StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars
#StarEmbed #TSFMs #variable stars #time series benchmarking #irregular sampling #heteroskedasticity #peta-scale data
📌 Key Takeaways
- StarEmbed is the first public benchmark focused on TSFMs evaluated on astronomical time series.
- Astronomical observations produce large, irregularly-sampled, heteroskedastic data sets that TSFMs have not been trained on.
- The benchmark provides standardized evaluation across variable star light‑curve tasks.
- The work was released as arXiv:2510.06200v3 (2025).
📖 Full Retelling
This paper introduces **StarEmbed**, a publicly available benchmark for evaluating Time Series Foundation Models (TSFMs) on astronomical data. The authors present the benchmark to address the lack of astronomical time series in the large corpora traditionally used to train TSFMs, noting that stellar observations yield *peta*-scale data characterized by irregular sampling and heteroskedastic noise. Published in the 2025 October revision of arXiv (v3), the work demonstrates how StarEmbed can rigorously measure model performance across a range of variable star light‑curve tasks, thereby providing a standardized evaluation platform for the astronomical community.
🏷️ Themes
Time series foundation models, Astronomical data, Variable stars, Benchmarking, Irregular sampling, Heteroskedasticity
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
arXiv:2510.06200v3 Announce Type: replace-cross
Abstract: Time series foundation models (TSFMs) are increasingly being adopted as highly-capable general-purpose time series representation learners. Although their training corpora are vast, they exclude astronomical time series data. Observations of stars produce peta-scale time series with unique challenges including irregular sampling and heteroskedasticity. We introduce StarEmbed, the first public benchmark for rigorous and standardized evalu
Read full article at source