Federated Learning: Privacy-Preserving Distributed Training
Abstract
Federated learning enables collaborative machine learning training across distributed devices without centralizing raw data. This paper examines privacy-preserving federated learning at scale using differential privacy and secure multi-party computation (MPC). We analyze the Federated Averaging (FedAvg) algorithm and its variants including FedProx, FedNova, and Scaffold for non-IID data distributions. Differential privacy mechanisms add calibrated noise to gradients, providing formal privacy guarantees with (ε,δ)-differential privacy where typical deployments use ε=2-8. Secure aggregation through MPC protocols enables encrypted gradient aggregation without revealing individual updates. We evaluate communication efficiency techniques including gradient compression, quantization, and selective parameter updates reducing bandwidth by 10-100×. Performance analysis across mobile keyboard prediction, medical imaging, and financial fraud detection demonstrates competitive accuracy within 1-5% of centralized training while preserving privacy. Implementation challenges include client heterogeneity, stragglers management, and Byzantine robustness. Our findings provide practical guidance for deploying federated learning in healthcare, finance, and edge computing applications requiring strong privacy protection.
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.