ML/Google Distinguished Lecture Series-Machine Learning Department - Carnegie Mellon University

Machine Learning Special Seminars

Fall 2016 Seminar
Date: November 15, 2016
Time: 4:30 PM - 5:30 PM
Location: 4405 Gates and Hillman Centers
Speaker: Csaba Szepesvari Professor, University of Alberta, Computing Science Department
Title: Stochastic linear bandits
Abstract: Learning and decision making often conflict: A decision maker who is uncertain about its environment may need to choose actions whose main benefit is to gain information rather to gain reward. The basic dilemma between exploration and exploitation is at the heart of bandit problems. In this talk I will focus on the so-called stochastic linear bandits where the payoff function is assumed to posses a linear structure, an assumption that proved to be extremely effective elsewhere in machine learning to facilitate generalization. Here we look into how the linear structure can be exploited in bandits. I will discuss questions like what are the limits of performance of bandit algorithms acting under the linearity assumption, how to design efficient and effective algorithms, or how to exploit additional information like sparsity.