Аннотация:The discovery of new astrophysical phenomena, as well as the study of already known transients, are among the mostexpected results of the next generation of large-scale astronomical sky surveys. Prompt data processing using modernmachine learning techniques allows us to quickly classify objects and provides information for their further study. For thelast three years the SNAD team is working on developing a pipeline where human expertise and modern machine learningalgorithms complement each other in the task of identifying unusual astronomical objects. In this paper, we applied theactive anomaly detection algorithm to the Zwicky Transient Facility Data Releases. As a result, we discovered 37 supernovacandidates, 21 of which are discovered for the first time.