When Can We Learn General-Sum Markov Games with a Large Number of Players Sample-Efficiently?
This work addresses the theoretical bottleneck of exponential sample complexity in multi-agent reinforcement learning for large-scale games, offering polynomial improvements that could enable more practical applications in domains like robotics or economics.
This paper tackles the problem of sample-efficient learning in multi-agent general-sum Markov games with many players, where existing methods have exponential sample complexity. It presents algorithms for learning coarse correlated equilibria and correlated equilibria with polynomial sample complexities in the number of actions per player, and for Markov potential games, it achieves an approximate Nash equilibrium with linear dependence on the total actions.
Multi-agent reinforcement learning has made substantial empirical progresses in solving games with a large number of players. However, theoretically, the best known sample complexity for finding a Nash equilibrium in general-sum games scales exponentially in the number of players due to the size of the joint action space, and there is a matching exponential lower bound. This paper investigates what learning goals admit better sample complexities in the setting of $m$-player general-sum Markov games with $H$ steps, $S$ states, and $A_i$ actions per player. First, we design algorithms for learning an $ε$-Coarse Correlated Equilibrium (CCE) in $\widetilde{\mathcal{O}}(H^5S\max_{i\le m} A_i / ε^2)$ episodes, and an $ε$-Correlated Equilibrium (CE) in $\widetilde{\mathcal{O}}(H^6S\max_{i\le m} A_i^2 / ε^2)$ episodes. This is the first line of results for learning CCE and CE with sample complexities polynomial in $\max_{i\le m} A_i$. Our algorithm for learning CE integrates an adversarial bandit subroutine which minimizes a weighted swap regret, along with several novel designs in the outer loop. Second, we consider the important special case of Markov Potential Games, and design an algorithm that learns an $ε$-approximate Nash equilibrium within $\widetilde{\mathcal{O}}(S\sum_{i\le m} A_i / ε^3)$ episodes (when only highlighting the dependence on $S$, $A_i$, and $ε$), which only depends linearly in $\sum_{i\le m} A_i$ and significantly improves over existing efficient algorithm in the $ε$ dependence. Overall, our results shed light on what equilibria or structural assumptions on the game may enable sample-efficient learning with many players.