Research map / Continual learning / Domain and prompt-based
Efficient Test-Time Model Adaptation without Forgetting
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 |
|---|---|---|---|
| Use entropy filtering to exclude high-entropy samples (ablation) | Tent-style entropy adaptation · ImageNet-C Gaussian noise, level 5 | error (%): 71.6 -> 65.7 | Helped |
| Use entropy filtering to exclude high-entropy samples (ablation) | Tent-style entropy adaptation · ImageNet-C Gaussian noise, level 5 | #Backwards: 50,000 -> 26,694 | Helped |
| Filter redundant samples using prediction diversity (ablation) | Entropy-filtered adaptation · ImageNet-C Gaussian noise, level 5 | error (%): 65.7 -> 65.0 | Helped |
| Filter redundant samples using prediction diversity (ablation) | Entropy-filtered adaptation · ImageNet-C Gaussian noise, level 5 | #Backwards: 26,694 -> 19,121 | Helped |
| Add Fisher weight regularization (ablation) | ETA · ImageNet-C Gaussian noise, level 5 | error (%): 65.0 -> 65.0 | No clear effect |
| Add Fisher weight regularization (ablation) | ETA · ImageNet-R | error (%): 54.5 -> 54.8 | Hurt |
| Add Fisher weight regularization (ablation) | ETA · CIFAR-10-C | average error (%): 19.4 -> 19.7 | Hurt |
| Select only low-entropy, non-redundant samples for adaptation | Tent · ImageNet-C Gaussian noise, level 5 | error (%): 71.6 -> 65.0 | Helped |
| Use EATA rather than Tent for mixed-shift adaptation | ResNet-50 · ImageNet-C mixed corruptions, level 5 | corruption accuracy (%): 2.3 -> 26.6 | Helped |
| Use EATA rather than Tent for lifelong adaptation | ResNet-50 · ImageNet-C lifelong adaptation | clean accuracy on ID data | Helped |
| Use a smaller entropy threshold E0 (ablation) | EATA · ImageNet-C | adaptation performance | Hurt |
| Use a larger entropy threshold E0 (ablation) | EATA · ImageNet-C | adaptation performance | Hurt |
Closest papers by meaning
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.91)
- Test-Time Model Adaptation with Only Forward Passes (ICML 2024 · similarity 0.90)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.90)
- ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains (NeurIPS 2025 · similarity 0.90)
- Ranked Entropy Minimization for Continual Test-Time Adaptation (ICML 2025 · similarity 0.89)
- Unified Entropy Optimization for Open-Set Test-Time Adaptation (CVPR 2024 · similarity 0.89)
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.