Model Exposure on One Device Endangers Others: Rethinking Edge Neural Network Security across Devices

Key points

• Neural networks (NNs) are increasingly deployed on edge devices such as mobile and IoT platforms, offering low latency and improved privacy compared to cloud-based deployment.1
• However, on-device models embedded in mobile applications can be unpacked and exposed to adversaries, enabling cross-node attacks in centrally distributed ecosystems, where adversarial examples crafted on one compromised device transfer effectively to other nodes sharing the same models.1
• The core idea of VeilNN is to introduce controlled model diversity while preserving original task functionality.1
• Specifically, VeilNN applies stealthy and irreversible obfuscation at both the architectural and parameter levels through network cell obfuscation and layer assembling, and is applicable to diverse NN architectures, including MLPs, CNNs, and Vision Transformers.1

Sources (1)

• [1]Model Exposure on One Device Endangers Others: Rethinking Edge Neural Network Security across Devices
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 06:45 PM

  To address this threat, we propose VeilNN, a defense framework that reduces adversarial transferability across on-device models via model obfuscation.

  Neural networks (NNs) are increasingly deployed on edge devices such as mobile and IoT platforms, offering low latency and improved privacy compared to cloud-based deployment.

Extractive summary: sentences quoted from the sources.