Article(id=1329456557160509589, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1329456279791186761, articleNumber=null, orderNo=null, doi=10.7544/issn1000-1239.202220933, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1668096000000, receivedDateStr=2022-11-11, revisedDate=1684166400000, revisedDateStr=2023-05-16, acceptedDate=null, acceptedDateStr=null, onlineDate=1794896743335, onlineDateStr=2026-11-17, pubDate=1711900800000, pubDateStr=2024-04-01, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1794896743335, onlineIssueDateStr=2026-11-17, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1794896743335, creator=13701087609, updateTime=1794896743335, updator=13701087609, issue=Issue{id=1329456279791186761, tenantId=1146029695717560320, journalId=1324759914146529478, year='2024', volume='61', issue='4', pageStart='809', pageEnd='1084', issueExtLink='null', onlineDate='null', pubDate='1711900800000', pubDateStr='2024-04-01', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1794896677204, creator='13701087609', updateTime=1794899232988, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext={EN=IssueExt(id=1329466999882756515, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1329456279791186761, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1329466999882756516, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1329456279791186761, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null, downloadFileDto=null}, startPage=856, endPage=876, ext={EN=ArticleExt(id=1329456557349253270, articleId=1329456557160509589, tenantId=1146029695717560320, journalId=1324759914146529478, language=EN, title=Research Progress of Few-Shot Learning Methods Based on Graph Neural Networks, columnId=1325489891789910943, journalTitle=Journal of Computer Research and Development, columnName=Artificial Intelligence, runingTitle=null, highlight=null, articleAbstract=

Few-shot learning (FSL) aims to learn to get a problem-solving model using a small number of samples. Under the trend of training models with big data, deep learning has gained success in many fields, but realistic scenarios often lack sufficient samples or labeled samples. Therefore, FSL becomes a promising research direction at present. Graph neural networks (GNN) have attracted great attention due to their excellent performance in many applications. In view of this, many methods try to use GNN for FSL. Currently there are few review researches related to FSL methods based on GNN, and there is a lack of division system and introductory work on this type of methods. We systematically compose the current work related to FSL based on GNN. The work outlines the basis and concepts of graph methods for FSL, broadly classifies them into four categories of methods based on node-based feature, edge-based feature, node-pair-based feature and class-level-based feature according to the basic ideas of the models. The research progress of the four methods is introduced as well. Then the experimental results of the commonly used few-shot datasets and representative models on these datasets are summarized, as well as the advantages and disadvantages of each type of methods. Finally, current status and challenges of the graph methods for FSL are introduced, and their future directions are prospected.

, authors=Jieyi Yang, Yihong Dong*, Jiangbo Qian, authorsList=Jieyi Yang, Yihong Dong, Jiangbo Qian, authorCompany=null, correspAuthors=Yihong Dong, authorNote=

born in 1999. Master candidate. Student member of CCF. Her main research interests include few-shot learning, graph neural network, and machine learning

born in 1969. PhD, professor, master supervisor. Member of CCF. His main research interests include big data, data mining, and artificial intelligence

born in 1974. PhD, professor, PhD supervisor. Senior member of CCF. His main research interests include machine learning, pattern recognition, and intelligent systems

, correspAuthorsNote=null, copyrightStatement=Copyright © 2024 Journal of Computer Research and Development. All rights reserved., copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, fund=null), CN=ArticleExt(id=1329456559547068589, articleId=1329456557160509589, tenantId=1146029695717560320, journalId=1324759914146529478, language=CN, title=基于图神经网络的小样本学习方法研究进展, columnId=1325489892104483745, journalTitle=计算机研究与发展, columnName=人工智能, runingTitle=null, highlight=null, articleAbstract=

小样本学习(few-shot learning,FSL)旨在利用少量样本学习得到解决问题的模型,为解决应用场景中样本量少或标注样本少的问题. 图神经网络(graph neural network,GNN)由于其在许多应用中的卓越性能引起了极大的关注,许多学者开始尝试利用图神经网络进行小样本学习,基于图神经网络的方法在小样本领域取得了卓越的成绩. 目前与基于图神经网络的小样本学习方法相关的综述性研究较少,缺乏该类方法的划分体系与介绍性工作,因此系统地梳理了当前基于图神经网络的小样本学习的相关工作:概括了小样本学习的图神经网络方法的概念,根据模型的基本思想将其划分为基于节点特征、基于边特征、基于节点对特征和基于类级特征的4类方法,介绍了这4类方法的研究进展;总结了目前常用的小样本数据集和代表性模型在这些数据集上的实验结果,归纳各类方法主要的研究内容和优劣势;最后概述了基于图神经网络的小样本学习方法的应用和面临的挑战,并展望其未发展方向.

, authors=杨洁祎, 董一鸿*, 钱江波, authorsList=杨洁祎, 董一鸿, 钱江波, authorCompany=null, correspAuthors=董一鸿, authorNote=

1999年生. 硕士研究生. CCF学生会员. 主要研究方向为小样本学习、图神经网络、机器学习

1969年生. 博士,教授,硕士生导师. CCF会员. 主要研究方向为大数据处理、数据挖掘、人工智能

1974年生. 博士,教授,博士生导师. CCF高级会员. 主要研究方向为机器学习、模式识别、智能系统

, correspAuthorsNote=
董一鸿()
, copyrightStatement=版权所有 © 2024 计算机研究与发展编辑部, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=RvgzrvfEe+w2uvffIRcPYw==, magXml=w/AnUIVVCWmmRbrhwaOycQ==, pdfUrl=null, pdf=PnCgUQVWdC0F6H1vAzyQVw==, pdfFileSize=2178420, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=cjvv6pxpj7uioecOkb+iyg==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=s78gmaaE5ck0RzubMBIBoQ==, mapNumber=null, fund=null)}, authors=[Author(id=1329456559903584437, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=17858878879@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329456559987470518, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456559903584437, language=EN, stringName=Jieyi Yang, firstName=Jieyi, middleName=null, lastName=Yang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"content":"

born in 1999. Master candidate. Student member of CCF. Her main research interests include few-shot learning, graph neural network, and machine learning

"}, bioImg=null, bioContent=

born in 1999. Master candidate. Student member of CCF. Her main research interests include few-shot learning, graph neural network, and machine learning

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329456560058773687, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456559903584437, language=CN, stringName=杨洁祎, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"img":"pkpkoM1HN11/2jXUEsg4qA==","content":"

1999年生. 硕士研究生. CCF学生会员. 主要研究方向为小样本学习、图神经网络、机器学习

"}, bioImg=pkpkoM1HN11/2jXUEsg4qA==, bioContent=

1999年生. 硕士研究生. CCF学生会员. 主要研究方向为小样本学习、图神经网络、机器学习

, aboutCorrespAuthor=null)}, companyList=null), Author(id=1329456560130076857, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=dongyihong@nbu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329456560260100282, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560130076857, language=EN, stringName=Yihong Dong, firstName=Yihong, middleName=null, lastName=Dong, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=null, bio={"content":"

born in 1969. PhD, professor, master supervisor. Member of CCF. His main research interests include big data, data mining, and artificial intelligence

"}, bioImg=null, bioContent=

born in 1969. PhD, professor, master supervisor. Member of CCF. His main research interests include big data, data mining, and artificial intelligence

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329456560339792059, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560130076857, language=CN, stringName=董一鸿, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=null, bio={"img":"PKUxcdppRxNm6kL0qg7ruA==","content":"

1969年生. 博士,教授,硕士生导师. CCF会员. 主要研究方向为大数据处理、数据挖掘、人工智能

"}, bioImg=PKUxcdppRxNm6kL0qg7ruA==, bioContent=

1969年生. 博士,教授,硕士生导师. CCF会员. 主要研究方向为大数据处理、数据挖掘、人工智能

, aboutCorrespAuthor=null)}, companyList=null), Author(id=1329456560440455357, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329456560524341438, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560440455357, language=EN, stringName=Jiangbo Qian, firstName=Jiangbo, middleName=null, lastName=Qian, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"content":"

born in 1974. PhD, professor, PhD supervisor. Senior member of CCF. His main research interests include machine learning, pattern recognition, and intelligent systems

"}, bioImg=null, bioContent=

born in 1974. PhD, professor, PhD supervisor. Senior member of CCF. His main research interests include machine learning, pattern recognition, and intelligent systems

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329456560599838911, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560440455357, language=CN, stringName=钱江波, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"img":"A6oQzcbvA095PSP+JHyWSQ==","content":"

1974年生. 博士,教授,博士生导师. CCF高级会员. 主要研究方向为机器学习、模式识别、智能系统

"}, bioImg=A6oQzcbvA095PSP+JHyWSQ==, bioContent=

1974年生. 博士,教授,博士生导师. CCF高级会员. 主要研究方向为机器学习、模式识别、智能系统

, aboutCorrespAuthor=null)}, companyList=null)], keywords=[Keyword(id=1329456560721473728, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, orderNo=1, keyword=few-shot learning), Keyword(id=1329456560796971201, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, orderNo=2, keyword=graph neural network), Keyword(id=1329456560864080066, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, orderNo=3, keyword=meta-learning), Keyword(id=1329456560939577539, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, orderNo=4, keyword=metric-learning), Keyword(id=1329456560998297796, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, orderNo=5, keyword=transfer-learning), Keyword(id=1329456561065406661, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, orderNo=1, keyword=小样本学习), Keyword(id=1329456561128321222, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, orderNo=2, keyword=图神经网络), Keyword(id=1329456561191235783, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, orderNo=3, keyword=元学习), Keyword(id=1329456561249956040, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, orderNo=4, keyword=度量学习), Keyword(id=1329456561312870601, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, orderNo=5, keyword=迁移学习)], refs=[Reference(id=1329456563686846692, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=null, journalName=null, refType=null, unstructuredReference=葛轶洲,刘恒,王言,等. 小样本困境下的深度学习图像识别综述[J]. 软件学报,2022,33(1):193−210, articleTitle=null, refAbstract=null), Reference(id=1329456563762344165, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=1, authorNames=null, journalName=null, refType=null, unstructuredReference=Ge Yizhou, Liu Heng, Wang Yan, et al. Survey on deep learning image recognition in dilemma of small samples[J]. Journal of Software, 2022, 33(1): 193−210 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456563825258726, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=2, authorNames=null, journalName=null, refType=null, unstructuredReference=李凡长,刘洋,吴鹏翔,等. 元学习研究综述[J]. 计算机学报,2021,44(2):422−446, articleTitle=null, refAbstract=null), Reference(id=1329456563892367591, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=3, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Fanchang, Liu Yang, Wu Pengxiang, et al. A survey on recent advance in meta-learning[J]. Chinese Journal of Computers, 2021, 44(2): 422−446 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456563963670760, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=4, authorNames=null, journalName=null, refType=null, unstructuredReference=Sun Qianru, Liu Yaoyao, Chua T, et al. Meta-transfer learning for few-shot learning [C] //Proc of the 32nd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 403−412, articleTitle=null, refAbstract=null), Reference(id=1329456564064334057, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=5, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang Peng, Liu Lingqiao, Shen Chunhua, et al. Multi-attention network for one shot learning [C] //Proc of the 30th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2017: 2721−2729, articleTitle=null, refAbstract=null), Reference(id=1329456564185968874, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=6, authorNames=null, journalName=null, refType=null, unstructuredReference=Khodadadeh S, Boloni L, Shan M. Unsupervised meta-learning for few-shot image classification [C] //Proc of the 33rd Annual Conf on Neural Information-Processing Systems. Cambrige, MA: MIT, 2019: 10132−10142, articleTitle=null, refAbstract=null), Reference(id=1329456564253077739, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=7, authorNames=null, journalName=null, refType=null, unstructuredReference=赵凯琳,靳小龙,王元卓. 小样本学习研究综述[J]. 软件学报,2021,32(2):349−369, articleTitle=null, refAbstract=null), Reference(id=1329456564315992300, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=8, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhao Kailin, Jin Xiaolong, Wang Yuanzhuo. Survey on few-shot learning[J]. Journal of Software, 2021, 32(2): 349−369 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456564408266989, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=7, rfOrder=9, authorNames=null, journalName=null, refType=null, unstructuredReference=Howard J, Ruder S. Universal language model fine-tuning for text classification[J]. arXiv preprint, arXiv: 1801. 06146, 2018, articleTitle=null, refAbstract=null), Reference(id=1329456564475375854, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=8, rfOrder=10, authorNames=null, journalName=null, refType=null, unstructuredReference=Royle J, Dorazio R, Link W. Analysis of multinomial models with unknown index using data augmentation[J]. Journal of Computational and Graphical Statistics, 2007, 16(1): 67−85, articleTitle=null, refAbstract=null), Reference(id=1329456564542484719, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=9, rfOrder=11, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu Zicheng, Li Siyuan, Wu Di, et al. AutoMix: Unveiling the power of mixup for stronger classifiers [C] //Proc of the 17th European Conf on Computer Vision. Berlin: Springer, 2022: 441−458, articleTitle=null, refAbstract=null), Reference(id=1329456564613787888, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=12, authorNames=null, journalName=null, refType=null, unstructuredReference=Kipf T, Welling M. Semi-supervised classification with graph convolutional networks[J]. arXiv preprint, arXiv: 1609. 02907, 2016, articleTitle=null, refAbstract=null), Reference(id=1329456564685091057, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=13, authorNames=null, journalName=null, refType=null, unstructuredReference=Blaes S, Burwick T. Few-shot learning in deep networks through global prototyping[J]. Neural Networks, 2017, 94: 159−172, articleTitle=null, refAbstract=null), Reference(id=1329456564743811314, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=12, rfOrder=14, authorNames=null, journalName=null, refType=null, unstructuredReference=Rahman S, Khan S, Porikli F. A unified approach for conventional zero-shot, generalized zero-shot, and few-shot learning[J]. IEEE Transactions on Image Processing, 2018, 27((11): ): 5652−5267, articleTitle=null, refAbstract=null), Reference(id=1329456564810920179, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=15, authorNames=null, journalName=null, refType=null, unstructuredReference=Cheng Yu, Yu Mo, Guo Xiaoxiao, et al. Few-shot learning with meta metric learners[J]. arXiv preprint, arXiv: 1901. 09890, 2019, articleTitle=null, refAbstract=null), Reference(id=1329456564869640436, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=14, rfOrder=16, authorNames=null, journalName=null, refType=null, unstructuredReference=马帅,刘建伟,左信. 图神经网络综述[J]. 计算机研究与发展,2022,59(1):47−80, articleTitle=null, refAbstract=null), Reference(id=1329456564949332213, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=14, rfOrder=17, authorNames=null, journalName=null, refType=null, unstructuredReference=Ma Shuai, Liu Jianwei, Zuo Xin. Survey of graph neural networks[J]. Journal of Computer Research and Development, 2022, 59(1): 47−80 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456565058384118, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=18, authorNames=null, journalName=null, refType=null, unstructuredReference=谢小杰,梁英,王梓森,等. 基于图卷积的异质网络节点分类方法[J]. 计算机研究与发展,2022,59(7):1470−1485, articleTitle=null, refAbstract=null), Reference(id=1329456565167436023, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=19, authorNames=null, journalName=null, refType=null, unstructuredReference=Xie Xiaojie, Liang Ying, Wang Zisen, et al. Node classification method for heterogeneous networks based on graph convolution[J]. Journal of Computer Research and Development, 2022, 59(7): 1470−1485 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456565230350584, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=20, authorNames=null, journalName=null, refType=null, unstructuredReference=任嘉睿,张海燕,朱梦涵,等. 基于元图卷积的异质网络嵌入学习算法[J]. 计算机研究与发展,2022,59(8):1683−1693, articleTitle=null, refAbstract=null), Reference(id=1329456565293265145, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Ren Jiarui, Zhang Haiyan, Zhu Menghan, et al. Embedding learning algorithm for heterogeneous networks based on metagram convolution[J]. Journal of Computer Research and Development, 2022, 59(8): 1683−1693 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456565389734138, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=22, authorNames=null, journalName=null, refType=null, unstructuredReference=Xu Shiyao, Xiang Yang. Frog-GNN: Multi-perspective aggregation based graph neural network for few-shot text classification[J]. Expert Systems with Applications, 2021, 176: 114795, articleTitle=null, refAbstract=null), Reference(id=1329456565482008827, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=23, authorNames=null, journalName=null, refType=null, unstructuredReference=Lyu Chen, Liu Weijie, Wang Ping. Few-shot text classification with edge-labeling graph neural network-based prototypical network [C] //Proc of the 28th Int Conf on Computational Linguistics. Stroudsburg, PA: ACL, 2020: 5547−552, articleTitle=null, refAbstract=null), Reference(id=1329456565549117692, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=24, authorNames=null, journalName=null, refType=null, unstructuredReference=Ding Ling, Chen Xiaojun, Xiang Yang. Negative-supervised capsule graph neural network for few-shot text classification[J]. Journal of Intelligent & Fuzzy Systems, 2021, 41(6): 6875−6887, articleTitle=null, refAbstract=null), Reference(id=1329456565624615165, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=25, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang Yuxiang, Li Wei, Zhang Mengmeng, et al. Dual graph cross-domain few-shot learning for hyperspectral image classification [C] //Proc of the 47th IEEE Int Conf on Acoustics Speech and Signal Processing. Piscataway, NJ: IEEE, 2022: 3573−3577, articleTitle=null, refAbstract=null), Reference(id=1329456565683335422, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=26, authorNames=null, journalName=null, refType=null, unstructuredReference=Song Xiaofa, Mao Mingyi, Qian Xiaohua. Auto-metric graph neural network based on a meta-learning strategy for the diagnosis of Alzheimer’s disease[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(8): 3141−3152, articleTitle=null, refAbstract=null), Reference(id=1329456565771415807, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=27, authorNames=null, journalName=null, refType=null, unstructuredReference=Xiong chao, Li wen, Wang Minghui, et al. Multi-dimensional edge features graph neural network on few-shot image classification[J]. IEEE Signal Processing Letters, 2021, 28: 573−577, articleTitle=null, refAbstract=null), Reference(id=1329456565863690496, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=28, authorNames=null, journalName=null, refType=null, unstructuredReference=刘颖,雷研博,范九伦,等. 基于小样本学习的图像分类技术综述[J]. 自动化学报,2021,47(2):297−315, articleTitle=null, refAbstract=null), Reference(id=1329456565934993665, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=29, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu Ying, Lei Yanbo, Fan Jiulun, et al. Survey on image classification technology based on small sample learning[J]. Acta Automatica Sinica, 2021, 47(2): 297−315 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456566027268354, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=30, authorNames=null, journalName=null, refType=null, unstructuredReference=徐冰冰,岑科廷,黄俊杰,等. 图卷积神经网络综述[J]. 计算机学报,2020,43(5):755−780, articleTitle=null, refAbstract=null), Reference(id=1329456566098571523, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=31, authorNames=null, journalName=null, refType=null, unstructuredReference=Xu Bingbing, Cen Keting, Huang Junjie, et al. A survey on graph convolutional neural networks[J]. Chinese Journal of Computers, 2020, 43(5): 755−780 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456566182457604, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=25, rfOrder=32, authorNames=null, journalName=null, refType=null, unstructuredReference=Nakamura A, Harada T. Revisiting fine-tuning for few-shot learning[J]. arXiv preprint, arXiv: 1910. 00216, 2019, articleTitle=null, refAbstract=null), Reference(id=1329456566270537989, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=26, rfOrder=33, authorNames=null, journalName=null, refType=null, unstructuredReference=Dhillon G S, Ghaudhari P, Ravichandran A, et al. A baseline for few-shot image classification [C/OL] //Proc of the 8th Int Conf on Learning Representation. Amsterdam: Elsevier, 2020 [2023-03-26].https://openreview.net/forum?id=rylXBkrYDS, articleTitle=null, refAbstract=null), Reference(id=1329456566379589894, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=27, rfOrder=34, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang Y, Girshick R, Hebert M, et al. Low-shot learning from imaginary data [C] //Proc of the 30th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2018: 7278−7286, articleTitle=null, refAbstract=null), Reference(id=1329456566446698759, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=28, rfOrder=35, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu Bo, Wang Xudong, Dixit M, et al. Feature space transfer for data augmentation [C] //Proc of the 30th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2018: 9090−9098, articleTitle=null, refAbstract=null), Reference(id=1329456566509613320, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=29, rfOrder=36, authorNames=null, journalName=null, refType=null, unstructuredReference=Jing Kunlei, Zhang Xinman, Yang Zhiyuan, et al. Feature augmentation learning for few-shot palmprint image recognition with unconstrained acquisition [C] //Proc of the 47th IEEE Int Conf on Acoustics Speech and Signal Processing. Piscataway, NJ: IEEE, 2022: 3323−3327, articleTitle=null, refAbstract=null), Reference(id=1329456566572527881, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=30, rfOrder=37, authorNames=null, journalName=null, refType=null, unstructuredReference=蒋留兵,周小龙,姜风伟,等. 基于改进匹配网络的单样本学习[J]. 系统工程与电子技术,2019,41(6):1210−1217, articleTitle=null, refAbstract=null), Reference(id=1329456566635442442, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=30, rfOrder=38, authorNames=null, journalName=null, refType=null, unstructuredReference=Jiang Liubing, Zhou Xiaolong, Jiang Fengwei, et al. One-shot learning based on improved matching network[J]. Systems Engineering and Electronics, 2019, 41(6): 1210−1217 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456566761271563, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=31, rfOrder=39, authorNames=null, journalName=null, refType=null, unstructuredReference=Snell J, Swersky K, Zemel R. Prototypical networks for few-shot learning [C] //Proc of the 31st Annual Conf on Neural InformationProcessing Systems. Cambrige, MA: MIT, 2017: 4077−4087, articleTitle=null, refAbstract=null), Reference(id=1329456566853546252, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=32, rfOrder=40, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Wenbin, Xu Jinglin, Huo Jing, et al. Distribution consistency based covariance metric networks for few-shot learning [C] //Proc of the 33rd AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2019: 8642−8649, articleTitle=null, refAbstract=null), Reference(id=1329456566916460813, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=33, rfOrder=41, authorNames=null, journalName=null, refType=null, unstructuredReference=Jiang Wen, Huang Kai, Geng Jie, et al. Multi-scale metric learning for few-shot learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 31(3): 1091−1102, articleTitle=null, refAbstract=null), Reference(id=1329456567004541198, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=34, rfOrder=42, authorNames=null, journalName=null, refType=null, unstructuredReference=<mixed-citation>Tian Yingjie, Zhao Xiaoxi, Huang Wei. Meta-learning approaches for learning-to-learn in deep learning: A survey [J]. Neurocomputing, 2022, 494: 203−223</mixed-citation>, articleTitle=null, refAbstract=null), Reference(id=1329456568606765327, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=34, rfOrder=43, authorNames=null, journalName=null, refType=null, unstructuredReference=Tian Yingjie,Zhao Xiaoxi,Huang Wei. Meta-learning approaches for learning-to-learn in deep learning:A survey [J]. Neurocomputing,2022,494:203−223, articleTitle=null, refAbstract=null), Reference(id=1329456568728400144, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=35, rfOrder=44, authorNames=null, journalName=null, refType=null, unstructuredReference=李维刚,甘平,谢璐,等. 基于样本对元学习的小样本图像分类方法[J]. 电子学报,2022,50(2):295−304, articleTitle=null, refAbstract=null), Reference(id=1329456568799703313, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=35, rfOrder=45, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Weigang, Gan Ping, Xie Lu, et al. A few-shot image classification method by pairwise-based meta learning[J]. Acta Electronica Sinica, 2022, 50(2): 295−304 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329456568900366610, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=36, rfOrder=46, authorNames=null, journalName=null, refType=null, unstructuredReference=Ravi S, Larochelle H. Optimization as a model for few-shot learning [C/OL] //Proc of the 4th Int Conf on Learning Representation. Amsterdam: Elsevier, 2016 [2023-03-26].https://openreview.net/pdf?id=rJY0-Kcll, articleTitle=null, refAbstract=null), Reference(id=1329456568967475475, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=37, rfOrder=47, authorNames=null, journalName=null, refType=null, unstructuredReference=Finn C, Abbeel P, Levine S. Model-agnostic meta-learning for fast adaptation of deep networks [C] //Proc of the 34th ACM Int Conf on Machine Learning. New York: ACM, 2017: 1126−1135, articleTitle=null, refAbstract=null), Reference(id=1329456569063944468, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=38, rfOrder=48, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Zhenguo, Zhuo Fengwei, Chen Fei, et al. Meta-SGD: Learning to learn quickly for few-shot learning[J]. arXiv preprint, arXiv: 1707. 09835, 2017, articleTitle=null, refAbstract=null), Reference(id=1329456569160413461, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=39, rfOrder=49, authorNames=null, journalName=null, refType=null, unstructuredReference=Vinyals O, Blundell C, Lilicrap T, et al. Matching networks for one shot learning [C] //Proc of the 30th Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2016: 3630−3638, articleTitle=null, refAbstract=null), Reference(id=1329456569227522326, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=40, rfOrder=50, authorNames=null, journalName=null, refType=null, unstructuredReference=Santoro A, Bartunov S, Botvinick M, et al. Meta-learning with memory-augmented neural networks [C] //Proc of the 33rd ACM Int Conf on Machine Learning. New York: ACM, 2016: 1842−1850, articleTitle=null, refAbstract=null), Reference(id=1329456569290436887, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=41, rfOrder=51, authorNames=null, journalName=null, refType=null, unstructuredReference=Brendel W, Bethge M. Approximating CNNs with bag-of-local-features models works surprisingly well on ImageNet [C/OL] //Proc of the 7th Int Conf on Learning Representation. Amsterdam: Elsevier, 2019 [2023-03-26].https://openreview.net/forum?id=rJY0-Kcll, articleTitle=null, refAbstract=null), Reference(id=1329456569378517272, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=42, rfOrder=52, authorNames=null, journalName=null, refType=null, unstructuredReference=Lai K, Zha Daochen, Zhou Kaixiong, et al. Policy-GNN: Aggregation optimization for graph neural networks [C] //Proc of the 26th ACM SIGKDD Conf on Knowledge Discovery and Data Mining. New York: ACM, 2020: 461−471, articleTitle=null, refAbstract=null), Reference(id=1329456569437237529, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=43, rfOrder=53, authorNames=null, journalName=null, refType=null, unstructuredReference=Garcia V, Bruna J. Few-shot learning with graph neural networks [C/OL] //Proc of the 6th Int Conf on Learning Representation. Amsterdam: Elsevier, 2018 [2023-03-26].https://openreview.net/pdf ?id=BJj6qGbRW, articleTitle=null, refAbstract=null), Reference(id=1329456569508540698, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=44, rfOrder=54, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu Yanbin, Lee J, Park M, et al. Learning to propagate labels: Transductive propagation network for few-shot learning [C/OL] //Proc of the 7th Int Conf on Learning Representation. Amsterdam: Elsevier, 2019 [2023-03-26].https://openreview.net/pdf?id=SyVuRiC5K7, articleTitle=null, refAbstract=null), Reference(id=1329456569584038171, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=45, rfOrder=55, authorNames=null, journalName=null, refType=null, unstructuredReference=Qiao Limeng, Shi Yemin, Li Jia, et al. Transductive episodic-wise adaptive metric for few-shot learning [C] //Proc of the 11th IEEE Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2019: 3602−3611, articleTitle=null, refAbstract=null), Reference(id=1329456569667924252, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=46, rfOrder=56, authorNames=null, journalName=null, refType=null, unstructuredReference=Angluin D, Smith C. Inductive inference: Theory and methods[J]. ACM Computing Surveys, 1983, 15(3): 237−269, articleTitle=null, refAbstract=null), Reference(id=1329456569743421725, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=47, rfOrder=57, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen Jinyin, Lin Xiang, Xiong Hui, et al. Smoothing adversarial training for GNN[J]. IEEE Transactions on Computational Social Systems, 2020, 8(3): 618−629, articleTitle=null, refAbstract=null), Reference(id=1329456569848279326, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=48, rfOrder=58, authorNames=null, journalName=null, refType=null, unstructuredReference=Cheng Hao, Zhou J, Tay W, et al. Attentive graph neural networks for few-shot learning [C] //Proc of the 5th IEEE Int Conf on Multimedia Information Processing and Retrieval. Piscataway, NJ: IEEE, 2022: 152−157, articleTitle=null, refAbstract=null), Reference(id=1329456569923776799, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=49, rfOrder=59, authorNames=null, journalName=null, refType=null, unstructuredReference=Velickovic P, Cucurull G, Casanova A, et al. Graph attention networks[J]. arXiv preprint, arXiv: 1710. 10903, 2017, articleTitle=null, refAbstract=null), Reference(id=1329456569986691360, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=50, rfOrder=60, authorNames=null, journalName=null, refType=null, unstructuredReference=<mixed-citation>Zhang Xu, Zhang Youjia, Zhang Zuyu, et al. Discriminative learning of imaginary data for few-shot classification [J]. Neurocomputing, 2022, 467: 406−417</mixed-citation>, articleTitle=null, refAbstract=null), Reference(id=1329456570049605921, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=50, rfOrder=61, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang Xu,Zhang Youjia,Zhang Zuyu,et al. Discriminative learning of imaginary data for few-shot classification [J]. Neurocomputing,2022,467:406−417, articleTitle=null, refAbstract=null), Reference(id=1329456570120909090, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=51, rfOrder=62, authorNames=null, journalName=null, refType=null, unstructuredReference=Xu Rui, Liu Baodi, Lu Xiaoping, et al. DMH-FSL: Dual-modal hypergraph for few-shot learning[J]. Neural Processing Letters, 2022, 54(2): 1317−1332, articleTitle=null, refAbstract=null), Reference(id=1329456570188017955, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=52, rfOrder=63, authorNames=null, journalName=null, refType=null, unstructuredReference=Yadati N, Nimishakavi M, Yadav P, et al. HyperGCN: A new method for training graph convolutional networks on hypergraphs [C] //Proc of the 33rd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2019: 1509−1520, articleTitle=null, refAbstract=null), Reference(id=1329456570255126820, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=53, rfOrder=64, authorNames=null, journalName=null, refType=null, unstructuredReference=Kim J, Kim T, Kim S, et al. Edge-labeling graph neural network for few-shot learning [C] //Proc of the 32nd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 11−20, articleTitle=null, refAbstract=null), Reference(id=1329456570334818597, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=54, rfOrder=65, authorNames=null, journalName=null, refType=null, unstructuredReference=Zuo Xibing, Yu Xuchun, Liu Bing, et al. FSL-EGNN: Edge-labeling graph neural network for hyperspectral image few-shot classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1−18, articleTitle=null, refAbstract=null), Reference(id=1329456570410316070, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=55, rfOrder=66, authorNames=null, journalName=null, refType=null, unstructuredReference=Xiong Chao, Li Wen, Liu Yun, et al. Multi-dimensional edge features graph neural network on few-shot image classification[J]. IEEE Signal Processing Letters, 2021, 28: 573−577, articleTitle=null, refAbstract=null), Reference(id=1329456570506785063, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=56, rfOrder=67, authorNames=null, journalName=null, refType=null, unstructuredReference=Tang Shixiang, Chen Dapeng, Bai Lei, et al. Mutual CRF-GNN for few-shot learning [C] //Proc of the 34th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 2329−2339, articleTitle=null, refAbstract=null), Reference(id=1329456570565505320, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=57, rfOrder=68, authorNames=null, journalName=null, refType=null, unstructuredReference=<mixed-citation>Bale T, Vale W. CRF and CRF receptors: Role in stress responsivity and other behaviors [J]. Annual Review of Pharmacology and Toxicology, 2004, 44: 525−557</mixed-citation>, articleTitle=null, refAbstract=null), Reference(id=1329456570620031273, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=57, rfOrder=69, authorNames=null, journalName=null, refType=null, unstructuredReference=Bale T,Vale W. CRF and CRF receptors:Role in stress responsivity and other behaviors [J]. Annual Review of Pharmacology and Toxicology,2004,44:525−557, articleTitle=null, refAbstract=null), Reference(id=1329456570687140138, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=58, rfOrder=70, authorNames=null, journalName=null, refType=null, unstructuredReference=Yang Lin, Li Liangliang, Zhang Zilun, et al. DPGNN: Distribution propagation graph network for few-shot learning [C] //Proc of the 33rd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 13387−13396, articleTitle=null, refAbstract=null), Reference(id=1329456570745860395, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=59, rfOrder=71, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang Bailin, Ling Hefei, Shen Jialie, et al. Mixture distribution graph network for few shot learning[J]. IEEE Transactions on Cognitive and Developmental Systems, 2022, 14(3): 892−901, articleTitle=null, refAbstract=null), Reference(id=1329456570812969260, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=60, rfOrder=72, authorNames=null, journalName=null, refType=null, unstructuredReference=Hu Yufan, Gao Junyu, Xu Changsheng. Learning dual-pooling graph neural networks for few-shot video classification[J]. IEEE Transactions on Multimedia, 2020, 23: 4285−4296, articleTitle=null, refAbstract=null), Reference(id=1329456570880078125, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=61, rfOrder=73, authorNames=null, journalName=null, refType=null, unstructuredReference=Mordeson J. Fuzzy mathematics [M]. Foundations of Image Understanding. Berlin: Springer, 2001: 95−125, articleTitle=null, refAbstract=null), Reference(id=1329456570942992686, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=62, rfOrder=74, authorNames=null, journalName=null, refType=null, unstructuredReference=Wei Tong, Hou Junlin, Feng Rui. Fuzzy graph neural network for few-shot learning [C/OL] //Proc of the 30th IEEE Int Joint Conf on Neural Networks. Piscataway, NJ: IEEE, 2020 [2023-03-26].https://openreview.net/pdf?id=BJj6qGbRW, articleTitle=null, refAbstract=null), Reference(id=1329456571005907247, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=63, rfOrder=75, authorNames=null, journalName=null, refType=null, unstructuredReference=Devlin J, Chang M, Lee K, et al. Bert: Pre-training of deep bidirectional transformers for language understanding [C] //Proc of the 17th Annual Conf of the North American chapter of the Association for Computational Linguistics. Stroudsburg, PA: ACL, 2019: 4171−4186, articleTitle=null, refAbstract=null), Reference(id=1329456571060433200, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=64, rfOrder=76, authorNames=null, journalName=null, refType=null, unstructuredReference=Ma Yuqing, Bai Shihao, An Shan, et al. Transductive relation-propagation network for few-shot Learning [C] //Proc of the 29th Int Joint Conf on Artificial Intelligence. San Francisco, CA: Morgan Kaufmann, 2020: 804−810, articleTitle=null, refAbstract=null), Reference(id=1329456571114959153, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=65, rfOrder=77, authorNames=null, journalName=null, refType=null, unstructuredReference=Ma Yuqing, Bai Shihao, Liu Wei, et al. Transductive relation-propagation with decoupling training for few-shot learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(11): 6652−6664, articleTitle=null, refAbstract=null), Reference(id=1329456571182068018, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=66, rfOrder=78, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen Cen, Li Kenli, Wei Wei, 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, articleTitle=null, refAbstract=null), Reference(id=1329456571244982579, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=67, rfOrder=79, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen Chaofan, Yang Xiaoshan, Xu Changsheng, et al. ECKPN: Explicit class knowledge propagation network for transductive few-shot learning [C] //Proc of the 34th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 6596−6605, articleTitle=null, refAbstract=null), Reference(id=1329456571312091444, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=68, rfOrder=80, authorNames=null, journalName=null, refType=null, unstructuredReference=Pennington J, Socher R, Manning C. Glove: Global vectors for word representation [C] //Proc of the 28th Int Conf on Empirical Methods in Natural Language Processing. Stroudsburg, PA: ACL, 2014: 1532−1543, articleTitle=null, refAbstract=null), Reference(id=1329456571383394613, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=69, rfOrder=81, authorNames=null, journalName=null, refType=null, unstructuredReference=Yu Tianyuan, He Sen, Song Yizhe, et al. Hybrid graph neural networks for few-shot learning [C] //Proc of the 36th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2022: 3179−3187, articleTitle=null, refAbstract=null), Reference(id=1329456571454697782, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=70, rfOrder=82, authorNames=null, journalName=null, refType=null, unstructuredReference=Deng Fei, Pu Shengliang, Chen Xuehong, et al. Hyperspectral image classification with capsule network using limited training samples[J]. Sensors, 2018, 18(9): 3153−3174, articleTitle=null, refAbstract=null), Reference(id=1329456573065310521, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=71, rfOrder=83, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang Ruiheng, Yang Shuo, Zhang Qi, et al. Graph-based few-shot learning with transformed feature propagation and optimal class allocation[J]. Neurocomputing, 2022, 470: 247−256, articleTitle=null, refAbstract=null), Reference(id=1329456573132419386, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=72, rfOrder=84, authorNames=null, journalName=null, refType=null, unstructuredReference=Tantawi A, Towsley G, Wolf J. Optimal allocation of multiple class resources in computer systems [C] //Proc of the 16th ACM SIGMETRICS Conf on Measurement and Modeling of Computer Systems. New York: ACM, 1988: 253−260, articleTitle=null, refAbstract=null), Reference(id=1329456573182751035, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=73, rfOrder=85, authorNames=null, journalName=null, refType=null, unstructuredReference=Cuturi M. Sinkhorn distances: Lightspeed computation of optimal transport [C] //Proc of the 27th Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2013: 2292−2300, articleTitle=null, refAbstract=null), Reference(id=1329456573245665596, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=74, rfOrder=86, authorNames=null, journalName=null, refType=null, unstructuredReference=Malalur P, Jaakkola T. Alignment based matching networks for one-shot classification and open-set recognition[J]. arXiv preprint, arXiv: 1903. 06538, 2019, articleTitle=null, refAbstract=null), Reference(id=1329456573333745981, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=75, rfOrder=87, authorNames=null, journalName=null, refType=null, unstructuredReference=Ren M, Triantafillou E, Ravi S, et al. Meta-learning for semi-supervised few-shot classification [C/OL]//Proc of the 6th Int Conf on Learning Representation. Amsterdam: Elsevier, 2018 [2023-03-26].https://openreview.net/pdf?id=HJcSzz-CZ, articleTitle=null, refAbstract=null), Reference(id=1329456573409243454, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=76, rfOrder=88, authorNames=null, journalName=null, refType=null, unstructuredReference=Cui Yin , Zhou Feng, Lin Yuanqing, et al. Fine-grained categorization and dataset bootstrapping using deep metric learning with humans in the loop [C] //Proc of the 29th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2016: 1153−1162, articleTitle=null, refAbstract=null), Reference(id=1329456573472158015, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=77, rfOrder=89, authorNames=null, journalName=null, refType=null, unstructuredReference=Singla S, Singla S, Feizi S. Improved deterministic l2 robustness on CIFAR-10 and CIFAR-100 [C/OL] //Proc of the 10th Int Conf on Learning Representation. Amsterdam: Elsevier, 2022 [2023-03-26].https://openreview.net/forum?id=tD7eCtaSkR, articleTitle=null, refAbstract=null), Reference(id=1329456573539266880, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=78, rfOrder=90, authorNames=null, journalName=null, refType=null, unstructuredReference=Deng J, Dong W, Socher R, et al. ImageNet: A large-scale hierarchical image database [C] //Proc of the 22nd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2009: 248−255, articleTitle=null, refAbstract=null), Reference(id=1329456573602181441, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=79, rfOrder=91, authorNames=null, journalName=null, refType=null, unstructuredReference=He Kaiming, Zhang Xianyu, Ren Shaoqing, et al. Deep residual learning for image recognition [C] //Proc of the 29th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2016: 770−778, articleTitle=null, refAbstract=null), Reference(id=1329456573669290306, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=80, rfOrder=92, authorNames=null, journalName=null, refType=null, unstructuredReference=Lin Guangfeng, Yang Ying, Fan Yindi, et al. High-order structure preserving graph neural network for few-shot learning[J]. arXiv preprint, arXiv: 2005. 14415, 2020, articleTitle=null, refAbstract=null), Reference(id=1329456573736399171, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=81, rfOrder=93, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang Yaqing, Yao Quanming, Kwok J, et al. Generalizing from a few examples: A survey on few-shot learning[J]. ACM Computing Surveys, 2020, 53(3): 1−34, articleTitle=null, refAbstract=null), Reference(id=1329456573795119428, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=82, rfOrder=94, authorNames=null, journalName=null, refType=null, unstructuredReference=Bendre N, Marin H, Najafirad P. Learning from few samples: A survey[J]. arXiv preprint, arXiv: 2007. 15484, 2020, articleTitle=null, refAbstract=null), Reference(id=1329456573866422597, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=83, rfOrder=95, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Na, Zhou Deyun, Shi Jiao, et al. Graph-based deep multitask few-shot learning for hyperspectral image classification[J]. Remote Sensing, 2022, 14(9): 2246−2267, articleTitle=null, refAbstract=null), Reference(id=1329456573937725767, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=84, rfOrder=96, authorNames=null, journalName=null, refType=null, unstructuredReference=Tong Xinyi, Yin Jihao, Han Bingnan, et al. Few-shot learning with attention-weighted graph convolutional networks for hyperspectral image classification [C] //Proc of the 27th IEEE Int Conf on Image Processing. Piscataway, NJ: IEEE, 2020: 1686−1690, articleTitle=null, refAbstract=null), Reference(id=1329456574021611848, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=85, rfOrder=97, authorNames=null, journalName=null, refType=null, unstructuredReference=Hong Danfeng, Gao Lianru, Yao Jing, et al. Graph convolutional networks for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(7): 5966−5978, articleTitle=null, refAbstract=null), Reference(id=1329456574080332105, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=86, rfOrder=98, authorNames=null, journalName=null, refType=null, unstructuredReference=Xiao Weiwei, Song Kechen, Liu Jie, et al. Graph embedding and optimal transport for few-shot classification of metal surface defect[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1−10, articleTitle=null, refAbstract=null), Reference(id=1329456574151635274, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=87, rfOrder=99, authorNames=null, journalName=null, refType=null, unstructuredReference=Bao Yanqi, Song Kechen, Liu Jie, et al. Triplet-graph reasoning network for few-shot metal generic surface defect segmentation[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1−11, articleTitle=null, refAbstract=null), Reference(id=1329456574214549835, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=88, rfOrder=100, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Feimo, Li Shuaibo, Fan Xinxin, et al. Structural attention enhanced continual meta-learning for graph edge labeling based few-shot remote sensing scene classification[J]. Remote Sensing, 2022, 14(3): 485−515, articleTitle=null, refAbstract=null), Reference(id=1329456574319407436, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=89, rfOrder=101, authorNames=null, journalName=null, refType=null, unstructuredReference=Yuan Zhengwu, Huang Wendong, Tang Chan, et al. Graph-based embedding smoothing network for few-shot scene classification of remote sensing images[J]. Remote Sensing, 2022, 14(5): 1161−1179, articleTitle=null, refAbstract=null), Reference(id=1329456574390710605, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=90, rfOrder=102, authorNames=null, journalName=null, refType=null, unstructuredReference=Guo Xinyu, Tian Bingjie, Tian Xuedong. HFGNN-proto: Hesitant fuzzy graph neural network-based prototypical network for few-shot text classification[J]. Electronics, 2022, 11(15): 2423−2437, articleTitle=null, refAbstract=null), Reference(id=1329456574453625166, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=91, rfOrder=103, authorNames=null, journalName=null, refType=null, unstructuredReference=Ma Ning, Bu Jiajun, Yang Jieyu, et al. Adaptive-step graph meta-learner for few-shot graph classification [C] //Proc of the 29th ACM Int Conf on Information &amp; Knowledge Management. New York: ACM, 2020: 1055−1064, articleTitle=null, refAbstract=null), Reference(id=1329456574524928335, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=92, rfOrder=104, authorNames=null, journalName=null, refType=null, unstructuredReference=Hospedales T, Antoniou A, Micaelli P, et al. Meta-learning in neural networks: A survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(9): 5149−5169, articleTitle=null, refAbstract=null), Reference(id=1329456574600425808, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=93, rfOrder=105, authorNames=null, journalName=null, refType=null, unstructuredReference=Pan Jiacheng, Lin Haocai, Dong Yihong, et al. MAMF-GCN: Multi-scale adaptive multi-channel fusion deep graph convolutional network for predicting mental disorder[J]. Computers in Biology and Medicine, 2022, 148: 105823, articleTitle=null, refAbstract=null), Reference(id=1329456574671728977, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=94, rfOrder=106, authorNames=null, journalName=null, refType=null, unstructuredReference=Wei Yinwei, Wang Xiang, Nie Liqiang, et al. MMGCN: Multi-modal graph convolution network for personalized recommendation of micro-video [C] //Proc of the 27th ACM Int Conf on Multimedia. New York: ACM, 2019: 1437−144, articleTitle=null, refAbstract=null), Reference(id=1329456574784975186, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=95, rfOrder=107, authorNames=null, journalName=null, refType=null, unstructuredReference=Tao Zhulin, Wei Yinwei, Wang Xiang, et al. MGAT: Multimodal graph attention network for recommendation[J]. Information Processing & Management, 2020, 57(5): 102277, articleTitle=null, refAbstract=null), Reference(id=1329456574843695443, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=96, rfOrder=108, authorNames=null, journalName=null, refType=null, unstructuredReference=Hsu K, Levine S, Finn C. Unsupervised learning via meta-learning [C/OL] //Proc of the 7th Int Conf on Learning Representation. Amsterdam: Elsevier, 2018 [2023-03-26].https://openreview.net/forum?id=r1My6sR9tX, articleTitle=null, refAbstract=null), Reference(id=1329456574902415700, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=97, rfOrder=109, authorNames=null, journalName=null, refType=null, unstructuredReference=Antoniou A, Storkey A. Assume, augment and learn: Unsupervised few-shot meta-learning via random labels and data augmentation[J]. arXiv preprint, arXiv: 1902. 09884, 2019, articleTitle=null, refAbstract=null), Reference(id=1329456574965330261, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=98, rfOrder=110, authorNames=null, journalName=null, refType=null, unstructuredReference=Veeriah V, Hessel M, Xu Zhongwen, et al. Discovery of useful questions as auxiliary tasks [C] // Proc of the 33rd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2019: 9306−9317, articleTitle=null, refAbstract=null), Reference(id=1329456575036633430, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=99, rfOrder=111, authorNames=null, journalName=null, refType=null, unstructuredReference=Zheng Zeyu, Oh J, Singh S. On learning intrinsic rewards for policy gradient methods [C] // Proc of the 32nd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2018: 4649−4659, articleTitle=null, refAbstract=null), Reference(id=1329456575095353687, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=100, rfOrder=112, authorNames=null, journalName=null, refType=null, unstructuredReference=Meier F, Kappler D, Schaal S. Online learning of a memory for learning rates [C] //Proc of the 4th IEEE Int Conf on Robotics and Automation. Piscataway, NJ: IEEE, 2018: 2425−2432, articleTitle=null, refAbstract=null), Reference(id=1329456575154073944, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=101, rfOrder=113, authorNames=null, journalName=null, refType=null, unstructuredReference=Tseng H, Lee H, Huang Jiabin, et al. Cross-domain few-shot classification via learned feature-wise transformation [C/OL] //Proc of the 9th Int Conf on Learning Representation. Amsterdam: Elsevier, 2020 [2023-03-26].https://openreview.net/forum?id=SJl5Np4tPr, articleTitle=null, refAbstract=null), Reference(id=1329456575221182809, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=102, rfOrder=114, authorNames=null, journalName=null, refType=null, unstructuredReference=Yun S, Jeong M, Kim R, et al. Graph transformer networks [C] // Proc of the 33rd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2019: 11960−11970, articleTitle=null, refAbstract=null), Reference(id=1329456575279903066, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=103, rfOrder=115, authorNames=null, journalName=null, refType=null, unstructuredReference=Jiang Bo, Zhao Kangkang, Tang jin, et al. RGTransformer: Region-graph transformer for image representation and few-shot classification[J]. IEEE Signal Processing Letters, 2022, 29: 792−796, articleTitle=null, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1329456559731617966, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, xref=1, ext=[AuthorCompanyExt(id=1329456559735812271, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, companyId=1329456559731617966, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, Zhejiang 315211), AuthorCompanyExt(id=1329456559744200880, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, companyId=1329456559731617966, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1宁波大学信息科学与工程学院 浙江宁波 315211)]), AuthorCompany(id=1329456559815504049, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, xref=2, ext=[AuthorCompanyExt(id=1329456559828086962, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, companyId=1329456559815504049, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Key Laboratory of Mobile Network Application Technology of Zhejiang province(Ningbo University), Ningbo, Zhejiang 315211), AuthorCompanyExt(id=1329456559836475571, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, companyId=1329456559815504049, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2浙江省移动网络应用技术重点实验室(宁波大学) 浙江宁波 315211)])], figs=[ArticleFig(id=1329456561421922506, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 1, caption=Illustration of task data setup, figureFileSmall=Zo6B/pz20ta9XRqqBBf8bg==, figureFileBig=cjvv6pxpj7uioecOkb+iyg==, tableContent=null), ArticleFig(id=1329456561493225675, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图1, caption=任务数据设置示意图, figureFileSmall=Zo6B/pz20ta9XRqqBBf8bg==, figureFileBig=cjvv6pxpj7uioecOkb+iyg==, tableContent=null), ArticleFig(id=1329456561811992780, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 2, caption=Process of training stage, figureFileSmall=d46AiQNyfZkUlVcOh3rFyw==, figureFileBig=7T+g9h3z+1qfCwr8unrQWA==, tableContent=null), ArticleFig(id=1329456561895878861, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图2, caption=训练阶段流程, figureFileSmall=d46AiQNyfZkUlVcOh3rFyw==, figureFileBig=7T+g9h3z+1qfCwr8unrQWA==, tableContent=null), ArticleFig(id=1329456561971376334, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 3, caption=Basic flow of FSL method based on GNN, figureFileSmall=0yxd/RFlB+6GcgG1LoLg8Q==, figureFileBig=3FJ7YSeNRIq8ipTfoaAkMw==, tableContent=null), ArticleFig(id=1329456562042679503, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图3, caption=基于GNN的FSL方法基本流程, figureFileSmall=0yxd/RFlB+6GcgG1LoLg8Q==, figureFileBig=3FJ7YSeNRIq8ipTfoaAkMw==, tableContent=null), ArticleFig(id=1329456562105594064, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 4, caption=The frame diagram of GNN-FSL model [43], figureFileSmall=cpkxVuPAQt/N7fK+bwnutw==, figureFileBig=A31tPeQtQ7xshdfbLeIyKw==, tableContent=null), ArticleFig(id=1329456562160120017, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图4, caption=GNN-FSL模型框架图[43], figureFileSmall=cpkxVuPAQt/N7fK+bwnutw==, figureFileBig=A31tPeQtQ7xshdfbLeIyKw==, tableContent=null), ArticleFig(id=1329456562214645970, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 5, caption=The framework of EGNN model[53], figureFileSmall=hXaYaGyG8qzU5XAX+YZwdg==, figureFileBig=PNUD27AQu/mZjzzH3J/FFw==, tableContent=null), ArticleFig(id=1329456562277560531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图5, caption=EGNN模型框架, figureFileSmall=hXaYaGyG8qzU5XAX+YZwdg==, figureFileBig=PNUD27AQu/mZjzzH3J/FFw==, tableContent=null), ArticleFig(id=1329456562332086484, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 6, caption=The frame diagram of Frog-GNN model[17], figureFileSmall=z0FUC30D7PSEIigzJxNH6A==, figureFileBig=ZbVIIUDFS9TNbdscnEXKnQ==, tableContent=null), ArticleFig(id=1329456562504052949, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图6, caption=Frog-GNN模型框架图[17], figureFileSmall=z0FUC30D7PSEIigzJxNH6A==, figureFileBig=ZbVIIUDFS9TNbdscnEXKnQ==, tableContent=null), ArticleFig(id=1329456562566967510, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 7, caption=The frame diagram of TPRN model[65], figureFileSmall=JXXdeqgwdEnrPu9F/MhyeQ==, figureFileBig=u0OPcraJWZLy6IBeZB2s/w==, tableContent=null), ArticleFig(id=1329456562629882071, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图7, caption=TPRN模型框架图[65], figureFileSmall=JXXdeqgwdEnrPu9F/MhyeQ==, figureFileBig=u0OPcraJWZLy6IBeZB2s/w==, tableContent=null), ArticleFig(id=1329456562705379544, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Figure 8, caption=The frame diagram of hierarchical GNN[66], figureFileSmall=Z24FyNj4JMW9SA4hKF2aCw==, figureFileBig=6k6+dwPQJNdwgZIrsHLaxQ==, tableContent=null), ArticleFig(id=1329456562768294105, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=图8, caption=分层GNN框架图[66], figureFileSmall=Z24FyNj4JMW9SA4hKF2aCw==, figureFileBig=6k6+dwPQJNdwgZIrsHLaxQ==, tableContent=null), ArticleFig(id=1329456562831208666, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Table 1, caption=

Comparison of Advantages and Disadvantages of Few-Shot Learning Methods

, figureFileSmall=null, figureFileBig=null, tableContent=
分类体系优点缺点
基于模型微调的方法方法简单,仅需要设计模型参数;
调整模块,而无需考虑模型构建.
当目标数据集与源数据集分布相差较大时,
将导致模型在目标数据集上过拟合.
基于数据增强的方法一定程度地缓解了模型过拟合问题,
通常专注于对数据属性的增强,
更高效地利用现有数据,
避免对模型的优化与调整.
可能会产生噪声数据或噪声特征,
难以达到最佳效果.
基于迁移学习的方法
基于度量学习的方法思想易于理解,可解释性较强,
便于计算和公式化,可扩展性强,
能够和其他FSL方法结合.
在样本量较少的情况下,简单地通过
距离度量的方法使得最终效果不佳.
基于元学习的方法受到人类学习的启发,使模型具备学习的
能力,能够学习到除训练之外的知识,并利用
已有知识,指导模型更快地适应新任务.
可解释性强,能够更准确地度量样本间关系.
模型复杂度较高、训练时间较长,
因此该类方法的时间成本与设备成本
较高,且模型效果仍有较大的提升空间.
基于GNN的方法GNN中节点分类的思想能够自然地应用于
小样本分类问题,且该类FSL方法性能较好.
不可避免地存在节点过平滑的
问题,由于该方法较为新颖,其
需要改进和发展的方面还有很多.
), ArticleFig(id=1329456562889928923, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=表1, caption=

小样本学习方法优缺点对比

, figureFileSmall=null, figureFileBig=null, tableContent=
分类体系优点缺点
基于模型微调的方法方法简单,仅需要设计模型参数;
调整模块,而无需考虑模型构建.
当目标数据集与源数据集分布相差较大时,
将导致模型在目标数据集上过拟合.
基于数据增强的方法一定程度地缓解了模型过拟合问题,
通常专注于对数据属性的增强,
更高效地利用现有数据,
避免对模型的优化与调整.
可能会产生噪声数据或噪声特征,
难以达到最佳效果.
基于迁移学习的方法
基于度量学习的方法思想易于理解,可解释性较强,
便于计算和公式化,可扩展性强,
能够和其他FSL方法结合.
在样本量较少的情况下,简单地通过
距离度量的方法使得最终效果不佳.
基于元学习的方法受到人类学习的启发,使模型具备学习的
能力,能够学习到除训练之外的知识,并利用
已有知识,指导模型更快地适应新任务.
可解释性强,能够更准确地度量样本间关系.
模型复杂度较高、训练时间较长,
因此该类方法的时间成本与设备成本
较高,且模型效果仍有较大的提升空间.
基于GNN的方法GNN中节点分类的思想能够自然地应用于
小样本分类问题,且该类FSL方法性能较好.
不可避免地存在节点过平滑的
问题,由于该方法较为新颖,其
需要改进和发展的方面还有很多.
), ArticleFig(id=1329456562957037788, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Table 2, caption=

Symbol Description

, figureFileSmall=null, figureFileBig=null, tableContent=
符号描述
Dbase训练集
Dnovel测试集
mbase训练集中样本数量
mnovel测试集中样本数量
$ {x_i} $样本i实例
$ {y_i} $样本i实际标签
$ {\overline y _i} $样本i预测标签
Cbase用于构建训练任务的类集合
Cnovel用于构建测试任务的类集合
N每轮任务中类的数量
K支持集中每一种类的样本数量
H查询集中每一种类的样本数量
$ {T_{{\text{train}}}} $训练任务
$ {T_{{\text{test}}}} $测试任务
$ \mathit{\theta } $图神经网络参数
$ \mathit{S} $支持集
Q查询集
L损失函数
Y样本标签矩阵
A邻接矩阵
D度矩阵
fi样本i对应的特征向量
ni样本i对应的节点表示
F样本特征向量矩阵
), ArticleFig(id=1329456563036729565, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=表2, caption=

符号说明

, figureFileSmall=null, figureFileBig=null, tableContent=
符号描述
Dbase训练集
Dnovel测试集
mbase训练集中样本数量
mnovel测试集中样本数量
$ {x_i} $样本i实例
$ {y_i} $样本i实际标签
$ {\overline y _i} $样本i预测标签
Cbase用于构建训练任务的类集合
Cnovel用于构建测试任务的类集合
N每轮任务中类的数量
K支持集中每一种类的样本数量
H查询集中每一种类的样本数量
$ {T_{{\text{train}}}} $训练任务
$ {T_{{\text{test}}}} $测试任务
$ \mathit{\theta } $图神经网络参数
$ \mathit{S} $支持集
Q查询集
L损失函数
Y样本标签矩阵
A邻接矩阵
D度矩阵
fi样本i对应的特征向量
ni样本i对应的节点表示
F样本特征向量矩阵
), ArticleFig(id=1329456563103838430, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Table 3, caption=

Statistics and Division of Datasets

, figureFileSmall=null, figureFileBig=null, tableContent=
数据集类别数量图像数量训练数量/验证数量/测试数量
MiniimageNet1006000064/16/20
TieredimageNet608779165351/97/160
CUB-200-201120011788100/50/50
CIFAR-FS1006000064/16/20
Stanford Cars19616185130/17/49
), ArticleFig(id=1329456563191918815, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=表3, caption=

数据集的统计与划分

, figureFileSmall=null, figureFileBig=null, tableContent=
数据集类别数量图像数量训练数量/验证数量/测试数量
MiniimageNet1006000064/16/20
TieredimageNet608779165351/97/160
CUB-200-201120011788100/50/50
CIFAR-FS1006000064/16/20
Stanford Cars19616185130/17/49
), ArticleFig(id=1329456563271610592, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Table 4, caption=

Accuracy of GNN-Based Few-Shot Image Classification Learning Method on Two Datasets %

, figureFileSmall=null, figureFileBig=null, tableContent=
方法模型主干MiniimageNetTieredImageNet

1-shot

5-shot

1-shot

5-shot
基于节点特征GNN-FSL[43]Conv450.3366.4154.9770.92
TPN[44]Conv455.51±0.8669.86±0.6557.53±0.9672.86±0.74
AGNN[48]Conv454.8169.8557.4772.29
AGNN*[48]Conv460.1472.4167.2379.55
DMH-FSL[51]Conv451.1867.86
文献[50]Conv461.9978.8157.8973.62
基于边缘特征EGNN[53]Conv452.8666.8570.98
EGNN*[53]Conv459.1876.3763.5280.15
MDEGNN*[22]Conv462.23±0.7378.83±0.67
MCGN[56]Conv457.89±0.8773.58 ± 0.8758.45 ± 0.5974.58 ± 0.84
MCGN*[56]Conv467.32±0.4383.03 ± 0.5471.21 ± 0.8585.98 ± 0.98
DPGN*[58]Conv466.01±0.3682.83±0.4169.43±0.4985.92±0.42
MDGN[59]Conv469.22±0.4685.35±0.72
Fuzzy GNN[62]Conv464.15±0.2880.08±0.3569.09±0.1584.13±0.18
基于节点对特征TPRN-T*[64]Conv457.84±0.5178.57±0.4459.26±0.5079.66±0.45
TPRN-D*[65]Conv462.98±0.5081.24±0.4261.01±0.4980.98±0.42
基于类级特征Hybrid GNN[69]Conv455.63±0.2272.48±0.1656.05±0.2172.82±0.18
Hierachical GNN[66]Conv460.03±0.5179.64±0.3664.32±0.4983.34±0.45
文献[71]Conv471.82±0.8883.04±0.5177.67±0.2787.98±1.01
ECKPN[67]Conv468.89±0.3483.59±0.4470.45±0.4886.74±0.42
), ArticleFig(id=1329456563342913761, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=表4, caption=

基于GNN的小样本图像分类学习方法在2个数据集上的准确率

, figureFileSmall=null, figureFileBig=null, tableContent=
方法模型主干MiniimageNetTieredImageNet

1-shot

5-shot

1-shot

5-shot
基于节点特征GNN-FSL[43]Conv450.3366.4154.9770.92
TPN[44]Conv455.51±0.8669.86±0.6557.53±0.9672.86±0.74
AGNN[48]Conv454.8169.8557.4772.29
AGNN*[48]Conv460.1472.4167.2379.55
DMH-FSL[51]Conv451.1867.86
文献[50]Conv461.9978.8157.8973.62
基于边缘特征EGNN[53]Conv452.8666.8570.98
EGNN*[53]Conv459.1876.3763.5280.15
MDEGNN*[22]Conv462.23±0.7378.83±0.67
MCGN[56]Conv457.89±0.8773.58 ± 0.8758.45 ± 0.5974.58 ± 0.84
MCGN*[56]Conv467.32±0.4383.03 ± 0.5471.21 ± 0.8585.98 ± 0.98
DPGN*[58]Conv466.01±0.3682.83±0.4169.43±0.4985.92±0.42
MDGN[59]Conv469.22±0.4685.35±0.72
Fuzzy GNN[62]Conv464.15±0.2880.08±0.3569.09±0.1584.13±0.18
基于节点对特征TPRN-T*[64]Conv457.84±0.5178.57±0.4459.26±0.5079.66±0.45
TPRN-D*[65]Conv462.98±0.5081.24±0.4261.01±0.4980.98±0.42
基于类级特征Hybrid GNN[69]Conv455.63±0.2272.48±0.1656.05±0.2172.82±0.18
Hierachical GNN[66]Conv460.03±0.5179.64±0.3664.32±0.4983.34±0.45
文献[71]Conv471.82±0.8883.04±0.5177.67±0.2787.98±1.01
ECKPN[67]Conv468.89±0.3483.59±0.4470.45±0.4886.74±0.42
), ArticleFig(id=1329456563430994146, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=EN, label=Table 5, caption=

Summary of Few-Shot Learning Methods Based on GNN

, figureFileSmall=null, figureFileBig=null, tableContent=
方法模型名称应用主要技术优点缺点
基于节点特征GNN-FSL[43]图像分类GCN1. 节点特征易获取、易利用.
2. 扩展性强,通过丰富GNN模型使节 点特征在图上进行高效的传播.
3. 易理解模型设计容易,可解释性强.
1. 对节点之间相关性的探究不足.
2. 通常利用节点特征间的相似度构建邻接矩阵,易导致不适当的特征聚合.
3. 存在过拟合的风险.
TPN[44]图像分类GCN 转导学习
AGNN[48]图像分类GCN 注意力机制
AMGNN[21]疾病诊断GCN 多模态融合
DMH-FSL[51]图像分类超图卷积
文献[50]图像分类GCN 幻觉网络
基于边缘特征EGNN[53]图像分类GCN1. 在节点特征的基础上考虑到对节点间的关系进行深入挖掘.
2. 精心设计的边缘特征既可以控制节点信息聚合的程度也可以建模节点间丰富的关系,效果提升明显.
3. 设计过的边特征可以缓解图上的过拟合问题.
1. 边缘特征的表示形式不够丰富.
2. 缺乏明确的初始边缘特征建模标准.
3. 设计合适的边缘特征更新网络较为困难. 算法复杂度较高.
文献[54]图像分类GCN
MDEGNN[55]图像分类GCN
HOSP-GNN[80]图像分类GCN 高阶关系
MCGN[56]图像分类GCN 混合高斯分布
DPGN[58]图像分类GCN 分布图
DPGNN[60]视频分类GCN 图池化
Fuzzy-GNN[62]图像分类GCN 模糊数学
基于节点对特征Frog-GNN[17]文本分类GCN1. 对节点关系进行显示建模,由此丰富了关系的表现形式.
2. 直接将关系在图上进行传播,能够学习到节点间的深入的关系.
3. 能够充分挖掘样本间相关性.
1. 关系节点的构建方法复杂.
2. 多种形式的信息在图上传播可能导致节点间过平滑.
3. 可解释性较差.
4. 算法复杂度较高.
5. 模型扩展性较差.
TPRN-T[64]图像分类GCN
TPRN-D[65]图像分类GCN 解耦学习
基于类级特征Hierachical GNN[66]图像分类GCN 图池化1. 结合类级知识能够为查询样本匹
配更高可信度的类级知识,对于5-shot的分类任务这类方法平均性能
提升明显.
2. 模型扩展性较强,改进方式较多.
1. 对于1-shot的分类任务,类级知识对模型性能的提升相当于失效.
ECKPN[67]图像分类GCN 图池化
多模态
NSCGNN[19]文本分类GCN 胶囊网络
动态路由算法
Hybrid GNN[69]图像分类GCN 原型网络
EGNN-prot[18]图像分类GCN 原型网络
文献[71]图像分类GCN 分布校准最优
传输(OPT)
), ArticleFig(id=1329456563510685923, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, language=CN, label=表5, caption=

基于GNN的小样本学习方法总结

, figureFileSmall=null, figureFileBig=null, tableContent=
方法模型名称应用主要技术优点缺点
基于节点特征GNN-FSL[43]图像分类GCN1. 节点特征易获取、易利用.
2. 扩展性强,通过丰富GNN模型使节 点特征在图上进行高效的传播.
3. 易理解模型设计容易,可解释性强.
1. 对节点之间相关性的探究不足.
2. 通常利用节点特征间的相似度构建邻接矩阵,易导致不适当的特征聚合.
3. 存在过拟合的风险.
TPN[44]图像分类GCN 转导学习
AGNN[48]图像分类GCN 注意力机制
AMGNN[21]疾病诊断GCN 多模态融合
DMH-FSL[51]图像分类超图卷积
文献[50]图像分类GCN 幻觉网络
基于边缘特征EGNN[53]图像分类GCN1. 在节点特征的基础上考虑到对节点间的关系进行深入挖掘.
2. 精心设计的边缘特征既可以控制节点信息聚合的程度也可以建模节点间丰富的关系,效果提升明显.
3. 设计过的边特征可以缓解图上的过拟合问题.
1. 边缘特征的表示形式不够丰富.
2. 缺乏明确的初始边缘特征建模标准.
3. 设计合适的边缘特征更新网络较为困难. 算法复杂度较高.
文献[54]图像分类GCN
MDEGNN[55]图像分类GCN
HOSP-GNN[80]图像分类GCN 高阶关系
MCGN[56]图像分类GCN 混合高斯分布
DPGN[58]图像分类GCN 分布图
DPGNN[60]视频分类GCN 图池化
Fuzzy-GNN[62]图像分类GCN 模糊数学
基于节点对特征Frog-GNN[17]文本分类GCN1. 对节点关系进行显示建模,由此丰富了关系的表现形式.
2. 直接将关系在图上进行传播,能够学习到节点间的深入的关系.
3. 能够充分挖掘样本间相关性.
1. 关系节点的构建方法复杂.
2. 多种形式的信息在图上传播可能导致节点间过平滑.
3. 可解释性较差.
4. 算法复杂度较高.
5. 模型扩展性较差.
TPRN-T[64]图像分类GCN
TPRN-D[65]图像分类GCN 解耦学习
基于类级特征Hierachical GNN[66]图像分类GCN 图池化1. 结合类级知识能够为查询样本匹
配更高可信度的类级知识,对于5-shot的分类任务这类方法平均性能
提升明显.
2. 模型扩展性较强,改进方式较多.
1. 对于1-shot的分类任务,类级知识对模型性能的提升相当于失效.
ECKPN[67]图像分类GCN 图池化
多模态
NSCGNN[19]文本分类GCN 胶囊网络
动态路由算法
Hybrid GNN[69]图像分类GCN 原型网络
EGNN-prot[18]图像分类GCN 原型网络
文献[71]图像分类GCN 分布校准最优
传输(OPT)
)], attaches=null, journal=Journal(id=1324759606947315908, delFlag=0, nameCn=计算机研究与发展, nameEn=Journal of Computer Research and Development, nameHistory1=null, nameHistory2=null, issn=1000-1239, eissn=null, cn=11-1777/TP, coden=null, periodic=0, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=k2l8zgqwstuxxevFX2qxVA==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Computer Research and Development, journalRemark=null, publicationField=null, createdTime=1793776903141, updatedTime=1793850698612, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=J, firstLetterEn=J, subjectCode=Electronic Information, subjectName=null, subjectCodeEn=Electronic Information, subjectNameEn=null, picCn=k2l8zgqwstuxxevFX2qxVA==, picEn=r6k0Ry+RD+c/VXj5QarmDw==, jcr=null, cjcr=null, exts=[JournalExt(id=1325069127710286688, language=CN, name=计算机研究与发展, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1793850698640, updatedTime=1793850698640, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1325069127777395553, language=EN, name=Journal of Computer Research and Development, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1793850698656, updatedTime=1793850698656, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1324759914146529478, websiteList=[Website(id=1325070300139892771, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN, language=CN, createTime=1793850978169, createBy=18614031015, updateTime=1793851040259, updateBy=18614031015, name=计算机研究与发展-中文, tplId=1146099689490845704, title=计算机研究与发展, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070893751349360, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=articleTextType, value=kx, createTime=1793851119697, updateTime=1793851119697, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893730377837, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=banner, value=null, createTime=1793851119692, updateTime=1793851119692, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893772320883, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=grayFlag, value=0, createTime=1793851119702, updateTime=1793851119702, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893721989228, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851119690, updateTime=1793851119690, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893784903797, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=minRunFlag, value=0, createTime=1793851119705, updateTime=1793851119705, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893742960751, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic, createTime=1793851119695, updateTime=1793851119695, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893780709492, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=silenceFlag, value=0, createTime=1793851119704, updateTime=1793851119704, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893738766446, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1793851119694, updateTime=1793851119694, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893755543665, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeColor, value=null, createTime=1793851119698, updateTime=1793851119698, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893763932274, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeStyle, value=null, createTime=1793851119700, updateTime=1793851119700, creator=18614031015, updator=18614031015)]), Website(id=1325070300580294698, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN, language=EN, createTime=1793850978274, createBy=18614031015, updateTime=1793851036638, updateBy=18614031015, name=计算机研究与发展-英文, tplId=1146101810881728533, title=Journal of Computer Research and Development, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070925288321148, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=articleTextType, value=kx, createTime=1793851127216, updateTime=1793851127216, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925263155321, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=banner, value=null, createTime=1793851127210, updateTime=1793851127210, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925309292671, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=grayFlag, value=0, createTime=1793851127221, updateTime=1793851127221, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925254766712, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851127208, updateTime=1793851127208, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925326069889, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=minRunFlag, value=0, createTime=1793851127225, updateTime=1793851127225, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925279932539, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic, createTime=1793851127214, updateTime=1793851127214, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925317681280, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=silenceFlag, value=0, createTime=1793851127223, updateTime=1793851127223, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925271543930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1793851127212, updateTime=1793851127212, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925292515453, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeColor, value=null, createTime=1793851127217, updateTime=1793851127217, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925300904062, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeStyle, value=null, createTime=1793851127219, updateTime=1793851127219, creator=18614031015, updator=18614031015)])], journalTitle=计算机研究与发展, weixinUrl=null, journalUrl=https://crad.ict.ac.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Computer Research and Development, journalPhotoCn=k2l8zgqwstuxxevFX2qxVA==, journalPhotoEn=r6k0Ry+RD+c/VXj5QarmDw==, journalFirstLetter=J, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202220933, detailUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/10.7544/issn1000-1239.202220933, pdfUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/PDF/10.7544/issn1000-1239.202220933, pdfUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/PDF/10.7544/issn1000-1239.202220933, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1711900800000, fullTextJson=null, articleText=null, reference=null)
收藏切换
基于图神经网络的小样本学习方法研究进展
杨洁祎 , 董一鸿 * , 钱江波
计算机研究与发展 | 人工智能 2024,61(4): 856-876
收起
收藏切换
计算机研究与发展 |人工智能 2024 , 61 (4) : 856 -876
基于图神经网络的小样本学习方法研究进展
全屏
[Author(id=1329456559903584437, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=17858878879@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329456559987470518, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456559903584437, language=EN, stringName=Jieyi Yang, firstName=Jieyi, middleName=null, lastName=Yang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"content":"

born in 1999. Master candidate. Student member of CCF. Her main research interests include few-shot learning, graph neural network, and machine learning

"}, bioImg=null, bioContent=

born in 1999. Master candidate. Student member of CCF. Her main research interests include few-shot learning, graph neural network, and machine learning

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329456560058773687, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456559903584437, language=CN, stringName=杨洁祎, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"img":"pkpkoM1HN11/2jXUEsg4qA==","content":"

1999年生. 硕士研究生. CCF学生会员. 主要研究方向为小样本学习、图神经网络、机器学习

"}, bioImg=pkpkoM1HN11/2jXUEsg4qA==, bioContent=

1999年生. 硕士研究生. CCF学生会员. 主要研究方向为小样本学习、图神经网络、机器学习

, aboutCorrespAuthor=null)}, companyList=null), Author(id=1329456560130076857, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=dongyihong@nbu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329456560260100282, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560130076857, language=EN, stringName=Yihong Dong, firstName=Yihong, middleName=null, lastName=Dong, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=null, bio={"content":"

born in 1969. PhD, professor, master supervisor. Member of CCF. His main research interests include big data, data mining, and artificial intelligence

"}, bioImg=null, bioContent=

born in 1969. PhD, professor, master supervisor. Member of CCF. His main research interests include big data, data mining, and artificial intelligence

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329456560339792059, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560130076857, language=CN, stringName=董一鸿, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=null, bio={"img":"PKUxcdppRxNm6kL0qg7ruA==","content":"

1969年生. 博士,教授,硕士生导师. CCF会员. 主要研究方向为大数据处理、数据挖掘、人工智能

"}, bioImg=PKUxcdppRxNm6kL0qg7ruA==, bioContent=

1969年生. 博士,教授,硕士生导师. CCF会员. 主要研究方向为大数据处理、数据挖掘、人工智能

, aboutCorrespAuthor=null)}, companyList=null), Author(id=1329456560440455357, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329456560524341438, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560440455357, language=EN, stringName=Jiangbo Qian, firstName=Jiangbo, middleName=null, lastName=Qian, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"content":"

born in 1974. PhD, professor, PhD supervisor. Senior member of CCF. His main research interests include machine learning, pattern recognition, and intelligent systems

"}, bioImg=null, bioContent=

born in 1974. PhD, professor, PhD supervisor. Senior member of CCF. His main research interests include machine learning, pattern recognition, and intelligent systems

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329456560599838911, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329456557160509589, authorId=1329456560440455357, language=CN, stringName=钱江波, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio={"img":"A6oQzcbvA095PSP+JHyWSQ==","content":"

1974年生. 博士,教授,博士生导师. CCF高级会员. 主要研究方向为机器学习、模式识别、智能系统

"}, bioImg=A6oQzcbvA095PSP+JHyWSQ==, bioContent=

1974年生. 博士,教授,博士生导师. CCF高级会员. 主要研究方向为机器学习、模式识别、智能系统

, aboutCorrespAuthor=null)}, companyList=null)]
杨洁祎 , 董一鸿* , 钱江波
作者信息
  • 1宁波大学信息科学与工程学院 浙江宁波 315211
  • 2浙江省移动网络应用技术重点实验室(宁波大学) 浙江宁波 315211
通讯作者:
董一鸿()
作者简介:

1999年生. 硕士研究生. CCF学生会员. 主要研究方向为小样本学习、图神经网络、机器学习

1969年生. 博士,教授,硕士生导师. CCF会员. 主要研究方向为大数据处理、数据挖掘、人工智能

1974年生. 博士,教授,博士生导师. CCF高级会员. 主要研究方向为机器学习、模式识别、智能系统

Research Progress of Few-Shot Learning Methods Based on Graph Neural Networks
Jieyi Yang , Yihong Dong* , Jiangbo Qian
Affiliations
  • 1Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, Zhejiang 315211
  • 2Key Laboratory of Mobile Network Application Technology of Zhejiang province(Ningbo University), Ningbo, Zhejiang 315211
  • born in 1999. Master candidate. Student member of CCF. Her main research interests include few-shot learning, graph neural network, and machine learning

    born in 1969. PhD, professor, master supervisor. Member of CCF. His main research interests include big data, data mining, and artificial intelligence

    born in 1974. PhD, professor, PhD supervisor. Senior member of CCF. His main research interests include machine learning, pattern recognition, and intelligent systems

出版时间: 2024-04-01 doi: 10.7544/issn1000-1239.202220933
文章导航
收藏切换

小样本学习(few-shot learning,FSL)旨在利用少量样本学习得到解决问题的模型,为解决应用场景中样本量少或标注样本少的问题. 图神经网络(graph neural network,GNN)由于其在许多应用中的卓越性能引起了极大的关注,许多学者开始尝试利用图神经网络进行小样本学习,基于图神经网络的方法在小样本领域取得了卓越的成绩. 目前与基于图神经网络的小样本学习方法相关的综述性研究较少,缺乏该类方法的划分体系与介绍性工作,因此系统地梳理了当前基于图神经网络的小样本学习的相关工作:概括了小样本学习的图神经网络方法的概念,根据模型的基本思想将其划分为基于节点特征、基于边特征、基于节点对特征和基于类级特征的4类方法,介绍了这4类方法的研究进展;总结了目前常用的小样本数据集和代表性模型在这些数据集上的实验结果,归纳各类方法主要的研究内容和优劣势;最后概述了基于图神经网络的小样本学习方法的应用和面临的挑战,并展望其未发展方向.

小样本学习  /  图神经网络  /  元学习  /  度量学习  /  迁移学习

Few-shot learning (FSL) aims to learn to get a problem-solving model using a small number of samples. Under the trend of training models with big data, deep learning has gained success in many fields, but realistic scenarios often lack sufficient samples or labeled samples. Therefore, FSL becomes a promising research direction at present. Graph neural networks (GNN) have attracted great attention due to their excellent performance in many applications. In view of this, many methods try to use GNN for FSL. Currently there are few review researches related to FSL methods based on GNN, and there is a lack of division system and introductory work on this type of methods. We systematically compose the current work related to FSL based on GNN. The work outlines the basis and concepts of graph methods for FSL, broadly classifies them into four categories of methods based on node-based feature, edge-based feature, node-pair-based feature and class-level-based feature according to the basic ideas of the models. The research progress of the four methods is introduced as well. Then the experimental results of the commonly used few-shot datasets and representative models on these datasets are summarized, as well as the advantages and disadvantages of each type of methods. Finally, current status and challenges of the graph methods for FSL are introduced, and their future directions are prospected.

few-shot learning  /  graph neural network  /  meta-learning  /  metric-learning  /  transfer-learning
杨洁祎, 董一鸿, 钱江波. 基于图神经网络的小样本学习方法研究进展. 计算机研究与发展, 2024 , 61 (4) : 856 -876 . DOI: 10.7544/issn1000-1239.202220933
Jieyi Yang, Yihong Dong, Jiangbo Qian. Research Progress of Few-Shot Learning Methods Based on Graph Neural Networks[J]. Journal of Computer Research and Development, 2024 , 61 (4) : 856 -876 . DOI: 10.7544/issn1000-1239.202220933
深度学习能够很好地实现复杂问题的学习,然而,深度学习需要大量带标注的训练数据[1]. 而在现实场景中,某些类别仅有少量标注数据. 与此相反,人类只需通过少量数据就能做到快速学习. 受到人类学习方式的启发[2],小样本学习(few-shot learning, FSL)[3-5]的概念被提出,即使在仅有少量数据或标记数据的情况下,也能学习到具有良好性能的模型.
小样本学习方法包括基于模型微调、基于数据增强和基于迁移学习3类方法[6]. 基于模型微调的主要策略是首先使用大规模数据集对模型进行预训练,然后针对只有少量训练样本的特定任务对模型进行微调. 然而该类模型存在过拟合[7]问题,即在实例不足的情况下无法建模新任务的数据分布. 基于数据增强的学习旨在应用数据增强[8]或数据混合[9]的技术来扩大训练集. 虽然数据增强的方法在有限的数据状态下缓解了过拟合问题,但在扩增数据的同时可能引入噪声数据,对模型效果产生消极影响. 基于迁移学习的方法是目前较为前沿的方法,可细分为基于度量学习、基于元学习(meta-learning, ML)和基于图神经网络(graph neural network, GNN)3类方法[5]. 基于度量学习的方法通过计算样本之间的距离,从而根据有标注样本确定未知样本类别. 但在样本数量较少的情况下,这类基于度量的思想使模型效果的提升遇到瓶颈. 元学习方法的出现使FSL更靠近人类的学习模式,以任务为训练单位,使模型开始具备学习的能力. 然而纯粹的元学习方法复杂度较高,且忽视了对FSL问题中样本间关系的探索. 图神经网络[10]能够通过消息传递机制迭代地执行来自邻居的特征聚合,因此可以表达数据实例间的复杂交互. 由于小样本学习算法已被证明需要充分利用支持集和查询集[11-13]之间的关系,因此GNN具有解决FSL问题的巨大潜力. GNN在节点分类、图分类和链接预测等应用中表现出良好的效果和可解释性[14],并成为一种广泛应用的图分析方法[15-16]. 现今,使用GNN来解决FSL问题已逐渐成为国内外的研究热点. 基于GNN的FSL方法遵循元学习的情景训练策略,将任务视为训练样本,任务中的数据实例能构成一个无向无环图,每个样本被视为图中节点. 如何准确有效地捕获样本间的关系是解决FSL问题的关键. 而GNN能够通过在图上进行消息传递来聚合更新信息,在仅有少量实例的任务中度量样本间距离并探索样本间的关系. 因此GNN被视为一种度量工具在FSL问题中的广泛应用.
基于GNN的FSL方法融合了度量学习和元学习的优点,相较于基于元学习的方法,GNN能够精准地捕捉样本的类间和类内关系. 相较于基于度量学习的方法,GNN模型能够挖掘节点间的潜在关系,从而提升模型性能. 基于GNN的FSL方法在FSL图像分类、文本分类[17-19]、视频分类[20]、疾病诊断[21]和疾病的相似性[22]等研究领域取得了一定的进展. 基于GNN的FSL方法旨在利用GNN能够在图上传播信息的特点,构建不同的特征信息并在图上传播,利用更新后的特征信息完成下游任务.
研究人员对FSL和GNN的相关工作分别进行了总结. 文献[6, 23]从不同的角度出发,分别构建了不同的FSL方法划分体系. 这些工作系统地归纳了各类FSL方法的基本思想,并对各类FSL方法中的经典模型进行了具体分析. 文献[14, 24]概括了GNN的不同框架,对各类GNN方法进行了详细介绍,并说明了GNN在不同实际场景中的应用. GNN作为近几年的热点模型被成功地应用于FSL问题,基于GNN的FSL方法较为新颖,且近2年出现了大量基于GNN的FSL方法. 然而以往的综述工作中缺乏对该类FSL方法的具体讨论与研究,因此本文对基于GNN的一系列FSL方法进行深入的研究,对不同方法进行系统分类,并详细地阐述了划分依据与各类方法的整体思想,主要贡献有3点:
1) 首次对基于GNN的FSL方法进行归纳总结,并给出了基于GNN的FSL范式,该范式明确地阐明了基于GNN的FSL目标是什么以及如何解决该类问题. 这一定义有助于确定基于GNN的FSL领域的研究目标.
2) 整理归纳了近几年涌现的基于GNN的FSL方法并对其进行系统地分类,首次给出了基于GNN的FSL的分类方法. 总结了不同类别方法的特点并深入地探讨对比了各类方法的优劣势.
3) 基于GNN的FSL方法的弱点,提出了基于GNN的FSL面临的挑战,并给出了其未来可能的探索方向,为解决FSL问题提供新的解决思路.
FSL领域的研究试图从有限的标记数据中提取信息,并对未知的数据做出准确的预测. 模型微调的方法是FSL方法中最直观且最有效的方法. 文献[25]提供了一种基于模型微调的方法,该方法通过在再训练阶段中采用更低的学习率,以及在微调过程中采用自适应的梯度优化器来实现优化的目的. 为了在微调阶段学习到更加适应目标数据域的知识,文献[26]将未标记的查询样本的熵作为正则项加入损失函数,通过利用无标签数据的信息来缓解可用样本量不充足的问题. 这一类基于微调的模型较为简单,但由于在实际场景中目标数据集与源数据集分布相差较大,采用模型微调的方法可能导致模型在目标数据集上发生过拟合.
基于数据增强方法借助辅助数据或辅助信息,对既有的小样本数据集进行扩充,以克服FSL中样本量不充足的困难. 文献[27]提出的数据增强方法结合了元学习的训练策略,该方法通过将现有图像的一些属性和特征变化迁移到新样本上来生成新数据图像,以达到扩充样本量的目的,并通过端到端联合训练的模式生成模型和分类算法. 文献[28]提出了一种特征迁移网络,用于描述物体姿态变化引起的运动轨迹变化. 该网络通过对数据样本的特征进行增强,即用物体的运动轨迹特征来扩充原有的物体姿态特征,以此达到数据增强的目的. 文献[29]提出了一种新的特征增强网络(feature augmentation network, FAN)用于对小样本的无约束掌纹图像识别,在缺乏其他外部信息的前提下,FAN可用于减少因无约束信息获取而导致的图像改变,其目的是解决由于图像噪声带来的模型效果不佳的问题. 此外FAN从少数支持样本中增强它们的特征表示,即通过对数据特征的强化来达到数据增强的目的. 基于数据增强的方法在一定程度上解决了仅用少量样本训练深度学习模型所产生的过拟合现象,但由于数据增强方法生成的数据为虚拟数据,生成的数据越多引入的噪声数据可能越多,因此模型的性能不随数据量正比增强,导致模型性能在得到了一定的提升后就停留在瓶颈期.
基于迁移学习的方法是目前较为前沿的方法. 其中基于度量方法的FSL模型,例如,匹配网络[30]结合了注意力机制和记忆机制,并利用情景训练的训练模式,将未标记的查询样本标签与各个类中的少量带标记支持样本的标签进行匹配,根据匹配结果从而确定未标记样本的标签. 原型网络[31]则学习了一个判别度量空间,通过计算查询样本和每个类的原型表示之间的距离来进行分类,每个类的原型表示通过计算该类支持样本的均值得到. 文献[32]利用协方差度量网络构造了一个嵌入的局部协方差表示来提取每个类的2阶统计信息,并定义了一个协方差矩阵来衡量查询样本和支持样本之间分布的一致性. MSML[33]模型提出了一种用于层次度量学习的多尺度关系生成网络,该模型中引入了特征金字塔结构,旨在结合高层语义特征和低层视觉特征,以及提取多尺度特征用于映射生成样本间的多尺度关系. 元学习[34]是一种模仿生物利用先前已有的知识,迅速学习新的未见事物能力的一种学习定式. 元学习的目标是使模型能够利用学习到的知识,快速适应未见过的新任务,这与实现人工智能的目标相契合[35]. 元学习作为机器学习领域中一个前沿的研究框架,被广泛地应用到小样本学习的任务中. Meta-learner LSTM[36]的随机梯度下降更新规则可解释为一个带有可训练参数的门控递归模型学习模型参数的更新规则. MAML[37]则寻求通过2阶梯度下降来进行适当的参数初始化,从而使模型在经过几步梯度下降后能够获得更好的泛化能力. 不同于深度学习模型的目标旨在通过梯度下降法得到最优的模型参数,MAML模型旨在得到能够从少量的数据中进行几步更新就可以得到具有较好分类效果的初始化参数. Meta-SGD[38]在MAML的基础上更进一步提出在一步内学习权重初始化、梯度更新方向和学习速率. 基于记忆的方法提取在元训练阶段获得的有价值的知识以协助元测试. 表1总结了各类FSL方法的优劣势,其中基于GNN的FSL方法是目前较为前沿且性能最好的一类,然而由于基于GNN的FSL是近几年才兴起的,其需要改进和发展的方面还有很多.
为了更好地理解基于GNN的FSL模型框架,本节首先简要介绍了相关的基本概念和专业术语,并给出了基于GNN的FSL方法的说明范式和一般流程. 为行文方便,对本文使用的符号进行说明,如表2所示.
基于GNN的FSL遵循元学习的情景训练策略[39-40]. 情景训练的核心思想是构造一系列FSL任务用于训练,训练目的是使模型能够快速适应并解决类似的新FSL任务. 训练任务用于训练FSL模型,相当于深度学习中的训练样本;测试任务则用于评估模型性能,即评估模型在新任务上的泛化能力.
情景训练将数据集划分为训练集$ {D_{{\text{base}}}} = \{ ({x_i},{y_i})_{i = 1}^{{m_{{\text{base}}}}},{y_i} \in {C_{{\text{base}}}}\} $和测试集$ {D_{{\text{novel}}}} = \{ ({x_i},{y_i})_{i = 1}^{{m_{{\text{novel}}}}}, {y_i} \in {C_{{\text{novel}}}}\} $,其中mbase和mnovel为Dbase和Dnovel中的样本数,Cbase和Cnovel分别是Dbase和Dnovel对应的类集合,且$ {C_{{\text{base}}}} \cap {C_{{\text{novel}}}} =\varnothing $. FSL旨在Dbase上学习一个模型,该模型能够很好地推广到未见过的测试集Dnovel. 情景训练分别在Dbase和Dnovel中采样一系列的任务作为FSL框架中的训练样本和测试样本. 其中每个分类任务由支持集和查询集构成,支持集$ S = \{ ({x_i},{y_i})\} _{i = 1}^{N \times K} $由N个类中抽取的K个样本( N-way K-shot设置)构成,这些样本作为标记实例. 从同样的N个类中抽取另外的H个样本作为查询集$ Q = \{ ({x_i},{y_i})\} _{i = N \times K + 1}^{N \times K + N \times H} $,这些样本为待预测的未标记样本.
设$ p(T) $为任务T的分布,情景训练从$ p(T) $中采样一系列训练任务$ {T_{{\text{train}}}}\sim p(T) $作为训练阶段样本,对于训练任务$ {T_{{\text{train}}}} $在特定于该任务的N个类的部分数据$ {D_{{T_{{\text{train}}}}}} $上操作,$ {C_{{T_{{\text{train}}}}}} = \{ {C_1},{C_2},…,{C_N}\} \in {C_{{\text{base}}}} $为这N个类的集合,训练任务中的支持集和查询集均来自$ {D_{{T_{{\text{train}}}}}} $. 每个任务中的数据采样如图1所示.
基于GNN的FSL通常将GNN作为分类器或特征适应模块,调整特征嵌入后再分类,GNN模型在整体训练过程中相当于元模型,所有任务的GNN模型共享元参数.
基于GNN的FSL方法通常将每个任务中的支持样本和查询样本进行全连接构图,构建得到的图输入到元GNN模块中. 其目的是基于节点分类的基本思想,利用GNN对查询样本进行分类,即可得到每个任务中查询样本的预测结果. 最终通过计算预测结果与观测结果的分类损失来更新元GNN模型中的模型参数$ \theta $. 如图2所示为基于GNN的FSL方法中,GNN模型的训练思想,其模型参数$ \theta $学习目标如式(1)所示.
$ \theta =\mathop {\arg \min }\limits_{\theta}{\sum_{T_{\mathrm{train}}\sim p(T)}} {{L_\theta }({T_{{\text{train}}}})} . $
在训练过程中,选取使得所有任务的总损失最小的模型参数作为元GNN模型的最优参数. 其中$ {L_\theta }({T_{{\text{train}}}}) $为模型在任务$ {T_{{\text{train}}}} $上的损失函数,其计算过程如式(2)所示.
$ \begin{gathered}{L_\theta }\left( {{T_{{\text{train}}}}} \right) = \displaystyle \sum\limits_{(x_{{T_{{\text{train}}}}}^i,y_{{T_{{\text{train}}}}}^i) \in D_{{T_{{\text{train}}}}}^Q} {{L_\theta }\left( {y_{{T_{{\text{train}}}}}^i,\overline y _{{T_{{\text{train}}}}}^i} \right)} . \\ \end{gathered} $
通过计算训练任务中查询样本的预测标签与观测标签的交叉熵损失作为该训练任务的分类损失,其中$ D_{{T_{{\text{train}}}}}^Q $为任务Ttrain中查询样本的集合,$ (x_{{T_{{\text{train}}}}}^i,y_{{T_{{\text{train}}}}}^i) $对应集合中某一查询样本,$ \overline y _{{T_{{\text{train}}}}}^i $和$ y_{{T_{{\text{train}}}}}^i $分别表示该样本的预测标签和实际观测标签. 基于GNN的FSL方法要求在训练任务上训练得到的模型能够泛化到测试任务$ {T_{{\text{test}}}}\sim p(T) $上. 测试任务中的类${C_{{T_{{\text{test}}}}}} = \{ {C_1}, {C_2},…, {C_N}\} \in {C_{{\text{novel}}}}$取自与训练任务类集合${C_{{\text{base}}}}$不相交的新类. 将训练任务中的元知识迁移到测试任务上,并在测试任务上评估模型性能. 与深度学习训练的“样本”概念类似,情景训练策略采用“任务”作为训练的基本单位.
在小样本学习中,GNN作为一个强大的后处理工具,能够获取鲁棒的特征表示.
假设$ {\boldsymbol{F}} = ({{\boldsymbol{f}}_1},{{\boldsymbol{f}}_2},…,{{\boldsymbol{f}}_{N \times K + N \times H}}) \in {\mathbb{R}^{(N \times K + N \times H) \times p}} $是分类任务$ {T_i} $中支持和查询样本经卷积神经网络(convolutional neuaral network,CNN)[41]提取的特征矩阵,其中p为特征维度,$ {{\boldsymbol{f}}_j}(j = 1,2,…,N \times K + N \times H) $则为任务$ {T_i} $中的某一实例j的特征向量. 给定GNN的初始特征输入$ {{\boldsymbol{F}}^0} = {\boldsymbol{F}} $及初始邻接矩阵输入$ {{\boldsymbol{A}}^0} = {\boldsymbol{A}} $,GNN在隐藏层中按层级传播如式(3)所示.
$ {{\boldsymbol{F}}^{l + 1}} = \sigma ({{\boldsymbol{D}}^{ - 1/2}}{{\boldsymbol{A}}^l}{\boldsymbol{DF}}) \text{,} $
其中l =0,1,…, L−1,$ \boldsymbol{D}^{ }=\text{diag}(d_1,d_2,\dots,d_n) $为视度矩阵,由$ {d_i} =\displaystyle \sum\limits_{j = 1}^n {a_{i,j}^l} $计算得到,$ \sigma $()通常为激活函数或可训练的特征转换函数.
基于GNN的FSL方法通过利用GNN挖掘图上更深入的结构关系来获取判别性的样本表示及更精确的样本间关系,预测查询样本的标签. 其预测整体过程如图3所示.
经特征提取器提取得到的样本特征通过构图模块后输入到GNN中,节点、边等图中特征通过特征聚合在图上传播,以此更新图中信息. 而后,基于更新后的特征信息推理查询样本的标签. 查询样本的最终预测结果如式(4)所示.
${ \bar{\boldsymbol{Y}} _Q} = M({{\boldsymbol{Y}}_S},{{\boldsymbol{F}}_Q},{{\boldsymbol{F}}_S}), $
其中M为基于GNN的模型,$ {{\boldsymbol{F}}_S} $和$ {{\boldsymbol{F}}_Q} $分别为支持集和查询集的样本特征矩阵,$ {{\boldsymbol{Y}}_S} $和$ {{ \bar{\boldsymbol{Y}} _Q}}$分别为支持集中样本的观测标签矩阵和查询集中样本的预测标签矩阵.
GNN能够自然地适用于FSL,GNN能够通过在有限实例构造的图上进行消息传递来聚合更新信息,探索样本间的关系,并学习得到具有代表性的节点特征表示.GNN的本质是聚合邻居信息[42],对于图中任意节点,其特征每更新一次,就聚合了更高一阶的邻居节点信息. 基于GNN的FSL方法利用GNN的这些特点能够学习具有判别性的节点表示以及探索图中查询节点和支持节点间丰富的关系.
根据模型利用图上不同粒度级别的特征进行预测,将基于GNN的FSL问题分为基于节点特征的方法、基于边缘特征的方法、基于节点对特征的方法和基于类级特征这4类方法.
基于节点特征的方法通常将卷积网络提取到的样本特征与其1维标签向量进行拼接得到节点特征向量,全连接构图后计算其邻接矩阵,输入到GNN模型,在节点特征更新的过程中将支持样本的标签信息传递给查询样本,最终通过查询节点特征确定查询样本类别.
GNN-FSL模型[43]首次提出利用GNN模型来解决FSL问题. 模型中的节点嵌入表示$ {\boldsymbol{n}}_i^{(0)} = (\varphi ({x_i}),h({y_i})) $,其中$ \varphi ({x_i}) $为卷积网络提取的特征向量,$ h({y_i}) $为节点的one-hot标签向量. 将支持节点与查询节点进行全连接构图后,第k层网络中节点i与节点j之间的边由式(5)构建得到:
$ {\varphi _{\tilde \theta}}({\boldsymbol{n}}_i^{(k)},{\boldsymbol{n}}_j^{(k)}) = ML{P_{\tilde \theta }}({\text{abs}}({\boldsymbol{n}}_i^{(k)} - {\boldsymbol{n}}_j^{(k)})), $
其中$ \varphi $是由神经网络参数化的对称函数,$\varphi $可从当前节点的特征表示中学习图的邻接矩阵,从在2个节点向量的绝对差异后堆叠多层感知器得到,$ \tilde\theta $为可学习参数. 模型通过GNN更新节点嵌入向量,进一步更新边向量,最终利用学习到的查询节点特征来评估查询样本所属的类别. 其模型框架图如图4所示. GNN-FSL模型将标签信息编码为节点特征的一部分,并利用图的消息传递机制在图上进行节点特征的传播. 将支持样本的标签信息通过信息传递的形式编码进与其实例特征相似的查询节点中,因此最终得到的查询节点特征中隐含了其可能的类别信息.
与标签传播的思想一致,TPN[44]模型直接利用图卷积(graph convolution network, GCN)显式地进行标签信息的传播,并在图卷积过程中融合标签信息. TPN首次在基于GCN的FSL学习中明确地建模转导推理[45]. 在归纳推理[46]中,模型对每个查询样本独立预测,利用支持样本标签推理查询样本标签. 而在转导推理中可将查询样本和支持样本一起用于推理,能够同时对整个查询集进行预测,而不是独立预测单个查询样本.
GNN-FSL和TPN这2种基于GNN的FSL模型证明了相较于传统的FSL方法,GNN在解决FSL问题上的优越性能. 但随着GNN层的叠加,模型不可避免地出现了节点过平滑和过拟合的问题[47]. AGNN(attentive GNN)[48]模型引入图注意力(graph attentive, GAT)[49]机制来缓解GNN中的节点过平滑问题. AGNN首先通过节点自注意力机制更新节点的样本特征和标签向量. 其次通过引入邻居节点注意力机制来限制参与聚合的邻居节点个数,通过稀疏性约束构建邻接矩阵,避免由于节点过度混合而造成节点特征趋于一致的问题. 使用层注意力机制保留并融合节点特征的早期“记忆信息”. AGNN的三重注意力机制有效地缓解了GNN的过平滑和过拟合问题.
基于节点特征的FSL问题通常只提取一个模态的信息作为节点特征. AMGNN(auto-metric GNN)[21]模型被用于阿尔兹海默症的诊断,该模型有效地融合多模态特征作为节点特征. AMGNN将样本的风险因子、认知评分等先验知识模态和磁共振成像特征模态进行拼接得到节点特征,并用样本的先验知识计算边缘概率矩阵来约束磁共振成像特征计算得到的邻接矩阵,从而融合样本的多模态信息,将更新后的节点特征输入到softmax层和归一化层后得到节点的最终标签. AMGNN在保证良好分类性能的同时提高了灵活性,促进了GNN向小样本的疾病诊断方向的发展.AMGNN将多模态GNN应用于FSL问题,文献[50]则提出利用多尺度GNN来对任务建模,该方法首先利用特征提取网络提取3种不同尺度的图像特征,将不同尺度的特征输入对应尺度的GNN网络中进行特征聚合,最终得到多尺度的预测结果,并用原始特征尺度的预测结果来限制其余2种尺度的预测结果,从而确保多尺度GNN预测结果的准确性.
文献[21, 43-45, 48-49]通过拼接样本特征和标签特征的方法融合了同一节点的2种视角信息作为节点特征,继而借助GNN探索样本节点之间的关系. 双模态超图小样本学习(dual-modal hypergraph FSL, DMH-FSL) [51]在此基础上,通过引入超图卷积[52]的思想,来模拟样本之间的高阶关系. 超图提供了一种灵活而自然的建模工具来表示样本间超越成对的关系. GCN中的每条边只能和2个顶点连接;而对于超图来说,其超边可以和任意数量的顶点连接. 每个边所包含的顶点数都是相同且为k时,该图被称为k阶超图. DMH-FSL首先使用特征模态超图构造了具有样本特征的关联矩阵,标签模态超图构造了带有样本标签的关联矩阵. 通过特征和标签2种模态来聚合节点特征嵌入,并利用超图卷积对不同模态的样本进行灵活聚合.
基于节点特征的方法将节点特征在图上传播,将支持节点的信息传递给查询节点,从而确定查询节点的类别. 该类方法通过拓展原始的GNN网络,使节点特征信息在图上高效传播,获取具有判别性的节点嵌入表示,利用最终学习到的节点嵌入确定查询节点类别.
不同于利用节点特征的方法,利用边缘特征的方法从另一方面进行考虑,对图中的边缘特征进行建模,利用边缘特征表示节点间的相关性. 相比之下,基于边缘特征的模型能够考虑样本间更丰富的关系与建模更灵活的信息交互形式. 这类方法通常仍采用紧密连接的图结构,将实例特征视为节点特征,通过不同的方法构建边缘初始特征及边缘更新网络,最终利用学习到的边特征表示样本间的相关性,从而对未标记样本进行分类.
文献[53]提出的EGNN模型由多个层组成,其中每一层都由1个节点更新块和1个边缘更新块组成. EGNN首先初始化边缘特征,使用2维的边特征向量,其中第1维表示相连的2个节点属于同一类的概率,即节点间相似性,第2维表示相连的2个节点不属于同一类的概率即节点间差异性. 如式(6)所示:
$ {\boldsymbol{e}}_{ij}^0 = \left\{\begin{aligned}&{(1,0),\qquad\quad{\text{if }}\;{y_i} = {y_j}\;{\text{and }}\;i,j \leqslant N \times K,{\text{ }}} \\& {(0,1),\qquad\quad{\text{if }}\;{y_i} \ne {y_j}\;{\text{and }}\;i,j \leqslant N \times K,} \\& {(0.5,0.5),\quad\;\;{\text{其他情况.}}} \end{aligned} \right. $
该边向量作为注意力权重聚合对应的邻居节点特征,得到更新后的节点向量. 使用可训练函数度量节点间的相似度和不相似度,分别作为自注意力系数更新边缘特征中第1维和第2维的值,以此更新边缘特征. 查询样本标签的预测则利用其最终的边缘特征与边相连的支持节点来确定. 图5为EGNN的整体框架图,其中边缘特征的强度用正方形中的颜色深度表示. 尽管每条边都有2维特征,但为了简单起见,图5中只描述了第1个维度. EGNN模型在小样本图像分类数据集上的优越性能证明了该模型的有效性. EGNN-FSL[54]模型将EGNN模型应用于高光谱图像(hyperspectral image, HSI)分类问题中,对HSI中像素之间的关联进行明确地考虑和探索,特别是像素之间的类内关联和类间关联.EGNN模型在进行节点和边缘特征更新时,同时将节点信息和节点间的关系在图上进行消息传递. 模型旨在最大化类内相似性和类间差异性以提高模型的预测性能. 在EGNN模型的基础上,MDEGNN(multi-dimensional EGNN)[55]使用多维特征作为边缘特征,所有的边缘特征被初始化成一个p维向量,若2个节点属于同一类,则它们之间的边特征是1个全1向量,若2个节点属于不同类,则它们之间的边特征是全0维向量,边缘特征初始化如式(7)所示.
$ {\boldsymbol{e}}_{ij}^0 = \left\{\begin{split} & (1,…,1) \in {\mathbb{R}^{1 \times p}},\qquad{\text{if }}\;{y_i} = {y_j}\;{\text{且 }} \;i,j \leqslant N \times K, \\& (0,…,0) \in {\mathbb{R}^{1 \times p}},\qquad{\text{if }}\;{y_i} \ne {y_j}\;{\text{且 }} \;i,j \leqslant N \times K, \\& (\frac{1}{{N \times K}},…,\frac{1}{{N \times K}}) \in {\mathbb{R}^{1 \times p}}{\text{,}} \qquad{\text{其他情况}}{\text{. }} \\ \end{split} \right.$
与EGNN类似,MDEGNN将多维边缘特征作为权值来聚合邻域节点信息. 边缘更新网络则利用更新后的节点特征生成多维相似度注意矩阵对多维边缘特征进行更新. 最后利用聚合网络对不同维数的边缘特征进行加权,得到最终的节点间关系表示. 基于边缘特征的方法通常基于节点的标签信息进行边缘特征的构建. 相较于基于图像特征建模的边,基于标签信息建模的边对异常样本点不敏感,且不会由于图像较为相似使得边缘特征较大而导致不适当的特征聚合. 然而标签信息忽略了特定于样本的特征信息,因此在标签的更新中通常会融合节点图像特征信息. 文献[56]提出一个将条件随机场(conditional random field, CRF)[57]和GCN结合的模型MCGN(mutual CRF-GNN),该模型利用样本标签信息和特征信息来构建边缘特征. MCGN构建了一个以标签和特征为条件的CRF,支持数据的标签和特征被 RF以有概率的方式推断GNN亲和力. 由此利用CRF实现了特征信息和标签信息的融合,即在特征空间利用标签特征相似度进行聚合. MCGN训练时同时关注CRF和GNN的输出,从而共同促进CRF和GNN.
由于节点标签和节点特征是最容易获得的节点信息,因此很容易度量得到节点间关系. 然而标签和节点特征信息都只关注于实例级信息而忽略了全局性信息,分布传播图网络(distribution propagation graph network, DPGN)[58]模型则通过引入节点的1-vs-n分布关系来构建节点关系,利用全局观测信息和单个实例信息来度量节点间相似度.DPGN提出了一个对偶图的架构:一个点图和一个分布图. 点图通过在实例上收集1-vs-n关系来生成分布图,而分布图通过传递每对例子之间的分布关系来细化点图. 通过点图和分布图交替更新相互传递信息的形式将节点特征和节点分布特征编码到边缘特征中,最终通过点图最后一层中相应的边缘特征及其相连的支持样本来确定与其相连查询样本的标签. 与DPGN利用分布特征的思想类似,MDGN模型[59]同时利用节点特征与节点分布信息对边缘特征进行建模. 不同的是MDGN通过混合高斯模型和期望最大算法来估算节点的分布特征,边缘特征由节点的分布特征相似度和实例特征相似度输入到多层感知器网络中得到. 大多数基于边缘特征的模型将支持节点和查询节点以全连接的方式关联,通过迭代更新边缘特征得到查询节点和支持节点属于同一类的概率作为边缘预测结果. 但事实上对于某查询节点,大部分支持节点与其不同类,这些支持节点传递的信息可能会引入噪声对其造成负面影响. 因此双池化图神经网络(dual pooling graph neural network,DPGNN)[60]设计了一种边缘池化模块,该模块能自适应地消除样本间的负面关系,模型能够成功地应用于小样本的视频分类任务. 边缘池化模块首先采用可训练的投影矩阵得到每条边的池化分数,对于支持节点$ {v_i} $和查询节点$ {v_q} $得到2节点间边的池化分数为$ {s_{i,q}} $. 为保留重要的边缘特征,基于池化分数来设计掩码,该掩码定义如式(8)所示.
$ {m}_{i,q}=\left\{\begin{array}{l}1\text{,}\qquad\text{if }\;top\_rank\text{(}{s}_{i,q}\text{)}\le k'\text{,}\\ 0\text{,}\qquad\text{其他情况},\end{array} \right.$
其中$ k^{'} $为保留的边缘数,即保留池化分数最大的$ k^{'} $条边. 将掩码$ {m_{i,q}} $和对应边相乘,以保留查询节点到与其相关的支持节点的边.
相较于基于节点特征的图网络,基于边缘特征的方法能够挖掘样本之间的相似性和差异性,将图驱动成一个更易于分类的结构. 然而这些模型缺乏一种关系归纳偏差,它决定了实体如何相互作用. 在FSL的GNN方法中,节点之间的关系是一个模糊而主观的概念,因此一些模型通常使用全连接的方式进行构图. 受模糊理论[61]启发,模糊图神经网络(fuzzy GNN,FGNN)[62]能够在每个情景中获得优越的关系归纳偏差,被用于FSL. FGNN首先利用节点特征相似度构建了边缘特征,并设计了一个模糊隶属度函数为节点分类实现了更精确的边特征建模. 隶属度函数由边缘特征来定义,其中$ \delta $>>$ \eta $,如式(9)所示.
$ u({e_{ij}}) = \left\{ {\begin{array}{*{20}{l}}{1,} \\ {{f_u}({e_{ij}}),} \\ {0,} \end{array}\begin{array}{*{20}{l}}{{\text{if }}\;{e_{ij}} > \delta , {\text{ }}} \\ {{\text{if }}\;\delta \geqslant {e_{ij}} \geqslant \eta , } \\ {{\text{其他情况}}. } \end{array}} \right. $
基于边缘特征的方法在考虑节点特征的基础上对边缘特征进行建模,通过这种方式能够更好地表示样本间关系. 边缘特征既可以表示节点间的相关性,也可以用于控制相邻节点的聚合程度. 基于边缘特征的方法通常在标签空间建模边缘特征,部分方法还结合了分布特征来丰富边缘特征的表示. 这类方法的重点在于边特征的初始化和更新网络的设计,如何有效准确地将样本间关系建模为边特征是这类方法的难点.
基于节点对特征的方法与基于边缘特征的方法旨在挖掘样本之间的关系. 但相较于利用边特征建模节点关系的方式,基于节点对特征的方法对样本间关系进行显式建模. 这类方法将节点构成节点对,节点对作为关系图中单个节点,节点对信息通过GNN在关系图上进行传播,节点对通常称为“关系节点”.
青蛙GNN(Frog-GNN)[17]是用于小样本文本分类任务的一种方法,Frog-GNN为避免信息丢失,保留了所有的查询支持对. 首先,从预训练的语言模型中提取文本表示,生成实例特征, 利用句子匹配任务的经验[63]提取成对实例特征,随后用对特征和实例特征初始化图中节点$ {\boldsymbol{v}}_i^{(0)} = \{ {\boldsymbol{p}}_i^{(0)},{\boldsymbol{s}}_i^{(0)},{\boldsymbol{q}}_i^{(0)}\} $,其中$ {\boldsymbol{p}}_i^0 $,$ {\boldsymbol{s}}_i^0 $,$ {\boldsymbol{q}}_i^0 $分别表示节点对特征、支持实例特征和查询实例特征.每对节点都能构成关系图中节点,并构成一个紧密连接的图. Frog-GNN采用多角度聚合策略,分别执行节点对特征更新和实例特征更新. 计算相邻关系节点中支持节点和查询节点的特征相似度作为聚合权重系数来聚合邻居关系节点中的节点对特征. 计算相邻关系节点中支持或查询节点间相似度作为聚合权重系数来聚合邻居关系节点中的支持节点特征或查询节点特征. Frog-GNN的整体模型框架图如图6所示. 模型将最终的关系节点输入到全连接网络中,得到表示相同类的查询—支持对得分. 根据查询—支持对的支持部分,按类别平均预测得分,计算出查询样本属于每个类的最终概率. Frog-GNN被提出用于小样本文本分类任务. TPRN-T[64]是另一种通过构建节点对特征来解决图上的节点分类问题的模型,且其节点对特征的构建更加复杂. TRPN-T将每个支持—查询对的实例特征拼接得到节点对特征并进行构图. 关系节点特征为$ {{\boldsymbol{v}}_i} = [{{\boldsymbol{g}}_{{\text{support}}}},{{\boldsymbol{g}}_{{{q}}}}] $,其中$ {{\boldsymbol{g}}_{{\text{support}}}} $为支持节点特征,$ {{\boldsymbol{g}}_{{\text{query}}}} $为查询节点特征. 由于支持节点主导了关系节点,为防止查询节点信息的丢失,TPRN-T进一步引入伪关系节点,让查询样本主动参与到图中,伪关系节点定义为$ {{\boldsymbol{v}}_q} = [{{\boldsymbol{g}}_{{q}}},{{\boldsymbol{g}}_{{q}}}] $. 由于相似的支持样本通常与同一查询样本有相似的关系,因此TPRN-T可利用支持样本之间的已知关系,在图上进行传播,得到最终需预测的支持—查询对关系. 关系节点的邻接矩阵由节点对中的主导的支持节点特征相似性度量得到. 支持集的类内共性和类间差异性在关系图上进行传播和聚合,在关系传播和特征聚合之后,对关系节点使用相似度函数来衡量支持—查询对之间的相似度水平,并将其量化为相似度分数. 与Frog-GNN类似,根据相似度分数及查询—支持对中的支持样本部分确定查询样本所属的类别. 为了弥合可见任务和不可见任务之间的鸿沟,在TPRN-T的基础上,TPRN-D[65]模型在训练时运用解耦策略,在保持特征嵌入模块的判别表示能力的同时,保证TRPN能更好地提取可见任务和不可见任务之间的语义鸿沟. TPRN的模型框架如图7所示.
基于节点对的方法通过构建关系节点,即支持—查询节点对的方法来建模节点间关系,并将关系表示在图上显式地传播,基于此探索更丰富的节点间关系. 该方法提出类似的支持节点与同一查询节点有着相似的关系,利用已知的支持—支持节点间关系在关系图上进行传播,得到消息聚合后的支持—查询对关系. 最终模型可直接根据支持—查询对关系和基于支持节点判断查询节点的类型. 基于节点对特征的方法能够灵活地应用节点间的相关性,聚合节点特征信息.
FSL受到人类学习的启发,人类能在仅有少量学习样本时学习区分这些样本所属类别的能力. 值得注意的是,人类可以很容易地从少数样本中学习到类级知识,例如当人类在学习如何区分手机和电脑这2类时,会通过学习更加抽象的类级概念来更好地掌握区分2类物品的能力. 换句话说,类级知识可以理解为原型,即该类中所有样本都应具有的普遍特性等. 但大多数现有的基于图的FSL方法旨在探索查询样本和支持样本之间的相关性从而忽略了从样本中提取类级知识. 这促使基于GNN的FSL方法考虑如何显式地学习更丰富的类知识,以指导图中查询节点的信息推理.
分层GNN[66]提出具有层次结构的GNN来探索类别中可能具有的层次结构. 对于FSL,从样本中提取类别的区别特征非常重要. 为解决这个问题,分层GNN模型[67]提取类内具有代表性的节点特征和保留类间难区分的节点特征即一些相似度高的类间节点进行深层学习. 分层GNN主要包含自下向上推理、自顶向下推理和跳过连接层3部分. 自下向上推理模块利用GNN探索节点间相关性并利用类内k近邻池化层为每个类保留离其类质心最近的k个节点,即提取具有类内共同特征的节点特征. 为保留类间的区别特征,分层GNN利用类间节点池化层保留相近的不同类来进一步地探索. 通过类内和类间池化层对支持节点进行下采样,提取样本中丰富的类级知识. 与自下向上推理模块中的下采样相反,自顶向下推理模块对下采样后的节点进行上采样恢复到原始个数. 相同级别的自下向上块和自顶向下块用跳过连接层连接,有效地融合了不同尺度的类级特征和节点特征. 最后,利用映射函数将节点特征映射为节点标签. 分层GNN的模型框架如图8所示.
与利用池化模块提取类级知识的方式不同,文献[67]提出利用显示类别知识传播网络(explicit class knowledge propagation network, ECKPN)来解决FSL问题. ECKPN由比较、压缩和校准3个模块组成. 首先使用比较模块来探索成对的样本关系,以学习实例级图中的样本表示. 随后,压缩模块利用可训练的赋值矩阵压缩实例图来生成类级图,即${{\boldsymbol{V}}_{\text{c}}} = {{\boldsymbol{P}}^{\text{T}}}{{\boldsymbol{V}}^{(l)}}$,其中$ {{\boldsymbol{V}}^{(l)}} \in {\mathbb{R}^{r_1 \times d}} $为实例图中的节点特征表示,$ r_1 $为实例图中节点个数,d为特征维度. $ {\boldsymbol{P}} \in {\mathbb{R}^{r_1 \times r}} $为赋值矩阵,r为类级图中节点个数. 其中每个元素$ {p_{uv}} $表示原始图中的节点u被分配给类级图中的节点v的概率. 将实例级图压缩为类级图,有助于获取类级的视觉知识,并对类间关系进行建模. 校准模块结合Glove[68]提取的标签语义嵌入,多模态地建模类间关系,获得具有区别性的类级知识表示,并将类级知识与实例级样本表示相结合,以指导查询样本的推理. 分层GNN和ECKPN利用池化的思想对原始实例图中的节点进行压缩获取具有高级表征的类级知识,并通过融合实例知识和类级知识获取丰富的样本表示. 混合GNN[69]则从不同角度出发,旨在解决类边界的样本重叠和少量样本离群、异常的问题. 混合GNN由实例GNN(instance GNN, IGNN)和原型GNN(prototypical GNN,PGNN)组成,通过减小类内样本距离和增大每个类原型间距离的方法对特征嵌入网络学习到的特征进行调整. PGNN在类原型上操作,旨在解决类边界的样本重叠问题. 每个类由类均值或类原型进行初始化,以提取初始类级特征. 类级特征在图上进行消息传递,以最小化类间重叠和增大类间距离为学习目标. 学习到的PGNN重新排列每个类原型的相对位置,使类特征变得更易分离. IGNN通过实例消息传递专注于异常样本识别和中和离群样本,以最减小类内样本距离为目标,调整样本中异常样本和离群样本特征. 最终通过计算PGNN和IGNN特征更新后的原型节点和查询节点间的欧氏距离,将查询样本分配给具有更高可信度的类. 除图像分类任务的FSL方法外,小样本文本分类任务的方法研究通常是基于类级特征进行的,即采用原型网络来学习广义类原型. 然而在编码过程中,一些方法关注支持实例和查询实例间的交互和匹配信息,忽略了关于类内相似性和类间差异性等类级信息. 文献[19]提出了一种负监督胶囊图神经网络(negative-supervised capsule GNN, NSCGNN),该网络明确地利用样本间相似性和差异性来进行文本再现,使同一类型的点彼此更近、不同类型的点彼此更远,产生具有判别性的类原型表示. NSCGNN同时学习实例特征表示和类级特征表示,采用具有负监督的GNN来学习所有支持和查询实例的特征表示. 类级表示过程受胶囊网络[70]的启发,将从GNN中提取的节点特征视为基本胶囊,将类表示视为高级胶囊,采用动态路由算法来学习类的原型表示.
EGNN-prot模型[18]是另一种用于解决小样本文本分类任务的方法,该模型将边缘标记图神经网络和原型网络结合,首先利用边缘标记图神经网络建模文档的类内相似性和类间不相似性. 不同于EGNN中通过边缘特征直接计算查询节点类别的方式,EGNN-prot引入原型网络,利用原型知识对未标记文本进行分类. 原型网络计算类中支持样本的平均向量作为该类的原型向量表示. 最终通过比较所有原型向量和查询向量之间的距离,将查询样本分类到最接近的原型向量所在类中. 与利用原型网络提取类级知识不同,文献[71]在分类阶段利用最优传输算法来确定类中心,间接地提取类级概念. 文献[72]利用最优类分配模块来确定查询节点类别,即使用sinkhorn算法[73]计算一个类分配矩阵,以最小的运输成本将未标记的查询样本分配到类中心,并交替更新类中心和类分配矩阵.
基于类级特征的方法有2种利用类级特征的形式:1)在GNN前提取初始类级特征,并将类级特征在图上传播,通过GNN学习得到构建的目标类级特征,以此达到提升模型性能的目的. 2)在分类阶段利用类级知识进行分类,即在GNN前不涉及类级特征,在GNN后使用训练好的样本特征提取原型特征作为所属类的参考标准,由此将查询样本分类到具有更高可信度的类别中.
在小样本图像分类任务中,一些标准数据集被广泛使用. MiniImageNet[74]和 TieredImageNet[75]是FSL领域中最常用的数据集,同时CUB[76], CIFAR-100[77], Stanford Cars等常用作细粒度小样本图像分类.
1) MiniImageNet. 该数据集是从ImageNet[78]分割得到的,包含ImageNet的100个类别,每个类别含有600张图片. 将100个类分为64类、16类、20类分别用于训练、验证和测试.
2) TieredImageNet. 该数据集是Ren等人[75]在2018年提出的数据集,是比MiniImageNet更大的属于ImageNet的子集,其类别高达608种.
3) CUB-200-2011(caltech-UCSD birds). 该数据集是一个鸟类图像数据集,包含200种鸟类,共计11788张图像. 通常将数据集划分为130类、20类、50类分别用于训练、验证和测试.
4) CIFAR-FS. 该数据集共100个类,每个类包含600个图像,分别包括500个训练图像和100个测试图像.CIFAR-100中的100个子类所属于20个父类,每个图像都带有一个子类标签和一个父类标签.
5) Stanford Cars. 该数据集一般用于细粒度图像分类任务,包括196类车的样本共计16185个图像. 数据集被划分为130类、17类、49类分别用于训练、验证和测试. 表3所示为各类数据集中的统计信息与数据集划分情况.
本文对FSL的GNN方法中的一些著名模型在小样本图像分类任务数据集MiniImageNet和TieredImageNet上的性能表现进行了总结. 所有方法都选取了5-way 1-shot和5-way 5-shot设置的结果进行展示.
FSL通常使用CNN作为特征提取网络,比如Conv-4[37], ResNet12[79],WRN等常用的特征提取网络. 具体来说,Conv-4和ResNet12网络将输入图像编码为128维的特征向量,Conv-4和ResNet12都由4个区块组成. Conv-4中的每个块都包含3×3的卷积块、1个批处理标准化和1个LeakyRelu激活函数. 而ResNet12中的每个残差块包含3个大小为3×3的卷积层,每个卷积层之后是一个2×2的最大池化层. 而网络将图像编码为64维的特征向量.
部分模型利用了归纳学习和转导学习的实验设置进行模型性能的评估,其中归纳学习从给定的训练数据或任务中学习共同的模式,并转移到未知的任务和数据中,而转导学习则直接从训练数据或任务推广到给定的测试数据.
各类方法的实验结果具体如表4所示. 表4选取了MiniimageNet和TieredImageNet数据集的实验结果作为对比参考. 在FSL的图方法中,最优的模型在5-shot的图像分类任务上能够达到约88%的准确率;这证明了基于GNN的FSL方法的优越性. 早期的基于GNN的FSL方法仅考虑在图上传播节点特征,且对于节点间关系的探索停留在利用计算样本相似度来构建邻接矩阵的阶段. 因此这类方法虽然在性能上有一定的提升,但其分类准确率仍处于较低水平. 基于边缘特征的方法总体上优于基于节点特征的方法. EGNN是最早的边缘标记图GNN,该模型在Mini-ImageNet数据集的1-shot和5-shot测试的准确率分别达到了59%和76%. 基于节点对特征和基于类级特征的模型效果整体处于较高水平. 这2类模型在5-shot设置时性能有明显的提升,因为节点对特征和类级特征在支持样本数量较多时有明显优势.
本节从主要研究内容、关键技术和具体应用等方面对近几年的基于GNN的FSL代表性研究算法进行了相关内容的整理,结果如表5所示.
在GNN中节点特征是最易获取的信息,因此大多数算法将节点特征在图上进行消息传递,从而将已知标签的支持节点信息传播到未知标签的查询节点上,利用学习到的查询节点特征确定其类别. 在基于节点特征方法的基础上,基于边特征的方法还需考虑边特征的构建方法,不同于将样本特征作为节点特征的直观表示方法,边缘特征的建模缺乏统一的标准. 边缘特征不仅可以表达样本间的关系,还可以控制消息聚合的程度. 因此设计合理的初始边缘特征和边缘特征更新网络非常关键. 通过计算节点间的相似度来构建边缘特征是最简单的方式,然而当2个不属于同类的样本其特征较为相似时,此时将导致相似度引导的节点间边特征值较大,从而导致信息发生不适当的信息聚合,因此边缘特征的构建通常在标签空间进行,因为标签信息固定且对离群样本不敏感. 基于节点特征的方法旨在如何利用GNN进行有效的节点信息传播,基于边缘特征的方法则关注边缘特征的学习希望最终的边缘特征应能反映节点间准确的关系. 类似于基于边缘特征的方法,基于节点对特征的方法旨在挖掘节点间的深层关系. 不同的是,基于节点对特征的方法不采用将“边”表示为节点关系的固有思想,而构建节点对作为“关系节点”,并将“关系节点”作为单个节点进行构图得到关系图. 将已知的支持—支持节点对关系在图上进行传播,以此推理支持—查询节点对关系,从而确定查询样本标签. 节点对的思想为基于GNN的FSL方法提供了一个新思路,并在小样本图像分类任务上有着突出的性能. 基于类级特征的方法模拟了人类能从学习中提取类级知识的行为,从实例图上提取类级特征构建类级图,类级特征是一类样本抽象出的概念,利用类级知识在图上进行传播从而提取类内相似性和类间差异性或利用类级知识进行查询样本的匹配. 基于类级特征的方法希望类特征能够提炼类别的概念,从而将查询样本匹配给更高可信度的类. 表5中这4类方法旨在利用不同粒度的特征进行查询样本的预测,本文对这4种方法做了总结,并且比较了它们的优点和缺点.
FSL在一定程度上解决了样本量少的问题,它能够减少数据收集工作和计算成本,或作为模拟人类学习的垫脚石[81]. 因此,许多现实世界的应用程序都涉及到FSL,在许多场景中也得到了应用.
基于GNN的FSL中最常用的数据集为图像分类数据,这是因为视觉信息很容易获取,并且得到了广泛的研究[82]. 而基于GNN的FSL方法也已经在小样本图像分类数据集MiniImageNet和TieredimageNet等数据集上获得了较高的准确率. 因此,研究人员开始探索更多的图像分类任务.
图像分类任务几乎覆盖了基于GNN的FSL方法的所有应用场景,由于图像分类问题的解决模式相对固定,解决思路相对简单,且很多图像分类任务都有与其相对应的图像数据集. 因此基于GNN的FSL方法在除ImgeNet数据集图像分类任务外的高光谱图分类[20, 54, 83-85]、金属表面分类[86-87]和遥感场景分类[88-89]等图像分类任务中也取得了一定的研究进展. 文献[20]提出了一种跨领域的对偶图FSL框架,试图通过将FSL与域对齐相结合来解决跨场景高光谱图分类任务. 当目标数据出现新类时,该模型分类准确率仍能稳定在较高水平. 缺陷分类对于金属表面的缺陷检测至关重要. 而现今由于良好优化的制造技术,在工业场景中收集缺陷图像相当困难,因此FSL对于金属表面缺陷检测具有重要意义. 文献[86]提出利用图嵌入与分布变换(graph embedding and distribution transformation, GEDT)模块和最优传输(optimal transport, OPT)模块来解决小样本金属表面分类任务. GEDT模块能够充分利用支持集和查询集中不同特征之间的关联信息,保证图嵌入结果分布的一致性. OPT模块以转导的方式实现小样本分类. 场景分类是遥感技术的一项基础性工作,具有广泛的应用前景. 随着在数据紧急情况下满足高方差情况的需求的增加,为大量的遥感场景进行标注可能需要耗费大量的人力,因此少镜头遥感场景分类研究越来越受到人们的关注,其重点是利用少量训练样本建立分类模型. 文献[88]提出了一种基于连续元学习原理的小样本场景分类算法,该算法通过在元训练或元测试阶段融合来自一系列任务的更多历史先验知识来增强任务间的相关性.
探究疾病预测领域的FSL方法促进了计算机辅助技术在疾病预测领域的实际应用. 一些计算机辅助技术往往基于传统的深度学习算法,然而,由于医学数据具有隐私保护性强以及标注成本高的特点,疾病领域的数据往往难以获取,缺乏大型的医学数据集. 疾病诊断是疾病预测领域中一项至关重要的技术,随着深度学习的蓬勃发展,许多计算机辅助诊断技术用于医学疾病的诊断,根据医学影像进行疾病诊断是计算机辅助技术最常见的一种诊断形式. 医学影像的产生来源于病例,而通常少量的病例使一些深度学习模型失效. GNN由于其能够融合多模态特征可以自然地将医学影像数据和受试者表型数据同时利用起来. 在疾病预测领域,利用GNN进行节点分类或者图分类来对受试者做出疾病预测的方法已经得到了广泛的运用. 文献[21]提出了一种用于疾病诊断的自动度量GNN模型,通过引入基于度量的元学习策略和设计了一个具有概率约束的GNN层进行节点多分类任务,实现独立测试的归纳学习. 该方法不仅缓解了当数据量较少时诊断准确率较低的问题,还克服了传统节点分类方法中固定大型图结构不适用于独立测试的问题.
除计算机视觉领域外,基于GNN的FSL方法在自然语言处理领域也取得了一定的进展,在自然语言处理领域中的文本分类[17-19, 90-91]任务中有较多涉及GNN的FSL学习方法. 随着深度学习和预训练语言模型的发展,自然语言处理领域中的各种研究方法已经取得了巨大的成功. 然而,这些模型取决于训练数据的数量和质量. 由于数据稀缺和语义复杂性,FSL在自然语言处理中仍然是一个具有挑战性的问题.
文本分类任务与图像分类任务同属于小样本分类问题,且文本数据也可以和图像数据一样映射到特征嵌入空间中,然后通过文本语义特征来建模样本之间的关系. 但相较于图像数据,文本数据的表现形式更加丰富,且文本之间的关系更加复杂. 文献[90]提出了一种新的犹豫模糊图神经网络模型(hesitant fuzzy GNN, HFGNN)来探索样本之间的多属性关系,并将HFGNN与原型网络结合,以实现小样本的文本分类. 文献[17]提出了一种基于多角度聚合的GNN, 分别关注支持和查询实例以及支持—查询节点对的更新和聚合,利用GNN将节点间关系显式地在图上传播,进行文本分类.
尽管近年来基于GNN的FSL已取得了一定进展,但仍面临着6个挑战.
1) 基于GNN的FSL方法遵循元学习的情景训练策略,因此模型的训练虽然在特定任务上只需要少量标注实例,但仍需大量训练任务作为“训练样本”,这致使“小样本学习”在一定程度上变成了伪命题,从本质上来看,与FSL的定义背道而驰[92]. 因此从根本上解决训练样本量的问题对于未来FSL的图方法是一个巨大的挑战.
2) 基于GNN的FSL方法通过构建一系列的训练任务作为训练样本用于训练,其目的是使模型能够解决类似的FSL任务,这些任务构建的图结构共享图模型的参数. 然而基于GNN的FSL方法中,GNN专注于任务内节点关系的建模,忽略了任务间的高阶关系. 由于每个任务的样本是随机采样的,即使在相同的图构建方式下,最终构建得到的图结构仍有特定于任务的度量范围与特征空间,而这与GNN模型的参数共享思想矛盾,使得在训练过程中很难找到一个最优的元参数. 如何构建适合所有任务的元GNN模型或建立任务间的高阶关系是基于GNN的FSL方法的一个难点.
3) 基于GNN的FSL方法在训练任务上训练GNN模型并使其能够泛化到测试任务. 虽然训练任务与测试任务来自相同数据集,但测试任务中的类是训练任务中未出现过的,测试任务与训练任务的样本特征分布跨度较大. 在元学习中,训练阶段的模型应用到测试任务时,模型会在支持样本上微调,而基于GNN的FSL中的图模型将任务间的样本统一建模为图网络,无法使用微调的方法,此时GNN作为度量函数的泛化能力将面临巨大的考验.
4) 基于GNN的FSL学习框架通常通过联合元学习与特征提取器以达到特征重用和快速学习的目的,但基于GNN的元学习分类器与特征提取器的训练本质上是矛盾的,因为分类器需快速适应不同的任务,而特征嵌入模块则不需要,因此这种联合元学习和特征嵌入模块的训练模式亟待改变. 若直接将特征嵌入模块与元学习分类器分离,可能导致提取到的特征与特定任务所需的特征有所偏离,因此如何校准特征嵌入模块与元学习分类器之间的特征分布差异值得考虑.
5) 在FSL的图方法中通常将GNN视为分类器,GNN的本质是进行消息传递,将支持样本标签传递给查询样本,此时GNN执行分类器的功能. 此外GNN也可以通过邻居节点聚合学习具有代表性的节点表示. 因此GNN是否能在FSL中执行除分类器以外的功能是值得考虑的问题. 突破GNN在整体训练框架中仅作为分类器的固有模式对于未来FSL的图方法研究工作是一项挑战.
6) GNN不可避免地存在过平滑问题,当任务内采样的样本属于同类但特征较为相似时,会导致边权较大而发生不适当的信息聚合,此时样本间的差异性信息可能会随着GNN网络的加深被中和,节点的特征趋于一致. 在基于GNN的FSL方法中,节点特征由于较为相似而丢失了具有区别性的信息,因此过平滑问题是模型性能欠佳的一大隐患. 在节点信息发生充分交互的前提下,如何缓解过平滑问题是基于GNN的FSL方法的又一难点.
由于真实世界中样本稀缺的问题,FSL越来越受到人们的重视. 基于图神经网络的FSL的分类问题已经在计算机视觉等领域有了深入的研究,通过对当前FSL的图方法研究进展的梳理,可以展望未来基于图的FSL的发展方向.
1) 先验知识的利用. 现有的基于GNN的FSL使用预训练的神经网络作为特征提取模块提取样本特征,因此特征提取模块仍需要大量数据用于训练. 未来可以尝试利用其他先验知识训练模型,或者合理利用无标注数据. 探索更多不依赖模型预训练、使用先验知识就能取得较好效果的方法.
2) 弥合任务间的差异性.GNN由于能够挖掘样本间的关系在FSL领域展现出优秀的性能,但基于GNN的FSL方法仅关注任务内问题的建模,而任务间的差异不仅影响了模型最优参数的确定也使模型由训练任务迁移到测试任务时性能大大降低,未来可以通过建模任务间的高阶关系来弥合任务间的差异性,建立适用于所有任务的全局一致的特征空间和度量标准可能是解决任务差异带来的问题的一个切入点. 在基于元学习的FSL方法中,对每个任务,元模型在支持集上微调后得到特定于该任务的模型,基于GNN的FSL方法可借鉴这种学习框架利用微调的方式间接地提高模型的泛化能力.
3) 弥合模块间差异性. 在基于GNN的FSL方法中,存在GNN模块特征与特征提取模块提取得到的特征分布不匹配问题,这严重影响了GNN的模型效率,因此对特征提取模块提取到的特征进行分布校准或特征对齐是解决该矛盾的有效途径. 例如可以通过高斯分布变换等方法来对分布进行校准. 考虑设计合理的分布校准模块,弥合特征网络与图网络的特征分布间差异可以作为未来基于GNN的FSL方法的突破点之一.
4) 应用场景和图结构的创新. 相较于传统的FSL方法,基于GNN的FSL方法性能好且可解释性强,但其应用场景较少. 在有相应小样本数据集的条件下,探索除图像分类外新场景中FSL的图方法利用是一个较佳的想法. 此外如何设计创新图网络模型值得进一步探究. 例如GNN可以很自然地进行多模态信息的融合[93-95],而现有FSL的图方法缺乏对多模态融合方法的探究,因此在未来的研究中可以考虑利用多模态信息挖掘更丰富的样本间关系,或构建具有多模态特征的节点特征来表示更准确的样本特征.
5) 更先进的FSL方法. 基于GNN的FSL需要一系列任务来进行元训练. 虽然这确实反映了人类的终身学习,但在一些应用程序中,此类任务族的数据可能不可用. 此外,对于某些FSL任务,缺乏足够的数据来构建一系列的任务. 已知的无监督元学习[5, 96-97]和单任务元学习方法[98-100]可以帮助缓解这种问题,未来基于GNN的FSL方法可结合这类更先进的元学习方法进一步地研究. 此外基于GNN的FSL方法可以寻求与其他FSL方法的融合. 数据作为深度学习的驱动力,对于模型的训练至关重要. 因此可通过融合数据增强的方法使基于GNN的FSL方法得到改进.
6) 跨领域的小样本学习. 在实际应用中,训练任务中的数据和测试任务中的数据可能来自不同的数据集,而无论是传统的FSL方法还是基于GCN的FSL方法,其中训练任务和测试任务中的数据均来自同一数据集,这限制了FSL方法的实际应用. 已知的跨域FSL方法[101]提供了一个很好的学习框架,到目前为止,针对小样本图像分类的领域自适应及领域泛化方法还很少. 因此,基于GNN进行跨领域的FSL研究值得进一步深入研究.
7) 前沿的图方法应用. 大部分基于GNN的FSL方法利用GCN模型来进行学习,而GCN仅仅是GNN中的一个分支,未来可以考虑使用一些前沿的图方法来解决FSL问题. 例如可以使用超图卷积来建模更复杂的节点间关系,在多模态数据下利用超图卷积能够更好地探索模态间的数据关系,更好地解决FSL问题. 除超图卷积外,还可以通过使用图transformer[102-103]等来提升图结构注意,在早期的FSL的图方法中已经尝试结合GAT来解决单纯图模型带来的过拟合和过平滑问题,但该方法并没有带来明显的性能提升. 未来可以考虑使用图强化学习、图transformer等更复杂的图模型来解决FSL问题.
从小样本的概念被提出,到基于元学习的FSL方法的兴起,再到GNN被用于解决FSL问题以及各类FSL方法在各个领域的广泛应用,FSL在理论和实践上经历了一个从无到有、从有到优化的过程. FSL的各类方法也在不断地完善和发展. 本文概括和提炼了目前主流的基于GNN的FSL方法,并对不同算法进行划分、归纳和总结,分析了基于GNN的FSL的应用以及未来发展方向. 基于GNN的FSL方法的出现为FSL问题的解决提供了一个全新的思路. 就模型性能而言,各类基于GNN的FSL方法已经超越了大多数过往的FSL方法,为FSL的研究进展做出了巨大的贡献. 即便如此,基于GNN的FSL方法仍有较大的改进空间. 在未来研究的道路上,还有更多优秀的FSL的图方法出现,为实现真正的FSL和发展人工智能技术贡献出一份力量.
作者贡献声明:杨洁祎负责资料收集、文献归纳整理、研究方案的构思和设计、论文撰写及修订;董一鸿提供研究思路、论文组织结构的设计、论文审阅及修订;钱江波负责论文审阅及修订.

参考文献 引证文献
排序方式:
1
葛轶洲,刘恒,王言,等. 小样本困境下的深度学习图像识别综述[J]. 软件学报,2022,33(1):193−210
Ge Yizhou, Liu Heng, Wang Yan, et al. Survey on deep learning image recognition in dilemma of small samples[J]. Journal of Software, 2022, 33(1): 193−210 (in Chinese)
2
李凡长,刘洋,吴鹏翔,等. 元学习研究综述[J]. 计算机学报,2021,44(2):422−446
Li Fanchang, Liu Yang, Wu Pengxiang, et al. A survey on recent advance in meta-learning[J]. Chinese Journal of Computers, 2021, 44(2): 422−446 (in Chinese)
3
Sun Qianru, Liu Yaoyao, Chua T, et al. Meta-transfer learning for few-shot learning [C] //Proc of the 32nd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 403−412
4
Wang Peng, Liu Lingqiao, Shen Chunhua, et al. Multi-attention network for one shot learning [C] //Proc of the 30th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2017: 2721−2729
5
Khodadadeh S, Boloni L, Shan M. Unsupervised meta-learning for few-shot image classification [C] //Proc of the 33rd Annual Conf on Neural Information-Processing Systems. Cambrige, MA: MIT, 2019: 10132−10142
6
赵凯琳,靳小龙,王元卓. 小样本学习研究综述[J]. 软件学报,2021,32(2):349−369
Zhao Kailin, Jin Xiaolong, Wang Yuanzhuo. Survey on few-shot learning[J]. Journal of Software, 2021, 32(2): 349−369 (in Chinese)
7
Howard J, Ruder S. Universal language model fine-tuning for text classification[J]. arXiv preprint, arXiv: 1801. 06146, 2018
8
Royle J, Dorazio R, Link W. Analysis of multinomial models with unknown index using data augmentation[J]. Journal of Computational and Graphical Statistics, 2007, 16(1): 67−85
9
Liu Zicheng, Li Siyuan, Wu Di, et al. AutoMix: Unveiling the power of mixup for stronger classifiers [C] //Proc of the 17th European Conf on Computer Vision. Berlin: Springer, 2022: 441−458
10
Kipf T, Welling M. Semi-supervised classification with graph convolutional networks[J]. arXiv preprint, arXiv: 1609. 02907, 2016
11
Blaes S, Burwick T. Few-shot learning in deep networks through global prototyping[J]. Neural Networks, 2017, 94: 159−172
12
Rahman S, Khan S, Porikli F. A unified approach for conventional zero-shot, generalized zero-shot, and few-shot learning[J]. IEEE Transactions on Image Processing, 2018, 27((11): ): 5652−5267
13
Cheng Yu, Yu Mo, Guo Xiaoxiao, et al. Few-shot learning with meta metric learners[J]. arXiv preprint, arXiv: 1901. 09890, 2019
14
马帅,刘建伟,左信. 图神经网络综述[J]. 计算机研究与发展,2022,59(1):47−80
Ma Shuai, Liu Jianwei, Zuo Xin. Survey of graph neural networks[J]. Journal of Computer Research and Development, 2022, 59(1): 47−80 (in Chinese)
15
谢小杰,梁英,王梓森,等. 基于图卷积的异质网络节点分类方法[J]. 计算机研究与发展,2022,59(7):1470−1485
Xie Xiaojie, Liang Ying, Wang Zisen, et al. Node classification method for heterogeneous networks based on graph convolution[J]. Journal of Computer Research and Development, 2022, 59(7): 1470−1485 (in Chinese)
16
任嘉睿,张海燕,朱梦涵,等. 基于元图卷积的异质网络嵌入学习算法[J]. 计算机研究与发展,2022,59(8):1683−1693
Ren Jiarui, Zhang Haiyan, Zhu Menghan, et al. Embedding learning algorithm for heterogeneous networks based on metagram convolution[J]. Journal of Computer Research and Development, 2022, 59(8): 1683−1693 (in Chinese)
17
Xu Shiyao, Xiang Yang. Frog-GNN: Multi-perspective aggregation based graph neural network for few-shot text classification[J]. Expert Systems with Applications, 2021, 176: 114795
18
Lyu Chen, Liu Weijie, Wang Ping. Few-shot text classification with edge-labeling graph neural network-based prototypical network [C] //Proc of the 28th Int Conf on Computational Linguistics. Stroudsburg, PA: ACL, 2020: 5547−552
19
Ding Ling, Chen Xiaojun, Xiang Yang. Negative-supervised capsule graph neural network for few-shot text classification[J]. Journal of Intelligent & Fuzzy Systems, 2021, 41(6): 6875−6887
20
Zhang Yuxiang, Li Wei, Zhang Mengmeng, et al. Dual graph cross-domain few-shot learning for hyperspectral image classification [C] //Proc of the 47th IEEE Int Conf on Acoustics Speech and Signal Processing. Piscataway, NJ: IEEE, 2022: 3573−3577
21
Song Xiaofa, Mao Mingyi, Qian Xiaohua. Auto-metric graph neural network based on a meta-learning strategy for the diagnosis of Alzheimer’s disease[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(8): 3141−3152
22
Xiong chao, Li wen, Wang Minghui, et al. Multi-dimensional edge features graph neural network on few-shot image classification[J]. IEEE Signal Processing Letters, 2021, 28: 573−577
23
刘颖,雷研博,范九伦,等. 基于小样本学习的图像分类技术综述[J]. 自动化学报,2021,47(2):297−315
Liu Ying, Lei Yanbo, Fan Jiulun, et al. Survey on image classification technology based on small sample learning[J]. Acta Automatica Sinica, 2021, 47(2): 297−315 (in Chinese)
24
徐冰冰,岑科廷,黄俊杰,等. 图卷积神经网络综述[J]. 计算机学报,2020,43(5):755−780
Xu Bingbing, Cen Keting, Huang Junjie, et al. A survey on graph convolutional neural networks[J]. Chinese Journal of Computers, 2020, 43(5): 755−780 (in Chinese)
25
Nakamura A, Harada T. Revisiting fine-tuning for few-shot learning[J]. arXiv preprint, arXiv: 1910. 00216, 2019
26
Dhillon G S, Ghaudhari P, Ravichandran A, et al. A baseline for few-shot image classification [C/OL] //Proc of the 8th Int Conf on Learning Representation. Amsterdam: Elsevier, 2020 [2023-03-26].https://openreview.net/forum?id=rylXBkrYDS
27
Wang Y, Girshick R, Hebert M, et al. Low-shot learning from imaginary data [C] //Proc of the 30th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2018: 7278−7286
28
Liu Bo, Wang Xudong, Dixit M, et al. Feature space transfer for data augmentation [C] //Proc of the 30th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2018: 9090−9098
29
Jing Kunlei, Zhang Xinman, Yang Zhiyuan, et al. Feature augmentation learning for few-shot palmprint image recognition with unconstrained acquisition [C] //Proc of the 47th IEEE Int Conf on Acoustics Speech and Signal Processing. Piscataway, NJ: IEEE, 2022: 3323−3327
30
蒋留兵,周小龙,姜风伟,等. 基于改进匹配网络的单样本学习[J]. 系统工程与电子技术,2019,41(6):1210−1217
Jiang Liubing, Zhou Xiaolong, Jiang Fengwei, et al. One-shot learning based on improved matching network[J]. Systems Engineering and Electronics, 2019, 41(6): 1210−1217 (in Chinese)
31
Snell J, Swersky K, Zemel R. Prototypical networks for few-shot learning [C] //Proc of the 31st Annual Conf on Neural InformationProcessing Systems. Cambrige, MA: MIT, 2017: 4077−4087
32
Li Wenbin, Xu Jinglin, Huo Jing, et al. Distribution consistency based covariance metric networks for few-shot learning [C] //Proc of the 33rd AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2019: 8642−8649
33
Jiang Wen, Huang Kai, Geng Jie, et al. Multi-scale metric learning for few-shot learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 31(3): 1091−1102
34
<mixed-citation>Tian Yingjie, Zhao Xiaoxi, Huang Wei. Meta-learning approaches for learning-to-learn in deep learning: A survey [J]. Neurocomputing, 2022, 494: 203−223</mixed-citation>
Tian Yingjie,Zhao Xiaoxi,Huang Wei. Meta-learning approaches for learning-to-learn in deep learning:A survey [J]. Neurocomputing,2022,494:203−223
35
李维刚,甘平,谢璐,等. 基于样本对元学习的小样本图像分类方法[J]. 电子学报,2022,50(2):295−304
Li Weigang, Gan Ping, Xie Lu, et al. A few-shot image classification method by pairwise-based meta learning[J]. Acta Electronica Sinica, 2022, 50(2): 295−304 (in Chinese)
36
Ravi S, Larochelle H. Optimization as a model for few-shot learning [C/OL] //Proc of the 4th Int Conf on Learning Representation. Amsterdam: Elsevier, 2016 [2023-03-26].https://openreview.net/pdf?id=rJY0-Kcll
37
Finn C, Abbeel P, Levine S. Model-agnostic meta-learning for fast adaptation of deep networks [C] //Proc of the 34th ACM Int Conf on Machine Learning. New York: ACM, 2017: 1126−1135
38
Li Zhenguo, Zhuo Fengwei, Chen Fei, et al. Meta-SGD: Learning to learn quickly for few-shot learning[J]. arXiv preprint, arXiv: 1707. 09835, 2017
39
Vinyals O, Blundell C, Lilicrap T, et al. Matching networks for one shot learning [C] //Proc of the 30th Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2016: 3630−3638
40
Santoro A, Bartunov S, Botvinick M, et al. Meta-learning with memory-augmented neural networks [C] //Proc of the 33rd ACM Int Conf on Machine Learning. New York: ACM, 2016: 1842−1850
41
Brendel W, Bethge M. Approximating CNNs with bag-of-local-features models works surprisingly well on ImageNet [C/OL] //Proc of the 7th Int Conf on Learning Representation. Amsterdam: Elsevier, 2019 [2023-03-26].https://openreview.net/forum?id=rJY0-Kcll
42
Lai K, Zha Daochen, Zhou Kaixiong, et al. Policy-GNN: Aggregation optimization for graph neural networks [C] //Proc of the 26th ACM SIGKDD Conf on Knowledge Discovery and Data Mining. New York: ACM, 2020: 461−471
43
Garcia V, Bruna J. Few-shot learning with graph neural networks [C/OL] //Proc of the 6th Int Conf on Learning Representation. Amsterdam: Elsevier, 2018 [2023-03-26].https://openreview.net/pdf ?id=BJj6qGbRW
44
Liu Yanbin, Lee J, Park M, et al. Learning to propagate labels: Transductive propagation network for few-shot learning [C/OL] //Proc of the 7th Int Conf on Learning Representation. Amsterdam: Elsevier, 2019 [2023-03-26].https://openreview.net/pdf?id=SyVuRiC5K7
45
Qiao Limeng, Shi Yemin, Li Jia, et al. Transductive episodic-wise adaptive metric for few-shot learning [C] //Proc of the 11th IEEE Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2019: 3602−3611
46
Angluin D, Smith C. Inductive inference: Theory and methods[J]. ACM Computing Surveys, 1983, 15(3): 237−269
47
Chen Jinyin, Lin Xiang, Xiong Hui, et al. Smoothing adversarial training for GNN[J]. IEEE Transactions on Computational Social Systems, 2020, 8(3): 618−629
48
Cheng Hao, Zhou J, Tay W, et al. Attentive graph neural networks for few-shot learning [C] //Proc of the 5th IEEE Int Conf on Multimedia Information Processing and Retrieval. Piscataway, NJ: IEEE, 2022: 152−157
49
Velickovic P, Cucurull G, Casanova A, et al. Graph attention networks[J]. arXiv preprint, arXiv: 1710. 10903, 2017
50
<mixed-citation>Zhang Xu, Zhang Youjia, Zhang Zuyu, et al. Discriminative learning of imaginary data for few-shot classification [J]. Neurocomputing, 2022, 467: 406−417</mixed-citation>
Zhang Xu,Zhang Youjia,Zhang Zuyu,et al. Discriminative learning of imaginary data for few-shot classification [J]. Neurocomputing,2022,467:406−417
51
Xu Rui, Liu Baodi, Lu Xiaoping, et al. DMH-FSL: Dual-modal hypergraph for few-shot learning[J]. Neural Processing Letters, 2022, 54(2): 1317−1332
52
Yadati N, Nimishakavi M, Yadav P, et al. HyperGCN: A new method for training graph convolutional networks on hypergraphs [C] //Proc of the 33rd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2019: 1509−1520
53
Kim J, Kim T, Kim S, et al. Edge-labeling graph neural network for few-shot learning [C] //Proc of the 32nd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 11−20
54
Zuo Xibing, Yu Xuchun, Liu Bing, et al. FSL-EGNN: Edge-labeling graph neural network for hyperspectral image few-shot classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1−18
55
Xiong Chao, Li Wen, Liu Yun, et al. Multi-dimensional edge features graph neural network on few-shot image classification[J]. IEEE Signal Processing Letters, 2021, 28: 573−577
56
Tang Shixiang, Chen Dapeng, Bai Lei, et al. Mutual CRF-GNN for few-shot learning [C] //Proc of the 34th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 2329−2339
57
<mixed-citation>Bale T, Vale W. CRF and CRF receptors: Role in stress responsivity and other behaviors [J]. Annual Review of Pharmacology and Toxicology, 2004, 44: 525−557</mixed-citation>
Bale T,Vale W. CRF and CRF receptors:Role in stress responsivity and other behaviors [J]. Annual Review of Pharmacology and Toxicology,2004,44:525−557
58
Yang Lin, Li Liangliang, Zhang Zilun, et al. DPGNN: Distribution propagation graph network for few-shot learning [C] //Proc of the 33rd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 13387−13396
59
Zhang Bailin, Ling Hefei, Shen Jialie, et al. Mixture distribution graph network for few shot learning[J]. IEEE Transactions on Cognitive and Developmental Systems, 2022, 14(3): 892−901
60
Hu Yufan, Gao Junyu, Xu Changsheng. Learning dual-pooling graph neural networks for few-shot video classification[J]. IEEE Transactions on Multimedia, 2020, 23: 4285−4296
61
Mordeson J. Fuzzy mathematics [M]. Foundations of Image Understanding. Berlin: Springer, 2001: 95−125
62
Wei Tong, Hou Junlin, Feng Rui. Fuzzy graph neural network for few-shot learning [C/OL] //Proc of the 30th IEEE Int Joint Conf on Neural Networks. Piscataway, NJ: IEEE, 2020 [2023-03-26].https://openreview.net/pdf?id=BJj6qGbRW
63
Devlin J, Chang M, Lee K, et al. Bert: Pre-training of deep bidirectional transformers for language understanding [C] //Proc of the 17th Annual Conf of the North American chapter of the Association for Computational Linguistics. Stroudsburg, PA: ACL, 2019: 4171−4186
64
Ma Yuqing, Bai Shihao, An Shan, et al. Transductive relation-propagation network for few-shot Learning [C] //Proc of the 29th Int Joint Conf on Artificial Intelligence. San Francisco, CA: Morgan Kaufmann, 2020: 804−810
65
Ma Yuqing, Bai Shihao, Liu Wei, et al. Transductive relation-propagation with decoupling training for few-shot learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(11): 6652−6664
66
Chen Cen, Li Kenli, Wei Wei, 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
67
Chen Chaofan, Yang Xiaoshan, Xu Changsheng, et al. ECKPN: Explicit class knowledge propagation network for transductive few-shot learning [C] //Proc of the 34th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 6596−6605
68
Pennington J, Socher R, Manning C. Glove: Global vectors for word representation [C] //Proc of the 28th Int Conf on Empirical Methods in Natural Language Processing. Stroudsburg, PA: ACL, 2014: 1532−1543
69
Yu Tianyuan, He Sen, Song Yizhe, et al. Hybrid graph neural networks for few-shot learning [C] //Proc of the 36th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2022: 3179−3187
70
Deng Fei, Pu Shengliang, Chen Xuehong, et al. Hyperspectral image classification with capsule network using limited training samples[J]. Sensors, 2018, 18(9): 3153−3174
71
Zhang Ruiheng, Yang Shuo, Zhang Qi, et al. Graph-based few-shot learning with transformed feature propagation and optimal class allocation[J]. Neurocomputing, 2022, 470: 247−256
72
Tantawi A, Towsley G, Wolf J. Optimal allocation of multiple class resources in computer systems [C] //Proc of the 16th ACM SIGMETRICS Conf on Measurement and Modeling of Computer Systems. New York: ACM, 1988: 253−260
73
Cuturi M. Sinkhorn distances: Lightspeed computation of optimal transport [C] //Proc of the 27th Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2013: 2292−2300
74
Malalur P, Jaakkola T. Alignment based matching networks for one-shot classification and open-set recognition[J]. arXiv preprint, arXiv: 1903. 06538, 2019
75
Ren M, Triantafillou E, Ravi S, et al. Meta-learning for semi-supervised few-shot classification [C/OL]//Proc of the 6th Int Conf on Learning Representation. Amsterdam: Elsevier, 2018 [2023-03-26].https://openreview.net/pdf?id=HJcSzz-CZ
76
Cui Yin , Zhou Feng, Lin Yuanqing, et al. Fine-grained categorization and dataset bootstrapping using deep metric learning with humans in the loop [C] //Proc of the 29th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2016: 1153−1162
77
Singla S, Singla S, Feizi S. Improved deterministic l2 robustness on CIFAR-10 and CIFAR-100 [C/OL] //Proc of the 10th Int Conf on Learning Representation. Amsterdam: Elsevier, 2022 [2023-03-26].https://openreview.net/forum?id=tD7eCtaSkR
78
Deng J, Dong W, Socher R, et al. ImageNet: A large-scale hierarchical image database [C] //Proc of the 22nd IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2009: 248−255
79
He Kaiming, Zhang Xianyu, Ren Shaoqing, et al. Deep residual learning for image recognition [C] //Proc of the 29th IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2016: 770−778
80
Lin Guangfeng, Yang Ying, Fan Yindi, et al. High-order structure preserving graph neural network for few-shot learning[J]. arXiv preprint, arXiv: 2005. 14415, 2020
81
Wang Yaqing, Yao Quanming, Kwok J, et al. Generalizing from a few examples: A survey on few-shot learning[J]. ACM Computing Surveys, 2020, 53(3): 1−34
82
Bendre N, Marin H, Najafirad P. Learning from few samples: A survey[J]. arXiv preprint, arXiv: 2007. 15484, 2020
83
Li Na, Zhou Deyun, Shi Jiao, et al. Graph-based deep multitask few-shot learning for hyperspectral image classification[J]. Remote Sensing, 2022, 14(9): 2246−2267
84
Tong Xinyi, Yin Jihao, Han Bingnan, et al. Few-shot learning with attention-weighted graph convolutional networks for hyperspectral image classification [C] //Proc of the 27th IEEE Int Conf on Image Processing. Piscataway, NJ: IEEE, 2020: 1686−1690
85
Hong Danfeng, Gao Lianru, Yao Jing, et al. Graph convolutional networks for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(7): 5966−5978
86
Xiao Weiwei, Song Kechen, Liu Jie, et al. Graph embedding and optimal transport for few-shot classification of metal surface defect[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1−10
87
Bao Yanqi, Song Kechen, Liu Jie, et al. Triplet-graph reasoning network for few-shot metal generic surface defect segmentation[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1−11
88
Li Feimo, Li Shuaibo, Fan Xinxin, et al. Structural attention enhanced continual meta-learning for graph edge labeling based few-shot remote sensing scene classification[J]. Remote Sensing, 2022, 14(3): 485−515
89
Yuan Zhengwu, Huang Wendong, Tang Chan, et al. Graph-based embedding smoothing network for few-shot scene classification of remote sensing images[J]. Remote Sensing, 2022, 14(5): 1161−1179
90
Guo Xinyu, Tian Bingjie, Tian Xuedong. HFGNN-proto: Hesitant fuzzy graph neural network-based prototypical network for few-shot text classification[J]. Electronics, 2022, 11(15): 2423−2437
91
Ma Ning, Bu Jiajun, Yang Jieyu, et al. Adaptive-step graph meta-learner for few-shot graph classification [C] //Proc of the 29th ACM Int Conf on Information &amp; Knowledge Management. New York: ACM, 2020: 1055−1064
92
Hospedales T, Antoniou A, Micaelli P, et al. Meta-learning in neural networks: A survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(9): 5149−5169
93
Pan Jiacheng, Lin Haocai, Dong Yihong, et al. MAMF-GCN: Multi-scale adaptive multi-channel fusion deep graph convolutional network for predicting mental disorder[J]. Computers in Biology and Medicine, 2022, 148: 105823
94
Wei Yinwei, Wang Xiang, Nie Liqiang, et al. MMGCN: Multi-modal graph convolution network for personalized recommendation of micro-video [C] //Proc of the 27th ACM Int Conf on Multimedia. New York: ACM, 2019: 1437−144
95
Tao Zhulin, Wei Yinwei, Wang Xiang, et al. MGAT: Multimodal graph attention network for recommendation[J]. Information Processing & Management, 2020, 57(5): 102277
96
Hsu K, Levine S, Finn C. Unsupervised learning via meta-learning [C/OL] //Proc of the 7th Int Conf on Learning Representation. Amsterdam: Elsevier, 2018 [2023-03-26].https://openreview.net/forum?id=r1My6sR9tX
97
Antoniou A, Storkey A. Assume, augment and learn: Unsupervised few-shot meta-learning via random labels and data augmentation[J]. arXiv preprint, arXiv: 1902. 09884, 2019
98
Veeriah V, Hessel M, Xu Zhongwen, et al. Discovery of useful questions as auxiliary tasks [C] // Proc of the 33rd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2019: 9306−9317
99
Zheng Zeyu, Oh J, Singh S. On learning intrinsic rewards for policy gradient methods [C] // Proc of the 32nd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2018: 4649−4659
100
Meier F, Kappler D, Schaal S. Online learning of a memory for learning rates [C] //Proc of the 4th IEEE Int Conf on Robotics and Automation. Piscataway, NJ: IEEE, 2018: 2425−2432
101
Tseng H, Lee H, Huang Jiabin, et al. Cross-domain few-shot classification via learned feature-wise transformation [C/OL] //Proc of the 9th Int Conf on Learning Representation. Amsterdam: Elsevier, 2020 [2023-03-26].https://openreview.net/forum?id=SJl5Np4tPr
102
Yun S, Jeong M, Kim R, et al. Graph transformer networks [C] // Proc of the 33rd Annual Conf on Neural Information Processing Systems. Cambrige, MA: MIT, 2019: 11960−11970
103
Jiang Bo, Zhao Kangkang, Tang jin, et al. RGTransformer: Region-graph transformer for image representation and few-shot classification[J]. IEEE Signal Processing Letters, 2022, 29: 792−796
2024年第61卷第4期
78
36
0
引用本文
BibTeX
文章信息
doi: 10.7544/issn1000-1239.202220933
  • 接收时间:2022-11-11
  • 首发时间:2026-11-17
  • 出版时间:2024-04-01
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2022-11-11
  • 修回日期:2023-05-16
基金
作者信息
    1宁波大学信息科学与工程学院 浙江宁波 315211
    2浙江省移动网络应用技术重点实验室(宁波大学) 浙江宁波 315211

通讯作者:

董一鸿()
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202220933
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏