Stackmix: a complementary mix algorithm

John Chen, Samarth Sinha, Anastasios Kyrillidis
Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, PMLR 180:326-335, 2022.

Abstract

Techniques combining multiple images as input/output have proven to be effective data augmentations for training convolutional neural networks. In this paper, we present StackMix: each input is presented as a concatenation of two images, and the label is the mean of the two one-hot labels. On its own, StackMix rivals other widely used methods in the “Mix” line of work. More importantly, unlike previous work, significant gains across a variety of benchmarks are achieved by combining StackMix with existing Mix augmentation, effectively mixing more than two images. E.g., by combining StackMix with CutMix, test error in the supervised setting is improved across a variety of settings over CutMix, including 0.8% on ImageNet, 3% on Tiny ImageNet, 2% on CIFAR-100, 0.5% on CIFAR-10, and 1.5% on STL-10. Similar results are achieved with Mixup. We further show that gains hold for robustness to common input corruptions and perturbations at varying severities with a 0.7% improvement on CIFAR-100-C, by combining StackMix with AugMix over AugMix. On its own, improvements with StackMix hold across different number of labeled samples on CIFAR-100, maintaining approximately a 2% gap in test accuracy –down to using only 5% of the whole dataset– and is effective in the semi-supervised setting with a 2% improvement with the standard benchmark Pi-model. Finally, we perform an extensive ablation study to better understand the proposed methodology.

Cite this Paper

BibTeX
@InProceedings{pmlr-v180-chen22b, title = {Stackmix: a complementary mix algorithm}, author = {Chen, John and Sinha, Samarth and Kyrillidis, Anastasios}, booktitle = {Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence}, pages = {326--335}, year = {2022}, editor = {Cussens, James and Zhang, Kun}, volume = {180}, series = {Proceedings of Machine Learning Research}, month = {01--05 Aug}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v180/chen22b/chen22b.pdf}, url = {https://proceedings.mlr.press/v180/chen22b.html}, abstract = {Techniques combining multiple images as input/output have proven to be effective data augmentations for training convolutional neural networks. In this paper, we present StackMix: each input is presented as a concatenation of two images, and the label is the mean of the two one-hot labels. On its own, StackMix rivals other widely used methods in the “Mix” line of work. More importantly, unlike previous work, significant gains across a variety of benchmarks are achieved by combining StackMix with existing Mix augmentation, effectively mixing more than two images. E.g., by combining StackMix with CutMix, test error in the supervised setting is improved across a variety of settings over CutMix, including 0.8% on ImageNet, 3% on Tiny ImageNet, 2% on CIFAR-100, 0.5% on CIFAR-10, and 1.5% on STL-10. Similar results are achieved with Mixup. We further show that gains hold for robustness to common input corruptions and perturbations at varying severities with a 0.7% improvement on CIFAR-100-C, by combining StackMix with AugMix over AugMix. On its own, improvements with StackMix hold across different number of labeled samples on CIFAR-100, maintaining approximately a 2% gap in test accuracy –down to using only 5% of the whole dataset– and is effective in the semi-supervised setting with a 2% improvement with the standard benchmark Pi-model. Finally, we perform an extensive ablation study to better understand the proposed methodology.} }
Endnote
%0 Conference Paper %T Stackmix: a complementary mix algorithm %A John Chen %A Samarth Sinha %A Anastasios Kyrillidis %B Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2022 %E James Cussens %E Kun Zhang %F pmlr-v180-chen22b %I PMLR %P 326--335 %U https://proceedings.mlr.press/v180/chen22b.html %V 180 %X Techniques combining multiple images as input/output have proven to be effective data augmentations for training convolutional neural networks. In this paper, we present StackMix: each input is presented as a concatenation of two images, and the label is the mean of the two one-hot labels. On its own, StackMix rivals other widely used methods in the “Mix” line of work. More importantly, unlike previous work, significant gains across a variety of benchmarks are achieved by combining StackMix with existing Mix augmentation, effectively mixing more than two images. E.g., by combining StackMix with CutMix, test error in the supervised setting is improved across a variety of settings over CutMix, including 0.8% on ImageNet, 3% on Tiny ImageNet, 2% on CIFAR-100, 0.5% on CIFAR-10, and 1.5% on STL-10. Similar results are achieved with Mixup. We further show that gains hold for robustness to common input corruptions and perturbations at varying severities with a 0.7% improvement on CIFAR-100-C, by combining StackMix with AugMix over AugMix. On its own, improvements with StackMix hold across different number of labeled samples on CIFAR-100, maintaining approximately a 2% gap in test accuracy –down to using only 5% of the whole dataset– and is effective in the semi-supervised setting with a 2% improvement with the standard benchmark Pi-model. Finally, we perform an extensive ablation study to better understand the proposed methodology.
APA
Chen, J., Sinha, S. & Kyrillidis, A.. (2022). Stackmix: a complementary mix algorithm. Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 180:326-335 Available from https://proceedings.mlr.press/v180/chen22b.html.

Related Material