Problem
Research questions and friction points this paper is trying to address.
Decentralized Federated Learning
selection bias
update staleness
partial reception
Age-of-Information
Inverse Probability Weighting and Age-of-Information Aggregation for Decentralized Federated Learning under Partial Reception
🤖 AI Summary
This work addresses selection bias induced by partial model reception and staleness caused by asynchronous node updates in decentralized federated learning over unreliable wireless networks. To tackle these challenges, the authors propose DFL-AA, a novel approach that uniquely integrates inverse probability weighting with Age of Information (AoI)-aware weighting. The former leverages online exponentially weighted moving average (EWMA) channel estimation to correct link-quality-induced biases, while the latter mitigates update staleness through AoI-aware aggregation without requiring global synchronization. Extensive experiments demonstrate that DFL-AA consistently outperforms state-of-the-art methods across varying packet loss rates, network scales, and heterogeneous wireless conditions, achieving both theoretical rigor and practical robustness.
📝 Abstract
Decentralized Federated Learning (DFL) over lossy wireless networks faces two key challenges: selection bias, where updates from poor-quality links are systematically underrepresented due to partial model reception, and update staleness, where asynchronous nodes contribute outdated information. We show that uniform gossip aggregation with local-fill reconstruction introduces persistent link-quality-induced bias, while completeness-based weighting further amplifies this effect. To address these challenges, we propose DFL-AA (Decentralized Federated Learning with Adaptive AoI-weighted Aggregation), which combines Inverse Probability Weighting with online EWMA-based channel estimation to correct selection bias and Age-of-Information-based weighting to mitigate staleness without requiring global synchronization. We theoretically show that DFL-AA removes link-quality distortion in expectation and experimentally demonstrate consistent improvements over state-of-the-art baselines across varying loss rates, network sizes, and heterogeneous wireless conditions.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Inverse Probability Weighting
Age-of-Information
Decentralized Federated Learning
Selection Bias Correction
Asynchronous Aggregation
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
