Model compression method and system based on sparse convolutional neural network, and related equipment

Abstract

The invention relates to the technical field of deep learning, particularly to a model compression method and system based on a sparse convolutional neural network, and related equipment, and aims toreduce resource occupation of a model. The compression method comprises the steps of performing sparse regularization training on a model to obtain a to-be-compressed model; calculating the importancescore of each filter according to the parameters of each convolution layer and the BN layer of the to-be-compressed model; setting an importance threshold according to the importance score and a preset pruning rate; cutting off the filters of which the importance scores are lower than the importance threshold and the BN layer parameters corresponding to the filters to obtain a pruned model; and finely adjusting the pruned model to ensure that the precision of the model is not lower than the preset precision. According to the method, on the premise that the model performance is not affected, large-amplitude compression of the model parameter quantity and the calculated quantity is achieved, the problem that resources are occupied by the model can be well solved, and the deep learning modelcan run on the edge computing equipment with limited resources.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a model compression method, system and related equipment based on a sparse convolutional neural network. Background technique [0002] Deep learning is developing rapidly and has been widely used in computer vision, speech recognition, natural language processing, automatic driving and other fields, and has great application prospects in edge devices of mobile terminals and embedded systems. [0003] As the scale of the deep neural network model continues to grow, training and running the deep learning model requires powerful computing power and a large amount of data storage space. However, the storage space and computing power of edge devices are limited, which makes the deep learning model in resource-constrained environments. Difficult to run on edge devices. Therefore, how to remove redundant parameters in the model and ensure the accuracy of the model is a problem th...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above-mentioned problems in the prior art, the present invention proposes a model compression method, system and related equipment based on sparse convolutional neural network. Under the premise of ensuring the effect of the model, the neural network model is compressed, and the model parameters are solved. It is difficult to run on resource-constrained edge devices due to the large amount of data and computation

Images