Poster: Adaptive Gradient Clipping with Personalized Differential Privacy for Heterogeneous Federated Learning
Abstract
We present GC-DP, a novel federated learning framework that enables personalized differential privacy by adaptively adjusting gradient clipping thresholds. Unlike traditional DP methods that apply a fixed clipping bound, GC-DP uses a proxy dataset to learn client-specific mappings from privacy budgets (ε) to optimal clipping thresholds (C^*), while also allowing each client to adjust its local clipping bound in real time based on the l2 norm of its local gradient. This dual adaptivity significantly improves the balance between privacy protection and model utility in heterogeneous FL settings.
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.