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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Recent studies on adversarial images have shown that they tend to leave theunderlying low-dimensional data manifold, making them significantly morechallenging for current models to make correct predictions. This so-calledoff-manifold conjecture has inspired a novel line of defenses againstadversarial attacks on images. In this study, we find a similar phenomenonoccurs in the contextualized embedding space induced by pretrained languagemodels, in which adversarial texts tend to have their embeddings diverge fromthe manifold of natural ones. Based on this finding, we propose TextualManifold-based Defense (TMD), a defense mechanism that projects text embeddingsonto an approximated embedding manifold before classification. It reduces thecomplexity of potential adversarial examples, which ultimately enhances therobustness of the protected model. Through extensive experiments, our methodconsistently and significantly outperforms previous defenses under variousattack settings without trading off clean accuracy. To the best of ourknowledge, this is the first NLP defense that leverages the manifold structureagainst adversarial attacks. Our code is available at\url{https://github.com/dangne/tmd}.

Textual Manifold-based Defense Against Natural Language Adversarial Examples