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

Deep Learning based AI systems have shown great promise in various domainssuch as vision, audio, autonomous systems (vehicles, drones), etc. Recentresearch on neural networks has shown the susceptibility of deep networks toadversarial attacks - a technique of adding small perturbations to the inputswhich can fool a deep network into misclassifying them. Developing defensesagainst such adversarial attacks is an active research area, with someapproaches proposing robust models that are immune to such adversaries, whileother techniques attempt to detect such adversarial inputs. In this paper, wepresent a novel statistical approach for adversarial detection in imageclassification. Our approach is based on constructing a per-class featuredistribution and detecting adversaries based on comparison of features of atest image with the feature distribution of its class. For this purpose, wemake use of various statistical distances such as ED (Energy Distance), MMD(Maximum Mean Discrepancy) for adversarial detection, and analyze theperformance of each metric. We experimentally show that our approach achievesgood adversarial detection performance on MNIST and CIFAR-10 datasetsirrespective of the attack method, sample size and the degree of adversarialperturbation.

Attack Agnostic Statistical Method for Adversarial Detection