Orateur
Description
Predictions suggest that the Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST) will discover ~six million Solar System objects (SSOs) in its decade of operations. The onset of activity in an SSO, such as coma or tail formation, signals a change in its physical environment that can also influence its orbit. Rapid identification of such activity is therefore essential both for understanding the physical processes driving SSO evolution and for improving orbital solutions through timely follow-up observations.
However, there are currently no dedicated tools designed to identify active SSOs within the LSST alert stream. I present ongoing work to develop a machine-learning pipeline that could potentially be part of the Fink broker ecosystem to flag candidate active SSOs. The approach combines anomaly detection with neural-network classifiers to identify alerts whose photometric or morphological features deviate from those of typical inactive asteroids. An active learning strategy is used to iteratively refine the model by prioritising uncertain or anomalous alerts for human labelling, enabling efficient training in a regime where labeled examples of active objects remain scarce.
This work aims to provide an early-warning capability for identifying active SSOs within the Fink alert stream, enabling rapid follow-up observations and improving our ability to study and track these dynamically evolving objects.