摘要
针对传统BP神经网络存在学习率设置不当与深度神经网络过拟合导致准确率不高的问题,提出改进的BP神经网络算法。该算法引入了drop-out机制来防止神经网络过拟合,并针对学习率设置不当的问题,将用指数衰减学习率代替传统BP神经网络中固定学习率。实验结果表明,改进后的BP神经网络相较于传统BP神经网络有效地提高了3.06%的测试准确率。
In order to solve the problems of improper learning rate setting and lowaccuracy caused by over-fitting of neural network in traditional BP deep neural network,an improved BP neural network algorithm is proposed.In this algorithm,drop-out mechanismis introduced to prevent neural network fromoverfitting,and in order to solve the problemof improper learning rate,exponential attenuated learning rate will be used instead of fixed learning rate in traditional BP neural network.the experimental results showthat compared with the traditional BP neural network,the improved BP neural network can effectively improve the test accuracy by 3.06%.
作者
吴鹏程
刘娅
李少夫
Wu Pengcheng;Liu Ya;Li Shaofu(College of Electrical and Electronic Engineering,Wuhan Polytechnic University,Wuhan 430023,China)
出处
《信息通信》
2019年第10期39-41,共3页
Information & Communications
基金
湖北省自然科学基金项目（No.2017CKB893）
作者简介
吴鹏程(1997-),男,学士,主要研究方向:图像识别;刘娅(1974-),女,副教授,主要研究方向:信息与通信系统。
