Identifying Scientifically Interesting Galaxies with Vision-Language Models
Abstract
Anomaly searches in large astronomical image datasets aim to identify rare objects for follow-up investigation. Yet the objects that are most statistically unusual are not necessarily those of greatest scientific relevance. We investigate whether pretrained vision-language models can directly assess scientific interest from natural-language instructions, enabling galaxy images to be ranked without labeled examples. We evaluate models on a 20,000-image Hubble Space Telescope sample that includes 167 reference anomalies previously identified through machine learning and expert inspection. By scoring images for scientific interest, Gemini 3.1 Flash-Lite places 98.2% of the reference anomalies within the highest-ranked ~2,000 images, at a present-day inference cost of approximately $5.05 per 10,000 images. We recover known gravitational lenses, jellyfish-galaxy candidates, and galaxies with visible jets, and release 102 high-scoring candidates for further investigation. These results show that an off-the-shelf vision-language model, without any training on the target data, offers a practical way to rank large image datasets by scientific relevance rather than rarity. The project code is available at https://github.com/TheCheng2005/Galaxy-Anomaly-Search/
Related
- Vision-Language Model Ensembles Achieve Human-Expert Accuracy for Galaxy Merger Classification
- SN2025aico: An Interesting Case Of 56Ni Mixing, Ejecta Asymmetries, and Dust Formation in a Type IIb Supernova
- Identifying heating processes in simulations with an entropy-based scheme: A single jet episode in a galaxy cluster
- On the mean profiles of radio pulsars. III. Identifying the mode
- A catalog of Gaia DR3 luminous blue variables and candidate luminous blue variables. Limits of Gaia DR3 light curves for identifying luminous blue variables
BibTeX
@misc{cheng2026identifying,
title = {{Identifying Scientifically Interesting Galaxies with Vision-Language Models}},
author = {Yixuan Cheng and Nolan Koblischke and Jo Bovy},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.37747},
eprint = {2610.37747},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.37747},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Identifying Scientifically Interesting Galaxies with Vision-Language Models
AU - Yixuan Cheng
AU - Nolan Koblischke
AU - Jo Bovy
PY - 2026
DA - 2026/10/27
PB - arXiv
DO - 10.48550/arxiv.2610.37747
UR - https://arxiv.org/abs/2610.37747
AB - Anomaly searches in large astronomical image datasets aim to identify rare objects for follow-up investigation. Yet the objects that are most statistically unusual are not necessarily those of greatest scientific relevance. We investigate whether pretrained vision-language models can directly assess scientific interest from natural-language instructions, enabling galaxy images to be ranked without labeled examples. We evaluate models on a 20,000-image Hubble Space Telescope sample that includes 167 reference anomalies previously identified through machine learning and expert inspection. By scoring images for scientific interest, Gemini 3.1 Flash-Lite places 98.2% of the reference anomalies within the highest-ranked ~2,000 images, at a present-day inference cost of approximately $5.05 per 10,000 images. We recover known gravitational lenses, jellyfish-galaxy candidates, and galaxies with visible jets, and release 102 high-scoring candidates for further investigation. These results show that an off-the-shelf vision-language model, without any training on the target data, offers a practical way to rank large image datasets by scientific relevance rather than rarity. The project code is available at https://github.com/TheCheng2005/Galaxy-Anomaly-Search/
ER -