Hugo Schmutz

About Me

I am a researcher in Statistics and Computer Science in the QARMA team of the Laboratory of Informatic and Systems in Marseille. I mostly focus on incomplete supervision theory for real-world applications.

Real-world datasets, due to their complexity, often involve incomplete supervision schemes when used in machine learning pipelines. Despite advancements in methodological frameworks for incomplete supervision, a significant gap remains in translating these methods to practical, scalable solutions for real-world applications. In particular, theoretical guarantees are frequently missing, yet they are essential for critical settings where large validation sets are not available.

From September 2020 to December 2023, I was a PhD student at the Université Côte d'Azur and at Centre Inria d'Université Côte d'Azur in the Maasai team, under the supervision of Olivier Humbert and Pierre-Alexandre Mattei, with funding from the 3IA Côte d'Azur. My thesis focused on predicting the response to immunotherapy in patients with lung cancer, in particular designing new "safe" semi-supervised learning methods. [Dissertation][Slides][Video]

Since May 2025, I am now a postdoc under the supervision of Thierry Artières and Hachem Kadri, with Amidex fundings working in collaboration with the Institut de Chimie Radicalaire on active learning and active testing for quantum chemistry, specifically on learning molecular potential energy surfaces and forces from geometric configuration.

Teaching 📓

Present: Aix-Marseille Université

BSc Informatic - Data Science

BSc Informatic - Machine learning introduction

MSc IAAA - Machine learning

Past: Université Côte d'Azur

BSc Biology - Informatic basics

MSc Data Science and AI - Deep Learning; Project supervision

DU "IA et santé" - Statistical learning; Machine learning[Slides]

MBBS - "ImAge": IA for medical images - Université Côte d'Azur & Université de Caen Normandie[Slides]

Articles

📄 H. Schmutz, O. Humbert, P.-A. Mattei, "Don't fear the unlabelled: safe deep semi-supervised learning via simple debiasing", ICLR 2023. [article]
[code]

📄 A. Sportisse, H. Schmutz, O. Humbert, C. Bouveyron, P.-A. Mattei, "Are labels informative in semi-supervised learning?--Estimating and leveraging the missing-data mechanism", ICML 2023. [article]

📄 F. Bergamin, P.-A. Mattei, J. Drachmann Havtorn, H. Sénétaire, H. Schmutz, L. Maaloe, S. Hauberg, J. Frellsen, "Model-agnostic out-of-distribution detection using combined statistical tests", AISTATS 2022. [article]

📄 V. Comte, H. Schmutz, D. Chardin, F. Orlhac, J. Darcourt, O. Humbert, "Development and validation of a radiomic model for the diagnosis of dopaminergic denervation on [18F] FDOPA PET/CT", EJNMMI 2022. [article]

📄 H. Schmutz, P.-A. Mattei, S. Contu, D. Chardin, O. Humbert, "18FDG PET/CT and Machine Learning for the prediction of lung cancer response to immunotherapy", EANM 2022. [article]

📄 E. Kiner, E. Willie, B. Vijaykumar, H. Schmutz, J. Chandler, G. LeGros, S. Mostafavi, D. Mathis, C. Benoist and the Immunological Genome Project consortium, "Gut CD4+ T cell phenotypes: a continuum molded by microbes, not by Th archetypes", Nature Immunology 2021. [article]

📄 A. Maslova, R. N. Ramirez, K. Ma, H. Schmutz, C. Wang, C. Fox, B. Ng, C. Benoist, S. Mostafavi and the Immunological Genome Project consortium, "Learning immune cell differentiation", PNAS 2022. [article]