基于二次Renyi熵的神经网络集成  

Neural Network Ensemble Based on Quadratic Renyi Entropy

作　　者：刘李飞[1] 邢红杰[1] 

机构地区：[1]河北大学数学与计算机学院,河北保定071002

出　　处：《电脑编程技巧与维护》2014年第6期68-71,共4页Computer Programming Skills & Maintenance

基　　金：国家自然科学基金项目(60903089);河北省自然科学基金项目(F2013201060)资助

摘　　要：神经网络集成能够有效提高单个神经网络的分类或回归性能。在基于香农熵的神经网络集成基础上,提出了基于二次Renyi熵的神经网络集成。与基于香农熵的神经网络集成相比,所提集成策略的最优组合权重可以解析求得,无需利用牛顿法迭代求取。在人工和标准数据集上的实验结果表明:与单个神经网络和基于香农熵的神经网络集成相比,基于二次Renyi熵的神经网络集成能够取得更优的性能。Neural network ensemble can enhance the classification and regression performance of single neural network. In this paper, the quadratic Renyi entropy based neural network ensemble is proposed basing on the Shannon entropy based neural network ensemble. In comparison with the Shannon entropy based neural network, the proposed ensemble strategy can analytically deduce the optimal combination weights rather than iteratively seek by the Netwon’ s method. Experimental results on several synthetic and benchmark data sets demonstrate that the quadratic Renyi entropy based neural network ensemble can obtain better performance compared to the single neural network and the Shannon entropy based neural network ensemble.

关 键 词：神经网络集成 多层感知器 二次Renyi熵 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]