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

Advances in Machine Learning (ML) have led to its adoption as an integralcomponent in many applications, including banking, medical diagnosis, anddriverless cars. To further broaden the use of ML models, cloud-based servicesoffered by Microsoft, Amazon, Google, and others have developed ML-as-a-servicetools as black-box systems. However, ML classifiers are vulnerable toadversarial examples: inputs that are maliciously modified can cause theclassifier to provide adversary-desired outputs. Moreover, it is known thatadversarial examples generated on one classifier are likely to cause anotherclassifier to make the same mistake, even if the classifiers have differentarchitectures or are trained on disjoint datasets. This property, which isknown as transferability, opens up the possibility of attacking black-boxsystems by generating adversarial examples on a substitute classifier andtransferring the examples to the target classifier. Therefore, the key toprotect black-box learning systems against the adversarial examples is to blocktheir transferability. To this end, we propose a training method that, as theinput is more perturbed, the classifier smoothly outputs lower confidence onthe original label and instead predicts that the input is "invalid". Inessence, we augment the output class set with a NULL label and train theclassifier to reject the adversarial examples by classifying them as NULL. Inexperiments, we apply a wide range of attacks based on adversarial examples onthe black-box systems. We show that a classifier trained with the proposedmethod effectively resists against the adversarial examples, while maintainingthe accuracy on clean data.

Blocking Transferability of Adversarial Examples in Black-Box Learning Systems