[2011.08367v1] Extreme Value Preserving Networksopen searchopen navigation menucontact arXivsubscribe to arXiv mailings

Recent evidence shows that convolutional neural networks (CNNs) are biased towards textures so that CNNs are non-robust to adversarial perturbations over textures, while traditional robust visual features like SIFT (scale-invariant feature transforms) are designed to be robust across a substantial range of affine distortion, addition of noise, etc with the mimic of human perception nature. This paper aims to leverage good properties of SIFT to renovate CNN architectures towards better accuracy and robustness. We borrow the scale-space extreme value idea from SIFT, and propose extreme value preserving networks (EVPNets). Experiments demonstrate that EVPNets can achieve similar or better accuracy than conventional CNNs, while achieving much better robustness on a set of adversarial attacks (FGSM,PGD,etc) even without adversarial training.

1 mentions: @agatan_
Date: 2020/11/18 06:53

Referring Tweets

@agatan_ 📝 SIFTはよく考えられていて、人間の知覚に近いロバスト性を持っているので、そこからアイディアを借りてCNNを再設計することでadversarial attackに強いモデルを作る、という話 t.co/h4nI7N7STU

Related Entries

Read more 5th Place Solution: Lions | Kaggle
0 users, 1 mentions 2019/10/08 15:48
Read more Kaggle Days Shop
0 users, 3 mentions 2020/05/27 11:20
Read more [2007.00649] Group Ensemble: Learning an Ensemble of ConvNets in a single ConvNetopen searchopen nav...
0 users, 2 mentions 2020/07/02 17:21
Read more Kaggle M5 Forecasting - Accuracy 42nd place solution - Speaker Deck
0 users, 3 mentions 2020/07/20 02:21
Read more Google Research Football with Manchester City F.C. | Kaggle
0 users, 8 mentions 2020/09/28 21:51