Special LAM/ML seminar: Marc Huertas-Company – Deep learning the physics of galaxy formation

The availability of statistical samples of galaxies over the past decade has resulted in a global picture of galaxy formation which is successfully translated into modern cosmological simulations that reproduce realistic galaxies.

However, most of the key physical processes that govern galaxy evolution are still largely unconstrained – as evidenced by the first JWST results at cosmic dawn – and, as such, are treated as subgrid physics in state of the art simulations. The data quality and complexity is progressing fast, which coupled with recent advances in AI, offers new opportunities to make progress in our understanding of the physics of galaxy formation.

Following a general introduction, in my talk, I will discuss recent results from our group aiming at extracting information from large and multi-modal deep surveys and connecting cosmological simulations and observations using a variety of modern deep learning methods.