The article discusses a new method for detecting anomalies when only a few types of anomalies are known. Previous research focused on either labeled or unlabeled samples, but this new approach deals with situations where only some anomalies are observed. The researchers developed a two-stage algorithm that effectively detects anomalies even when not all types are seen. Their method works well with different datasets and doesn't require specific settings, showing promising results in both limited and comprehensive anomaly observations.