The « UltraBioLearn » project
Machine learning for medical applications, in particular supervised machine learning has a number of constraints that are hard to fulfill: results must be interpretable by clinicians; of high quality since they involve patient health; they must also be reproducible and trustworthy. Modern machine learning techniques also rely heavily on the quality of training databases and of human annotations. At the same time, the private lives of patients must be respected and so medical data access is restricted. The combination of all these factors make machine learning, in particular deep learning very difficult in practice.
This project proposes to research innovative solutions to these questions, in particular by leveraging semi-supervised learning using generative, graph-based and certifiable networks, in the context of predicting patient response to cancer treatments.
The project involves the IR4M lab of University Paris-Sud for the medical and medical imaging side and the CVN at CentraleSupelec for the computer vision and machine learning side.