Authors: Mingyi Zhou, Xiang Gao, Jing Wu, Kui Liu, Hailong Sun, Li Li

Published on: February 08, 2024

Impact Score: 8.3

Arxiv code: Arxiv:2402.05493

Summary

• What is new: Introducing a new Reverse Engineering framework (REOM) that allows white-box attacks on on-device deep learning models, showcasing significantly higher efficiency and effectiveness over previous black-box attack methods.
• Why this is important: On-device deep learning models in mobile apps are vulnerable to attacks, but existing methods only support less effective black-box attacks due to lack of support for gradient computing.
• What the research proposes: The REOM framework systematically transforms compiled TFLite models into a debuggable format, enabling white-box attacks that were previously impractical.
• Results: REOM effectively transformed 244 TFLite models for white-box attacks, proving to be far more efficient by achieving higher success rates with much smaller attack perturbations.

Technical Details

Technological frameworks used: REOM, TFLite, ONNX

Models used: Deep Learning models

Data used: nan

Potential Impact

Mobile app developers, cybersecurity fields, deep learning application markets.

Want to implement this idea in a business?

We have generated a startup concept here: ShieldNet.