Prof. Georgios Tsaousoglou, Technical University of Denmark

Title: Shaping Incentives to Tame Strategic Learning in Multi-Agent Systems

Abstract: Many socio-technical systems are governed by autonomous agents, each coming with its own objective which is not necessarily aligned with system-level goals. In such settings, we do not get to directly design the agents, and coordination must instead be achieved by shaping the incentives faced by the agents. This motivates a mechanism-design perspective on control of learning-enabled multi-agent systems. We study coordination mechanisms for selfish learning agents interacting in partially observable Markov games, where a coordinator designs protocols for information exchange, resource allocation, and reward or credit assignment. Agents strategically adapt their behavior through learning, aiming to optimize individual performance. To align individual incentives with a social objective, we propose a class of critical-value-based mechanisms that induce coordinated behavior without requiring direct control over agents’ policies. Analytic results are derived for a stylized setting, providing insight into equilibrium structure and incentive alignment. Simulation results for use cases with (deep) reinforcement learning agents controlling resources in power systems demonstrate that incentive design can teach coordinated, jointly efficient behavior to distributed learners.

Brief bio: Georgios Tsaousoglou received his PhD from the National Technical University of Athens in 2019. He then joined the Eindhoven University of Technology on a Marie-Curie Fellowship for 2 years. Since 2022, he is with the Department of Applied Mathematics and Computer Science, Technical University of Denmark (DTU), where he now serves as an Associate Professor. He teaches Decision-making Under Uncertainty and Algorithmic Game Theory. His research interests include sequential decision-making, multi-agent systems, and AI surveillance, with applications to systems operating over critical infrastructure. In 2024, he received the Outstanding Reviewer Award from IEEE Transactions on Smart Grid and he currently serves as an Associate Editor for the same journal. In 2026, he received the Villum Young Investigator Award, to form a group for his project MELISANDRE: “Market equilibria likelihood and safety assessment in AI-driven energy systems”.