Research map / Multi-agent systems and games / LLM agents
Multi-Agent Discussion Mechanism for Natural Language Generation
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 |
|---|---|---|---|
| use multi-agent discussion with 3 rounds instead of single-agent conditional copy (ablation) | multi-agent GRU encoder-decoder · RotoWire | BLEU: 14.49 -> 15.88 | Helped |
| increase discussion rounds from 3 to 5 (ablation) | multi-agent discussion · RotoWire | BLEU: 15.88 -> 14.50 | Hurt |
| use 1 discussion round instead of more (ablation) | multi-agent discussion · RotoWire | perplexity: 7.59 (5 rounds) -> 7.41 (1 round) | Helped |
| increase discussion rounds from 1 to 3 (ablation) | multi-agent discussion · MS COCO | CIDEr: 58.4 -> 66.6 | Helped |
| increase discussion rounds from 3 to 5 (ablation) | multi-agent discussion · MS COCO | CIDEr: 66.6 -> 56.0 | Hurt |
| replace layer-wise encoder communication with decoder-step discussion (ablation) | multi-agent discussion · CNN/DailyMail | ROUGE-1: 41.11 -> 44.26 | Helped |
| replace layer-wise encoder communication with decoder-step discussion (ablation) | multi-agent discussion · CNN/DailyMail | ROUGE-2: 18.21 -> 18.16 | No clear effect |
| use discussion with 3 rounds over single-agent encoders (ablation) | multi-agent discussion · MS COCO | BLEU-4: 21.9 -> 22.6 | Helped |
Closest papers by meaning
- Chain of Agents: Large Language Models Collaborating on Long-Context Tasks (NeurIPS 2024 · similarity 0.87)
- Belief-Calibrated Multi-Agent Consensus Seeking for Complex NLP Tasks (NeurIPS 2025 · similarity 0.86)
- Communicating Activations Between Language Model Agents (ICML 2025 · similarity 0.85)
- Latent Collaboration in Multi-Agent Systems (ICML 2026 · similarity 0.85)
- The Consensus Game: Language Model Generation via Equilibrium Search (ICLR 2024 · similarity 0.85)
- ToolACE-MT: Non-Autoregressive Generation for Agentic Multi-Turn Interaction (ICLR 2026 · similarity 0.85)
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.