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

Although fast adversarial training provides an efficient approach forbuilding robust networks, it may suffer from a serious problem known ascatastrophic overfitting (CO), where multi-step robust accuracy suddenlycollapses to zero. In this paper, we for the first time decouple single-stepadversarial examples into data-information and self-information, which revealsan interesting phenomenon called "self-fitting". Self-fitting, i.e., thenetwork learns the self-information embedded in single-step perturbations,naturally leads to the occurrence of CO. When self-fitting occurs, the networkexperiences an obvious "channel differentiation" phenomenon that someconvolution channels accounting for recognizing self-information becomedominant, while others for data-information are suppressed. In this way, thenetwork can only recognize images with sufficient self-information and losesgeneralization ability to other types of data. Based on self-fitting, weprovide new insights into the existing methods to mitigate CO and extend CO tomulti-step adversarial training. Our findings reveal a self-learning mechanismin adversarial training and open up new perspectives for suppressing differentkinds of information to mitigate CO.

Investigating Catastrophic Overfitting in Fast Adversarial Training: A Self-fitting Perspective