基于特征空间增强重放和偏差校正的类增量学习方法

doi:10.16451/j.cnki.issn1003-6059.202408006

摘要
图/表
参考文献
相关文章 (14)

全文: PDF (1031 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要网络不断学习新的知识时会遭受灾难性遗忘,增量学习方法可通过存储少量旧数据重放以实现增量学习的可塑性与稳定性的平衡.然而,存储旧任务的数据会有内存限制及隐私泄露的问题.针对该问题,文中提出基于特征空间增强重放和偏差校正的类增量学习方法,用于缓解灾难性遗忘.首先,每类存储一个中间层特征均值作为其代表的原型,并冻结低层特征提取网络,避免原型“漂移”.在增量学习阶段,存储的原型通过几何平移变换增强重放的方式维持先前任务的决策边界.然后,通过偏差校正为每个任务学习分类权重,进一步纠正方法分类偏向于新任务的问题.在4个基准数据集上的实验表明文中方法性能较优.

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	孙晓鹏
	余璐
	徐常胜

关键词 ：类增量学习, 持续学习, 灾难性遗忘, 特征表示, 特征增强

Abstract：The problem of catastrophic forgetting arises when the network learns new knowledge continuously. Various incremental learning methods are proposed to solve this problem and one mainstream approach is to balance the plasticity and stability of incremental learning through storing a small amount of old data and replaying it. However, storing data from old tasks can lead to memory limitations and privacy breaches. To address this issue, a class-incremental learning method based on feature space augmented replay and bias correction is proposed to alleviate catastrophic forgetting. Firstly, the mean feature of an intermediate layer for each class is stored as its representative prototype and the low-level feature extraction network is frozen to prevent prototype drift. In the incremental learning stage, the stored prototypes are enhanced and replayed through geometric translation transformation to maintain the decision boundaries of the previous task. Secondly, bias correction is proposed to learn classification weights for each task, further correcting the problem of model classification bias towards new tasks. Experiments on four benchmark datasets show that the proposed method outperforms the state-of-the-art algorithms.

Key words： Class Incremental Learning Continuous Learning Catastrophic Forgetting Feature Representation Feature Enhancement

收稿日期: 2024-06-20

ZTFLH:	TP183
	TP391.41

基金资助:国家自然科学基金项目(No.62202331)资助

通讯作者: 余璐,博士,副教授,主要研究方向为持续学习、表征学习.E-mail:luyu@email.tjut.edu.cn.

作者简介: 孙晓鹏,硕士研究生,主要研究方向为持续学习.E-mail:xiaopeng12138@stud.tjut.edu.cn. 徐常胜,博士,研究员.主要研究方向为多媒体分析/索引/检索、模式识别、计算机视觉.E-mail:csxu@nlpr.ia.ac.cn.

引用本文:

孙晓鹏, 余璐, 徐常胜. 基于特征空间增强重放和偏差校正的类增量学习方法[J]. 模式识别与人工智能, 2024, 37(8): 729-740. SUN Xiaopeng, YU Lu, XU Changsheng. Class-Incremental Learning Method Based on Feature Space Augmented Replay and Bias Correction. Pattern Recognition and Artificial Intelligence, 2024, 37(8): 729-740.

链接本文:

http://manu46.magtech.com.cn/Jweb_prai/CN/10.16451/j.cnki.issn1003-6059.202408006 或 http://manu46.magtech.com.cn/Jweb_prai/CN/Y2024/V37/I8/729

[1] HE K M, ZHANG X Y, REN S Q, et al. Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification // Proc of the IEEE International Conference on Computer Vision . Washington, USA: IEEE, 2015: 1026-1034.
[2] KIRKPATRICK J, PASCANU R, RABINOWITZ N, et al. Overcoming Catastrophic Forgetting in Neural Networks. Proceedings of the National Academy of Sciences, 2017, 114(13): 3521-3526.
[3] KEMKER R, MCCLURE M, ABITINO A, et al. Measuring Catastrophic Forgetting in Neural Networks. Proceedings of the AAAI Conference on Artificial Intelligence, 2018, 32(1): 3390-3398.
[4] LI G P, XU Y, DING J, et al. Toward Generic and Controllable Attacks Against Object Detection. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62. DOI: 10.1109/TGRS.2024.3417958.
[5] JEEVESWARAN K, BHAT P S, ZONOOZ B, et al. BiRT: Bio-inspired Replay in Vision Transformers for Continual Learning. Journal of Machine Learning Research, 2023, 202: 14817-14835.
[6] CHEN X W, CHANG X B.Dynamic Residual Classifier for Class Incremental Learning // Proc of the IEEE/CVF International Conference on Computer Vision. Washington, USA: IEEE, 2023: 18697-18706.
[7] 周大蔚,汪福运,叶翰嘉,等.基于深度学习的类别增量学习算法综述.计算机学报, 2023, 46(8): 1577-1605.
(ZHOU D W, WANG F Y, YE H J,et al. Deep Learning for Class-Incremental Learning: A Survey. Chinese Journal of Compu-ters, 2023, 46(8): 1577-1605.)
[8] MASANA M, LIU X L, TWARDOWSKI B, et al. Class-Incremental Learning: Survey and Performance Evaluation on Image Classification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(5): 5513-5533.
[9] ZHOU D W, WANG Q W, QI Z H, et al. Deep Class-Incremental Learning: A Survey[C/OL].[2024-05-19]. https://arxiv.org/pdf/2302.03648.
[10] GOU J P, YU B S, MAYBANK S J, et al. Knowledge Distillation: A Survey. International Journal of Computer Vision, 2021, 129: 1789-1819.
[11] LI Z Z, HOIEM D.Learning without Forgetting. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(12): 2935-2947.
[12] ZHU K, ZHAI W, CAO Y, et al. Self-Sustaining Representation Expansion for Non-exemplar Class-Incremental Learning // Proc of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2022: 9286-9295.
[13] LIU Y Y, SCHIELE B, SUN Q R.Adaptive Aggregation Networks for Class-Incremental Learning // Proc of the IEEE/CVF Confe-rence on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2021: 2544-2553.
[14] REBUFFI S A, KOLESNIKOV A, SPERL G, et al. ICaRL: Incremental Classifier and Representation Learning // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2017: 5533-5542.
[15] CASTRO F M, MARÍN-JIMÉNEZ M J, GUIL N, et al. End-to-End Incremental Learning // Proc of the European Conference on Computer Vision. Berlin, Germany: Springer, 2018: 241-257.
[16] WU Y, CHEN Y P, WANG L J, et al. Large Scale Incremental Learning // Proc of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2019: 374-382.
[17] SHIN H, LEE J K, KIM J, et al. Continual Learning with Deep Generative Replay // Proc of the 31st International Conference on Neural Information Processing Systems. Cambridge,USA: MIT Press, 2017: 2994-3003.
[18] ZHAI M Y, CHEN L, TUNG F, et al. Lifelong GAN: Continual Learning for Conditional Image Generation // Proc of the IEEE/CVF International Conference on Computer Vision. Washington, USA: IEEE, 2019: 2759-2768.
[19] XIANG Y, FU Y, JI P, et al. Incremental Learning Using Conditional Adversarial Networks // Proc of the IEEE/CVF International Conference on Computer Vision. Washington, USA: IEEE, 2019: 6618-6627.
[20] GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al.Ge-nerative Adversarial Nets // Proc of the 27th International Confe-rence on Neural Information Processing Systems. Cambridge,USA: MIT Press, 2014, II: 2672-2680.
[21] KEMKER R, KANAN C.FearNet: Brain-Inspired Model for Incremental Learning[C/OL]. [2024-05-19].https://arxiv.org/pdf/1711.10563v1.
[22] ZHU F, ZHANG X Y, WANG C, et al. Prototype Augmentation and Self-Supervision for Incremental Learning // Proc of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2021: 5867-5876.
[23] SHI W X, YE M.Prototype Reminiscence and Augmented Asy-mmetric Knowledge Aggregation for Non-exemplar Class-Incremental Learning // Proc of the IEEE/CVF International Conference on Computer Vision. Washington, USA: IEEE, 2023: 1772-1781.
[24] PETIT G, POPESCU A, SCHINDLER H, et al. FeTrIL: Feature Translation for Exemplar-Free Class-Incremental Learning // Proc of the IEEE/CVF Winter Conference on Applications of Computer Vision. Washington, USA: IEEE, 2023: 3900-3909.
[25] HOU S H, PAN X Y, LOY C C, et al. Learning a Unified Classifier Incrementally via Rebalancing // Proc of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2019: 831-839.
[26] ZEILER M D, FERGUS R.Visualizing and Understanding Convolutional Networks // Proc of the 13th European Conference on Computer Vision. Berlin, Germany: Springer, 2014: 818-833.
[27] KRIZHEVSKY A. Learning Multiple Layers of Features from Tiny Images[EB/OL]. [2024-05-19]. http://www.cs.toronto.edu/~kriz/learning-features-2009-TR.pdf.
[28] LE Y, YANG X. Tiny ImageNet Visual Recognition Challenge[EB/OL]. [2024-05-19]. http://vision.stanford.edu/teaching/cs231n/reports/2015/pdfs/yle_project.pdf.
[29] DENG J, DONG W, SOCHER R, et al. ImageNet: A Large-Scale Hierarchical Image Database // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2009: 248-255.
[30] BOSSARD L, GUILLAUMIN M, VAN GOOL L.Food-101-Mining Discriminative Components with Random Forests // Proc of the 13th European Conference on Computer Vision. Berlin, Germany: Springer, 2014: 446-461.
[31] ALJUNDI R.Continual Learning in Neural Networks[C/OL]. [2024-05-19].https://arxiv.org/pdf/1910.02718.
[32] SMITH J, HSU Y C, BALLOCH J, et al. Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning // Proc of the IEEE/CVF International Conference on Computer Vision. Washington, USA: IEEE, 2021: 9354-9364.
[33] ZHU F, CHENG Z, ZHANG X Y, et al. Class-Incremental Lear-ning via Dual Augmentation[C/OL].[2024-05-19].https://proceedings.neurips.cc/paper/2021/file/77ee3bc58ce560b86c2b59363281e914-Paper.pdf.
[34] HE K M, ZHANG X Y, REN S Q, et al. Deep Residual Learning for Image Recognition // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2016: 770-778.
[35] YAN S P, XIE J W, HE X M.DER: Dynamically Expandable Representation for Class Incremental Learning // Proc of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2021: 3013-3022.