Speaker
Description
The Rubin Observatory is detecting an unprecedented number of transients each night, exceeding the classification capacity of all existing spectroscopic resources my several orders of magnitude. As a result, robust photometric classifications will be essential both for assembling complete samples of different transient subtypes and for identifying events that warrant spectroscopic follow-up.
In this talk, we introduce a family of highly performant, real-time hierarchical classifiers designed for ZTF and LSST alert streams. These models incorporate images, multi-band light curves, and contextual metadata, to deliver reliable, high-level classifications within seconds of the first alert. Our hierarchical approach is unique in its ability to make classifications at several levels of granularity based on the available information, making it possible, for the first time, to produce high-level classifications at early times, and refine those classifications as more observations become available.
Our models achieve strong performance on real ZTF data, reaching >95% accuracy on binary Transient vs Persistent classification after the first day. For a more granular task distinguishing CVs, AGNs, and supernova subtypes, our models achieve >85% accuracy 128 days after first detection with our multimodal model delivering performance that far exceeds the light-curve only models, especially at early times. On the ELAsTiCC datasets of simulated LSST observations, we achieve state-of-the-art performance with a macro F1 score of 0.88. These results demonstrate that multi-modal, hierarchical classifiers can deliver classifications at LSST scale, supporting both real-time triaging and population studies from the earliest alerts.