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 has introduced attacks that extract the architecture informationof deep neural networks (DNN), as this knowledge enhances an adversary'scapability to conduct black-box attacks against the model. This paper presentsthe first in-depth security analysis of DNN fingerprinting attacks that exploitcache side-channels. First, we define the threat model for these attacks: ouradversary does not need the ability to query the victim model; instead, sheruns a co-located process on the host machine victim's deep learning (DL)system is running and passively monitors the accesses of the target functionsin the shared framework. Second, we introduce DeepRecon, an attack thatreconstructs the architecture of the victim network by using the internalinformation extracted via Flush+Reload, a cache side-channel technique. Oncethe attacker observes function invocations that map directly to architectureattributes of the victim network, the attacker can reconstruct the victim'sentire network architecture. In our evaluation, we demonstrate that an attackercan accurately reconstruct two complex networks (VGG19 and ResNet50) havingobserved only one forward propagation. Based on the extracted architectureattributes, we also demonstrate that an attacker can build a meta-model thataccurately fingerprints the architecture and family of the pre-trained model ina transfer learning setting. From this meta-model, we evaluate the importanceof the observed attributes in the fingerprinting process. Third, we propose andevaluate new framework-level defense techniques that obfuscate our attacker'sobservations. Our empirical security analysis represents a step towardunderstanding the DNNs' vulnerability to cache side-channel attacks.

Security Analysis of Deep Neural Networks Operating in the Presence of Cache Side-Channel Attacks