I’m Etienne Lehembre, a postdoctoral researcher currently working in data science and machine learning. My background is focused on theoretical computer science, but I’m currently looking at algorithms considering the mathematical relations inside structured data. I worked on subgraphs and chemical data during my thesis, but I’m currently getting on board with sequential data and forecasting models.
My PhD thesis, was part of the ANR project InvolvD. I designed methods aiming to expose the ambiguities hidden behind the expert’s prior. I aslo took an interest in the leverages available to guide an expert through a dataset they don’t know in the most robust possible manner. These work led me to study the evaluation process of the interactive pattern mining methods in order to improve them.
Nowadays, studying reinforcement and machine learning in the scope of the Junon project. In this setting, I work with hydro-geologist on the integration of their expert knowledge in a machine learning model. In order to take into account the expert knowledge, I developed a novel neural network architecture using the physical equations to modify the latent space, leading to better, and more explainable, predictive performance.