Distributed methods for generalized Nash equi librium learning

Generalised Nash equilibrium (GNE) problems model multi-agent systems with shared coupling constraints, arising in applications such as energy networks, economic markets, and distributed control. However, computing GNEs in a distributed manner remains challenging due to global coupling constraints and limited information exchange among agents. In this talk, we present distributed methods for learning variational GNEs under two information settings: (i) communication-based approaches, where agents exchange information over a network, and (ii) payoff-based approaches, where agents rely only on local cost observations. The proposed algorithms combine consensus, projection, and primal–dual techniques to handle coupling linear constraints without centralised coordination. We provide convergence guarantees under monotonicity assumptions and discuss extensions to general constraints, highlighting the trade-offs between information availability and convergence performance.

Join at imt.lu/aula1