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

It has been shown that most machine learning algorithms are susceptible toadversarial perturbations. Slightly perturbing an image in a carefully chosendirection in the image space may cause a trained neural network model tomisclassify it. Recently, it was shown that physical adversarial examplesexist: printing perturbed images then taking pictures of them would stillresult in misclassification. This raises security and safety concerns.  However, these experiments ignore a crucial property of physical objects: thecamera can view objects from different distances and at different angles. Inthis paper, we show experiments that suggest that current constructions ofphysical adversarial examples do not disrupt object detection from a movingplatform. Instead, a trained neural network classifies most of the picturestaken from different distances and angles of a perturbed image correctly. Webelieve this is because the adversarial property of the perturbation issensitive to the scale at which the perturbed picture is viewed, so (forexample) an autonomous car will misclassify a stop sign only from a small rangeof distances.  Our work raises an important question: can one construct examples that areadversarial for many or most viewing conditions? If so, the construction shouldoffer very significant insights into the internal representation of patterns bydeep networks. If not, there is a good prospect that adversarial examples canbe reduced to a curiosity with little practical impact.

NO Need to Worry about Adversarial Examples in Object Detection in Autonomous Vehicles