Memorizing Normality to Detect Anomaly: Memory-Augmented Deep Autoencoder for Unsupervised Anomaly Detection

By Dong Gong, Lingqiao Liu, Vuong Le, Budhaditya Saha, Moussa Reda Mansour, Svetha Venkatesh, Anton van den Hengel. IEEE International Conference on Computer Vision, 2019

1,733 citations recorded by the source corpus.

Deep autoencoder has been extensively used for anomaly detection. Training on the normal data, the autoencoder is expected to produce higher reconstruction error for the abnormal inputs than the normal ones, which is adopted as a criterion for identifying anomalies. However, this assumption does not always hold in practice. It has been observed that sometimes the autoencoder "generalizes" so well that it can also reconstruct anomalies well, leading to the miss detection of anomalies. To mitigate this drawback for autoencoder based anomaly detector, we propose to augment the autoencoder with a memory module and develop an improved autoencoder called memory-augmented autoencoder, i.e. MemAE. Given an input, MemAE firstly obtains the encoding from the encoder and then uses it as a query to retrieve the most relevant memory items for reconstruction. At the training stage, the memory contents are updated and are encouraged to represent the prototypical elements of the normal data. At the test stage, the learned memory will be fixed, and the reconstruction is obtained from a few selected memory records of the normal data. The reconstruction will thus tend to be close to a normal sample. Thus the reconstructed errors on anomalies will be strengthened for anomaly detection. MemAE is free of assumptions on the data type and thus general to be applied to different tasks. Experiments on various datasets prove the excellent generalization and high effectiveness of the proposed MemAE.

Subfields: Artificial Intelligence, Computer Networks and Communications

View Paper: https://www.semanticscholar.org/paper/d65eb30e5f0d2013fd5e4f45d1413bc2969ee803

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Networks and Communications

Citations in the corpus

SimpleNet: A Simple Network for Image Anomaly Detection and Localization (2023, methodology): “Some methods [10] utilize generative models such as auto-encoders and generative adversarial networks [11] to encode and reconstruct normal data.”
Generalized Out-of-Distribution Detection: A Survey (2021, positive): “AE/VAE-based Models Apart from the standard combination of reconstruction-error and AE/VAE models [111], [112], other methods use more sophisticated strategies such as reconstructing by memorized normality [115], [116], adapting model architectures [117], and partial/conditional reconstruction [89],”
PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization (2020, background): “At the image level, the simplest approach is to take the reconstructed error as an anomaly score [10] but additional information from the latent space”
Spatio-Temporal Meta-Graph Learning for Traffic Forecasting (2022, positive): “…further enhance its distinguishing power for diverse scenarios on different roads over time, we regulate the memory parameters with two constraints (Gong et al. 2019; Park, Noh, and Ham 2020), including a contrastive loss L 1 and a consistency loss L 1 , denoted by: (9) where T denotes the total…”
Sub-Image Anomaly Detection with Deep Pyramid Correspondences (2020, neutral): “TSC [23] StackRNN [23] AE-Conv3D [35] MemAE [12] AE(2D) [16] ITAE [19] SPADE”
EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latencies (2023, methodology): “[20] Dong Gong, Lingqiao Liu, Vuong Le, Budhaditya Saha, Moussa Reda Mansour, Svetha Venkatesh, and Anton Van Den Hengel.”
Learning Memory-Guided Normality for Anomaly Detection (2020, methodology): “Following the experimental protocol in [22, 8, 26], we measure the average area under curve (AUC) by computing the area under the receiver operation characteristics (ROC) with varying threshold values for abnormality scores.”
The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection (2021, background): “There exist various extensions to CAEs such as memory-augmented (Gong et al. 2019) or variational autoencoders (VAEs) (Kingma and Welling 2014).”
Same Same But DifferNet: Semi-Supervised Defect Detection with Normalizing Flows (2020, background): “[11] tackle the generalization problem by employing memory modules which can be seen as a discretized latent space.”
A Hybrid Video Anomaly Detection Framework via Memory-Augmented Flow Reconstruction and Flow-Guided Frame Prediction (2021, methodology): “For the memory modules, we adopt a similar implementation of [8].”
Unsupervised Deep Anomaly Detection for Multi-Sensor Time-Series Signals (2021, methodology): “works have tried to add Memory module [24] or Gaussian Mixture Model (GMM) [23].”
Self-Training Multi-Sequence Learning with Transformer for Weakly Supervised Video Anomaly Detection (2022, background): “There are three main paradigms: unsupervised VAD (Gong et al. 2019; Cai et al. 2021), weakly supervised VAD (Zhong et al. 2019), and supervised VAD (Liu and Ma 2019; Wan et al. 2021).”
OpenGAN: Open-Set Recognition via Open Data Generation (2021, neutral): “To recognize open-set examples, they resort to post-hoc functions like density estimation [59, 57], uncertainty modeling [17, 29], and input image reconstruction [39, 20, 14, 50].”
MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection (2021, methodology): “VAD aims to detect anomaly events in a given video and has been researched for years[9, 29, 14, 7, 15, 13, 12, 32, 31, 5, 24].”