Matches in SemOpenAlex for { <https://semopenalex.org/work/W2996564207> ?p ?o ?g. }
- W2996564207 abstract "Adversarial training, which minimizes the maximal risk for label-preserving input perturbations, has proved to be effective for improving the generalization of language models. In this work, we propose a novel adversarial training algorithm - FreeLB, that promotes higher robustness and invariance in the embedding space, by adding adversarial perturbations to word embeddings and minimizing the resultant adversarial risk inside different regions around input samples. To validate the effectiveness of the proposed approach, we apply it to Transformer-based models for natural language understanding and commonsense reasoning tasks. Experiments on the GLUE benchmark show that when applied only to the finetuning stage, it is able to improve the overall test scores of BERT-based model from 78.3 to 79.4, and RoBERTa-large model from 88.5 to 88.8. In addition, the proposed approach achieves state-of-the-art test accuracies of 85.39% and 67.32% on ARC-Easy and ARC-Challenge. Experiments on CommonsenseQA benchmark further demonstrate that FreeLB can be generalized and boost the performance of RoBERTa-large model on other tasks as well." @default.
- W2996564207 created "2019-12-26" @default.
- W2996564207 creator A5023777406 @default.
- W2996564207 creator A5026746295 @default.
- W2996564207 creator A5066666034 @default.
- W2996564207 creator A5072804127 @default.
- W2996564207 creator A5082943001 @default.
- W2996564207 date "2020-04-30" @default.
- W2996564207 modified "2023-09-24" @default.
- W2996564207 title "FreeLB: Enhanced Adversarial Training for Language Understanding" @default.
- W2996564207 cites W131533222 @default.
- W2996564207 cites W1599016936 @default.
- W2996564207 cites W1871489475 @default.
- W2996564207 cites W1892947258 @default.
- W2996564207 cites W2095705004 @default.
- W2996564207 cites W2130158090 @default.
- W2996564207 cites W2251939518 @default.
- W2996564207 cites W2396767181 @default.
- W2996564207 cites W2606964149 @default.
- W2996564207 cites W2787708942 @default.
- W2996564207 cites W2794325560 @default.
- W2996564207 cites W2799007037 @default.
- W2996564207 cites W2799194071 @default.
- W2996564207 cites W2890894339 @default.
- W2996564207 cites W2895595267 @default.
- W2996564207 cites W2942836458 @default.
- W2996564207 cites W2962713901 @default.
- W2996564207 cites W2962784628 @default.
- W2996564207 cites W2962872506 @default.
- W2996564207 cites W2963080814 @default.
- W2996564207 cites W2963126845 @default.
- W2996564207 cites W2963207607 @default.
- W2996564207 cites W2963266340 @default.
- W2996564207 cites W2963310665 @default.
- W2996564207 cites W2963341956 @default.
- W2996564207 cites W2963389226 @default.
- W2996564207 cites W2963661177 @default.
- W2996564207 cites W2963748441 @default.
- W2996564207 cites W2963823140 @default.
- W2996564207 cites W2963846996 @default.
- W2996564207 cites W2963854351 @default.
- W2996564207 cites W2963969878 @default.
- W2996564207 cites W2963995027 @default.
- W2996564207 cites W2964159205 @default.
- W2996564207 cites W2964207259 @default.
- W2996564207 cites W2964222271 @default.
- W2996564207 cites W2964253222 @default.
- W2996564207 cites W2965373594 @default.
- W2996564207 cites W2970049488 @default.
- W2996564207 cites W2970317235 @default.
- W2996564207 cites W2970597249 @default.
- W2996564207 cites W2978670439 @default.
- W2996564207 cites W3035281947 @default.
- W2996564207 cites W806995027 @default.
- W2996564207 hasPublicationYear "2020" @default.
- W2996564207 type Work @default.
- W2996564207 sameAs 2996564207 @default.
- W2996564207 citedByCount "1" @default.
- W2996564207 countsByYear W29965642072020 @default.
- W2996564207 crossrefType "proceedings-article" @default.
- W2996564207 hasAuthorship W2996564207A5023777406 @default.
- W2996564207 hasAuthorship W2996564207A5026746295 @default.
- W2996564207 hasAuthorship W2996564207A5066666034 @default.
- W2996564207 hasAuthorship W2996564207A5072804127 @default.
- W2996564207 hasAuthorship W2996564207A5082943001 @default.
- W2996564207 hasConcept C104317684 @default.
- W2996564207 hasConcept C119599485 @default.
- W2996564207 hasConcept C119857082 @default.
- W2996564207 hasConcept C127413603 @default.
- W2996564207 hasConcept C13280743 @default.
- W2996564207 hasConcept C134306372 @default.
- W2996564207 hasConcept C137293760 @default.
- W2996564207 hasConcept C154945302 @default.
- W2996564207 hasConcept C165801399 @default.
- W2996564207 hasConcept C177148314 @default.
- W2996564207 hasConcept C185592680 @default.
- W2996564207 hasConcept C185798385 @default.
- W2996564207 hasConcept C193221554 @default.
- W2996564207 hasConcept C195324797 @default.
- W2996564207 hasConcept C205649164 @default.
- W2996564207 hasConcept C2779439875 @default.
- W2996564207 hasConcept C33923547 @default.
- W2996564207 hasConcept C37736160 @default.
- W2996564207 hasConcept C41008148 @default.
- W2996564207 hasConcept C41608201 @default.
- W2996564207 hasConcept C55493867 @default.
- W2996564207 hasConcept C63479239 @default.
- W2996564207 hasConcept C66322947 @default.
- W2996564207 hasConceptScore W2996564207C104317684 @default.
- W2996564207 hasConceptScore W2996564207C119599485 @default.
- W2996564207 hasConceptScore W2996564207C119857082 @default.
- W2996564207 hasConceptScore W2996564207C127413603 @default.
- W2996564207 hasConceptScore W2996564207C13280743 @default.
- W2996564207 hasConceptScore W2996564207C134306372 @default.
- W2996564207 hasConceptScore W2996564207C137293760 @default.
- W2996564207 hasConceptScore W2996564207C154945302 @default.
- W2996564207 hasConceptScore W2996564207C165801399 @default.
- W2996564207 hasConceptScore W2996564207C177148314 @default.
- W2996564207 hasConceptScore W2996564207C185592680 @default.
- W2996564207 hasConceptScore W2996564207C185798385 @default.