With Tingting Zhu at the CHI Lab, IBME, University of Oxford, and the clinicians Simon Knight and Laura Wingfield, I contribute to the construction of a computerised decision support system (CDSS) to help surgeons and patients understand if a given organ offer is a suitable one, or if it is preferable to wait for another one.

More precisely, given more than twenty years of data provided by the National Institute for Health and Care Research (NIHR) with ethical approval, we develop machine learning models predicting transplant outcomes (graft failure and patient death) in the case of an accepted offer, or telling what could happen in the case of a denied one. These models are supported by interpretability methods.

During this industrial PhD (CIFRE) between Télécom SudParis (SAMOVAR) and Nokia Bell Labs, I worked on predicting alarms in networks via space-time pattern matching and machine learning. For this, I benefited from the supervision of François Desbouvries, Anne Bouillard, Marc-Olivier Buob, and Yohan Petetin.

This thesis is two-fold. On the one hand, we proposed a structure, called DIG-DAG, able to store online causality chains observed within log. On the other hand, we compared analytically the expressivity of two popular generative models: Hidden Markov Models and Recurrent Neural Networks.

Apart from the theoretical work, my thesis shows a strong applied component. Indeed, Anne, Marc-Olivier and I implemented our DIG-DAG related algorithms into a Python 3 module, which has been at the core of a collaboration with Nokia’s business units.

Please follow this link for more details about my thesis.