CDS&E: Machine Learning for Star Cluster Classification

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

This is a pilot project to compare and test human visual inspection versus machine learning (ML) and computer vision (CV) methods for identifying young star clusters (YSC) in high resolution images of galaxies. It is the first step towards exploring and optimizing the ML methods so as to build a tool capable of automatic search, classification, and shape measurement of YSC. This study will provide a launch-pad for the full project. It seems likely that ML tools will be the only viable way to handle the vast 'Big Data' databases becoming common in astronomy. An integrated educational component includes summer research for undergraduate students, including a valuable introduction to 'Big Data' issues. Initial tests will be performed on the two closest galaxies to our own Milky Way (M31 and M33), and then extended to M51 and NGC628, which are further away from us. These are well-studied galaxies for which high-fidelity catalogs already exist, which are available for comparison and calibration. The chosen test galaxies have very different cluster populations, and thus represent key testbeds to validate both the standard (human-based) approach and the future ML approach being developed. ML/CV algorithms to be explored and tested on these images include very deep convolutional neural networks, which will be adapted to provide collective classifications of star clusters. The human-based approach is currently the 'industry standard', and its validation will provide a more secure footing for future investigations of the physics of star formation in external galaxies. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.