{"630874":{"#nid":"630874","#data":{"type":"event","title":"ML@GT Seminar: Daniel Russo, Columbia University","body":[{"value":"\u003Cp\u003EML@GT invites you to a seminar by Daniel Russo, an assistant professor at Columbia Business School.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Ca href=\u0022https:\/\/docs.google.com\/forms\/d\/e\/1FAIpQLScO_LcGzGuE8reXCZ8v03hknbkweLf20TQ-H5VggfFK0LKB4w\/viewform\u0022\u003ERSVP here\u003C\/a\u003E\u003C\/p\u003E\r\n\r\n\u003Ch3\u003ETalk Title\u003C\/h3\u003E\r\n\r\n\u003Cp\u003EGlobal Optimality Guarantees for Policy Gradient Methods\u003C\/p\u003E\r\n\r\n\u003Ch3\u003EAbstract\u003C\/h3\u003E\r\n\r\n\u003Cp\u003EPolicy gradients methods are perhaps the most widely used class of reinforcement learning algorithms.\u0026nbsp; These methods apply to complex, poorly understood, control problems by performing stochastic gradient descent over a parameterized class of polices. Unfortunately, due to the multi-period nature of the objective, policy gradient algorithms face non-convex optimization problems and can get stuck in suboptimal local minima even for extremely simple problems. This talk with discus structural properties \u0026ndash; shared by several canonical control problems \u0026ndash; that guarantee the policy gradient objective function has no suboptimal stationary points despite being non-convex. Time permitting, I\u0026rsquo;ll also discuss (1) convergence rates that follow as a consequence of this theory and (2) consequences of this theory for policy gradient performed with highly expressive policy classes.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E* This talk is based on ongoing joint work with Jalaj Bhandari.\u003C\/p\u003E\r\n\r\n\u003Ch3\u003EBio\u003C\/h3\u003E\r\n\r\n\u003Cp\u003ERusso joined the\u0026nbsp;\u003Ca href=\u0022https:\/\/www8.gsb.columbia.edu\/faculty-research\/divisions\/decision-risk-operations\u0022\u003EDecision, Risk, and Operations division\u003C\/a\u003E\u0026nbsp;of the Columbia Business School as an assistant professor in Summer 2017. Prior to joining Columbia, he\u0026nbsp;spent one great year as an assistant professor in the MEDS department at Northwestern\u0026#39;s Kellogg School of Management and one year at Microsoft Research in New England as Postdoctoral Researcher. Russo recieved his\u0026nbsp;Ph.D. from Stanford University in 2015, where he\u0026nbsp;was advised by\u0026nbsp;\u003Ca href=\u0022http:\/\/engineering.stanford.edu\/profile\/bvr\u0022\u003EBenjamin Van Roy\u003C\/a\u003E. In 2011 Russo recieved his\u0026nbsp;\u0026nbsp;BS in Mathematics and Economics from the University of Michigan.\u003C\/p\u003E\r\n\r\n\u003Cp\u003ERusso\u0026#39;s research lies at the intersection of statistical machine learning and sequential decision-making, and contributes to the fields of online optimization, reinforcement learning, and sequential design of experiments. He is\u0026nbsp;interested in the design and analysis of algorithms that learn over time to make increasingly effective decisions through interacting with a poorly understood environment.\u003C\/p\u003E\r\n","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":"","field_summary_sentence":[{"value":"ML@GT invites you to a seminar by Daniel Russo, an assistant professor at Columbia University\u0027s business school. "}],"uid":"34773","created_gmt":"2020-01-10 15:21:49","changed_gmt":"2020-02-27 15:47:38","author":"ablinder6","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2020-03-11T13:15:00-04:00","event_time_end":"2020-03-11T14:15:00-04:00","event_time_end_last":"2020-03-11T14:15:00-04:00","gmt_time_start":"2020-03-11 17:15:00","gmt_time_end":"2020-03-11 18:15:00","gmt_time_end_last":"2020-03-11 18:15:00","rrule":null,"timezone":"America\/New_York"},"extras":[],"groups":[{"id":"47223","name":"College of Computing"},{"id":"37041","name":"Computational Science and Engineering"},{"id":"1299","name":"GVU Center"},{"id":"589608","name":"Machine Learning"},{"id":"576481","name":"ML@GT"},{"id":"431631","name":"OMS"},{"id":"50877","name":"School of Computational Science and Engineering"},{"id":"50875","name":"School of Computer Science"},{"id":"50876","name":"School of Interactive Computing"}],"categories":[],"keywords":[],"core_research_areas":[],"news_room_topics":[],"event_categories":[{"id":"1795","name":"Seminar\/Lecture\/Colloquium"}],"invited_audience":[{"id":"78761","name":"Faculty\/Staff"},{"id":"177814","name":"Postdoc"},{"id":"78771","name":"Public"},{"id":"174045","name":"Graduate students"},{"id":"78751","name":"Undergraduate students"}],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003EKyla Hanson\u003C\/p\u003E\r\n\r\n\u003Cp\u003Ekhanson@cc.gatech.edu\u003C\/p\u003E\r\n","format":"limited_html"}],"email":[],"slides":[],"orientation":[],"userdata":""}}}