FedLF: Layer-Wise Fair Federated Learning

Authors

  • Zibin Pan The Chinese University of Hong Kong, Shenzhen The Shenzhen Institute of Artificial Intelligence and Robotics for Society
  • Chi Li The Chinese University of Hong Kong, Shenzhen Shenzhen Research Institute of Big Data
  • Fangchen Yu The Chinese University of Hong Kong, Shenzhen
  • Shuyi Wang The Chinese University of Hong Kong, Shenzhen The Shenzhen Institute of Artificial Intelligence and Robotics for Society
  • Haijin Wang The Chinese University of Hong Kong, Shenzhen The Shenzhen Institute of Artificial Intelligence and Robotics for Society
  • Xiaoying Tang The Chinese University of Hong Kong, Shenzhen The Shenzhen Institute of Artificial Intelligence and Robotics for Society The Guangdong Provincial Key Laboratory of Future Networks of Intelligence
  • Junhua Zhao The Chinese University of Hong Kong, Shenzhen The Shenzhen Institute of Artificial Intelligence and Robotics for Society

DOI:

https://doi.org/10.1609/aaai.v38i13.29368

Keywords:

ML: Distributed Machine Learning & Federated Learning, ML: Applications, PEAI: Bias, Fairness & Equity

Abstract

Fairness has become an important concern in Federated Learning (FL). An unfair model that performs well for some clients while performing poorly for others can reduce the willingness of clients to participate. In this work, we identify a direct cause of unfairness in FL - the use of an unfair direction to update the global model, which favors some clients while conflicting with other clients’ gradients at the model and layer levels. To address these issues, we propose a layer-wise fair Federated Learning algorithm (FedLF). Firstly, we formulate a multi-objective optimization problem with an effective fair-driven objective for FL. A layer-wise fair direction is then calculated to mitigate the model and layer-level gradient conflicts and reduce the improvement bias. We further provide the theoretical analysis on how FedLF can improve fairness and guarantee convergence. Extensive experiments on different learning tasks and models demonstrate that FedLF outperforms the SOTA FL algorithms in terms of accuracy and fairness. The source code is available at https://github.com/zibinpan/FedLF.

Published

2024-03-24

How to Cite

Pan, Z., Li, C., Yu, F., Wang, S., Wang, H., Tang, X., & Zhao, J. (2024). FedLF: Layer-Wise Fair Federated Learning. Proceedings of the AAAI Conference on Artificial Intelligence, 38(13), 14527-14535. https://doi.org/10.1609/aaai.v38i13.29368

Issue

Section

AAAI Technical Track on Machine Learning IV