Neural-network-model compression method, system and device and readable storage medium

Abstract

The invention discloses a neural-network-model compression method, system and device and a computer-readable storage medium. The method includes: utilizing a neural-network clipping method to clip a to-be-clipped neural network model to obtain a to-be-quantized neural network model; utilizing an INQ algorithm to quantize the to-be-quantized neural network model to obtain a to-be-stored neural network model; and using a compression format to store the to-be-stored neural network model. Therefore, according to the neural-network-model compression method provided by the embodiment of the invention, the neural network model is cropped, the INQ algorithm is used at the same time to quantify the same after cropping, thus a model size can be reduced in a case of effectively guaranteeing no loss of model precision after compression, thus the problem of excessive consumed resources can be solved, and calculation can be accelerated.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, more specifically, to a neural network model compression method, system, device and computer-readable storage medium. Background technique [0002] In today's era, whether it is in daily life or in the Internet world, there is a word that cannot be avoided, AI (Artificial Intelligence), that is, artificial intelligence. The application of AI has penetrated into many aspects, such as face recognition, speech recognition, text processing, Go games, game battles, automatic driving, picture beautification, lip reading, and even simulation of stratum fractures. In many aspects, its accuracy and ability to deal with problems have surpassed that of humans, so it has a very broad application prospect and imagination space. Among the algorithm technologies in the field of AI, deep learning has attracted widespread attention from academia and industry since it won the ImageNet competition wi...

AI Technical Summary

Problems solved by technology

[0007] Although the usual model compression method reduces the size of the model and stores the model parameters in the form of a sparse matrix, the accuracy of the model inevitably decreases

In addition, there are also compression methods that adopt the method of retraining the compressed model, which reduces the loss of model accuracy, but the computing performance of the model reasoning prediction is significantly reduced.

Images