DFDG: Adaptive federated learning for dynamic graph-based traffic forecasting

Muhammad Usman

doi:10.1016/j.knosys.2025.114019

DFDG: Adaptive federated learning for dynamic graph-based traffic forecasting

2025-06-30

0

PRE

AI

DOI：10.1016/j.knosys.2025.114019

Original

Original request for help

Save

Abstract

En 中文

• Proposes DFDG, a federated framework integrating dynamic GNNs, KANODE, and privacy-preserving Fourier encoding. • Achieves up to 18% lower error than prior methods on three real-world traffic datasets. • Delivers sub-3-second inference and scales to 21 asynchronous participants. • Ablation studies confirm the importance of temporal modeling and adaptive aggregation. • Code and evaluation tools will be released upon manuscript acceptance.

DFDG: Adaptive federated learning for dynamic graph-based traffic forecasting

Abstract

Journal

Organization