Mixing Bayesian Techniques for Effective Real-time Classification of Astronomical Transients

Abstract

With the recent advent of time domain astronomy through various surveys several approaches at classification of transient s are being tried. Choosing relatively interesting and rarer transients for follow-up is important since following all transients being detected per night is not possible given the limited resources available. In addition, the classification needs to be carried out using minimal number of observations available in order to catch some of the more interesting objects. We present details on two such classification methods: (1) using Bayesian networks with colors and contextual information, and (2) using Gaussian Process Regression and light-curves. Both can be carried out in real-time and from a very small number of epochs. In order to improve classification i.e. narrow down number of competing classes, it is important to combine as many different classifiers as possible. We mention how this can be accomplished using a higher order fusion network.