Inverse Learning in Mixed-Individual Mean Field Games
Abstract
Understanding how individuals and populations respond to incentives is essential for designing effective policies. In continuous-time stochastic models of large interacting populations, mean field games (MFGs) describe non-cooperative agents, whereas mean field control (MFC) provides a fully cooperative benchmark, interpreted here as fully altruistic behavior. Mixed-individual MFGs interpolate between these two extremes through a parameter that captures the degree of altruism. For regulators and policymakers, however, intrinsic altruism and other structural parameters, such as individual labor costs, are typically unobservable. To address this challenge, we develop an inverse learning framework for mixed-individual MFGs that recovers latent altruism and labor cost parameters from noisy observations. Numerical experiments demonstrate the feasibility and accuracy of the proposed method. Our results highlight the potential of inverse MFG methods to uncover latent preference structures in large populations, with implications for incentive design, empirical behavioral modeling, and data-driven policy analysis. (Based on joint work with Haoyang Cao, Gökçe Dayanıklı and Leyan Su).
Register to attend ↗The talk takes place on Zoom and is open to the public. Registration is usually valid for multiple future talks.
About the speaker
I am an Assistant Professor in the Department of Statistical Sciences at the University of Toronto. My research focuses on stochastic optimization and stochastic differential equations, with applications to mathematical finance, market equilibrium, and machine learning. My current interests include portfolio optimization, asset pricing under transaction costs and liquidity frictions, risk-constrained decision-making, deep and generative learning for financial markets, and mean field games. I have also worked on crowdsourcing, dimensionality reduction, compressed sensing, and sublinear-time algorithms.