Variable Star Classification with a Multiple-input Neural Network

In this experiment, we created a Multiple-Input Neural Network, consisting of Convolutional and Multi-layer Networks. With setup the selected highest-performing neural network was able to distinguish variable stars based on visual characteristics their light curves, while taking also into account additional numerical information (e.g. period, reddening-free brightness) differentiate visually similar curves. The trained tested OGLE-III data using all observation fields, phase-folded curves period data. yielded accuracies 89--99\% for most main classes (Cepheids, $\delta$ Scutis, eclipsing binaries, RR Lyrae stars, Type-II Cepheids), only first-overtone Anomalous Cepheids had an accuracy 45\%. To counteract large confusion between RRab added brightness as new input from LMC field were retained have fixed distance. change improved network's result almost 80\%. Overall, Multiple-input Network method developed by our team is promising alternative existing classification methods.