NOMAI : A real-time photometric classifier for superluminous supernovae identification

E. Russeil ,, R. Lunnan, J. Peloton, S. Schulze, P. J. Pessi, D. Perley, J. Sollerman, A. Gkini, Y. Hu, T.-W. Chen, E. C. Bellm, T. X. Chen, B. Rusholme · 2026

Read NOMAI : A real-time photometric classifier for superluminous supernovae identification by E. Russeil ,, R. Lunnan, J. Peloton, S. Schulze, P. J. Pessi, D. Perley, J. Sollerman, A. Gkini, Y. Hu, T.-W. Chen, E. C. Bellm, T. X. Chen, B. Rusholme (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-14761.txt.

Introduction

Russeil and 12 others built NOMAI, a classifier that spots superluminous supernova candidates from photometric alerts alone, live inside the Fink broker. It works from ZTF light curves spanning at least 30 days, using features fitted with SALT2 and the blackbody-based Rainbow framework, and needs no spectroscopic redshift. Trained on 5,280 labeled sources including 225 confirmed superluminous supernovae, it reached 66 percent completeness and 58 percent purity. Running nightly since December 18, 2025, it recovered 22 of 24 active superluminous supernovae in its first two months, and adaptation to the Rubin Observatory survey is planned.

Catalog record

CREATORE. Russeil ,, R. Lunnan, J. Peloton, S. Schulze, P. J. Pessi, D. Perley, J. Sollerman, A. Gkini, Y. Hu, T.-W. Chen, E. C. Bellm, T. X. Chen, B. Rusholme
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56919
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.14761

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

NOMAI : A real-time photometric classifier for superluminous supernovae identification E. Russeil ,, R. Lunnan, J. Peloton, S. Schulze, P. J. Pessi, D. Perley, J. Sollerman, A. Gkini, Y. Hu, T.-W. Chen, E. C. Bellm, T. X. Chen, B. Rusholme Abstract Superluminous supernovae (SLSNe) are one of the most luminous stellar explosions known, yet they remain poorly understood. Because they are intrinsically rare, efficiently identifying them in the large alert streams produced by modern time-domain surveys is essential for building larger observational samples and enabling spectroscopic follow-up. We present NOMAI , a machine learning classifier designed to identify SLSN candidates directly from photometric alerts in the ZTF stream, using light curves accumulated over at least 30 days. It does not require any spectroscopic redshift and is running in real time within the Fink broker. ZTF light curves are transformed into a set of physically motivated features derived primarily from model-fitting procedures using SALT2 and Rainbow, a blackbody-based multi-band fitting framework. These features are used to train an XGBoost classifier on a curated dataset of labeled ZTF sources constructed using literature samples of SLSNe, along with TNS and internal ZTF labeled sources. The final training dataset contains 5280 unique sources, including 225 spectroscopically classified SLSNe. On the training sample, the classifier reaches 66% completeness and 58% purity. Deployed within the Fink broker, NOMAI has been running continuously since 18/12/2025 on the ZTF alert stream and publicly reports SLSN candidates every night by automatically posting them to dedicated communication channels. Based on this, we also report the first two-month as an evaluation period, where the classifier successfully recovered 22 of the 24 active SLSNe reported on the Transient Name Server. The achieved performances, particularly the high completeness in real-time operations, demonstrate that the classifier provides a valuable tool for experts to efficiently scan the alert stream and identify promising candidates. In the near future, NOMAI is intended to be adapted to operate on the Legacy Survey of Space and Time conducted by the Vera C. Rubin Observatory, which is expected to uncover an unprecedented number of transients, making machine learning based photometric classification essential. A science module for the Fink broker 1 Introduction Superluminous supernovae (SLSNe) are rare stellar explosions whose peak luminosities exceed those of standard SNe by one to two orders of magnitude ( 66 ; 26 ; 27 ) . Since their identification as a distinct population ( 66 ) , SLSNe have attracted considerable attention due to their extreme energetics. Although they display very diverse photometric properties, their light curves are typically evolving more slowly than those of normal SNe and often display blue colors ( 50 ; 1 ; 16 ; 13 ; 32 ; 62 , e.g.,) . Because of their high luminosities, they are detectable at large cosmological distances, making them valuable probes for the high-redshift universe. Despite significant observational and theoretical progress over the past decade, the physical mechanisms powering these explosions remain debated. Several scenarios have been proposed, including magnetar spin-down ( 58 ; 44 ; 80 ; 76 ) , interaction with dense circumstellar material ( 11 ; 71 ; 77 ; 14 ) , long-term fallback accretion of material onto a black hole ( 19 ; 56 ) , or pair-instability explosions ( 22 ; 3 ; 67 ; 79 ; 36 ) . Further constraining these models requires larger samples of well-observed events spanning a broad range of environments and redshifts. However, SLSNe are intrinsically rare ( 64 ; 60 ; 25 , e.g.,) and only represent a small fraction of the overall supernova population. Identifying them efficiently in wide-field surveys is therefore a key step toward improving our understanding of these extreme transients. Modern time-domain surveys are now producing unprecedented volumes of transient alerts. The Zwicky Transient Facility ( 5 ; 34 ; 18 ; 51 , ZTF, ) currently delivers hundreds of thousands of alerts per night, and the recent Vera C. Rubin Observatory will increase this number to ten million through the Legacy Survey of Space and Time ( 42 , LSST, ) . In this context, the classical approach of spectroscopic follow-up for every candidate becomes impossible. Only a small fraction of transient discoveries can realistically be spectroscopically classified, making automated photometric classification an essential part of modern astronomy. The alert streams produced by modern surveys require dedicated infrastructures capable of processing and distributing transient classifications in real time. Alert brokers ( 57 ; 54 ; 69 ; 52 ; 78 ; 43 ) 1 1 1 https://pitt-broker.readthedocs.io/en/latest/# have emerged to fulfill this role by ingesting survey alerts, enriching them with additional information, and providing value-added products to the community. In this context, machine-learning classifiers can naturally be integrated within alert brokers to operate directly on the incoming data stream, and multiple works have explored the specific task of classifying SLSN light curves, although some classification pipelines are still developed and evaluated outside broker infrastructures. Some approaches rely on multi-class classifiers in which SLSNe constitute one of several transient classes, such as the ALeRCE data broker ( 69 ) , or Superphot+ ( 17 ) running on the ANTARES broker ( 52 ) , both of which nightly classifies alerts from the ZTF data stream. Other pipelines are specifically designed to identify SLSN events among the broader population of transients. The ”NEural Engine for Discovering Luminous Events” ( 70 , NEEDLE,) , hosted within the Lasair broker ( 78 ) , uses a hybrid neural network to process both the photometry and the telescope images to classify SLSN and TDE sources. ”Finding Luminous and Exotic Extragalactic Transients” ( 29 ; 28 , FLEET, ) attempts to classify SLSNe by relying on host galaxy association and photometric feature extraction to train a random forest classifier. Finally, by combining SuperRAENN ( 75 ) , a recurrent autoencoder architecture, with Superphot ( 74 ) , a versatile feature-extraction pipeline based on light-curve model fitting, 40 developed a classifier designed to identify SLSNe in Pan-STARRS1 data ( 72 ) . Furthermore, other works have focused on anticipating the LSST era using simulated datasets such as the Photometric LSST Astronomical Time-series Classification Challenge ( 38 , PLAsTiCC,) to train SLSN classification models ( 65 ; 9 ; 48 ) . Similarly, more recent works using the Extended LSST Astronomical Time-Series Classification Challenge ( 46 , ELAsTiCC,) have explored analogous problems using a more realistic LSST-like data stream format ( 23 ; 10 ) . Among the brokers, Fink ( 54 ) has been operating on the ZTF public stream since late 2019, while also helping prepare for the upcoming Rubin era. It provides an infrastructure that allows the scientific community to deploy dedicated “science modules”, which operate on the alert stream to compute additional features and classifications. This framework makes it possible to run machine-learning models continuously on the incoming alerts and distribute their outputs in near real time. In this work, we present NOMAI , a machine-learning classifier designed to identify SLSN candidates directly from ZTF photometric alerts and deploy it within the Fink broker as a science module. The classifier relies on physically motivated features extracted from transient light curves without requiring a spectroscopic redshift, and is optimized to

Read the complete text