Simulation Research on Influence of Neural Networks Scale on its Performance

Abstract

When using deep learning methods to deal with specific problems, the structure and configuration complexity of neural networks model have a great impact on the performance of the system. This paper constructed a multi-layer perceptual neural networks model basing on classification problem of film and television reviews. The paper also analyzed the forward and back propagation, processing algorithm and key parameter optimization method, and built the simulation environment. By adjusting the network depth and the number of neurons in each layer, the paper analyzed the influence of the complexity of the neural networks model on the network performance. Then the paper gave the analysis results under the condition of different model scale. Finally the paper purported conclusion in the last section that the performance of the neural networks model changes greatly with number of neurons in each layer changed, but the performance changes little with network depth changed. It can provide an important reference for two classification problems in different fields.