Trusted and Resilient User-controlled Scalable Transformation for (Privacy-Preserving) Machine Learning (TrustML)

• Unité de recherche: INNOSUISSE
• Numéro de projet: 121.294 IP-ICT
• Titre du projet anglais: Trusted and Resilient User-controlled Scalable Transformation for (Privacy-Preserving) Machine Learning (TrustML)

Mots-clé (Anglais)

ICT, IT security and cryptography

Résumé des résultats (Abstract) (Anglais)

TrustML is a privacy-preserving framework for secure inference and training of Machine Learning models in the cloud that will offer a widely applicable privacy-enhancing solution. We specifically focus on its application to cloud-based privacy-preserving large language models.