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

Existing works show that augmenting the training data of pre-trained languagemodels (PLMs) for classification tasks fine-tuned via parameter-efficientfine-tuning methods (PEFT) using both clean and adversarial examples canenhance their robustness under adversarial attacks. However, this adversarialtraining paradigm often leads to performance degradation on clean inputs andrequires frequent re-training on the entire data to account for new, unknownattacks. To overcome these challenges while still harnessing the benefits ofadversarial training and the efficiency of PEFT, this work proposes a novelapproach, called AdpMixup, that combines two paradigms: (1) fine-tuning throughadapters and (2) adversarial augmentation via mixup to dynamically leverageexisting knowledge from a set of pre-known attacks for robust inference.Intuitively, AdpMixup fine-tunes PLMs with multiple adapters with both cleanand pre-known adversarial examples and intelligently mixes them up in differentratios during prediction. Our experiments show AdpMixup achieves the besttrade-off between training efficiency and robustness under both pre-known andunknown attacks, compared to existing baselines on five downstream tasks acrosssix varied black-box attacks and 2 PLMs. All source code will be available.

Adapters Mixup: Mixing Parameter-Efficient Adapters to Enhance the Adversarial Robustness of Fine-tuned Pre-trained Text Classifiers