AstroAI Workshop 2025
Paul Gregory
A Novel and Scalable Transformer-based Classifier to Automatically Process Millions of TESS Light Curves
Presenter: Paul Gregory
Title: A Novel and Scalable Transformer-based Classifier to Automatically Process Millions of TESS Light Curves
Date/Time: Thursday, July 10th, 2:15 - 2:30 PM
Abstract: Photometric missions such as Kepler and TESS are generating millions of light curves spanning nearly the entire sky, offering unprecedented opportunities to study stellar variability and advance our understanding of the universe. In this data-rich environment, machine learning has emerged as a powerful tool to efficiently and accurately process these light curves and classify them by variability. In this work, we introduce a novel Transformer-based architecture for variability classification that also integrates bi-directional LSTMs and CNNs to encode the light curve. By directly processing time series data without requiring feature engineering, this approach simplifies the analysis pipeline. Trained on Kepler and TESS data, the model achieves 94.26% accuracy on a Kepler holdout set and 89.48% accuracy on TESS. Training on both data sets notably improves our accuracy, demonstrating the model’s ability to scale performance with larger and more diverse training data. Our results illustrate the model’s robustness and accuracy, highlighting the effectiveness of deep learning for variable classification. In this talk I will be presenting our novel methodology as well as our current work deploying the model to deliver stellar variability classifications for millions of stars.