BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Turkey BEGIN:DAYLIGHT DTSTART:20380119T061407 TZOFFSETFROM:+0300 TZOFFSETTO:+0300 RRULE:FREQ=YEARLY;BYDAY=3TU;BYMONTH=1 TZNAME:+03 END:DAYLIGHT BEGIN:STANDARD DTSTART:20160907T000000 TZOFFSETFROM:+0300 TZOFFSETTO:+0300 RRULE:FREQ=YEARLY;BYDAY=1WE;BYMONTH=9 TZNAME:+03 END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20200106T232212Z UID:88DBF5A5-62FD-44C1-80E4-DEACD18FBBFB DTSTART;TZID=Turkey:20191227T124000 DTEND;TZID=Turkey:20191227T144000 DESCRIPTION:27 December 2019 (12:40): IEEE AP/MTT/EMC/ED Turkey Seminar Ser ies (S.64)\n\nSpeaker: Asst. Prof. Cem Tekin\, Bilkent University\n\nTopic : "Exploration and Exploitation in Complex Domains: Learning with Multiple Objectives and Contexts"\n\nLocation: Middle East Technical University\, Ankara\, Turkey\n\nAbstract: Delivering personalized medicine for treatmen t of complex diseases\, discovering and recommending interesting articles for a particular user from a huge set of documents and optimising hyper-pa rameters of deep learning architectures given a particular dataset all req uire context-driven learning of optimal decisions over time. Efficient onl ine learning in these domains is a formidable task at least for two reason s. The first is that the contexts often arrive from a variety of sources a nd have diverse features\, so that integrating what is learned is subject to the curse of dimensionality. The second is that existence of multiple a nd possibly conflicting objectives often makes it impossible to simultaneo usly optimize each objective. In this talk\, I will present recent pieces of multi-objective and contextual bandit models addressing these challenge s. In particular\, I will explain how to tradeoff conflicting objectives u sing notions of lexicographic and Pareto optimality in an online learning setting and context-dependent multi-dimensional rewards. I will discuss ho w the well-known notion of regret can be extended to capture a rich set of multi-dimensional performance metrics\, and present learning algorithms w ith sublinear regret guarantees.\n\nBio: Cem Tekin is an Assistant Profess or in the Department of Electrical and Electronics Engineering\, Bilkent U niversity\, Ankara\, Turkey. He received PhD in Electrical Engineering: Sy stems\, MS in Mathematics and MSE in Electrical Engineering: Systems\, fro m the University of Michigan in 2013\, 2011 and 2010\, respectively. He re ceived BS degree in Electrical and Electronics Engineering from Middle Eas t Technical University\, Ankara\, Turkey in 2008. From 2013 to 2015\, he w as a postdoctoral scholar in Electrical Engineering Department\, UCLA. Cem has authored or coauthored over 50 research papers\, 5 book chapters and a research monograph. He received the Fred W. Ellersick Award for the best paper in MILCOM 2009 and the Science Academy Association of Turkey Distin guished Young Scientist (BAGEP) Award and Parlar Foundation Young Investig ator Award in 2019. His research interests include bandit problems\, machi ne learning for personalized medicine\, multi-agent systems\, stream minin g\, influence maximization and cognitive communications.\n\nSpeaker(s): As st. Prof. Cem Tekin\, \n\nAnkara\, Ankara\, Türkiye LOCATION:Ankara\, Ankara\, Türkiye ORGANIZER:ozergul@metu.edu.tr SEQUENCE:1 SUMMARY:IEEE AP/MTT/EMC/ED TURKEY CHAPTER SEMINAR SERIES -- SEMINAR 64 URL;VALUE=URI:https://events.vtools.ieee.org/m/217593 X-ALT-DESC:Description: <br /><p><strong>27 December 2019 (12:40): &nbsp\;I EEE AP/MTT/EMC/ED Turkey Seminar Series (S.64)</strong></p>\n<p>Speaker: A sst. Prof. Cem Tekin\, Bilkent University</p>\n<p>Topic: "Exploration and Exploitation in Complex Domains: Learning with Multiple Objectives and Con texts"</p>\n<p>Location:&nbsp\;Middle East Technical University\, Ankara\, Turkey</p>\n<p>Abstract: Delivering personalized medicine for treatment o f complex diseases\, discovering and recommending interesting articles for a particular user from a huge set of documents and optimising hyper-param eters of deep learning architectures given a particular dataset all requir e context-driven learning of optimal decisions over time. Efficient online learning in these domains is a formidable task at least for two reasons. The first is that the contexts often arrive from a variety of sources and have diverse features\, so that integrating what is learned is subject to the curse of dimensionality. The second is that existence of multiple and possibly conflicting objectives often makes it impossible to simultaneousl y optimize each objective. In this talk\, I will present recent pieces of multi-objective and contextual bandit models addressing these challenges. In particular\, I will explain how to tradeoff conflicting objectives usin g notions of lexicographic and Pareto optimality in an online learning set ting and context-dependent multi-dimensional rewards. I will discuss how t he well-known notion of regret can be extended to capture a rich set of mu lti-dimensional performance metrics\, and present learning algorithms with sublinear regret guarantees.</p>\n<p>Bio: Cem Tekin is an Assistant Profe ssor in the Department of Electrical and Electronics Engineering\, Bilkent University\, Ankara\, Turkey. He received PhD in Electrical Engineering: Systems\, MS in Mathematics and MSE in Electrical Engineering: Systems\, f rom the University of Michigan in 2013\, 2011 and 2010\, respectively. He received BS degree in Electrical and Electronics Engineering from Middle E ast Technical University\, Ankara\, Turkey in 2008. From 2013 to 2015\, he was a postdoctoral scholar in Electrical Engineering Department\, UCLA. C em has authored or coauthored over 50 research papers\, 5 book chapters an d a research monograph. He received the Fred W. Ellersick Award for the be st paper in MILCOM 2009 and the Science Academy Association of Turkey Dist inguished Young Scientist (BAGEP) Award and Parlar Foundation Young Invest igator Award in 2019. His research interests include bandit problems\, mac hine learning for personalized medicine\, multi-agent systems\, stream min ing\, influence maximization and cognitive communications.</p> END:VEVENT END:VCALENDAR