增强碰撞体算法优化的自编码神经网络  

Autoencoder Neural Network Optimized by Enhanced Colliding Bodies Optimization

作　　者：冯文 陈志国[1] 傅毅[1] 王凯宇 FENG Wen;CHEN Zhi-guo;FU Yi;WANG Kai-yu(School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《小型微型计算机系统》2019年第4期721-725,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61502203)资助;江苏省自然科学基金项目(BK20150122)资助;江苏省高等学校自然科学基金项目资助

摘　　要：自编码神经网络是神经网络中常见的网络,自编码网络常用于数据降维,特征学习,数据去噪等.传统的自编码网络采用的BP策略,为了应对BP算法固有的一些不足,本文结合了增强碰撞体算法的优点,提出了一种增强碰撞体算法优化的自编码网络.将自编码的代价函数和softmax分类器的误差函数加权求和作为该算法的评价函数,利用增强碰撞体算法分别优化自编码网络和softmax分类器的参数.实验结果表明与其他算法和粒子群算法优化的自编码算法相比,该方法在邮件分类上取得了较好效果,与极限学习机优化的栈式自编码相比在UCI的一些公共数据库上取得了较好的分类效果.Autoencoder neural network are often used in neural network,it is used in dimension reduced,feature learning,data denoising and so on. BP algorithm has been used in traditional neural network,but there are some shortages with BP. To solve these shortages,in this paper we combine the advantages of Enhanced colliding Bodies Optimization with Autoencoder neural network. The Autoencoder neural network optimized by Enhanced Colliding Bodies Optimization has presented. We used weighted summation of cost function for Autoencoder network and error function for softmax as evaluation function to optimize the weights of Autoencoder network and Softmax respectively which optimized by Enhanced Colliding Bodies Optimization. The results of experimental show that compared with other methods and Autoencoder network based on PSO,this method has better results in Email classification. and get better result than Extreme Learning Machine( ELM) Optimized Deep Autoencoder Classification Algorithm in some public database of UCI.

关 键 词：自编码网络 神经网络 增强碰撞体优化算法 分类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]