Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols
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 straight-through Gumbel-softmax instead of REINFORCE | Referential-game agents · MSCOCO 2014 validation | communication success | Helped |
| Use straight-through Gumbel-softmax instead of REINFORCE | Referential-game agents · MSCOCO 2014 training | updates to convergence: five-fold more updates | Helped |
| Increase maximum message length | ST-GS referential-game agents · MSCOCO 2014 | updates to convergence | Helped |
| Use REINFORCE instead of ST-GS | Referential-game agents with maximum message length L = 1 · MSCOCO 2014 | communication success | Helped |
| Use greedy argmax encoding instead of sampling | Learned communication protocol · MSCOCO 2014 validation | communication success | Helped |
| Learn the temperature instead of using vanilla ST-GS (ablation) | ST-GS referential-game agents · MSCOCO 2014 validation | communication success: +0.8% | Helped |
| Add KL regularization toward a natural-language model (ablation) | Learned communication protocol · MSCOCO 2014 validation | communication success (%): 95.65 -> 52.51 | Hurt |
| Add KL regularization toward a natural-language model (ablation) | Learned communication protocol · MSCOCO 2014 validation | omission score: 0.193 -> 0.258 | Helped |
| Add communication loss to image-caption training (ablation) | Image captioning model · Image captioning (MSCOCO) | BLEU-2: 0.436 -> 0.435 | No clear effect |
| Add communication loss to image-caption training (ablation) | Image captioning model · Image captioning (MSCOCO) | CIDEr: 0.594 -> 0.590 | No clear effect |
| Add communication loss to image-caption training (ablation) | Image captioning model · Image captioning (MSCOCO) | average length: 12.85 -> 13.93 | Helped |
Closest papers by meaning
- Emergence of Linguistic Communication from Referential Games with Symbolic and Pixel Input (ICLR 2018 · similarity 0.93)
- Multi-Agent Cooperation and the Emergence of (Natural) Language (ICLR 2017 · similarity 0.92)
- Lewis's Signaling Game as beta-VAE For Natural Word Lengths and Segments (ICLR 2024 · similarity 0.89)
- Multi-agent KTO: Enhancing Strategic Interactions of Large Language Model in Language Game (NeurIPS 2025 · similarity 0.89)
- A compressive-expressive communication framework for compositional representations (NeurIPS 2025 · similarity 0.88)
- Improving Cooperation in Language Games with Bayesian Inference and the Cognitive Hierarchy (AAAI 2025 · similarity 0.88)
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.