Research map / Multi-agent systems and games / Human-AI and goal-conditioned
PatchGame: Learning to Signal Mid-level Patches in Referential Games
Kamal Gupta, Gowthami Somepalli, Anubhav Gupta, Vinoj Jayasundara, Matthias Zwicker, Abhinav Shrivastava
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 PatchGame's listener vision representations instead of PatchVAE (S=32) representations | ImageNet k-NN classification · ImageNet val | Top-1 accuracy: 21.3 ($\pm$ 0.5) -> 30.3 ($\pm$ 0.5) | Improved |
| Use PatchGame's listener vision representations instead of MoCo-v2 representations | ImageNet k-NN classification · ImageNet val | Top-1 accuracy: 36.8 ($\pm$ 0.2) -> 30.3 ($\pm$ 0.5) | Hurt |
| Use PatchGame's listener representations instead of PatchVAE representations | Pascal VOC transfer · Pascal VOC | mAP-50: 54.2 -> 61.3 | Improved |
| Use PatchGame's PatchSymbol instead of PatchVAE symbols | Visual BoW image classification · ImageNet val | Top-1 accuracy: 1.02 ($\pm$ 0.07) -> 1.28 ($\pm$ 0.06) | Improved |
| Remove color jitter (ablation) | PatchGame · ImageNet train/val | Top-1 accuracy (k-NN): 26.3 -> 23.2 | Hurt |
| Remove random resized crops (ablation) | PatchGame · ImageNet train/val | Top-1 accuracy (k-NN): 26.3 -> 22.9 | Hurt |
| Increase batch size from 128 to 256 (ablation) | PatchGame · ImageNet train/val | Top-1 accuracy: 47.2 -> 49.1 | Improved |
| Increase batch size from 256 to 512 (ablation) | PatchGame · ImageNet train/val | Top-1 accuracy: 49.1 -> 49.9 | Improved |
| Use a larger vocabulary or longer messages (ablation) | PatchGame communication game · ImageNet validation communication task | Top-1 communication accuracy | Improved |
Closest papers by meaning
- Learning to Communicate and Solve Visual Blocks-World Tasks (AAAI 2019 · similarity 0.86)
- Learning Multi-Object Positional Relationships via Emergent Communication (AAAI 2024 · similarity 0.86)
- Emergent Communication of Generalizations (NeurIPS 2021 · similarity 0.86)
- Modeling Conceptual Understanding in Image Reference Games (NeurIPS 2019 · similarity 0.86)
- Iterated Learning Improves Compositionality in Large Vision-Language Models (CVPR 2024 · similarity 0.85)
- CodeV: Code with Images for Faithful Visual Reasoning via Tool-Aware Policy Optimization (CVPR 2026 · similarity 0.83)
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.
