作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2006, Vol. 32 ›› Issue (21): 177-179,. doi: 10.3969/j.issn.1000-3428.2006.21.062

• 人工智能及识别技术 • 上一篇    下一篇

基于支持向量的分层并行筛选训练样本方法

文益民   

  1. (湖南工业职业技术学院信息工程系,长沙 410007)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2006-11-05 发布日期:2006-11-05

Method for Flatted and Parallel Training Samples Selection Based on Support Vectors

WEN Yimin   

  1. (Department of Information Engineering, Hunan Industry Polytechnic, Changsha 410007)
  • Received:1900-01-01 Revised:1900-01-01 Online:2006-11-05 Published:2006-11-05

摘要: 基于支持向量能够代表训练集分类特征的特点,该文提出了一种基于支持向量的分层并行筛选训练样本的机器学习方法。该方法按照分而治之的思想将原分类问题分解成若干子问题,将训练样本的筛选过程分解成级联的2个层次。每层采用并行方法提取各训练集中的支持向量,这些被提取的支持向量将作为下一层的训练样本,各层训练集中的非支持向量通过学习被逐步筛选掉。为了保证问题的一致性,引入了交叉合并规则,仿真实验结果表明该方法在保证分类器推广能力的情况下,缩短了支持向量机的训练时间,减少了支持向量的数目。

关键词: 分而治之, 训练样本筛选, 支持向量机, 交叉合并规则

Abstract: In order to handle large-scale classification problems, this paper presents a machine learning method for hierarchically and parallel training samples selection, based on the characteristics of support vectors that represent the classification information of training data. In this method, according to the principle of divide and conquer, the original classification problem is divided into several small sub-problems. After that, the training procedure is separated into two cascade phases. In each phase, all of the sub-problems are processed, their support vectors are extracted and so the non-support-vectors are filtered out step by step. In order to keep the consistency, a cross-merger principle is introduced. The simulation results indicate that the method speeds up training while maintaining the generalization accuracy of support vector machines(SVMs), and reduces the number of support vectors (SVs).

Key words: Divide and conquer, Training sample selection, Support vector machines, Cross-merger rules

中图分类号: