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

Despite the great success achieved in machine learning (ML), adversarialexamples have caused concerns with regards to its trustworthiness: A smallperturbation of an input results in an arbitrary failure of an otherwiseseemingly well-trained ML model. While studies are being conducted to discoverthe intrinsic properties of adversarial examples, such as their transferabilityand universality, there is insufficient theoretic analysis to help understandthe phenomenon in a way that can influence the design process of MLexperiments. In this paper, we deduce an information-theoretic model whichexplains adversarial attacks as the abuse of feature redundancies in MLalgorithms. We prove that feature redundancy is a necessary condition for theexistence of adversarial examples. Our model helps to explain some majorquestions raised in many anecdotal studies on adversarial examples. Our theoryis backed up by empirical measurements of the information content of benign andadversarial examples on both image and text datasets. Our measurements showthat typical adversarial examples introduce just enough redundancy to overflowthe decision making of an ML model trained on corresponding benign examples. Weconclude with actionable recommendations to improve the robustness of machinelearners against adversarial examples.

One Bit Matters: Understanding Adversarial Examples as the Abuse of Redundancy