Improved Deep Learning Optimization Algorithm Based on Variable Speed Integral Control

The performance of neural network is closely related to the optimization algorithm. An improved SWATS algorithm is designed to solve the non-convergence problem and the local optimum oscillation problem. Adamod algorithm and a piecewise function are introduced into the SWATS algorithm to improve the neural network performance. Experiments show that the proposed algorithm in this paper is applied to different model structures and different data sets. Compared with the SWATS algorithm, our algorithm has higher accuracy and converges more quickly and stably.