This seminar is part of the Smart Cities Event Series, which focuses on the future of urban systems and the data science and AI-driven technologies shaping them.
The DSI Smart Cities Center supports research that addresses the challenges of aging infrastructure while advancing innovations in smart grids, intelligent transportation systems, and networked sensing technologies that enable real-time monitoring and decision-making.
Registration Request Form: Registration will be prioritized for Columbia faculty and affiliated scholars. If you are a postdoctoral researcher, student, or external guest, you may be waitlisted to attend until closer to the event date. All who submit on the below form will receive a confirmation email and a calendar hold if your registration is approved. Thank you again for your interest in attending!
_______
Date: Tuesday, March 26, 2026 (12:00 PM – 1:00 PM ET)
Location: Northwest Corner Building, DSI Suite, Armen Avanessians Conference Room, 14th Floor
Address: 550 West 120th Street, New York, NY 10027
Speaker: Minyi Huang, Professor, School of Mathematics and Statistics, Carleton University
Title: Mean Field Games Over Large Networks
Abstract:
As an active research area for about two decades, mean field game (MFG) theory provides a powerful tool to tackle large-population non-cooperative dynamic games with individual-mass interactions. This talk describes how to extend MFG theory to models with subpopulations (i.e., agent clusters) distributed over large networks: (i) dense networks, (ii) moderately sparse networks, and (iii) very sparse networks.
The analysis of the dense case (i) has been developed based on graphon theory (Lovasz, 2012), leading to the so-called graphon mean field games (GMFGs). For the second case (ii), by suitable scaling along the graph sequence, we define the limit edge distribution called graphexon, and further introduce a probability kernel to characterize network connections seen at a given node. Then the GMFG theory can be extended to this case. For the very sparse case (iii), we consider a lattice structure to model interactions of neighboring clusters of agents. We apply scaling of diffusion type to obtain a meaningful limit, leading to the so-called Laplexion dynamics, as a hydrodynamic limit of the lattice model. For each case, the limit model will be used to construct decentralized strategies for the actual finite population, which are shown be an asymptotic Nash equilibrium.
(Based on recent work with P. E. Caines, T. Chen, and S. Gao)