Research map / Label-efficient and robust learning / Test-time adaptation
Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering
Yongyi Su, Xun Xu, Kui Jia
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 anchored clustering alone (all test samples update cluster statistics) (ablation) | TTAC N-O baseline · CIFAR10-C | error rate (%): 29.15 -> 14.32 | Improved |
| add pseudo label filtering alone (self-training on filtered labels) (ablation) | no-adaptation baseline · CIFAR10-C | error rate (%): 29.15 -> 15.00 | Improved |
| combine pseudo label filtering with anchored clustering (ablation) | anchored clustering · CIFAR10-C | error rate (%): 14.32 -> 11.33 | Improved |
| add global feature alignment with KL-Divergence on top of anchored clustering and filtering (ablation) | TTAC N-O · CIFAR10-C | error rate (%): 11.33 -> 10.94 | Improved |
| use KL-Divergence instead of L2 distance for global feature alignment (ablation) | TTAC N-M · CIFAR10-C | error rate (%): 11.87 -> 10.8 | Improved |
| replace L2 alignment with KL-Divergence in the TTT++ code (ablation) | TTT++ (Y-M) · CIFAR10 level 5 snow | error rate (%): 9.85 -> 8.43 | Improved |
| add contrastive branch (modify source training) to TTAC (ablation) | TTAC · CIFAR10-C | error rate (%): 10.94 -> 10.69 | Improved |
| maintain a 4096-sample test queue instead of only the current batch | TTAC · CIFAR10-C | error rate (%): 11.91 -> 10.88 | Improved |
| maintain a test queue instead of only the current batch | SHOT · CIFAR10-C | error rate (%): 15.18 -> 13.96 | Improved |
| increase update epochs on the queue from 1 to 4 | TTAC · CIFAR10-C | error rate (%): 10.88 -> 9.96 | Improved |
| increase update epochs from 4 to 6 | TTAC · CIFAR10-C | average error rate (%): 10.94 -> 10.98 | No clear effect |
| update cluster statistics with all pseudo-labeled samples (no filtering) (ablation) | TTAC · CIFAR10-C | average error rate (%): 10.94 -> 13.00 | 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 4 of 4 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 0.89)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.88)
- Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization (AAAI 2024 · similarity 0.88)
- Unraveling Batch Normalization for Realistic Test-Time Adaptation (AAAI 2024 · similarity 0.87)
- Evaluation of Test-Time Adaptation Under Computational Time Constraints (ICML 2024 · similarity 0.87)
- Improved Self-Training for Test-Time Adaptation (CVPR 2024 · 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.
