Dear colleagues, as part of Rubin Prompt Data Product Pipeline, we have prototyped a Real/Bogus model based on convolutional neural networks, which was used to produce DP1 data and is currently in use to deliver the first alerts to the alert brokers. This model was presented at the Rubin session Update on Science Pipelines at the Rubin Community Workshop 2025. As noted in this session, the performance of the model is limited by the availability of accurate labels on real data and this causes a differential performance on different types of objects, and a particularly low performance on variable stars. To address this challenge, we are collecting an extensive dataset of stamps and issuing a Zooniverse campaign to collect labels. We are gathering the data during SV, and we expect to go public with the Zooniverse project in October/November 2025. We encourage the community to participate, and we will post updates here, including announcements of release date.
3 Likes