2024 workshop program
Improving Stellar Age Estimations with Normalizing Flows
The work
Abstract
Understanding the ages of stars is crucial for unraveling the history and evolution of our galaxy. Current methods for estimating stellar ages from the spectra often struggle with providing precise uncertainties and are limited by the parameter space covered by the training data. This work introduces a novel approach using normalizing flows, a type of deep generative model, to estimate stellar ages with a focus on improving the accuracy and descriptive power of uncertainties. Unlike standard convolutional neural network techniques, normalizing flows enable the recovery of likelihood distributions for the ages of individual stars, offering a richer and more informative perspective on uncertainty. This method not only yields age estimations for a wide array of stars but also intrinsically accounts for the coverage and density of the training data, ensuring that the resulting uncertainties are reflective of both the inherent noise in the data and its representativeness. To expand the training parameter space, we have incorporated age determinations from cluster stars and are exploring the use of mixing models for stars in the upper giant branch. Using this method with the upcoming milky way mapper data release we hope to have the most comprehensive stellar age catalog for galactic archaeology produced.