Matches in SemOpenAlex for { <https://semopenalex.org/work/W3170160219> ?p ?o ?g. }
Showing items 1 to 94 of
94
with 100 items per page.
- W3170160219 abstract "Pre-trained contextualized language models (PrLMs) have led to strong performance gains in downstream natural language understanding tasks. However, PrLMs can still be easily fooled by adversarial word substitution, which is one of the most challenging textual adversarial attack methods. Existing defence approaches suffer from notable performance loss and complexities. Thus, this paper presents a compact and performance-preserved framework, Anomaly Detection with Frequency-Aware Randomization (ADFAR). In detail, we design an auxiliary anomaly detection classifier and adopt a multi-task learning procedure, by which PrLMs are able to distinguish adversarial input samples. Then, in order to defend adversarial word substitution, a frequency-aware randomization process is applied to those recognized adversarial input samples. Empirical results show that ADFAR significantly outperforms those newly proposed defense methods over various tasks with much higher inference speed. Remarkably, ADFAR does not impair the overall performance of PrLMs. The code is available at https://github.com/LilyNLP/ADFAR" @default.
- W3170160219 created "2021-06-22" @default.
- W3170160219 creator A5036050911 @default.
- W3170160219 creator A5055860272 @default.
- W3170160219 creator A5082114171 @default.
- W3170160219 date "2021-05-30" @default.
- W3170160219 modified "2023-09-24" @default.
- W3170160219 title "Defending Pre-trained Language Models from Adversarial Word Substitutions Without Performance Sacrifice" @default.
- W3170160219 cites W2163455955 @default.
- W3170160219 cites W2251939518 @default.
- W3170160219 cites W2289952147 @default.
- W3170160219 cites W2460937040 @default.
- W3170160219 cites W2737504179 @default.
- W3170160219 cites W2767962654 @default.
- W3170160219 cites W2768282280 @default.
- W3170160219 cites W2803392236 @default.
- W3170160219 cites W2946345909 @default.
- W3170160219 cites W2947415936 @default.
- W3170160219 cites W2949128310 @default.
- W3170160219 cites W2951467133 @default.
- W3170160219 cites W2962818281 @default.
- W3170160219 cites W2963109131 @default.
- W3170160219 cites W2963341956 @default.
- W3170160219 cites W2965373594 @default.
- W3170160219 cites W2971296908 @default.
- W3170160219 cites W2971970905 @default.
- W3170160219 cites W2982756474 @default.
- W3170160219 cites W2998230451 @default.
- W3170160219 cites W3015189805 @default.
- W3170160219 cites W3035736465 @default.
- W3170160219 cites W3105604018 @default.
- W3170160219 doi "https://doi.org/10.48550/arxiv.2105.14553" @default.
- W3170160219 hasPublicationYear "2021" @default.
- W3170160219 type Work @default.
- W3170160219 sameAs 3170160219 @default.
- W3170160219 citedByCount "0" @default.
- W3170160219 crossrefType "posted-content" @default.
- W3170160219 hasAuthorship W3170160219A5036050911 @default.
- W3170160219 hasAuthorship W3170160219A5055860272 @default.
- W3170160219 hasAuthorship W3170160219A5082114171 @default.
- W3170160219 hasBestOaLocation W31701602191 @default.
- W3170160219 hasConcept C119857082 @default.
- W3170160219 hasConcept C137293760 @default.
- W3170160219 hasConcept C138885662 @default.
- W3170160219 hasConcept C154945302 @default.
- W3170160219 hasConcept C162324750 @default.
- W3170160219 hasConcept C177264268 @default.
- W3170160219 hasConcept C187736073 @default.
- W3170160219 hasConcept C195324797 @default.
- W3170160219 hasConcept C199360897 @default.
- W3170160219 hasConcept C204321447 @default.
- W3170160219 hasConcept C2776214188 @default.
- W3170160219 hasConcept C2776760102 @default.
- W3170160219 hasConcept C2780451532 @default.
- W3170160219 hasConcept C37736160 @default.
- W3170160219 hasConcept C41008148 @default.
- W3170160219 hasConcept C41895202 @default.
- W3170160219 hasConcept C90805587 @default.
- W3170160219 hasConcept C95623464 @default.
- W3170160219 hasConceptScore W3170160219C119857082 @default.
- W3170160219 hasConceptScore W3170160219C137293760 @default.
- W3170160219 hasConceptScore W3170160219C138885662 @default.
- W3170160219 hasConceptScore W3170160219C154945302 @default.
- W3170160219 hasConceptScore W3170160219C162324750 @default.
- W3170160219 hasConceptScore W3170160219C177264268 @default.
- W3170160219 hasConceptScore W3170160219C187736073 @default.
- W3170160219 hasConceptScore W3170160219C195324797 @default.
- W3170160219 hasConceptScore W3170160219C199360897 @default.
- W3170160219 hasConceptScore W3170160219C204321447 @default.
- W3170160219 hasConceptScore W3170160219C2776214188 @default.
- W3170160219 hasConceptScore W3170160219C2776760102 @default.
- W3170160219 hasConceptScore W3170160219C2780451532 @default.
- W3170160219 hasConceptScore W3170160219C37736160 @default.
- W3170160219 hasConceptScore W3170160219C41008148 @default.
- W3170160219 hasConceptScore W3170160219C41895202 @default.
- W3170160219 hasConceptScore W3170160219C90805587 @default.
- W3170160219 hasConceptScore W3170160219C95623464 @default.
- W3170160219 hasLocation W31701602191 @default.
- W3170160219 hasOpenAccess W3170160219 @default.
- W3170160219 hasPrimaryLocation W31701602191 @default.
- W3170160219 hasRelatedWork W2035959783 @default.
- W3170160219 hasRelatedWork W2883550961 @default.
- W3170160219 hasRelatedWork W3107474891 @default.
- W3170160219 hasRelatedWork W3115399078 @default.
- W3170160219 hasRelatedWork W3118032223 @default.
- W3170160219 hasRelatedWork W3119555019 @default.
- W3170160219 hasRelatedWork W3173996958 @default.
- W3170160219 hasRelatedWork W3184371396 @default.
- W3170160219 hasRelatedWork W3185852197 @default.
- W3170160219 hasRelatedWork W4287548963 @default.
- W3170160219 isParatext "false" @default.
- W3170160219 isRetracted "false" @default.
- W3170160219 magId "3170160219" @default.
- W3170160219 workType "article" @default.