Mean field games on sparse graphs using graphons, Lp graphons and graphexes
±õ²Ô´Ú´Ç°ù³¾²¹±ôÌý³§²â²õ³Ù±ð³¾²õÌý³§±ð³¾¾±²Ô²¹°ùÌý(±õ³§³§)
Centre for Intelligent Machines (CIM) and Groupe d'Etudes et de Recherche
±ð²ÔÌý´¡²Ô²¹±ô²â²õ±ðÌý»å±ð²õÌý¶Ù±ð³¦¾±²õ¾±´Ç²Ô²õÌý(³Ò·¡¸é´¡¶Ù)
Speaker: Kai Cui
²Ñ±ð±ð³Ù¾±²Ô²µÌý±õ¶Ù:Ìý845Ìý1388Ìý1004ÌýÌýÌýÌýÌýÌýÌý
±Ê²¹²õ²õ³¦´Ç»å±ð:Ìý³Õ±õ³§³§
** Note that this is a hybrid event.
** This seminar will be projected at McConnell 437 at Ï㽶ÊÓƵ.
Abstract: In this talk, we consider mean field games on graphs in discrete time. Here, each node corresponds to a single agent, and agent interaction is through their neighborhoods. We begin by reiterating graphon mean field games on dense graphs. There, we show a computational reduction thereof to standard mean field games, as well as a propagation of chaos to motivate the limiting system. We also extend results to sparser graphs via Lp graphons, and more recently via graphex mean field games for significantly more realistic, sparse graphs. Finally, we briefly give an outlook on our recent reinforcement learning algorithms for mean field control, which could be an avenue for future graphical extensions.
Bio: I am a fifth year PhD candidate at the Self-Organizing Systems Lab under supervision of Professor Heinz Koeppl at Technische Universität Darmstadt. My research focuses on multi-agent reinforcement learning, mean-field games and applications thereof. Prior to the PhD studies, I received my MSc degrees in Computer Science as well as Electrical Engineering and Information Technology at Technische Universität Darmstadt.
