DFedReweighting: A Unified Framework for Objective-Oriented Reweighting in Decentralized Federated Learning

2025-12-12

AI Summary

Addressing the dual challenges of insufficient fairness and weak Byzantine-robustness in decentralized federated learning (DFL), this paper proposes a goal-oriented dynamic reweighting aggregation framework. The framework employs a lightweight auxiliary dataset to evaluate target performance—such as group fairness or Byzantine-robust accuracy—at the end of each communication round, and dynamically assigns node aggregation weights via gradient sensitivity and distribution-aware mechanisms. It is the first to achieve a unified, plug-and-play, multi-objective-compatible reweighting scheme, with theoretical guarantees of linear convergence. Extensive experiments on multiple benchmark datasets demonstrate that the method reduces the Fairness Gap by 23.6% and maintains a test accuracy of 89.4% under Byzantine attacks—significantly outperforming the baseline (61.2%). These results establish new state-of-the-art performance in balancing fairness and robustness in DFL.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Natural Language Processing: Safety and Robustness
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Decentralized federated learning (DFL) has recently emerged as a promising paradigm that enables multiple clients to collaboratively train machine learning model through iterative rounds of local training, communication, and aggregation without relying on a central server which introduces potential vulnerabilities in conventional Federated Learning. Nevertheless, DFL systems continue to face a range of challenges, including fairness, robustness, etc. To address these challenges, we propose extbf{DFedReweighting}, a unified aggregation framework designed to achieve diverse objectives in DFL systems via a objective-oriented reweighting aggregation at the final step of each learning round. Specifically, the framework first computes preliminary weights based on extit{target performance metric} obtained from auxiliary dataset constructed using local data. These weights are then refined using extit{customized reweighting strategy}, resulting in the final aggregation weights. Our results from the theoretical analysis demonstrate that the appropriate combination of the target performance metric and the customized reweighting strategy ensures linear convergence. Experimental results consistently show that our proposed framework significantly improves fairness and robustness against Byzantine attacks in diverse scenarios. Provided that appropriate target performance metrics and customized reweighting strategy are selected, our framework can achieve a wide range of desired learning objectives.

Problem

• Improves fairness and robustness in decentralized federated learning
• Addresses challenges like Byzantine attacks without a central server
• Enables diverse learning objectives via objective-oriented reweighting aggregation

Innovation

• Decentralized federated learning with objective-oriented reweighting aggregation
• Uses target performance metrics and customized reweighting strategies
• Ensures linear convergence, improves fairness and robustness