SPM-FL: A Federated Learning Privacy-Protection Mechanism Based on Local Differential Privacy
Year: 2024
Type: article
Abstract: Federated learning (FL) is a popular distributed machine learning approach that avoids data transfer to third parties by sharing and computing model parameters through a central server, thereby protecting client privacy. However, model weights and other information can still be analyzed and leaked. Currently, many methods in FL apply local differential privacy (LDP) to protect model parameters, but model accuracy significantly declines with smaller privacy budgets and varying numbers of clients. To address this issue, we propose the Symmetric Piecewise Mechanism (SPM) to perturb local model weight parameters before aggregation. First, we design an LDP mechanism that satisfies ε-differential privacy to ensure algorithmic privacy. Second, we introduce a variance-constrained mechanism suitable for FL, maintaining effectiveness with smaller privacy budgets and different client numbers. Finally, we compare the proposed mechanism with other algorithms on three public datasets in terms of model accuracy, variance, and privacy protection. Theoretical proofs demonstrate that our mechanism achieves the minimum variance and highest model accuracy under any privacy budget, showing superior performance.
Source: Electronics
Authors: Zhiyan Chen, Hong Zheng
Institution: Changchun University of Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,728
OpenAlex ID: W4403485134