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

Researchers have recently found that Self-Supervised Learning (SSL) isvulnerable to backdoor attacks. The attacker can embed hidden SSL backdoors viaa few poisoned examples in the training dataset and maliciously manipulate thebehavior of downstream models. To defend against SSL backdoor attacks, afeasible route is to detect and remove the poisonous samples in the trainingset. However, the existing SSL backdoor defense method fails to detect thepoisonous samples precisely. In this paper, we propose to erase the SSLbackdoor by cluster activation masking and propose a novel PoisonCAM method.After obtaining the threat model trained on the poisoned dataset, our methodcan precisely detect poisonous samples based on the assumption that masking thebackdoor trigger can effectively change the activation of a downstreamclustering model. In experiments, our PoisonCAM achieves 96% accuracy forbackdoor trigger detection compared to 3% of the state-of-the-art method onpoisoned ImageNet-100. Moreover, our proposed PoisonCAM significantly improvesthe performance of the trained SSL model under backdoor attacks compared to thestate-of-the-art method. Our code will be available athttps://github.com/LivXue/PoisonCAM.

Erasing Self-Supervised Learning Backdoor by Cluster Activation Masking