Research map / Label-efficient and robust learning / Recognition and long-tail
Roadblocks for Temporarily Disabling Shortcuts and Learning New Knowledge
Hongjing Niu, Hanting Li, Feng Zhao, Bin Li
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 Roadblocks instead of LDD | debiasing model · CMNIST 0.5% minority | accuracy: 65.22 -> 66.64 | Improved |
| use Roadblocks instead of LDD | debiasing model · CMNIST 1.0% minority | accuracy: 81.73 -> 82.04 | Improved |
| use Roadblocks instead of LDD | debiasing model · CMNIST 2.0% minority | accuracy: 84.79 -> 84.93 | No clear effect |
| use Roadblocks instead of LDD | debiasing model · CMNIST 5.0% minority | accuracy: 89.66 -> 88.65 | Hurt |
| use Roadblocks instead of prior-based ReBias | debiasing model · CMNIST (0.5% to 5.0% minority) | accuracy: 70.47 -> 66.64 at 0.5%; 96.96 -> 88.65 at 5.0% | Hurt |
| use Roadblocks instead of Vanilla training | CNN · CMNIST 0.5% minority | accuracy: 35.19 -> 66.64 | Improved |
| use Roadblocks instead of Vanilla training | CNN · CMNIST 5.0% minority | accuracy: 82.17 -> 88.65 | Improved |
| use Roadblocks instead of LDD | debiasing model · CelebA unbiased test | accuracy: 64.82 -> 79.73 | Improved |
| use Roadblocks instead of LDD | debiasing model · CelebA bias-conflicting test | accuracy: 47.63 -> 64.06 | Improved |
| use Roadblocks instead of LDD | debiasing model · BAR bias-conflicting test | accuracy: 63.50 -> 69.51 | Improved |
| use Roadblocks instead of EnD | debiasing model · CelebA unbiased test | accuracy: 75.93 -> 79.73 | Improved |
| use Roadblocks instead of Vanilla training | CNN · CelebA bias-conflicting test | accuracy: 33.75 -> 64.06 | Improved |
Closest papers by meaning
- On the Foundations of Shortcut Learning (ICLR 2024 · similarity 0.87)
- Improved Training Technique for Shortcut Models (NeurIPS 2025 · similarity 0.83)
- Navigate Beyond Shortcuts: Debiased Learning Through the Lens of Neural Collapse (CVPR 2024 · similarity 0.82)
- Do ImageNet-trained Models Learn Shortcuts? The Impact of Frequency Shortcuts on Generalization (CVPR 2025 · similarity 0.81)
- SCISSOR: Mitigating Semantic Bias through Cluster-Aware Siamese Networks for Robust Classification (ICML 2025 · similarity 0.81)
- Adaptive Shortcut Debiasing for Online Continual Learning (AAAI 2024 · similarity 0.80)
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.
