Characterising extragalactic Pre-Main-Sequence Stars with Machine and Deep Learning Techniques

Victor Ksoll

Poster -- Star-forming regions

Deep photometric surveys with the Hubble Space Telescope, such as the Hubble Tarantula Treasury Project (HTTP), which covers the region of 30 Doradus down to the half solar mass limit, and Measuring Young Stars in Space and Time (MYSST), covering the entire bubble of the star forming complex N44, provide unprecedented coverage of entire active star forming regions. We use the deep stellar catalogues of HTTP and MYSST to identify all the pre-main-sequence (PMS) stars of the respective regions. The photometric distinction of these stars from more evolved populations is, however, not a trivial task due to several factors that alter their colour-magnitude diagram positions. To overcome this hurdle, we employ Machine Learning Classification techniques, including Random Forests and Support Vector Machines (SVM), on the HTTP and MYSST surveys to unveil their PMS stellar content. Our methodology consists of 1) carefully selecting the most probable low-mass PMS stellar population of a prominent star forming cluster within the observed fields, 2) using these samples to train classification algorithms to build predictive models for PMS stars and 3) applying these models to identify the most probable PMS content across the entire observed regions. We further develop an Invertible Neural Network (INN) in order to predict the fundamental physical parameters of age and mass of the identified young stars and evaluate the spatial variations of these parameters across the entire star forming complexes.