What question did this study set out to answer?

The aim is to enhance multiagent reinforcement learning algorithms to limit the impact of adversarial attacks on cooperative networks.

Resilient Multiagent Reinforcement Learning With Function Approximation

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial shows improved performance of agents under adversarial attacks, suggesting robustness in cooperative settings.

The aim is to enhance multiagent reinforcement learning algorithms to limit the impact of adversarial attacks on cooperative networks.
Proposed a resilient consensus-based actor-critic algorithm for decentralized agents.
Agents communicate parameter vectors and estimate team-average rewards.
Analysis demonstrated convergence of cooperative agents' estimates in the presence of Byzantine agents.
Estimates of cooperative agents converge to a bounded consensus value with probability one when there are at most H Byzantine agents.
Proved that agent policies converge with probability one to a bounded neighborhood around a stationary point.
Showed the robustness of performance even with adversarial agent interference.

Ye et al. (2024) studied this question.

View Full Paper

Ask AI

Bookmark