Papers of the day   All papers

Adversarial Examples Improve Image Recognition

Comments

Quoc Le: 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.

3 replies, 569 likes


Mingxing Tan: Can adversarial examples improve image recognition? Check out our recent work: AdvProp, achieving ImageNet top-1 accuracy 85.5% (no extra data) with adversarial examples! Arxiv: https://arxiv.org/abs/1911.09665 Checkpoints: https://git.io/JeopW https://t.co/bAu054LGt2

2 replies, 376 likes


Daisuke Okanohara: NN training with adversarial examples doesn't improve the generalization ability because of the distribution gap between clean and adversarial ones. Only using different BNs for clean and adversarial can solve this problem, achieving new SOTA ImageNet acc https://arxiv.org/abs/1911.09665

0 replies, 19 likes


arXiv CS-CV: Adversarial Examples Improve Image Recognition http://arxiv.org/abs/1911.09665

0 replies, 4 likes


akira: https://arxiv.org/abs/1911.09665 Significantly improved accuracy of ImageNet and ImageNet with noise using adversarial samples. By separating BN from adversarial one and normal data, changes in distribution due to data mixing are prevented. Simple but quite powerful. https://t.co/jGFNaUIkha

1 replies, 3 likes


arXiv CS-CV: Adversarial Examples Improve Image Recognition http://arxiv.org/abs/1911.09665

0 replies, 2 likes


Content

Found on Nov 25 2019 at https://arxiv.org/pdf/1911.09665.pdf

PDF content of a computer science paper: Adversarial Examples Improve Image Recognition