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

Recent work suggests that representations learned by adversarially robustnetworks are more human perceptually-aligned than non-robust networks via imagemanipulations. Despite appearing closer to human visual perception, it isunclear if the constraints in robust DNN representations match biologicalconstraints found in human vision. Human vision seems to rely ontexture-based/summary statistic representations in the periphery, which havebeen shown to explain phenomena such as crowding and performance on visualsearch tasks. To understand how adversarially robustoptimizations/representations compare to human vision, we performed apsychophysics experiment using a set of metameric discrimination tasks where weevaluated how well human observers could distinguish between images synthesizedto match adversarially robust representations compared to non-robustrepresentations and a texture synthesis model of peripheral vision (Texforms).We found that the discriminability of robust representation and texture modelimages decreased to near chance performance as stimuli were presented fartherin the periphery. Moreover, performance on robust and texture-model imagesshowed similar trends within participants, while performance on non-robustrepresentations changed minimally across the visual field. These resultstogether suggest that (1) adversarially robust representations captureperipheral computation better than non-robust representations and (2) robustrepresentations capture peripheral computation similar to currentstate-of-the-art texture peripheral vision models. More broadly, our findingssupport the idea that localized texture summary statistic representations maydrive human invariance to adversarial perturbations and that the incorporationof such representations in DNNs could give rise to useful properties likeadversarial robustness.

Finding Biological Plausibility for Adversarially Robust Features via Metameric Tasks