Journal of Computer Applications ›› 2015, Vol. 35 ›› Issue (12): 3472-3476.DOI: 10.11772/j.issn.1001-9081.2015.12.3472

• Artificial intelligence • Previous Articles     Next Articles

Stopping criterion of active learning for scenario of single-labeling mode

YANG Ju1, LI Qingwen1, YU Hualong1,2   

  1. 1. School of Computer Science and Engineering, Jiangsu University of Science and Technology, Zhenjiang Jiangsu 212003, China;
    2. College of Automation, Southeast University, Nanjing Jiangsu 210095, China
  • Received:2015-06-15 Revised:2015-08-16 Online:2015-12-10 Published:2015-12-10

适用于单轮单样例标注场景的主动学习停止准则

杨菊1, 李青雯1, 于化龙1,2   

  1. 1. 江苏科技大学计算机科学与工程学院, 江苏镇江 212003;
    2. 东南大学自动化学院, 南京 210095
  • 通讯作者: 于化龙(1982-),男,黑龙江哈尔滨人,副教授,博士,主要研究方向:机器学习、数据挖掘
  • 作者简介:杨菊(1990-),女,江苏南通人,硕士研究生,主要研究方向:机器学习、主动学习;李青雯(1992-),女,江苏仪征人,硕士研究生,主要研究方向:机器学习、图像处理。
  • 基金资助:
    国家自然科学基金资助项目(61305058);江苏省自然科学基金资助项目(BK20130471);中国博士后科学基金资助项目(2013M540404);江苏省博士后基金资助项目(1401037B);江苏省普通高校研究生科研创新计划项目。

Abstract: In order to solve the problem that selected accuracy stopping criterion can only be applied in the scenario of batch mode-based active learning, an improved stopping criterion for single-labeling mode was proposed. The matching relationship between each predicted label and the corresponding real label existing in a pre-designed number of learning rounds was used to approximately estimate and calculate the selected accuracy. The higher the match quality was, the higher the selected accuracy was. Then, the variety of selected accuracy could be monitored by moving a sliding-time window. Active learning would stop when the selected accuracy was higher than a pre-designed threshold. The experiments were conducted on 6 baseline data sets with active learning algorithm based on Support Vector Machine (SVM) classifier for indicating the effectiveness and feasibility of the proposed criterion. The experimental results show that when pre-designing an appropriate threshold, active learning can stop at the right time. The proposed method expands the applications of selected accuracy stopping criterion and improves its practicability.

Key words: active learning, selected accuracy stopping criterion, single instance labeling on each round, sliding-time window, Support Vector Machine (SVM)

摘要: 针对现有的选择精度主动学习停止准则仅适用于批量样例标注场景这一问题,提出了一种适用于单轮单样例标注场景的改进的选择精度停止准则。该准则通过监督自本轮起前溯的固定学习轮次内的预测标记与真实标记间的匹配关系,对选择精度进行近似的评估计算,匹配度越高则选择精度越高,继而利用滑动时间窗实时监测该选择精度的变化,若当其高于事先设定的阈值,则停止主动学习算法的运行。以基于支持向量机的主动学习方法为例,通过6个基准数据集对该准则的有效性与可行性进行了验证,结果表明当选取合适的阈值时,该准则能找到主动学习停止的合理时机。该方法扩大了选择精度停止准则的适用范围,提升了其实用性。

关键词: 主动学习, 选择精度停止准则, 单轮单样例标注, 滑动时间窗, 支持向量机

CLC Number: