Logo StarEmbed

Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars

Weijian Li, Hong-Yu Chen†, Qinjie Lin, Nabeel Rehemtulla†, Ved G. Shah, Dennis Wu, Adam A. Miller, Han Liu,

†Corresponding to: charlie.chen@u.northwestern.edu, nabeelr@u.northwestern.edu,
arXiv 🤗 Dataset Code Leaderboard SkAI Institute Examples

Overview

StarEmbed is the first benchmark to test modern time series foundation models (TSFMs) on astronomical data. The ultimate goal of this project is to explore the potential of TSFMs in astronomical data analysis. These models are powerful tools trained on massive datasets, but until now they have never been evaluated on stellar light curves in a systematic way. For evaluation, we compare these models to traditional astrophysics methods that extracts light curve features on (1) embedding quality, (2) light curve classification, and (3) out-of-distribution detection.

StarEmbed Diagram

Contributions:

  • Benchmarking. StarEmbed is the first benchmark to test modern time series foundation models (TSFMs) on astronomical data.
  • Open Source. We publish a comprehensive evaluation framework and a large-scale dataset for benchmarking TSFMs on astronomical data.
  • Multidisciplinary. This project is done by a collaboration between the Astronomy and Computer Science departments at Northwestern University, promoting AI4Science in Astronomy.

News

🎉 [Oct 2025]
StarEmbed I. is released on arXiv!!!

🎉 [Oct 2025]
Nabeel presented “StarEmbed: Benchmarking time-series foundation models on unified datasets of variable star light curves” at SkAI’s Works in Progress series

🎉 [Oct 2025]
Hong-yu is going to present our project "StarEmbed-GPT: Toward a Foundation Model for General-purpose Inference on Variable Stars", at 2025 Open Accelerated Computing Summit.

🎉 [Sep 2025]
Hong-yu gives a talk at Open SkAI 2025, Chicago. "StarEmbed-GPT: Toward a foundation model for general-purpose inference on variable stars".

🎉 [Sep 2025]
Prof. Adam Miller is invited to give a keynote review: "A Billion Stars and Galaxies – Oh My! AstroAI Needs SkAI" at the 6th Global Research Platform Workshop , Chicago, IL.

🎉 [Sep 2025]
Weijian and Qinjie successfully obtain their PhD degree and graduate from Northwestern University! Congrats!! 🎓🎓🎓

🎉 [Aug 2025]
Prof. Adam Miller is invited to give a presentation: "Automating Discovery & Improving Classification With (and Without) Foundation Models" at Multimessenger Astronomy in the Era of Foundational AI, Vanderbilt University, Nashville, TN

🎉 [Jul 2025]
Nabeel is invited to sit on an AI-Astro panel at the Rutgers Summer Transients Soiree

🎉 [May 2025]
Weijian is invited to present “StarEmbed: A benchmark for evaluating pre-trained light-curve embeddings in variable star classification” at SkAI’s Works in Progress series.

🎉 [May 2025]
Nabeel is invited to speak at the Foundation Models for Astronomy conference at the Flatiron Institute’s Center for Computational Astrophysics.

🎉 [Jan 2025]
Prof. Adam Miller, Prof. Han Liu, and Weijian are invited to present “A Universal Forecaster for Astronomical Light Curves (and Other Out-of-domain Time-Series Data)” at SkAI Hub.

BibTeX


            @misc{li2025starembedbenchmarkingtimeseries,
      title={StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars}, 
      author={Weijian Li and Hong-Yu Chen and Qinjie Lin and Nabeel Rehemtulla and Ved G. Shah and Dennis Wu and Adam A. Miller and Han Liu},
      year={2025},
      eprint={2510.06200},
      archivePrefix={arXiv},
      primaryClass={astro-ph.SR},
      url={https://arxiv.org/abs/2510.06200}, 
}