Bastien Vialla (Orange Labs, Caen)

In the Franco-German collaborative project CRYPTECS, Orange Innovation explores privacy-preserving technologies for industrial applications. A prime use-case is detecting compromised computers through network traffic analysis. We have developed efficient AI models tailored for this. Our aim is to deploy these models while safeguarding both the model and network data, making Fully Homomorphic Encryption (FHE) an ideal theoretical solution.

This presentation is at the cross-section of cybersecurity, AI and cryptography. We will delve into the broader context of the use-case, the AI models used, the adaptations required in training to accommodate FHE, the consequential impact on prediction quality, and overall performance metrics.