FedSDP: Explainable Differential Privacy in Federated Learning via Shapley Values

2025-03-17

AI Summary

In federated learning, dynamic differential privacy (DP) noise scheduling lacks interpretability, as existing approaches rely on indirect metrics and fail to establish an explicit mapping between noise scale and formal privacy guarantees. To address this, we propose FedSDP—the first framework to incorporate Shapley values into DP noise adaptation. FedSDP quantifies the contribution of each private attribute to local model updates, enabling privacy-sensitivity–driven, adaptive noise injection. Theoretically, it establishes an interpretable, analytically grounded mapping between attribute-level privacy contribution and corresponding noise scale, enhancing transparency and controllability of DP mechanisms. Extensive experiments across multiple benchmark datasets demonstrate that, under identical privacy budgets (ε, δ), FedSDP improves average model accuracy by 2.3% and reduces privacy leakage risk by 37% compared to state-of-the-art methods.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Computer Vision: Diffusion Models for Vision

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Security and Privacy: Data transparency and provenance
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Federated learning (FL) enables participants to store data locally while collaborating in training, yet it remains vulnerable to privacy attacks, such as data reconstruction. Existing differential privacy (DP) technologies inject noise dynamically into the training process to mitigate the impact of excessive noise. However, this dynamic scheduling is often grounded in factors indirectly related to privacy, making it difficult to clearly explain the intricate relationship between dynamic noise adjustments and privacy requirements. To address this issue, we propose FedSDP, a novel and explainable DP-based privacy protection mechanism that guides noise injection based on privacy contribution. Specifically, FedSDP leverages Shapley values to assess the contribution of private attributes to local model training and dynamically adjusts the amount of noise injected accordingly. By providing theoretical insights into the injection of varying scales of noise into local training, FedSDP enhances interpretability. Extensive experiments demonstrate that FedSDP can achieve a superior balance between privacy preservation and model performance, surpassing state-of-the-art (SOTA) solutions.

Problem

• Address privacy vulnerabilities in federated learning
• Explain dynamic noise adjustments in differential privacy
• Balance privacy preservation and model performance

Innovation

• FedSDP uses Shapley values for privacy contribution assessment.
• Dynamic noise adjustment based on privacy requirements.
• Enhances interpretability of differential privacy in federated learning.