Neural network construction system and method based on variable length gene genetic algorithm, and storage medium

Abstract

The invention discloses a neural network construction system and method based on a variable length genetic algorithm, and the construction system comprises a training set generation module, a network initialization module, a network training module, a network updating module, a network optimization module, a network selection module, an iteration number updating module and a network generation module which are connected in sequence. According to the construction method of the scheme, the general thought is that variable-length chromosomes are generated in the initialization step, and BN components are randomly added to specific genes of all the chromosomes. And training on a training set to obtain a decoded neural network structure expression corresponding to each chromosome, and selecting progeny chromosomes. Subsequently, in the crossover step, the length of the crossover chromosome is not fixed, and growth and contraction strategies are utilized to generate progeny chromosomes. Then, filial generation chromosome selection is completed through traditional variation and environment selection operation, the above steps are repeated, and the obtained optimal chromosome is decoded into a corresponding neural network architecture.

Description

technical field [0001] The invention relates to the technical field of computer network structure search in deep learning, in particular to a neural network construction system, method and storage medium based on a variable-length genetic algorithm. Background technique [0002] With the development of deep learning, convolutional neural network (CNN) is widely used in computer vision tasks. Compared with standard neural networks, CNN's feature sharing mechanism helps CNN reduce the number of weights and variables that need to be learned, thereby reducing the complexity of the network and improving the image generalization ability. Thanks to the great efforts of medical and data engineering experts and well-designed CNNs, significant progress has been made in automatically analyzing CT images and predicting whether they are positive for COVID-19, reducing the burden of analyzing CT images for medical professionals. [0003] Although deep learning technology has been well de...

AI Technical Summary

Problems solved by technology

The neural network architecture designed based on these data sets will bring model bias, resulting in inaccurate application to other data sets

Images