基于联合对抗训练的鲁棒度量迁移
收稿日期: 2023-01-08
网络出版日期: 2023-03-28
基金资助
国家重点研发计划资助项目(2021YFC3300200);湖北珞珈实验室开放基金资助项目(220100014);国家自然科学基金资助项目(62002090);国家自然科学基金资助项目(62276195);新加坡国家研究基金资助项目(NRF2020NRF-CG001-027)
Robust metric transfer using joint adversarial training
Received date: 2023-01-08
Online published: 2023-03-28
迁移度量学习旨在从强大且可靠的距离度量中迁移知识来改善目标度量的效果, 这些度量往往来自于学习目标相关的任务. 现有的迁移度量学习算法仅关注于如何迁移知识, 而这些知识容易过拟合到源域中. 首先研究如何在源域中训练一个适合于迁移的源域度量, 然后设计了一种通用的深度异质迁移算法来进行高效的迁移学习. 值得注意的是, 将源域度量以联合对抗学习的方式进行训练, 再以深度神经网络的方式将其参数化表示并对其进行迁移. 迁移中通过表征模仿的方式来学习源域度量中的知识, 这种方式允许源域和目标域中的知识来自于异质域. 此外, 严格限制目标度量网络的大小, 使得目标网络更够进行高效的推理计算. 在人脸识别数据集上的实验展现了本方法的有效性.
杨乾成, 罗勇, 胡晗, 周昕, 杜博, 陶大程 . 基于联合对抗训练的鲁棒度量迁移[J]. 上海大学学报(自然科学版), 2023 , 29(1) : 1 -9 . DOI: 10.12066/j.issn.1007-2861.2460
Transfer metric learning (TML) aims to improve the metric learning in target domains by transferring knowledge from those related tasks where the distance metrics are strong and reliable. Existing TML approaches focus on only transferring the source metric knowledge, which is often prone to overfitting to the source domain. In this study, we train a source metric that is appropriate for transfer and then design a general deep TML method for effective metric transfer. In particular, we propose learning the source metric parameterized by a deep neural network through joint adversarial training and then transferring the metric to the target domain by embedding imitation, which allows the inputs of source and target domains to be heterogeneous. Besides, we restrict the size of the target metric network to be small so that the inference is efficient in the target domain. Finally, the results of applying the proposed method to a popular face verification application demonstrate its effectiveness.
| [1] | Liu W Y, Wen Y D, Yu Z D, et al. SphereFace: deep hypersphere embedding for face recognition[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017: 212-220. |
| [2] | Brattoli B, Roth K, Ommer B. MIC: mining interclass characteristics for improved metric learning[C]// Proceedings of the International Conference on Computer Vision (ICCV). 2019: 7999-8008. |
| [3] | Sanakoyeu A, Tschernezki V, Bü, et al. Divide and conquer the embedding space for metric learning[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2019: 471-480. |
| [4] | Ojala T, PietikÄINEN M, MÄENPÄÄ T. Multiresolution gray-scale and rotation invariant texture classification with local binary patterns[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24: 971-987. |
| [5] | Yang Q C, Luo Y, Hu H, et al. Robust metric boosts transfer[C]// IEEE 24th International Workshop on Multimedia Signal Processing (MMSP). 2022:1-6. |
| [6] | Wang X, Han X, Huang W, et al. Multi-similarity loss with general pair weighting for deep metric learning[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2019: 5022-5030. |
| [7] | Xuan H, Stylianou A, Pless R. Improved embeddings with easy positive triplet mining[C]// IEEE Winter Conference on Applications of Computer Vision (WACV). 2020: 2463-2471 |
| [8] | Wang F, Liu W, Liu H, et al. Additive margin softmax for face verification[J]. IEEE Signal Processing Letters, 2018, 25: 926-930. |
| [9] | Wen Y D, Zhang K P, Li Z F, et al. A discriminative feature learning approach for deep face recognition[C]// European Conference on Computer Vision (ECCV). 2016: 499-515. |
| [10] | Guo Y, Shi H, Kumar A, et al. SpotTune: transfer learning through adaptive fine-tuning[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2019: 4805-4814, |
| [11] | Yu L, Yazici V O, Liu X L, et al. Learning metrics from teachers: compact networks for image embedding[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2019: 2907-2916. |
| [12] | Luo Y, Wen Y G, Liu T L, et al. Transferring knowledge fragments for learning distance metric from a heterogeneous domain[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41: 1013-1026. |
| [13] | Wang C, Mahadevan S. Heterogeneous domain adaptation using manifold alignment[C]// Proceedings of the 22nd International Joint Conference on Artificial Intelligence (IJCAI). 2011: 1541-1546. |
| [14] | Zhou J T, Tsang I W, Pan S J, et al. Heterogeneous domain adaptation for multiple classes[C]// Proceedings of the 17th International Conference on Artificial Intelligence and Statistics (AISTATS). 2014: 1095-1103. |
| [15] | Ilyas A, Santurkar S, Tsipras D, et al. Adversarial examples are not bugs, they are features[C]// Proceedings of Advances in Neural Information Processing Systems (NIPS). 2019: 125-136. |
| [16] | Utrera F, Kravitz E, Erichson N B, et al. Adversarially-trained deep nets transfer better: illustration on image classification[EB/OL]. (2020-07-11)[2023-01-10]. https://arxiv.org/abs/2007.05869. |
| [17] | Qi G J, Aggarwal C C, Huang T S. Transfer learning of distance metrics by cross-domain metric sampling across heterogeneous spaces[C]// Proceedings of the 12th SIAM International Conference on Data Mining. 2012: 528-539. |
| [18] | Zhang X, Zhou X, Lin M, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[EB/OL]. (2017-07-04)[2023-01-03]. https://arxiv.org/pdf/1707.01083.pdf. |
| [19] | Goodfellow I J, Shlens J, Szegedy C. Explaining and harnessing adversarial examples[EB/OL]. (2014-12-20)[2023-01-03]. https://arxiv.org/pdf/1412.6572.pdf. |
| [20] | Madry A, Makelov A, Schmidt L, et al. Towards deep learning models resistant to adversarial attacks[C]// Proceedings of 6th International Conference on Learning Representations (ICLR). 2018: 1-28. |
| [21] | Huang G B, Ramesh M, Berg T, et al. Labeled faces in the wild: a database for studying face recognition in unconstrained environments[R/OL]. (2021-07-13) [2023-01-30] https://people.cs.umass.edu/~elm/papers/lfw.pdf. |
| [22] | Dai D X, Kroeger T, Timofte L, et al. Metric imitation by manifold transfer for efficient vision applications[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2015: 3527-3536. |
/
| 〈 |
|
〉 |
