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 present a novel framework for generating adversarial benchmarks toevaluate the robustness of image classification models. Our framework allowsusers to customize the types of distortions to be optimally applied to images,which helps address the specific distortions relevant to their deployment. Thebenchmark can generate datasets at various distortion levels to assess therobustness of different image classifiers. Our results show that theadversarial samples generated by our framework with any of the imageclassification models, like ResNet-50, Inception-V3, and VGG-16, are effectiveand transferable to other models causing them to fail. These failures happeneven when these models are adversarially retrained using state-of-the-arttechniques, demonstrating the generalizability of our adversarial samples. Weachieve competitive performance in terms of net $L_2$ distortion compared tostate-of-the-art benchmark techniques on CIFAR-10 and ImageNet; however, wedemonstrate our framework achieves such results with simple distortions likeGaussian noise without introducing unnatural artifacts or color bleeds. This ismade possible by a model-based reinforcement learning (RL) agent and atechnique that reduces a deep tree search of the image for model sensitivity toperturbations, to a one-level analysis and action. The flexibility of choosingdistortions and setting classification probability thresholds for multipleclasses makes our framework suitable for algorithmic audits.

Benchmark Generation Framework with Customizable Distortions for Image Classifier Robustness