A Gated Few-shot Learning Model For Anomaly Detection

Anomaly detection, as one of the most important problems in the domain of network and service management, has been widely studied in statistics and machine learning. The supervised methods with plenty of labeled data have achieved great success in anomaly detection, but cannot integrate new anomaly types. In this paper, we propose a few-shot learning model for anomaly detection. Our model is trained with labeled data, and is then tested in terms of its ability to learn how to detect new types, given examples of the unseen classes. Due to a gap between the known anomaly data and unseen anomaly data, we designed a gated network structure to tackle the imbalanced data problem, to which we added a gate structure to aggregate known anomaly types and unknown types. We evaluated our proposed method based on the anomaly dataset NSL-KDD and our experimental results show that the proposed method achieved the state-of-the-art results in few-shot settings.