Multi-Agent Learning under Uncertainty: Equilibria, Robustness & Stability
Event details
| Date | 16.03.2026 |
| Hour | 11:15 › 12:00 |
| Speaker | Kyriakos Lotidis Ph.D. candidate in Operations Research Stanford University, USA. |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
Abstract:
Modern AI systems, from autonomous vehicles to digital marketplaces, operate in decentralized environments characterized by strategic interaction, noisy feedback, and imperfect information. Motivated by these challenges, we study the interplay between robustness, learning, and uncertainty in continuous games. We first investigate which equilibria persist under perturbations and introduce a structural notion of equilibrium robustness based on the geometry of the underlying game. We then analyze the long-run behavior of adaptive agents learning in such environments, identifying conditions under which robust equilibria emerge as stable outcomes of the learning dynamics. When classical convergence fails, we further examine the stochastic behavior of these dynamics, studying when they admit a stationary distribution and how it concentrates around equilibrium. Together, these results contribute to our understanding of how equilibrium stability and learning dynamics interact under uncertainty, highlighting both the strengths and limitations of regularization-based learning in multi-agent systems.
Biography:
Kyriakos Lotidis is a final-year Ph.D. candidate in Operations Research at Stanford University. His research lies at the intersection of game theory, online learning, and stochastic methods, with a particular focus on the stability and convergence of learning algorithms in multi-agent systems. He also holds an M.Sc. in Statistics from Stanford University and a B.Sc. in Electrical and Computer Engineering from the National Technical University of Athens.
Modern AI systems, from autonomous vehicles to digital marketplaces, operate in decentralized environments characterized by strategic interaction, noisy feedback, and imperfect information. Motivated by these challenges, we study the interplay between robustness, learning, and uncertainty in continuous games. We first investigate which equilibria persist under perturbations and introduce a structural notion of equilibrium robustness based on the geometry of the underlying game. We then analyze the long-run behavior of adaptive agents learning in such environments, identifying conditions under which robust equilibria emerge as stable outcomes of the learning dynamics. When classical convergence fails, we further examine the stochastic behavior of these dynamics, studying when they admit a stationary distribution and how it concentrates around equilibrium. Together, these results contribute to our understanding of how equilibrium stability and learning dynamics interact under uncertainty, highlighting both the strengths and limitations of regularization-based learning in multi-agent systems.
Biography:
Kyriakos Lotidis is a final-year Ph.D. candidate in Operations Research at Stanford University. His research lies at the intersection of game theory, online learning, and stochastic methods, with a particular focus on the stability and convergence of learning algorithms in multi-agent systems. He also holds an M.Sc. in Statistics from Stanford University and a B.Sc. in Electrical and Computer Engineering from the National Technical University of Athens.
Practical information
- General public
- Free
Organizer
- Prof. Maryam Kamgarpour