English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Robust multi-agent reinforcement learning (MARL) necessitates resilience touncertain or worst-case actions by unknown allies. Existing max-minoptimization techniques in robust MARL seek to enhance resilience by trainingagents against worst-case adversaries, but this becomes intractable as thenumber of agents grows, leading to exponentially increasing worst-casescenarios. Attempts to simplify this complexity often yield overly pessimisticpolicies, inadequate robustness across scenarios and high computationaldemands. Unlike these approaches, humans naturally learn adaptive and resilientbehaviors without the necessity of preparing for every conceivable worst-casescenario. Motivated by this, we propose MIR2, which trains policy in routinescenarios and minimize Mutual Information as Robust Regularization.Theoretically, we frame robustness as an inference problem and prove thatminimizing mutual information between histories and actions implicitlymaximizes a lower bound on robustness under certain assumptions. Furtheranalysis reveals that our proposed approach prevents agents from overreactingto others through an information bottleneck and aligns the policy with a robustaction prior. Empirically, our MIR2 displays even greater resilience againstworst-case adversaries than max-min optimization in StarCraft II, Multi-agentMujoco and rendezvous. Our superiority is consistent when deployed inchallenging real-world robot swarm control scenario. See code and demo videosin Supplementary Materials.

MIR2: Towards Provably Robust Multi-Agent Reinforcement Learning by Mutual Information Regularization