Machine Learning Techniques Meet Binaries

Abstract

We briefly review the various machine learning methods and discuss how they can be used in efficient identification and analysis of spectroscopic binary stars. They can be treated as complementary to conventional methods, and we argue that some amount of human oversight is always needed and in fact highly beneficial when employing machine learning. We propose that a general dimensionality reduction technique can serve to diagnose and classify a given data set, and in case of GALAH spectra, our method quite effectively reveals a population of SB2 and SB3 systems. Once identified, the binary spectra can be analysed with the help of generative models, which can be constructed using machine learning techniques such as The Cannon and The Payne. Furthermore, in the case of spectroscopically unresolved multiple stars, we can recover the multiple contributions to an observed spectrum by reversing the process and proceeding from analysis to identification. Show more