BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260920T174240EDT-0294ufaCcF@132.216.98.100
DTSTAMP:20260920T214240Z
DESCRIPTION:\n	ISS Informal Systems Seminar\n\n	Speaker: Mohamad Kazem Shiran
 i Faradonbeh – University of Georgia\, United States \n\n\n \n\n\n	\n\n	Pres
 entation on YouTube.\n\n	Abstract: We focus on learning from a single traje
 ctory to control linear dynamical systems that evolve as stochastic differ
 ential equations. Reinforcement learning policies will be presented for st
 abilizing unknown systems\, and for minimizing quadratic cost functions. F
 irst\, fast and reliable stabilization algorithms that utilize Bayesian le
 arning methods will be discussed. Then\, we propose effective policies tha
 t can balance the exploration and exploitation\, in a manner similar to Ep
 silon-Greedy or Thompson Sampling. Theoretical analyses showing regret bou
 nds that grow with the square-root of time and with the number of paramete
 rs will be provided\, together with experiments for different real systems
 . Further fundamental limitations will be discussed as well.\n\n	\n	Bio: Moh
 amad Kazem Shirani Faradonbeh received his PhD in statistics from the Univ
 ersity of Michigan in 2017\, and his BSc in electrical engineering from Sh
 arif University of Technology in 2012. He was a postdoc with the Informati
 cs Institute and with the Department of Statistics at the University of Fl
 orida\, and a fellow in the Simons Institute for the Theory of Computing a
 t the University of California - Berkeley. From 2020 at the University of 
 Georgia\, he is an assistant professor of Data Science with the Department
  of Statistics and with the Institute for Artificial Intelligence.\n\n
DTSTART:20230627T143000Z
DTEND:20230627T153000Z
LOCATION:CA\, ZOOM
SUMMARY:Continuous-Time Linear-Quadratic Reinforcement Learning
URL:https://www.mcgill.ca/cim/channels/event/continuous-time-linear-quadrat
 ic-reinforcement-learning-351621
END:VEVENT
END:VCALENDAR