Convolutional neural network model parameter processing method and system

Abstract

The invention is suitable for the technical field of artificial intelligence, and provides a convolutional neural network model parameter processing method and system. The method comprises the following steps of extracting weight information of each data layer of a first convolutional neural network model; calculating a threshold of each data layer through a threshold calculation method; carrying out zero setting on weights through a zero setting method so as to obtain a second convolutional neural network model, and recording the total number of the zero-set weights; calculating a weight zero-setting rate; training and testing the second convolutional neural network model, and recording a test result; judging whether the second convolutional neural network model satisfies requirements or not according to the test result and the weight zero-setting rate; if the judging result is positive, outputting the second convolutional neural network model; and if the judging result is negative, returning to the former operations. According to the method and system provided by the invention, the convolutional neural network model parameters are sparse, so that the calculated amount of the convolutional neural network models during the application is decreased and the calculation efficiency is improved.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to a convolutional neural network model parameter processing method and system. Background technique [0002] Convolutional neural network is an efficient recognition algorithm widely used in the field of pattern recognition and image processing in recent years. Its weight sharing network structure makes it more similar to biological neural network, which reduces the complexity of the network. As the input of the network directly, it avoids the complex feature extraction and data reconstruction process in traditional algorithms. [0003] The Convolutional neural network (CNN) model contains a large number of parameters, which not only makes it prone to overfitting problems in training, but more importantly, so many parameters make its application a huge amount of calculation and computational complexity. The problem of inefficiency. Contents of the inv...

AI Technical Summary

Problems solved by technology

[0004] In view of this, an embodiment of the present invention provides a convolutional neural network model parameter processing method and system to solve the huge amount of calculation and low calculation efficiency caused by too many model parameters of the convolutional neural network model in the prior art. The problem

Images