Astronomical Data Approximation Based on Neural Network Models
2021-10-28, 12:00–12:15, Grand Ballroom

Modern astronomical telescopes observe millions of transients every night. Light curves of these objects help to deduce their physical and phenomenological properties. A light curve is a flux time series measured in one or several photometric passbands. Each passband might have a different number of observations and various time gaps between them, thus complicating further analysis. To overcome these difficulties, we propose new methods for light curve approximation based on neural network models.

In this study, we use The Photometric LSST Astronomical Time Series Classification Challenge (PLAsTiCC) dataset. We apply shallow neural networks, bayesian neural networks, and normalizing flows to approximate observations for a single object and facilitate analysis of the light curves. We show that the approximation quality of the proposed methods significantly outperform the existing approaches based on gaussian processes. We also perform two physical analyses: supernovae type Ia classification and intensity peak estimation. For both these problems, convolutional neural networks are trained on approximated light curves. The results show that the proposed methods help to improve the quality of supernovae type identification and increase the accuracy of the intensity peak estimation compared with the gaussian processes model.


Understanding and improving machine learning