f708f064faaf32a43e4d3c784e6af9ea-AuthorFeedback.pdf

Weagree in the sense that there is definitely a'prior' induced with mixup (e.g. Furthermore, thepaper'manifold mixup' addresses thisconcern andshowsthatmixupbehaves29 differently to other regularisation schemes, and is competitive with strong baselines (though only uses mixup with30 k =2).