Article(id=1329435688237379963, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1329435571803501430, articleNumber=null, orderNo=null, doi=10.7544/issn1000-1239.202550799, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1762099200000, receivedDateStr=2025-11-03, revisedDate=1780502400000, revisedDateStr=2026-06-04, acceptedDate=1782057600000, acceptedDateStr=2026-06-22, onlineDate=1794891767795, onlineDateStr=2026-11-17, pubDate=1793462400000, pubDateStr=2026-11-01, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1794891767795, onlineIssueDateStr=2026-11-17, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1794891767795, creator=13701087609, updateTime=1794891767795, updator=13701087609, issue=Issue{id=1329435571803501430, tenantId=1146029695717560320, journalId=1324759914146529478, year='2026', volume='63', issue='11', pageStart='2767', pageEnd='3031', issueExtLink='null', onlineDate='null', pubDate='1793462400000', pubDateStr='2026-11-01', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=1, specialIssue=null, createTime=1794891740036, creator='13701087609', updateTime=1794891740036, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext=null, issueFiles=null, downloadFileDto=null}, startPage=3017, endPage=3031, ext={EN=ArticleExt(id=1329435688421929340, articleId=1329435688237379963, tenantId=1146029695717560320, journalId=1324759914146529478, language=EN, title=Enhanced Balanced Incremental Distillation Network for Debiased Dermatological Disease Diagnosis, columnId=1325489891789910943, journalTitle=Journal of Computer Research and Development, columnName=Artificial Intelligence, runingTitle=null, highlight=null, articleAbstract=

Deep learning technology has been increasingly empowering the practice of dermatological disease diagnosis, leading to a steady improvement in its diagnostic performance. However, such data-driven deep learning models are plagued by severe bias issues. If systems with these biases are deployed in clinical applications, they are bound to cause unfair treatment towards specific groups, such as underrepresented groups. Ensuring the fairness of decision-making has become a critical and urgent problem that needs to be addressed. Existing research has indicated that enhancing fairness often comes at the cost of reduced diagnostic performance. Therefore, to achieve a better trade-off between diagnostic performance and fairness, while simultaneously alleviating the unfair treatment of underrepresented groups, we propose an enhanced balanced incremental distillation network (EBID-Net). Specifically, with the aid of balanced memory, the model leverages representative demographic groups to assist underrepresented groups in knowledge learning, while being incrementally trained by integrating distributions across different groups. Additionally, we incorporate global information into the contextual attention mechanism to capture correlated interactions between features across different spatial locations, thereby obtaining more robust feature representations during incremental learning. Experimental results show that EBID-Net outperforms other methods in fairness criteria and in the trade-off between fairness and diagnostic performance.

, authors=Yiqin Luo1, Tianlong Gu1, *, Fengrui Hao1, Liang Chang2, authorsList=Yiqin Luo, Tianlong Gu, Fengrui Hao, Liang Chang, authorCompany=null, correspAuthors=Tianlong Gu, authorNote=

born in 1991. PhD. Student member of CCF. Her main research interests include fair machine learning and trustworthy artificial intelligence

born in 1964. PhD, professor. Senior member of CCF. His main research interests include formal methods, trustworthy artificial intelligence, artificial intelligence safety, and artificial intelligence ethics

born in 1995. PhD. Student member of CCF. His main research interests include fair machine learning and trustworthy artificial intelligence

born in 1980. PhD, professor. Senior member of CCF. His main research interests include data and knowledge engineering, trustworthy artificial intelligence, artificial intelligence ethics, and data privacy protection

, correspAuthorsNote=null, copyrightStatement=Copyright © 2026 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=1329435691936756113, articleId=1329435688237379963, tenantId=1146029695717560320, journalId=1324759914146529478, language=CN, title=面向皮肤病诊断去偏的增强型平衡增量蒸馏网络, columnId=1325489892104483745, journalTitle=计算机研究与发展, columnName=人工智能, runingTitle=null, highlight=null, articleAbstract=

深度学习技术日益赋能皮肤病诊断应用,使其诊断性能稳步提升。然而,这类数据驱动的深度学习模型存在严重的偏见问题。若将存在偏见的系统部署于临床应用,势必会对特定群体(如非代表性群体)造成不公平对待。确保决策的公平性已成为一个至关重要且亟待解决的问题。现有研究表明,提升公平性往往以降低诊断性能为代价。因此,为了在诊断性能和公平性之间取得更优权衡,同时减轻对非代表性群体的不公平对待,提出一个增强型平衡增量蒸馏网络(enhanced balanced incremental distillation network,EBID-Net)。具体而言,该模型借助平衡记忆内存整合不同群体的分布开展增量训练,促使代表性群体辅助非代表性群体学习知识。此外,将全局信息融入上下文注意力机制,捕捉不同空间位置特征间的交互关系,从而在增量学习过程中获得更稳健的特征表示。实验结果显示,EBID-Net在公平性以及诊断性能和公平性间的权衡方面均优于其他方法。

, authors=罗义琴1, 古天龙1, *, 郝峰锐1, 常亮2, authorsList=罗义琴, 古天龙, 郝峰锐, 常亮, authorCompany=null, correspAuthors=古天龙, authorNote=

1991年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

1964年生。博士,教授。CCF高级会员。主要研究方向为形式化方法、可信人工智能、人工智能安全、人工智能伦理

1995年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

1980年生。博士,教授。CCF高级会员。主要研究方向为数据与知识工程、可信人工智能、人工智能伦理、数据隐私保护

, correspAuthorsNote=
古天龙()
, copyrightStatement=版权所有 © 2026 计算机研究与发展编辑部, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=o0gwENIpX542UPGbQC2UMg==, magXml=OnwP8CCcd9nyfx3NUSlTMw==, pdfUrl=null, pdf=uLm+N6iSynBQjobA3K41Gw==, pdfFileSize=4410929, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=O/kbzlNi3/4ujPyKEABnBA==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=jNpMv85H8vFHi/0cvOnm1Q==, mapNumber=null, fund=null)}, authors=[Author(id=1329435692301660569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=lyq_ah@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329435692372963739, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692301660569, language=EN, stringName=Yiqin Luo, firstName=Yiqin, middleName=null, lastName=Luo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632, bio={"content":"

born in 1991. PhD. Student member of CCF. Her main research interests include fair machine learning and trustworthy artificial intelligence

"}, bioImg=null, bioContent=

born in 1991. PhD. Student member of CCF. Her main research interests include fair machine learning and trustworthy artificial intelligence

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435692431683996, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692301660569, language=CN, stringName=罗义琴, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632, bio={"img":"+ZkvzXHgExEW4Po6XK/h5w==","content":"

1991年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

"}, bioImg=+ZkvzXHgExEW4Po6XK/h5w==, bioContent=

1991年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692154859922, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=1, ext=[AuthorCompanyExt(id=1329435692163248531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632), AuthorCompanyExt(id=1329435692171637140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632)])]), Author(id=1329435692511375774, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=gutianlong@jnu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329435692578484640, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692511375774, language=EN, stringName=Tianlong Gu, firstName=Tianlong, middleName=null, lastName=Gu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632, bio={"content":"

born in 1964. PhD, professor. Senior member of CCF. His main research interests include formal methods, trustworthy artificial intelligence, artificial intelligence safety, and artificial intelligence ethics

"}, bioImg=null, bioContent=

born in 1964. PhD, professor. Senior member of CCF. His main research interests include formal methods, trustworthy artificial intelligence, artificial intelligence safety, and artificial intelligence ethics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435692641399201, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692511375774, language=CN, stringName=古天龙, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632, bio={"img":"XvnoLojK7lVwQFaeJXIIfA==","content":"

1964年生。博士,教授。CCF高级会员。主要研究方向为形式化方法、可信人工智能、人工智能安全、人工智能伦理

"}, bioImg=XvnoLojK7lVwQFaeJXIIfA==, bioContent=

1964年生。博士,教授。CCF高级会员。主要研究方向为形式化方法、可信人工智能、人工智能安全、人工智能伦理

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692154859922, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=1, ext=[AuthorCompanyExt(id=1329435692163248531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632), AuthorCompanyExt(id=1329435692171637140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632)])]), Author(id=1329435692700119459, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=1329435692771422629, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692700119459, language=EN, stringName=Fengrui Hao, firstName=Fengrui, middleName=null, lastName=Hao, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632, bio={"content":"

born in 1995. PhD. Student member of CCF. His main research interests include fair machine learning and trustworthy artificial intelligence

"}, bioImg=null, bioContent=

born in 1995. PhD. Student member of CCF. His main research interests include fair machine learning and trustworthy artificial intelligence

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435692825948582, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692700119459, language=CN, stringName=郝峰锐, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632, bio={"img":"9eDGRdBEJGiYnrSrPyLWRA==","content":"

1995年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

"}, bioImg=9eDGRdBEJGiYnrSrPyLWRA==, bioContent=

1995年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692154859922, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=1, ext=[AuthorCompanyExt(id=1329435692163248531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632), AuthorCompanyExt(id=1329435692171637140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632)])]), Author(id=1329435692893057448, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, orderNo=3, 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=1329435692960166314, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692893057448, language=EN, stringName=Liang Chang, firstName=Liang, middleName=null, lastName=Chang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Guangxi Key Laboratory of Trusted Software (Guilin University of Electronic Technology), Guilin, Guangxi 541004, bio={"content":"

born in 1980. PhD, professor. Senior member of CCF. His main research interests include data and knowledge engineering, trustworthy artificial intelligence, artificial intelligence ethics, and data privacy protection

"}, bioImg=null, bioContent=

born in 1980. PhD, professor. Senior member of CCF. His main research interests include data and knowledge engineering, trustworthy artificial intelligence, artificial intelligence ethics, and data privacy protection

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435693027275179, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692893057448, language=CN, stringName=常亮, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2广西可信软件重点实验室(桂林电子科技大学) 广西桂林 541004, bio={"img":"qqLoq37wAqHAa+xqVRjTlQ==","content":"

1980年生。博士,教授。CCF高级会员。主要研究方向为数据与知识工程、可信人工智能、人工智能伦理、数据隐私保护

"}, bioImg=qqLoq37wAqHAa+xqVRjTlQ==, bioContent=

1980年生。博士,教授。CCF高级会员。主要研究方向为数据与知识工程、可信人工智能、人工智能伦理、数据隐私保护

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692238746005, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=2, ext=[AuthorCompanyExt(id=1329435692247134614, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692238746005, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Guangxi Key Laboratory of Trusted Software (Guilin University of Electronic Technology), Guilin, Guangxi 541004), AuthorCompanyExt(id=1329435692255523223, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692238746005, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2广西可信软件重点实验室(桂林电子科技大学) 广西桂林 541004)])])], keywords=[Keyword(id=1329435693119549868, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, orderNo=1, keyword=fairness), Keyword(id=1329435693186658733, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, orderNo=2, keyword=dermatological disease), Keyword(id=1329435693245378990, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, orderNo=3, keyword=incremental learning), Keyword(id=1329435693304099247, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, orderNo=4, keyword=trade-off), Keyword(id=1329435693358625200, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, orderNo=5, keyword=attention mechanism), Keyword(id=1329435693421539761, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, orderNo=1, keyword=公平), Keyword(id=1329435693488648626, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, orderNo=2, keyword=皮肤病), Keyword(id=1329435693551563187, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, orderNo=3, keyword=增量学习), Keyword(id=1329435693622866356, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, orderNo=4, keyword=权衡), Keyword(id=1329435693689975221, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, orderNo=5, keyword=注意力机制)], refs=[Reference(id=1329435697905250764, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Karimkhani C, Dellavalle R P, Coffeng L E, et al. Global skin disease morbidity and mortality: An update from the global burden of disease study 2013[J]. JAMA Dermatology, 2017, 153(5): 406−412, articleTitle=null, refAbstract=null), Reference(id=1329435697963971021, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=null, journalName=null, refType=null, unstructuredReference=医疗大数据应用技术国家工程实验室(中南大学)皮肤疾病大数据工作委员会, 中华医学会皮肤性病学分会皮肤肿瘤研究中心, 中国医师协会皮肤科医师分会皮肤肿瘤亚专业委员会. 皮肤病流行病学研究专家共识[J]. 中华皮肤科杂志, 2020, 53(12): 951−961, articleTitle=null, refAbstract=null), Reference(id=1329435698047857102, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=National Engineering Laboratory for Medical Big Data Application Technology (Central South University), Skin Disease Big Data Working Committee, Skin Tumor Research Center, Chinese Society of Dermatology, Chinese Medical Association, Sub-specialty Committee of Skin Tumor, Dermatology Branch, Chinese Medical Doctor Association. Expert consensus on the methodology of dermato-epidemiological investigations[J]. Chinese Journal of Dermatology, 2020, 53(12): 951−961 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435698144326095, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=3, authorNames=null, journalName=null, refType=null, unstructuredReference=陈宇聪, 董晓华, 梁泽民, 等. 皮肤病智能辅助诊疗系统的研究与设计[J]. 中国卫生信息管理杂志, 2025, 22(3): 443−449, articleTitle=null, refAbstract=null), Reference(id=1329435698224017872, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Chen Yucong, Dong Xiaohua, Liang Zemin, et al. Research and design of intelligent auxiliary diagnosis and treatment system for skin diseases[J]. Chinese Journal of Health Informatics and Management, 2025, 22(3): 443−449 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435698291126737, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=耿蕾蕾, 许振熙, 纪则轩, 等. 基于拓扑先验和双分支视觉Mamba的皮肤病变分割[J]. 计算机研究与发展, 2026, 63(5): 1290−1304, articleTitle=null, refAbstract=null), Reference(id=1329435698379207122, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=6, authorNames=null, journalName=null, refType=null, unstructuredReference=Geng Leilei, Xu Zhenxi, Ji Zexuan, et al. Skin lesion segmentation based on topological prior and dual-branch vision Mamba[J]. Journal of Computer Research and Development, 2026, 63(5): 1290−1304 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435698475676115, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=7, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu Yuan, Jain A, Eng C, et al. A deep learning system for differential diagnosis of skin diseases[J]. Nature Medicine, 2020, 26(6): 900−908, articleTitle=null, refAbstract=null), Reference(id=1329435698572145108, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Papachristou P, Söderholm M, Pallon J, et al. Evaluation of an artificial intelligence-based decision support for the detection of cutaneous melanoma in primary care: A prospective real-life clinical trial[J]. British Journal of Dermatology, 2024, 191(1): 125−133, articleTitle=null, refAbstract=null), Reference(id=1329435698639253973, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Huang Kai, Wu Xian, Li Yixin, et al. Artificial intelligence–based psoriasis severity assessment: Real-world study and application[J]. Journal of Medical Internet Research, 2023, 25: e44932, articleTitle=null, refAbstract=null), Reference(id=1329435698702168534, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Lin Mingquan, Li Tianhao, Yang Yifan, et al. Improving model fairness in image-based computer-aided diagnosis[J]. Nature Communications, 2023, 14(1): 6261, articleTitle=null, refAbstract=null), Reference(id=1329435698773471703, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Seth P, Pai A K. Does the fairness of your pre-training hold up? Examining the influence of pre-training techniques on skin tone bias in skin lesion classification[C]//Proc of the 13th IEEE/CVF Winter Conf on Applications of Computer Vision. Piscataway, NJ: IEEE, 2024: 570−577, articleTitle=null, refAbstract=null), Reference(id=1329435698836386264, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=范卓娅, 孟小峰. 算法公平与公平计算[J]. 计算机研究与发展, 2023, 60(9): 2048−2066, articleTitle=null, refAbstract=null), Reference(id=1329435698903495129, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=13, authorNames=null, journalName=null, refType=null, unstructuredReference=Fan Zhuoya, Meng Xiaofeng. Algorithmic fairness and fairness computing[J]. Journal of Computer Research and Development, 2023, 60(9): 2048−2066 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435698970603994, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=14, authorNames=null, journalName=null, refType=null, unstructuredReference=Zong Yongshuo, Yang Yongxin, Hospedales T. MEDFAIR: Benchmarking fairness for medical imaging[J]. arXiv preprint, arXiv: 2210.01725, 2022, articleTitle=null, refAbstract=null), Reference(id=1329435699041907163, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=12, rfOrder=15, authorNames=null, journalName=null, refType=null, unstructuredReference=Burgon A, Zhang Yuhang, Petrick N, et al. Bias amplification to facilitate the systematic evaluation of bias mitigation methods[J]. IEEE Journal of Biomedical and Health Informatics, 2025, 29(2): 1444−1454, articleTitle=null, refAbstract=null), Reference(id=1329435699100627420, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=16, authorNames=null, journalName=null, refType=null, unstructuredReference=Abdulqawy A, Sallam E, Elkholy A. UnBias: Unveiling bias implications in deep learning models for healthcare applications[J]. IEEE Journal of Biomedical and Health Informatics, 2025, 29(2): 1464−1471, articleTitle=null, refAbstract=null), Reference(id=1329435699176124893, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Mehrabi N, Morstatter F, Saxena N, et al. A survey on bias and fairness in machine learning[J]. ACM Computing Surveys, 2021, 54(6): 1−35, articleTitle=null, refAbstract=null), Reference(id=1329435699239039454, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=Churamani N, Kara O, Gunes H. Domain-incremental continual learning for mitigating bias in facial expression and action unit recognition[J]. IEEE Transactions on Affective Computing, 2022, 14(4): 3191−3206, articleTitle=null, refAbstract=null), Reference(id=1329435699310342623, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=19, authorNames=null, journalName=null, refType=null, unstructuredReference=Mccloskey M, Cohen N J. Catastrophic Interference in Connectionist Networks: The Sequential Learning Problem[M]//Psychology of Learning and Motivation. Amsterdam: Academic, 1989: 109−165, articleTitle=null, refAbstract=null), Reference(id=1329435699373257184, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=20, authorNames=null, journalName=null, refType=null, unstructuredReference=Xu Zikang, Zhao Shang, Quan Quan, et al. FairAdaBN: Mitigating unfairness with adaptive batch normalization and its application to dermatological disease classification[C]//Proc of the 26th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2023: 307−317, articleTitle=null, refAbstract=null), Reference(id=1329435699436171745, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Yehao, Yao Ting, Pan Yingwei, et al. Contextual Transformer networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(2): 1489−1500, articleTitle=null, refAbstract=null), Reference(id=1329435699507474914, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=22, authorNames=null, journalName=null, refType=null, unstructuredReference=Tang Xiaoya, Zhang Bodong, Ho M M, et al. Duoformer: Leveraging hierarchical representations by local and global attention vision transformer[J]. arXiv preprint, arXiv: 2506.12982, 2025, articleTitle=null, refAbstract=null), Reference(id=1329435699570389475, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=23, authorNames=null, journalName=null, refType=null, unstructuredReference=Chowdary G J, Yin Zhaozheng. Med-former: A Transformer based architecture for medical image classification[C]//Proc of the 27th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2024: 448−457, articleTitle=null, refAbstract=null), Reference(id=1329435699633304036, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=24, authorNames=null, journalName=null, refType=null, unstructuredReference=Van de Ven G M, Tuytelaars T, Tolias A S. Three types of incremental learning[J]. Nature Machine Intelligence, 2022, 4(12): 1185−1197, articleTitle=null, refAbstract=null), Reference(id=1329435699692024293, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=25, authorNames=null, journalName=null, refType=null, unstructuredReference=韩亚楠, 刘建伟, 罗雄麟. 连续学习研究进展[J]. 计算机研究与发展, 2022, 59(6): 1213−1239, articleTitle=null, refAbstract=null), Reference(id=1329435699754938854, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=26, authorNames=null, journalName=null, refType=null, unstructuredReference=Han Yanan, Liu Jianwei, Luo Xionglin. Research progress of continual learning[J]. Journal of Computer Research and Development, 2022, 59(6): 1213−1239 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435699822047719, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=27, authorNames=null, journalName=null, refType=null, unstructuredReference=Smith J S, Karlinsky L, Gutta V, et al. Coda-prompt: Continual decomposed attention-based prompting for rehearsal-free continual learning[C]//Proc of the 36th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2023: 11909−11919, articleTitle=null, refAbstract=null), Reference(id=1329435699889156584, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=28, authorNames=null, journalName=null, refType=null, unstructuredReference=Shen Xiaoyu, Zhang Tao, Wu Hao, et al. An adaptive domain-incremental framework with knowledge replay and domain alignment for specific emitter identification[J]. IEEE Transactions on Information Forensics and Security, 2025, 20: 3519−3533, articleTitle=null, refAbstract=null), Reference(id=1329435699977236969, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=25, rfOrder=29, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen Jinpeng, Cong R, Luo Yuxuan, et al. Replay without saving: Prototype derivation and distribution rebalance for class-incremental semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(6): 4699−4716, articleTitle=null, refAbstract=null), Reference(id=1329435700040151530, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=26, rfOrder=30, authorNames=null, journalName=null, refType=null, unstructuredReference=Weng Zhi, Lu Yongzheng, Zheng Zhiqiang, et al. Few-shot class-incremental learning for classification and object detection: A survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(4): 2924−2945, articleTitle=null, refAbstract=null), Reference(id=1329435700103066091, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=27, rfOrder=31, authorNames=null, journalName=null, refType=null, unstructuredReference=古天龙, 李龙, 常亮, 等. 公平机器学习: 概念, 分析与设计[J]. 计算机学报, 2022, 45(5): 1018−1051, articleTitle=null, refAbstract=null), Reference(id=1329435700161786348, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=27, rfOrder=32, authorNames=null, journalName=null, refType=null, unstructuredReference=Gu Tianlong, Li Long, Chang Liang, et al. Fair machine learning: Concept, analysis, and design[J]. Chinese Journal of Computers, 2022, 45(5): 1018−1051 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435700220506605, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=28, rfOrder=33, authorNames=null, journalName=null, refType=null, unstructuredReference=陈晋音, 陈奕芃, 陈一鸣, 等. 面向深度学习的公平性研究综述[J]. 计算机研究与发展, 2021, 58(2): 264−280, articleTitle=null, refAbstract=null), Reference(id=1329435700279226862, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=28, rfOrder=34, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen Jinyin, Chen Yipeng, Chen Yiming, et al. Fairness research on deep learning[J]. Journal of Computer Research and Development, 2021, 58(2): 264−280 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435700388278767, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=29, rfOrder=35, authorNames=null, journalName=null, refType=null, unstructuredReference=Seyyed-Kalantari L, Liu Guanxiong, McDermott M, et al. CheXclusion: Fairness gaps in deep chest X-ray classifiers[C]//Proc of the Pacific Symp (BIOCOMPUTING 2021). Singapore: World Scientific, 2020: 232−243, articleTitle=null, refAbstract=null), Reference(id=1329435700446999024, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=30, rfOrder=36, authorNames=null, journalName=null, refType=null, unstructuredReference=Joshi N, Burlina P. AI fairness via domain adaptation[J]. arXiv preprint, arXiv: 2104.01109, 2021, articleTitle=null, refAbstract=null), Reference(id=1329435700514107889, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=31, rfOrder=37, authorNames=null, journalName=null, refType=null, unstructuredReference=Yao Ruichen, Cui Ziteng, Li Xiaoxiao, et al. Improving fairness in image classification via sketching[J]. arXiv preprint, arXiv: 2211.00168, 2022, articleTitle=null, refAbstract=null), Reference(id=1329435700597993970, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=32, rfOrder=38, authorNames=null, journalName=null, refType=null, unstructuredReference=Puyol-Antón E, Ruijsink B, Piechnik S K, et al. Fairness in cardiac MR image analysis: An investigation of bias due to data imbalance in deep learning based segmentation[C]//Proc of the 24th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2021: 413−423, articleTitle=null, refAbstract=null), Reference(id=1329435700660908531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=33, rfOrder=39, authorNames=null, journalName=null, refType=null, unstructuredReference=Yao Siqiong, Dai Fang, Sun Peng, et al. Enhancing the fairness of AI prediction models by Quasi-Pareto improvement among heterogeneous thyroid nodule population[J]. Nature Communications, 2024, 15(1): 1958, articleTitle=null, refAbstract=null), Reference(id=1329435700719628788, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=34, rfOrder=40, authorNames=null, journalName=null, refType=null, unstructuredReference=Wu Yawen, Zeng Dewen, Xu Xiaowei, et al. Fairprune: Achieving fairness through pruning for dermatological disease diagnosis[C]//Proc of the 25th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2022: 743−753, articleTitle=null, refAbstract=null), Reference(id=1329435700782543349, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=35, rfOrder=41, authorNames=null, journalName=null, refType=null, unstructuredReference=Kong Qingpeng, Chiu Chinghao, Zeng Dewen, et al. Achieving fairness through channel pruning for dermatological disease diagnosis[C]//Proc of the 27th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2024: 24−34, articleTitle=null, refAbstract=null), Reference(id=1329435700908372470, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=36, rfOrder=42, authorNames=null, journalName=null, refType=null, unstructuredReference=Jin Ruinan, Deng Wenlong, Chen Minghui, et al. Debiased noise editing on foundation models for fair medical image classification[C]//Proc of the 27th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2024: 164−174, articleTitle=null, refAbstract=null), Reference(id=1329435700962898423, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=37, rfOrder=43, 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=1329435701046784504, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=37, rfOrder=44, 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=1329435701139059193, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=38, rfOrder=45, authorNames=null, journalName=null, refType=null, unstructuredReference=Castro F M, Marín-Jiménez M J, Guil N, et al. End-to-end incremental learning[C]//Proc of the 15th European Conf on Computer Vision. Berlin: Springer, 2018: 233−248, articleTitle=null, refAbstract=null), Reference(id=1329435701197779450, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=39, rfOrder=46, authorNames=null, journalName=null, refType=null, unstructuredReference=周大蔚, 汪福运, 叶翰嘉, 等. 基于深度学习的类别增量学习算法综述[J]. 计算机学报, 2023, 46(8): 1577−1605, articleTitle=null, refAbstract=null), Reference(id=1329435701260694011, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=39, rfOrder=47, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhou Dawei, Wang Fuyun, Ye Hanjia, et al. Deep learning for class-incremental learning: A survey[J]. Chinese Journal of Computers, 2023, 46(8): 1577−1605 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329435701327802876, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=40, rfOrder=48, authorNames=null, journalName=null, refType=null, unstructuredReference=Zemel R, Wu Y, Swersky K, et al. Learning fair representations[C]//Proc of the 21st Int Conf on Machine Learning. New York: PMLR, 2013: 325−333, articleTitle=null, refAbstract=null), Reference(id=1329435701403300349, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=41, rfOrder=49, authorNames=null, journalName=null, refType=null, unstructuredReference=Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]// Proc of the 29th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2016: 2818−2826, articleTitle=null, refAbstract=null), Reference(id=1329435701478797822, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=42, rfOrder=50, authorNames=null, journalName=null, refType=null, unstructuredReference=Groh M, Harris C, Daneshjou R, et al. Towards transparency in dermatology image datasets with skin tone annotations by experts, crowds, and an algorithm[J]. Proceedings of the ACM on Human-Computer Interaction, 2022, 6(CSCW2): 1−26, articleTitle=null, refAbstract=null), Reference(id=1329435701541712383, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=43, rfOrder=51, authorNames=null, journalName=null, refType=null, unstructuredReference=Tschandl P, Rosendahl C, Kittler H. The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions[J]. Scientific Data, 2018, 5(1): 1−9, articleTitle=null, refAbstract=null), Reference(id=1329435701604626944, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=44, rfOrder=52, authorNames=null, journalName=null, refType=null, unstructuredReference=Codella N C F, Gutman D, Celebi M E, et al. Skin lesion analysis toward melanoma detection: A challenge at the 2017 international symposium on biomedical imaging (ISBI), hosted by the international skin imaging collaboration (ISIC)[C]//Proc of the 15th IEEE Int Symp on Biomedical Imaging. Piscataway, NJ: IEEE, 2018: 168−172, articleTitle=null, refAbstract=null), Reference(id=1329435701671735809, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=45, rfOrder=53, authorNames=null, journalName=null, refType=null, unstructuredReference=Combalia M, Codella N C, Rotemberg V, et al. Bcn20000: Dermoscopic lesions in the wild[J]. arXiv preprint, arXiv: 1908.02288, 2019, articleTitle=null, refAbstract=null), Reference(id=1329435701734650370, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=46, rfOrder=54, authorNames=null, journalName=null, refType=null, unstructuredReference=Hardt M, Price E, Srebro N. Equality of opportunity in supervised learning[J/OL]. Advances in Neural Information Processing Systems, 2016[2026-05-23]. https://proceedings.neurips.cc/paper_files/paper/2016/hash/6a9659feb1216f14f7384ba499518b38-Abstract.html, articleTitle=null, refAbstract=null), Reference(id=1329435701793370627, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=47, rfOrder=55, authorNames=null, journalName=null, refType=null, unstructuredReference=Sagawa S, Koh P W, Hashimoto T B, et al. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization[J]. arXiv preprint, arXiv: 1911.08731, 2019, articleTitle=null, refAbstract=null), Reference(id=1329435701860479492, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=48, rfOrder=56, authorNames=null, journalName=null, refType=null, unstructuredReference=Tartaglione E, Barbano C A, Grangetto M. End: Entangling and disentangling deep representations for bias correction[C]//Proc of the 34th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 13508−13517, articleTitle=null, refAbstract=null), Reference(id=1329435701923394053, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=49, rfOrder=57, authorNames=null, journalName=null, refType=null, unstructuredReference=Quadrianto N, Sharmanska V, Thomas O. Discovering fair representations in the data domain[C]//Proc of the 32nd IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 8227−8236, articleTitle=null, refAbstract=null), Reference(id=1329435701982114310, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=50, rfOrder=58, authorNames=null, journalName=null, refType=null, unstructuredReference=Jung S, Lee D, Park T, et al. Fair feature distillation for visual recognition[C]//Proc of the 34th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 12115−12124, articleTitle=null, refAbstract=null), Reference(id=1329435702040834567, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=51, rfOrder=59, authorNames=null, journalName=null, refType=null, unstructuredReference=Chiu Chinghao, Chung Haowei, Chen Yujen, et al. Toward fairness through fair multi-exit framework for dermatological disease diagnosis[C]//Proc of the 26th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2023: 97−107, articleTitle=null, refAbstract=null), Reference(id=1329435702099554824, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=52, rfOrder=60, authorNames=null, journalName=null, refType=null, unstructuredReference=Xu Gelei, Duan Yuying, Liu Zheyuan, et al. Incorporating rather than eliminating: Achieving fairness for skin disease diagnosis through group-specific experts[C]//Proc of the 28th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2025: 284−294, articleTitle=null, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1329435692154859922, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=1, ext=[AuthorCompanyExt(id=1329435692163248531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632), AuthorCompanyExt(id=1329435692171637140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632)]), AuthorCompany(id=1329435692238746005, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=2, ext=[AuthorCompanyExt(id=1329435692247134614, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692238746005, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Guangxi Key Laboratory of Trusted Software (Guilin University of Electronic Technology), Guilin, Guangxi 541004), AuthorCompanyExt(id=1329435692255523223, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692238746005, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2广西可信软件重点实验室(桂林电子科技大学) 广西桂林 541004)])], figs=[ArticleFig(id=1329435693815804342, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Figure 1, caption=The main framework of EBID-Net, figureFileSmall=EhFPAq94LOvo3K+JQSCEZw==, figureFileBig=O/kbzlNi3/4ujPyKEABnBA==, tableContent=null), ArticleFig(id=1329435696240112055, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=图1, caption=EBID-Net主要框架, figureFileSmall=EhFPAq94LOvo3K+JQSCEZw==, figureFileBig=O/kbzlNi3/4ujPyKEABnBA==, tableContent=null), ArticleFig(id=1329435696441438648, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Figure 2, caption=Illustration of EConAtt, figureFileSmall=NpJv+iD+7dBIRCmvqSSEow==, figureFileBig=1TEaXxps2AIgo5shc6DHCQ==, tableContent=null), ArticleFig(id=1329435696508547513, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=图2, caption=EConAtt示意图, figureFileSmall=NpJv+iD+7dBIRCmvqSSEow==, figureFileBig=1TEaXxps2AIgo5shc6DHCQ==, tableContent=null), ArticleFig(id=1329435696575656378, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Figure 3, caption=Impact of distillation loss, figureFileSmall=L47VKfmASfumjmLM2mvciA==, figureFileBig=3AtmlVfb+mEbU1zDxOraog==, tableContent=null), ArticleFig(id=1329435696634376635, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=图3, caption=蒸馏损失的影响, figureFileSmall=L47VKfmASfumjmLM2mvciA==, figureFileBig=3AtmlVfb+mEbU1zDxOraog==, tableContent=null), ArticleFig(id=1329435696705679804, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Figure 4, caption=Impact of statistical parity disparity loss, figureFileSmall=vOjEfHbJxR2C/fXDq94rKQ==, figureFileBig=QfU5l3jJgqtm/25nWzihVA==, tableContent=null), ArticleFig(id=1329435696760205757, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=图4, caption=统计均等差异损失的影响, figureFileSmall=vOjEfHbJxR2C/fXDq94rKQ==, figureFileBig=QfU5l3jJgqtm/25nWzihVA==, tableContent=null), ArticleFig(id=1329435696823120318, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Figure 5, caption=Impact of memory size, figureFileSmall=ypTU8UmjjSy4NRQsDxTmLQ==, figureFileBig=koi6r9pSz1TIcU1ADmJrlQ==, tableContent=null), ArticleFig(id=1329435696890229183, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=图5, caption=记忆内存大小的影响, figureFileSmall=ypTU8UmjjSy4NRQsDxTmLQ==, figureFileBig=koi6r9pSz1TIcU1ADmJrlQ==, tableContent=null), ArticleFig(id=1329435696957338048, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Figure 6, caption=Attention visualization results of dermatological disease images with different classes on ISIC 2019 dataset, figureFileSmall=jGKo44sHFbRO44XuoCgjvA==, figureFileBig=+kZ0DfEpwcN25RaAN48EPw==, tableContent=null), ArticleFig(id=1329435697045418433, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=图6, caption=ISIC 2019数据集上不同类别皮肤病图像的注意力可视化结果, figureFileSmall=jGKo44sHFbRO44XuoCgjvA==, figureFileBig=+kZ0DfEpwcN25RaAN48EPw==, tableContent=null), ArticleFig(id=1329435697099944386, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Table 1, caption=

Results of Diagnostic Performance, Fairness and Trade-off of Various Methods on Fitzpatrick17k Dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
方法肤色诊断性能 /%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
ResNet18深色82.86±0.01382.97±0.01281.80±0.0131.03±0.1019.88±1.46810.38±1.7910.00±0.0000.00±0.0000.00±0.000
浅色75.46±0.00376.64±0.00174.94±0.003
总体77.68±0.00378.62±0.00477.13±0.002
Resampling[32]深色83.38±0.00983.60±0.00882.63±0.0070.88±0.16711.01±1.19111.66±1.37015.66±0.155−10.67±0.113−11.59±0.125
浅色76.18±0.00877.16±0.00575.68±0.003
总体78.28±0.00679.17±0.00577.86±0.003
GroupDRO[47]深色80.38±0.00779.87±0.01179.72±0.0080.94±0.1799.23±0.3349.77±0.3986.58±0.1734.74±0.0334.07±0.037
浅色74.54±0.00874.71±0.01174.07±0.009
总体76.25±0.00776.32±0.01175.83±0.008
EnD[48]深色79.90±0.00379.74±0.00379.45±0.0031.05±0.1428.31±1.5678.67±1.753−5.26±0.13312.53±0.16313.19±0.173
浅色72.99±0.00772.71±0.01072.59±0.008
总体75.10±0.00474.91±0.00774.74±0.005
HSIC[49]深色78.76±0.00579.34±0.00678.69±0.0060.81±0.0528.06±0.6518.20±0.72018.68±0.05215.46±0.06318.03±0.066
浅色73.89±0.00274.95±0.00374.15±0.002
总体75.35±0.00376.32±0.00475.56±0.003
MFD[50]深色82.72±0.00482.87±0.00381.87±0.0020.90±0.0459.34±0.3089.38±0.33412.73±0.0435.89±0.03310.06±0.034
浅色76.10±0.00276.96±0.00176.12±0.002
总体78.02±0.00178.81±0.00177.94±0.001
ME[51]深色80.71±0.00280.71±0.00280.32±0.0020.95±0.01211.25±0.32411.77±0.3495.40±0.011−16.52±0.033−16.11±0.034
浅色73.49±0.00274.25±0.00273.67±0.001
总体75.59±0.00176.20±0.00175.68±0.001
FairAdaBN[17]深色77.50±0.00678.67±0.00577.05±0.0050.66±0.0787.10±0.2027.37±0.09028.75±0.07121.25±0.02322.18±0.005
浅色70.26±0.00572.23±0.00469.89±0.002
总体72.36±0.00574.24±0.00472.14±0.003
QP-Net[33]深色79.13±0.00677.84±0.00477.96±0.0040.74±0.0608.41±1.1898.66±1.14624.25±0.05610.68±0.12312.34±0.113
浅色72.46±0.00371.61±0.00671.73±0.004
总体74.40±0.00273.56±0.00373.66±0.002
SCP[35]深色80.96±0.00181.54±0.00480.73±0.0020.76±0.0438.45±0.8308.85±0.81325.41±0.05313.71±0.07313.91±0.068
浅色75.30±0.01376.22±0.01675.15±0.012
总体77.06±0.00977.88±0.01276.92±0.008
DNE[36]深色80.65±0.00379.91±0.00180.06±0.0020.69±0.1078.82±0.4969.26±0.54430.40±0.1058.44±0.0478.47±0.050
浅色73.97±0.00373.82±0.00773.81±0.005
总体75.91±0.00375.72±0.00575.74±0.004
FairMoE[52]深色75.99±0.01976.25±0.02075.52±0.0230.79±0.0267.68±0.3127.86±0.34415.45±0.03214.74±0.01816.78±0.017
浅色70.17±0.01170.45±0.01669.69±0.019
总体71.83±0.01372.26±0.01871.49±0.020
EBID-Net
(本文)
深色78.52±0.00277.51±0.00577.65±0.0040.59±0.1076.92±0.1567.22±0.21039.43±0.10826.96±0.01227.44±0.019
浅色74.26±0.00374.75±0.00574.37±0.004
总体75.38±0.00375.61±0.00575.37±0.004
), ArticleFig(id=1329435697192219075, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=表1, caption=

不同方法在Fitzpatrick17k数据集上的诊断性能、公平性及权衡结果

, figureFileSmall=null, figureFileBig=null, tableContent=
方法肤色诊断性能 /%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
ResNet18深色82.86±0.01382.97±0.01281.80±0.0131.03±0.1019.88±1.46810.38±1.7910.00±0.0000.00±0.0000.00±0.000
浅色75.46±0.00376.64±0.00174.94±0.003
总体77.68±0.00378.62±0.00477.13±0.002
Resampling[32]深色83.38±0.00983.60±0.00882.63±0.0070.88±0.16711.01±1.19111.66±1.37015.66±0.155−10.67±0.113−11.59±0.125
浅色76.18±0.00877.16±0.00575.68±0.003
总体78.28±0.00679.17±0.00577.86±0.003
GroupDRO[47]深色80.38±0.00779.87±0.01179.72±0.0080.94±0.1799.23±0.3349.77±0.3986.58±0.1734.74±0.0334.07±0.037
浅色74.54±0.00874.71±0.01174.07±0.009
总体76.25±0.00776.32±0.01175.83±0.008
EnD[48]深色79.90±0.00379.74±0.00379.45±0.0031.05±0.1428.31±1.5678.67±1.753−5.26±0.13312.53±0.16313.19±0.173
浅色72.99±0.00772.71±0.01072.59±0.008
总体75.10±0.00474.91±0.00774.74±0.005
HSIC[49]深色78.76±0.00579.34±0.00678.69±0.0060.81±0.0528.06±0.6518.20±0.72018.68±0.05215.46±0.06318.03±0.066
浅色73.89±0.00274.95±0.00374.15±0.002
总体75.35±0.00376.32±0.00475.56±0.003
MFD[50]深色82.72±0.00482.87±0.00381.87±0.0020.90±0.0459.34±0.3089.38±0.33412.73±0.0435.89±0.03310.06±0.034
浅色76.10±0.00276.96±0.00176.12±0.002
总体78.02±0.00178.81±0.00177.94±0.001
ME[51]深色80.71±0.00280.71±0.00280.32±0.0020.95±0.01211.25±0.32411.77±0.3495.40±0.011−16.52±0.033−16.11±0.034
浅色73.49±0.00274.25±0.00273.67±0.001
总体75.59±0.00176.20±0.00175.68±0.001
FairAdaBN[17]深色77.50±0.00678.67±0.00577.05±0.0050.66±0.0787.10±0.2027.37±0.09028.75±0.07121.25±0.02322.18±0.005
浅色70.26±0.00572.23±0.00469.89±0.002
总体72.36±0.00574.24±0.00472.14±0.003
QP-Net[33]深色79.13±0.00677.84±0.00477.96±0.0040.74±0.0608.41±1.1898.66±1.14624.25±0.05610.68±0.12312.34±0.113
浅色72.46±0.00371.61±0.00671.73±0.004
总体74.40±0.00273.56±0.00373.66±0.002
SCP[35]深色80.96±0.00181.54±0.00480.73±0.0020.76±0.0438.45±0.8308.85±0.81325.41±0.05313.71±0.07313.91±0.068
浅色75.30±0.01376.22±0.01675.15±0.012
总体77.06±0.00977.88±0.01276.92±0.008
DNE[36]深色80.65±0.00379.91±0.00180.06±0.0020.69±0.1078.82±0.4969.26±0.54430.40±0.1058.44±0.0478.47±0.050
浅色73.97±0.00373.82±0.00773.81±0.005
总体75.91±0.00375.72±0.00575.74±0.004
FairMoE[52]深色75.99±0.01976.25±0.02075.52±0.0230.79±0.0267.68±0.3127.86±0.34415.45±0.03214.74±0.01816.78±0.017
浅色70.17±0.01170.45±0.01669.69±0.019
总体71.83±0.01372.26±0.01871.49±0.020
EBID-Net
(本文)
深色78.52±0.00277.51±0.00577.65±0.0040.59±0.1076.92±0.1567.22±0.21039.43±0.10826.96±0.01227.44±0.019
浅色74.26±0.00374.75±0.00574.37±0.004
总体75.38±0.00375.61±0.00575.37±0.004
), ArticleFig(id=1329435697267716548, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Table 2, caption=

Results of Diagnostic Performance, Fairness and Trade-off of Various Methods on ISIC 2019 Dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
方法性别诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
ResNet18男性80.61±0.00580.76±0.00680.36±0.0060.81±0.0255.73±0.7725.50±0.9360.00±0.000.00±0.000.00±0.00
女性83.02±0.00583.19±0.00582.92±0.005
总体81.69±0.00581.91±0.00681.58±0.006
Resampling[32]男性82.02±0.00182.30±0.00181.99±0.0010.87±0.0465.86±0.5485.73±0.475−5.77±0.057−1.10±0.096−2.89±0.087
女性83.47±0.00183.76±0.00283.36±0.001
总体82.69±0.00182.99±0.00182.65±0.001
GroupDRO[47]男性80.02±0.01079.64±0.01479.45±0.0120.78±0.0636.08±0.4365.90±0.7283.17±0.085−6.99±0.071−8.15±0.129
女性81.99±0.00881.84±0.01081.70±0.008
总体80.92±0.00980.68±0.01280.53±0.010
EnD[48]男性78.90±0.00478.76±0.00778.58±0.0070.86±0.0335.18±0.4544.71±0.493−8.45±0.0377.73±0.07412.44±0.085
女性81.69±0.00581.38±0.00981.26±0.008
总体80.17±0.00480.00±0.00779.85±0.007
HSIC[49]男性80.18±0.00380.52±0.00380.11±0.0030.64±0.0615.22±0.2575.34±0.12220.07±0.0317.51±0.0411.52±0.018
女性81.14±0.00481.45±0.00481.10±0.004
总体80.60±0.00380.96±0.00480.59±0.004
MFD[50]男性79.48±0.00779.55±0.00979.30±0.0070.72±0.0225.15±0.4355.10±0.3829.59±0.0318.78±0.0795.69±0.075
女性81.64±0.00281.70±0.00381.61±0.002
总体80.45±0.00480.57±0.00680.41±0.005
ME[51]男性80.53±0.00580.45±0.00680.40±0.0050.59±0.0084.63±0.4554.34±0.61026.13±0.00918.11±0.07820.06±0.110
女性81.33±0.00481.14±0.00681.15±0.005
总体80.85±0.00480.78±0.00680.75±0.005
FairAdaBN[17]男性76.52±0.00577.14±0.00576.57±0.0050.60±0.0054.25±0.2503.87±0.20820.98±0.00420.53±0.0424.28±0.036
女性78.37±0.00378.91±0.00378.33±0.003
总体77.32±0.00277.98±0.00377.41±0.002
QP-Net[33]男性78.64±0.01178.08±0.01678.17±0.0140.86±0.0634.38±0.0654.26±0.115−10.00±0.07320.09±0.00919.13±0.010
女性79.40±0.01178.74±0.01978.94±0.017
总体78.90±0.01178.39±0.01778.50±0.015
SCP[35]男性80.65±0.00180.77±0.00180.59±0.0010.65±0.0454.27±0.1364.03±0.16919.27±0.05625.35±0.02226.66±0.029
女性82.80±0.00583.00±0.00782.82±0.005
总体81.63±0.00281.83±0.00481.65±0.003
DNE[36]男性80.39±0.00280.62±0.00280.44±0.0020.60±0.0244.10±0.0984.39±0.13915.18±0.03127.64±0.01619.37±0.024
女性81.88±0.00182.01±0.00181.89±0.001
总体81.08±0.00181.28±0.00181.14±0.001
FairMoE[52]男性81.62±0.00281.70±0.00381.21±0.0030.60±0.0423.81±0.2143.82±0.36126.84±0.05234.01±0.03731.11±0.064
女性82.71±0.00382.84±0.00282.36±0.002
总体82.10±0.00282.24±0.00181.76±0.002
EBID-Net
(本文)
男性80.17±0.00480.16±0.00580.05±0.0050.56±0.02413.65±0.37663.50±0.44129.47±0.03134.85±0.06634.97±0.080
女性81.06±0.00281.13±0.00381.00±0.003
总体80.56±0.00380.62±0.00380.50±0.003
), ArticleFig(id=1329435697351602629, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=表2, caption=

不同方法在ISIC 2019数据集上的诊断性能、公平性及权衡结果

, figureFileSmall=null, figureFileBig=null, tableContent=
方法性别诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
ResNet18男性80.61±0.00580.76±0.00680.36±0.0060.81±0.0255.73±0.7725.50±0.9360.00±0.000.00±0.000.00±0.00
女性83.02±0.00583.19±0.00582.92±0.005
总体81.69±0.00581.91±0.00681.58±0.006
Resampling[32]男性82.02±0.00182.30±0.00181.99±0.0010.87±0.0465.86±0.5485.73±0.475−5.77±0.057−1.10±0.096−2.89±0.087
女性83.47±0.00183.76±0.00283.36±0.001
总体82.69±0.00182.99±0.00182.65±0.001
GroupDRO[47]男性80.02±0.01079.64±0.01479.45±0.0120.78±0.0636.08±0.4365.90±0.7283.17±0.085−6.99±0.071−8.15±0.129
女性81.99±0.00881.84±0.01081.70±0.008
总体80.92±0.00980.68±0.01280.53±0.010
EnD[48]男性78.90±0.00478.76±0.00778.58±0.0070.86±0.0335.18±0.4544.71±0.493−8.45±0.0377.73±0.07412.44±0.085
女性81.69±0.00581.38±0.00981.26±0.008
总体80.17±0.00480.00±0.00779.85±0.007
HSIC[49]男性80.18±0.00380.52±0.00380.11±0.0030.64±0.0615.22±0.2575.34±0.12220.07±0.0317.51±0.0411.52±0.018
女性81.14±0.00481.45±0.00481.10±0.004
总体80.60±0.00380.96±0.00480.59±0.004
MFD[50]男性79.48±0.00779.55±0.00979.30±0.0070.72±0.0225.15±0.4355.10±0.3829.59±0.0318.78±0.0795.69±0.075
女性81.64±0.00281.70±0.00381.61±0.002
总体80.45±0.00480.57±0.00680.41±0.005
ME[51]男性80.53±0.00580.45±0.00680.40±0.0050.59±0.0084.63±0.4554.34±0.61026.13±0.00918.11±0.07820.06±0.110
女性81.33±0.00481.14±0.00681.15±0.005
总体80.85±0.00480.78±0.00680.75±0.005
FairAdaBN[17]男性76.52±0.00577.14±0.00576.57±0.0050.60±0.0054.25±0.2503.87±0.20820.98±0.00420.53±0.0424.28±0.036
女性78.37±0.00378.91±0.00378.33±0.003
总体77.32±0.00277.98±0.00377.41±0.002
QP-Net[33]男性78.64±0.01178.08±0.01678.17±0.0140.86±0.0634.38±0.0654.26±0.115−10.00±0.07320.09±0.00919.13±0.010
女性79.40±0.01178.74±0.01978.94±0.017
总体78.90±0.01178.39±0.01778.50±0.015
SCP[35]男性80.65±0.00180.77±0.00180.59±0.0010.65±0.0454.27±0.1364.03±0.16919.27±0.05625.35±0.02226.66±0.029
女性82.80±0.00583.00±0.00782.82±0.005
总体81.63±0.00281.83±0.00481.65±0.003
DNE[36]男性80.39±0.00280.62±0.00280.44±0.0020.60±0.0244.10±0.0984.39±0.13915.18±0.03127.64±0.01619.37±0.024
女性81.88±0.00182.01±0.00181.89±0.001
总体81.08±0.00181.28±0.00181.14±0.001
FairMoE[52]男性81.62±0.00281.70±0.00381.21±0.0030.60±0.0423.81±0.2143.82±0.36126.84±0.05234.01±0.03731.11±0.064
女性82.71±0.00382.84±0.00282.36±0.002
总体82.10±0.00282.24±0.00181.76±0.002
EBID-Net
(本文)
男性80.17±0.00480.16±0.00580.05±0.0050.56±0.02413.65±0.37663.50±0.44129.47±0.03134.85±0.06634.97±0.080
女性81.06±0.00281.13±0.00381.00±0.003
总体80.56±0.00380.62±0.00380.50±0.003
), ArticleFig(id=1329435697443877318, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Table 3, caption=

Impact of Training Order and Attention Mechanism on Diagnostic Performance, Fairness and Trade-off on Fitzpatrick17k Dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
训练顺序注意力机制肤色诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
深色—浅色w/o深色77.3175.8576.300.737.277.3124.9922.2825.44
浅色73.4173.7973.46
总体74.4774.4374.32
w/ ConAtt深色77.5377.0577.090.687.177.2831.2124.6627.09
浅色74.9375.9275.11
总体75.5376.2775.72
w/ EConAtt深色78.2976.9577.290.626.726.9336.4828.6629.91
浅色73.9574.1073.94
总体75.1074.9974.95
浅色—深色w/o深色81.0081.5481.031.3312.1713.32−33.97−28.03−33.17
浅色70.8272.1571.29
总体73.9175.0974.33
w/ ConAtt深色81.9082.4481.941.3711.2012.22−37.42−17.77−22.14
浅色70.9171.9771.33
总体74.2575.2474.64
w/ EConAtt深色80.4181.0480.400.9811.7412.460.91−22.76−23.97
浅色72.0473.1172.33
总体74.6275.5874.87
), ArticleFig(id=1329435697523569095, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=表3, caption=

Fitzpatrick17k数据集上训练顺序与注意力机制对诊断性能、公平性及权衡关系的影响

, figureFileSmall=null, figureFileBig=null, tableContent=
训练顺序注意力机制肤色诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
深色—浅色w/o深色77.3175.8576.300.737.277.3124.9922.2825.44
浅色73.4173.7973.46
总体74.4774.4374.32
w/ ConAtt深色77.5377.0577.090.687.177.2831.2124.6627.09
浅色74.9375.9275.11
总体75.5376.2775.72
w/ EConAtt深色78.2976.9577.290.626.726.9336.4828.6629.91
浅色73.9574.1073.94
总体75.1074.9974.95
浅色—深色w/o深色81.0081.5481.031.3312.1713.32−33.97−28.03−33.17
浅色70.8272.1571.29
总体73.9175.0974.33
w/ ConAtt深色81.9082.4481.941.3711.2012.22−37.42−17.77−22.14
浅色70.9171.9771.33
总体74.2575.2474.64
w/ EConAtt深色80.4181.0480.400.9811.7412.460.91−22.76−23.97
浅色72.0473.1172.33
总体74.6275.5874.87
), ArticleFig(id=1329435697594872264, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Table 4, caption=

Impact of Training Order and Attention Mechanism on Diagnostic Performance, Fairness and Trade-off on ISIC 2019 Dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
训练顺序注意力机制性别诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
女性—男性w/o男性79.8679.4680.460.673.993.8515.0828.1627.79
女性80.4680.8280.41
总体79.8980.3179.92
w/ ConAtt男性79.7579.9779.700.573.883.7127.9030.5630.81
女性80.9781.2080.94
总体80.2880.5580.29
w/ EConAtt男性80.7380.8180.680.563.573.5229.9736.8035.10
女性81.2681.4181.24
总体80.9681.0980.95
男性—女性w/o男性78.0778.1077.990.925.415.62−16.902.25−5.51
女性80.0480.0280.01
总体78.9779.0178.95
w/ ConAtt男性78.8979.2878.910.865.155.25−8.577.722.14
女性80.7881.1680.88
总体79.7380.1779.85
w/ EConAtt男性78.3878.9478.510.9244.855.26−15.9412.992.00
女性81.3481.7581.39
总体79.7680.2779.88
), ArticleFig(id=1329435697670369737, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=表4, caption=

ISIC 2019数据集上训练顺序与注意力机制对诊断性能、公平性及权衡关系的影响

, figureFileSmall=null, figureFileBig=null, tableContent=
训练顺序注意力机制性别诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
女性—男性w/o男性79.8679.4680.460.673.993.8515.0828.1627.79
女性80.4680.8280.41
总体79.8980.3179.92
w/ ConAtt男性79.7579.9779.700.573.883.7127.9030.5630.81
女性80.9781.2080.94
总体80.2880.5580.29
w/ EConAtt男性80.7380.8180.680.563.573.5229.9736.8035.10
女性81.2681.4181.24
总体80.9681.0980.95
男性—女性w/o男性78.0778.1077.990.925.415.62−16.902.25−5.51
女性80.0480.0280.01
总体78.9779.0178.95
w/ ConAtt男性78.8979.2878.910.865.155.25−8.577.722.14
女性80.7881.1680.88
总体79.7380.1779.85
w/ EConAtt男性78.3878.9478.510.9244.855.26−15.9412.992.00
女性81.3481.7581.39
总体79.7680.2779.88
), ArticleFig(id=1329435697741672906, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=EN, label=Table 5, caption=

Ablation Study Results of Balanced Updating Strategy

, figureFileSmall=null, figureFileBig=null, tableContent=
数据集平衡更新属性诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
Fitzpatrick17kw/o深色76.5573.8574.671.1511.5011.59−17.17−21.91−17.18
浅色72.2670.0270.78
总体73.3971.2271.98
w/深色78.2976.9577.290.626.726.9336.4828.6629.91
浅色73.9574.1073.94
总体75.1074.9974.95
ISIC 2019w/o男性79.3579.5179.340.796.215.570.44−10.39−3.29
女性81.1381.3781.03
总体80.0480.3980.13
w/男性80.7380.8180.680.563.573.5229.9736.8035.10
女性81.2681.4181.24
总体80.9681.0980.95
), ArticleFig(id=1329435697812976075, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, language=CN, label=表5, caption=

平衡更新策略的消融实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
数据集平衡更新属性诊断性能/%公平性权衡
精确率↑召回率↑F1值↑EOpp0↓EOpp1↓EOdd↓$ FAT{E}_{EOpp0} $↑$ FAT{E}_{EOpp1} $↑$ FAT{E}_{EOdd} $↑
Fitzpatrick17kw/o深色76.5573.8574.671.1511.5011.59−17.17−21.91−17.18
浅色72.2670.0270.78
总体73.3971.2271.98
w/深色78.2976.9577.290.626.726.9336.4828.6629.91
浅色73.9574.1073.94
总体75.1074.9974.95
ISIC 2019w/o男性79.3579.5179.340.796.215.570.44−10.39−3.29
女性81.1381.3781.03
总体80.0480.3980.13
w/男性80.7380.8180.680.563.573.5229.9736.8035.10
女性81.2681.4181.24
总体80.9681.0980.95
)], 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/issn1000-1239.202550799, detailUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/10.7544/issn1000-1239.202550799, pdfUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/PDF/10.7544/issn1000-1239.202550799, pdfUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/PDF/10.7544/issn1000-1239.202550799, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1793462400000, fullTextJson=null, articleText=null, reference=null)
收藏切换
面向皮肤病诊断去偏的增强型平衡增量蒸馏网络
罗义琴 1 , 古天龙 1, * , 郝峰锐 1 , 常亮 2
计算机研究与发展 | 人工智能 2026,63(11): 3017-3031
收起
收藏切换
计算机研究与发展 |人工智能 2026 , 63 (11) : 3017 -3031
面向皮肤病诊断去偏的增强型平衡增量蒸馏网络
全屏
[Author(id=1329435692301660569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=lyq_ah@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329435692372963739, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692301660569, language=EN, stringName=Yiqin Luo, firstName=Yiqin, middleName=null, lastName=Luo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632, bio={"content":"

born in 1991. PhD. Student member of CCF. Her main research interests include fair machine learning and trustworthy artificial intelligence

"}, bioImg=null, bioContent=

born in 1991. PhD. Student member of CCF. Her main research interests include fair machine learning and trustworthy artificial intelligence

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435692431683996, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692301660569, language=CN, stringName=罗义琴, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632, bio={"img":"+ZkvzXHgExEW4Po6XK/h5w==","content":"

1991年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

"}, bioImg=+ZkvzXHgExEW4Po6XK/h5w==, bioContent=

1991年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692154859922, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=1, ext=[AuthorCompanyExt(id=1329435692163248531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632), AuthorCompanyExt(id=1329435692171637140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632)])]), Author(id=1329435692511375774, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=gutianlong@jnu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329435692578484640, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692511375774, language=EN, stringName=Tianlong Gu, firstName=Tianlong, middleName=null, lastName=Gu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632, bio={"content":"

born in 1964. PhD, professor. Senior member of CCF. His main research interests include formal methods, trustworthy artificial intelligence, artificial intelligence safety, and artificial intelligence ethics

"}, bioImg=null, bioContent=

born in 1964. PhD, professor. Senior member of CCF. His main research interests include formal methods, trustworthy artificial intelligence, artificial intelligence safety, and artificial intelligence ethics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435692641399201, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692511375774, language=CN, stringName=古天龙, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632, bio={"img":"XvnoLojK7lVwQFaeJXIIfA==","content":"

1964年生。博士,教授。CCF高级会员。主要研究方向为形式化方法、可信人工智能、人工智能安全、人工智能伦理

"}, bioImg=XvnoLojK7lVwQFaeJXIIfA==, bioContent=

1964年生。博士,教授。CCF高级会员。主要研究方向为形式化方法、可信人工智能、人工智能安全、人工智能伦理

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692154859922, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=1, ext=[AuthorCompanyExt(id=1329435692163248531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632), AuthorCompanyExt(id=1329435692171637140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632)])]), Author(id=1329435692700119459, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, 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=1329435692771422629, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692700119459, language=EN, stringName=Fengrui Hao, firstName=Fengrui, middleName=null, lastName=Hao, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632, bio={"content":"

born in 1995. PhD. Student member of CCF. His main research interests include fair machine learning and trustworthy artificial intelligence

"}, bioImg=null, bioContent=

born in 1995. PhD. Student member of CCF. His main research interests include fair machine learning and trustworthy artificial intelligence

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435692825948582, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692700119459, language=CN, stringName=郝峰锐, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632, bio={"img":"9eDGRdBEJGiYnrSrPyLWRA==","content":"

1995年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

"}, bioImg=9eDGRdBEJGiYnrSrPyLWRA==, bioContent=

1995年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692154859922, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=1, ext=[AuthorCompanyExt(id=1329435692163248531, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632), AuthorCompanyExt(id=1329435692171637140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692154859922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1可信人工智能教育部工程研究中心(暨南大学) 广州 510632)])]), Author(id=1329435692893057448, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, orderNo=3, 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=1329435692960166314, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692893057448, language=EN, stringName=Liang Chang, firstName=Liang, middleName=null, lastName=Chang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Guangxi Key Laboratory of Trusted Software (Guilin University of Electronic Technology), Guilin, Guangxi 541004, bio={"content":"

born in 1980. PhD, professor. Senior member of CCF. His main research interests include data and knowledge engineering, trustworthy artificial intelligence, artificial intelligence ethics, and data privacy protection

"}, bioImg=null, bioContent=

born in 1980. PhD, professor. Senior member of CCF. His main research interests include data and knowledge engineering, trustworthy artificial intelligence, artificial intelligence ethics, and data privacy protection

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329435693027275179, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, authorId=1329435692893057448, language=CN, stringName=常亮, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2广西可信软件重点实验室(桂林电子科技大学) 广西桂林 541004, bio={"img":"qqLoq37wAqHAa+xqVRjTlQ==","content":"

1980年生。博士,教授。CCF高级会员。主要研究方向为数据与知识工程、可信人工智能、人工智能伦理、数据隐私保护

"}, bioImg=qqLoq37wAqHAa+xqVRjTlQ==, bioContent=

1980年生。博士,教授。CCF高级会员。主要研究方向为数据与知识工程、可信人工智能、人工智能伦理、数据隐私保护

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329435692238746005, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, xref=2, ext=[AuthorCompanyExt(id=1329435692247134614, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692238746005, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Guangxi Key Laboratory of Trusted Software (Guilin University of Electronic Technology), Guilin, Guangxi 541004), AuthorCompanyExt(id=1329435692255523223, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329435688237379963, companyId=1329435692238746005, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2广西可信软件重点实验室(桂林电子科技大学) 广西桂林 541004)])])]
罗义琴1 , 古天龙1, * , 郝峰锐1, 常亮2
作者信息
  • 1可信人工智能教育部工程研究中心(暨南大学) 广州 510632
  • 2广西可信软件重点实验室(桂林电子科技大学) 广西桂林 541004
通讯作者:
古天龙()
作者简介:

1991年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

1964年生。博士,教授。CCF高级会员。主要研究方向为形式化方法、可信人工智能、人工智能安全、人工智能伦理

1995年生。博士。CCF学生会员。主要研究方向为公平机器学习、可信人工智能

1980年生。博士,教授。CCF高级会员。主要研究方向为数据与知识工程、可信人工智能、人工智能伦理、数据隐私保护

Enhanced Balanced Incremental Distillation Network for Debiased Dermatological Disease Diagnosis
Yiqin Luo1 , Tianlong Gu1, * , Fengrui Hao1, Liang Chang2
Affiliations
  • 1Engineering Research Center of Trustworthy AI (Jinan University), Ministry of Education, Guangzhou 510632
  • 2Guangxi Key Laboratory of Trusted Software (Guilin University of Electronic Technology), Guilin, Guangxi 541004
  • born in 1991. PhD. Student member of CCF. Her main research interests include fair machine learning and trustworthy artificial intelligence

    born in 1964. PhD, professor. Senior member of CCF. His main research interests include formal methods, trustworthy artificial intelligence, artificial intelligence safety, and artificial intelligence ethics

    born in 1995. PhD. Student member of CCF. His main research interests include fair machine learning and trustworthy artificial intelligence

    born in 1980. PhD, professor. Senior member of CCF. His main research interests include data and knowledge engineering, trustworthy artificial intelligence, artificial intelligence ethics, and data privacy protection

出版时间: 2026-11-01 doi: 10.7544/issn1000-1239.202550799
文章导航
收藏切换

深度学习技术日益赋能皮肤病诊断应用,使其诊断性能稳步提升。然而,这类数据驱动的深度学习模型存在严重的偏见问题。若将存在偏见的系统部署于临床应用,势必会对特定群体(如非代表性群体)造成不公平对待。确保决策的公平性已成为一个至关重要且亟待解决的问题。现有研究表明,提升公平性往往以降低诊断性能为代价。因此,为了在诊断性能和公平性之间取得更优权衡,同时减轻对非代表性群体的不公平对待,提出一个增强型平衡增量蒸馏网络(enhanced balanced incremental distillation network,EBID-Net)。具体而言,该模型借助平衡记忆内存整合不同群体的分布开展增量训练,促使代表性群体辅助非代表性群体学习知识。此外,将全局信息融入上下文注意力机制,捕捉不同空间位置特征间的交互关系,从而在增量学习过程中获得更稳健的特征表示。实验结果显示,EBID-Net在公平性以及诊断性能和公平性间的权衡方面均优于其他方法。

公平  /  皮肤病  /  增量学习  /  权衡  /  注意力机制

Deep learning technology has been increasingly empowering the practice of dermatological disease diagnosis, leading to a steady improvement in its diagnostic performance. However, such data-driven deep learning models are plagued by severe bias issues. If systems with these biases are deployed in clinical applications, they are bound to cause unfair treatment towards specific groups, such as underrepresented groups. Ensuring the fairness of decision-making has become a critical and urgent problem that needs to be addressed. Existing research has indicated that enhancing fairness often comes at the cost of reduced diagnostic performance. Therefore, to achieve a better trade-off between diagnostic performance and fairness, while simultaneously alleviating the unfair treatment of underrepresented groups, we propose an enhanced balanced incremental distillation network (EBID-Net). Specifically, with the aid of balanced memory, the model leverages representative demographic groups to assist underrepresented groups in knowledge learning, while being incrementally trained by integrating distributions across different groups. Additionally, we incorporate global information into the contextual attention mechanism to capture correlated interactions between features across different spatial locations, thereby obtaining more robust feature representations during incremental learning. Experimental results show that EBID-Net outperforms other methods in fairness criteria and in the trade-off between fairness and diagnostic performance.

fairness  /  dermatological disease  /  incremental learning  /  trade-off  /  attention mechanism
罗义琴, 古天龙, 郝峰锐, 常亮. 面向皮肤病诊断去偏的增强型平衡增量蒸馏网络. 计算机研究与发展, 2026 , 63 (11) : 3017 -3031 . DOI: 10.7544/issn1000-1239.202550799
Yiqin Luo, Tianlong Gu, Fengrui Hao, Liang Chang. Enhanced Balanced Incremental Distillation Network for Debiased Dermatological Disease Diagnosis[J]. Journal of Computer Research and Development, 2026 , 63 (11) : 3017 -3031 . DOI: 10.7544/issn1000-1239.202550799
皮肤病是全球最常见的健康问题之一。然而,皮肤病的准确诊断仍颇具挑战性。该诊断过程需要整合多种视觉线索,如病变外观、大小分布、颜色、鳞屑特征和排列等[1]。此外,皮肤病在全球范围内的发病率日益攀升,加之皮肤科医生短缺,促使研究人员探索更高效的诊断方法[2]。人工智能辅助诊断在提升全科医生诊断能力方面展现出巨大潜力[3-4]。例如,Liu等人[5]开发了一个深度学习系统,该系统能够诊断26种常见皮肤病,其准确率可与获得认证的皮肤科医生相媲美。在一项真实世界临床试验中,Papachristou等人[6]证明,人工智能驱动的决策支持工具在检测黑色素瘤方面具有较高的诊断性能,表明其对初级保健医生具有重要的临床价值。此外,Huang等人[7]还成功开发了一种用于评估实际临床环境中银屑病严重程度的模型。
尽管这类数据驱动的深度学习模型前景广阔,但高度依赖数据质量,因而极易受到数据偏见的影响[8-10]。研究人员发现,这些模型在疾病分类时往往会过度关注人口统计学特征(如肤色和性别)[11]。这种倾向可能导致模型出现偏见,具体表现为不同人口统计学亚组间的诊断性能存在显著差异[12-13]。在临床应用中部署此类存在偏见的系统,将不可避免地对某些特定人口统计学群体造成不公平对待。鉴于医疗应用中对公平性的迫切需求以及医学数据的独特性,探索有效策略以减轻医疗人工智能中的此类偏见显得至关重要。本研究聚焦于人工智能辅助医学图像诊断,特别是皮肤病诊断中的偏见缓解问题。
收集平衡且具有代表性的数据是防止偏见的最有效方法,但在实际操作中,这仍是最具挑战性的任务之一。为了在数据分布不平衡的情况下保证公平性,研究人员在机器学习生命周期的不同阶段提出了各种方法[9,14]。这些方法的核心原则是从数据分布的内在不平衡中学习并适应。具体而言,其实现途径为调整数据分布以实现平衡,或者优化算法本身。事实上,域增量学习也遵循这一原则:不断融入新信息,同时保留先前获取的知识,从而促进不同域的平衡学习[15]。
基于此,本文提出一个面向皮肤病诊断的公平域增量学习网络EBID-Net(enhanced balanced incremental distillation network)。该网络结合重放、知识蒸馏以及扩展的上下文注意力机制实现不同域的平衡学习。在增量学习过程中,随着新信息的融入,模型会逐渐丢失已学习的旧知识[16]。为应对该挑战,EBID-Net利用旧模型作为教师模型来提取关键响应信息,帮助新模型在学习新知识的同时,最大限度降低遗忘旧知识的风险。然而,由于提取的响应信息中新旧数据域之间存在不平衡,这一策略可能会引入额外的偏见。针对该问题,本文首先将一半记忆内存用于存储域平衡样本以缓解旧知识的遗忘。其次,将剩余记忆内存用于存储目标类平衡样本,并结合统计均等差异约束,以缓解目标类在不同域之间的预测偏见。研究表明,公平增强方法在提升模型公平性的过程中,往往会导致模型诊断性能下降[17]。为缓解这一问题,本文引入上下文注意力(contextual attention,ConAtt)机制[18],以增强模型对判别性特征的捕获能力。然而,ConAtt更侧重于局部信息交互。鉴于全局信息与局部信息在模型决策过程中均具有重要作用[19-20],本文将全局信息融入ConAtt,进一步保证模型性能。本研究的主要贡献如下:
1)提出了EBID-Net,该网络基于感知的数据分布变化来平衡不同群体的学习。
2)设计了一个平衡记忆组件,其由类平衡记忆模块和域平衡记忆模块组成,并集成差异知识蒸馏模块以促进知识迁移并强化公平性约束。
3)扩展了ConAtt机制,将全局信息融入局部静态上下文模型中以捕捉不同空间位置特征间的交互。
4)在2个皮肤病学数据集上的实验表明,EBID-Net在确保公平性的同时,在诊断性能和公平性之间实现了更好的权衡。
域增量学习是增量学习的一种重要应用场景[21],致力于在任务恒定的情况下,应对样本属性分布的动态变化[15]。增量学习面临的主要挑战之一是灾难性遗忘,即深度神经网络在获取新知识时容易遗忘已学的知识,导致难以识别旧域中的样本[22]。为解决这一问题,研究人员提出了3种主要策略:正则化[23]、重放[24-25]以及网络扩展[26]。
正则化方法通过限制权重变化来防止旧知识的过度遗忘;重放方法利用少量旧数据巩固模型记忆;网络扩展方法则通过动态调整网络架构并增加参数来适应新知识,从而避免遗忘。在正则化方法中,域增量学习已被证明能有效平衡不同领域数据的学习过程[15]。基于此,本研究探索了基于重放机制的方法在缓解皮肤病诊断不公平性方面的应用。
在机器学习算法的整个生命周期中,预处理、处理中及后处理方法均能用于增强模型决策的公平性[14,27-28]。
预处理方法倾向于修改数据以防止预测模型中潜在的歧视。例如,通过数据集组合[29]构建公平数据集;利用生成对抗网络[30]或草图模型[31]生成额外的图像;直接对训练集进行重采样[32]。不过,鉴于医疗数据的稀缺性和珍贵性,此类方法需要投入大量精力来确保数据的质量和多样性。
处理中方法尝试通过修改和改变学习算法来增强模型训练过程中的公平性[17,33]。FairAdaBN(fair adaptive batch normalization)[17]通过共享部分网络参数动态整合来自子群体的共同信息。QP-Net(Quasi-Pareto net)[33]基于准帕累托改进方法,结合多任务学习和域自适应,提升模型在非代表性群体中的表现。与之不同,本文旨在利用代表性子群体的学习能力来增强模型公平性。
后处理方法作为补救策略,通过对已训练模型或待预测数据实施特定处理,以消除训练中可能存在的潜在偏见,以及训练过程中可能引入的不公平性因素[14]。剪枝策略是一种有效的改进方法,在医疗应用中更受青睐[34-35]。例如,SCP(sensitive channel pruning)[35]提出了基于软最近邻损失的通道剪枝框架,该框架通过选择性修剪对敏感属性敏感的信道实现公平性。此外,还存在其他优化策略,如DNE(debiased noise editing)[36]生成噪声来掩盖医学图像中敏感属性与疾病标签间的虚假相关性,从而降低模型对敏感属性的依赖。
在域增量学习过程中,数据分批次输入模型,每个批次对应新域或新域组。模型依据当前批次输入的数据训练,完成训练后再进入下一阶段。新域指当前批次的数据,旧域则涵盖已学习的所有数据。训练集与测试集的数据域同步更新,模型训练后需在混合测试集上评估,确保既学习新域知识又不遗忘旧域知识。本文中涉及的域由敏感属性(或人口统计属性)定义。对于每个属性,数据被划分为不同的域组。例如,性别划分为男性域和女性域。
设定一个医学成像数据集$ S $,其中包含$ N $个样本。数据集由3部分组成:输入图像$ X $、敏感属性$ A $以及真实病症标签$ Y $。本文的核心目标是在确保模型诊断性能的同时,极力缩小不同属性域在性能上的差距。
EBID-Net的架构如图1所示,主要由特征提取模块(ResNet和扩展的上下文注意力机制)、平衡记忆更新模块(balanced memory updating module,BMU-Module)和差异知识蒸馏模块(disparity knowledge distillation module,DKD-Module)组成。每一个训练阶段,学生模型结合新域样本和内存中的旧域样本协同训练。教师模型仅提供学生模型训练过程中需要的关于旧域的知识而不更新参数。每一阶段训练结束后平衡更新内存,并复制当前模型作为下一阶段的教师模型。需要注意的是,在初始的第1阶段,记忆内存为空,无教师模型。
注意力机制能够从图像中提取关键特征,进而保障模型性能[37]。在传统自注意力机制里,所有成对的查询-键(query-key)关系均是在孤立状态下独立学习的。为探索这些关系中丰富的上下文信息,ConAtt机制利用输入键之间的上下文信息引导注意力矩阵的学习,以此增强基于2维特征图的自注意力学习在视觉表征学习中的效能[18]。ConAtt在计算过程大量采用1×1卷积,这使其更倾向于捕捉局部空间关系。然而,在模型决策过程中,全局信息同样重要。因此,为了增强全局-局部交互,本文扩展了ConAtt,将全局信息引入局部静态上下文建模过程,形成了扩展的上下文注意力(extended contextual attention,EConAtt),如图2所示。
在EConAtt中,输入特征为$ {\boldsymbol{X}}_{\mathrm{in}}\in {\mathbb{R}}^{H\times W\times C} $,并定义了3个变量:键(${\boldsymbol{K}} $)、查询(${\boldsymbol{Q}} $)和值(${\boldsymbol{V}} $)($ \boldsymbol{K}={\boldsymbol{X}}_{\mathrm{in}} $;$ \boldsymbol{Q}= {\boldsymbol{X}}_{\mathrm{in}} $;$ \boldsymbol{V}={\boldsymbol{X}}_{\mathrm{in}}{\boldsymbol{W}}_{\mathrm{v}} $,其中的$ {\boldsymbol{W}}_{\mathrm{v}} $为嵌入矩阵)。首先,对$ \boldsymbol{K} $应用一个$ k\times k $的分组卷积,以提取包含局部上下文信息的特征表示$ {\boldsymbol{K}}^{1} $,这与局部信息的静态建模类似。
接下来,拼接$ {\boldsymbol{K}}^{1} $与输入特征$ {\boldsymbol{X}}_{\mathrm{in}} $的全局平均池化(global average pooling,GAP)的输出,从而引入全局特征$ \boldsymbol{G} $。随后,将$ \boldsymbol{Q} $与$ {\boldsymbol{K}}^{1} $的拼接结果与全局特征$ \boldsymbol{G} $进行逐元素相加并经过2个连续的卷积层得到2维注意力图。然后,将得到的注意力图与$ \boldsymbol{V} $相乘,获得动态上下文建模特征图。最后,通过融合局部静态上下文建模与全局静态上下文建模,生成输出特征图。
$ \boldsymbol{G}=Concat\left(GAP\left(\boldsymbol{K}{\boldsymbol{W}}_{\mathrm{k}}\right),GAP\left({\boldsymbol{X}}_{\mathrm{in}}\right)\right) \;, $
$ \boldsymbol{D}=Addition\left(Concat\left(\boldsymbol{K}{\boldsymbol{W}}_{\mathrm{k}},\boldsymbol{Q}\right),\boldsymbol{G}\right){\boldsymbol{W}}_{\boldsymbol{\theta }}{\boldsymbol{W}}_{\boldsymbol{\delta }} \;, $
$ {\boldsymbol{X}}_{\text{out}}=Fusion(\boldsymbol{D},\boldsymbol{V}\circledast \boldsymbol{D}) \;, $
其中$ {\boldsymbol{W}}_{\mathrm{k}} $,$ {\boldsymbol{W}}_{\boldsymbol{\theta }} $,$ {\boldsymbol{W}}_{\boldsymbol{\delta }} $均为嵌入矩阵。
在增量学习中,为防止灾难性遗忘,部分研究引入记忆内存组件[38]以保留少量旧样本用于新域的学习[39]。由于记忆内存中存储的旧域样本数量有限,因此在当前训练阶段中,旧域训练样本数量会远少于新域训练样本。为减轻因训练数据分布不平衡而产生的额外偏见,通常会在训练阶段结束时应用平衡微调组件[38]。该组件采用平衡采样技术生成一个平衡的子集,并使用较小的学习率对模型进行微调。
本文整合增量记忆内存组件与平衡微调组件的优势,提出了一种平衡记忆内存组件设计方案。考虑到数据集中敏感属性域和目标类别均存在分布不平衡的情况,采用平衡记忆内存组件由内存大小相同的目标类别平衡记忆模块$ {M}^{\mathrm{ta}} $和敏感属性域平衡记忆模块$ {M}^{\mathrm{s}\text{e}} $组成。该设计旨在实现更为平衡且有效的模型训练,从而提高模型的公平性和诊断性能。
$ {M}^{\text{ta}}=\left\{\left(x_{i}^{\mathrm{ta}},a_{i}^{\mathrm{ta}},y_{i}^{\mathrm{ta}}\right)_{i}^{B}\right\},\left| {X}_{y}\right| =B/2|U| \;, $
$ {M}^{\text{se}}=\left\{\left(x_{i}^{\text{se}},a_{i}^{\text{se}},y_{i}^{\mathrm{se}}\right)_{i}^{B}\right\},\left| {X}_{a}\right| =B/2|R| \;, $
其中$ {X}_{y} $表示目标类别为$ y $的子集,$ {X}_{a} $表示属性为$ a $的子集,$ B $为记忆容量大小,$ U $和$ R $分别对应目标类别集合与敏感属性域集合。
算法1. 平衡记忆内存更新。
输入:第i个学习阶段$ {T}_{i} $,阶段$ {T}_{i} $新域数据集$ X{B}_{i} $,阶段$ {T}_{i} $记忆内存中的旧域数据集$ X{M}_{i} $,目标类别数量$ |U| $,记忆内存大小$ B $,阶段$ {T}_{i} $旧域数量$ {R}_{i-1} $和新域数量$ {R}_{i} $,阶段$ {T}_{i} $目标类别平衡记忆内存$ M_{i}^{\mathrm{ta}} $,$ {T}_{i} $阶段敏感属性域平衡记忆内存$ M_{i}^{\mathrm{se}} $;
输出:样本集$ X{M}_{i+1} $。
① for $ y $ in $ U $
②  $ XB_{i}^{y}\leftarrow 选择样本(X{B}_{i},y) $;
③  $ M_{i}^{\mathrm{ta},y}\leftarrow 选择样本(M_{i}^{\mathrm{ta}},y) $;
④  $ {n}_{\mathrm{ta}}=(B/2|U|) $;
⑤  $ count=|M_{i}^{\mathrm{ta},y}| $;
⑥ for 元素 in $ XB_{i}^{y} $
⑦   $ count++ $;
⑧  以概率$ \min \;(1,{n}_{\mathrm{ta}}/count) $将元素存入$ M_{i}^{\mathrm{ta},y} $ (蓄水池采样);
⑨ end for
⑩ end for
⑪ $ M_{i+1}^{\mathrm{ta}}=M_{i}^{\mathrm{ta},0}\cup M_{i}^{\mathrm{ta},2}\cup … \cup M_{i}^{\mathrm{ta},(|U|-1)} $;
⑫ for $ a $ in $ {R}_{i-1}\cup {R}_{i} $
⑬  $ {n}_{\mathrm{se}}=B/2(|{R}_{i-1}|+|{R}_{i}|) $;
⑭ if $ {a}\in {R}_{i-1} $
⑮   $ M_{i}^{\mathrm{se},a}=\{x\in M_{i}^{\mathrm{se}}|域(x)==a\} $;
⑯   $ M_{i}^{\mathrm{se},a}\leftarrow 随机采样(M_{i}^{\mathrm{se},a},{n}_{\mathrm{se}}) $;
⑰ else
⑱   $ XB_{i}^{a}=\{x\in X{B}_{i}|域(x)==a\} $;
⑲   $ M_{i}^{\mathrm{se},a}\leftarrow 随机采样(XB_{i}^{a},{n}_{\mathrm{se}}) $;
⑳ end if
㉑ end for
㉒ $ M_{i+1}^{\mathrm{se}}=M_{i}^{\mathrm{se},0}\cup M_{i}^{\mathrm{se},1}\cup … \cup M_{i}^{\mathrm{se},(|{R}_{i-1}|+|{R}_{i}|-1)} $;
㉓ return $ M_{i+1}^{\mathrm{ta}}\cup M_{i+1}^{\mathrm{se}} $。
来自同一类别的样本在特征空间中应呈现出相似的嵌入表示。因此,学生模型产生的响应信息应与教师模型中同一类别的响应信息保持一致。蒸馏损失用于衡量学生模型与教师模型输出分布之间的差异,促使学生模型的输出分布向教师模型对齐。在本文设定中,每个新学习阶段开始时,学生模型与教师模型的参数初始化为相同状态。然而,随着学生模型开始接触并学习新域数据,其参数会迭代更新。为确保学生模型在有效学习新域知识的同时保留旧域知识,本文利用蒸馏损失来缓解遗忘并促进知识迁移。
在此约束中,特别指定输入数据为$ {M}^{\text{se}} $。$ {M}^{\text{se}} $中的数据在已学习域间保持平衡。这为学生模型构建了一个仅包含旧域且数据分布平衡的学习环境,使其能够更好地巩固已学知识。通过这种方式确保学生模型保留对旧知识的记忆,从而促进知识的有效迁移与整合。蒸馏损失的计算公式为:
$ {\mathcal{L}}_{\text{dis}}=-\displaystyle\sum _{i=1}^{N}\displaystyle\sum _{j=1}^{U}{q}_{{\mathrm{t}}}(x_{i,j}^{\prime})\ln \left({q}_{{\mathrm{s}}}(x_{i,j}^{\prime})\right) \;, $
$ {q}_{{\mathrm{t}}}(x_{i,j}^{\prime})=\frac{{\left({q}_{{\mathrm{t}}}({x}_{i,j})\right)}^{\tfrac{1}{T}}}{\displaystyle\sum _{i}{\left({q}_{{\mathrm{t}}}({x}_{i,j})\right)}^{\tfrac{1}{T}}} \;, $
$ {q}_{{\mathrm{s}}}(x_{i,j}^{\prime})=\frac{{\left({q}_{{\mathrm{s}}}({x}_{i,j})\right)}^{\tfrac{1}{T}}}{\displaystyle\sum _{i}{\left({q}_{{\mathrm{s}}}({x}_{i,j})\right)}^{\tfrac{1}{T}}} \;, $
其中$ N $表示样本数量,$ U $表示目标类别数,$ T $表示蒸馏温度系数(默认设置为2),$ {q}_{{\mathrm{t}}}({x}_{i,j}) $和$ {q}_{{\mathrm{s}}}({x}_{i,j}) $分别表示第$ i $样本为$ j $类病症的教师模型和学生模型的预测概率。
本文的主要目标是增强学生模型的公平性。因此,需要保证模型能够公平地对待不同的敏感属性域。统计均等性差异(statistical parity disparity)特指在模型预测为正类(或目标类)的实例中,不同属性群体所占比例的差异[40]。为此,本文引入统计均等性差异损失作为公平性约束条件。该损失函数旨在最小化模型对不同属性群体目标类预测结果的差异。为进一步提升学生模型的学习能力,此约束中的输入数据不仅包含当前域数据,还纳入了来自$ {M}^{\mathrm{ta}} $的数据。$ {M}^{\mathrm{ta}} $中的数据在目标类间平衡。统计均等性差异损失的计算公式为:
$ {\mathcal{L}}_{\mathrm{spd}} = \displaystyle\sum _{y=1}^{U}{\left\|{\mathbb{E}}_{{{\boldsymbol{X}}_{i}}\sim {{S}_{A=0}}}\left| \left(q\left({{\boldsymbol{X}}}_{i}\right)=y\right)-{\mathbb{E}}_{{\boldsymbol{X}_{i}}\sim {{S}_{A=1}}}\right| \left(q\left({{\boldsymbol{X}}}_{i}\right)=y\right)\right\|}^{2}\;。 $
综上所述,EBID-Net的总训练目标表示为
$ \mathcal{L}={\mathcal{L}}_{\mathrm{lsce}}+\alpha {\mathcal{L}}_{\mathrm{dis}}+\beta {\mathcal{L}}_{\mathrm{spd}} \;, $
其中权重因子$ \alpha $和$ \beta $分别用于控制知识迁移与公平性约束的强度。标签平滑[41]是神经网络中常用的一种正则化方法,旨在防止模型过拟合。标签平滑交叉熵损失用于衡量输出期望概率分布与标签平滑后实际概率分布之间的差异。因此,损失函数$ {\mathcal{L}}_{\mathrm{lsce}} $负责整体性能优化,其计算方式为:
$ {\mathcal{L}}_{\mathrm{lsce}}=-\frac{1}{N}\displaystyle\sum _{i=1}^{N}\displaystyle\sum _{j=1}^{U}\tilde{p}({x}_{i,j})\ln \left(\tilde{q}({x}_{i,j})\right) \;, $
$ \tilde{p}({x}_{i,j})=\left\{\begin{aligned}&1-\epsilon ,\;\; \text{if}\;i=j,\\&\epsilon /U-1,\;\;其他。\end{aligned}\right.$
其中$ \tilde{p}({x}_{i,j}) $和$ \tilde{q}({x}_{i,j}) $分别表示经标签平滑处理后的真实概率与预测概率。$ \epsilon $是一个较小的常数(设$\epsilon = 0.01$),用于防止概率出现极端值(即1或0)。
在公平皮肤病诊断领域的研究中,数据集的选择至关重要,其需要具备高质量的图像和属性信息以及广泛的认可度。基于此,本文与现有研究一致,选用了 Fitzpatrick17k[42]与 ISIC 2019[43-45]这2个数据集。
Fitzpatrick17k 数据集是皮肤病诊断领域内重要的通用公开数据集。它包含16577张皮肤病学图像,这些图像涉及9个诊断类别,且每张图像都附有Fitzpatrick皮肤类型标签。皮肤类型从Ⅰ~Ⅵ共分为6个等级,等级越低,肤色越浅。数据处理时,本文保留了带有皮肤类型标签的图像,并将Ⅰ~Ⅲ型皮肤设置为浅色皮肤,Ⅳ~Ⅵ型皮肤设置为深色皮肤。最终,浅色图像与深色图像的比例约为7∶3。值得注意的是,尽管浅色皮肤图像数量较多,但在标准条件下训练的模型,在深色皮肤图像上的准确率反而更高,因此选择深色群体为代表性群体。
ISIC 2019数据集是国际皮肤影像协作组织主办的皮肤影像分析挑战赛的竞赛数据集。它包含25331张皮肤病学图像,涉及8个诊断类别。本文排除了没有性别标签的图像,并将性别作为敏感属性。处理后,女性与男性图像的比例约为4.5∶5.5。使用数据集训练的基准模型在女性图像上表现出更高的准确率,因此选择女性群体为代表性群体。
为了全面评估模型性能,本文从诊断性能、公平性以及诊断性能和公平性的权衡3个维度进行评估。在评估诊断性能时,采用的指标为精确率、召回率和F1值。关于公平指标,本文引入平等机会(equal opportunity,EOpp)[46]和均衡概率(equalized odds,EOdd)[46]。基于实际标签,平等机会考虑了EOpp0和EOpp1这2种情况。
$ EOpp0= |P(\hat{Y}=0\mid Y=0,A=1)-P(\hat{Y}=0\mid Y=0,A=0)|\;,$
$ EOpp1= |P(\hat{Y}=1\mid Y=1,A=1)-P(\hat{Y}=1\mid Y=1,A=0)|\;,$
$ \begin{split} EOdd= &|P(\hat{Y}=1\mid Y=y,A=1)-\\&P(\hat{Y}=1\mid Y=y,A=0)|,\;y\in \{0,1\}\;,\end{split}$
其中$ A $,$ Y $,$ \hat{Y} $分别表示敏感属性、真实标签和预测标签。
为了评估性能和公平性之间的权衡关系,本文采用FATE[17](fairness-accuracy trade-off efficiency)指标。该指标可量化公平性标准化提升与诊断性能降低之间的权衡情况。其公式如下:
$ FAT{E}_{FC}=\frac{AC{C}_{\mathrm{e}}-AC{C}_{\mathrm{b}}}{AC{C}_{\mathrm{b}}}-\lambda \frac{F{C}_{\mathrm{e}}-F{C}_{\mathrm{b}}}{F{C}_{\mathrm{b}}} \;, $
其中$ FC $取值为EOpp0,EOpp1,EOdd中的1种。$ ACC $代表诊断性能,后续采用精确率进行比较分析。下标$ \mathrm{e} $和$ \mathrm{b} $用于区分公平增强模型和基线模型。$ \lambda $是加权因子,用于调整用户预定义的公平性要求。在实验评估中,$ \lambda $设置为1.0。
与文献[17]的处理方式一致,本文将数据集按照 6∶2∶2的比例随机划分为训练集、验证集和测试集。随后,采用与文献[32]相同的数据增强技术,具体包括随机翻转、旋转和缩放操作。图像裁剪为224×224像素大小。本文用ResNet18作为主干网络,并使用AdamW优化器开展训练。其中,批量大小设置为128,学习率设置为1E−4,缓冲区大小设置为600。对于不同的数据集,系数设置有所不同。在Fitzpatrick17k数据集中,$ \alpha $和$ \beta $分别设置为0.6和1。对于ISIC 2019数据集,系数$ \alpha $和$ \beta $分别设置为0.8和1.25。所有实验均在NVIDIA RTX A5000显卡上进行,所有FATE指标以ResNet18作为基线进行评估。
EBID-Net将与基准(ResNet18)和其他公平性增强模型进行比较。具体包括:重采样(resampling)[32],GroupDRO(group distributionally robust optimization)[47],EnD(entangling and disentangling)[48],HSIC(Hilbert-Schmidt norm)[49],MFD(MMD-based fair distillation)[50],ME[51],FairAdaBN[17],QP-Net[33],SCP[35],DNE[36],FairMoE[52]。
1)Fitzpatrick17k数据集上的实验结果
表1展示了在Fitzpatrick17k数据集上的诊断性能、公平性指标及二者的权衡效果。首先,EBID-Net在所有公平性指标上均优于其他对比方法。其次,尽管与最优诊断性能结果相比EBID-Net存在轻微差距,但相较于公平性表现次优的FairAdaBN模型,其在精确率、召回率和F1值方面均有小幅提升。上述结果表明,在该数据集上,公平性的提升未以显著牺牲诊断性能为代价,二者之间的冲突较弱。在诊断性能与公平性的权衡方面,EBID-Net在所有指标上均取得最高值。这些结果表明,EBID-Net在保持诊断性能的同时,有效提升了模型的公平性。
2)ISIC 2019数据集上的实验结果
与Fitzpatrick17k数据集相比,ISIC 2019数据集多了约9000张图像,这增加了模型同时保证诊断性能和公平性的难度。尽管ISIC 2019数据集中敏感属性的分布更为平衡,但如表2所示,各方法在该数据集上的性能表现并不优于在Fitzpatrick17k数据集上的表现。此外,还有多个公平增强模型在公平性指标上表现逊于基础模型。上述现象表明,数据规模的扩大不仅增加了优化难度,也加剧了公平性与诊断性能之间的冲突。在此环境下,EBID-Net取得了最优的公平性表现。FairMoE模型分别在3项公平性指标上获得次优结果。但值得注意的是,EBID-Net的诊断性能略低于多数对比模型。这一现象表明,在该数据集上,公平性与诊断性能之间存在明显的权衡关系:公平性的显著提升以诊断性能的轻微下降为代价。尽管如此,在评估权衡指标时,EBID-Net仍取得了最优结果。
1)训练顺序的影响
实验结果表明,增量学习的有效性与其训练顺序密切相关。如表3和表4所示,采用“深色—浅色”训练顺序时,模型在不同域上的性能表现及平衡效果均优于“浅色—深色”训练顺序。类似地,与“男性—女性”训练顺序相比,采用“女性—男性”训练顺序时,模型的性能及跨域平衡性更佳。在此情况下,深色群体和女性群体可被视为代表性群体。上述规律表明,从代表性群体到非代表性群体的训练顺序,有助于实现公平性与诊断性能的权衡优化。
本文的核心设计理念在于:从代表性群体到非代表性群体的训练过程中学习与决策相关的知识。在数据分布不平衡的数据集中,数据量较大的域群体通常被视为代表性群体。然而,对于Fitzpatrick17k数据集而言,深色图像的实际数量少于浅色图像。因此,域群体是否具有代表性,并不仅仅取决于该域群体的数据量。
2)EConAtt的影响
注意力机制能增强输出特征图的表达能力。如表3和表4 所示,与未使用注意力机制的 EBID-Net 相比,使用注意力机制的 EBID-Net 在整体及跨域诊断性能上均有一定提升。这表明特征表达能力的增强对下游任务具有正向促进作用,且有利于公平性与诊断性能之间权衡。此外,使用EConAtt的 EBID-Net 在“深色—浅色”“女性—男性”训练顺序下,于非代表性域中取得了更优的诊断性能,公平性也得到提升。这证明了引入全局信息对知识迁移的有效性。
3)蒸馏损失的影响
在总目标损失中,蒸馏损失系数$ \alpha $用于控制蒸馏损失对整体损失的贡献。图3(a)~(c)分别展示了在Fitzpatrick17k数据集上,不同$ \alpha $值对诊断性能、公平性及二者权衡效果的影响。实验结果显示,蒸馏损失能够间接影响公平性与诊断性能的权衡关系。随着$ \alpha $逐渐增大至0.6,包括精确率、召回率和F1值在内的诊断性能指标均呈现缓慢同步提升的趋势。公平性指标EOpp0,EOpp1,EOdd也同步改善,并达到最优值。相应地,图3(c)中的权衡性能也达到最佳。此外,图3(d)~(f)展示了在ISIC 2019数据集上的实验结果。随着$ \alpha $逐渐增大至0.8,其变化趋势与Fitzpatrick17k数据集上观察到的趋势一致。
然而,当$ \alpha $继续增大时,网络性能不再提升。这是因为蒸馏损失在网络中的主要作用是缓解旧知识的遗忘。如图3(a)所示,当$ \alpha $从0.6增大至1.4时,模型的诊断性能呈现轻微下降趋势。同时,如图3(b)所示,公平性指标开始急剧下降。此时,联合损失中负责分类的交叉熵损失和负责公平性的统计均等差异损失的贡献均减小,自然导致模型的分类性能和公平性均下降。因此,模型的权衡性能也迅速下降。值得注意的是,ISIC 2019数据集上的实验结果与Fitzpatrick17k数据集上的结果高度一致。这表明,蒸馏损失对模型性能、公平性以及权衡的影响在不同数据集上具有普适性。
4)统计均等差异损失的影响
在网络中,统计均等差异损失的主要作用是确保不同属性类别间的公平性。在总目标损失中,系数$ \beta $用于调节统计均等差异损失对整体损失的影响。对于Fitzpatrick17k数据集,如图4(a)所示,随着系数$ \beta $的变化,模型的诊断性能保持相对稳定。而对于ISIC 2019数据集,如图4(d)所示,诊断性能在$ \beta $增大初期保持相对稳定,随后略有下降。
2个数据集的公平性并未随着$ \beta $的增大而持续改善,而是在达到最优值后呈现先增强后减弱的趋势。因此,统计差异损失所施加的约束虽能提升公平性,但若约束过于严格,则这种提升效果可能受到限制。特别地,与其他2项公平性指标相比,EOpp0在达到最优结果后的下降幅度并不显著。综合考虑诊断性能与公平性,模型的权衡效果如图4(c)(f)所示,呈现先上升后下降的趋势。这验证了公平性与诊断性能之间的权衡关系,当公平约束强度处于合理区间时,二者达到最优权衡,过弱或过强的公平约束均会导致综合性能下降。
5)平衡记忆更新的影响
大多数现有的记忆内存更新策略在数据更新过程中会优先保留代表性样本,这往往导致记忆数据分布出现严重失衡。此类分布失衡现象可能在模型训练阶段引入额外的偏见。如表5所示,采用平衡更新策略后,模型在公平性和权衡效果2方面均有所提升。这些结果表明,通过平衡更新策略约束数据分布,能够有效缓解潜在偏见,进而优化权衡指标。
6)记忆内存大小的影响
本节评估了记忆内存大小对模型性能的影响。内存大小被划分为5个等级,最小为200,最大为1000,间隔为200。
如图5所示,随着内存的持续增大,模型的诊断性能保持相对稳定。这种稳定性可能归因于蒸馏损失对灾难性遗忘问题的有效缓解。在公平性方面,随着内存增大,3项指标的值均呈下降趋势。当内存从600增至1000时,ISIC 2019数据集的公平性指标出现轻微波动。内存的持续增大提供了更多旧属性样本,从而为统计均等差异损失提供了更丰富的信息。因此,增大内存有助于提升公平性。在此过程中,诊断性能与公平性的权衡效果是先提升后趋于稳定。这表明适度的内存大小足以支撑公平性与诊断性能的协同优化,过度增大内存虽能维持公平性,但对综合权衡效果的增益有限。
增量技术确保特定域不会主导EBID-Net,并有效地将已学习的特征迁移至后续训练过程中。如图6所示,在第2阶段结果中,模型对男性和女性受试者的病变区域均实现了精准聚焦,印证了EBID-Net的跨域平衡学习能力。对比男性受试者在第2阶段与第1阶段的注意力可视化结果可知,增量学习后模型对病变区域的关注度明显提升,表明域增量学习机制能够驱动知识从代表性域向非代表性域的有效迁移。
灾难性遗忘是指模型在引入新任务后,其在已学习任务上的性能出现明显下降的现象。对比女性受试者第2阶段与第1阶段的结果可知,增量学习后模型对病变区域的关注度仅有轻微衰减,且对关键区域的注意力保持稳定。这表明EBID-Net有效缓解了灾难性遗忘问题。
本文提出了一个增强型平衡增量蒸馏网络(EBID-Net)以缓解偏见问题。在2个皮肤病学数据集上的实验表明,EBID-Net在提升公平性及权衡方面具有潜力与有效性。EBID-Net的另一优势在于,其能够利用在代表性域中所学的知识提升非代表性域的诊断性能,从而改善在涉及某些属性数据缺失的应用中出现性能不佳的情况。未来工作中,我们将进一步优化模型性能,并在多样化临床场景中验证其有效性,从而为构建公平、高效的医疗人工智能系统贡献力量。
罗义琴提出了算法思路和实验方案,完成实验并撰写论文;古天龙提出指导意见并修改论文;郝峰锐协助完成论文修改;常亮提出指导意见。

参考文献 引证文献
排序方式:
1
Karimkhani C, Dellavalle R P, Coffeng L E, et al. Global skin disease morbidity and mortality: An update from the global burden of disease study 2013[J]. JAMA Dermatology, 2017, 153(5): 406−412
2
医疗大数据应用技术国家工程实验室(中南大学)皮肤疾病大数据工作委员会, 中华医学会皮肤性病学分会皮肤肿瘤研究中心, 中国医师协会皮肤科医师分会皮肤肿瘤亚专业委员会. 皮肤病流行病学研究专家共识[J]. 中华皮肤科杂志, 2020, 53(12): 951−961
National Engineering Laboratory for Medical Big Data Application Technology (Central South University), Skin Disease Big Data Working Committee, Skin Tumor Research Center, Chinese Society of Dermatology, Chinese Medical Association, Sub-specialty Committee of Skin Tumor, Dermatology Branch, Chinese Medical Doctor Association. Expert consensus on the methodology of dermato-epidemiological investigations[J]. Chinese Journal of Dermatology, 2020, 53(12): 951−961 (in Chinese)
3
陈宇聪, 董晓华, 梁泽民, 等. 皮肤病智能辅助诊疗系统的研究与设计[J]. 中国卫生信息管理杂志, 2025, 22(3): 443−449
Chen Yucong, Dong Xiaohua, Liang Zemin, et al. Research and design of intelligent auxiliary diagnosis and treatment system for skin diseases[J]. Chinese Journal of Health Informatics and Management, 2025, 22(3): 443−449 (in Chinese)
4
耿蕾蕾, 许振熙, 纪则轩, 等. 基于拓扑先验和双分支视觉Mamba的皮肤病变分割[J]. 计算机研究与发展, 2026, 63(5): 1290−1304
Geng Leilei, Xu Zhenxi, Ji Zexuan, et al. Skin lesion segmentation based on topological prior and dual-branch vision Mamba[J]. Journal of Computer Research and Development, 2026, 63(5): 1290−1304 (in Chinese)
5
Liu Yuan, Jain A, Eng C, et al. A deep learning system for differential diagnosis of skin diseases[J]. Nature Medicine, 2020, 26(6): 900−908
6
Papachristou P, Söderholm M, Pallon J, et al. Evaluation of an artificial intelligence-based decision support for the detection of cutaneous melanoma in primary care: A prospective real-life clinical trial[J]. British Journal of Dermatology, 2024, 191(1): 125−133
7
Huang Kai, Wu Xian, Li Yixin, et al. Artificial intelligence–based psoriasis severity assessment: Real-world study and application[J]. Journal of Medical Internet Research, 2023, 25: e44932
8
Lin Mingquan, Li Tianhao, Yang Yifan, et al. Improving model fairness in image-based computer-aided diagnosis[J]. Nature Communications, 2023, 14(1): 6261
9
Seth P, Pai A K. Does the fairness of your pre-training hold up? Examining the influence of pre-training techniques on skin tone bias in skin lesion classification[C]//Proc of the 13th IEEE/CVF Winter Conf on Applications of Computer Vision. Piscataway, NJ: IEEE, 2024: 570−577
10
范卓娅, 孟小峰. 算法公平与公平计算[J]. 计算机研究与发展, 2023, 60(9): 2048−2066
Fan Zhuoya, Meng Xiaofeng. Algorithmic fairness and fairness computing[J]. Journal of Computer Research and Development, 2023, 60(9): 2048−2066 (in Chinese)
11
Zong Yongshuo, Yang Yongxin, Hospedales T. MEDFAIR: Benchmarking fairness for medical imaging[J]. arXiv preprint, arXiv: 2210.01725, 2022
12
Burgon A, Zhang Yuhang, Petrick N, et al. Bias amplification to facilitate the systematic evaluation of bias mitigation methods[J]. IEEE Journal of Biomedical and Health Informatics, 2025, 29(2): 1444−1454
13
Abdulqawy A, Sallam E, Elkholy A. UnBias: Unveiling bias implications in deep learning models for healthcare applications[J]. IEEE Journal of Biomedical and Health Informatics, 2025, 29(2): 1464−1471
14
Mehrabi N, Morstatter F, Saxena N, et al. A survey on bias and fairness in machine learning[J]. ACM Computing Surveys, 2021, 54(6): 1−35
15
Churamani N, Kara O, Gunes H. Domain-incremental continual learning for mitigating bias in facial expression and action unit recognition[J]. IEEE Transactions on Affective Computing, 2022, 14(4): 3191−3206
16
Mccloskey M, Cohen N J. Catastrophic Interference in Connectionist Networks: The Sequential Learning Problem[M]//Psychology of Learning and Motivation. Amsterdam: Academic, 1989: 109−165
17
Xu Zikang, Zhao Shang, Quan Quan, et al. FairAdaBN: Mitigating unfairness with adaptive batch normalization and its application to dermatological disease classification[C]//Proc of the 26th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2023: 307−317
18
Li Yehao, Yao Ting, Pan Yingwei, et al. Contextual Transformer networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(2): 1489−1500
19
Tang Xiaoya, Zhang Bodong, Ho M M, et al. Duoformer: Leveraging hierarchical representations by local and global attention vision transformer[J]. arXiv preprint, arXiv: 2506.12982, 2025
20
Chowdary G J, Yin Zhaozheng. Med-former: A Transformer based architecture for medical image classification[C]//Proc of the 27th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2024: 448−457
21
Van de Ven G M, Tuytelaars T, Tolias A S. Three types of incremental learning[J]. Nature Machine Intelligence, 2022, 4(12): 1185−1197
22
韩亚楠, 刘建伟, 罗雄麟. 连续学习研究进展[J]. 计算机研究与发展, 2022, 59(6): 1213−1239
Han Yanan, Liu Jianwei, Luo Xionglin. Research progress of continual learning[J]. Journal of Computer Research and Development, 2022, 59(6): 1213−1239 (in Chinese)
23
Smith J S, Karlinsky L, Gutta V, et al. Coda-prompt: Continual decomposed attention-based prompting for rehearsal-free continual learning[C]//Proc of the 36th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2023: 11909−11919
24
Shen Xiaoyu, Zhang Tao, Wu Hao, et al. An adaptive domain-incremental framework with knowledge replay and domain alignment for specific emitter identification[J]. IEEE Transactions on Information Forensics and Security, 2025, 20: 3519−3533
25
Chen Jinpeng, Cong R, Luo Yuxuan, et al. Replay without saving: Prototype derivation and distribution rebalance for class-incremental semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(6): 4699−4716
26
Weng Zhi, Lu Yongzheng, Zheng Zhiqiang, et al. Few-shot class-incremental learning for classification and object detection: A survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(4): 2924−2945
27
古天龙, 李龙, 常亮, 等. 公平机器学习: 概念, 分析与设计[J]. 计算机学报, 2022, 45(5): 1018−1051
Gu Tianlong, Li Long, Chang Liang, et al. Fair machine learning: Concept, analysis, and design[J]. Chinese Journal of Computers, 2022, 45(5): 1018−1051 (in Chinese)
28
陈晋音, 陈奕芃, 陈一鸣, 等. 面向深度学习的公平性研究综述[J]. 计算机研究与发展, 2021, 58(2): 264−280
Chen Jinyin, Chen Yipeng, Chen Yiming, et al. Fairness research on deep learning[J]. Journal of Computer Research and Development, 2021, 58(2): 264−280 (in Chinese)
29
Seyyed-Kalantari L, Liu Guanxiong, McDermott M, et al. CheXclusion: Fairness gaps in deep chest X-ray classifiers[C]//Proc of the Pacific Symp (BIOCOMPUTING 2021). Singapore: World Scientific, 2020: 232−243
30
Joshi N, Burlina P. AI fairness via domain adaptation[J]. arXiv preprint, arXiv: 2104.01109, 2021
31
Yao Ruichen, Cui Ziteng, Li Xiaoxiao, et al. Improving fairness in image classification via sketching[J]. arXiv preprint, arXiv: 2211.00168, 2022
32
Puyol-Antón E, Ruijsink B, Piechnik S K, et al. Fairness in cardiac MR image analysis: An investigation of bias due to data imbalance in deep learning based segmentation[C]//Proc of the 24th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2021: 413−423
33
Yao Siqiong, Dai Fang, Sun Peng, et al. Enhancing the fairness of AI prediction models by Quasi-Pareto improvement among heterogeneous thyroid nodule population[J]. Nature Communications, 2024, 15(1): 1958
34
Wu Yawen, Zeng Dewen, Xu Xiaowei, et al. Fairprune: Achieving fairness through pruning for dermatological disease diagnosis[C]//Proc of the 25th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2022: 743−753
35
Kong Qingpeng, Chiu Chinghao, Zeng Dewen, et al. Achieving fairness through channel pruning for dermatological disease diagnosis[C]//Proc of the 27th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2024: 24−34
36
Jin Ruinan, Deng Wenlong, Chen Minghui, et al. Debiased noise editing on foundation models for fair medical image classification[C]//Proc of the 27th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2024: 164−174
37
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)
38
Castro F M, Marín-Jiménez M J, Guil N, et al. End-to-end incremental learning[C]//Proc of the 15th European Conf on Computer Vision. Berlin: Springer, 2018: 233−248
39
周大蔚, 汪福运, 叶翰嘉, 等. 基于深度学习的类别增量学习算法综述[J]. 计算机学报, 2023, 46(8): 1577−1605
Zhou Dawei, Wang Fuyun, Ye Hanjia, et al. Deep learning for class-incremental learning: A survey[J]. Chinese Journal of Computers, 2023, 46(8): 1577−1605 (in Chinese)
40
Zemel R, Wu Y, Swersky K, et al. Learning fair representations[C]//Proc of the 21st Int Conf on Machine Learning. New York: PMLR, 2013: 325−333
41
Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]// Proc of the 29th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2016: 2818−2826
42
Groh M, Harris C, Daneshjou R, et al. Towards transparency in dermatology image datasets with skin tone annotations by experts, crowds, and an algorithm[J]. Proceedings of the ACM on Human-Computer Interaction, 2022, 6(CSCW2): 1−26
43
Tschandl P, Rosendahl C, Kittler H. The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions[J]. Scientific Data, 2018, 5(1): 1−9
44
Codella N C F, Gutman D, Celebi M E, et al. Skin lesion analysis toward melanoma detection: A challenge at the 2017 international symposium on biomedical imaging (ISBI), hosted by the international skin imaging collaboration (ISIC)[C]//Proc of the 15th IEEE Int Symp on Biomedical Imaging. Piscataway, NJ: IEEE, 2018: 168−172
45
Combalia M, Codella N C, Rotemberg V, et al. Bcn20000: Dermoscopic lesions in the wild[J]. arXiv preprint, arXiv: 1908.02288, 2019
46
Hardt M, Price E, Srebro N. Equality of opportunity in supervised learning[J/OL]. Advances in Neural Information Processing Systems, 2016[2026-05-23]. https://proceedings.neurips.cc/paper_files/paper/2016/hash/6a9659feb1216f14f7384ba499518b38-Abstract.html
47
Sagawa S, Koh P W, Hashimoto T B, et al. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization[J]. arXiv preprint, arXiv: 1911.08731, 2019
48
Tartaglione E, Barbano C A, Grangetto M. End: Entangling and disentangling deep representations for bias correction[C]//Proc of the 34th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 13508−13517
49
Quadrianto N, Sharmanska V, Thomas O. Discovering fair representations in the data domain[C]//Proc of the 32nd IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 8227−8236
50
Jung S, Lee D, Park T, et al. Fair feature distillation for visual recognition[C]//Proc of the 34th IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 12115−12124
51
Chiu Chinghao, Chung Haowei, Chen Yujen, et al. Toward fairness through fair multi-exit framework for dermatological disease diagnosis[C]//Proc of the 26th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2023: 97−107
52
Xu Gelei, Duan Yuying, Liu Zheyuan, et al. Incorporating rather than eliminating: Achieving fairness for skin disease diagnosis through group-specific experts[C]//Proc of the 28th Int Conf on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2025: 284−294
2026年第63卷第11期
96
46
0
引用本文
BibTeX
文章信息
doi: 10.7544/issn1000-1239.202550799
  • 接收时间:2025-11-03
  • 首发时间:2026-11-17
  • 出版时间:2026-11-01
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-11-03
  • 修回日期:2026-06-04
  • 录用日期:2026-06-22
基金
作者信息
    1可信人工智能教育部工程研究中心(暨南大学) 广州 510632
    2广西可信软件重点实验室(桂林电子科技大学) 广西桂林 541004

通讯作者:

古天龙()
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202550799
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