Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Task-Driven Modular Networks for Zero-Shot Compositional Learning
Senthil Purushwalkam, Maximilian Nickel, Abhinav Gupta, Marc'Aurelio Ranzato
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 |
|---|---|---|---|
| add task-driven gatings (ablation) | TMN · MIT-States | Top-1 val AUC x100: 3.2 -> 3.5 | Improved |
| add task-driven gatings (ablation) | TMN · UT-Zappos | Top-1 val AUC x100: 32.7 -> 36.8 | Improved |
| add joint extraction of image and pair features (embedding at input rather than output) (ablation) | TMN without task-driven gatings · MIT-States | Top-1 val AUC x100: 0.8 -> 3.2 | Improved |
| add joint extraction of image and pair features (embedding at input rather than output) (ablation) | TMN without task-driven gatings · UT-Zappos | Top-1 val AUC x100: 20.1 -> 32.7 | Improved |
| add ConceptDrop regularization (ablation) | TMN · MIT-States | Top-1 val AUC x100: 3.3 -> 3.5 | Improved |
| add ConceptDrop regularization (ablation) | TMN · UT-Zappos | Top-1 val AUC x100: 35.7 -> 36.8 | Improved |
| increase layers from 1 to 3 (24 modules) (ablation) | TMN · MIT-States val | Val AUC x100: 2.50 -> 3.51 | Improved |
| increase layers from 3 to 5 (24 modules) (ablation) | TMN · MIT-States val | Val AUC x100: 3.51 -> 3.24 | Hurt |
| increase modules per layer from 12 to 24 (3 layers) (ablation) | TMN · MIT-States val | Val AUC x100: 3.23 -> 3.51 | Improved |
| increase modules per layer from 12 to 30 (5 layers) (ablation) | TMN · MIT-States val | Val AUC x100: 3.48 -> 3.19 | Hurt |
| use TMN instead of RedWine | RedWine · MIT-States test | Top-1 test AUC x100: 2.4 -> 2.9 | Improved |
| use TMN instead of RedWine | RedWine · UT-Zappos test | Top-1 test AUC x100: 27.1 -> 29.3 | Improved |
Closest papers by meaning
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.88)
- Scaling can lead to compositional generalization (NeurIPS 2025 · similarity 0.88)
- Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.88)
- Beyond Image Classification: A Video Benchmark and Dual-Branch Hybrid Discrimination Framework for Compositional Zero-Shot Learning (CVPR 2025 · similarity 0.88)
- Unlocking Compositional Generalization in Continual Few-Shot Learning (NeurIPS 2026 · similarity 0.87)
- Compositional Few-Shot Class-Incremental Learning (ICML 2024 · 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.
