Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Emergent Communication for Rules Reasoning
Yuxuan Guo, Yifan Hao, Rui Zhang, Enshuai Zhou, Zidong Du, Xishan Zhang, Xinkai Song, Yuanbo Wen, Yongwei Zhao, Xuehai Zhou, Jiaming Guo, Qi Yi and 5 more
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 two-stage curriculum training (speaker pre-training on held-out rules) (ablation) | Reasoning Game agents · rule-RAVEN N=20/30/40 | train accuracy: ~0.9 vs low accuracy similar to message-blocked | Improved |
| generate candidates with rule-based distract rules instead of I-RAVEN style (ablation) | rule-RAVEN vs I-RAVEN · message-blocked listener | listener train accuracy (overfitting): ~0.9 -> ~0.6 | Improved |
| increase attribute values from 20 to 30 or 40 in I-RAVEN (ablation) | I-RAVEN · message-blocked listener | overfitting | No clear effect |
| measure Rule-Message vs Panel-Message Topsim | emerged language · ID split N=20 | Topsim: 0.1148 -> 0.2382 | Improved |
| increase attribute value diversity N from 20 to 40 | Reasoning Game agents · ID accuracy | accuracy: 0.9492 -> 0.9194 | No clear effect |
| generalize to unseen rule combinations (Inpo-ood vs ID) | emerged language · N=20 | accuracy: 0.9492 -> 0.9473 | No clear effect |
| rules seen on other attributes (Expo-ood-L1 vs Expo-ood-L2) | emerged language · N=20 | accuracy: 0.4921 -> 0.6590 | Improved |
| harder generalization from Inpo-ood to Expo-ood-L1 to Expo-ood-L2 | emerged language · averaged over N | accuracy: 0.92 -> 0.58 -> 0.46 | Hurt |
| transfer emerged language from source N=20 to target N=80 | new listeners · task transfer | train accuracy: 0.9199 | No clear effect |
| emerged language vs random mapping for transfer | new listeners · task transfer curves | convergence | Improved |
| emerged language vs ideal rule-encoding language for transfer | new listeners · task transfer curves | convergence speed | No clear effect |
| longer messages (M=6) and larger vocabulary (V=30) (ablation) | Reasoning Game agents · ID accuracy N=20 | accuracy: 0.9539 -> 0.9522 | No clear effect |
Closest papers by meaning
- Emergent Communication for Numerical Concepts Generalization (AAAI 2024 · similarity 0.87)
- Eliciting Reasoning in Language Models with Cognitive Tools (NeurIPS 2025 · similarity 0.87)
- Neural Reasoning about Agents’ Goals, Preferences, and Actions (AAAI 2024 · similarity 0.87)
- GAM-Agent: Game-Theoretic and Uncertainty-Aware Collaboration for Complex Visual Reasoning (NeurIPS 2025 · similarity 0.87)
- Playing ZendoWorld: Challenging AI Agents on Active Visual Concept Induction (NeurIPS 2026 · similarity 0.87)
- Benefits and Limitations of Communication in Multi-Agent Reasoning (ICLR 2026 · 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.
