Multi-Metric Client Activation Method for Fast and Accurate Federated Learning

Jihyun Lim, Tuo Zhang, Sunwoo Lee

ACM Transactions on Intelligent Systems and Technology (TIST), 2026

This work proposes a multi-metric bias-based client activation method for Federated Learning, prioritizing clients whose local loss and gradient norm best represent the global dataset, that accelerates convergence and improves generalization under strongly non-IID settings, outperforming both uniform random sampling and state-of-the-art biased client activation methods on standard benchmarks.

BibTeX

@article{lim2026multi,
  title={Multi-Metric Client Activation Method for Fast and Accurate Federated Learning},
  author={Lim, Jihyun and Zhang, Tuo and Lee, Sunwoo},
  journal={ACM Transactions on Intelligent Systems and Technology},
  volume={17},
  number={2},
  pages={1--27},
  year={2026},
  publisher={ACM New York, NY}
}