Research map / Label-efficient and robust learning / Recognition and long-tail
MaskTune: Mitigating Spurious Correlations by Forcing to Explore
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 |
|---|---|---|---|
| mask xGradCAM-identified discriminative regions instead of random regions (ablation) | MaskTune · biased MNIST | accuracy: 59.06 -> 98.26 | Helped |
| add MaskTune single-epoch masked finetuning to ERM | ResNet-50 · CelebA | worst-group accuracy: 47.2 -> 78.0 | Helped |
| add MaskTune finetuning to ERM | ResNet-50 · CelebA | average accuracy: 95.6 -> 91.3 | Hurt |
| add MaskTune finetuning to ERM | ResNet-50 · Waterbirds (corrected) | worst-group accuracy: 80.8 -> 86.4 | Helped |
| add MaskTune finetuning to baseline | ResNet-50 · Background Challenge Only-FG | accuracy: 85.6 -> 88.1 | Helped |
| add MaskTune finetuning to baseline | ResNet-50 · Background Challenge Original | accuracy: 96.3 -> 95.6 | Hurt |
| ensemble ERM and MaskTune models for abstention | ResNet32 · CIFAR-10 selective classification 95% coverage | error (%): 8.74 -> 6.54 | Helped |
| ensemble with MaskTune vs best prior at 90% coverage | ResNet32 · CIFAR-10 / Cats vs. Dogs 90% coverage | error (%): 4.67 -> 4.74 (CIFAR-10) | No clear effect |
| iterate masking twice with accumulative masks (two spurious features) (ablation) | MaskTune · biased MNIST two patches | accuracy: 24.68 -> 97.01 | Helped |
| non-accumulative iterative masking instead of accumulative (ablation) | MaskTune · biased MNIST two patches | accuracy: 97.01 -> 30.69 (2 iterations) | Hurt |
| apply MaskTune to ImageNet with no known spurious features | ResNet-50 · ImageNet | top-1 accuracy: 78.862 -> 78.213 | No clear effect |
Closest papers by meaning
- Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples (CVPR 2025 · similarity 0.87)
- Masking meets Supervision: A Strong Learning Alliance (CVPR 2025 · similarity 0.86)
- NeuronTune: Towards Self-Guided Spurious Bias Mitigation (ICML 2025 · similarity 0.86)
- Severing Spurious Correlations with Data Pruning (ICLR 2025 · similarity 0.86)
- Complexity Matters: Feature Learning in the Presence of Spurious Correlations (ICML 2024 · similarity 0.84)
- Understanding and Enhancing Mask-Based Pretraining towards Universal Representations (NeurIPS 2025 · 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.