Achieving Fairness Across Local and Global Models in Federated Learning

TL;DR

This work tackles fairness in federated learning under data heterogeneity and private sensitive attributes by introducing EquiFL, which adds a fairness term weighted by $\mu$ to each client’s local objective and employs a coordination mechanism that prevents bias propagation during aggregation. The method selectively updates model parameters, copying only weight matrices from the global model while keeping the predictor layer and biases local, thereby preserving local personalization and reducing output-level leakage of sensitive information. Empirical results on Adult, COMPAS, Heritage Health, and a healthcare case study show that EquiFL achieves a favorable accuracy-fairness balance and uniform performance distribution across clients, outperforming several baselines in both local and global fairness metrics. The work demonstrates practical impact for sensitive domains such as healthcare, where equitable treatment effect estimation and robust generalization across institutions are critical, while outlining avenues for extending privacy guarantees via differential privacy and deployment in dynamic FL settings.

Abstract

Achieving fairness across diverse clients in Federated Learning (FL) remains a significant challenge due to the heterogeneity of the data and the inaccessibility of sensitive attributes from clients' private datasets. This study addresses this issue by introducing \texttt{EquiFL}, a novel approach designed to enhance both local and global fairness in federated learning environments. \texttt{EquiFL} incorporates a fairness term into the local optimization objective, effectively balancing local performance and fairness. The proposed coordination mechanism also prevents bias from propagating across clients during the collaboration phase. Through extensive experiments across multiple benchmarks, we demonstrate that \texttt{EquiFL} not only strikes a better balance between accuracy and fairness locally at each client but also achieves global fairness. The results also indicate that \texttt{EquiFL} ensures uniform performance distribution among clients, thus contributing to performance fairness. Furthermore, we showcase the benefits of \texttt{EquiFL} in a real-world distributed dataset from a healthcare application, specifically in predicting the effects of treatments on patients across various hospital locations.

Figures (5)

• Figure 1: Comparison of the $\Delta$-DP of the local and the global models for a 3 client setting on the Adult dataset in (a) and (b). Figure (a) shows imposing local fairness doesn't guarantee a fair global model, and figure (b) shows that imposing global fairness doesn't ensure fairness in local models. Figure (c) shows the change in the local fairness of a single client before and after the aggregation step in FL highlighting how the aggregation step might propagate bias.
• Figure 2: An overview of the proposed method showing $N$ clients, a server and the key steps in the procedure.
• Figure 3: Change in performance and fairness with varying $\mu$ - the weight of the fairness term.
• Figure 4: Accuracy and $\Delta$-DP distributions across 100 clients trained using EquiFL on Adult dataset.
• Figure 5: Distribution of patients by race across 3 locations in the ICH dataset.

Theorems & Definitions (4)

• Definition 2.1: Local Fairness
• Definition 2.2: Global Fairness
• Definition 2.3: Demographic Parity
• Definition 2.4: Equal Opportunity