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

Traditional classification algorithms assume that training and test data comefrom similar distributions. This assumption is violated in adversarialsettings, where malicious actors modify instances to evade detection. A numberof custom methods have been developed for both adversarial evasion attacks androbust learning. We propose the first systematic and general-purpose retrainingframework which can: a) boost robustness of an \emph{arbitrary} learningalgorithm, in the face of b) a broader class of adversarial models than anyprior methods. We show that, under natural conditions, the retraining frameworkminimizes an upper bound on optimal adversarial risk, and show how to extendthis result to account for approximations of evasion attacks. Extensiveexperimental evaluation demonstrates that our retraining methods are nearlyindistinguishable from state-of-the-art algorithms for optimizing adversarialrisk, but are more general and far more scalable. The experiments also confirmthat without retraining, our adversarial framework dramatically reduces theeffectiveness of learning. In contrast, retraining significantly boostsrobustness to evasion attacks without significantly compromising overallaccuracy.

A General Retraining Framework for Scalable Adversarial Classification