Research map / Label-efficient and robust learning / Test-time adaptation
MemBN: Robust Test-Time Adaptation via Batch Norm with Statistics Memory
Juwon Kang, Nayeong Kim, Jungseul Ok, Suha Kwak
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 statistics memory of size 128 with reset (vs TBN) (ablation) | TBN · CIFAR10-C continual | avg error rate (%): 33.09 -> 15.07 | Improved |
| add memory reset to a 128-size memory (ablation) | MemBN memory · CIFAR10-C continual | avg error rate (%): 17.40 -> 15.07 | Improved |
| add adaptive weighting between source and memory statistics (ablation) | MemBN · CIFAR10-C continual | avg error rate (%): 15.07 -> 12.18 | Improved |
| add adaptive weighting between source and memory statistics (ablation) | MemBN · CIFAR100-C continual | avg error rate (%): 40.17 -> 36.80 | Improved |
| use memory of 128 without reset instead of EMA memory of size 1 (ablation) | MemBN · CIFAR10-C continual, batch size 200 | error rate (%): 14.63 -> 21.05 | Hurt |
| replace in-batch statistics with MemBN statistics | TENT · ImageNet-C batch size 1 | error rate (%): 99.90 -> 67.88 | Improved |
| add MemBN to EATA | EATA · ImageNet-C | avg error rate (%): 77.15 -> 58.42 | Improved |
| add MemBN to EATA | EATA · ImageNet-C batch size 1 | error rate (%): 99.86 -> 73.72 | No clear effect |
| use MemBN instead of TTN without any post-training phase | TTN · CIFAR10-C single domain | avg error rate (%): 13.89 -> 12.48 | Improved |
| use MemBN at large batch size 200 vs TTN | TTN · CIFAR10-C single domain, batch size 200 | error rate (%): 11.67 -> 12.68 | Hurt |
| add MemBN to TENT in segmentation | TENT · Cityscapes to BDD-100K, batch size 1 | mIoU: 40.78 -> 50.94 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- TeSLA — TeSLA: Test-Time Self-Learning With Automatic Adversarial Augmentation (CVPR 2023) · better on 1 of 1 shared result
- NOTE — NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation (NeurIPS 2022) · better on 1 of 1 shared result
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 1 of 1 shared result
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
- Unraveling Batch Normalization for Realistic Test-Time Adaptation (AAAI 2024 · similarity 0.92)
- Buffer layers for Test-Time Adaptation (NeurIPS 2025 · similarity 0.91)
- MedBN: Robust Test-Time Adaptation against Malicious Test Samples (CVPR 2024 · similarity 0.88)
- SNAP: Low-Latency Test-Time Adaptation with Sparse Updates (NeurIPS 2025 · similarity 0.88)
- Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization (AAAI 2024 · similarity 0.88)
- A Layer Selection Approach to Test Time Adaptation (AAAI 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.
