Understanding Robustness of Transformers for Image Classification

Deep Convolutional Neural Networks (CNNs) have long been the architecture of choice for computer vision tasks. Recently, Transformer-based architectures like Vision Transformer (ViT) have matched or even surpassed ResNets for image classification. However, details of the Transformer architecture ±such as the use of non-overlapping patches± lead one to wonder whether these networks are as robust. In this paper, we perform an extensive study of a variety of different measures of robustness of ViT models and compare the findings to ResNet baselines. We investigate robustness to input perturbations as well as robustness to model perturbations. We find that when pre-trained with a sufficient amount of data, ViT models are at least as robust as the ResNet counterparts on a broad range of perturbations. We also find that Transformers are robust to the removal of almost any single layer, and that while activations from later layers are highly correlated with each other, they nevertheless play an important role in classification.

[1]  Andrew Zisserman,et al.  Very Deep Convolutional Networks for Large-Scale Image Recognition , 2014, ICLR.

[2]  Jürgen Schmidhuber,et al.  Highway and Residual Networks learn Unrolled Iterative Estimation , 2016, ICLR.

[3]  Kilian Q. Weinberger,et al.  Densely Connected Convolutional Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[4]  Martin Jaggi,et al.  On the Relationship between Self-Attention and Convolutional Layers , 2019, ICLR.

[5]  D. Song,et al.  The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization , 2020, 2021 IEEE/CVF International Conference on Computer Vision (ICCV).

[6]  Kunihiko Fukushima,et al.  Neocognitron: A Self-Organizing Neural Network Model for a Mechanism of Visual Pattern Recognition , 1982 .

[7]  Dumitru Erhan,et al.  Going deeper with convolutions , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[8]  Georg Heigold,et al.  An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale , 2021, ICLR.

[9]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[10]  Dawn Song,et al.  Natural Adversarial Examples , 2019, 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[11]  Quoc V. Le,et al.  EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks , 2019, ICML.

[12]  Bo Chen,et al.  MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications , 2017, ArXiv.

[13]  Omer Levy,et al.  Are Sixteen Heads Really Better than One? , 2019, NeurIPS.

[14]  Omer Levy,et al.  Improving Transformer Models by Reordering their Sublayers , 2020, ACL.

[15]  Matthieu Cord,et al.  Training data-efficient image transformers & distillation through attention , 2020, ICML.

[16]  Mohit Bansal,et al.  LXMERT: Learning Cross-Modality Encoder Representations from Transformers , 2019, EMNLP.

[17]  Chen Sun,et al.  Revisiting Unreasonable Effectiveness of Data in Deep Learning Era , 2017, 2017 IEEE International Conference on Computer Vision (ICCV).

[18]  Andreas Loukas,et al.  Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth , 2021, ICML.

[19]  Yoshua Bengio,et al.  Gradient-based learning applied to document recognition , 1998, Proc. IEEE.

[20]  Enhua Wu,et al.  Squeeze-and-Excitation Networks , 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[21]  Abhinav Gupta,et al.  Non-local Neural Networks , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[22]  Lucas Beyer,et al.  Big Transfer (BiT): General Visual Representation Learning , 2020, ECCV.

[23]  知秀 柴田.  5分で分かる!? 有名論文ナナメ読み：Jacob Devlin et al. : BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding , 2020 .

[24]  Fedor Moiseev,et al.  Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned , 2019, ACL.

[25]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[26]  Cordelia Schmid,et al.  VideoBERT: A Joint Model for Video and Language Representation Learning , 2019, 2019 IEEE/CVF International Conference on Computer Vision (ICCV).

[27]  Stefan Lee,et al.  ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks , 2019, NeurIPS.

[28]  Jonathon Shlens,et al.  Explaining and Harnessing Adversarial Examples , 2014, ICLR.

[29]  Vladlen Koltun,et al.  Exploring Self-Attention for Image Recognition , 2020, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[30]  Fahad Shahbaz Khan,et al.  Transformers in Vision: A Survey , 2021, ACM Comput. Surv..

[31]  Mark Chen,et al.  Language Models are Few-Shot Learners , 2020, NeurIPS.

[32]  Lukasz Kaiser,et al.  Attention is All you Need , 2017, NIPS.

[33]  Matthias Bethge,et al.  ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness , 2018, ICLR.

[34]  Marcus Rohrbach,et al.  12-in-1: Multi-Task Vision and Language Representation Learning , 2020, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[35]  Yoshua Bengio,et al.  Show, Attend and Tell: Neural Image Caption Generation with Visual Attention , 2015, ICML.

[36]  Ahmed El Kholy,et al.  UNITER: Learning UNiversal Image-TExt Representations , 2019, ECCV 2020.

[37]  Ming-Wei Chang,et al.  BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , 2019, NAACL.

[38]  Serge J. Belongie,et al.  Residual Networks Behave Like Ensembles of Relatively Shallow Networks , 2016, NIPS.

[39]  Mark Chen,et al.  Generative Pretraining From Pixels , 2020, ICML.

[40]  Joan Bruna,et al.  Intriguing properties of neural networks , 2013, ICLR.

[41]  Jitendra Malik,et al.  Deformable part models are convolutional neural networks , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[42]  Georg Heigold,et al.  Object-Centric Learning with Slot Attention , 2020, NeurIPS.

[43]  Alec Radford,et al.  Scaling Laws for Neural Language Models , 2020, ArXiv.

[44]  Dawn Song,et al.  Pretrained Transformers Improve Out-of-Distribution Robustness , 2020, ACL.

[45]  Razvan Pascanu,et al.  Relational inductive biases, deep learning, and graph networks , 2018, ArXiv.

[46]  Alice Caplier,et al.  Are Adversarial Robustness and Common Perturbation Robustness Independent Attributes ? , 2019, 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW).

[47]  Thomas G. Dietterich,et al.  Benchmarking Neural Network Robustness to Common Corruptions and Perturbations , 2018, ICLR.

[48]  Aleksander Madry,et al.  Towards Deep Learning Models Resistant to Adversarial Attacks , 2017, ICLR.

[49]  Aleksander Madry,et al.  Exploring the Landscape of Spatial Robustness , 2017, ICML.

[50]  Ilya Sutskever,et al.  Learning Transferable Visual Models From Natural Language Supervision , 2021, ICML.