Paper 2024/639

Computational Attestations of Polynomial Integrity Towards Verifiable Machine Learning

Abstract

Machine-learning systems continue to advance at a rapid pace, demonstrating remarkable utility in various fields and disciplines. As these systems continue to grow in size and complexity, a nascent industry is emerging which aims to bring machine-learning-as-a-service (MLaaS) to market. Outsourcing the operation and training of these systems to powerful hardware carries numerous advantages, but challenges arise when needing to ensure privacy and the correctness of work carried out by a potentially untrusted party. Recent advancements in the discipline of applied zero-knowledge cryptography, and probabilistic proof systems in general, have led to a means of generating proofs of integrity for any computation, which in turn can be efficiently verified by any party, in any place, at any time. In this work we present the application of a non-interactive, plausibly-post-quantum-secure, probabilistically-checkable argument system utilized as an efficiently verifiable guarantee that a privacy mechanism was irrefutably applied to a machine-learning model during the training process. That is, we prove the correct training of a differentially-private (DP) linear regression over a dataset of 60,000 samples on a single machine in 55 minutes, verifying the entire computation in 47 seconds. To our knowledge, this result represents the fastest known instance in the literature of provable-DP over a dataset of this size. Finally, we show how this task can be run in parallel, leading to further dramatic reductions in prover and verifier runtime complexity. We believe this result constitutes a key stepping-stone towards end-to-end private MLaaS.