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 become ubiquitous due to their performanceon prediction and classification problems. However, they face a variety ofthreats as their usage spreads. Model extraction attacks, which steal DNNs,endanger intellectual property, data privacy, and security. Previous researchhas shown that system-level side-channels can be used to leak the architectureof a victim DNN, exacerbating these risks. We propose two DNN architectureextraction techniques catering to various threat models. The first techniqueuses a malicious, dynamically linked version of PyTorch to expose a victim DNNarchitecture through the PyTorch profiler. The second, called EZClone, exploitsaggregate (rather than time-series) GPU profiles as a side-channel to predictDNN architecture, employing a simple approach and assuming little adversarycapability as compared to previous work. We investigate the effectiveness ofEZClone when minimizing the complexity of the attack, when applied to prunedmodels, and when applied across GPUs. We find that EZClone correctly predictsDNN architectures for the entire set of PyTorch vision architectures with 100%accuracy. No other work has shown this degree of architecture predictionaccuracy with the same adversarial constraints or using aggregate side-channelinformation. Prior work has shown that, once a DNN has been successfullycloned, further attacks such as model evasion or model inversion can beaccelerated significantly.

EZClone: Improving DNN Model Extraction Attack via Shape Distillation from GPU Execution Profiles