基于粒子群优化的深度随机神经网络  

Deep random neural network based on particle swarm optimization

作　　者：凌青华[1,2] 宋余庆[1] 韩飞[1] LING Qinghua;SONG Yuqing;HAN Fei(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212013,China;School of Computer Science,Jiangsu University of Science and Technology,Zhenjiang 212003,China)

机构地区：[1]江苏大学计算机科学与通信工程学院,镇江212013 [2]江苏科技大学计算机学院,镇江212003

出　　处：《江苏科技大学学报（自然科学版）》2020年第1期69-74,共6页Journal of Jiangsu University of Science and Technology:Natural Science Edition

基　　金：江苏省普通高校研究生科研创新计划项目(KYLX_1056)。

摘　　要：针对传统多层随机神经网络性能不稳定问题,提出了一类利用粒子群优化算法来优化各层权值的深度随机网络方法.该方法利用粒子群优化算法,结合网络的输入输出灵敏度信息,逐层对自动编码器的输入层权值进行优化,通过改善自动编码器的性能来改善多层随机神经网络的性能.最后利用粒子群优化方法,对整个网络的权值作适当优化,进一步提高深度随机神经网络的性能.相对于传统深度学习算法,该方法在保持收敛精度的基础上降低了时间开销;相对于传统深度随机神经网络,该方法在增加时间开销基础上提高了收敛精度,从而较好地平衡了时间复杂度和收敛精度.A deep random network based on particle swarm optimization(PSO)is proposed to optimize the weight of each layer in this paper.Combined with the input-to-output sensitivity information of the auto encoder,this method uses particle swarm optimization algorithm to optimize the weights of the input layer of the auto encoder layer by layer.Thus,the performance of the multi-layer random neural network will be improved by improving the performance of the auto encoder.Finally,the weights of the whole network are moderately optimized by PSO to further improve the performance of the deep random neural network.Compared with the traditional deep learning algorithm,the proposed method maintains the convergence accuracy and reduces the time cost.Compared with the traditional deep random neural networks,the proposed method improves the convergence accuracy on the basis of increasing the time cost,and the time complexity and precision are well balanced.

关 键 词：随机神经网络 深度自动编码器 粒子群优化 超限学习机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]