Open AkiraTOSEI opened 3 years ago
paper proposes MADAO, data augmentation method that uses the outer loop to increase the validation score like a meta-learning. Because it is not possible to differentiate categorical distribution, they use Gumbel softmax.
https://arxiv.org/abs/2006.07965
TL;DR
paper proposes MADAO, data augmentation method that uses the outer loop to increase the validation score like a meta-learning. Because it is not possible to differentiate categorical distribution, they use Gumbel softmax.
Why it matters:
Paper URL
https://arxiv.org/abs/2006.07965
Submission Dates(yyyy/mm/dd)
Authors and institutions
Methods
Results
Comments