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

Fast adversarial training (FAT) effectively improves the efficiency ofstandard adversarial training (SAT). However, initial FAT encounterscatastrophic overfitting, i.e.,the robust accuracy against adversarial attackssuddenly and dramatically decreases. Though several FAT variants spare noeffort to prevent overfitting, they sacrifice much calculation cost. In thispaper, we explore the difference between the training processes of SAT and FATand observe that the attack success rate of adversarial examples (AEs) of FATgets worse gradually in the late training stage, resulting in overfitting. TheAEs are generated by the fast gradient sign method (FGSM) with a zero or randominitialization. Based on the observation, we propose a prior-guided FGSMinitialization method to avoid overfitting after investigating severalinitialization strategies, improving the quality of the AEs during the wholetraining process. The initialization is formed by leveraging historicallygenerated AEs without additional calculation cost. We further provide atheoretical analysis for the proposed initialization method. We also propose asimple yet effective regularizer based on the prior-guided initialization,i.e.,the currently generated perturbation should not deviate too much from theprior-guided initialization. The regularizer adopts both historical and currentadversarial perturbations to guide the model learning. Evaluations on fourdatasets demonstrate that the proposed method can prevent catastrophicoverfitting and outperform state-of-the-art FAT methods. The code is releasedat https://github.com/jiaxiaojunQAQ/FGSM-PGI.

Prior-Guided Adversarial Initialization for Fast Adversarial Training