I develop deep learning methods to decode the structure and function of the human brain — from detecting rare white matter lesions in adolescents to mapping the neural circuits that make brain stimulation work.

Visiting Researcher at UCSF (Rauschecker–Sugrue Lab). Trained in materials engineering and AI at École Polytechnique, I apply multi-scale modeling and representation learning to clinical neuroimaging.

Research

Weak Supervision for White Matter Lesion Detection in Adolescents

A Swin UNETR pipeline trained under extreme class imbalance to detect leukoaraiosis from T1w/T2w pairs across 11,868 adolescents — without a single voxel-level annotation.

Connectivity Fingerprint Analysis of DBS Stimulation Sites

SIFT2-weighted tractography fingerprints from 116 stimulation configurations in 5 DBS patients, tested against the destination network convergence hypothesis and validated with seven ML classifiers.

Multi-Scale Temporal Masking and Distribution Regularization for Self-Supervised fMRI Learning

Two extensions to Brain-JEPA: a hierarchical masking scheme at short, medium, and long temporal horizons, and a VICReg covariance regularization term that prevents representational collapse in latent fMRI space.

Explorations

Conversations with ResearchersA Spotify podcast where I invite researchers working at the intersection of AI and science.
Listen on Spotify →