Deep Boosting
发布人: 系统管理员   发布时间: 2015-11-17   浏览次数: 177

1122Deep Boosting


讲座题目:Deep Boosting

主讲人:Mehryar Mohri 教授

主持人:羊丹平 教授



主办单位:数学系 科技处

报告人简介:Mehryar Mohri is a professor of computer science and mathematics at the Courant Institute of Mathematical Sciences at New York University and a Research Consultant at Google since 2004.

Prior to joining NYU, Prof. Mohri served as a Department Head and a Technology Leader at AT&T Labs-Research, earlier AT&T Bell Labs, where he worked and supervised research in several areas including text and speech processing, machine learning, and algorithms. He taught for about a year at both Ecole Polytechnique and the University of Paris 7 and have held visiting professorship positions at several institutions, including Google Research, Ecole Normale Superieure (ENS Ulm), and Institut des Hautes Etudes Scientifiques (IHES).

His research interests cover a number of different areas: machine learning, algorithms and theory, automata theory, speech processing, natural language processing, computational biology, and the design of general-purpose software libraries. He has extensively published conference and journal papers in all of these areas.

His research in learning theory and algorithms has been used in a variety of applications. His work on automata theory and algorithms has served as the foundation for several applications in language processing, with several of my algorithms used in virtually all spoken-dialog and speech recognitions systems used in the United States. He has also co-authored several software libraries widely used in research and academic labs.

He has contributed to the organization of numerous conferences and workshops, including as co-chair for COLT 2010. He is on the Editorial or advisory boards of several journals including the Journal of Machine Learning Research, Machine Learning journal, and the Journal of Automata, Languages and Combinatorics.

He is co-author of the machine learning textbook Foundations of Machine Learning used in graduate courses on machine learning in several universities and corporate research laboratories.

He received his B.S from École Polytechnique in 1987, his M.S. in computer science and applied mathematics from École Normale Supérieure in 1989 and his Ph.D. in 1993 from the University of Paris 7.


This talk discusses a new ensemble learning algorithm, DeepBoost, which can use as a base classifier set deep decision trees, or other rich families. Extensive experiments show that DeepBoost consistently outperforms AdaBoost, Logistic Regression, and their L1-regularized variants. The key to the success of the algorithm is a capacity-conscious selection criterion for the hypotheses forming the ensemble, which is grounded in a new theoretical foundation with several significant implications. The theory developed is quite general and leads a new model selection framework, Voted Risk Minimization, which can guide the design of a variety of other learning algorithms such as Structural Maxent or Deep Cascade models.