arXiv 2019

Unsupervised Data Augmentation

arXiv 2019 google-research/uda

Unlike previous methods that use random noise such as Gaussian noise or dropout noise, UDA has a small twist in that it makes use of harder and more realistic noise generated by state-of-the-art data augmentation methods.

DATA AUGMENTATION SENTIMENT ANALYSIS TEXT CLASSIFICATION