Deep Perturbation Learning: Enhancing the Network Performance via Image Perturbations

Zifan Song, Xiao Gong, Guosheng Hu, Cairong Zhao
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:32273-32287, 2023.

Abstract

Image perturbation technique is widely used to generate adversarial examples to attack networks, greatly decreasing the performance of networks. Unlike the existing works, in this paper, we introduce a novel framework Deep Perturbation Learning (DPL), the new insights into understanding image perturbations, to enhance the performance of networks rather than decrease the performance. Specifically, we learn image perturbations to amend the data distribution of training set to improve the performance of networks. This optimization w.r.t data distribution is non-trivial. To approach this, we tactfully construct a differentiable optimization target w.r.t. image perturbations via minimizing the empirical risk. Then we propose an alternating optimization of the network weights and perturbations. DPL can easily be adapted to a wide spectrum of downstream tasks and backbone networks. Extensive experiments demonstrate the effectiveness of our DPL on 6 datasets (CIFAR-10, CIFAR100, ImageNet, MS-COCO, PASCAL VOC, and SBD) over 3 popular vision tasks (image classification, object detection, and semantic segmentation) with different backbone architectures (e.g., ResNet, MobileNet, and ViT).

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-song23c, title = {Deep Perturbation Learning: Enhancing the Network Performance via Image Perturbations}, author = {Song, Zifan and Gong, Xiao and Hu, Guosheng and Zhao, Cairong}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {32273--32287}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/song23c/song23c.pdf}, url = {https://proceedings.mlr.press/v202/song23c.html}, abstract = {Image perturbation technique is widely used to generate adversarial examples to attack networks, greatly decreasing the performance of networks. Unlike the existing works, in this paper, we introduce a novel framework Deep Perturbation Learning (DPL), the new insights into understanding image perturbations, to enhance the performance of networks rather than decrease the performance. Specifically, we learn image perturbations to amend the data distribution of training set to improve the performance of networks. This optimization w.r.t data distribution is non-trivial. To approach this, we tactfully construct a differentiable optimization target w.r.t. image perturbations via minimizing the empirical risk. Then we propose an alternating optimization of the network weights and perturbations. DPL can easily be adapted to a wide spectrum of downstream tasks and backbone networks. Extensive experiments demonstrate the effectiveness of our DPL on 6 datasets (CIFAR-10, CIFAR100, ImageNet, MS-COCO, PASCAL VOC, and SBD) over 3 popular vision tasks (image classification, object detection, and semantic segmentation) with different backbone architectures (e.g., ResNet, MobileNet, and ViT).} }
Endnote
%0 Conference Paper %T Deep Perturbation Learning: Enhancing the Network Performance via Image Perturbations %A Zifan Song %A Xiao Gong %A Guosheng Hu %A Cairong Zhao %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-song23c %I PMLR %P 32273--32287 %U https://proceedings.mlr.press/v202/song23c.html %V 202 %X Image perturbation technique is widely used to generate adversarial examples to attack networks, greatly decreasing the performance of networks. Unlike the existing works, in this paper, we introduce a novel framework Deep Perturbation Learning (DPL), the new insights into understanding image perturbations, to enhance the performance of networks rather than decrease the performance. Specifically, we learn image perturbations to amend the data distribution of training set to improve the performance of networks. This optimization w.r.t data distribution is non-trivial. To approach this, we tactfully construct a differentiable optimization target w.r.t. image perturbations via minimizing the empirical risk. Then we propose an alternating optimization of the network weights and perturbations. DPL can easily be adapted to a wide spectrum of downstream tasks and backbone networks. Extensive experiments demonstrate the effectiveness of our DPL on 6 datasets (CIFAR-10, CIFAR100, ImageNet, MS-COCO, PASCAL VOC, and SBD) over 3 popular vision tasks (image classification, object detection, and semantic segmentation) with different backbone architectures (e.g., ResNet, MobileNet, and ViT).
APA
Song, Z., Gong, X., Hu, G. & Zhao, C.. (2023). Deep Perturbation Learning: Enhancing the Network Performance via Image Perturbations. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:32273-32287 Available from https://proceedings.mlr.press/v202/song23c.html.

Related Material