BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260514T141025EDT-8899KanU6I@132.216.98.100
DTSTAMP:20260514T181025Z
DESCRIPTION:Virtual Informal Systems Seminar (VISS) Centre for Intelligent 
 Machines (CIM) and Groupe d'Etudes et de Recherche en Analyse des Decision
 s (GERAD)\n\n\n	Zoom Link\n		Meeting ID: 910 7928 6959        \n		Passcode: VIS
 S\n		\n		Speaker:  Kai Cui\, PhD candidate\, Technical University of Darmstadt
 \n		\n		Abstract: The recent mean field game (MFG) formalism facilitates other
 wise intractable computation of approximate Nash equilibria in many-agent 
 settings. In this paper\, we consider discrete-time finite MFGs subject to
  finite-horizon objectives. We show that all discrete-time finite MFGs wit
 h non-constant fixed point operators fail to be contractive as typically a
 ssumed in existing MFG literature\, barring convergence via fixed point it
 eration. Instead\, we incorporate entropy-regularization and Boltzmann pol
 icies into the fixed point iteration. As a result\, we obtain provable con
 vergence to approximate fixed points where existing methods fail\, and rea
 ch the original goal of approximate Nash equilibria. All proposed methods 
 are evaluated with respect to their exploitability\, on both instructive e
 xamples with tractable exact solutions and high-dimensional problems where
  exact methods become intractable. In high-dimensional scenarios\, we appl
 y established deep reinforcement learning methods and empirically combine 
 fictitious play with our approximations.\n		\n		Bio: Kai Cui is a PhD candidat
 e at the Bioinspired Communication Systems Lab under supervision of Profes
 sor Heinz Koeppl at Technical University of Darmstadt (Germany). Prior to 
 joining BCS\, he also received BSc and MSc degrees in Computer Science as 
 well as Electrical Engineering and Information Technology at Technical Uni
 versity of Darmstadt. His current research interests are multi-agent syste
 ms\, reinforcement learning\, mean field games and UAV swarms.\n\n
DTSTART:20210326T140000Z
DTEND:20210326T150000Z
LOCATION:CA\, ZOOM
SUMMARY:Approximately Solving Mean Field Games via Entropy-Regularized Deep
  Reinforcement Learning
URL:https://www.mcgill.ca/cim/channels/event/approximately-solving-mean-fie
 ld-games-entropy-regularized-deep-reinforcement-learning-329072
END:VEVENT
END:VCALENDAR