Classification via Auxiliary Information: Formalism and Application to Classification of Astronomical Time Series

• Activity Number: 485
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Beatriz Etchegaray*+ and Chad Schafer and Peter Freeman
• Companies: and Carnegie Mellon Univeristy and Carnegie Mellon University
• Keywords: classification ; variable sources

Abstract

Upcoming astronomical surveys will require real-time classification of sources with varying brightness. Manual classification using spectroscopy has been the preferred method in Astronomy, but the large amounts of non-spectroscopic data of future surveys makes classification difficult. For each candidate source, we consider brightness measurements in time or light curves. Classification based on these light curves alone is challenging because the variability in the time series is related in complex ways to the underlying physical processes that generate the data. We propose a classification scheme that combines (1) the physical knowledge of the relationship between the type of object and spectroscopic information with (2) the noisy time series that we will observe. The underlying idea is that the spectra can accurately predict the type of source, thus we can learn auxiliary features from the spectra that can both accurately predict the type and can be well predicted by the light curves using standard regression techniques. The proposed methodology is illustrated with an analysis of simulated multiband light curves derived from real spectra using the filters of the LSST.