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

We propose a novel and effective purification based adversarial defensemethod against pre-processor blind white- and black-box attacks. Our method iscomputationally efficient and trained only with self-supervised learning ongeneral images, without requiring any adversarial training or retraining of theclassification model. We first show an empirical analysis on the adversarialnoise, defined to be the residual between an original image and its adversarialexample, has almost zero mean, symmetric distribution. Based on thisobservation, we propose a very simple iterative Gaussian Smoothing (GS) whichcan effectively smooth out adversarial noise and achieve substantially highrobust accuracy. To further improve it, we propose Neural Contextual IterativeSmoothing (NCIS), which trains a blind-spot network (BSN) in a self-supervisedmanner to reconstruct the discriminative features of the original image that isalso smoothed out by GS. From our extensive experiments on the large-scaleImageNet using four classification models, we show that our method achievesboth competitive standard accuracy and state-of-the-art robust accuracy againstmost strong purifier-blind white- and black-box attacks. Also, we propose a newbenchmark for evaluating a purification method based on commercial imageclassification APIs, such as AWS, Azure, Clarifai and Google. We generateadversarial examples by ensemble transfer-based black-box attack, which caninduce complete misclassification of APIs, and demonstrate that our method canbe used to increase adversarial robustness of APIs.

NCIS: Neural Contextual Iterative Smoothing for Purifying Adversarial Perturbations