Latent Feature Decentralization Loss for One-Class Anomaly Detection

Abstract

Anomaly detection is essential for many real-world applications, such as video surveillance, disease diagnosis, and visual inspection. With the development of neural networks, many neural networks have been used for anomaly detection by learning the distribution of normal data. However, they are vulnerable to distinguishing abnormalities when the normal and abnormal images are not significantly different. To mitigate this problem, we propose a novel loss function for one-class anomaly detection: decentralization loss. The main goal of the proposed method is to cause the latent feature of the encoder to disperse over the manifold space, such that the decoder can generate images similar to those in a normal class for any input. To this end, a decentralization term designed based on the dispersion measure for latent vectors is also added to the existing mean-squared error loss. To design a general solution for various datasets, we restrict the latent space by designing a decentralization loss term-based upper bound of the dispersion measure. As intended, a model trained with the proposed decentralization loss function disperses vectors on the manifold space and generates constant images. Consequently, the reconstruction error increases when the given test image is unknown. Experiments conducted on various datasets verify that the proposed function improves detection performance improved by about 1 % while reducing training time by 48 %, without any structural changes in the conventional autoencoder.

Improving Novelty Detection using the Reconstructions of Nearest Neighbours

“…We show that AEs do not perform particularly well on CIFAR-10. This is expected, as images from the same class in contain substantially different Model MNIST CIFAR-10 F-MNIST GANomaly [8] 0.753 0.532 0.679 Skip-GAN [20] 0.492 0.629 0.515 OC-GAN [9] 0.683 0.510 0.678 VAE [21] 0.515 0.497 0.521 AnoGAN [3] 0.632 0.434 0.510 EGBAD [36] 0.656 0.496 0.500 DKNN [18] 0.791 0.714 0.746 Ours 0.921 0.560 0.763 Model MNIST CIFAR-10 F-MNIST MVTec-AD GANomaly [8] 0.965 0.695 0.906 0.762 OC-GAN [9] 0.975 0.657 0.924 0.756 AnoGAN [3] 0.912 0.618 0.817 0.600 LFD [28] 0.977 -0.927 0.777 CBiGAN [37] ---0.770 CAVGA-D u [34] 0.986 0.737 0.885 -DKNN 3 [18] 0.917 0.890 0.938 0.750 Ours 0.974 0.658 0.941 0.783 pixel-level information. For example the aeroplane class contains images of both the cockpit of a grounded Boeing 747 as well a fighter-jet photographed from the side-view in mid-flight.…”

“…In [9] and [28] the limitations of using AEs for one-class novelty detection are demonstrated. They show that when an AE is trained on the relatively complex 8-class from the MNIST dataset [46], the AE is able to implicitly learn the representations of digit classes such as the 1, 3, 6 and 7.…”

“…A common problem with using autoencoding methods for novelty detection is that AEs can generalise to unseen classes thereby performing poorly as novelty detectors [28]. In [9], this issue is addressed by placing a classifier in the training path of a multi-discriminator based autoencoder, which results in a fairly complicated and costly training procedure.…”

Anomaly Detection in 3D Point Clouds using Deep Geometric Descriptors

“…Ehret et al (2019) and Pang et al (2021) give comprehensive overviews. Some of the existing methods are trained from scratch with random weight initialization, in particular, those based on convolutional autoencoders (AEs) (Bergmann et al, 2019;Hong and Choe, 2020;Liu et al, 2020;Venkataramanan et al, 2020; or generative adversarial networks (GANs) (Carrara et al, 2021;Potter et al, 2020;Schlegl et al, 2019).…”

Prognostics-Based Reliability Optimization for Consumer Healthcare Devices

“…Any issues related to system design such as failure related to prolonged usage and overheating due to inadequate ventilation detected using appropriate data mining from usage profile [19]. Among the dataset related to device operation, data mining is implemented in facilitating model computation of failure prognostics for extracting operation data and to identify any failure patterns that might exhibit during normal operation under different environments [20] PHM facilitates fault avoidance and any necessary followup actions; it also assists with future product development for enhanced reliability. The maintenance data can be efficiently archived using the Self-Organizing Map (SOM) technique for document clustering [21].…”