[论文解读] PAC to the Future: Zero-Knowledge Proofs of PAC Private Systems

Guilhem Repetto, Nojan Sheybani+2 more|arXiv (Cornell University)|Feb 12, 2026

Cryptography and Data Security被引用 0

一句话总结

tldr: 论文提出一个框架，将 PAC Privacy 与零知识证明相结合，在外包计算中利用非交互式 zk-STARK 实现可验证、后量子安全的隐私保护。

ABSTRACT

Privacy concerns in machine learning systems have grown significantly with the increasing reliance on sensitive user data for training large-scale models. This paper introduces a novel framework combining Probably Approximately Correct (PAC) Privacy with zero-knowledge proofs (ZKPs) to provide verifiable privacy guarantees in trustless computing environments. Our approach addresses the limitations of traditional privacy-preserving techniques by enabling users to verify both the correctness of computations and the proper application of privacy-preserving noise, particularly in cloud-based systems. We leverage non-interactive ZKP schemes to generate proofs that attest to the correct implementation of PAC privacy mechanisms while maintaining the confidentiality of proprietary systems. Our results demonstrate the feasibility of achieving verifiable PAC privacy in outsourced computation, offering a practical solution for maintaining trust in privacy-preserving machine learning and database systems while ensuring computational integrity.

研究动机与目标

• Motivate privacy concerns in large-scale ML and data systems outsourcing.
• Introduce a framework that verifies correct PAC privacy noise is applied in trustless environments.
• Provide non-interactive, post-quantum secure zero-knowledge proofs to attest correct computation.
• Demonstrate feasibility and practicality of verifiable PAC privacy in cloud-based ML and database tasks.

提出的方法

• Extend PAC Privacy with zero-knowledge proofs to certify correct noise application.
• Use non-interactive zk-STARKs to avoid trusted setups and enable public verifiability.
• Define two deterministic functions f_h and f_PAC to enable verifiable noise computation.
• Provide a workflow where the server proves correct noise covariance and PAC-privacy implementation without revealing Σ.
• Implement and evaluate mechanisms such as K-means, SVM, and database statistics within a RISC-Zero framework.

Figure 2. Evolution of the centroids after some iterations of the Risc-Zero implementation of $4$ -means ( $K=4$ )

实验结果

研究问题

• RQ1 Can zero-knowledge proofs certify that PAC privacy noise is correctly computed and applied in outsourced computations?
• RQ2 How can post-quantum secure, non-interactive ZKPs enable verifiable privacy guarantees without exposing proprietary parameters?
• RQ3 What is the practicality and overhead of implementing PAC privacy with zk-STARKs in ML and database workloads?

主要发现

• Demonstrate feasibility of verifiable PAC privacy in outsourced computation using zk-STARKs.
• Show that noise for PAC privacy can be generated and proven without revealing private parameters like Σ.
• Illustrate near-plaintext utility with minimal overhead for proof generation in ML and database tasks.
• Adapt existing mechanisms (K-means, SVM, database statistics) to a PAC-private, ZKP-enabled workflow.
• Provide an end-to-end framework leveraging RISC-Zero to realize non-interactive, transparent proofs.