Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Learning Policy Representations in Multiagent Systems
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 |
|---|---|---|---|
| combine imitation and agent identification objectives (ablation) | policy representations · RoboSumo weak generalization | IICR: 0.25 -> 0.22 | Helped |
| combine imitation and agent identification objectives (ablation) | policy representations · RoboSumo weak generalization | accuracy: 0.67 -> 0.73 | Helped |
| combine imitation and agent identification objectives (ablation) | policy representations · RoboSumo strong generalization | accuracy: 0.56 -> 0.56 | No clear effect |
| use imitation objective without agent identification (ablation) | policy representations · RoboSumo weak generalization | IICR: 0.25 -> 0.24 | Helped |
| use imitation objective without agent identification (ablation) | policy representations · RoboSumo strong generalization | accuracy: 0.56 -> 0.60 | Helped |
| use agent identification objective without imitation (ablation) | policy representations · ParticleWorld weak generalization | IICR: 0.58 -> 0.50 | Helped |
| use agent identification objective without imitation (ablation) | policy representations · ParticleWorld strong generalization | IICR: 0.86 -> 0.82 | Helped |
| condition the speaker policy on listener representations | MADDPG speaker policy · ParticleWorld test | average reward: -18.99 -> -17.20 | Helped |
| condition the speaker policy on listener representations | MADDPG speaker policy · ParticleWorld train | average reward: -11.66 -> -11.77 | Hurt |
| use Emb-Hyb rather than Emb-Im listener representations (ablation) | MADDPG speaker policy · ParticleWorld test | average reward: -17.75 -> -17.20 | Helped |
| use offline rather than online embeddings | RoboSumo embedding-conditioned policy · RoboSumo held-out opponents | win rate | Helped |
| provide an incorrect all-zero or random-opponent embedding (ablation) | RoboSumo embedding-conditioned policy · RoboSumo held-out opponents | win rate | Hurt |
Closest papers by meaning
- Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments (NeurIPS 2017 · similarity 0.90)
- Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games (NeurIPS 2024 · similarity 0.89)
- Multi-agent cooperation through learning-aware policy gradients (ICLR 2025 · similarity 0.88)
- Make-An-Agent: A Generalizable Policy Network Generator with Behavior-Prompted Diffusion (NeurIPS 2024 · similarity 0.87)
- Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspective (NeurIPS 2025 · similarity 0.87)
- Enhancing Cooperative Multi-Agent Reinforcement Learning with State Modelling and Adversarial Exploration (ICML 2025 · similarity 0.87)
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.