Paper 2025/572

Zinnia: Expressive, Efficient Zero-Knowledge Framework for General-Purpose Data Analytics

Abstract

Data analytics is a powerful tool for uncovering patterns and generating insights. However, once a claim is made based on data analysis, its audience must either trust the analyst or re-execute the analysis (often on private or proprietary data) to verify its correctness. This reliance raises significant concerns about transparency and trust in the analytics process. Zero-Knowledge Proofs (ZKPs), a cryptographic technique, offer a principled solution by enabling analysts to produce proofs of correctness without revealing the underlying data. This paradigm, known as verifiable computation, allows any verifier to check the validity of the analysis result solely from the proof. In this paper, we introduce Zinnia, a expressive and efficient ZKP framework designed for general-purpose data analytics. Zinnia provides a high-level domain-specific language (DSL) for encoding analytics workflows and a symbolic execution engine that reasons about the programs and compiles them into optimized ZKP circuits. Together, these components support rich language features such as data-dependent control flow (e.g., dynamic loops, recursion, early exits), real-valued arithmetic, non-linear functions and multidimensional array manipulations. We evaluate Zinnia's usability through real-world case studies and a user study, and benchmark its performance across diverse analytics tasks. Zinnia achieves up to $5.8\times$ speedup over zkVMs and produces ZKP circuits that are $19.3\%$ smaller than those generated by existing zero-knowledge programming languages.