计算机工程 ›› 2012, Vol. 38 ›› Issue (9): 205-207.doi: 10.3969/j.issn.1000-3428.2012.09.062

• 人工智能及识别技术 • 上一篇    下一篇

基于快速估计的相关向量机优化算法

赵 榈,苏一丹,覃 华   

  1. (广西大学计算机与电子信息学院,南宁 530004)
  • 收稿日期:2011-07-12 出版日期:2012-05-05 发布日期:2012-05-05
  • 作者简介:赵 榈(1986-),男,硕士研究生,主研方向:数据挖掘,网络安全;苏一丹,教授、博士;覃 华,副教授、博士
  • 基金项目:
    国家自然科学基金资助项目(61063032)

Optimized Algorithm of Relevance Vector Machine Based on Rapid Estimation

ZHAO Lv, SU Yi-dan, QIN Hua   

  1. (School of Computer & Electronics Information, Guangxi University, Nanning 530004, China)
  • Received:2011-07-12 Online:2012-05-05 Published:2012-05-05

摘要: 针对相关向量机在大规模数据集上训练速度较慢的问题,提出一种基于快速估计的相关向量机优化算法。利用阈值系数、约减最大上限并结合迭代估计,对训练样本的超参进行快速预估计,去除训练集中大量的非相关向量,减小训练样本规模,减少训练时间。在UCI等数据集上的实验结果表明,该算法在保持训练精度的同时具有更快的训练速度。

关键词: 相关向量机, 阈值, 最大上限, 迭代估计, 超参

Abstract: To further improve the efficiency of training on large sets in Relevance Vector Machine(RVM), this paper proposes an optimized algorithm based on rapid estimation. That is preprocessing the training set and excluding the non-relevance vector with iterative estimation integrated by threshold factor and maximum reduction toplimit, then decreases the sample scale and reduces the training time. Experimental results on UCI datasets illustrate that this algorithm has faster training rate while maintaining training precision.

Key words: Relevance Vector Machine(RVM), threshold, maximum toplimit, iterative estimation, hyperparameter

中图分类号: