Convolutional neural network model pruning method based on structure redundancy detection

Abstract

The invention provides a convolutional neural network model pruning method based on structure redundancy detection, and the method comprises the steps: firstly training a convolutional neural networkmodel on a training set, and evaluating the convolutional neural network model on a verification set; and then trying to prune different substructures of the neural network model and finely adjustingother structures to detect the redundancy of the pruned substructures, and in each iteration, if the pruned neural network model cannot obtain most lost precision again through fine adjustment, restoring the pruned structure; repeating the method until the redundancy of all substructures of the model is checked; and finally, reinitializing and training the pruned model on the training set and theverification set to obtain a final optimization model. According to the method, the resource consumption of the neural network is reduced, and the problem that an existing model pruning method is complex to implement is solved.

Description

technical field [0001] The invention belongs to a compression technology in the technical field of deep convolutional neural network models, and in particular relates to a convolutional neural network model pruning method based on structural redundancy detection, in which the convolutional neural network model is pruned through the model pruning method Compress and accelerate model derivation. Background technique [0002] Deep learning has achieved great success in areas such as computer vision, speech, and natural language processing. It has been applied in image recognition, target detection, semantic segmentation, pedestrian detection, pedestrian re-identification, face detection, face recognition, speech recognition, language translation and other tasks, and achieved good results. The deep convolutional neural network model occupies a lot of hard disk storage, memory bandwidth and computing resources, and there is a trend towards deeper network layers and more network ...

AI Technical Summary

Problems solved by technology

[0007] The present invention aims at the shortcomings of the convolutional neural network model pruning method in the prior art, and provides a convolutional neural network model pruning method based on structural redundancy detection. The method adopts three steps to reduce the resource consumption of the neural network, Improve the complex problem of the existing model pruning method, thus simplifying the complexity of the model pruning method and improving the pruning performance

Images