There have been many efforts to apply methods from machine learning and statistics to make discoveries in astrophysics and throughout the physical sciences. While it is clear that the use of these methods has advanced our science goals, I will argue that these collaborations can also advance research in machine learning. My research program is focused on the discovery and characterization of planetary systems orbiting stars throughout the Milky Way and the excitement surrounding this field has encouraged productive interdisciplinary collaborations. In this talk, I will present some examples from my research where these collaborations have enabled new discoveries in astrophysics and inspired the development of new methods for machine learning. In particular, I will discuss two methods (and their open source implementations) that I have developed in collaboration with applied mathematicians for scaling Gaussian Process inference to the large astronomical datasets. My research program also emphasizes the development of well-tested and user-friendly open source implementations of these methods, as well as the educational materials needed for scientists to most effectively benefit from these developments. I will discuss how this open source work and the practices commonly used in open source communities have benefited my research group.
Daniel Foreman-Mackey joined the [Simons] Foundation in 2017 as an associate research scientist at the Center for Computational Astrophysics. His research focuses on developing novel data-analysis methods and applying them to astronomical survey datasets. Recently, he has been using a combination of data-driven and physically motivated time-series models to discover and characterize new exoplanets using observations from NASA’s Kepler and K2 missions. Before joining the center, Foreman-Mackey received his Ph.D. from New York University in 2015 and then spent two years as a NASA Sagan Postdoctoral Fellow at the University of Washington.