On-Device Realistic Test-Time Adaptation via Bias-Resistant Statistical Alignment

Bai, Haojie; Chen, Aiguo; Dai, Ruiting; Rong, Yijia; Wang, Zirui; Liu, Jiaxin; Li, Kexin; Dustdar, Schahram

doi:10.24963/ijcai.2026/101

On-Device Realistic Test-Time Adaptation via Bias-Resistant Statistical Alignment

Haojie Bai, Aiguo Chen, Ruiting Dai, Yijia Rong, Zirui Wang, Jiaxin Liu, Kexin Li, Schahram Dustdar

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 899-907.

https://doi.org/10.24963/ijcai.2026/101

Keywords

Computer Vision: Applications and Systems
Computer Vision: Machine learning for vision
Computer Vision: Efficiency and Optimization

Abstract

Test-Time Adaptation (TTA) aims to adapt pretrained models to unseen test data, which is crucial for resource-constrained edge devices that must handle distribution shifts on the fly without human supervision. However, conventional TTA methods often fail in realistic scenarios characterized by continuous shifts and severe class imbalances while incurring prohibitive memory overheads. To enable continuous adaptation to test data on edge devices under a realistic TTA setting, we propose a Bias-Resistant Online Statistical Alignment (BOSA) method. BOSA facilitates unbiased adaptation via a discrepancy-aware statistical alignment mechanism integrated with a class-balanced memory bank. To ensure memory efficiency, we further design a saliency-guided activation sparsification and gradient reconstruction scheme, which drastically reduces memory overhead without sacrificing gradient integrity. Extensive evaluations demonstrate that BOSA achieves superior accuracy with a compact memory consumption compared to state-of-the-art methods under realistic TTA settings.

Haojie Bai, Aiguo Chen, Ruiting Dai, Yijia Rong, Zirui Wang, Jiaxin Liu, Kexin Li, and Schahram Dustdar. 2026. On-Device Realistic Test-Time Adaptation via Bias-Resistant Statistical Alignment. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 899–907. https://doi.org/10.24963/ijcai.2026/101

@inproceedings{ijcai2026p101,
  title     = {On-Device Realistic Test-Time Adaptation via Bias-Resistant Statistical Alignment},
  author    = {Bai, Haojie and Chen, Aiguo and Dai, Ruiting and Rong, Yijia and Wang, Zirui and Liu, Jiaxin and Li, Kexin and Dustdar, Schahram},
  booktitle = {Proceedings of the Thirty-Fifth International Joint Conference on
               Artificial Intelligence, {IJCAI-26}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Diego Calvanese},
  pages     = {899--907},
  year      = {2026},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2026/101},
  url       = {https://doi.org/10.24963/ijcai.2026/101},
}

TY  - CPAPER
TI  - On-Device Realistic Test-Time Adaptation via Bias-Resistant Statistical Alignment
AU  - Bai, Haojie
AU  - Chen, Aiguo
AU  - Dai, Ruiting
AU  - Rong, Yijia
AU  - Wang, Zirui
AU  - Liu, Jiaxin
AU  - Li, Kexin
AU  - Dustdar, Schahram
T2  - Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2026
SP  - 899
EP  - 907
DO  - 10.24963/ijcai.2026/101
UR  - https://doi.org/10.24963/ijcai.2026/101
ER  -