I can't see it but I can Fine-tune it: On Encrypted Fine-tuning of Transformers using Fully Homomorphic Encryption

TL;DR

The paper tackles privacy concerns in fine-tuning transformers by introducing BlindTuner, a system that performs end-to-end training on FHE-encrypted data. It leverages DEiT for data-efficient feature extraction and CKKS-based FHE to enable encrypted fine-tuning, using Nesterov optimization and encrypted matrix multiplications. Across MNIST, CIFAR-10, DermaMNIST, and Face Mask Detection, BlindTuner achieves accuracy comparable to unencrypted baselines while delivering substantial speedups (up to 1.5x–600x) over prior privacy-preserving methods. The work demonstrates practical viability of privacy-preserving MLaaS for vision transformers and outlines a path toward broader deployment on sensitive data domains.

Abstract

In today's machine learning landscape, fine-tuning pretrained transformer models has emerged as an essential technique, particularly in scenarios where access to task-aligned training data is limited. However, challenges surface when data sharing encounters obstacles due to stringent privacy regulations or user apprehension regarding personal information disclosure. Earlier works based on secure multiparty computation (SMC) and fully homomorphic encryption (FHE) for privacy-preserving machine learning (PPML) focused more on privacy-preserving inference than privacy-preserving training. In response, we introduce BlindTuner, a privacy-preserving fine-tuning system that enables transformer training exclusively on homomorphically encrypted data for image classification. Our extensive experimentation validates BlindTuner's effectiveness by demonstrating comparable accuracy to non-encrypted models. Notably, our findings highlight a substantial speed enhancement of 1.5x to 600x over previous work in this domain.

Figures (2)

• Figure 1: Overview of the proposed end-to-end BlindTuner
• Figure 2: BlindTuner performance on various datasets