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 (DNNs) are known to have a fundamental sensitivity toadversarial attacks, perturbations of the input that are imperceptible tohumans yet powerful enough to change the visual decision of a model.Adversarial attacks have long been considered the "Achilles' heel" of deeplearning, which may eventually force a shift in modeling paradigms.Nevertheless, the formidable capabilities of modern large-scale DNNs havesomewhat eclipsed these early concerns. Do adversarial attacks continue to posea threat to DNNs?  Here, we investigate how the robustness of DNNs to adversarial attacks hasevolved as their accuracy on ImageNet has continued to improve. We measureadversarial robustness in two different ways: First, we measure the smallestadversarial attack needed to cause a model to change its object categorizationdecision. Second, we measure how aligned successful attacks are with thefeatures that humans find diagnostic for object recognition. We find thatadversarial attacks are inducing bigger and more easily detectable changes toimage pixels as DNNs grow better on ImageNet, but these attacks are alsobecoming less aligned with features that humans find diagnostic forrecognition. To better understand the source of this trade-off, we turn to theneural harmonizer, a DNN training routine that encourages models to leveragethe same features as humans to solve tasks. Harmonized DNNs achieve the best ofboth worlds and experience attacks that are detectable and affect features thathumans find diagnostic for recognition, meaning that attacks on these modelsare more likely to be rendered ineffective by inducing similar effects on humanperception. Our findings suggest that the sensitivity of DNNs to adversarialattacks can be mitigated by DNN scale, data scale, and training routines thatalign models with biological intelligence.

Adversarial alignment: Breaking the trade-off between the strength of an attack and its relevance to human perception