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

Crafting adversarial examples has become an important technique to evaluatethe robustness of deep neural networks (DNNs). However, most existing worksfocus on attacking the image classification problem since its input space iscontinuous and output space is finite.  In this paper, we study the much more challenging problem of craftingadversarial examples for sequence-to-sequence (seq2seq) models, whose inputsare discrete text strings and outputs have an almost infinite number ofpossibilities. To address the challenges caused by the discrete input space, wepropose a projected gradient method combined with group lasso and gradientregularization. To handle the almost infinite output space, we design somenovel loss functions to conduct non-overlapping attack and targeted keywordattack. We apply our algorithm to machine translation and text summarizationtasks, and verify the effectiveness of the proposed algorithm: by changing lessthan 3 words, we can make seq2seq model to produce desired outputs with highsuccess rates. On the other hand, we recognize that, compared with thewell-evaluated CNN-based classifiers, seq2seq models are intrinsically morerobust to adversarial attacks.

Seq2Sick: Evaluating the Robustness of Sequence-to-Sequence Models with Adversarial Examples