Optimization method and optimization device of convolution neural network

Abstract

The invention provides an optimization method of a convolution neural network. The optimization method comprises steps of setting a shortcut connection on an added layer of the convolution neural network and acquiring residual error mapping corresponding to the shortcut connection; according to the residual error mapping, determining expectation mapping corresponding to the shortcut connection; and replacing a layer corresponding to the shortcut connection with the expectation mapping, and carrying out convolution neural network model prediction. According to the invention, parameters of the added layer can be effectively reduced; data circulation between networks is allowed to be quite smooth; and improvement of prediction precision and prediction speed of the model is facilitated.

Description

technical field [0001] The invention belongs to the field of artificial neural networks, and in particular relates to an optimization method and device for a convolutional neural network. Background technique [0002] Convolutional Neural Network (English name is Convolutional Neural Network, English abbreviation is CNN) is a kind of artificial neural network, which has become a research hotspot in the field of speech analysis and image recognition. The weight sharing network structure of the convolutional neural network is similar to the biological neural network, which effectively reduces the complexity of the network model and reduces the number of weights. [0003] With the development of CNN network, especially the proposal of VGG (English full name visual geometry group, Chinese full name: visual geometry group) convolutional neural network, the increase of network layers has become an important research direction of convolutional neural network. However, as the numbe...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a method for optimizing convolutional neural networks to solve the problem in the prior art that the number of parameters of the convolutional neural network increases rapidly due to the increase in the number of network layers, which affects the prediction accuracy and prediction speed of the system.

Images