http://arxiv.org/abs/1907.04690
Upcoming astronomical surveys such as the Large Synoptic Survey Telescope (LSST) will rely on photometric classification to identify the majority of the transients and variables that they discover. We present a set of techniques for photometric classification that can be applied even when the training set of spectroscopically-confirmed objects is heavily biased towards bright, low-redshift objects. Using Gaussian process regression to model arbitrary light curves in all bands simultaneously, we “augment” the training set by generating new versions of the original light curves covering a range of redshifts and observing conditions. We train a boosted decision tree classifier on features extracted from the augmented light curves, and we show how such a classifier can be designed to produce classifications that are independent of the redshift distributions of objects in the training sample. Our classification algorithm was the best-performing among the 1,094 models considered in the blinded phase of the Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC), scoring 0.468 on the organizers’ logarithmic-loss metric with flat weights for all object classes in the training set, and achieving an AUC of 0.957 for classification of Type Ia supernovae. Our results suggest that future surveys relying on photometric classification should focus their spectroscopic efforts on typing large numbers of well-observed, intermediate redshift transients instead of attempting to type a sample of light curves that is directly representative of the full dataset. All of the algorithms described in this paper are implemented in the avocado software package.
K. Boone
Thu, 11 Jul 19
49/62
Comments: 24 pages, 9 figures
You must be logged in to post a comment.