Research map / Multi-agent systems and games / Games and equilibria
Conservative Offline Policy Adaptation in Multi-Agent Games
Chengjie Wu, Pingzhong Tang, Jun Yang, Yujing Hu, Tangjie Lv, Changjie Fan, Chongjie Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add conservative dataset-constrained self-play instead of BC-First proxy | BC-First · Maze case 1 | test reward: -4 -> 1 | Improved |
| remove exploitation (Self-Play, i.e. CSP without dataset) (ablation) | CSP · Maze case 2 | test reward: 16 -> 1 | Hurt |
| use CSP instead of pure dataset-following | Dataset-Only · Maze case 1 | test reward: -64 -> 1 | Improved |
| use CSP instead of BC-First | BC-First · MPE predator-prey, adversaries, 10R | episode reward: 21.3 -> 34.6 | Improved |
| increase dataset quality 10R to 100S | BC-First · MPE good agent | episode reward: -29.8 -> -18.5 | Improved |
| increase dataset quality 10R to 100S | CSP · MPE good agent | episode reward: -16.7 -> -17.1 | No clear effect |
| use MAIL imitation instead of CSP | CSP · MPE adversaries, 10R | episode reward: 34.6 -> 4.3 | Hurt |
| use CSP instead of BC-First | BC-First · MuJoCo YouShallNotPassHumans target 2 | winning rate: 0.46 -> 0.65 | Improved |
| use CSP instead of Self-Play | Self-Play · MuJoCo YouShallNotPassHumans target 3 | episode reward: 318 -> 1203 | Improved |
| use CSP instead of BC-First | BC-First · Google Football 3vs1 defender target 1 | winning rate: 0.64 -> 0.9 | Improved |
| use CSP instead of BC-First | BC-First · Google Football 3vs1 attacker target 1 | winning rate: 0.83 -> 0.81 | No clear effect |
| use BC-First (non-conservative) instead of Self-Play on MPE adversaries 100S | Self-Play · MPE adversaries, 100S | episode reward: 29.7 -> 27.8 | Hurt |
Closest papers by meaning
- From Static Policies to Adaptive Priors in Offline Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
- Offline Multi-Agent Reinforcement Learning via In-Sample Sequential Policy Optimization (AAAI 2025 · similarity 0.89)
- An Offline Adaptation Framework for Constrained Multi-Objective Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Bayesian Design Principles for Offline-to-Online Reinforcement Learning (ICML 2024 · similarity 0.89)
- A2PO: Towards Effective Offline Reinforcement Learning from an Advantage-aware Perspective (NeurIPS 2024 · similarity 0.89)
- Constrained Exploitability Descent: An Offline Reinforcement Learning Method for Finding Mixed-Strategy Nash Equilibrium (ICML 2025 · similarity 0.89)
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.
