AI Paper Atlas

Research map / Multi-agent systems and games / Cooperative multi-agent RL

Disentangling Sources of Risk for Distributional Multi-Agent Reinforcement Learning

Kyunghwan Son, Junsu Kim, Sungsoo Ahn, Roben D Delos Reyes, Yung Yi, Jinwoo Shin

ICML 2022 · Multi-agent systems and games › Cooperative multi-agent RL · approach: MARL benchmarks and grouping

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
set DMIX risk sensitivity from neutral to seeking (ablation)DMIX · Stochastic two-step matrix gametest reward: -0.02 -> 0.03No clear effect
disentangle cooperative (seeking) and environmental (neutral) risk using DRIMA instead of DMIXDRIMA · Stochastic two-step matrix gametest reward: 0.03 -> 6.98Improved
disentangle risk sources in DRIMA versus eight baselinesDRIMA · SMAC explorative and dilemmatic scenariosbattle won meanImproved
cooperative risk-neutral instead of seeking in DRIMA (environment seeking) (ablation)DRIMA · SMAC MMM2-v1battle won meanHurt
environmental risk neutral or averse instead of seeking in DRIMA (ablation)DRIMA · SMAC MMM2-v1battle won meanHurt
risk-sensitive variants of DDN and DMIX (quantile range [0, 0.25] averse, [0.75, 1] seeking) versus DRIMA (ablation)DDN and DMIX · SMAC MMM2-v1battle won meanHurt
vary environmental risk level (seeking, neutral, averse) on easy tasks (ablation)DRIMA · 5m_vs_6m, 3s_vs_5z, 8m_vs_9mbattle won meanNo clear effect
environmental risk-averse setting (ablation)DRIMA · 8m_vs_9m-v1win rateImproved
control risk sensitivity with random or adversarial noisy agentsDRIMA · SMAC 3s_vs_5z and 5m_vs_6m noisybattle won meanImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy