In the current Lambda-Cold-Dark-Matter paradigm of the Universe, galaxies, such as our own Milky Way, were formed through persistent merging with smaller systems in the last 14 billion years since the Big Bang. This vigorous merging history of the Milky Way is expected to leave detailed fossil records imprinted on the distribution of stars in the Milky Way. Thanks to the many technological breakthroughs in the last decade, such fossil records are now routinely observed through large-scale astronomical surveys. For example, the launch of the European Gaia mission, which provides us the positions and velocities for billions of stars, was a watershed moment for enabling the identification of these prehistoric clues of the Milky Way’s history. Cutting-edge hydrodynamic simulations further concur with this picture; these cosmological simulations have demonstrated how the merging history of galaxies can indeed leave distinct fingerprints on the galaxies: how disperse stars are, the shape of the Galactic “halo,” and the properties of various Galactic “streams” in the Galactic halo.
However, despite the promising prospects and abundance of data, most comparisons between the simulations and observations have remained qualitative. Even though we can now routinely “forward model” the Universe with cutting-edge hydrodynamic simulations and regularly observe different pieces of fossils in our quest to perform Galactic “Archaeology,” we still lack the robust statistical tool to make detailed “backward” inferences on the merging history of a galaxy. This project will explore machine learning to make “likelihood-free” inferences. We will analyze the cutting-edge cosmological simulations, study the relationship between the merging history and the output galaxy with ML, and then apply the ML model to the Gaia data. This project aims to rigorously (and probabilistically) quantify the merging history of the Milky Way.
Develop machine learning tools that can learn from cosmological simulations and infer the merging history of a galaxy, such as the Milky Way, through its observable.
Python programming (Pytorch, Tensorflow) and experience in machine learning and data science are essential.
machine learning, likelihood-free inference, astronomy