Bridgeable

Mathematical analysis of neural networks
A_Pesquet_300_400
© Pesquet

Project: BRIDinG thE gAp Between iterative proximaL methods and nEural networks (Bridgeable)

Project leader: Jean-Christophe Pesquet – OPIS team – Inria Saclay - Île-de-France

Activity scope: 3D medical imaging (in collaboration with GE Healthcare), analysis of data from the field of energy and the environment (in collaboration with the IFPEN), modelling electric motors (Schneider Electric). Aim: to develop a mathematical solution to the problems encountered by computer models of neural networks. This project deals with the explainability of how these ‘black boxes’ work: what basic factors cause these tools to operate and why are some better than others? It also focuses on their reliability (assessing their fragility when faced with disturbances), comparing them with other learning methods.