学术报告:Margin Theory for Machine Learning

演讲人:王立威 北京大学信息科学技术学院教授

时 间:2013年1月10日上午10:00

地 点:北京大学计算机科学技术研究所大楼(北京大学东南门对面)一楼报告厅

报告题目:Margin Theory for Machine Learning

摘要:Margin is a central concept in machine learning. It is a measure of confidence of the learning result. The margin theory upper bounds the performance of a learning algorithm in terms of margins, which greatly extends the classical VC and PAC theory for machine learning. In this talk, I am going to review the margin theories for boosting and SVM respectively. I will talk about the debate on the margin-theoretic explanation of boosting as well as how different notions of margins are related to the generalization performance. In particular, I show that our concept of Equilibrium margin induces sharp bounds for boosting. Then I will turn to the margin theory for SVM. Previously it is widely believed that margin bounds are dimensionality independent. However, I show a dimensionality dependent margin bound which is strictly sharper than previously well known dimensionality independent margin bound.

王立威 北京大学信息科学技术学院教授。于清华大学电子工程系获本科和硕士学位,北京大学数学学院获博士学位。自2005年起在北京大学信息学院任教。主要研究兴趣为机器学习理论。在机器学习顶级会议NIPS, COLT, ICML和顶级期刊JMLR, IEEE Trans. PAMI发表论文多篇。其中2008年发表于机器学习理论最高会议COLT的论文On the Margin Explanation of Boosting Algorithms是中国大陆学者在该会议上的首篇论文。2010年入选AI's 10 to Watch,是首位获得该奖项的亚洲学者。2012年获得国家自然科学基金优秀青年基金;入选新世纪优秀人才。目前任中国计算机学会模式识别与人工智能专委会委员。担任Journal of Computer Science and Technology (JCST)等期刊编委。

CLOSE

没有了 下一篇