Deep neural network model pruning method, system and device and medium

Abstract

The invention discloses a deep neural network model pruning method, system and device and a medium, and the method comprises the steps: carrying out the sparse training of a to-be-pruned model, and obtaining a sparse model; wherein the to-be-pruned model is a deep neural network model of a depth separable convolution; the depth separable convolution comprises a depth-wise convolution and a point-wise convolution, and the depth separable convolution comprises a depth-wise convolution and a point-wise convolution; on the basis of an importance evaluation result of a weight absolute value of each channel in point-wise convolution, pruning a convolutional layer channel of the sparse model to obtain a pruned network model; and performing fine tuning training on the weight of the pruned network model, and outputting the fine-tuned network model, thereby obtaining the pruned deep neural network model. According to the method, the sparsification of the point-wise convolution weight is realized, the network precision is ensured, and meanwhile, the calculation amount and the parameter amount of the model can be effectively reduced.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and particularly relates to a deep neural network model pruning method, system, equipment and medium. Background technique [0002] At present, artificial intelligence technology with convolutional neural network as the core technology has made a series of breakthroughs, and it is gradually applied to weapon equipment and various types of spacecraft to achieve accurate detection of satellite-based on-orbit targets and intelligent target recognition of missiles. Strike, autonomous obstacle avoidance, mission planning and other applications; but the deep neural network models that the above applications rely on usually have many model parameters and a large amount of calculation; in most practical application scenarios, the computing unit of the neural network model in AI application embedded devices is in The volume and power consumption are often limited, resulting in too many mod...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the technical problems existing in the prior art, the present invention provides a deep neural network model pruning method, system, equipment and medium to solve the problem that most of the existing model pruning methods are aimed at the traditional convolution, and there is no depth-capable Technical issues with pruning methods for deep network models that separate convolutions

Images