| 基于概率神经网络和K-L散度的样例选择 |
| |
| 引用本文： | 翟俊海,李 畅,李 塔,王熙照.基于概率神经网络和K-L散度的样例选择[J].计算机应用研究,2014,31(1):63-65. |
| |
| 作者姓名： | 翟俊海 李 畅 李 塔 王熙照 |
| |
| 作者单位： | 河北大学 数学与计算机学院 河北省机器学习与计算智能重点实验室, 河北 保定 071002 |
| |
| 基金项目： | 国家自然科学基金资助项目（61170040）; 河北省自然科学基金资助项目（F2013201220, F2013201110）; 河北大学自然科学基金资助项目（2011-228043）; 河北大学教育教学改革研究项目（JX07-Y-27） |
| |
| 摘 要： | 提出了一种基于概率神经网络和K-L散度的样例选择算法。该算法利用概率神经网络估计训练样例的概率分布, 利用K-L散度作为启发式来进行样例选择, 用该方法选出的样例大多分布在分类边界附近。与五个著名的样例选择算法CNN、ENN、RNN、MCS和ICF进行了实验比较, 实验结果显示, 算法的选择比更低, 训练出分类器具有更好的泛化能力, 提出的方法是有效的。
|
| 关 键 词： | 概率神经网络 样例选择 K-L散度 最近邻分类 |
Instance selection based on probabilistic neural network and K-L divergence |
| |
| ZHAI Jun-hai,LI Chang,LI T,WANG Xi-zhao.Instance selection based on probabilistic neural network and K-L divergence[J].Application Research of Computers,2014,31(1):63-65. |
| |
| Authors: | ZHAI Jun-hai LI Chang LI T WANG Xi-zhao |
| |
| Affiliation: | Hebei Province Key Laboratory of Machine Learning & Computational Intelligence, College of Mathematics & Computer Science, Hebei University, Baoding Hebei 071002, China |
| |
| Abstract: | This paper proposed an instance selection method based on probabilistic neural network and K-L divergence. Firstly, it employed the probabilistic neural network to estimate the probabilistic distribution of training samples, used the K-L divergence as heuristic to select instances, and distributed most of the selected instances with the proposed method near the class boundary. It experimentally compared the proposed method with five famous instance algorithms which were CNN, ENN, RNN, MCS and ICF, much lower selection ratios could be achieved and better generalization ability could be obtained with the classifier trained with the selected instances. The experimental results show that the proposed method is effective and efficient. |
| |
| Keywords: | probabilistic neural network instance selection K-L divergence nearest neighbor classification |
|
| 点击此处可从《计算机应用研究》浏览原始摘要信息 |
|
点击此处可从《计算机应用研究》下载全文 |
|
