Brief Biography
I have a strong scientific background, with a particular emphasis on mathematics, statistical learning, and computer vision. I am currently finalizing my PhD on Neural Radiance Fields (NeRFs) and generative modeling through a collaboration between the Criteo AI Lab in Paris and the Laboratoire d’Informatique, Signaux et Systèmes (I3S) in Sophia Antipolis. My research focuses on bridging 2D foundation models and 3D scene representations to improve 3D reconstruction and generation. My work has been published at major machine learning and computer vision venues, including CVPR, ICLR, and TMLR.
Background and Position
I was born in Alsace, France, where I completed my early studies before attending the classes préparatoires in mathematics and physics at Lycée Kléber. I then pursued engineering studies at CentraleSupélec near Paris. During this time, I completed research internships in computer science laboratories, including one in a public Danish research lab and another at Schlumberger, where I developed a strong interest in machine learning and deep learning. I later completed the Mathematics, Vision, Learning (MVA) Master’s program at École Normale Supérieure Paris-Saclay, specializing in machine learning and computer vision. I am currently pursuing an industrial PhD jointly between Criteo, a technology company specializing in online marketing, and the I3S laboratory, a robotics research unit operated by CNRS and Université Côte d’Azur.
Research Areas
My research explores how advances in 2D foundation models can be leveraged to improve scalable 3D reconstruction and generation. I study how latent representations learned in image models can be transferred to Neural Radiance Fields to build more efficient and scalable 3D representations. By designing representations that are structurally compatible with image models, such as latent NeRFs and tri-planar representations, I enable the direct integration of strong 2D priors into 3D learning. More recently, my work investigates how generative image models can be used to create immersive and navigable 3D worlds.
Selected Publications
- Bringing NeRFs to the Latent Space: Inverse Graphics Autoencoder
Venue: ICLR 2025
Project page: https://ig-ae.github.io/ - Fused-Planes: Why Train a Thousand Tri-Planes When You Can Share?
Venue: ICLR 2026
Project page: https://fused-planes.github.io/ - RefinedFields: Radiance Fields Refinement for Planar Scene Representations
Venue: TMLR
Project page: https://refinedfields.github.io/ - SphericalDreamer
Project page coming soon.