Fully Byzantine-Resilient Distributed Multi-Agent Q-Learning Over Compromised Networks
arXiv·medium signal
Addresses Byzantine-resilient distributed multi-agent reinforcement learning where agents must collaboratively learn optimal value functions over a compromised communication network. Existing resilient MARL approaches only guarantee approximate convergence; this method achieves full convergence guarantees despite arbitrary Byzantine failures — critical for deploying multi-agent systems in adversarial environments.