English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Unsupervised Anomaly detection (AD) requires building a notion of normalcy,distinguishing in-distribution (ID) and out-of-distribution (OOD) data, usingonly available ID samples. Recently, large gains were made on this task for thedomain of natural images using self-supervised contrastive feature learning asa first step followed by kNN or traditional one-class classifiers for featurescoring. Learned representations that are non-uniformly distributed on the unithypersphere have been shown to be beneficial for this task. We go a stepfurther and investigate how the \emph {geometrical compactness} of the IDfeature distribution makes isolating and detecting outliers easier, especiallyin the realistic situation when ID training data is polluted (i.e. ID datacontains some OOD data that is used for learning the feature extractorparameters). We propose novel architectural modifications to theself-supervised feature learning step, that enable such compact distributionsfor ID data to be learned. We show that the proposed modifications can beeffectively applied to most existing self-supervised objectives, with largegains in performance. Furthermore, this improved OOD performance is obtainedwithout resorting to tricks such as using strongly augmented ID images (e.g. by90 degree rotations) as proxies for the unseen OOD data, as these impose overlyprescriptive assumptions about ID data and its invariances. We performextensive studies on benchmark datasets for one-class OOD detection and showstate-of-the-art performance in the presence of pollution in the ID data, andcomparable performance otherwise. We also propose and extensively evaluate anovel feature scoring technique based on the angular Mahalanobis distance, andpropose a simple and novel technique for feature ensembling during evaluationthat enables a big boost in performance at nearly zero run-time cost.

No Shifted Augmentations (NSA): compact distributions for robust self-supervised Anomaly Detection