Information-Theoretic Foundation for Federated Learning Privacy
Abstract
Federated learning (FL) offers a promising approach to training machine learning models on decentralized data while preserving data privacy. However, existing privacy-preserving techniques frequently rely on approximations and heuristics, often lacking rigorous mathematical justification. This work presents a novel information-theoretic framework for analyzing and controlling information leakage in federated learning. We leverage concepts from source coding and channel capacity to derive tight bounds on the information revealed by local model updates. These bounds provide provable privacy guarantees, offering a significant advancement over existing methods. The core contribution lies in establishing a solid, mathematically grounded foundation for FL privacy, facilitating the design of more robust and efficient privacy-preserving algorithms. We demonstrate the framework's applicability through theoretical analysis and illustrate its potential for optimizing privacy-utility trade-offs within FL systems. ---
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.