FUDGE: Controlled Text Generation With Future Discriminators

Posted on Thu, Jul 22, 2021 NLP TLDR논문리뷰

Do You Even Need Attention? A Stack of Feed-Forward Layers Does Surprisingly Well on ImageNet

ViT에서 Transformer Attention을 단순한 FF Layer로 바꿨는데 성능이 비슷. 79.9(ViT) vs 77.9(FF Layer only)

Posted on Mon, May 24, 2021 TLDR논문리뷰 CV

FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders

ICLR2021, PLM(BERT)에 추가 모듈 붙이고, Contrastive learning + Regualizer로 Debiased된 output 추출하는 방법론.

Posted on Mon, May 24, 2021 NLP TLDR논문리뷰