101 ways in which machine learning can help your research (or can it?)

Machine learning applications to physical sciences have seen a huge diffusion in recent years, with astrophysics not being immune to the appealing features of such powerful and expressive models. These algorithms promise to be the only possible answer to the big-data challenges that lie in front of us thanks to facilities like Euclid and the Square Kilometre Array Observatory, to name a few. In this talk, I will present recent applications of deep learning techniques to accelerate Bayesian analyses for next-generation surveys, as well as to emulate summary statistics like cosmological power spectra and peculiar velocity covariances, with a critical focus on the advantages and limitations of such methods over traditional techniques. I will also discuss the black-box nature of deep learning models, and present how we can interpret them combining representation learning and information theory. As I will not cover all recent trends in machine learning applications to astrophysics, attendees are welcome to bring their own problems!