Unsupervised data augmentation (UDA) from Google...better than mixup

[jamesrequa]

I also shared earlier here

Good readings 2019

Very exciting new paper “Unsupervised Data Augmentation” just out today, if the claims are accurate Works with all types of data including text and images. On the IMDb text classification dataset, with only 20 labeled examples, UDA outperforms the state-of-the-art model trained on 25,000 labeled examples. On standard semi-supervised learning benchmarks, CIFAR-10 with 4,000 examples and SVHN with 1,000 examples, UDA outperforms all previous approaches and reduces more than 30% of…

@LessW2020

outperforms pretty much all other data augmentation including mixup for cases where labelled data is small…

They actually compared UDA with an alternative method of mixup called mixmixup which I couldn’t find any info on – did anyone else find it?. UPDATE: here is the mixmixup paper for those interested: https://openreview.net/pdf?id=r1gp1jRN_4

Overall, I don’t think this paper gave enough credit to mixup (I now use mixup as default in all of my projects – the results are that good) and in fact I think they could potentially be used together for even better performance. Anyway, UDA is still exciting nonetheless if the claims are accurate.