Skip to yearly menu bar Skip to main content


Poster

Optimistic Verifiable Training by Controlling Hardware Nondeterminism

Megha Srivastava · Simran Arora · Dan Boneh

East Exhibit Hall A-C #4508
[ ] [ Project Page ]
Wed 11 Dec 11 a.m. PST — 2 p.m. PST

Abstract:

The increasing compute demands of AI systems has led to the emergence of services that train models on behalf of clients lacking necessary resources. However, ensuring correctness of training and guarding against potential training-time attacks, such as data poisoning and backdoors, poses challenges. Existing works on verifiable training largely fall into two classes: proof-based systems, which can be difficult to scale, and ``optimistic'' methods that consider a trusted third-party auditor who replicates the training process. A key challenge with the latter is that hardware nondeterminism between GPU types during training prevents an auditor from replicating the training process exactly, and such schemes are therefore non-robust. We propose a method that combines training in a higher precision than the target model, rounding after intermediate computation steps, and storing rounding decisions based on an adaptive thresholding procedure, to successfully control for nondeterminism. Across three different NVIDIA GPUs (A40, Titan XP, RTX 2080 Ti), we achieve exact training replication at FP32 precision for both full-training and fine-tuning of ResNet-50 (23M) and GPT-2 (117M) models. Our verifiable training scheme significantly decreases the storage and time costs compared to proof-based systems.

Live content is unavailable. Log in and register to view live content