Geneviève Robin
I’m a CNRS researcher affiliated to MAP5, Université Paris Cité. I develop Statistics and Machine Learning methods, with a particular emphasis on applications to high-dimensional, multimodal and complex health-care data. Previously, I worked for the French AI for Health company Owkin where I led the Exploratory Research Program. Some of my representative recent projects are highlighted below.
Generalizable Machine Learning for digital pathology
In computational pathology, a common task is to analyze digitally scanned tissue samples to predict endpoints that enhance disease understanding and improve patient care. A key requirement for deploying these models in real-world settings is their robustness to variations in slide preparation (e.g., staining) and digitization (e.g., scanner differences), ensuring reliability across different clinical centers, which can be significantly improved by relying on robust, state-of-the art Foundation Models associated with robust model calibration. Another important aspect is the integration of multimodal data combining histology with, multi-omics data, which can be done, e.g., using Multiple Instance Learning.
- Distilling foundation models for robust models in digital pathology
- Robust sensitivity control in digital pathology
- Multiple instance learning for multimodal patient-level predictions
Molecular dynamics for Machine Learning
Bayesian inference for large-scale datasets is of great interest for its applications in machine learning. It also offers a framework for knowledge transfer between numerical Langevin diffusion methods developed in statistical physics, and those that are used in statistical learning. One the one hand, we used a statistical physics perspective to analyze a Langevin diffusion process, offering some insights on how the diffusion coefficient to accelerate sampling. On the other hand, we investigated how Generative ML models can be leveraged to improve the sampling pf transition paths in Molecular dynamics.