Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedFair^3: Unlocking Threefold Fairness in Federated Learning

Simin Javaherian, Sanjeev Panta, Shelby Williams, Md Sirajul Islam, Li Chen

TL;DR

The paper tackles fairness in federated learning amid heterogeneous clients by proposing FedFair3, a threefold fairness framework that optimizes fair client participation, equitable round-robin participation, and fair accuracy distribution. It achieves this through a resource-aware, probabilistic client selection that incorporates client features (data size, energy, round duration, power) and a global resource budget, building on q-fairness ideas. Empirically, FedFair3 attains substantially lower accuracy variance across clients on both IID and non-IID data and reduces wall-clock training time without sacrificing global accuracy, outperforming baselines such as AFL, FedProx, Oort, and FedAvg. The method enhances participation willingness and efficiency in large-scale FL deployments, with practical impact for real-world edge networks and heterogeneous device ecosystems.

Abstract

Federated Learning (FL) is an emerging paradigm in machine learning without exposing clients' raw data. In practical scenarios with numerous clients, encouraging fair and efficient client participation in federated learning is of utmost importance, which is also challenging given the heterogeneity in data distribution and device properties. Existing works have proposed different client-selection methods that consider fairness; however, they fail to select clients with high utilities while simultaneously achieving fair accuracy levels. In this paper, we propose a fair client-selection approach that unlocks threefold fairness in federated learning. In addition to having a fair client-selection strategy, we enforce an equitable number of rounds for client participation and ensure a fair accuracy distribution over the clients. The experimental results demonstrate that FedFair^3, in comparison to the state-of-the-art baselines, achieves 18.15% less accuracy variance on the IID data and 54.78% on the non-IID data, without decreasing the global accuracy. Furthermore, it shows 24.36% less wall-clock training time on average.

FedFair^3: Unlocking Threefold Fairness in Federated Learning

TL;DR

The paper tackles fairness in federated learning amid heterogeneous clients by proposing FedFair3, a threefold fairness framework that optimizes fair client participation, equitable round-robin participation, and fair accuracy distribution. It achieves this through a resource-aware, probabilistic client selection that incorporates client features (data size, energy, round duration, power) and a global resource budget, building on q-fairness ideas. Empirically, FedFair3 attains substantially lower accuracy variance across clients on both IID and non-IID data and reduces wall-clock training time without sacrificing global accuracy, outperforming baselines such as AFL, FedProx, Oort, and FedAvg. The method enhances participation willingness and efficiency in large-scale FL deployments, with practical impact for real-world edge networks and heterogeneous device ecosystems.

Abstract

Federated Learning (FL) is an emerging paradigm in machine learning without exposing clients' raw data. In practical scenarios with numerous clients, encouraging fair and efficient client participation in federated learning is of utmost importance, which is also challenging given the heterogeneity in data distribution and device properties. Existing works have proposed different client-selection methods that consider fairness; however, they fail to select clients with high utilities while simultaneously achieving fair accuracy levels. In this paper, we propose a fair client-selection approach that unlocks threefold fairness in federated learning. In addition to having a fair client-selection strategy, we enforce an equitable number of rounds for client participation and ensure a fair accuracy distribution over the clients. The experimental results demonstrate that FedFair^3, in comparison to the state-of-the-art baselines, achieves 18.15% less accuracy variance on the IID data and 54.78% on the non-IID data, without decreasing the global accuracy. Furthermore, it shows 24.36% less wall-clock training time on average.
Paper Structure (13 sections, 9 equations, 2 figures, 3 tables, 1 algorithm)

This paper contains 13 sections, 9 equations, 2 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background and Related works
  3. Preliminaries of Federated Learning
  4. Fairness and Efficiency Related Works
  5. Motivation
  6. Methodology
  7. FedFair³ Algorithm
  8. Fairness Qualification
  9. Experimental Result
  10. Experimental setup
  11. Results and Analysis
  12. Limitation and Future Work
  13. Conclusion

Figures (2)

  • Figure 1: A toy example of comparison of local accuracy (Oort vs. FedFair3) considering 6 clients in each round.
  • Figure 2: The variance of accuracy of FedFair3 versus the FedAvg, AFL, FedProx and Oort.