Real-time detection of anomalies in large-scale transient surveys

• DOI: 10.1093/mnras/stac2582
• Publication Date: 2022-09-20T11:11:40Z
• Authors: Daniel Muthukrishna; Kaisey S Mandel; Michelle Lochner; Sara Webb; Gautham Narayan
• Keywords: Transient (computer programming)

ABSTRACT New time-domain surveys, such as the Vera C. Rubin Observatory Legacy Survey of Space and Time, will observe millions transient alerts each night, making standard approaches visually identifying new interesting transients infeasible. We present two novel methods automatically detecting anomalous light curves in real-time. Both are based on simple idea that if from a known population can be accurately modelled, any deviations model predictions likely anomalies. The first modelling approach is probabilistic neural network built using Temporal Convolutional Networks (TCNs) second an interpretable Bayesian parametric transient. demonstrate our methods’ ability to provide anomaly scores function time Zwicky Transient Facility. show flexibility networks, attribute makes them powerful tool for many regression tasks, what less suitable detection when compared with model. able identify anomalies respect common supernova classes high precision recall scores, achieving area under precision-recall above 0.79 most rare kilonovae, tidal disruption events, intermediate luminosity transients, pair-instability supernovae. Our improves over lifetime curves. framework, used conjunction classifiers, enable fast prioritized followup unusual large-scale surveys.