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

This paper studies model-inversion attacks, in which the access to a model isabused to infer information about the training data. Since its firstintroduction, such attacks have raised serious concerns given that trainingdata usually contain privacy-sensitive information. Thus far, successfulmodel-inversion attacks have only been demonstrated on simple models, such aslinear regression and logistic regression. Previous attempts to invert neuralnetworks, even the ones with simple architectures, have failed to produceconvincing results. We present a novel attack method, termed the generativemodel-inversion attack, which can invert deep neural networks with high successrates. Rather than reconstructing private training data from scratch, weleverage partial public information, which can be very generic, to learn adistributional prior via generative adversarial networks (GANs) and use it toguide the inversion process. Moreover, we theoretically prove that a model'spredictive power and its vulnerability to inversion attacks are indeed twosides of the same coin---highly predictive models are able to establish astrong correlation between features and labels, which coincides exactly withwhat an adversary exploits to mount the attacks. Our extensive experimentsdemonstrate that the proposed attack improves identification accuracy over theexisting work by about 75\% for reconstructing face images from astate-of-the-art face recognition classifier. We also show that differentialprivacy, in its canonical form, is of little avail to defend against ourattacks.

The Secret Revealer: Generative Model-Inversion Attacks Against Deep Neural Networks