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

Deep neural networks are susceptible to adversarially crafted, small andimperceptible changes in the natural inputs. The most effective defensemechanism against these examples is adversarial training which constructsadversarial examples during training by iterative maximization of loss. Themodel is then trained to minimize the loss on these constructed examples. Thismin-max optimization requires more data, larger capacity models, and additionalcomputing resources. It also degrades the standard generalization performanceof a model. Can we achieve robustness more efficiently? In this work, weexplore this question from the perspective of knowledge transfer. First, wetheoretically show the transferability of robustness from an adversariallytrained teacher model to a student model with the help of mixup augmentation.Second, we propose a novel robustness transfer method called Mixup-BasedActivated Channel Maps (MixACM) Transfer. MixACM transfers robustness from arobust teacher to a student by matching activated channel maps generatedwithout expensive adversarial perturbations. Finally, extensive experiments onmultiple datasets and different learning scenarios show our method can transferrobustness while also improving generalization on natural images.

MixACM: Mixup-Based Robustness Transfer via Distillation of Activated Channel Maps