A Convolutional Neural Network Structure Search Method and System Based on Evolutionary Algorithm

Abstract

The invention discloses a convolution neural network structure searching method and a system based on an evolutionary algorithm. The method comprises the steps of inputting a data set and setting preset parameters to obtain an initial population. By the controller TC as the main thread popping the initial population into the queue q and opening the queue manager tq and the message manager tm, After the queue manager TQ is turned on, the untrained chromosomes in the queue Q are pop-up and decoded, and a worker manager TW as an independent temporary thread is started to calculate the fitness forthe training, Through the cooperation of controller TC, queue manager TQ, worker manager TW and message manager TM, the parallel search of convolution neural network structure based on evolutionary algorithm is completed and the best model is output. The invention can realize automatic modeling, parameter adjustment and training for a given data set, has the advantages of high performance, large-scale, flow-chart and good certainty, and is particularly suitable for deployment and implementation on a high-performance computer cluster.

Description

technical field [0001] The invention relates to the field of model design of deep learning, in particular to a convolutional neural network structure search method and system based on an evolutionary algorithm, which is a solution for automatic modeling, parameter adjustment and training for a given data set. Background technique [0002] Due to the rapid development of computer software and hardware, the computing power and storage space of equipment have been greatly improved, and many large-scale data sets have emerged. These conditions make it very effective to use a deep convolutional neural network to process large-scale image data sets. Among them, excellent models such as AlexNet, VGGNet, ResNet, and GoogleNet have emerged. Each model sets a fixed operator combination structure. The current convolutional neural network mainly includes operators such as convolution, pooling, and full connection. Many operators need to set some precise parameters. For example, a convol...

AI Technical Summary

Problems solved by technology

[0007] The technical problem to be solved by the present invention: In order to solve the shortcomings of manual design models that cannot be scaled, streamlined, and uncertain, it provides a convolutional neural network structure search method and system based on evolutionary algorithms, which can target a given data set Realize automatic modeling, parameter adjustment and training, with the advantages of high performance, scale, process, and good determinism, especially suitable for deployment and implementation on high-performance computer clusters

Images