BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Computation and Learning of Equilibria in Markov Games
DTSTART:20250820T130000
DTEND:20250820T150000
DTSTAMP:20260501T002404Z
UID:681d482006fbdb68c2b8d109bc676c5a9b83e76538a0aee3e4c62e52
CATEGORIES:Conferences - Seminars
DESCRIPTION:Philip Jordan\nEDIC candidacy exam\nExam president: Prof. Mika
  Göös\nThesis advisor: Prof. Maryam Kamgarpour\nCo-examiner: Prof. Nicol
 as Flammarion\n\nAbstract\nMarkov games [1] provide a framework for modell
 ing multi-agent RL problems. While computing Nash equilibria is hard even 
 in the less general case of normal-form games [2]\, for certain subclasses
  of Markov games (e.g. zero-sum\, and potential games)\, a variety of lear
 ning algorithms have been proposed [3].\nA problem I have been studying in
  my first year is about (Markov) games with coupling constraints. Such gam
 es may e.g. model multi-agent RL with joint safety constraints. In particu
 lar\, I am interested in existence and computation/learning of Nash equili
 bria in such constrained games. These topics have received significantly l
 ess attention than their unconstrained counterparts\, and known results fo
 r both existence and computation come with strong assumptions -- such as c
 onvexity of the joint feasible set -- which we aim to relax.\n\nSelected p
 apers\n[1] Shapley\, Lloyd S. "Stochastic games." Proceedings of the natio
 nal academy of sciences 39.10 (1953): 1095-1100. https://www.jstor.org/st
 able/88799\n[2] Daskalakis\, Constantinos\, Paul W. Goldberg\, and Christo
 s H. Papadimitriou. "The complexity of computing a Nash equilibrium." Comm
 unications of the ACM 52.2 (2009): 89-97. https://dl.acm.org/doi/abs/10.1
 145/1461928.1461951 (for extended version\, see https://people.csail.mit
 .edu/costis/journal_ver10.pdf)\n[3] Ding\, Dongsheng\, et al. "Independent
  policy gradient for large-scale markov potential games: Sharper rates\, f
 unction approximation\, and game-agnostic convergence." International Conf
 erence on Machine Learning. PMLR\, 2022. https://proceedings.mlr.press/v1
 62/ding22b/ding22b.pdf\n 
LOCATION:ME C2 405 https://plan.epfl.ch/?room==ME%20C2%20405
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR