中国计量大学学报2017，Vol.28Issue(3)：352-358,7.DOI:10.3969/j.issn.2096-2835.2017.03.014
一种加权误差最小化的深度信念网络优化技术
A deep belief network optimization technique based on weighted error minimization
吴强 1杨小兵1
作者信息
- 1. 中国计量大学信息工程学院 ,浙江杭州310018
- 折叠
摘要
Abstract
The traditional deep belief network model lacks parallel and effective algorithm to determine the number of network layers and the number of hidden neurons . Most of the experiments chose them by experience .This makes the model training difficult .This paper chose the number of network layers according to the size of the reconstruction error and the product of the error classification rate of the verification set (weighted errors) .After the network layers were determined ,the number of hidden neurons was selected according to the reconstruction error by the incremental method or dichotomy method . The experimental results show that the number of model network layers and the number of hidden neurons with the method can improve the accuracy of model classification or prediction .关键词
深度信念网络/网络层数/神经元数目/重构误差/加权误差Key words
deep belief network/network layers/number of neurons/reconstruction error/weighted error分类
信息技术与安全科学引用本文复制引用
吴强,杨小兵..一种加权误差最小化的深度信念网络优化技术[J].中国计量大学学报,2017,28(3):352-358,7.
