摘要
针对相关向量机的性能易受到奇异值影响的情况,提出了一种增强相关向量机稳健性的方法。其主要思想如下:首先用原始训练数据训练相关向量机;然后,利用某种准则,从原始数据中挑选一些样本,用其预测值代替输出变量值;随后,用改变后的训练样本重新训练相关向量机。这个过程可重复几次。数据试验表明,较之相关向量机和变分稳健相关向量机,新算法对奇异值更加不敏感。
Relevance vector machine（RVM）is seriously affected by examples with gross error.A method is proposed to enhance the robustness of RVM.Its main idea is as follows：after training a RVM on the original data set,the target values with large prediction errors are replaced by the predictive values of the RVM,and next RVM is trained on the new training data set.This process is done several times recursively.Some experiments are conducted and the results demonstrated the proposed approach is less sensitive to outliers in comparison with RVM and variational robust relevance vector machine（VRRVM）,while reserving the sparseness of RVM.
出处
《软件》
2012年第6期1-5,共5页
Software
基金
陕西省教育厅2009年科学技术研究计划基金(项目编号:09JK615)
关键词
人工智能
支撑向量机
相关向量机
稀疏性
稳健性
奇异值
Artificial Intelligence
Support Vector Machine
Relevance Vector Machine
Sparseness
Robustness
Outlier
