Research map / Label-efficient and robust learning / Recognition and long-tail
Towards Context-Agnostic Learning Using Synthetic Data
Charles Jin, Martin Rinard
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 PGD refinement to random-context synthetic training (ablation) | Context-agnostic classifier trained from one synthetic example per class · GTSRB test | accuracy (%): 72.1 -> 86.4 | Improved |
| Add previous-image context bias correction to random-context training (ablation) | Context-agnostic classifier trained from Picto · GTSRB test | accuracy (%): 72.1 -> 87.3 | Improved |
| Combine bias correction with PGD refinement (ablation) | Context-agnostic classifier trained from Picto · GTSRB test | accuracy (%): 72.0 -> 95.9 | Improved |
| Add PGD refinement to random-context synthetic training (ablation) | Context-agnostic classifier trained from Digit · MNIST | accuracy (%): 88.3 -> 89.7 | Improved |
| Add previous-image context bias correction to random-context training (ablation) | Context-agnostic classifier trained from Digit · MNIST | accuracy (%): 81.9 -> 89.2 | Improved |
| Combine bias correction with PGD refinement (ablation) | Context-agnostic classifier trained from Digit · MNIST | accuracy (%): 81.9 -> 90.2 | Improved |
| Add PGD refinement to random-context synthetic training (ablation) | Context-agnostic classifier trained from Omnifont · Omniglot 20-way | accuracy (%): 69.8 -> 90.8 | Improved |
| Add previous-image context bias correction to random-context training (ablation) | Context-agnostic classifier trained from Omnifont · Omniglot 20-way | accuracy (%): 71.9 -> 80.5 | Improved |
| Combine bias correction with PGD refinement (ablation) | Context-agnostic classifier trained from Omnifont · Omniglot 20-way | accuracy (%): 71.9 -> 92.2 | Improved |
| Reuse random backgrounds as contexts (ablation) | Synthetic classifier trained from Picto · GTSRB test | accuracy (%): 72.0 -> 72.1 | No clear effect |
| Reuse random backgrounds as contexts (ablation) | Synthetic classifier trained from Omnifont · Omniglot 20-way | accuracy (%): 71.9 -> 69.8 | Hurt |
| Remove bias correction from the full method (ablation) | Context-agnostic classifier trained from Picto · GTSRB test | accuracy (%): 95.9 -> 86.4 | Hurt |
Closest papers by meaning
- Provably Improving Generalization of Few-shot models with Synthetic Data (ICML 2025 · similarity 0.87)
- Real-Fake: Effective Training Data Synthesis Through Distribution Matching (ICLR 2024 · similarity 0.86)
- Dataset Distillation for Pre-Trained Self-Supervised Vision Models (NeurIPS 2025 · similarity 0.86)
- The Unmet Promise of Synthetic Training Images: Using Retrieved Real Images Performs Better (NeurIPS 2024 · similarity 0.85)
- Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations (NeurIPS 2024 · similarity 0.85)
- Scaling Laws of Synthetic Images for Model Training ... for Now (CVPR 2024 · similarity 0.84)
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.
