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

The vulnerability to slight input perturbations is a worrying yet intriguingproperty of deep neural networks (DNNs). Despite many previous works studyingthe reason behind such adversarial behavior, the relationship between thegeneralization performance and adversarial behavior of DNNs is still littleunderstood. In this work, we reveal such relation by introducing a metriccharacterizing the generalization performance of a DNN. The metric can bedisentangled into an information-theoretic non-robust component, responsiblefor adversarial behavior, and a robust component. Then, we show by experimentsthat current DNNs rely heavily on optimizing the non-robust component inachieving decent performance. We also demonstrate that current state-of-the-artadversarial training algorithms indeed try to robustify the DNNs by preventingthem from using the non-robust component to distinguish samples from differentcategories. Also, based on our findings, we take a step forward and point outthe possible direction for achieving decent standard performance andadversarial robustness simultaneously. We believe that our theory could furtherinspire the community to make more interesting discoveries about therelationship between standard generalization and adversarial generalization ofdeep learning models.

Understanding Adversarial Behavior of DNNs by Disentangling Non-Robust and Robust Components in Performance Metric