ML Security PapersDFedReweighting: A Unified Framework for Objective-Oriented Reweighting in Decentralized Federated Learning
Kaichuang Zhang 1, Wei Yin 2, Jinghao Yang 2, Ping Xu 3
Published on arXiv
2512.12022
Data Poisoning Attack
OWASP ML Top 10 — ML02
Key Finding
DFedReweighting significantly improves fairness and robustness against Byzantine attacks across diverse DFL scenarios while guaranteeing linear convergence
DFedReweighting
Novel technique introduced
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 \textbf{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 \textit{target performance metric} obtained from auxiliary dataset constructed using local data. These weights are then refined using \textit{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.
Key Contributions
- DFedReweighting: a unified reweighting aggregation framework for decentralized FL that simultaneously addresses fairness and Byzantine robustness objectives
- Two-stage weight computation: target performance metric from auxiliary data followed by customized reweighting strategy to produce final aggregation weights
- Theoretical convergence analysis showing linear convergence guarantees under appropriate metric and strategy selection
🛡️ Threat Analysis
Byzantine attacks in federated learning — malicious clients sending arbitrary/corrupted model updates to degrade the global model — are the primary adversarial threat model this framework defends against. The paper proposes a Byzantine-fault-tolerant aggregation protocol (DFedReweighting) as a direct defense, which falls squarely under ML02 (robust aggregation against data/gradient poisoning by malicious FL participants).