Twitter | Pretraživanje | |
Quoc Le 25. stu
AdvProp: One weird trick to use adversarial examples to reduce overfitting. Key idea is to use two BatchNorms, one for normal examples and another one for adversarial examples. Significant gains on ImageNet and other test sets.
Prikaži podatke · Reply Retweet Označi sa "sviđa mi se"
Quoc Le
Many of us tried to use adversarial examples as data augmentation and observed a drop in accuracy. And it seems that simply using two BatchNorms overcomes this mysterious drop in accuracy.
13:35 - 25. stu 2019.
Reply Retweet Označi sa "sviđa mi se" More
Quoc Le 25. stu
Odgovor korisniku/ci @quocleix
As a data augmentation method, adversarial examples are more general than other image processing techniques. So I expect AdvProp to be useful everywhere (language, structured data etc.), not just image recognition.
Prikaz razgovora · Reply Retweet Označi sa "sviđa mi se"
Quoc Le 26. stu
Odgovor korisniku/ci @quocleix
AdvProp improves accuracy for a wide range of image models, from small to large. But the improvement seems bigger when the model is larger.
Prikaz razgovora · Reply Retweet Označi sa "sviđa mi se"
Quoc Le 26. stu
Odgovor korisniku/ci @wightmanr
Pretrained checkpoints in Pytorch: h/t to
Prikaz razgovora · Reply Retweet Označi sa "sviđa mi se"