Active anomaly detection for time-domain discoveries

We present the first evidence that adaptive learning techniques can boost discovery of unusual objects within astronomical light curve data sets. Our method follows an active strategy where algorithm chooses which potentially improve learner if additional information about them is provided. This new subsequently used to update machine model, allowing its accuracy evolve with each information. For case anomaly detection, aims maximize number scientifically interesting anomalies presented expert by slightly modifying weights a traditional Isolation Forest (IF) at iteration. In order demonstrate potential such techniques, we apply Active Anomaly Discovery (AAD) 2 sets: simulated curves from PLAsTiCC challenge and real Open Supernova Catalog. compare AAD results those static IF. both methods, performed detailed analysis for all ~2% highest scores. show that, in scenario, was able identify ~80\% more true than result algorithms play central role search physics era large scale sky surveys.