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 neural networks have proven remarkably effective at solving manyclassification problems, but have been criticized recently for two majorweaknesses: the reasons behind their predictions are uninterpretable, and thepredictions themselves can often be fooled by small adversarial perturbations.These problems pose major obstacles for the adoption of neural networks indomains that require security or transparency. In this work, we evaluate theeffectiveness of defenses that differentiably penalize the degree to whichsmall changes in inputs can alter model predictions. Across multiple attacks,architectures, defenses, and datasets, we find that neural networks trainedwith this input gradient regularization exhibit robustness to transferredadversarial examples generated to fool all of the other models. We also findthat adversarial examples generated to fool gradient-regularized models foolall other models equally well, and actually lead to more "legitimate,"interpretable misclassifications as rated by people (which we confirm in ahuman subject experiment). Finally, we demonstrate that regularizing inputgradients makes them more naturally interpretable as rationales for modelpredictions. We conclude by discussing this relationship betweeninterpretability and robustness in deep neural networks.

Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradients