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 studies show that deep neural networks (DNNs) are vulnerable tobackdoor attacks. A backdoor DNN model behaves normally with clean inputs,whereas outputs attacker's expected behaviors when the inputs contain apre-defined pattern called a trigger. However, in some tasks, the attackercannot know the exact target that shows his/her expected behavior, because thetask may contain a large number of classes and the attacker does not have fullaccess to know the semantic details of these classes. Thus, the attacker iswilling to attack multiple suspected targets to achieve his/her purpose. Inlight of this, in this paper, we propose the M-to-N backdoor attack, a newattack paradigm that allows an attacker to launch a fuzzy attack bysimultaneously attacking N suspected targets, and each of the N targets can beactivated by any one of its M triggers. To achieve a better stealthiness, werandomly select M clean images from the training dataset as our triggers foreach target. Since the triggers used in our attack have the same distributionas the clean images, the inputs poisoned by the triggers are difficult to bedetected by the input-based defenses, and the backdoor models trained on thepoisoned training dataset are also difficult to be detected by the model-baseddefenses. Thus, our attack is stealthier and has a higher probability ofachieving the attack purpose by attacking multiple suspected targetssimultaneously in contrast to prior backdoor attacks. Extensive experimentsshow that our attack is effective against different datasets with variousmodels and achieves high attack success rates (e.g., 99.43% for attacking 2targets and 98.23% for attacking 4 targets on the CIFAR-10 dataset) whenpoisoning only an extremely small portion of the training dataset (e.g., lessthan 2%). Besides, it is robust to pre-processing operations and can resiststate-of-the-art defenses.

M-to-N Backdoor Paradigm: A Stealthy and Fuzzy Attack to Deep Learning Models