From stellar light to astrophysical insight: automating variable star research with machine learning

TL;DR

This paper reviews how machine learning automates the analysis of large stellar variability datasets from space missions, enabling new scientific discoveries in asteroseismology and exoplanet research.

Contribution

It provides a comprehensive overview of recent machine learning advances applied to stellar variability, from data cleaning to classification and parameter inference.

Findings

Machine learning enhances automated variability classification.

Representation learning improves feature extraction from stellar data.

Foundation models open new avenues for time-domain astronomy.

Abstract

Large-scale photometric surveys are revolutionizing astronomy by delivering unprecedented amounts of data. The rich data sets from missions such as the NASA Kepler and TESS satellites, and the upcoming ESA PLATO mission, are a treasure trove for stellar variability, asteroseismology and exoplanet studies. In order to unlock the full scientific potential of these massive data sets, automated data-driven methods are needed. In this review, I illustrate how machine learning is bringing asteroseismology toward an era of automated scientific discovery, covering the full cycle from data cleaning to variability classification and parameter inference, while highlighting the recent advances in representation learning, multimodal datasets and foundation models. This invited review offers a guide to the challenges and opportunities machine learning brings for stellar variability research and how it could help unlock new frontiers in time-domain astronomy.