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

Real-world natural language processing systems need to be robust to humanadversaries. Collecting examples of human adversaries for training is aneffective but expensive solution. On the other hand, training on syntheticattacks with small perturbations - such as word-substitution - does notactually improve robustness to human adversaries. In this paper, we propose anadversarial training framework that uses limited human adversarial examples togenerate more useful adversarial examples at scale. We demonstrate theadvantages of this system on the ANLI and hate speech detection benchmarkdatasets - both collected via an iterative, adversarialhuman-and-model-in-the-loop procedure. Compared to training only on observedhuman attacks, also training on our synthetic adversarial examples improvesmodel robustness to future rounds. In ANLI, we see accuracy gains on thecurrent set of attacks (44.1%$\,\to\,$50.1%) and on two future unseen rounds ofhuman generated attacks (32.5%$\,\to\,$43.4%, and 29.4%$\,\to\,$40.2%). In hatespeech detection, we see AUC gains on current attacks (0.76 $\to$ 0.84) and afuture round (0.77 $\to$ 0.79). Attacks from methods that do not learn thedistribution of existing human adversaries, meanwhile, degrade robustness.

Break it, Imitate it, Fix it: Robustness by Generating Human-Like Attacks