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

Despite achieving remarkable success in various domains, recent studies haveuncovered the vulnerability of deep neural networks to adversarialperturbations, creating concerns on model generalizability and new threats suchas prediction-evasive misclassification or stealthy reprogramming. Amongdifferent defense proposals, stochastic network defenses such as random neuronactivation pruning or random perturbation to layer inputs are shown to bepromising for attack mitigation. However, one critical drawback of currentdefenses is that the robustness enhancement is at the cost of noticeableperformance degradation on legitimate data, e.g., large drop in test accuracy.This paper is motivated by pursuing for a better trade-off between adversarialrobustness and test accuracy for stochastic network defenses. We proposeDefense Efficiency Score (DES), a comprehensive metric that measures the gainin unsuccessful attack attempts at the cost of drop in test accuracy of anydefense. To achieve a better DES, we propose hierarchical random switching(HRS), which protects neural networks through a novel randomization scheme. AHRS-protected model contains several blocks of randomly switching channels toprevent adversaries from exploiting fixed model structures and parameters fortheir malicious purposes. Extensive experiments show that HRS is superior indefending against state-of-the-art white-box and adaptive adversarialmisclassification attacks. We also demonstrate the effectiveness of HRS indefending adversarial reprogramming, which is the first defense againstadversarial programs. Moreover, in most settings the average DES of HRS is atleast 5X higher than current stochastic network defenses, validating itssignificantly improved robustness-accuracy trade-off.

Protecting Neural Networks with Hierarchical Random Switching: Towards Better Robustness-Accuracy Trade-off for Stochastic Defenses