The Milky Way is the only large spiral galaxy in which we can study a significant fraction of the stellar population on a star-by-star basis, building up an understanding of the workings of the galaxy from individual stars to the scale of the galaxy itself. As such, it offers a unique laboratory for answering some of the most important questions in astrophysics. What is the distribution of dark matter on small scales, and what can we learn about the nature of dark matter from its small-scale distribution? What is the star formation history of a typical large spiral galaxy, and how does it depend on the interaction of the galaxy with its environment?
The field of Milky Way dynamics and structure is undergoing a transformation, due both to the availability of revolutionary new datasets and significant methodological innovations, many of which are drawn from or inspired by the field of machine learning. The European Space Agency's Gaia telescope is gathering data of unprecedented precision on the positions and kinematics of over a billion stars in the Milky Way. At the same time, new ground-based spectroscopic surveys, such as the Sloan Digital Sky Survey V Milky Way Mapper, are providing detailed information on the chemical composition of millions of stars. These new datasets have opened up an entirely new view of the Milky Way. We are learning that the Milky Way is a highly dynamic system, which has experienced multiple major mergers with smaller galaxies throughout its lifetime and which shows signs of recent perturbation (e.g., by the Sagittarius dwarf galaxy and the Magellanic clouds).
In order to handle this flood of highly precise data, astronomers have increasingly looked to mathematical and computational tools drawn from or inspired by the field of machine learning. Driven by more terrestrial problems, such as facial recognition, self-driving cars, and machine translation, the field of machine learning has developed a large number of powerful computational tools over the last decade, such as convolutional neural networks, normalizing flows, autoencoders, and transformers, and has also driven the development of significantly more powerful hardware (primarily GPUs) to implement these new computational methods. Astronomers have begun to apply these methods to astrophysical problems, particularly in the field of Milky Way dynamics and structure. A few examples are the use of neural networks to flexibly parameterize the gravitational potential of the Milky Way; the use of symplectic normalizing flows to directly learn canonical transformations between Cartesian phase space and action-angle coordinates from observational data; the use of neural networks to model the structure of stellar streams; and the application of approximate Gaussian processes to map the density of dust throughout the Milky Way. Machine learning continues to rapidly innovate new computational techniques, making it at times difficult for astronomers to keep up.
The goal of this workshop is to bring together a group of astronomers who are interested in applying these new computational techniques to the field of Milky Way dynamics and structure, in order to facilitate the exchange and development of new ideas in this space. We believe 2024 will be the right time to hold such a workshop, as many new datasets have recently become available (e.g., Gaia Data Release 3) or will soon become available (e.g., the SDSS-V Milky Way Mapper), and as the application of tools from machine learning to Milky Way dynamics and structure has proliferated over the last few years. The goal of this Ringberg workshop is to inspire new collaborations, seeded by a mix of talks, breakout sessions, and unstructured discussions.