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) and natural language processing (NLP) systemshave developed rapidly and have been widely used in various real-world fields.However, they have been shown to be vulnerable to backdoor attacks.Specifically, the adversary injects a backdoor into the model during thetraining phase, so that input samples with backdoor triggers are classified asthe target class. Some attacks have achieved high attack success rates on thepre-trained language models (LMs), but there have yet to be effective defensemethods. In this work, we propose a defense method based on deep model mutationtesting. Our main justification is that backdoor samples are much more robustthan clean samples if we impose random mutations on the LMs and that backdoorsare generalizable. We first confirm the effectiveness of model mutation testingin detecting backdoor samples and select the most appropriate mutationoperators. We then systematically defend against three extensively studiedbackdoor attack levels (i.e., char-level, word-level, and sentence-level) bydetecting backdoor samples. We also make the first attempt to defend againstthe latest style-level backdoor attacks. We evaluate our approach on threebenchmark datasets (i.e., IMDB, Yelp, and AG news) and three style transferdatasets (i.e., SST-2, Hate-speech, and AG news). The extensive experimentalresults demonstrate that our approach can detect backdoor samples moreefficiently and accurately than the three state-of-the-art defense approaches.

BDMMT: Backdoor Sample Detection for Language Models through Model Mutation Testing