Research map / Multi-agent systems and games / Games and equilibria
Mastering the Game of No-Press Diplomacy via Human-Regularized Reinforcement Learning and Planning
Anton Bakhtin, David J Wu, Adam Lerer, Jonathan Gray, Athul Paul Jacob, Gabriele Farina, Alexander H Miller, Noam Brown
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 |
|---|---|---|---|
| Use RL-DiL-piKL (DiL-piKL in self-play and test-time search) instead of DORA's RM | Diplodocus vs DORA · Agent population | Score against population (%): 29% vs 13% | Improved |
| Use RL-DiL-piKL instead of DORA's RM in self-play and search | Diplodocus vs DORA · Human tournament | Elo: 181 vs -20 | Improved |
| Sample lambda from a distribution (DiL-piKL) with regularized self-play, versus unregularized self-play (DNVI/DORA) (ablation) | RL agents, human prediction accuracy · Held-out human games | Human move prediction accuracy | Improved |
| Regularized self-play (DiL-piKL) instead of unregularized self-play with BC init (DNVI) (ablation) | RL training · Agent population | Score against population (%): 28% vs 15% | Improved |
| Do not feed policy proposal network back to self-play (DNVI-NPU) relative to DNVI (ablation) | DNVI · Agent population | Score against population (%): 20% vs 15% | Improved |
| Sampling lambda from a distribution in DiL-piKL during planning vs fixed lambda piKL | piKL planning · Discussion/design claim | Score | No clear effect |
| Best response to BC (BRBot) as a human-tournament baseline versus Diplodocus agents | BRBot vs Diplodocus · Human tournament | Elo: 119 vs 152-181 | Hurt |
Closest papers by meaning
- DipLLM: Fine-Tuning LLM for Strategic Decision-making in Diplomacy (ICML 2025 · similarity 0.86)
- Outbidding and Outbluffing Elite Humans: Mastering Liar’s Poker via Self-Play and Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- Learning in Stackelberg Mean Field Games: A Non-Asymptotic Analysis (NeurIPS 2025 · similarity 0.86)
- A Minimaximalist Approach to Reinforcement Learning from Human Feedback (ICML 2024 · similarity 0.86)
- Learning Diverse Risk Preferences in Population-Based Self-Play (AAAI 2024 · similarity 0.86)
- Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games (NeurIPS 2024 · 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.
