EUVIMEDEUVIMED entdecken
Crossref

Decentralized Edge Learning Under Weak Smoothness: A Normalized Random Walk Gradient Approach

Tao Sun

IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 22619-22631

DOI: 10.1109/tmc.2026.3724227 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Tao Sun. Die Arbeit nennt 55 Literaturangaben. Gefördert von: National Natural Science Foundation of China, NUDT Foundational Research Funding.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Random walk algorithms are particularly well-suited for decentralized distributed edge computing due to their inherent decentralization and low computational requirements for edge devices. However, existing random walk training algorithms typically rely on uniform gradient Lipschitz smoothness assumptions, which are often difficult to satisfy in practical scenarios. To address this challenge, we propose a novel random walk algorithm based on gradient normalization (N-RWGD) for weak smooth problems. Our method tackles the issue of delayed gradient differences without assuming uniform Lipschitz gradients and under Markovian sampling conditions, making it highly robust in edge computing environments where gradients may be very large. Through extensive theoretical analysis and experimental validation, we demonstrate that the proposed algorithm outperforms existing random walk methods in terms of stability and performance. Our work offers new insights and solutions for decentralized edge computing challenges.

Förderung

  • National Natural Science Foundation of China
  • NUDT Foundational Research Funding

Zitieren

APA

Sun, T. (2026). Decentralized Edge Learning Under Weak Smoothness: A Normalized Random Walk Gradient Approach. IEEE Transactions on Mobile Computing, 25(12), 22619-22631. https://doi.org/10.1109/tmc.2026.3724227

Vancouver

Sun T. Decentralized Edge Learning Under Weak Smoothness: A Normalized Random Walk Gradient Approach. IEEE Transactions on Mobile Computing. 2026;25(12):22619-22631. doi:10.1109/tmc.2026.3724227

BibTeX
@article{Sun2026,
  author = {Sun, Tao},
  title = {Decentralized Edge Learning Under Weak Smoothness: A Normalized Random Walk Gradient Approach},
  journal = {IEEE Transactions on Mobile Computing},
  year = {2026},
  volume = {25},
  number = {12},
  pages = {22619--22631},
  doi = {10.1109/tmc.2026.3724227}
}
Kostenlos starten →