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

Neural Image Classifiers are effective but inherently hard to interpret andsusceptible to adversarial attacks. Solutions to both problems exist, amongothers, in the form of counterfactual examples generation to enhanceexplainability or adversarially augment training datasets for improvedrobustness. However, existing methods exclusively address only one of theissues. We propose a unified framework leveraging image-to-image translationGenerative Adversarial Networks (GANs) to produce counterfactual samples thathighlight salient regions for interpretability and act as adversarial samplesto augment the dataset for more robustness. This is achieved by combining theclassifier and discriminator into a single model that attributes real images totheir respective classes and flags generated images as "fake". We assess themethod's effectiveness by evaluating (i) the produced explainability masks on asemantic segmentation task for concrete cracks and (ii) the model's resilienceagainst the Projected Gradient Descent (PGD) attack on a fruit defectsdetection problem. Our produced saliency maps are highly descriptive, achievingcompetitive IoU values compared to classical segmentation models despite beingtrained exclusively on classification labels. Furthermore, the model exhibitsimproved robustness to adversarial attacks, and we show how the discriminator's"fakeness" value serves as an uncertainty measure of the predictions.

Counterfactual Image Generation for adversarially robust and interpretable Classifiers