Mathematical Results and Challenges in Learning Theory

posted Jul 9, 2010, 10:33 AM by Glenna Buford

Ingrid Daubechies
Princeton University

AWM Emmy Noether Lecture
January 2006
San Antonio, Texas

Abstract. One of the most widespread applications of learning theory is in ubiquitous search engines, which have to (and do!) classify enormous databases according to (almost) arbitrary criteria. Computer scientists have developed powerful algorithms for these very high-dimensional problems, which typically cannot be tackled by gradient-descent or similar optimization methods. These algorithms and the problems they attack provide very interesting mathematical challenges. The talk will discuss in particular the widely applied AdaBoost algorithm and its properties, as well as some variants. It will review joint work with Cynthia Rudin and Rob Schapire (co-inventor, with Freund, of AdaBoost, for which they were awarded the 2003 Gödel prize.)