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/

Zendy Plus

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

Zendy Tools

Zendy Open

Contrastive learning (CL) has recently been applied to adversarial learningtasks. Such practice considers adversarial samples as additional positive viewsof an instance, and by maximizing their agreements with each other, yieldsbetter adversarial robustness. However, this mechanism can be potentiallyflawed, since adversarial perturbations may cause instance-level identityconfusion, which can impede CL performance by pulling together differentinstances with separate identities. To address this issue, we propose to treatadversarial samples unequally when contrasted, with an asymmetric InfoNCEobjective ($A-InfoNCE$) that allows discriminating considerations ofadversarial samples. Specifically, adversaries are viewed as inferior positivesthat induce weaker learning signals, or as hard negatives exhibiting highercontrast to other negative samples. In the asymmetric fashion, the adverseimpacts of conflicting objectives between CL and adversarial learning can beeffectively mitigated. Experiments show that our approach consistentlyoutperforms existing Adversarial CL methods across different finetuning schemeswithout additional computational cost. The proposed A-InfoNCE is also a genericform that can be readily extended to other CL methods. Code is available athttps://github.com/yqy2001/A-InfoNCE.

Adversarial Contrastive Learning via Asymmetric InfoNCE