EUVIMEDEUVIMED entdecken
Crossref

Responsive Test-Time Model Adaptation for Mobile Applications via Runtime-Efficient Sparse Updates

Cheng Fang, Bin Guo, Sicong Liu, Zimu Zhou, Jiaqi Tang, Ke Ma, Shiyan Luo, Geyang Song, Zhiwen Yu

IEEE Transactions on Mobile Computing · 2026-09 · 25 (9) · 14448-14466

DOI: 10.1109/tmc.2026.3680282 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-09. Verfasst von Cheng Fang und 8 weiteren Autor:innen. Die Arbeit nennt 58 Literaturangaben. Gefördert von: National Science Fund for Distinguished Young Scholars, National Natural Science Foundation of China, CityU APRC.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Adapting the on-device deep learning (DL) models to continual and unpredictable domain shifts is crucial for mobile applications like autonomous driving and augmented reality, which require seamless user experiences in ever-changing environments. Test-time adaptation (TTA) offers a promising solution that fine-tunes models with unlabeled real-time data just before making predictions. However, TTA introduces a significant challenge, its forward-backward-reforward pipeline increases latency, compromising responsiveness in time-sensitive mobile scenarios. In this paper, we present AdaShadow, a responsive test-time adaptation framework designed for non-stationary mobile data distributions and resource dynamics. AdaShadow focuses on selectively updating only adaptation-critical layers, minimizing the latency impact. While this idea has been explored in general on-device training, the unsupervised and real-time nature of TTA presents unique challenges in assessing layer importance, predicting latency, and planning layer updates efficiently. AdaShadow addresses these challenges through a backpropagation-free importance assessor for rapid identification of critical layers, a unit-based runtime latency predictor that accounts for resource constraints, and an online layer update scheduler to ensure prompt retraining. Furthermore, a memory I/O-aware computation reuse scheme optimizes the reforward pass to further reduce delays. Our evaluations show that AdaShadow reduces adaptation latency by up to 3.7× (achieving ms-level response time) and improves accuracy by up to 25.4%, all while maintaining low memory and energy costs across CNN, and ViT models, outperforming state-of-the-art methods.

Förderung

  • National Science Fund for Distinguished Young Scholars
  • National Natural Science Foundation of China
  • CityU APRC

Zitieren

APA

Fang, C., Guo, B., Liu, S., Zhou, Z., Tang, J., Ma, K., Luo, S., Song, G., & Yu, Z. (2026). Responsive Test-Time Model Adaptation for Mobile Applications via Runtime-Efficient Sparse Updates. IEEE Transactions on Mobile Computing, 25(9), 14448-14466. https://doi.org/10.1109/tmc.2026.3680282

Vancouver

Fang C, Guo B, Liu S, Zhou Z, Tang J, Ma K, et al. Responsive Test-Time Model Adaptation for Mobile Applications via Runtime-Efficient Sparse Updates. IEEE Transactions on Mobile Computing. 2026;25(9):14448-14466. doi:10.1109/tmc.2026.3680282

BibTeX
@article{Fang2026,
  author = {Fang, Cheng and Guo, Bin and Liu, Sicong and Zhou, Zimu and Tang, Jiaqi and Ma, Ke and Luo, Shiyan and Song, Geyang and Yu, Zhiwen},
  title = {Responsive Test-Time Model Adaptation for Mobile Applications via Runtime-Efficient Sparse Updates},
  journal = {IEEE Transactions on Mobile Computing},
  year = {2026},
  volume = {25},
  number = {9},
  pages = {14448--14466},
  doi = {10.1109/tmc.2026.3680282}
}
Kostenlos starten →