Research map / Multi-agent systems and games / Cooperative multi-agent RL
Attention-Based Recurrence for Multi-Agent Reinforcement Learning under Stochastic Partial Observability
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace the true state with attention-based joint memory representations | QMIX-based AERIAL · Dec-Tiger | average return: about -15 -> about zero | Helped |
| add self-attention over agents' memory representations (ablation) | AERIAL (no attention) · Dec-Tiger | average return: about -8 -> about zero | Helped |
| replace state-based CTDE with attention-based joint memory representations | QMIX-based AERIAL · MessySMAC | win rate | Helped |
| add self-attention over agents' memory representations (ablation) | AERIAL (no attention) · MessySMAC | win rate | Helped |
| use raw joint history concatenated with the true state instead of learned memory representations (ablation) | AERIAL · MessySMAC | win rate | Hurt |
| replace the true state with joint memory representations | AERIAL · Original SMAC | win rate | No clear effect |
| use QPLEX instead of QMIX as the factorization operator (ablation) | AERIAL · SMAC and MessySMAC | performance | No clear effect |
| increase observation-negation probability | AERIAL and state-based CTDE methods · MessySMAC | normalized win rate | Hurt |
| increase the number of initial random steps | AERIAL and state-based CTDE methods · MessySMAC | normalized win rate | Hurt |
| use attention-based recurrence rather than state-based CTDE | AERIAL · MessySMAC with φ ≥ 15% or K ≥ 10 | win rate | Helped |
Closest papers by meaning
- Deep Decentralized Multi-task Multi-Agent Reinforcement Learning under Partial Observability (ICML 2017 · similarity 0.89)
- SMACv2: An Improved Benchmark for Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2023 · similarity 0.88)
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (NeurIPS 2024 · similarity 0.87)
- Rainbow Delay Compensation: A Multi-Agent Reinforcement Learning Framework for Mitigating Observation Delays (NeurIPS 2025 · similarity 0.87)
- Learning Decentralized Partially Observable Mean Field Control for Artificial Collective Behavior (ICLR 2024 · similarity 0.86)
- Independent Learning of Nash Equilibria in Partially Observable Markov Potential Games with Decoupled Dynamics (NeurIPS 2026 · similarity 0.86)
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.