Model-Free Learning in Karma Economy Dynamic Population Games

A recent paper on arXiv (2605.11042) explores model-free equilibrium learning within Karma economies, which are equitable non-monetary resource distribution systems represented as Dynamic Population Games (DPGs). Current computational methods for DPGs require complete understanding of the game model and function centrally, which poses challenges in practical scenarios where agents possess only individual experiences. This study examines the integration of a new agent into a Karma DPG that is already at a Stationary Nash Equilibrium (SNE) and its ability to learn a policy through Deep Q-Networks (DQN) without prior model knowledge. It also derives a suboptimality bound for the policy learned, leveraging recent advancements in DQN convergence.