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

Substantial research works have shown that deep models, e.g., pre-trainedmodels, on the large corpus can learn universal language representations, whichare beneficial for downstream NLP tasks. However, these powerful models arealso vulnerable to various privacy attacks, while much sensitive informationexists in the training dataset. The attacker can easily steal sensitiveinformation from public models, e.g., individuals' email addresses and phonenumbers. In an attempt to address these issues, particularly the unauthorizeduse of private data, we introduce a novel watermarking technique via abackdoor-based membership inference approach named TextMarker, which cansafeguard diverse forms of private information embedded in the training textdata. Specifically, TextMarker only requires data owners to mark a small numberof samples for data copyright protection under the black-box access assumptionto the target model. Through extensive evaluation, we demonstrate theeffectiveness of TextMarker on various real-world datasets, e.g., marking only0.1% of the training dataset is practically sufficient for effective membershipinference with negligible effect on model utility. We also discuss potentialcountermeasures and show that TextMarker is stealthy enough to bypass them.

Watermarking Text Data on Large Language Models for Dataset Copyright