基于原型增强的元学习分类模型

doi:10.3778/j.issn.1002-8331.2310-0320

摘要/Abstract

摘要： 元学习旨在利用已有的知识经验快速获取新知识、适应新任务，是解决小样本分类问题常用的方法之一。现有的元学习方法在特征提取时通常会忽略任务中支持集样本和查询集样本之间的关系，从而无法获得最具鉴别性的特征，导致类原型不可靠。因此，提出一种基于原型增强的元学习分类模型。该模型主要由两部分组成：特征表示模块与原型修正模块。针对现有方法特征利用不足的问题，特征表示模块利用注意力机制捕捉支持集和查询集样本间的交互信息并更新其特征表示；而针对数据稀缺问题，原型修正模块利用部分查询集的无标签样本扩充支持集，进而对原型位置进行迭代修正。在mini-ImageNet和tiered-ImageNet数据集上的实验结果表明，该模型的分类准确率与其他元学习方法相比有较为显著的提升。

关键词: 原型增强, 元学习, 小样本学习, 图像分类

Abstract: Meta-learning aims to utilize existing knowledge and experience to quickly acquire new knowledge and adapt to new tasks. It is one of the commonly used methods to solve few-shot classification problems. Existing meta-learning methods usually ignore the relationship between support set and query set in feature extraction, thus failing to obtain the most discriminative features and leading to unreliable class prototypes. Therefore, this paper proposes a prototype enhanced meta-learning classification model. The model consists of two main components: a feature representation module and a prototype modification module. To address the problem of underutilization of features in existing methods, the feature representation module utilizes attention mechanism to capture the relationship between support set and query set, and update their feature representations. While to address the problem of data scarcity, the prototype modification module utilizes unlabeled samples from query set to expand the support set, and then iteratively corrects the positions of prototypes. Experimental results on mini-ImageNet and tiered-ImageNet datasets show that the classification accuracy of proposed model has significantly improved compared with other meta-learning methods.

Key words: prototype enhancement, meta-learning, few-shot learning, image classification

翟文茜, 李凡长. 基于原型增强的元学习分类模型[J]. 计算机工程与应用, 2025, 61(6): 273-281.

ZHAI Wenxi, LI Fanzhang. Meta-Learning Classification Model Based on Prototype Enhanced[J]. Computer Engineering and Applications, 2025, 61(6): 273-281.

参考文献

[1] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
[2] DING Y, ZHANG Z L, ZHAO X F, et al. Deep hybrid: multi-graph neural network collaboration for hyperspectral image classification[J]. Defence Technology, 2023, 23: 164-176.
[3] LI F, ZENG A, LIU S, et al. Lite DETR: an interleaved multi-scale encoder for efficient detr[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 18558-18567.
[4] YAN J, LIU Y, SUN J, et al. Cross modal transformer: towards fast and robust 3D object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023: 18268-18278.
[5] ALAYRAC J B, DONAHUE J, LUC P, et al. Flamingo: a visual language model for few-shot learning[C]//Advances in Neural Information Processing Systems, 2022: 23716-23736.
[6] KIM D, KIM J, CHO S, et al. Universal few-shot learning of dense prediction tasks with visual token matching[C]//Proceedings of the Eleventh International Conference on Learning Representations, 2022.
[7] RAVI S, LAROCHELLE H. Optimization as a model for few-shot learning[C]//Proceedings of the International Conference on Learning Representations, 2016.
[8] HOSPEDALES T, ANTONIOU A, MICAELLI P, et al. Meta-learning in neural networks: a survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(9): 5149-5169.
[9] RAJENDRAN J, IRPAN A, JANG E. Meta-learning requires meta-augmentation[C]//Advances in Neural Information Processing Systems, 2020: 5705-5715.
[10] YAO H, WANG Y, WEI Y, et al. Meta-learning with an adaptive task scheduler[C]//Advances in Neural Information Processing Systems, 2021: 7497-7509.
[11] RAMALHO T, GARNELO M. Adaptive posterior learning: few-shot learning with a surprise-based memory module[J]. arXiv:1902.02527, 2019.
[12] 刘兵, 杨娟, 汪荣贵, 等. 结合记忆与迁移学习的小样本学习[J]. 计算机工程与应用, 2022, 58(19): 242-249.
LIU B, YANG J, WANG R G, et al. Memory-based transfer learning for few-shot learning[J]. Computer Engineering and Applications, 2022, 58(19): 242-249.
[13] WANG W, DUAN L, WANG Y, et al. MMT: cross domain few-shot learning via meta-memory transfer[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(12): 15018-15035.
[14] SNELL J, SWERSKY K, ZEMEL R. Prototypical networks for few-shot learning[C]//Advances in Neural Information Processing Systems, 2017.
[15] SIMON C, KONIUSZ P, NOCK R, et al. Adaptive subspaces for few-shot learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 4136-4145.
[16] ZHOU Y, YU L. Few-shot learning via weighted prototypes from graph structure[J]. Pattern Recognition Letters, 2023, 176: 230-235.
[17] FINN C, ABBEEL P, LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[C]//Proceedings of the International Conference on Machine Learning, 2017: 1126-1135.
[18] ZHANG Y, ZUO X, ZHENG X, et al. Improving metric-based few-shot learning with dynamically scaled softmax loss[J]. Image and Vision Computing, 2023, 140: 104860.
[19] JIA J, FENG X, YU H. Few-shot classification via efficient meta-learning with hybrid optimization[J]. Engineering Applications of Artificial Intelligence, 2024, 127: 107296.
[20] SANTORO A, BARTUNOV S, BOTVINICK M, et al. Meta-learning with memory-augmented neural networks[C]//Proceedings of the International Conference on Machine Learning, 2016: 1842-1850.
[21] YE H J, HU H, ZHAN D C, et al. Few-shot learning via embedding adaptation with set-to-set functions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 8808-8817.
[22] HUANG W, YUAN Z, YANG A, et al. TAE-net: task-adaptive embedding network for few-shot remote sensing scene classification[J]. Remote Sensing, 2021, 14(1): 111.
[23] LEE Sanghyuk, LEE Seunghyun, SONG B C. Efficient meta-learning through task-specific pseudo labelling[J]. Electronics, 2023, 12(13): 2757.
[24] ZHANG H, QUE H, REN J, et al. Transductive semantic knowledge graph propagation for zero-shot learning[J]. Journal of the Franklin Institute, 2023, 360(17): 13108-13125.
[25] AHMED M, SERAJ R, ISLAM S M S. The k-means algorithm: a comprehensive survey and performance evaluation[J]. Electronics, 2020, 9(8): 1295.
[26] GUO Q, YIN Z, WANG P. An improved three-way k-means algorithm by optimizing cluster centers[J]. Symmetry, 2022, 14(9): 1821.
[27] LIM J Y, LIM K M, LEE C P, et al. SSL-ProtoNet: self-supervised learning prototypical networks for few-shot learning[J]. Expert Systems with Applications, 2024, 238: 122173.
[28] VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching networks for one shot learning[C]//Advances in Neural Information Processing Systems, 2016.
[29] REN M, TRIANTAFILLOU E, RAVI S, et al. Meta-learning for semi-supervised few-shot classification[J]. arXiv:1803. 00676, 2018.
[30] RUSSAKOVSKY O, DENG J, SU H, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115: 211-252.
[31] LAKE B, SALAKHUTDINOV R, GROSS J, et al. One shot learning of simple visual concepts[C]//Proceedings of the Annual Meeting of the Cognitive Science Society, 2011.
[32] CHEN Y, LIU Z, XU H, et al. Meta-baseline: exploring simple meta-learning for few-shot learning[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 9062-9071.
[33] CHEN C, LI K, WEI W, et al. Hierarchical graph neural networks for few-shot learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 32(1): 240-252.
[34] ZHOU F, ZHANG L, WEI W. Meta-generating deep attentive metric for few-shot classification[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(10): 6863-6873.