Nothing Special   »   [go: up one dir, main page]
Skip to main content

Advertisement

Springer Nature Link
Log in

BO-Aug: learning data augmentation policies via Bayesian optimization

  • Published:
Applied Intelligence Aims and scope Submit manuscript

Abstract

Data augmentation has been an essential technique to increase the amount and diversity of datasets, thus improving deep learning models. To pursue further performance, several automated data augmentation approaches have recently been proposed to find data augmentation policies automatically. However, there are still some key issues that deserve further exploration, i.e., a precise policy search space definition, the instructive policy evaluation method, and the low computational cost of policy search. In this paper, we propose a novel method named BO-Aug that attempts to solve the above issues. Empirical verification on three widely used image classification datasets shows that the proposed method can achieve state-of-the-art or comparable performance compared with advanced automated data augmentation methods, with a relatively low cost. Our code is available at https://github.com/Zhangcx19/BO-Aug.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Subscribe and save

Springer+ Basic
$34.99 /Month
  • Get 10 units per month
  • Download Article/Chapter or eBook
  • 1 Unit = 1 Article or 1 Chapter
  • Cancel anytime
Subscribe now

Buy Now

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3

Similar content being viewed by others

Explore related subjects

Discover the latest articles, news and stories from top researchers in related subjects.

Notes

  1. https://pillow.readthedocs.io/en/stable/

  2. https://tensorflow.google.cn/

References

  1. Cubuk ED, Zoph B, Mane D, Vasudevan V, Le QV (2019) Autoaugment: learning augmentation policies from data. In: Proceedings of the 32nd IEEE conference on computer vision and pattern recognition, pp 113–123

  2. Ho D, Liang E, Chen X, Stoica I, Abbeel P (2019) Population based augmentation: efficient learning of augmentation policy schedules. In: Proceedings of the 36th international conference on machine learning, pp 2731–2741

  3. Lim S, Kim I, Kim T, Kim C, Kim S (2019) Fast autoaugment. In: Proceedings of the 33rd advances in neural information processing systems, pp 6665–6675

  4. Takase T, Karakida R, Asoh H (2021) Self-paced data augmentation for training neural networks. Neurocomputing 442:296–306

    Article  Google Scholar 

  5. Bandara K, Hewamalage H, Liu Y-H, Kang Y, Bergmeir C (2021) Improving the accuracy of global forecasting models using time series data augmentation. Pattern Recogn 120:108148

    Article  Google Scholar 

  6. Wang Y, Wei X, Tang X, Shen H, Ding L (2020) Cnn tracking based on data augmentation. Knowl-Based Syst 194:105594

    Article  Google Scholar 

  7. Leng Y, Zhao W, Lin C, Sun C, Wang R, Yuan Q, Li D (2020) Lda-based data augmentation algorithm for acoustic scene classification. Knowl-Based Syst 195:105600

    Article  Google Scholar 

  8. Yun S, Han D, Oh SJ, Chun S, Choe J, Yoo Y (2019) Cutmix: regularization strategy to train strong classifiers with localizable features. In: Proceedings of the 17th international conference on computer vision, pp 6023–6032

  9. Tolstikhin IO, Houlsby N, Kolesnikov A, Beyer L, Zhai X, Unterthiner T, Yung J, Steiner A, Keysers D, Uszkoreit J et al (2021) Mlp-mixer: an all-mlp architecture for vision. Adv Neural Inf Process Syst, vol 34

  10. Sohn K, Berthelot D, Carlini N, Zhang Z, Zhang H, Raffel C, Cubuk ED, Kurakin A, Li C (2020) Fixmatch: simplifying semi-supervised learning with consistency and confidence. In: Proceedings of the 34th advances in neural information processing systems, pp 596–608

  11. Zoph B, Ghiasi G, Lin T, Cui Y, Liu H, Cubuk ED, Le Q (2020) Rethinking pre-training and self-training. In: Proceedings of the 34th advances in neural information processing systems, pp 3833–3845

  12. Xie Q, Dai Z, Hovy EH, Luong T, Le Q (2020) Unsupervised data augmentation for consistency training. In: Proceedings of the 34th advances in neural information processing systems, pp 6256–6268

  13. Caron M, Misra I, Mairal J, Goyal P, Bojanowski P, Joulin A (2020) Unsupervised learning of visual features by contrasting cluster assignments. In: Proceedings of the 34th advances in neural information processing systems, pp 9912–9924

  14. He K, Fan H, Wu Y, Xie S, Girshick RB (2020) Momentum contrast for unsupervised visual representation learning. In: Proceedings of the 34th advances in neural information processing systems, pp 9726–9735

  15. Misra I, Maaten LVD (2020) Self-supervised learning of pretext-invariant representations. In: Proceedings of the 33rd IEEE conference on computer vision and pattern recognition, pp 6706–6716

  16. Chen T, Kornblith S, Swersky K, Norouzi M, Hinton GE (2020) Big self-supervised models are strong semi-supervised learners. In: Proceedings of the 34th advances in neural information processing systems, pp 22243–22255

  17. Chen T, Kornblith S, Norouzi M, Hinton GE (2020) A simple framework for contrastive learning of visual representations. In: Proceedings of the 37th international conference on machine learning, pp 1597–1607

  18. Hu J, Shen L, Sun G (2018) Squeeze-and-excitation networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 7132–7141

  19. Tan M, Le Q (2019) Efficientnet: rethinking model scaling for convolutional neural networks. In: International conference on machine learning, pp 6105–6114

  20. DeVries T, Taylor GW (2017) Improved regularization of convolutional neural networks with cutout. arXiv:1708.04552

  21. Zhong Z, Zheng L, Kang G, Li S, Yang Y (2020) Random erasing data augmentation. In: Proceedings of the AAAI conference on artificial intelligence, vol 34, pp 13001–13008

  22. Zhang H, Cisse M, Dauphin YN, Lopez-paz D (2018) Mixup: beyond empirical risk minimization. In: International conference on learning representations

  23. Chinbat V, Bae S-H (2022) Ga3n: generative adversarial autoaugment network. Pattern Recogn 127:108637

    Article  Google Scholar 

  24. dos Santos Tanaka FHK, Aranha C (2019) Data augmentation using gans. Proc Mach Learn Res 1:16

    Google Scholar 

  25. Songyan Liu J-GH, Haiyun Guo X u, Zhao EA (2020) A novel data augmentation scheme for pedestrian detection with attribute preserving gan. Neurocomputing 401:123–132

    Article  Google Scholar 

  26. DeVries T, Taylor GW (2017) Dataset augmentation in feature space. In: Proceedings of the 34th international conference on machine learning, workshop track

  27. Wang Y, Pan X, Song S, Zhang H, Huang G, Wu C (2019) Implicit semantic data augmentation for deep networks. Adv Neural Inf Process Syst, vol 32

  28. Pham H, Guan M, Zoph B, Le Q, Dean J (2018) Efficient neural architecture search via parameters sharing. In: International conference on machine learning, pp 4095–4104

  29. Ding Z, Chen Y, Li N, Zhao D, Sun Z, Chen CP (2021) Bnas: efficient neural architecture search using broad scalable architecture. IEEE Trans Neural Netw Learn Syst

  30. Wei C, Niu C, Tang Y, Wang Y, Hu H, Liang J (2022) Npenas: neural predictor guided evolution for neural architecture search. IEEE Trans Neural Netw Learn Syst

  31. Chen Y, Gao R, Liu F, Zhao D (2021) Modulenet: knowledge-inherited neural architecture search. IEEE Trans Cybern

  32. Cubuk ED, Zoph B, Shlens J, Le QV (2020) Randaugment: practical automated data augmentation with a reduced search space. In: Proceedings of the 33rd IEEE conference on computer vision and pattern recognition, pp 3008–3017

  33. Naghizadeh A, Abavisani M, Metaxas DN (2020) Greedy autoaugment. Pattern Recogn Lett 138:624–630

    Article  Google Scholar 

  34. Tian K, Lin C, Sun M, Zhou L, Yan J, Ouyang W (2020) Improving auto-augment via augmentation-wise weight sharing. Adv Neural Inf Process Syst 33:19088–19098

    Google Scholar 

  35. Cui J, Yang B (2017) Survey on bayesian optimization methodology and applications. In: Journal of software, pp 176–198

  36. Lindauer M, Eggensperger K, Feurer M, Biedenkapp A, Deng D, Benjamins C, Ruhkopf T, Sass R, Hutter F (2022) Smac3: a versatile bayesian optimization package for hyperparameter optimization. J Mach Learn Res 23(54):1–9

    MATH  Google Scholar 

  37. Imani M, Ghoreishi SF (2021) Graph-based bayesian optimization for large-scale objective-based experimental design. IEEE Trans Neural Netw Learn Syst

  38. Du L, Gao R, Suganthan PN, Wang DZ (2022) Bayesian optimization based dynamic ensemble for time series forecasting. Inf Sci 591:155–175

    Article  Google Scholar 

  39. Turner R, Eriksson D, McCourt M, Kiili J, Laaksonen E, Xu Z, Guyon I (2021) Bayesian optimization is superior to random search for machine learning hyperparameter tuning: analysis of the black-box optimization challenge 2020. In: NeurIPS 2020 competition and demonstration track, pp 3–26

  40. Williams CK, Rasmussen CE (2006) Gaussian Processes for Machine Learning. MIT Press Cambridge, MA

  41. Krizhevsky A, Hinton G (2009) Learning multiple layers of features from tiny images. Technical report, Citeseer

  42. Netzer Y, Wang T, Coates A, Bissacco A, Wu B, Ng AY (2011) Reading digits in natural images with unsupervised feature learning. In: Proceedings of the 28th neural information processing systems workshop on deep learning and unsupervised feature learning

  43. Deng J, Dong W, Socher R, Li L-J, Li K, Fei-Fei L (2009) Imagenet: a large-scale hierarchical image database. In: IEEE conference on computer vision and pattern recognition, pp 248–255

  44. Zagoruyko S, Komodakis N (2016) Wide residual networks. Proceedings of the 27th British machine vision conference, pp 8701–8712

  45. Gastaldi X (2017) Shake-shake regularization. arXiv:1705.07485

Download references

Acknowledgements

This work was supported by the National Key R&D Program of China under Grant Nos. 2021ZD0112501 and 2021ZD0112502; the National Natural Science Foundation of China under Grant Nos. 62172185 and 61876069; Jilin Province Key Scientific and Technological Research and Development Project under Grant Nos. 20180201067GX and 20180201044GX; Jilin Province Natural Science Foundation under Grant No. 20200201036JC; and China Postdoctoral Science Foundation funded project under Grant No.2021M701388.

Author information

Authors and Affiliations

  1. College of Computer Science and Technology, Jilin University, No.2699, Qianjin Street, Chaoyang District, Changchun City, 130012, Jilin Province, China

    Chunxu Zhang, Ximing Li, Zijian Zhang, Jiaxu Cui & Bo Yang

  2. Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, No.2699, Qianjin Street, Chaoyang District, Changchun City, 130012, Jilin Province, China

    Chunxu Zhang, Ximing Li, Zijian Zhang, Jiaxu Cui & Bo Yang

Authors
  1. Chunxu Zhang
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Ximing Li
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Zijian Zhang
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Jiaxu Cui
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Bo Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Bo Yang.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendices

Appendix A: Image transformation operations used for DA policies search

All available image transformation functions during the search process are listed below. These functions accept an image and corresponding operation parameters as input, and output a transformed image. The range of magnitudes for each operation is shown in the third column. Some transformations do not use magnitude information (e.g., Invert and Equalize).

Table 8 Image transformation operations used for DA policies search
Full size table

Appendix B: Mapping relationship between operation type value of policy vector and operation types

For each policy, we regard its first dimension as the two DA operation types. Considering that there are 14 DA operations in total, we use a number between 0 and 196 to represent two operations. Here we list the mapping relationship of the value and corresponding operation types.

Table 9 Mapping relationship between operation type value of policy vector and operation types
Full size table

Appendix C: Hyperparameter configuration in the experiments

Once BO-Aug finds the optimal DA policies, we will validate the policies’ performance on different target models based on several datasets. Here we give the learning rate and weight decay values during target model training.

Table 10 Hyperparameter configuration in the experiments
Full size table

Appendix D: Policies found on reduced CIFAR-10

Table 11 Policies found on reduced CIFAR-10
Full size table

Appendix E: Policies found on reduced SVHN

Table 12 Policies found on reduced SVHN
Full size table

Appendix F: Policies found on reduced TinyImagenet

Table 13 Policies found on reduced TinyImagenet
Full size table

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Zhang, C., Li, X., Zhang, Z. et al. BO-Aug: learning data augmentation policies via Bayesian optimization. Appl Intell 53, 8978–8993 (2023). https://doi.org/10.1007/s10489-022-03790-z

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10489-022-03790-z

Keywords

Search

Navigation