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

In recent years, several unsupervised, "contrastive" learning algorithms invision have been shown to learn representations that perform remarkably well ontransfer tasks. We show that this family of algorithms maximizes a lower boundon the mutual information between two or more "views" of an image where typicalviews come from a composition of image augmentations. Our bound generalizes theInfoNCE objective to support negative sampling from a restricted region of"difficult" contrasts. We find that the choice of negative samples and viewsare critical to the success of these algorithms. Reformulating previouslearning objectives in terms of mutual information also simplifies andstabilizes them. In practice, our new objectives yield representations thatoutperform those learned with previous approaches for transfer toclassification, bounding box detection, instance segmentation, and keypointdetection. % experiments show that choosing more difficult negative samplesresults in a stronger representation, outperforming those learned with IR, LA,and CMC in classification, bounding box detection, instance segmentation, andkeypoint detection. The mutual information framework provides a unifyingcomparison of approaches to contrastive learning and uncovers the choices thatimpact representation learning.

On Mutual Information in Contrastive Learning for Visual Representations