CoDA: Contrast-Enhancing and Diversity-Promoting Data Augmentation for NLU

This video explains an interesting new paper for applying Data Augmentation to NLP. I thought the interpretation of Data Augmentation as constructing local neighborhoods was very interesting and like the way the MoCo loss compliments this intuition. I hope you found this video useful and it inspires your interest in Data Augmentation!

Content Links:
Paper: https://arxiv.org/pdf/2010.08670.pdf
On the Measure of Intelligence: https://arxiv.org/pdf/1911.01547.pdf
EDA: https://arxiv.org/pdf/1901.11196.pdf
AutoAugment: https://arxiv.org/pdf/1805.09501.pdf
MoCo: https://arxiv.org/pdf/1911.05722.pdf

Chapters
0:00 Beginning
1:10 Research Questions
1:50 High-Level Idea
3:30 Data Aug for Text
4:57 Stacking Ideas
5:28 The Measure of Intelligence
8:53 Augmentations Explored
11:16 MoCo quick recap
12:42 GLUE Benchmark
14:06 Results
18:40 Thoughts on Local Neighborhoods

Видео CoDA: Contrast-Enhancing and Diversity-Promoting Data Augmentation for NLU канала Connor Shorten