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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Backdoors and adversarial examples are the two primary threats currentlyfaced by deep neural networks (DNNs). Both attacks attempt to hijack the modelbehaviors with unintended outputs by introducing (small) perturbations to theinputs. Backdoor attacks, despite the high success rates, often require astrong assumption, which is not always easy to achieve in reality. Adversarialexample attacks, which put relatively weaker assumptions on attackers, oftendemand high computational resources, yet do not always yield satisfactorysuccess rates when attacking mainstream black-box models in the real world.These limitations motivate the following research question: can model hijackingbe achieved more simply, with a higher attack success rate and more reasonableassumptions? In this paper, we propose CleanSheet, a new model hijacking attackthat obtains the high performance of backdoor attacks without requiring theadversary to tamper with the model training process. CleanSheet exploitsvulnerabilities in DNNs stemming from the training data. Specifically, our keyidea is to treat part of the clean training data of the target model as"poisoned data," and capture the characteristics of these data that are moresensitive to the model (typically called robust features) to construct"triggers." These triggers can be added to any input example to mislead thetarget model, similar to backdoor attacks. We validate the effectiveness ofCleanSheet through extensive experiments on 5 datasets, 79 normally trainedmodels, 68 pruned models, and 39 defensive models. Results show that CleanSheetexhibits performance comparable to state-of-the-art backdoor attacks, achievingan average attack success rate (ASR) of 97.5% on CIFAR-100 and 92.4% on GTSRB,respectively. Furthermore, CleanSheet consistently maintains a high ASR, whenconfronted with various mainstream backdoor defenses.

Hijacking Attacks against Neural Networks by Analyzing Training Data