LoGoFair: Post-Processing for Local and Global Fairness in Federated Learning

2025-03-21

AI Summary

Existing fair federated learning (FL) methods struggle to simultaneously ensure local (client-level) and global (system-level) fairness—particularly for sensitive attributes such as gender—under statistical heterogeneity and model-agnostic settings, suffering from theoretical limitations in fairness guarantees. Method: We propose the first Bayesian-optimal post-processing framework that unifies local and global fairness modeling; further, we design a model-agnostic federated co-optimization mechanism integrating probabilistic calibration, constrained optimization, and distributed gradient negotiation to achieve joint risk minimization. Results: Extensive experiments on three real-world datasets demonstrate that our method maintains high accuracy while improving local and global fairness metrics by 23.6% and 19.4% on average, respectively—significantly outperforming current state-of-the-art fair FL approaches.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Search and Optimization: Local Search
• Computer Vision: Bias, Fairness & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Federated learning (FL) has garnered considerable interest for its capability to learn from decentralized data sources. Given the increasing application of FL in decision-making scenarios, addressing fairness issues across different sensitive groups (e.g., female, male) in FL is crucial. Current research often focuses on facilitating fairness at each client's data (local fairness) or within the entire dataset across all clients (global fairness). However, existing approaches that focus exclusively on either local or global fairness fail to address two key challenges: ( extbf{CH1}) Under statistical heterogeneity, global fairness does not imply local fairness, and vice versa. ( extbf{CH2}) Achieving fairness under model-agnostic setting. To tackle the aforementioned challenges, this paper proposes a novel post-processing framework for achieving both Local and Global Fairness in the FL context, namely LoGoFair. To address CH1, LoGoFair endeavors to seek the Bayes optimal classifier under local and global fairness constraints, which strikes the optimal accuracy-fairness balance in the probabilistic sense. To address CH2, LoGoFair employs a model-agnostic federated post-processing procedure that enables clients to collaboratively optimize global fairness while ensuring local fairness, thereby achieving the optimal fair classifier within FL. Experimental results on three real-world datasets further illustrate the effectiveness of the proposed LoGoFair framework.

Problem

• Address local and global fairness in federated learning
• Balance accuracy and fairness under statistical heterogeneity
• Achieve fairness in model-agnostic federated settings

Innovation

• Bayes optimal classifier under fairness constraints
• Model-agnostic federated post-processing procedure
• Balances local and global fairness in FL