Making neural networks secure and reliable

Project: Security of AI for Defense Applications (SAIDA)

Project leader: Teddy Furon – Linkmedia team – Inria Rennes - Bretagne Atlantique Activity scope: computer-aided diagnosis, defence (target detection).

Aim: to study what it is that makes neural networks vulnerable to malevolent falsifications; to develop tools for detecting falsified images and for cleaning images. Ultimately, the goal is to adapt the process used in creating neural networks in order to prevent them from being deceived by such disturbances.