自动化学报2017，Vol.43Issue(1)：20-39,20.DOI:10.16383/j.aas.2017.c160149
Rademacher复杂度在统计学习理论中的研究：综述
Researches on Rademacher Complexities in Statistical Learning Theory：A Survey
摘要
Abstract
Measuring the complexity of a hypothesis space plays a crucial role in statistical learning theory. Unlike those data-independent complexities, Rademacher complexity, which is data-dependent, can attain a much more compact generalization representation. In recent years, Rademacher complexity has attracted more attention and found broad applications in the development of statistical learning theory. Because of its importance, in this paper we review several complexity measures of function classes and their relations with Rademacher complexities. Next, we describe the techniques of Rademacher complexities in generalization analysis. Then, we present the recent researches of Rademacher complexity learning bounds for independent and identical distribution (i.i.d.) and non-independent and identical distribution (non-i.i.d.). Finally, we discuss the potential issues and possible directions of Rademacher complexities in statistical learning theory.关键词
机器学习/统计学习理论/泛化界/Rademacher复杂度Key words
Machine learning/statistical learning theory/generalization bounds/Rademacher complexities引用本文复制引用
吴新星,张军平..Rademacher复杂度在统计学习理论中的研究：综述[J].自动化学报,2017,43(1):20-39,20.基金项目
国家自然科学基金(61673118,61273299),上海浦江人才计划(16PJ D009),上海市人才发展资金(201629)资助Supported by National Natural Science Foundation of China (61673118,61273299), Shanghai Pujiang Program (16PJD009), and Shanghai Talents Development Funds (201629) （61673118,61273299）
