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 has shown impressive performance on challenging perceptualtasks and has been widely used in software to provide intelligent services.However, researchers found deep neural networks vulnerable to adversarialexamples. Since then, many methods are proposed to defend against adversariesin inputs, but they are either attack-dependent or shown to be ineffective withnew attacks. And most of existing techniques have complicated structures ormechanisms that cause prohibitively high overhead or latency, impractical toapply on real software.  We propose DAFAR, a feedback framework that allows deep learning models todetect/purify adversarial examples in high effectiveness and universality, withlow area and time overhead. DAFAR has a simple structure, containing a victimmodel, a plug-in feedback network, and a detector. The key idea is to importthe high-level features from the victim model's feature extraction layers intothe feedback network to reconstruct the input. This data stream forms afeedback autoencoder. For strong attacks, it transforms the imperceptibleattack on the victim model into the obvious reconstruction-error attack on thefeedback autoencoder directly, which is much easier to detect; for weakattacks, the reformation process destroys the structure of adversarialexamples. Experiments are conducted on MNIST and CIFAR-10 data-sets, showingthat DAFAR is effective against popular and arguably most advanced attackswithout losing performance on legitimate samples, with high effectiveness anduniversality across attack methods and parameters.

DAFAR: Defending against Adversaries by Feedback-Autoencoder Reconstruction