BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Sham Kakade (University of Washington\, Seattle) DTSTART:20200616T190000Z DTEND:20200616T200000Z DTSTAMP:20260423T021020Z UID:OptMLStat/4 DESCRIPTION:Title: Representation\, Modeling\, and Gradient Based Optimization in Reinforce ment Learning\nby Sham Kakade (University of Washington\, Seattle) as part of Online Seminar of Mathematical Foundations of Data Science\n\n\nAb stract\nReinforcement learning is now the dominant paradigm for how an age nt learns to interact with the world. The approach has lead to successes r anging across numerous domains\, including game playing and robotics\, and it holds much promise in new domains\, from self driving cars to interact ive medical applications. Some of the central challenges are:\n\n- Represe ntational learning: does having a good representation of the environment p ermit efficient reinforcement learning?\n\n- Modeling: should we explicitl y build a model of our environment or\, alternatively\, should we directly learn how to act?\n\n- Optimization: in practice\, deployed algorithms of ten use local search heuristics. Can we provably understand when these ap proaches are effective and provide faster and more robust alternatives?\n\ nThis talk will survey a number of results on these basic questions. Throu ghout\, we will highlight the interplay of theory\, algorithm design\, an d practice.\n LOCATION:https://researchseminars.org/talk/OptMLStat/4/ END:VEVENT END:VCALENDAR