Robustmix: Improving Robustness by Regularizing the Frequency Bias of Deep Nets
Not on the map. Why?
Abstract (arXiv)
Deep networks have achieved impressive results on a range of well-curated benchmark datasets. Surprisingly, their performance remains sensitive to perturbations that have little effect on human performance. In this work, we propose a novel extension of Mixup called Robustmix that regularizes networks to classify based on lower-frequency spatial features. We show that this type of regularization improves robustness on a range of benchmarks such as Imagenet-C and Stylized Imagenet. It adds little computational overhead and, furthermore, does not require a priori knowledge of a large set of image transformations. We find that this approach further complements recent advances in model architecture and data augmentation, attaining a state-of-the-art mCE of 44.8 with an EfficientNet-B8 model and RandAugment, which is a reduction of 16 mCE compared to the baseline.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 31 of its 37 references that are in the map. First posted on arXiv 6 Apr 2023.
Builds on
31 of 37 references.
References · 31
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
- 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,617
- 2014Explaining and Harnessing Adversarial Examplescited by 7,287
- 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
- 2013Intriguing properties of neural networkscited by 5,114
- 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 4,993
- 2019CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Featurescited by 2,875
- 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,857
- 2019RandAugment: Practical automated data augmentation with a reduced search spacecited by 2,261
- 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,842
- 2017Accurate, Large Minibatch SGD: Training ImageNet in 1 Hourcited by 1,479
19 more
- 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,430
- 2019AugMix: A Simple Data Processing Method to Improve Robustness and Uncertaintycited by 897
- 2016Universal adversarial perturbationscited by 774
- 2019Big Transfer (BiT): General Visual Representation Learningcited by 766
- 2020FDA: Fourier Domain Adaptation for Semantic Segmentationcited by 540
- 2019Making Convolutional Networks Shift-Invariant Againcited by 390
- 2014Deep Speech: Scaling up end-to-end speech recognitioncited by 354
- 2019High Frequency Component Helps Explain the Generalization of Convolutional Neural Networkscited by 310
- 2020Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixupcited by 260
- 2021Vision Transformers are Robust Learnerscited by 208
- 2021Towards Robust Vision Transformercited by 137
- 2019Learning Robust Representations by Projecting Superficial Statistics Outcited by 123
- 2021Fair Mixup: Fairness via Interpolationcited by 109
- 2018Low Frequency Adversarial Perturbationcited by 80
- 2018DARTS: Deceiving Autonomous Cars with Toxic Signscited by 53
- 2016Lets keep it simple, Using simple architectures to outperform deeper and more complex architecturescited by 24
- 2019MetaMixUp: Learning Adaptive Interpolation Policy of MixUp with Meta-Learningcited by 21
- 2020Addressing Neural Network Robustness with Mixup and Targeted Labeling Adversarial Trainingcited by 10
- 2021Pose Adaptive Dual Mixup for Few-Shot Single-View 3D Reconstructioncited by 5
Built on by
Citing papers per year: 2023: 1; 2024: 1; 2025: 1.
3 papers.