Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Dynamic population-based meta-learning for multi-agent communication with natural language
Abhinav Gupta, Marc Lanctot, Angeliki Lazaridou
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 dynamic population-based meta-learning instead of prior methods | Referential game agents · MSCOCO image game test captions | BLEU: 28.4 ± 0.32 -> 29.3 ± 0.25 | Improved |
| Use dynamic population-based meta-learning instead of prior methods | Referential game agents · IWSLT'14 English-German text game | BLEU: 23.9 ± 0.21 -> 25.2 ± 0.16 | Improved |
| Use top-k sampling instead of greedy decoding | Pretrained text-game speaker · IWSLT'14 English-German text game | BLEU: 20.8 ± 0.06 -> 24.6 ± 0.05 | Improved |
| Use dynamic population-based meta-learning rather than Pretrained | Referential game agents · Flickr8k cross-task test (trained on MSCOCO) | referential accuracy: 59.7 ± 0.9 -> 70.2 ± 1.8 | Improved |
| Use dynamic population-based meta-learning rather than Pretrained | Referential game agents · Flickr8k within-task test | referential accuracy: 65.3 ± 0.3 -> 78.9 ± 1.4 | Improved |
| Add interaction with the meta-agent when building the agent population (ablation) | Dynamic population-based meta-learning · MSCOCO image game test | referential accuracy | Improved |
| Train the meta-agent iteratively instead of resetting its weights and training from scratch (ablation) | Dynamic population-based meta-learning · MSCOCO image game test | referential accuracy | Improved |
| Integrate interactive and supervised learning in separate phases rather than the full method's grounding setup (ablation) | Dynamic population-based meta-learning · MSCOCO image game test | referential accuracy | No clear effect |
| Use FOMAML or Reptile instead of MAML (ablation) | Dynamic population-based meta-learning · MSCOCO image game | referential accuracy | No clear effect |
Closest papers by meaning
- Adaptive Collaboration with Humans: Metacognitive Policy Optimization for Multi-Agent LLMs with Continual Learning (ICLR 2026 · similarity 0.89)
- Language Grounded Multi-agent Reinforcement Learning with Human-interpretable Communication (NeurIPS 2024 · similarity 0.89)
- LLM-Assisted Semantically Diverse Teammate Generation for Efficient Multi-agent Coordination (ICML 2025 · similarity 0.89)
- Learning to Deliberate: Meta-policy Collaboration for Agentic LLMs with Multi-agent Reinforcement Learning (AAAI 2026 · similarity 0.88)
- Emergent Coordination in Multi-Agent Language Models (ICLR 2026 · similarity 0.88)
- Evaluating Generalization Capabilities of LLM-Based Agents in Mixed-Motive Scenarios Using Concordia (NeurIPS 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.
