| Abstract: |
Machine learning algorithms provide powerful new windows into complicated and high-dimensional datasets. Of particular interests are normalizing flows, which estimate the phase space density of data in an unsupervised manner. These new techniques arrive concurrently with an era of Big Data in astrophysics, as large sky surveys provide a wealth of data about the structure of our own Galaxy and the Universe beyond. In this talk, I apply normalizing flows to Gaia data, which measures the position and velocity of the nearest 1.5 billion bright stars (30 million with full 6D kinematic! s). With the phase space density output from the flow algorithm, we can directly solve the Boltzmann Equation to obtain a model-free and unbinned acceleration and mass density fields within 3 kpc of the Sun. Our approach does not require symmetry assumptions, allowing novel tests of the equilibrium assumptions in the Boltzmann Equation. We find a local dark matter density of 0.47 GeV/cm^3, and a density profile that is consistent with a generalized NFW. Time permitting I will discuss other uses of flow architectures in Gaia data, including searching for stellar streams and generating synthetic astronomical datasets. |
