When: October, 14, 2019 @ 12:30pm Where: PAB, 6th Floor, eScience Studio, Seminar Rm.

Deep Generative Models of Galaxy Morphology and Application to Deblending

The upcoming generation of wide-field optical surveys which includes LSST will aim to shed some much needed light on the physical nature of dark energy and dark matter by mapping the Universe in great detail and on an unprecedented scale. However, with the increase in data quality also comes a significant increase in data complexity, bringing new and outstanding challenges at all levels of the scientific analysis. In this talk, I will illustrate how deep generative models, combined with physical modeling, can be used to address some of these challenges at the image processing level, specifically by providing data-driven priors of galaxy morphology.

I will first describe how to build such generative models from corrupted and heterogeneous data, i.e. when the training set contains varying observing  conditions (in terms of noise, seeing, or even instruments). Once trained, sampling from these models produces realistic galaxy light profiles, which can then be used in survey emulation, for the purpose of validating and/or calibrating data reduction pipelines.

Even more interestingly, these models can be seen as priors on galaxy morphologies and used as such as part of standard Bayesian inference techniques to solve astronomical inverse problems ranging from deconvolution to deblending galaxy images. I will present how combining these deep morphology priors with a physical forward model of observed blended scenes allows us to address the deblending problem in a physically motivated and interpretable way.  

About François Lanusse

I am currently a postdoctoral follow at the Berkeley Center for Cosmological Physics (BCCP) and with the Foundation of Data Analysis (FODA) institute at UC Berkeley, where I conduct my research at the intersection between cosmology and machine learning.

Previously, I was a postdoctoral researcher in the McWilliams Center for Cosmology at Carnegie Mellon University, where I was working with Prof. Rachel Mandelbaum on weak gravitational lensing measurements and systematics, and also interacted with both Statistics and Machine Learning departments here at CMU.

I did my PhD in the CosmoStat laboratory of CEA Saclay near Paris, France, under the supervision of Jean-Luc Starck. My PhD work focused on the application of sparse regularization techniques to solve ill-posed inverse problems in a cosmological context.

Before that, I received a Master’s degree in fundamental physics from Paris-Sud University as well as a Master’s degree in fundamental and applied mathematics from Paul Verlaine University in Metz (France). I am also a Supélec engineer, from one of France’s top grandes écoles, where I specialized in Robotics and Interactive Systems.

You can find my academic CV here.