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

Recently, it has been shown that deep neural networks (DNN) are subject toattacks through adversarial samples. Adversarial samples are often craftedthrough adversarial perturbation, i.e., manipulating the original sample withminor modifications so that the DNN model labels the sample incorrectly. Giventhat it is almost impossible to train perfect DNN, adversarial samples areshown to be easy to generate. As DNN are increasingly used in safety-criticalsystems like autonomous cars, it is crucial to develop techniques for defendingsuch attacks. Existing defense mechanisms which aim to make adversarialperturbation challenging have been shown to be ineffective. In this work, wepropose an alternative approach. We first observe that adversarial samples aremuch more sensitive to perturbations than normal samples. That is, if we imposerandom perturbations on a normal and an adversarial sample respectively, thereis a significant difference between the ratio of label change due to theperturbations. Observing this, we design a statistical adversary detectionalgorithm called nMutant (inspired by mutation testing from softwareengineering community). Our experiments show that nMutant effectively detectsmost of the adversarial samples generated by recently proposed attackingmethods. Furthermore, we provide an error bound with certain statisticalsignificance along with the detection.

Detecting Adversarial Samples for Deep Neural Networks through Mutation Testing