Research map / Label-efficient and robust learning / Test-time adaptation
FRET: Feature Redundancy Elimination for Test Time Adaptation
Linjing You, Jiabao Lu, Xiayuan Huang, Xiangli Nie
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 the attention-part representation loss L_R(A) alone to the source model (ablation) | Source ResNet-18 · OfficeHome (ResNet-18) | accuracy: 0.30% | Improved |
| add the redundancy-part representation loss L_R(R) alone (ablation) | Source ResNet-18 · OfficeHome (ResNet-18) | accuracy: 1.11% | Improved |
| add the attention-part prediction loss L_P(A) alone (ablation) | Source ResNet-18 · OfficeHome (ResNet-18) | accuracy: 1.15% | Improved |
| add the redundancy-part prediction loss L_P(R) alone (ablation) | Source ResNet-18 · OfficeHome (ResNet-18) | accuracy: 0.06% | No clear effect |
| add attention-part L_R (A) to the other components (ablation) | G-FRET · OfficeHome (ResNet-18) | accuracy: 63.74 -> 63.81 | Improved |
| add redundancy-part L_R (R) to the other components (ablation) | G-FRET · OfficeHome (ResNet-18) | accuracy: 63.28 -> 63.81 | Improved |
| add attention-part L_P (A) to the other components (ablation) | G-FRET · OfficeHome (ResNet-18) | accuracy: 63.24 -> 63.81 | Improved |
| add redundancy-part L_P (R) to the other components (ablation) | G-FRET · OfficeHome (ResNet-18) | accuracy: 63.77 -> 63.81 | Improved |
| combine attention and redundancy components rather than using either alone (ablation) | G-FRET · OfficeHome (ResNet-18) | accuracy: 1.28% and 1.17% -> 1.80% | Improved |
| replace S-FRET by G-FRET (add GCN contrastive learning) under long-tail label shift | S-FRET · CIFAR-100-C-LT (imbalance factor 100) | accuracy: 46.62 -> 50.27 | Improved |
| increase imbalance factor from 1 to 100 | S-FRET · CIFAR-100-C-LT | accuracy: 48.53 -> 46.62 | Hurt |
| apply G-FRET on top of source model | ResNet-18 · PACS | accuracy: 6.67% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TIPI — TIPI: Test Time Adaptation With Transformation Invariance (CVPR 2023) · better on 3 of 3 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
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.88)
- Test-time Correlation Alignment (ICML 2025 · similarity 0.88)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.87)
- L-TTA: Lightweight Test-Time Adaptation Using a Versatile Stem Layer (NeurIPS 2024 · similarity 0.87)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.86)
- Feature-Based Instance Neighbor Discovery: Advanced Stable Test-Time Adaptation in Dynamic World (NeurIPS 2025 · 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.
