Research map / Label-efficient and robust learning / Test-time adaptation
CS-TTA: Preserving Concept Sensitivity in Test-Time Adaptation
Junah Jung, YeonGyu Han, Chang Min Park, Dongheon Lee
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 CS-TTA regularizer | Tent · Waterbirds | worst-group accuracy: 56.2 -> 65.8 | Improved |
| add CS-TTA regularizer | DeYO · Waterbirds | worst-group accuracy: 63.8 -> 68.5 | Improved |
| add CS-TTA regularizer | five TTA baselines · CheXpert->MIMIC-CXR | AUC: +1.4-1.8 points | Improved |
| add CS-TTA regularizer | five TTA baselines · CIFAR-10-C and ImageNet-C | accuracy: +0.6-1.4 points | Improved |
| adapt with Tent, EATA or CoTTA (no concept regularization) | source ResNet-50 · Waterbirds | worst-group accuracy: 58.1 -> 55.1 (CoTTA) | Hurt |
| replace concept-selective loss with blind L2 on all activations (ablation) | Tent+CS-TTA · Waterbirds | worst-group accuracy: 65.8 -> 57.5 | Improved |
| select random concepts instead of object/background concepts (ablation) | Tent+CS-TTA · Waterbirds | worst-group accuracy: 65.8 -> 58.2 | Improved |
| raise lambda from 0.1 to 10.0 (ablation) | Tent+CS-TTA · Waterbirds | worst-group accuracy: 65.8 -> 63.5 | Hurt |
| lower lambda from 0.1 to 0.01 (ablation) | Tent+CS-TTA · Waterbirds | worst-group accuracy: 65.8 -> 62.3 | Hurt |
| regularize at early layer2 instead of layer4 (ablation) | Tent+CS-TTA · Waterbirds | worst-group accuracy: 65.8 -> 62.8 | Hurt |
| use CLIP pseudo-label concepts instead of annotations (ablation) | Tent+CS-TTA · Waterbirds | worst-group accuracy: 65.8 -> 64.5 | Hurt |
| use second-order exact sensitivity instead of first-order surrogate (ablation) | Tent+CS-TTA · Waterbirds | worst-group accuracy: 65.8 -> 66.0 | No clear effect |
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 3 of 3 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
- SCAP: Transductive Test-Time Adaptation via Supportive Clique-based Attribute Prompting (CVPR 2025 · similarity 0.87)
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.87)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.86)
- SURGEON: Memory-Adaptive Fully Test-Time Adaptation via Dynamic Activation Sparsity (CVPR 2025 · similarity 0.86)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.86)
- TEA: Test-time Energy Adaptation (CVPR 2024 · similarity 0.86)
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.
