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

Self-supervised learning (SSL) has become a popular method for generatinginvariant representations without the need for human annotations. Nonetheless,the desired invariant representation is achieved by utilising prior onlinetransformation functions on the input data. As a result, each SSL framework iscustomised for a particular data type, e.g., visual data, and furthermodifications are required if it is used for other dataset types. On the otherhand, autoencoder (AE), which is a generic and widely applicable framework,mainly focuses on dimension reduction and is not suited for learning invariantrepresentation. This paper proposes a generic SSL framework based on aconstrained self-labelling assignment process that prevents degeneratesolutions. Specifically, the prior transformation functions are replaced with aself-transformation mechanism, derived through an unsupervised training processof adversarial training, for imposing invariant representations. Via theself-transformation mechanism, pairs of augmented instances can be generatedfrom the same input data. Finally, a training objective based on contrastivelearning is designed by leveraging both the self-labelling assignment and theself-transformation mechanism. Despite the fact that the self-transformationprocess is very generic, the proposed training strategy outperforms a majorityof state-of-the-art representation learning methods based on AE structures. Tovalidate the performance of our method, we conduct experiments on four types ofdata, namely visual, audio, text, and mass spectrometry data, and compare themin terms of four quantitative metrics. Our comparison results indicate that theproposed method demonstrate robustness and successfully identify patternswithin the datasets.

A Generic Self-Supervised Framework of Learning Invariant Discriminative Features