PhD in bioinformatics. Eleven years of applied machine learning, eight of them at Mila (Quebec AI Institute). Now Lead Data Scientist.
I work on the part most projects get wrong: deciding what "good" means before choosing a model. Datasets, benchmarks and evaluation protocols that match the decision the model actually serves. Domains so far: power line inspection, hyperspectral medical imaging, organ transplantation, diabetic retinopathy screening, water demand forecasting, industrial anomaly detection.
Expert annotated anomaly detection dataset and evaluation protocol, built with Hydro-Québec. 4,798 images, 6,023 annotated anomalies. NeurIPS 2024, Datasets and Benchmarks. Co-first author.
Exhaustive rule mining for the early prediction of chronic kidney disease from metabolomics and multi-source data. Rules a clinician can read, at the same level of prediction as the global baseline. PLOS One 2016, ICMLA 2015 (oral). First author.
Deep learning for kidney graft survival on censored data, US national transplant registry. Wasserstein metric for time-to-event analysis. PMLR 2021, second author. I led the earlier stages of this line as first author (arXiv 2017, ICML workshop 2018).
Full list on Google Scholar.
Co-author on two award winning papers. One of the general organizers of the Joint AI for Social Good workshop at NeurIPS 2019, chair roles at ICML 2019, ICLR 2019 and NeurIPS 2018. Reviewer for the American Journal of Transplantation and PLOS One. One of six scientific experts teaching the SDS-230 MOOC on data science and health (IVADO, Mila, IRIC), under the scientific direction of Yoshua Bengio.
Based in Montréal.