Toward Efficient and Robust Federated Learning with Heterogeneity
Abstract
Federated learning enables collaborative model training across decentralized data sources, preserving privacy by keeping data local and leveraging edge devices' computational power. Due to its strong privacy-preserving properties, federated learning is widely adopted in various real-world applications. However, scaling federated learning remains challenging due to issues like statistical heterogeneity, client staleness, adversarial attacks, and label noise. This dissertation introduces three novel frameworks—\emph{Polaris}, \emph{AsyncFilter}, and \emph{FedLID}—to address these challenges and enhance the robustness, scalability, and performance of federated learning. Asynchronous federated learning improves training efficiency compared to synchronous approaches, yet its advantages are often limited by statistical heterogeneity and staleness. Hence, we present \emph{Polaris}, a theoretically grounded client selection strategy for asynchronous settings, balancing data quality and client response times through a geometric optimization framework. Besides, the decentralized nature of federated learning, especially in asynchronous settings, increases its vulnerability to poisoning attacks, where malicious clients can compromise the global model with manipulated updates. To mitigate this, we propose \emph{AsyncFilter}, a plug-and-play defense module that detects and filters out poisoned updates in real-time, safeguarding model integrity during training. Additionally, in real-world…
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.