Poster abstract details
Deep cadence-invariant classification of variable stars.
Abstract
During the last decade, considerable effort has been made to perform automatic classification of variable stars using machine learning techniques. Traditionally, light curves are represented as a vector of statistical descriptors or features used as input for many algorithms. Features are computationally expensive, cannot be updated quickly and for large datasets can take up to days. Its computation cannot scale for large-scale datasets, which is expected to obtain from the LSST. Previous work has been done to develop unsupervised feature extraction algorithms for light curves. In this work, we propose an algorithm to automatically learn a representation of light curves that is
invariant to the cadence of the observations. We propose a series of architectures based on Recurrent Neural Networks and test them in automatic classification scenarios. Our method uses minimal preprocessing, can be updated with a low computational cost for new data and can scale for more massive datasets. We test our method in two surveys: OGLE-III and the cross-match with VVV. We obtain accuracies +75% in the majority of subclasses and +95% in classes. We compare our results with the Random Forest algorithm in variable stars classification.
invariant to the cadence of the observations. We propose a series of architectures based on Recurrent Neural Networks and test them in automatic classification scenarios. Our method uses minimal preprocessing, can be updated with a low computational cost for new data and can scale for more massive datasets. We test our method in two surveys: OGLE-III and the cross-match with VVV. We obtain accuracies +75% in the majority of subclasses and +95% in classes. We compare our results with the Random Forest algorithm in variable stars classification.