遗传优化分布式BP算法

Genetic Optimization Distributed BP Algorithm

针对传统BP算法训练深度学习模型易受模型初始参数影响,训练效率较低的问题,基于遗传算法进行优化,提出了遗传优化分布式BP算法。首先,分析了BP算法和遗传算法的基本原理。然后,结合分布式训练的特点,提出了遗传优化分布式BP算法,确定了训练策略。最后,对遗传优化分布式BP算法和传统BP算法的训练效率进行了对比实验分析。结果表明,遗传优化分布式BP算法不受模型初始参数的影响,相对传统BP算法实现了较高的训练效率。

In this paper,an optimized distributed training algorithm based on genetic algorithm is proposed,which used to solve the problem of easy affected by the initial parameters and low training efficiency of Traditional BP algorithm.Firstly,the basic principles of BP algorithm and genetic algorithm are analyzed.Then,considered with the characteristics of distributed training,the genetic optimization distributed BP algorithm is proposed,and the training strategy is determined.Finally,the training efficiency of genetic optimization distributed BP algorithm is compared with that of traditional BP algorithm.The results show that genetic optimization distributed BP algorithm is not affected by the initial parameters of the model,and achieves higher training efficiency than the traditional BP algorithm.