Statistical and Machine Learning for Petascale Astronomy

Following the success of the pioneering Sloan Digital Sky Survey, the astronomical community has made significant, new investments in large, photometric surveys. Currently, one of the most important such surveys is the Dark Energy Survey, an international collaboration that is mapping a large fraction of the Southern sky in multiple bands and whose data are processed and archived at NCSA. In the next decade, the petascale Large Synoptic Survey Telescope effort, which was recently ranked as the highest priority upcoming astronomy project, will chart the Universe with unprecedented reach and detail. But the ability of the astronomical community to fully capitalize on these rich data will be limited unless we transform our approach to precisely extract all information from these massive data. By developing and accelerating new statistical and machine learning approaches, we are generating more accurate, probabilistic source classification algorithms, which greatly increases the size of the stellar samples used to map the structure of the Milky Way galaxy and the Local Group, and the size and purity of galaxy samples used to probe the structures that make up the Cosmos.