集成学习中的多样性度量  

Diversity measures in ensemble learning

作　　者：孙博[1] 王建东[1] 陈海燕[1] 王寅同[1] 

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京210016

出　　处：《控制与决策》2014年第3期385-395,共11页Control and Decision

基　　金：国家自然科学基金重点项目(61139002);江苏省博士后计划项目(1301013A);中央高校基本科研业务费专项基金项目(NS2012134;NZ2013306)

摘　　要：在集成学习中,基分类器之间的多样性对于解释多分类器系统的工作机理和构造有效的集成系统具有重要的作用,但至今仍没有统一的度量多样性的方法.首先总结介绍常用的多样性度量方法,阐述每种方法评估多样性的角度和方式;然后从对多样性新的解释和度量、多样性度量在选择性集成中的应用、多样性度量和集成学习精度的关系3个方面探讨多样性度量的研究进展;最后给出关于多样性度量进一步的研究方向.Diversity among base classifiers plays an important role in ensemble learning for illustrating the working mechanism of multiple classifier systems and constructing effective ensemble systems. However, at present, there doesn’t exist a widely accepted diversity measure. Firstly, some commonly used diversity measures are summarized, and the perspective adopted by each measure is illustrated when evaluating the diversity. Then, the research progresses of diversity measures are investigated in the following three aspects： The recently proposed interpretations and measures for diversity, the application of diversity measures in selective ensemble, and the relationship between diversity measures and ensemble accuracy. Finally, several directions for future research are given.

关 键 词：集成学习 多样性度量 精度 选择性集成 泛化性能 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]