language processing and automatic speech recognition. DL combines lower modules for classifier output and raw features input to produce new features at hierarchy higher layer. Deep Auto Encoder (DAE) is a DL aims to represent data to be utilized for rebuilding and classification. It is considered as one of the powerful algorithms in DL that gives higher accuracy and best performance. The proposed method in this work is based on using DAE and Genetic Algorithm (GA) through applying split-training and merging algorithms for DL. First, the network is divided into two initialized networks using DAE. Second, both of these networks were merged using GA. This proposed approach was evaluated based on the Mixed National Institute of Standards and Technology (MNIST) dataset and the obtained results showed that it achieve a higher performance and lower error rate in the classification.