Research map / Label-efficient and robust learning / Test-time adaptation
Test-Time Training with Self-Supervision for Generalization under Distribution Shifts
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 |
|---|---|---|---|
| Adapt shared features on each test sample using rotation prediction | ResNet-26 · CIFAR-10-C level 5, Gaussian noise | error (%): 49.4 -> 45.6 | Helped |
| Keep the self-supervised updates across samples in an online stream | TTT · CIFAR-10-C level 5, Gaussian noise | error (%): 45.6 -> 25.8 | Helped |
| Keep the self-supervised updates across samples in an online stream | TTT · CIFAR-10-C level 5, original | error (%): 7.9 -> 8.2 | Hurt |
| Keep the self-supervised updates across samples in an online stream | TTT · ImageNet-C level 5, Gaussian noise | accuracy (%): 3.1 -> 26.3 | Helped |
| Keep the self-supervised updates across samples in an online stream | TTT · ImageNet-C level 5, original | accuracy (%): 69.0 -> 68.8 | Hurt |
| Adapt shared features on each test sample using rotation prediction | ResNet-26 · CIFAR-10.1 | error (%): 16.7 -> 15.9 | Helped |
| Keep the self-supervised updates across samples in an online stream | TTT · VID-Robust | accuracy (%): 63.8 -> 64.3 | Helped |
| Use rotation prediction for test-time adaptation | CIFAR ResNet · Airplane class in CIFAR classes from VID-Robust | accuracy (%): 67.9 -> 70.2 | No clear effect |
| Use rotation prediction for test-time adaptation | CIFAR ResNet · CIFAR classes in VID-Robust, average | accuracy (%): 42.4 -> 45.2 | Helped |
| Use a task-specific self-supervised update rather than a fixed jointly trained model | CIFAR-10-C · Corruptions across levels 1-5 | error (%) | Helped |
| Use online updates rather than a fixed jointly trained model | CIFAR-10-C · Corruptions across levels 1-5 | error (%): >10%; 0.2% | Mixed |
| Freeze Batch Normalization layers during test-time training (ablation) | TTT with Batch Normalization · CIFAR-10-C | error (%) | Hurt |
Closest papers by meaning
- Protected Test-Time Adaptation via Online Entropy Matching: A Betting Approach (NeurIPS 2024 · similarity 0.88)
- Improved Self-Training for Test-Time Adaptation (CVPR 2024 · similarity 0.87)
- Improving self-training under distribution shifts via anchored confidence with theoretical guarantees (NeurIPS 2024 · similarity 0.86)
- Self-Bootstrapping for Versatile Test-Time Adaptation (ICML 2025 · similarity 0.86)
- Learning Neural Networks with Distribution Shift: Efficiently Certifiable Guarantees (ICLR 2025 · similarity 0.85)
- Ranked Entropy Minimization for Continual Test-Time Adaptation (ICML 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.