Joint Estimation of Cosmic Shear, PSF, and Galaxy Morphologies
The goal of my PhD thesis is to develop a hierarchical probabilistic model of the observed Euclid images combining physical models with Deep Learning components accounting for unknowns factors. In particular, I aim to build a forward model of Euclid field of views accounting for the PSF, cosmic shear, and galaxy morphology. Fitting this model to observed exposures is a theoretically optimal way to jointly estimate the cosmic shear field and perform the calibration.
Variational Inference and Hierarchical models
So far, solving such inference problem at scale was intractable. I am very interested in efficient optimization-based inference approaches, such as Variational Inference, replacing expensive Markov Chain Monte Carlo methods, to solve this problem.
- Probabilistic Mapping of Dark Matter by Neural Score Matching
Benjamin Remy, François Lanusse, Zaccharie Ramzi, Jia Liu, Niall Jeffrey and Jean-Luc Starck.
Machine Learning and the Physical Sciences Workshop, NeurIPS 2020.
(arXiv, code, poster)
- Denoising Score-Matching for Uncertainty Quantification in Inverse Problems
Zaccharie Ramzi, Benjamin Remy, François Lanusse, Jean-Luc Starck and Philippe Ciuciu
Deep Learning and Inverse Problems Workshop, NeurIPS 2020.
- IN2P3/IRFU Machine Learning Workshop, March 17th, online. Slides
- Euclid Workshop on Machine Learning and Deep learning. December, 14th 2020, online
- Denoising Score Matching for Uncertainty Quantification in Inverse Problems: Application to gravitational lensing and Magnetic Resonance Imaging, with Zaccharie Ramzi. Machine Learning Club, Nov 18th 2020, online.