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

As the will to deploy neural networks models on embedded systems grows, andconsidering the related memory footprint and energy consumption issues, findinglighter solutions to store neural networks such as weight quantization and moreefficient inference methods become major research topics. Parallel to that,adversarial machine learning has risen recently with an impressive andsignificant attention, unveiling some critical flaws of machine learningmodels, especially neural networks. In particular, perturbed inputs calledadversarial examples have been shown to fool a model into making incorrectpredictions. In this article, we investigate the adversarial robustness ofquantized neural networks under different threat models for a classicalsupervised image classification task. We show that quantization does not offerany robust protection, results in severe form of gradient masking and advancesome hypotheses to explain it. However, we experimentally observe poortransferability capacities which we explain by quantization value shiftphenomenon and gradient misalignment and explore how these results can beexploited with an ensemble-based defense.

Impact of Low-bitwidth Quantization on the Adversarial Robustness for Embedded Neural Networks