CSAIL Event Calendar: Previous Series
An overview of agnostic learning
Speaker: Adam Kalai , Microsoft Research New England
Agnostic Learning (Kearns, Schapire and Sellie '92; Haussler '90) is a computational model of learning in which little or no assumptions are made about the true function being learned. Consequently, agnostic learning algorithms also tolerate arbitrary and, in particular, realistic noise. In this overview, I will place agnostic learning in the context of many traditional concepts in machine learning, such as Valiant's PAC model (1984), Fourier learning, Support Vector Machines, Decision Trees, (Inter)active Learning, and Boosting. No prior learning knowledge will be assumed.