Few-Shot Anomaly Detection Using Deep Generative Models for Grouped Data

Abstract

There exists a great demand for automatic anomaly detection in industrial world. The anomaly has been defined as a group of samples that rarely or never appears. Given a type of products, one has to collect numerous samples and train an anomaly detector. When one diverts a model trained with old types of products with sufficient inventory to the new type, one can detect anomalies of the new type before a production line is established. However, because of the definition of the anomaly, a typical anomaly detector considers the new type of products anomalous even if it is consistent with the standard. Given the above practical demand, this study propose a novel problem setting, few-shot anomaly detection, where an anomaly detector trained in source domains is adapted to a small set of target samples without full retraining. Then, we tackle this problem using a hierarchical probabilistic model based on deep learning. Our empirical results on toy and real-world datasets demonstrate that the proposed model detects anomalies in a small set of target samples successfully.

Teacher-Student Network for 3D Point Cloud Anomaly Detection with Few Normal Samples

“…As for the few-shot based methods, in [29], a multilevel VAE is proposed to separate domain-level features from samplelevel features for domain generalization with few samples. In [24], a novel prior-driven model is created to achieve an end-to-end differentiable learning of fine-grained anomaly score by utilizing a small number of labeled anomalies with a Gaussian prior.…”

Few-shot pump anomaly detection via Diff-WRN-based model-agnostic meta-learning strategy

“…18 The DL models, such as convolutional neural network (CNN), 19 recurrent neural network (RNN), 20 deep auto encoder (DAE), 21 and generative adversarial networks (GAN), 22 display ever-improving feature distillation capabilities. 23 In general, existing FSL approaches, such as data augmentation, model architecture, and algorithm design, use three types of learning strategy: (1) metric learning, 24 (2) meta-learning, 25 and (3) graph network. 26,27 The first one evaluates the distance metrics using a modeling process; the allure of meta-learning is its rapid adaptability to new tasks, which can construct a general model; and the graph network investigates the label structures between the support set and the query set by generating embedding models.…”