BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Dylan Foster (MIT) DTSTART:20200925T150500Z DTEND:20200925T160500Z DTSTAMP:20260423T024542Z UID:sss/6 DESCRIPTION:Title: Sepa rating Estimation from Decision Making in Contextual Bandits\nby Dylan Foster (MIT) as part of Stochastics and Statistics Seminar Series\n\n\nAb stract\nThe contextual bandit is a sequential decision making problem in w hich a learner repeatedly selects an action (e.g.\, a news article to disp lay) in response to a context (e.g.\, a user’s profile) and receives a r eward\, but only for the action they selected. Beyond the classic explore- exploit tradeoff\, a fundamental challenge in contextual bandits is to dev elop algorithms that can leverage flexible function approximation to model similarity between contexts\, yet have computational requirements compara ble to classical supervised learning tasks such as classification and regr ession. To this end\, we provide the first universal and optimal reduction from contextual bandits to online regression. We show how to transform an y oracle for online regression with a given value function class into an a lgorithm for contextual bandits with the induced policy class\, with no ov erhead in runtime or memory requirements. Conceptually\, our results show that it is possible to provably separate estimation and decision making in to separate algorithmic building blocks\, and that this can be effective b oth in theory and in practice. Time permitting\, I will discuss extensions of these techniques to more challenging reinforcement learning problems.\ n LOCATION:https://researchseminars.org/talk/sss/6/ END:VEVENT END:VCALENDAR