From Medical Imaging to Climate: Machine Learning for the Good of Society

Summary:
By establishing collaborations with experts in specific fields, we can use the power of artificial intelligence to help make diagnoses, and also to find concrete and local solutions for our companies. I will first talk about my doctoral work in medical imaging and non-invasive cardiac techniques for surgery prediction at Inria Sophia-Antipolis. In collaboration with cardiologists, I developed a customization of a non-invasive cardiac electrophysiological model and generated a simulated database for training a scattered Bayesian regression. Although the need for machine learning is still very important in medicine, I also realized that it was almost absent in other critical areas such as climate and computer sustainability. During my post-doctoral internship (CNRS Saclay and University of Colorado), I worked on the prediction of hurricane tracks from image-based meteorological measurements. We developed a convolutional neural network fusion, combining past trajectory data and reanalysis of atmospheric images (3D wind and pressure fields). A comparison with current forecast models shows that deep methods could provide a valuable and complementary forecast.