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

A recent popular approach to out-of-distribution (OOD) detection is based ona self-supervised learning technique referred to as contrastive learning. Thereare two main variants of contrastive learning, namely instance and classdiscrimination, targeting features that can discriminate between differentinstances for the former, and different classes for the latter.  In this paper, we aim to understand the effectiveness and limitation ofexisting contrastive learning methods for OOD detection. We approach this in 3ways. First, we systematically study the performance difference between theinstance discrimination and supervised contrastive learning variants indifferent OOD detection settings. Second, we study which in-distribution (ID)classes OOD data tend to be classified into. Finally, we study the spectraldecay property of the different contrastive learning approaches and examine howit correlates with OOD detection performance. In scenarios where the ID and OODdatasets are sufficiently different from one another, we see that instancediscrimination, in the absence of fine-tuning, is competitive with supervisedapproaches in OOD detection. We see that OOD samples tend to be classified intoclasses that have a distribution similar to the distribution of the entiredataset. Furthermore, we show that contrastive learning learns a feature spacethat contains singular vectors containing several directions with a highvariance which can be detrimental or beneficial to OOD detection depending onthe inference approach used.

Understanding the properties and limitations of contrastive learning for Out-of-Distribution detection