Advances in AI and Differentiable Computing for the Analysis of Cosmological Galaxy Surveys

As we move towards the next generation of galaxy surveys, our field is facing outstanding challenges at all levels of scientific analysis, from pixel-level data reduction to cosmological inference. In this talk, I will present how the most recent techniques in AI and associated computational frameworks impact our ability to make sense of these increasingly large and complex datasets.

New AI architectures, particularly Foundation Models trained through self-supervised learning, are poised to significantly transform how Deep Learning is deployed on the extremely large and complex datasets typical of modern cosmological galaxy surveys such as the Euclid mission and the Vera C. Rubin Observatory LSST survey. These powerful AI models facilitate lower-level data processing and enhance our ability to interpret rich, multimodal observational data (images, spectra, timeseries) from different instruments and different observatories. I will discuss our recent work on building such multimodal foundation models, highlighting their potential to streamline and optimize tasks like redshift estimation, morphology classification, and performing physical inference on individual astronomical objects.

At the other end of the processing chain, I will discuss the emergence of new cosmological inference methods which replace traditional analytic theoretical models in favor of full simulation models integrated into high-dimensional Bayesian inference frameworks. These probabilistic, differentiable models not only enable a more precise modeling of cosmological surveys, but also streamline the joint analysis of multiple cosmological probes. On this topic, I will present our work on building the necessary simulation frameworks, as well as developing efficient MCMC sampling methods that make inference tractable in the 10^9 dimensions involved in this problem. 

--

Dr. Lanusse is a CNRS researcher in observational cosmology and machine learning at the Astrophysics Department of CEA Paris-Saclay (France), and a guest researcher at the Flatiron Institute in New York City. He received his PhD in cosmology and inverse problems in 2015 in Paris, and further developed an interdisciplinary expertise in Deep Learning for cosmology as a postdoctoral researcher at Carnegie Mellon University (2015-2018) and UC Berkeley (2018-2019) through multiple collaborations with their respective Machine Learning and Statistics Departments. He is now broadly interested in developing scientific applications of state of the art Deep Learning techniques, by combining concepts of bayesian inference, deep neural networks, and physical forward modeling.