Scientific direction Development of key enabling technologies
Transfer of knowledge to industry

PostDocs : selection by topics

Technological challenges >> Cyber security : hardware and sofware
1 proposition(s).

See all positions

Privacy in embedding-based neural networks by means of homomorphic encryption

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire composants logiciels pour la Sûreté et la Sécurité des Systèmes

01-01-2020

PsD-DRT-20-0021

renaud.sirdey@cea.fr

Cybersécurité : hardware et software (.pdf)

AI presently emerges as the killer application of homomorphic encryption or FHE. Indeed, this kind of cryptography, which allows to perform general calculations directly over encrypted data, has the potential of bringing privacy-by-construction for either or both user or model data, depending on the application scenario. In the longer term, FHE may also help protect training data, unleashing new usages in training data sharing and collaborative AI model building. In this context, the present postdoctoral offer aims at investigating the practical relevance of homomorphic encryption in the case of a specific kind of neural networks, the so-called embedding-based networks, which, for intrinsic reasons, both are favorable to good homomorphic execution performances and enjoy a wide spectrum of applications. Thus, this postdoctorate will study the theoretical and practical aspects cropping up in several FHE integration scenarios and will also lead to prototyping work on a best-in-class open-source speech recognition system using an embedding-based network.

Download the offer (.zip)

See all positions