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) have been shown to tolerate "brain damage":cumulative changes to the network's parameters (e.g., pruning, numericalperturbations) typically result in a graceful degradation of classificationaccuracy. However, the limits of this natural resilience are not wellunderstood in the presence of small adversarial changes to the DNN parameters'underlying memory representation, such as bit-flips that may be induced byhardware fault attacks. We study the effects of bitwise corruptions on 19 DNNmodels---six architectures on three image classification tasks---and we showthat most models have at least one parameter that, after a specific bit-flip intheir bitwise representation, causes an accuracy loss of over 90%. We employsimple heuristics to efficiently identify the parameters likely to bevulnerable. We estimate that 40-50% of the parameters in a model might lead toan accuracy drop greater than 10% when individually subjected to suchsingle-bit perturbations. To demonstrate how an adversary could take advantageof this vulnerability, we study the impact of an exemplary hardware faultattack, Rowhammer, on DNNs. Specifically, we show that a Rowhammer enabledattacker co-located in the same physical machine can inflict significantaccuracy drops (up to 99%) even with single bit-flip corruptions and noknowledge of the model. Our results expose the limits of DNNs' resilienceagainst parameter perturbations induced by real-world fault attacks. Weconclude by discussing possible mitigations and future research directionstowards fault attack-resilient DNNs.

Terminal Brain Damage: Exposing the Graceless Degradation in Deep Neural Networks Under Hardware Fault Attacks