Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Emergent Communication of Generalizations
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 |
|---|---|---|---|
| Replace single-object reference with shared-input set communication | ShapeWorld agents · ShapeWorld | AMI: 0.04 (0.00) -> 0.59 (0.08) | Helped |
| Replace single-object reference with generalization communication from separate examples | ShapeWorld agents · ShapeWorld | AMI: 0.04 (0.00) -> 0.66 (0.07) | Helped |
| Replace single-object reference with shared-input set communication | Birds agents · Birds | AMI: 0.05 (0.00) -> 0.17 (0.02) | Helped |
| Replace single-object reference with generalization communication from separate examples | Birds agents · Birds | AMI: 0.05 (0.00) -> 0.26 (0.02) | Helped |
| Communicate larger sets of targets (ablation) | Setref and concept games · ShapeWorld and Birds | topographic rho: ShapeWorld setref: ρ = 0.39, p = 0.029; concept: ρ = 0.75, p < 10^{-5}; Birds setref: ρ = 0.90, p < 10^{-7}; concept: ρ | Helped |
| Increase communication bandwidth (ablation) | Reference, setref, and concept agents · ShapeWorld and Birds | accuracy | Helped |
| Increase communication bandwidth (ablation) | Reference, setref, and concept agents · ShapeWorld and Birds | information-theoretic systematicity | Hurt |
| Train setref and concept agents rather than reference agents | Communication agents · ShapeWorld and Birds | accuracy on unseen concepts: ShapeWorld: 98 -> 87 (Setref), 98 -> 75 (Concept); Birds: 89 -> 78 (Setref), 89 -> 73 (Concept) | Hurt |
| Train reference agents and evaluate zero-shot on setref or concept games | Reference agents · ShapeWorld and Birds, zero-shot game transfer | accuracy: ShapeWorld: 56 (Setref), 50 (Concept); Birds: 64 (Setref), 56 (Concept) | Hurt |
| Train setref agents and evaluate zero-shot on concept games | Setref agents · ShapeWorld and Birds, zero-shot game transfer | accuracy: ShapeWorld: 90; Birds: 75 | Mixed |
| Use ACRe rather than the closest-concept message baseline | ShapeWorld setref and concept languages · Held-out concepts | student accuracy: Setref: 56 (0.7) -> 65 (3.4); Concept: 56 (1.3) -> 70 (4.0) | Helped |
| Use ACRe rather than random teacher messages | ShapeWorld setref and concept languages · Held-out concepts | student accuracy: Setref: 50 (0.3) -> 65 (3.4); Concept: 50 (0.4) -> 70 (4.0) | Helped |
Closest papers by meaning
- Learning Multi-Object Positional Relationships via Emergent Communication (AAAI 2024 · similarity 0.91)
- Emergent Communication for Numerical Concepts Generalization (AAAI 2024 · similarity 0.90)
- CtD: Composition through Decomposition in Emergent Communication (ICLR 2025 · similarity 0.89)
- A compressive-expressive communication framework for compositional representations (NeurIPS 2025 · similarity 0.89)
- Speaking Your Language: Spatial Relationships in Interpretable Emergent Communication (NeurIPS 2024 · similarity 0.89)
- Lewis's Signaling Game as beta-VAE For Natural Word Lengths and Segments (ICLR 2024 · 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.