Optimization Method of Deep Learning Algorithm Based on Extreme Learning Machine

The Traditional global optimization algorithms is generally explored in whole deep belief network (DBN) with consuming amount of time, while the optimal method based on gradient used in DBN tends to local optimization. For solving this issue, in this paper, we propose a novel method which use extreme learning machine (ELM) into the training procedure of DBN to accelerate the learning process of DBN. Extensive experiment on the three widely available datasets demonstrate that our proposal effectively improves the accuracy and efficiency of the learning process of DBN, the result shows that our approach has a higher learning speed with a reliable learning accuracy.