Abstract: Deep neural networks are widely used in computer vision and speech processing, but the large number of parameters and computation of deep neural network models make it difficult to deploy them in embedded applications with limited resources. The network pruning can effectively compress and accelerate the deep neural network models without affecting the computational accuracy, and therefore has become a hot research topic. This paper discusses the problems and theories related to pruning of deep neural networks, and summarizes the pruning methods for deep neural networks in recent years, classifies the current main pruning methods, and provides the future development of the network architecture search combination.