I am an associate astronomer working at the "Laboratoire d'Astrophysique de Marseille" (LAM) of the Aix-Marseille University in the South of France.
Currently I am quite involved in the Euclid mission which allows me to work with beautiful data such as this image of the Perseus cluster. I am also presently co-leading the Galaxies, Stars, and Cosmology team (GECO) at LAM.
I am an observational cosmologist, which means I try to understand how the Universe formed and evolves by comparing theoretical models with real observational data. My main tool for this is gravitational lensing, one of the coolest prediction of general relativity. Massive objects — such as galaxies — locally bend space-time. As light from distant background galaxies travels through this warped space, its path is deflected, subtly distorting the apparent shapes of those galaxies. By measuring these tiny shape distortions, we can effectively "weigh" the matter responsible for the bending, including dark matter, which emits no light and can only be traced through its gravitational effects. This technique, applied statistically over millions of galaxies, is known as cosmic shear. It lets us map the invisible distribution of dark matter across the sky and probe dark energy and the growth of cosmic structure over time — some of the biggest open questions in modern cosmology.
My research contributes to this effort in three main directions: measuring galaxy shape distortions as precisely as possible, developing higher-order statistical tools to extract more cosmological information from lensing data, and applying deep learning to both image analysis and cosmological inference.
Shear calibration
Measuring galaxy shapes accurately is the foundation of weak gravitational lensing — and it's harder than it sounds. Atmospheric blurring, telescope optics, and noise all distort galaxy images in ways that mimic or hide the lensing signal. To reach the precision required by surveys like Euclid, shear measurement algorithms must be tested and calibrated on realistic image simulations, where the "true" input distortion is known in advance. I currently lead this calibration effort within the Euclid Consortium. We produce highly realistic simulated images that reproduce the telescope's instrumental effects and survey conditions and process them through the same pipeline used for actual observations. Comparing the measured shear to the known input shear lets us quantify and correct for biases in our measurement algorithms, which is essential for obtaining reliable, unbiased cosmological results.
Higher-order weak lensing statistics
I co-lead the HOWLS (Higher Order Weak Lensing Statistics) team within Euclid. Traditional cosmic shear analyses rely on the power spectrum or two-point correlation functions, which fully capture the information in a Gaussian random field. But the late-time growth of structure is non-linear, especially on small scales, and produces non-Gaussian features — filaments, clusters, voids — that these standard estimators simply miss. Our goal is to extract this extra information using complementary statistics such as peak counts, one-point probability distribution functions, higher-order moments, Minkowski functionals, and scattering transforms. Together, these methods can recover 2–3 times more cosmological information than standard two-point statistics alone, significantly sharpening our constraints on dark matter and dark energy.
Deep learning
Finally, I am the Principal Investigator of the PISCO (PIxelS to COsmology) ANR project, which asks a more radical question: instead of compressing images into hand-designed statistics, can we extract cosmological information directly from the pixels? This project uses deep learning tools — convolutional neural networks and neural density estimators — to learn the shear signal directly from galaxy images, and to infer cosmological parameters from shear maps, bypassing traditional summary statistics altogether.
I supervise students and postdocs at every level and welcome motivated candidates — feel free to get in touch.