Physical Society Colloquium

Explorations in Deep Learning for Astrophysics

Siamak Ravanbakhsh

School of Computer Science
McGill University

With the growing number of sky surveys, the sheer size of observational data is creating a crucial role for deep learning in astronomy and cosmology. This talk explores some of these applications, ranging from predicting physical constants to fast cosmological simulations to the identification of rare events. Throughout the talk, I will emphasize the choice of data structure and show symmetry-based deep models can accommodate non-standard data structures such as point sets and spherical data that are widespread in astronomy.