Parallelization method of BP neural network optimized by genetic algorithm based on spark

Abstract

The invention provides a parallelization method of a BP neural network optimized by genetic algorithm based on spark. The method comprises the following steps: performing global evolution optimizing on a weight value of the BP neural network through the adoption of a spark parallelization programming model improved genetic algorithm, and obtaining an optimized neural network initial weight value after performing a certain number of evolution iterations, and iterating through the adoption of the parallelized BP neural network algorithm, finally outputting a network structure. In the training process, the multi-node parallel processing can be performed in each stage, the convergence speed of the BP neural network is greatly promoted, and the training efficiency is improved.

Description

technical field [0001] The invention relates to the field of machine learning algorithm parallelization, in particular to a BP neural network parallelization method optimized by a genetic algorithm based on a spark distributed computing framework. Background technique [0002] BP (Back Propagation) neural network was proposed by a team of scientists headed by Rinehart and McClelland in 1986. It is a multi-layer feed-forward network trained by the error back propagation algorithm. The main idea of ​​BP neural network, including forward propagation signal and back propagation error. In the forward propagation process, the input signal is passed to the output layer after being processed by the hidden layer. If the output value is not equal to the expected value and is greater than the acceptable range of error, enter the error backpropagation process. The error is transmitted to the input layer through the hidden layer for error adjustment. By continuously adjusting the weig...

AI Technical Summary

Problems solved by technology

[0004] The BP neural network algorithm has the shortcomings of slow convergence speed and easy to fall into local minimum points. Based on the BP neural network optimized by the genetic algorithm, the weight and threshold of the BP network are first optimized by the genetic algorithm to improve the convergence rate of the network and overcome the tendency to fall into the local minimum. local minimum deficiency

[0005] The traditional BP neural network training method is to serially process data sets on a single machine, but with the rapid development of the information society, the amount of data that needs to be mined has increased sharply, reaching the level of massive data, so the traditional BP neural network The neural network training method will have big problems when dealing with massive data sets, such as taking a long time, or even insufficient memory to train, etc.

Images