Authors: Raphael Joud, Pierre-Alain Moellic, Simon Pontie, Jean-Baptiste Rigaud

Published on: November 02, 2023

Impact Score: 8.38

Arxiv code: Arxiv:2311.01344

Summary

• What is new: A methodology for extracting architecture information from neural network models via EM side-channel traces, focusing on edge devices.
• Why this is important: The security of AI systems is at risk due to the possibility of model extraction through side-channel leakages.
• What the research proposes: Using simple pattern recognition analysis to extract architecture information from MLP and CNN models on Cortex-M7 microcontrollers.
• Results: The complexity of extracting architecture information is relatively low, highlighting the need for effective protection mechanisms.

Technical Details

Technological frameworks used: ARM CMSIS-NN

Models used: MLP and CNN

Data used: EM side-channel traces

Potential Impact

AI security services, edge device manufacturers, and companies relying on proprietary AI models for competitive advantage

Want to implement this idea in a business?

We have generated a startup concept here: SecureAI Guard.