Agentic Active Learning Meets Visual Embeddings: Finding Anomalies among 370 000 Variable Stars from ASAS-SN

Abstract

Unusual light-curve morphologies can point to rare physical configurations or new phenomena, but automatic searches for anomalies are often dominated by artifacts. Separating genuine anomalies from false positives has traditionally required manual vetting, which does not scale to modern surveys. We present an active learning framework for detecting anomalies in samples of periodic variable stars, with the vetting delegated to multimodal large language model agents. The initial ranking comes from…