UniDrop: A Simple yet Effective Technique to Improve Transformer without Extra Cost

Abstract

Transformer architecture achieves great success in abundant natural language processing tasks. The over-parameterization of the Transformer model has motivated plenty of works to alleviate its overfitting for superior performances. With some explorations, we find simple techniques such as dropout, can greatly boost model performance with a careful design. Therefore, in this paper, we integrate different dropout techniques into the training of Transformer models. Specifically, we propose an approach named UniDrop to unite three different dropout techniques from fine-grain to coarse-grain, i.e., feature dropout, structure dropout, and data dropout. Theoretically, we demonstrate that these three dropouts play different roles from regularization perspectives. Empirically, we conduct experiments on both neural machine translation and text classification benchmark datasets. Extensive results indicate that Transformer with UniDrop can achieve around 1.5 BLEU improvement on IWSLT14 translation tasks, and better accuracy for the classification even using strong pre-trained RoBERTa as backbone.

CipherDAug: Ciphertext based Data Augmentation for Neural Machine Translation

“…By contrast, CipherDAug trains a single model, and improves the baseline transformer by 2.9 BLEU points on IWSLT14 De→En and about 2.2 BLEU points on the smaller datasets. (Zhu et al, 2020) 1x(+BERT) 36.11 MAT (Fan et al, 2020) 0.9x 36.22 UniDrop (Wu et al, 2021b) 1x 36.88 R-DROP (Liang et al, 2021) 1x 37.25 BiBERT (Xu et al, 2021) 1x(+BERT) 37.50…”

“…Macaron Net (Lu* et al, 2020) 35.40 BERT Fuse (Zhu et al, 2020) 36.11 MAT (Fan et al, 2020) 36.22 UniDrop (Wu et al, 2021b) 36.88 R-DROP (Liang et al, 2021) 37.25 BiBERT (Xu et al, 2021) 37.50…”

CipherDAug: Ciphertext based Data Augmentation for Neural Machine Translation

“…Macaron Net (Lu* et al, 2020) 35.40 BERT Fuse (Zhu et al, 2020) 36.11 MAT 36.22 UniDrop (Wu et al, 2021b) 36.88 R-DROP (Liang et al, 2021) 37.25 BiBERT (Xu et al, 2021) 37.50…”