CASIA OpenIR  > 学术期刊  > 自动化学报
基于拓扑一致性对抗互学习的知识蒸馏
赖轩; 曲延云; 谢源; 裴玉龙
Source Publication自动化学报
ISSN0254-4156
2023
Volume49Issue:1Pages:102-110
Abstract针对基于互学习的知识蒸馏方法中存在模型只关注教师网络和学生网络的分布差异,而没有考虑其他的约束条件,只关注了结果导向的监督,而缺少过程导向监督的不足,提出了一种拓扑一致性指导的对抗互学习知识蒸馏方法 (Topology-guided adversarial deep mutual learning, TADML).该方法将教师网络和学生网络同时训练,网络之间相互指导学习,不仅采用网络输出的类分布之间的差异,还设计了网络中间特征的拓扑性差异度量.训练过程采用对抗训练,进一步提高教师网络和学生网络的判别性.在分类数据集CIFAR10、CIFAR100和Tiny-ImageNet及行人重识别数据集Market1501上的实验结果表明了TADML的有效性, TADML取得了同类模型压缩方法中最好的效果.
Keyword互学习 生成对抗网络 特征优化 知识蒸馏
DOI10.16383/j.aas.c200665
Citation statistics
Document Type期刊论文
Identifierhttp://ir.ia.ac.cn/handle/173211/56212
Collection学术期刊_自动化学报
Recommended Citation
GB/T 7714
赖轩,曲延云,谢源,等. 基于拓扑一致性对抗互学习的知识蒸馏[J]. 自动化学报,2023,49(1):102-110.
APA 赖轩,曲延云,谢源,&裴玉龙.(2023).基于拓扑一致性对抗互学习的知识蒸馏.自动化学报,49(1),102-110.
MLA 赖轩,et al."基于拓扑一致性对抗互学习的知识蒸馏".自动化学报 49.1(2023):102-110.
Files in This Item: Download All
File Name/Size DocType Version Access License
AAS-CN-2020-0665.pdf(2838KB)期刊论文出版稿开放获取CC BY-NC-SAView Download
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[赖轩]'s Articles
[曲延云]'s Articles
[谢源]'s Articles
Baidu academic
Similar articles in Baidu academic
[赖轩]'s Articles
[曲延云]'s Articles
[谢源]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[赖轩]'s Articles
[曲延云]'s Articles
[谢源]'s Articles
Terms of Use
No data!
Social Bookmark/Share
File name: AAS-CN-2020-0665.pdf
Format: Adobe PDF
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.