Skip to content
Menu
ls -1F

Research

Publications

  • security-for-distributed-machine-learning.md

    Security for Distributed Machine Learning

    Laurent Gomez, Tianchi Yu, Patrick Duverger

    SECRYPT · 20th International Conference on Security and Cryptography

    Abstract

    With the adoption of IoT-like technologies, industrials aim to enhance the business value of their physical assets and improve their operational efficiency. However, IoT devices alone tend to strain enterprise systems with a sheer volume of unstructured and unfiltered data. To overcome this challenge, endowing (smart) devices with AI-based capabilities can significantly enhance enterprise system capabilities. However, deploying AI-based capabilities on potentially insecure edge hardware and platforms introduces new security risks, including AI model theft, poisoning, and data leaks. This paradigm shift necessitates the protection of distributed AI applications and data. In this paper, we propose a solution for safeguarding the Intellectual Property and data privacy of ML-based software. We utilize hardware-assisted Privacy Enhancing Technologies, specifically Trusted Execution Environments. We evaluate the effectiveness of our approach in the context of ML-based motion detection in CCTV cameras. This work is part of a co-innovation project with the Smart City of Antibes, France.

in_progress/

Current directions

  • System and hardware security

    Memory protection and trustworthy systems.

  • Hardware–software co-design

    Security at the boundary between system software and hardware.

  • AI infrastructure and heterogeneous computing

    GPU execution, inference runtimes, and numerical correctness.

    Related experiments

If you’re working on related questions, I’d be happy to exchange ideas, compare experiments, or discuss a possible collaboration.

Get in touch