基于类边界样本生成与微调的类增量学习初始阶段训练方法

doi:10.19678/j.issn.1000-3428.0070459

摘要/Abstract

摘要： 近年来,众多研究工作聚焦于类增量学习(CIL)领域,其中一种常见的策略是在模型的初始化训练阶段结束后,对特征提取器的参数进行冻结处理,以此来维持已习得知识的稳定性。这种策略依赖于在初始训练阶段构建一个鲁棒的特征提取器,然而在面对初始阶段数据量有限的场景时,这种方法可能无法充分发挥其潜力,导致增量学习效果欠佳。为了优化增量学习的初始训练环节,应对初始阶段数据量有限的挑战,设计了一套两步骤训练方法用于初始阶段的特征提取器训练,称为基于类边界样本生成与微调的类增量学习初始阶段训练方法。该方法包括生成对抗训练和微调训练两个步骤。生成对抗训练采用生成对抗网络(GAN)来生成类边界附近的样本,使用对比学习迫使类中心远离这些生成样本,从而增强不同类别间的特征的区分度并提升模型的泛化表现。微调训练阶段结合交叉熵损失和自监督损失微调模型,提升对初始任务基类的分类准确率。经过训练后的特征提取器用作其他增量学习方法的起点,该方法在CIFAR-100数据集上50个基类5个增量任务设置下将FeCAM的平均准确率提升了3.8百分点,将FeTrIL的平均准确率提升了2.2百分点。

关键词: 持续学习, 类增量学习, 对比学习, 生成对抗网络, 微调

Abstract: Several recent studies have focused on the field of Class Incremental Learning (CIL). A common strategy involves freezing the parameters of the feature extractor after the initial training phase of the model for maintaining the stability of the learned knowledge. This strategy relies on building a robust feature extractor during the initial training phase. However, this method may not reach its full potential in scenarios with limited data during the initial phase, resulting in poor incremental learning. To optimize the initial training in incremental learning and cope with the challenge of limited data at the initial stage, a two-step training method that is the initial phase training method of CIL based on class boundary sample generation and fine-tuning, is proposed for the initial stage of feature extractor training. The method comprises two steps: generative adversarial training and fine-tuning training. Generative adversarial training uses a generative adversarial network to generate samples near the boundary of each class and uses contrastive learning to force the center of the class away from these generated samples for enhancing the discrimination of features between different classes and improving the generalization performance of the model. In the fine-tuning stage, cross-entropy loss and self-supervised loss are combined to fine-tune the model and improve the classification accuracy on the initial task. The resulting trained feature extractor is used as the starting point for other incremental learning methods. The method improves the average accuracy of FeCAM by 3.8 percentage points and FeTrIL by 2.2 percentage points on the CIFAR-100 dataset with 50 base classes and five incremental tasks.

Key words: continual learning, Class-Incremental Learning (CIL), contrastive learning, Generative Adversarial Network (GAN), fine-tuning

中图分类号:

TP391

郑宇濠, 李建华. 基于类边界样本生成与微调的类增量学习初始阶段训练方法[J]. 计算机工程, 2026, 52(7): 86-93.

ZHENG Yuhao, LI Jianhua. Initial Phase Training Method of Class Incremental Learning Based on Class Boundary Sample Generation and Fine-Tuning[J]. Computer Engineering, 2026, 52(7): 86-93.

https://www.ecice06.com/CN/Y2026/V52/I7/86

参考文献

[1] ROBINS A. Catastrophic forgetting, rehearsal and pseudorehearsal[J]. Connection Science, 1995, 7(2): 123-146.
[2] LI Z Z, HOIEM D. Learning without forgetting[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(12): 2935-2947.
[3] HOU S H, PAN X Y, LOY C C, et al. Learning a unified classifier incrementally via rebalancing[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2020: 831-839.
[4] ZHU F, ZHANG X Y, WANG C, et al. Prototype augmentation and self-supervision for incremental learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2021: 5867-5876.
[5] BEYER L, ZHAI X H, ROYER A, et al. Knowledge distillation: a good teacher is patient and consistent[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2022: 10915-10924.
[6] GOSWAMI D, LIU Y Y, TWARDOWSKI B, et al. FeCAM: exploiting the heterogeneity of class distributions in exemplar-free continual learning[C]//Proceedings of the Advances in Neural Information Processing Systems. New Orleans, USA: Neural Information Processing Systems Foundation, Inc. (NeurIPS), 2023: 6582-6595.
[7] GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial networks[J]. Communications of the ACM, 2020, 63(11): 139-144.
[8] PETIT G, POPESCU A, SCHINDLER H, et al. FeTrIL: feature translation for exemplar-free class-incremental learning[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. Washington D.C., USA: IEEE Press, 2023: 3900-3909.
[9] TUKEY J W. Exploratory data analysis[M]. Boston, USA: Addison Wesley Publishing Company, 1977.
[10] CHA H, LEE J, SHIN J. Co²L: contrastive continual learning[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Washington D.C., USA: IEEE Press, 2022: 9496-9505.
[11] CHEN T, KORNBLITH S, NOROUZI M, et al. A simple framework for contrastive learning of visual representations[EB/OL].[2024-09-11]. https://arxiv.org/abs/2002.05709.
[12] SHI Y J, ZHOU K Q, LIANG J, et al. Mimicking the Oracle: an initial phase decorrelation approach for class incremental learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2022: 16701-16710.
[13] CHEN X L, HE K M. Exploring simple Siamese representation learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2021: 15745-15753.
[14] TOMASEV N, BICA I, MCWILLIAMS B, et al. Pushing the limits of self-supervised ResNets: can we outperform supervised learning without labels on ImageNet?[EB/OL].[2024-09-11]. https://arxiv.org/abs/2201.05119.
[15] HE K M, FAN H Q, WU Y X, et al. Momentum contrast for unsupervised visual representation learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2020: 9726-9735.
[16] TIAN Y L, SUN C, POOLE B, et al. What makes for good views for contrastive learning?[EB/OL].[2024-09-11]. https://arxiv.org/abs/2005.10243.
[17] CRESWELL A, WHITE T, DUMOULIN V, et al. Generative adversarial networks: an overview[J]. IEEE Signal Processing Magazine, 2018, 35(1): 53-65.
[18] SALIMANS T, GOODFELLOW I, ZAREMBA W, et al. Improved techniques for training GANs[EB/OL].[2024-09-11]. https://arxiv.org/abs/1606.03498.
[19] KURACH K, LUCIC M, ZHAI X H, et al. A large-scale study on regularization and normalization in GANs[EB/OL].[2024-09-11]. https://arxiv.org/abs/1807.04720.
[20] MUKHERJEE S, ASNANI H, LIN E, et al. ClusterGAN: latent space clustering in generative adversarial networks[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto, USA: AAAI Press, 2019: 4610-4617.
[21] KANG M, PARK J. ContraGAN: contrastive learning for conditional image generation[EB/OL].[2024-09-11]. https://arxiv.org/abs/2006.12681.
[22] KIRKPATRICK J, PASCANU R, RABINOWITZ N, et al. Overcoming catastrophic forgetting in neural networks[J]. Proceedings of the National Academy of Sciences of the United States of America, 2017, 114(13): 3521-3526.
[23] BELOUADAH E, POPESCU A. DeeSIL: deep-shallow incremental learning[C]//Proceedings of Computer Vision Workshops. Berlin, Germany: Springer, 2019: 151-157.
[24] LIU Y, PARISOT S, SLABAUGH G, et al. More classifiers, less forgetting: a generic multi-classifier paradigm for incremental learning[C]//Proceedings of the 16th European Conference on Computer Vision. Berlin, Germany: Springer, 2020: 699-716.
[25] ZHU F, CHENG Z, ZHANG X Y, et al. Class-incremental learning via dual augmentation[J]. Advances in Neural Information Processing Systems, 2021, 34: 14306-14318.
[26] ZHU K, ZHAI W, CAO Y, et al. Self-sustaining representation expansion for non-exemplar class-incremental learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2022: 9286-9295.
[27] MIYATO T, KATAOKA T, KOYAMA M, et al. Spectral normalization for generative adversarial networks[EB/OL].[2024-09-11]. https://arxiv.org/abs/1802.05957.
[28] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2016: 770-778.
[29] LIN Z, SEKAR V, FANTI G, et al. Why spectral normalization stabilizes GANs: analysis and improvements[EB/OL].[2024-09-11]. https://arxiv.org/abs/2009.02773.
[30] 郭祥振, 李思潼, 卢锐, 等. 基于多任务联合注意力的结肠息肉分割网络[J]. 计算机工程, 2024, 50(2): 327-336. GUO X Z, LI S T, LU R, et al. Colon polyp segmentation network based on multi-task joint attention[J]. Computer Engineering, 2024, 50(2): 327-336.(in Chinese)
[31] HEUSEL M, RAMSAUER H, UNTERTHINER T, et al. GANs trained by a two time-scale update rule converge to a local Nash equilibrium[J]. Advances in Neural Information Processing Systems, 2017, 30, 6626-6637.

选择文件类型/文献管理软件名称

选择包含的内容