Quick evolution method for optimized deep convolution neural network structure

Abstract

The invention discloses a quick evolution method for an optimized deep convolution neural network structure. The method comprises the following steps that: 1) utilizing an evolution algorithm based onGNP (Gene Network Coding) to effectively construct a nonlinear CNN (Convolutional Neural Network) structure, and carrying out mutation on various hyper-parameters of the CNN structure to search an optimal CNN hyper-parameter combination; 2) in an evolution process, designing a multi-objective network structure evaluation method, simultaneously taking classification accuracy and the complexity degree of the classifier as an optimization objective so as to aim at generating the CNN classifier with high classification accuracy and a simple structure; and 3) putting forwarding an incremental training method, and carrying out filial generation CNN structure training on the basis of a previous generation of CNN structure. By use of the method, the training frequency of the model can be reduced,and the time complexity of an algorithm is lowered.

Description

technical field [0001] The invention relates to a rapid evolution method for optimizing the structure of a deep convolutional neural network. Background technique [0002] With the rapid development of science and technology, the era of big data has arrived. Deep learning uses deep neural network (DNN) as a model, and has achieved remarkable results in many key areas of artificial intelligence, such as image recognition, enhanced learning, and semantic analysis. As a typical DNN structure, convolutional neural network (CNN) can effectively extract the hidden layer features of images and accurately classify images, and has been widely used in the field of image recognition in recent years. In 1998, LeCun et al. proposed the LeNet-5 convolutional network structure, which is considered a milestone in the history of deep learning. LeNet-5 can recognize handwritten images composed of 32×32 pixels, but due to the relatively simple structure of LeNet-5 and the lack of computing p...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the shortcomings of the existing evolutionary CNN structure algorithm, such as high time complexity and a single evaluation index for the CNN model, the present invention provides a rapid evolution of an optimized deep convolutional neural network structure with low time complexity and reasonable evaluation indexes. method, using the GNP-based evolutionary algorithm to effectively construct a nonlinear CNN network structure, and mutate various hyperparameters of the CNN structure to find the optimal combination of CNN hyperparameters; in the evolution process, the algorithm designs a multi- The network structure evaluation method of the target can effectively simplify the network structure and enable the network structure to achieve better classification results; finally, the algorithm also proposes the concept of incremental training, and carries out the sub-generation CNN structure on the basis of the previous generation CNN structure. training, this method can effectively reduce the training times of the model and reduce the time complexity of the algorithm

Images