Deep Anomaly Detection by Residual Adaptation

Deecke, Lucas; Ruff, Lukas; Vandermeulen, Robert A.; Bilen, Hakan

doi:10.48550/arxiv.2010.02310

Cited by 3 publications

(3 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Despite such negative samples may not coincide with ground-truth anomalies, such contrasting can be beneficial for learning characteristic representations of normal concepts. Moreover, the combination of Outlier Exposure and expressive representations with the hyperphere classifier have shown exceptional results for Deep AD on images (Ruff et al, 2020a;Deecke et al, 2020).…”

Section: −(1 − Ymentioning

confidence: 99%

Neural Contextual Anomaly Detection for Time Series

Carmona¹,

Aubet²,

Flunkert³

et al. 2021

Preprint

View full text Add to dashboard Cite

We introduce Neural Contextual Anomaly Detection (NCAD), a framework for anomaly detection on time series that scales seamlessly from the unsupervised to supervised setting, and is applicable to both univariate and multivariate time series. This is achieved by effectively combining recent developments in representation learning for multivariate time series, with techniques for deep anomaly detection originally developed for computer vision that we tailor to the time series setting. Our window-based approach facilitates learning the boundary between normal and anomalous classes by injecting generic synthetic anomalies into the available data. Moreover, our method can effectively take advantage of all the available information, be it as domain knowledge, or as training labels in the semi-supervised setting. We demonstrate empirically on standard benchmark datasets that our approach obtains a state-of-the-art performance in these settings. * Equal contribution. † Work done while working at AWS AI Labs.Preprint. Under review.

show abstract

Section: −(1 − Ymentioning

confidence: 99%

Neural Contextual Anomaly Detection for Time Series

Carmona¹,

Aubet²,

Flunkert³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…They further show that empirical gains can only be guaranteed when anomalies are used for OE that differ strongly from the normal class instead. Nevertheless, gains have been reported by Deecke et al [13], that make use of the HSC objective in conjunction with OE to learn a modulation of frozen features by means of newly introduced, residual adaptation layers. Liznerski et al [35] propose to fine-tune intermediate features extracted from a VGG network using their Fully Convolutional Data Descriptor (FCDD) objective together with OE.…”

Section: Learning Anomaly Detection From Scratchmentioning

confidence: 99%

“…However, due to catastrophic forgetting, feature representations cannot be easily fine-tuned to the datasets at hand. Furthermore, proposed methods that tackle this problem [40,13,38,35,48] currently do not integrate the strong Gaussian prior which can be induced based on ties between deep generative and discriminative modeling [32]. This prior has been confirmed in [42] to enhance AD results in the transfer learning setting in conjunction with fixed ImageNet representations.…”

Section: Introductionmentioning

confidence: 99%

Transfer Learning Gaussian Anomaly Detection by Fine-tuning Representations

Rippel,

Chavan,

Lei

et al. 2021

Preprint

View full text Add to dashboard Cite

Current state-of-the-art Anomaly Detection (AD) methods exploit the powerful representations yielded by large-scale ImageNet training. However, catastrophic forgetting prevents the successful fine-tuning of pre-trained representations on new datasets in the semi/unsupervised setting, and representations are therefore commonly fixed.In our work, we propose a new method to fine-tune learned representations for AD in a transfer learning setting. Based on the linkage between generative and discriminative modeling, we induce a multivariate Gaussian distribution for the normal class, and use the Mahalanobis distance of normal images to the distribution as training objective. We additionally propose to use augmentations commonly employed for vicinal risk minimization in a validation scheme to detect onset of catastrophic forgetting.Extensive evaluations on the public MVTec AD dataset reveal that a new state of the art is achieved by our method in the AD task while simultaneously achieving AS performance comparable to prior state of the art. Further, ablation studies demonstrate the importance of the induced Gaussian distri-

show abstract