Searching for Changing-state AGNs in Massive Data Sets. I. Applying Deep Learning and Anomaly-detection Techniques to Find AGNs with Anomalous Variability Behaviors

The classic classification scheme for Active Galactic Nuclei (AGNs) was recently challenged by the discovery of so-called changing-state (changing-look) AGNs (CSAGNs). physical mechanism behind this phenomenon is still a matter open debate and samples are too small serendipitous nature to provide robust answers. In order tackle problem, we need design methods that able detect AGN right in act changing-state. Here present an anomaly detection (AD) technique designed identify light curves with anomalous behaviors massive datasets. main aim CSAGN at different stages transition, but it can also be used more general purposes, such as cleaning datasets variability analyses. We from Zwicky Transient Facility data release 5 (ZTF DR5), containing sample 230,451 classes. ZTF DR5 were modeled Variational Recurrent Autoencoder (VRAE) architecture, allowed us obtain set attributes VRAE latent space describes behaviour our sample. These then features Isolation Forest (IF) algorithm, detector "one class" kind problem. reconstruction errors IF score select 8,809 anomalies. anomalies dominated bogus candidates, 75 promising candidates.