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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Deep Neural Networks (DNNs) are vulnerable to deliberately craftedadversarial examples. In the past few years, many efforts have been spent onexploring query-optimisation attacks to find adversarial examples of eitherblack-box or white-box DNN models, as well as the defending countermeasuresagainst those attacks. In this work, we explore vulnerabilities of DNN modelsunder the umbrella of Man-in-the-Middle (MitM) attacks, which has not beeninvestigated before. From the perspective of an MitM adversary, theaforementioned adversarial example attacks are not viable anymore. First, suchattacks must acquire the outputs from the models by multiple times beforeactually launching attacks, which is difficult for the MitM adversary inpractice. Second, such attacks are one-off and cannot be directly generalisedonto new data examples, which decreases the rate of return for the attacker. Incontrast, using generative models to craft adversarial examples on the fly canmitigate the drawbacks. However, the adversarial capability of the generativemodels, such as Variational Auto-Encoder (VAE), has not been extensivelystudied. Therefore, given a classifier, we investigate using a VAE decoder toeither transform benign inputs to their adversarial counterparts or decodeoutputs from benign VAE encoders to be adversarial examples. The proposedmethod can endue more capability to MitM attackers. Based on our evaluation,the proposed attack can achieve above 95% success rate on both MNIST andCIFAR10 datasets, which is better or comparable with state-of-the-artquery-optimisation attacks. At the meantime, the attack is 104 times fasterthan the query-optimisation attacks.

Man-in-the-Middle Attacks against Machine Learning Classifiers via Malicious Generative Models