Astronomy Colloquium

  Despite incredible maturity, physical stellar models do not match all the detailed properties of stellar spectra, and different models compared to different sections of stellar spectra return different stellar parameter and abundance estimates.  At the same time, we have enormous data sets containing high-resolution, high signal-to-noise spectra of tens of thousands of stars.  I discuss ways to build data-driven stellar models, permitting stellar parameter and chemical abundance estimation without any explicit use of physical models.  I will show successes with the SDSS-III APOGEE data.  Time permitting, I will also talk about data-driven approaches in other areas, such as the search for exoplanets in Kepler and K2 data.