Distributed Differentially Private Data Analytics via Secure Sketching

2024-11-30

IACR Cryptology ePrint Archive

AI Summary

To resolve the inherent trade-off in distributed differential privacy—namely, the single-point failure risk of the central model versus the excessive noise of the local model—this paper proposes a linear transformation intermediate model. Clients perform privacy-preserving computation within a trusted execution environment (TEE) that executes only publicly known matrix multiplication, while decentralized coordination is achieved via efficient secure multi-party computation. This model establishes, for the first time, a provably sound theoretical bridge between central and local models, inheriting the expressive power of the former and the trust minimization of the latter. Theoretically, it incurs significantly less noise than the local model, with estimation error independent of the number of clients. Empirically, on private low-rank approximation and private ridge regression tasks, it achieves superior data utility, reduced communication overhead, and improved privacy budget efficiency—thereby advancing privacy, utility, and computational efficiency simultaneously.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Search and Optimization: Distributed Search
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

We introduce the linear-transformation model, a distributed model of differentially private data analysis. Clients have access to a trusted platform capable of applying a public matrix to their inputs. Such computations can be securely distributed across multiple servers using simple and efficient secure multiparty computation techniques. The linear-transformation model serves as an intermediate model between the highly expressive central model and the minimal local model. In the central model, clients have access to a trusted platform capable of applying any function to their inputs. However, this expressiveness comes at a cost, as it is often prohibitively expensive to distribute such computations, leading to the central model typically being implemented by a single trusted server. In contrast, the local model assumes no trusted platform, which forces clients to add significant noise to their data. The linear-transformation model avoids the single point of failure for privacy present in the central model, while also mitigating the high noise required in the local model. We demonstrate that linear transformations are very useful for differential privacy, allowing for the computation of linear sketches of input data. These sketches largely preserve utility for tasks such as private low-rank approximation and private ridge regression, while introducing only minimal error, critically independent of the number of clients.

Problem

• Distributed differentially private data analysis model
• Balances central and local model trade-offs
• Enables private computations with minimal error

Innovation

• Distributed differentially private data analytics
• Secure sketching via linear transformations
• Mitigates noise and single point failure