基于不变风险最小化的模型无关跨域小样本学习框架

doi:10.11896/jsjkx.250900009

计算机科学 ›› 2026, Vol. 53 ›› Issue (6A): 250900009-8.doi: 10.11896/jsjkx.250900009

基于不变风险最小化的模型无关跨域小样本学习框架

安悦瑄^1,3, 赵星宇^2,3

1 河海大学计算机与软件学院南京 211100
2 南京航空航天大学计算机科学与技术学院南京 211106
3 新一代人工智能技术与交叉应用教育部重点实验室(东南大学) 南京 211189

出版日期:2026-06-16 发布日期:2026-06-12
通讯作者: 赵星宇(zhaoxy@nuaa.edu.cn)
作者简介:(anyx@hhu.edu.cn)
基金资助:
国家自然科学基金(62506117,62506163);中央高校基本科研业务费专项资金(B250201040,aiia-25-01,aiia-25-03);江苏省自然科学基金(BK20251408);中国博士后科学基金(2025M774282)

Model-agnostic Cross-domain Few-shot Learning Framework Based on Invariant Risk Minimization

AN Yuexuan^1,3, ZHAO Xingyu^2,3

1 College of Computer Science and Software Engineering,Hohai University,Nanjing 211100,China
2 College of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China
3 Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Application(Southeast University),Ministry of Education,China,Nanjing 211189,China

Published:2026-06-16 Online:2026-06-12
About author:AN Yuexuan,born in 1993,Ph.D,is a member of CCF(No.X529M).Her main research interests include machine learning and pattern recognition.
ZHAO Xingyu,born in 1994,Ph.D,is a member of CCF(No.62464M).His main research interests include machine learning and data mining.
Supported by:
National Natural Science Foundation of China(62506117,62506163),Fundamental Research Funds for the Central Universities(B250201040,aiia-25-01,aiia-25-03),Natural Science Foundation of Jiangsu Province,China(BK20251408) and China PostdoctoralScience Foundation(2025M774282).

摘要/Abstract

摘要： 小样本学习(Few-Shot Learning,FSL)旨在通过少量标注样本构建高效预测模型,以降低对海量标注数据的依赖,提升模型的学习效率和实用价值。然而,当测试域与训练域之间存在显著分布差异时,传统方法往往因领域偏移(Domain Shift)而性能大幅下降。现有的针对领域迁移场景的小样本学习的方法大多依赖于特定的模型结构或对齐策略,无法有效与现有的方法结合来提升泛化能力,且难以平衡任务相关特征与领域不变特征的学习。针对以上缺陷,提出一种模型无关的跨域小样本学习框架。该框架基于不变风险最小化(Invariant Risk Minimization,IRM)策略,可以与多种现有的小样本学习方法相结合,使这些模型能够有效学习到样本的领域不变特征,从而显著增强其跨域预测性能。多个基准数据集上的实验证明了所提框架的有效性。

关键词: 小样本学习, 跨域学习, 不变风险最小化, 元学习, 数据有效性

Abstract: Few-Shot Learning(FSL) aims to build efficient predictive models using only a small number of labeled samples,thereby reducing the reliance on large-scale annotated data and improving the learning efficiency and practical value of models.How-ever,when there is a significant distribution shift between the test domain and the training domain,traditional methods often suffer a severe performance drop due to domain shift.Existing few-shot learning methods designed for domain generalization scenarios mostly rely on specific model architectures or alignment strategies,making them difficult to integrate with other methods to enhance generalization capabilities.Moreover,they often struggle to balance the learning of task-relevant features and domain-inva-riant features.To address these issues,this paper proposes the Model-agnostic Cross-domain Few-shot Learning framework based on the strategy of Invariant Risk Minimization(IRM).This framework can be integrated with various existing few-shot learning methods,enabling these models to effectively learn domain-invariant features from samples,thereby significantly improving their cross-domain predictive performance.Experiments on multiple benchmark datasets demonstrate the effectiveness of the proposed framework.

Key words: Few-shot learning, Cross-domain learning, Invariant risk minimization, Meta-learning, Data efficiency

中图分类号:

TP391

安悦瑄, 赵星宇. 基于不变风险最小化的模型无关跨域小样本学习框架[J]. 计算机科学, 2026, 53(6A): 250900009-8. https://doi.org/10.11896/jsjkx.250900009

AN Yuexuan, ZHAO Xingyu. Model-agnostic Cross-domain Few-shot Learning Framework Based on Invariant Risk Minimization[J]. Computer Science, 2026, 53(6A): 250900009-8. https://doi.org/10.11896/jsjkx.250900009

参考文献

[1] LIU S,BAI X,GAO X.Few-Shot Learning Method Based on Symmetric Convolutional Block Network and Prototype Calibration[J].Computer Science,2024,51(11):182-190.
[2] CHEN Y,GAO Y,LU C,et al.Saliency Mask Mixup for Few-shot Image Classification[J].Computer Science,2025,52(6):256-263.
[3] TANG R,XIAO T,CHI Z,et al.Few-shot Image Classification Based on Pseudo-label Dependence Enhancement and Noise Interference Reduction[J].Computer Science,2024,51(8):152-159.
[4] LU L,CUI X,TAN Z,et al.MedOptNet:Meta-Learning Framework for Few-Shot Medical Image Classification[J].IEEE/ACM Transactions on Computational Biology and Bioinforma-tics,2024,21(4):725-736.
[5] LI R,LI M,LIU W,et al.UniMatch:Universal Matching from Atom to Task for Few-Shot Drug Discovery[C]//Proceedings of the International Conference on Learning Representations.2025.
[6] LI S,YANG J,HE H,et al.Few-Shot Scenario Testing for Autonomous Vehicles Based on Neighborhood Coverage and Similarity[C]//Proceedings of the IEEE Intelligent Vehicles Symposium.2024:620-626.
[7] SNELL J,SWERSKY K,ZEMEL R.Prototypical Networks for Few-shot Learning[C]//Proceedings ofthe Annual Conference on Neural Information Processing Systems.2017:4077-4087.
[8] FINN C,ABBEEL P,LEVINE S.Model-Agnostic Meta-Lear-ning for Fast Adaptation of Deep Networks.[C]//Proceedings of International Conference on Machine Learning.2017:1126-1135.
[9] AN Y,XUE H,ZHAO X,et al.From Instance to Metric Calibration:A Unified Framework for Open-World Few-Shot Lear-ning[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(8):9757-9773.
[10] ZHANG J,YANG K,SHI H,et al.Behind Every Domain There is a Shift:Adapting Distortion-Aware Vision Transformers for Panoramic Semantic Segmentation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2024,46(12):8549-8567.
[11] LUO L,HU S,CHEN L.Beyond Batch Learning:GlobalAwareness Enhanced Domain Adaptation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2025,47(6):4473-4488.
[12] HASSANPOUR Z,SEYDI V.A Survey on Adversarial Domain Adaptation[J].Neural Processing Letters,2023,55(3):2429-2469.
[13] DING X,DENG A,DENG M,et al.Synergistic Progressive Adversarial Network for Domain Adaptation in Rotating Machinery[J].IEEE Transactions on Instrumentation and Measurement,2025,74:1-11.
[14] MIAO W,HAN W,GENG J,et al.Hierarchical Feature Progressive Alignment Network for Remote Sensing Image Scene Classification in Multitarget Domain Adaptation[J].IEEE Transactions on Geoscience and Remote Sensing,2024,62:1-13.
[15] WU J,FANG Y.Dual-View Global and Local Category-Attentive Domain Alignment for Unsupervised Conditional AdversarialDomain Adaptation[J].Neural Networks,2025,185:107129.
[16] GILLES M,FURMANS K,RAYYES R.MetaMVUC:Active Learning for Sample-Efficient Sim-to-Real Domain Adaptation in Robotic Grasping[J].IEEE Robotics and Automation Letters,2025,10(4):3644-3651.
[17] VETTORUZZO A,BOUGUELIA M R,RÖGNVALDSSON T.Meta-Learning for Efficient Unsupervised Domain Adaptation[J].Neurocomputing,2024,574:127264.
[18] ARJOVSKY M,BOTTOU L,GULRAJANI I,et al.Invariant Risk Minimization[J/OL].https://arxiv.org/abs/1907.02893,2019.
[19] AN Y,XUE H,ZHAO X,et al.Conditional Self-SupervisedLearning for Few-Shot Classification[C]//Proceedings of the International Joint Conference on Artificial Intelligence.2021:2140-2146.
[20] AN Y,XUE H,ZHAO X,et al.Leveraging Bilateral Correlations for Multi-Label Few-Shot Learning[J].IEEE Transactions on Neural Networks and Learning Systems,2025,36(4):6816-6828.
[21] YE H,HU H,ZHAN D,et al.ew-Shot Learning via Embedding Adaptation with Set-to-Set Functions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2020:8805-8814.
[22] MANGLA P,KUMARI N,SINHA A,et al.Charting the Right Manifold:Manifold Mixup for Few-shot Learning[C]//Procee-dings of the IEEE/CVF Winter Conference on Applications of Computer Vision.2020:2207-2216.
[23] PEREZ E,STRUB F,DE V H,et al.FiLM:Visual Reasoning with a General Conditioning Layer[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2018:3942-3951.
[24] AN Y,ZHAO X,XUE H.Learning to Learn from Corrupted Data for Few-Shot Learning[C]//Proceedings of the International Joint Conference on Artificial Intelligence.2023:3423-3431.
[25] GLOROT X,BORDES A,BENGIO Y.Deep Sparse RectifierNeural Networks[C]//Proceedings of the International Confe-rence on Artificial Intelligence and Statistics.2011:315-323.
[26] IOFFE S,SZEGEDY C.Batch Normalization:Accelerating Deep Network Training by Reducing Internal Covariate Shift[C]//Proceedings of International Conference on Machine Learning.2015:448-456.
[27] VERMA V,LAMB A,BECKHAM C,et al.Manifold Mixup:Better Representations by Interpolating Hidden States[C]//Proceedings of International Conference on Machine Learning.2019:6438-6447.
[28] GIDARIS S,SINGH P,KOMODAKIS N.Unsupervised Representation Learning by Predicting Image Rotations[C]//Procee-dings of the International Conference on Learning Representations.2018.
[29] BERTINETTO L,HENRIQUES J,TORR P,et al.Meta-Lear-ning with Differentiable Closed-Form Solvers[C]//Proceedings of the International Conference on Learning Representations.2019.
[30] WAH C,BRANSON S,WELINDER P,et al.The Caltech-Ucsd Birds-200-2011 Dataset[EB/OL].https://www.visi-on.caltech.edu/datasets/cub_200_2011/,2011.
[31] VINYALS O,BLUNDELL C,LILLICRAP T,et al.MatchingNetworks for One Shot Learning[C]//Proceedings of the AnnualConference on Neural Information Processing Systems.2016:3630-3638.
[32] REN M,TRIANTAFILLOU E,RAVI S,et al.Meta-Learning for Semi-Supervised Few-Shot Classification[C]//Proceedings of the International Conference on Learning Representations.2018.
[33] KHOSLA A,JAYADEVAPRAKASH N,YAO B,et al.NovelDataset for Fine-Grained Image Categorization:Stanford Dogs[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshop on Fine-grained VisualCategorization.2011:1-2.
[34] SU J,MAJI S,HARIHARAN B.When Does Self-supervision Improve Few-Shot Learning?[C]//Proceedings of the Euro-pean Conference on Computer Vision.2020:645-666.
[35] PATTERSON G,XU C,SU H,et al.The SUN Attribute Database:Beyond Categories for Deeper Scene Understanding[J].International Journal of Computer Vision,2014,108:59-81.
[36] HUANG S,ZHANG M,KANG Y,et al.Attributes-Guided and Pure-Visual Attention Alignment for Few-Shot Recognition[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2021:7840-7847.
[37] LECUN Y,CORTES C,BURGES C.The MNIST Hand-written Digit Database[EB/OL].http://yann.lecun.com/exdb/mnist.
[38] CLANUWAT T,BOBER-IRIZAR M,KITAMOTO A,et al.Deep Learning for Classical Japanese Literature[J/OL].https://arxiv.org/abs/1812.01718.
[39] XIAO H,RASUL K,VOLLGRAF R.Fashion-MNIST:a Novel Image Dataset for Benchmarking Machine Learning Algorithms[J/OL].https://arxiv.org/abs/1708.07747,2017.
[40] KINGMA D,BA J.Adam:Adam:A Method for Stochastic Optimization[C]//Proceedings ofthe International Conference on Learning Representations.2015.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

基于不变风险最小化的模型无关跨域小样本学习框架

Model-agnostic Cross-domain Few-shot Learning Framework Based on Invariant Risk Minimization

PDF (PC)

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 0

Metrics

本文评价

推荐阅读 0