Speed Up Federated Learning in Heterogeneous Environments: A Dynamic Tiering Approach

Abstract

Heterogeneity in edge device compute and communication capacities introduces severe straggler delays in federated optimization. We introduce a dynamic tiering methodology grouping clients adaptively by throughput and latency, cutting federated training completion wall-clock time by over 80% without sacrificing model convergence or global classification accuracy.

@article{sajjadi2024tiering,
  title     = {Speed Up Federated Learning in Heterogeneous Environments: A Dynamic Tiering Approach},
  author    = {Sajjadi Mohammadabadi, Seyed Mahmoud and Zawad, Syed and Yan, Feng and Yang, Lei},
  journal   = {IEEE Internet of Things Journal},
  volume    = {12},
  number    = {5},
  pages     = {5026--5035},
  year      = {2025},
  publisher = {IEEE},
  doi       = {10.1109/JIOT.2024.3468511}
}

Author Profile