Research map / Label-efficient and robust learning / Domain adaptation and generalization
Selecting Data Augmentation for Simulating Interventions
Maximilian Ilse, Jakub M Tomczak, Patrick Forré
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 |
|---|---|---|---|
| Apply random rotations selected by SDA | ERM on Rotated MNIST · Rotated MNIST average | accuracy: 88.3 -> 95.9 | Improved |
| Apply hue and translation augmentation selected by SDA | ERM on Colored MNIST · Colored MNIST test | accuracy: 68.7 $\pm$ 0.9 -> 74.1 $\pm$ 0.9 | Improved |
| Apply color perturbations selected by SDA | ERM on PACS · PACS average | accuracy: 73.4 -> 74.9 | Improved |
| Apply selected color perturbations | ERM on PACS · PACS cartoon | accuracy: 71.7 -> 68.49 | Hurt |
| Apply selected color perturbations | ERM on PACS · PACS photo | accuracy: 91.1 -> 88.35 | Hurt |
| Use all listed augmentations instead of SDA-selected augmentations (ablation) | SDA on Rotated MNIST · Rotated MNIST average | accuracy: 95.9 -> 70.5 | Hurt |
| Use all listed augmentations instead of SDA-selected augmentations (ablation) | SDA on Colored MNIST · Colored MNIST test | accuracy: 74.1 -> 65.4 | Hurt |
| Use all listed augmentations instead of SDA-selected augmentations (ablation) | SDA on PACS · PACS average | accuracy: 74.9 -> 58.8 | Hurt |
| Expand the rotation range from [-15°, 15°] to [0°, 359°] (ablation) | Rotation augmentation selection on Rotated MNIST · Rotated MNIST domain-classifier selection | domain accuracy: 92.60 $\pm$ 0.98 -> 51.70 $\pm$ 2.21 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MetaReg — MetaReg: Towards Domain Generalization using Meta-Regularization (NeurIPS 2018) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- An Analysis of Causal Effect Estimation using Outcome Invariant Data Augmentation (NeurIPS 2025 · similarity 0.89)
- Do causal predictors generalize better to new domains? (NeurIPS 2024 · similarity 0.88)
- Identifying Representations for Intervention Extrapolation (ICLR 2024 · similarity 0.86)
- PEER Pressure: Model-to-Model Regularization for Single Source Domain Generalization (CVPR 2025 · similarity 0.86)
- ACAMDA: Improving Data Efficiency in Reinforcement Learning through Guided Counterfactual Data Augmentation (AAAI 2024 · similarity 0.85)
- First-Order Manifold Data Augmentation for Regression Learning (ICML 2024 · similarity 0.85)
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.
