Research map / Label-efficient and robust learning / Test-time adaptation
Improving Test-Time Adaptation via Shift-Agnostic Weight Regularization and Nearest Source Prototypes
Sungha Choi, Seunghan Yang, Seokeon Choi, Sungrack Yun
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 |
|---|---|---|---|
| add SWR to main entropy loss at lr 1e-3 (ablation) | Main (entropy) TTA · CIFAR-100-C, ResNet-50 | avg error (%): 75.70 -> 37.76 | Improved |
| add NSP auxiliary task to main loss (ablation) | Main (entropy) TTA · CIFAR-100-C, ResNet-50 | avg error (%): 39.44 -> 37.55 | Improved |
| add NSP on top of SWR (ablation) | Main + SWR · CIFAR-100-C, ResNet-50 | avg error (%): 37.76 -> 35.65 | Improved |
| add self-supervised aux loss to NSP entropy loss (ablation) | Main + SWR + NSP · CIFAR-100-C, ResNet-50 | avg error (%): 35.80 -> 35.65 | Improved |
| attach projector (depth 2) before NSP (ablation) | SWR+NSP · CIFAR-100-C, ResNet-50 | avg error (%): 36.34 -> 35.65 | Improved |
| attach projector (depth 2) before NSP (ablation) | SWR+NSP · CIFAR-10-C, ResNet-50 | avg error (%): 12.52 -> 12.95 | Hurt |
| train projector with color, crop and blur transforms instead of none (ablation) | NSP projector · CIFAR-100-C, ResNet-50 | avg error (%): 37.88 -> 35.65 | Improved |
| freeze projector during adaptation instead of fine-tuning (ablation) | SWR+NSP · CIFAR-100-C, ResNet-50 | avg error (%): 35.96 -> 35.65 | Improved |
| update theta* each step instead of freezing it at source parameters (ablation) | SWR · CIFAR-10-C, WRN-28-10 | avg error (%): 18.30 -> 15.70 | Improved |
| build prototypes from projection z instead of classifier weights (ablation) | NSP · CIFAR-100-C, ResNet-50 | avg error (%): 39.89 -> 35.65 | Improved |
| raise SWR weight lambda_r from 10 to 250 (ablation) | SWR+NSP · CIFAR-100-C, ResNet-50 | avg error (%): 53.01 -> 35.65 | Improved |
| raise SWR weight lambda_r from 250 to 1000 (ablation) | SWR+NSP · CIFAR-100-C, ResNet-50 | avg error (%): 35.65 -> 37.43 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 7 of 7 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
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.88)
- Stationary Latent Weight Inference for Unreliable Observations from Online Test-Time Adaptation (ICML 2024 · similarity 0.88)
- Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization (AAAI 2024 · similarity 0.87)
- Test-Time Model Adaptation with Only Forward Passes (ICML 2024 · similarity 0.87)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.87)
- Bayesian Weight Enhancement with Steady-State Adaptation for Test-time Adaptation in Dynamic Environments (ICML 2025 · similarity 0.87)
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.
