StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars

Weijian Li, Hong-Yu Chen, Nabeel Rehemtulla, Ved G. Shah, Dongho Kim, Dennis Wu, Qinjie Lin, Adam A. Miller, Han Liu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:66289-66329, 2026.

Abstract

Current time series foundation model (TSFM) training corpora largely omit data with certain complexities like irregular temporal sampling. Astronomical time series of stellar fluxes (“light curves”) are available in immense quantities and exhibit irregular sampling, multiple variates, and heteroskedasticity. We introduce $\texttt{StarEmbed}$, the first public benchmark for light curves comprised of real observations of $\sim$40,000 stars across seven classes and evaluations in clustering, classification, and out-of-distribution (OOD) source detection. We benchmark TSFMs with differing architecture and training strategies as well as domain-specific transformers. Our results demonstrate that the $\texttt{Chronos}$ family, despite being pre-trained on regularly sampled non-astronomical data, yields state-of-the-art (SOTA) performance in light curve clustering and OOD detection. While no TSFM strictly surpasses the classification performance of the long-established domain baseline, they do demonstrate excellent generalization abilities. $\texttt{StarEmbed}$ marks a step toward universal light curve embeddings and improved TSFM performance on challenging data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-li26t, title = {{S}tar{E}mbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars}, author = {Li, Weijian and Chen, Hong-Yu and Rehemtulla, Nabeel and Shah, Ved G. and Kim, Dongho and Wu, Dennis and Lin, Qinjie and Miller, Adam A. and Liu, Han}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {66289--66329}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/li26t/li26t.pdf}, url = {https://proceedings.mlr.press/v306/li26t.html}, abstract = {Current time series foundation model (TSFM) training corpora largely omit data with certain complexities like irregular temporal sampling. Astronomical time series of stellar fluxes (“light curves”) are available in immense quantities and exhibit irregular sampling, multiple variates, and heteroskedasticity. We introduce $\texttt{StarEmbed}$, the first public benchmark for light curves comprised of real observations of $\sim$40,000 stars across seven classes and evaluations in clustering, classification, and out-of-distribution (OOD) source detection. We benchmark TSFMs with differing architecture and training strategies as well as domain-specific transformers. Our results demonstrate that the $\texttt{Chronos}$ family, despite being pre-trained on regularly sampled non-astronomical data, yields state-of-the-art (SOTA) performance in light curve clustering and OOD detection. While no TSFM strictly surpasses the classification performance of the long-established domain baseline, they do demonstrate excellent generalization abilities. $\texttt{StarEmbed}$ marks a step toward universal light curve embeddings and improved TSFM performance on challenging data.} }
Endnote
%0 Conference Paper %T StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars %A Weijian Li %A Hong-Yu Chen %A Nabeel Rehemtulla %A Ved G. Shah %A Dongho Kim %A Dennis Wu %A Qinjie Lin %A Adam A. Miller %A Han Liu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-li26t %I PMLR %P 66289--66329 %U https://proceedings.mlr.press/v306/li26t.html %V 306 %X Current time series foundation model (TSFM) training corpora largely omit data with certain complexities like irregular temporal sampling. Astronomical time series of stellar fluxes (“light curves”) are available in immense quantities and exhibit irregular sampling, multiple variates, and heteroskedasticity. We introduce $\texttt{StarEmbed}$, the first public benchmark for light curves comprised of real observations of $\sim$40,000 stars across seven classes and evaluations in clustering, classification, and out-of-distribution (OOD) source detection. We benchmark TSFMs with differing architecture and training strategies as well as domain-specific transformers. Our results demonstrate that the $\texttt{Chronos}$ family, despite being pre-trained on regularly sampled non-astronomical data, yields state-of-the-art (SOTA) performance in light curve clustering and OOD detection. While no TSFM strictly surpasses the classification performance of the long-established domain baseline, they do demonstrate excellent generalization abilities. $\texttt{StarEmbed}$ marks a step toward universal light curve embeddings and improved TSFM performance on challenging data.
APA
Li, W., Chen, H., Rehemtulla, N., Shah, V.G., Kim, D., Wu, D., Lin, Q., Miller, A.A. & Liu, H.. (2026). StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:66289-66329 Available from https://proceedings.mlr.press/v306/li26t.html.

Related Material