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

Randomized smoothing is currently a state-of-the-art method to construct acertifiably robust classifier from neural networks against $\ell_2$-adversarialperturbations. Under the paradigm, the robustness of a classifier is alignedwith the prediction confidence, i.e., the higher confidence from a smoothedclassifier implies the better robustness. This motivates us to rethink thefundamental trade-off between accuracy and robustness in terms of calibratingconfidences of a smoothed classifier. In this paper, we propose a simpletraining scheme, coined SmoothMix, to control the robustness of smoothedclassifiers via self-mixup: it trains on convex combinations of samples alongthe direction of adversarial perturbation for each input. The proposedprocedure effectively identifies over-confident, near off-class samples as acause of limited robustness in case of smoothed classifiers, and offers anintuitive way to adaptively set a new decision boundary between these samplesfor better robustness. Our experimental results demonstrate that the proposedmethod can significantly improve the certified $\ell_2$-robustness of smoothedclassifiers compared to existing state-of-the-art robust training methods.

SmoothMix: Training Confidence-calibrated Smoothed Classifiers for Certified Robustness