Learning theory pursues a principled mathematical understanding of how computer algorithms can learn from examples. Over the last 20 years, learning theory has inspired the design of many influential and useful machine learning techniques. Two important trends in modern learning theory are statistical supervised learning and game-theoretic online prediction. In this tutorial, I will introduce these settings and give a taste of the typical algorithms and analysis techniques in each setting. I will discuss empirical risk minimization, sample complexity bounds, online convex optimization and regret analysis. The tutorial is intended for a general computer science audience, with basic knowledge in elementary probability theory and linear algebra.
Ofer Dekel is a researcher at Microsoft Research in Redmond. He received his PhD in Computer Science from the Hebrew University of Jerusalem. His research interests include machine learning, statistical learning theory, online prediction, stochastic optimization, and Internet search. He is currently involved in the effort to improve the relevance of the Bing search engine using machine learning techniques.
Pizza will be served in the theory lab (CSE 306) right before the talk.