Predicting Young Stellar Ages with Machine Learning

Studying the environments of star forming regions is critical to our understanding of early stellar evolution, but historically, distinguishing young stars from evolved field population has been difficult. Gaia DR2 now allows for vastly improved membership constraints of young populations through spatial and kinematical clustering, However, clustering cannot identify kinematically peculiar stars, such as those that have been ejected through n-body interactions. Stellar ages, which can be extrapolated on an individual basis from photometric measurements, are a more general alternative to astrometric clustering in identifying cluster membership. An alternate way to identify members of star forming regions is through extrapolation of the ages from the position of the stars on the HR diagram. This process is computationally challenging, not in the least due to mismatch between the theoretical pre-main sequence isochrones and the real data. We present a deep learning method using Gaia DR2 and 2MASS photometry to predict stellar ages for the pre-main-sequence stars. Our model, consisting of two cascading convolution neural networks, is trained to first identify young stars from the field based on the empirical training sample of populations with known ages. After classifying the evolutionary stage of input stars, the model predicts ages for all sources categorized as pre-main-sequence. Through this analysis, we explore the age gradients in the nearby star forming regions.