Applying multimodal learning to Classify transient Detections Early (AppleCiDEr) I: Data set, methods, and infrastructure

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Junell, Alexandra, Sasli, Argyro, Nunes, Felipe Fontinele, Xu, Maojie, Border, Benny, Rehemtulla, Nabeel, Rizhko, Mariia, Qin, Yu-Jing, Laz, Theophile Jegou Du, Calloch, Antoine Le, Chaudhary, Sushant Sharma, Wu, Shaowei, Sollerman, Jesper, Sravan, Niharika, Groom, Steven L., Hale, David, Kasliwal, Mansi M., Purdum, Josiah, Wold, Avery, Graham, Matthew J., Coughlin, Michael W.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908463548858368
author Junell, Alexandra
Sasli, Argyro
Nunes, Felipe Fontinele
Xu, Maojie
Border, Benny
Rehemtulla, Nabeel
Rizhko, Mariia
Qin, Yu-Jing
Laz, Theophile Jegou Du
Calloch, Antoine Le
Chaudhary, Sushant Sharma
Wu, Shaowei
Sollerman, Jesper
Sravan, Niharika
Groom, Steven L.
Hale, David
Kasliwal, Mansi M.
Purdum, Josiah
Wold, Avery
Graham, Matthew J.
Coughlin, Michael W.
author_facet Junell, Alexandra
Sasli, Argyro
Nunes, Felipe Fontinele
Xu, Maojie
Border, Benny
Rehemtulla, Nabeel
Rizhko, Mariia
Qin, Yu-Jing
Laz, Theophile Jegou Du
Calloch, Antoine Le
Chaudhary, Sushant Sharma
Wu, Shaowei
Sollerman, Jesper
Sravan, Niharika
Groom, Steven L.
Hale, David
Kasliwal, Mansi M.
Purdum, Josiah
Wold, Avery
Graham, Matthew J.
Coughlin, Michael W.
contents Modern time-domain surveys like the Zwicky Transient Facility (ZTF) and the Legacy Survey of Space and Time (LSST) generate hundreds of thousands to millions of alerts, demanding automatic, unified classification of transients and variable stars for efficient follow-up. We present AppleCiDEr (Applying Multimodal Learning to Classify Transient Detections Early), a novel framework that integrates four key data modalities (photometry, image cutouts, metadata, and spectra) to overcome limitations of single-modality classification approaches. Our architecture introduces (i) two transformer encoders for photometry, (ii) a multimodal convolutional neural network (CNN) with domain-specialized metadata towers and Mixture-of-Experts fusion for combining metadata and images, and (iii) a CNN for spectra classification. Training on ~ 30,000 real ZTF alerts, AppleCiDEr achieves high accuracy, allowing early identification and suggesting follow-up for rare transient spectra. The system provides the first unified framework for both transient and variable star classification using real observational data, with seamless integration into brokering pipelines, demonstrating readiness for the LSST era.
format Preprint
id arxiv_https___arxiv_org_abs_2507_16088
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Applying multimodal learning to Classify transient Detections Early (AppleCiDEr) I: Data set, methods, and infrastructure
Junell, Alexandra
Sasli, Argyro
Nunes, Felipe Fontinele
Xu, Maojie
Border, Benny
Rehemtulla, Nabeel
Rizhko, Mariia
Qin, Yu-Jing
Laz, Theophile Jegou Du
Calloch, Antoine Le
Chaudhary, Sushant Sharma
Wu, Shaowei
Sollerman, Jesper
Sravan, Niharika
Groom, Steven L.
Hale, David
Kasliwal, Mansi M.
Purdum, Josiah
Wold, Avery
Graham, Matthew J.
Coughlin, Michael W.
Instrumentation and Methods for Astrophysics
High Energy Astrophysical Phenomena
Modern time-domain surveys like the Zwicky Transient Facility (ZTF) and the Legacy Survey of Space and Time (LSST) generate hundreds of thousands to millions of alerts, demanding automatic, unified classification of transients and variable stars for efficient follow-up. We present AppleCiDEr (Applying Multimodal Learning to Classify Transient Detections Early), a novel framework that integrates four key data modalities (photometry, image cutouts, metadata, and spectra) to overcome limitations of single-modality classification approaches. Our architecture introduces (i) two transformer encoders for photometry, (ii) a multimodal convolutional neural network (CNN) with domain-specialized metadata towers and Mixture-of-Experts fusion for combining metadata and images, and (iii) a CNN for spectra classification. Training on ~ 30,000 real ZTF alerts, AppleCiDEr achieves high accuracy, allowing early identification and suggesting follow-up for rare transient spectra. The system provides the first unified framework for both transient and variable star classification using real observational data, with seamless integration into brokering pipelines, demonstrating readiness for the LSST era.
title Applying multimodal learning to Classify transient Detections Early (AppleCiDEr) I: Data set, methods, and infrastructure
topic Instrumentation and Methods for Astrophysics
High Energy Astrophysical Phenomena
url https://arxiv.org/abs/2507.16088