Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Emergent Communication: Generalization and Overfitting in Lewis Games
Mathieu Rita, Corentin Tallec, Paul Michel, Jean-Bastien Grill, Olivier Pietquin, Emmanuel Dupoux, Florian Strub
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 |
|---|---|---|---|
| early-stopped listener instead of continuous listener (ablation) | Lewis game agents · Reconstruction Lewis game | Generalization: 0.58 -> 0.95 | Improved |
| early-stopped listener instead of continuous listener (ablation) | Lewis game agents · Reconstruction Lewis game | Compositionality: 0.22 -> 0.39 | Improved |
| top partial listener (tuned N_step) instead of continuous listener (ablation) | Lewis game agents · Reconstruction Lewis game | Test co-adaptation loss: 4.64 -> 0.97 | Improved |
| add dropout to listener (ablation) | Continuous listener · Reconstruction Lewis game | Generalization: 0.58 -> 0.64 | Improved |
| remove layer norm from listener (ablation) | Continuous listener · Reconstruction Lewis game | Generalization: 0.58 -> 0.70 | Improved |
| add weight decay to listener (ablation) | Continuous listener · Reconstruction Lewis game | Generalization: 0.58 -> 0.72 | Improved |
| remove layer norm and add weight decay (ablation) | Continuous listener · Reconstruction Lewis game | Generalization: 0.58 -> 0.87 | Improved |
| dropout on listener (ablation) | Continuous listener · Reconstruction Lewis game | Test co-adaptation loss: 4.64 -> 4.86 | No clear effect |
| early-stopped listener on CelebA 1/100 subset (ablation) | Discrimination game · CelebA 1/100 | Generalization: 0.39 -> 0.69 | Improved |
| early-stopped listener on CelebA full data (ablation) | Discrimination game · CelebA 1/1 | Generalization: 0.94 -> 0.97 | Improved |
| early-stopped listener on ImageNet 1/100 subset (ablation) | Discrimination game · ImageNet 1/100 | Generalization: 0.51 -> 0.64 | Improved |
| early-stopped listener on ImageNet full data (ablation) | Discrimination game · ImageNet 1/1 | Generalization: 0.96 -> 0.98 | Improved |
Closest papers by meaning
- Lewis's Signaling Game as beta-VAE For Natural Word Lengths and Segments (ICLR 2024 · similarity 0.91)
- Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols (NeurIPS 2017 · similarity 0.91)
- Emergent Communication with Conversational Repair (ICLR 2024 · similarity 0.91)
- Emergent Communication of Generalizations (NeurIPS 2021 · similarity 0.91)
- Semantics and Spatiality of Emergent Communication (NeurIPS 2024 · similarity 0.87)
- A compressive-expressive communication framework for compositional representations (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.
