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 Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Adversarial training serves as one of the most popular and effective methodsto defend against adversarial perturbations. However, most defense mechanismsonly consider a single type of perturbation while various attack methods mightbe adopted to perform stronger adversarial attacks against the deployed modelin real-world scenarios, e.g., $\ell_2$ or $\ell_\infty$. Defending againstvarious attacks can be a challenging problem since multi-perturbationadversarial training and its variants only achieve suboptimal robustnesstrade-offs, due to the theoretical limit to multi-perturbation robustness for asingle model. Besides, it is impractical to deploy large models in somestorage-efficient scenarios. To settle down these drawbacks, in this paper wepropose a novel multi-perturbation adversarial training framework,parameter-saving adversarial training (PSAT), to reinforce multi-perturbationrobustness with an advantageous side effect of saving parameters, whichleverages hypernetworks to train specialized models against a singleperturbation and aggregate these specialized models to defend against multipleperturbations. Eventually, we extensively evaluate and compare our proposedmethod with state-of-the-art single/multi-perturbation robust methods againstvarious latest attack methods on different datasets, showing the robustnesssuperiority and parameter efficiency of our proposed method, e.g., for theCIFAR-10 dataset with ResNet-50 as the backbone, PSAT saves approximately 80\%of parameters with achieving the state-of-the-art robustness trade-offaccuracy.

Parameter-Saving Adversarial Training: Reinforcing Multi-Perturbation Robustness via Hypernetworks