Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information
Recently, deep metric learning (DML) has achieved great success. Some existing DML methods propose adaptive sample mining strategies, which learn to weight the samples, leading to interesting performance. However, these methods suffer from a small memory (e.g., one training batch), limiting their ef...
id |
ftunivsthongkong:oai:repository.hkust.edu.hk:1783.1-121522 |
---|---|
record_format |
openpolar |
spelling |
ftunivsthongkong:oai:repository.hkust.edu.hk:1783.1-121522 2023-05-15T16:01:28+02:00 Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information Jiang, Xiruo Liu, Sheng Dai, Xili Hu, Guosheng Huang, Xingguo Yao, Yazhou Xie, Guo-Sen Shao, Ling 2022 https://repository.hkust.edu.hk/ir/Record/1783.1-121522 https://doi.org/10.1109/TNNLS.2022.3202571 http://lbdiscover.ust.hk/uresolver?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rfr_id=info:sid/HKUST:SPI&rft.genre=article&rft.issn=2162-237X&rft.volume=&rft.issue=&rft.date=2022&rft.spage=1&rft.aulast=Jiang&rft.aufirst=&rft.atitle=Deep+Metric+Learning+Based+on+Meta-Mining+Strategy+With+Semiglobal+Information&rft.title=IEEE+Transactions+on+Neural+Networks+and+Learning+Systems http://www.scopus.com/record/display.url?eid=2-s2.0-85139440557&origin=inward http://gateway.isiknowledge.com/gateway/Gateway.cgi?GWVersion=2&SrcAuth=LinksAMR&SrcApp=PARTNER_APP&DestLinkType=FullRecord&DestApp=WOS&KeyUT=000859867000001 English eng Institute of Electrical and Electronics Engineers Inc. https://repository.hkust.edu.hk/ir/Record/1783.1-121522 IEEE Transactions on Neural Networks and Learning Systems, 15 September 2022, article number 9893740, p. 1-14 2162-237X 2162-2388 https://doi.org/10.1109/TNNLS.2022.3202571 http://lbdiscover.ust.hk/uresolver?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rfr_id=info:sid/HKUST:SPI&rft.genre=article&rft.issn=2162-237X&rft.volume=&rft.issue=&rft.date=2022&rft.spage=1&rft.aulast=Jiang&rft.aufirst=&rft.atitle=Deep+Metric+Learning+Based+on+Meta-Mining+Strategy+With+Semiglobal+Information&rft.title=IEEE+Transactions+on+Neural+Networks+and+Learning+Systems http://www.scopus.com/record/display.url?eid=2-s2.0-85139440557&origin=inward http://gateway.isiknowledge.com/gateway/Gateway.cgi?GWVersion=2&SrcAuth=LinksAMR&SrcApp=PARTNER_APP&DestLinkType=FullRecord&DestApp=WOS&KeyUT=000859867000001 Adaptation models Deep metric learning (DML) Dictionaries Measurement Meta-learning Neural networks Sample weighting Self-supervised learning Task analysis Training Article 2022 ftunivsthongkong https://doi.org/10.1109/TNNLS.2022.3202571 2022-11-11T01:08:11Z Recently, deep metric learning (DML) has achieved great success. Some existing DML methods propose adaptive sample mining strategies, which learn to weight the samples, leading to interesting performance. However, these methods suffer from a small memory (e.g., one training batch), limiting their efficacy. In this work, we introduce a data-driven method, meta-mining strategy with semiglobal information (MMSI), to apply meta-learning to learn to weight samples during the whole training, leading to an adaptive mining strategy. To introduce richer information than one training batch only, we elaborately take advantage of the validation set of meta-learning by implicitly adding additional validation sample information to training. Furthermore, motivated by the latest self-supervised learning, we introduce a dictionary (memory) that maintains very large and diverse information. Together with the validation set, this dictionary presents much richer information to the training, leading to promising performance. In addition, we propose a new theoretical framework that can formulate pairwise and tripletwise metric learning loss functions in a unified framework. This framework brings new insights to society and facilitates us to generalize our MMSI to many existing DML methods. We conduct extensive experiments on three public datasets, CUB200-2011, Cars-196, and Stanford Online Products (SOP). Results show that our method can achieve the state of the art or very competitive performance. Our source codes have been made available at https://github.com/NUST-Machine-Intelligence-Laboratory/MMSI. IEEE Article in Journal/Newspaper DML The Hong Kong University of Science and Technology: HKUST Institutional Repository IEEE Transactions on Neural Networks and Learning Systems 1 14 |
institution |
Open Polar |
collection |
The Hong Kong University of Science and Technology: HKUST Institutional Repository |
op_collection_id |
ftunivsthongkong |
language |
English |
topic |
Adaptation models Deep metric learning (DML) Dictionaries Measurement Meta-learning Neural networks Sample weighting Self-supervised learning Task analysis Training |
spellingShingle |
Adaptation models Deep metric learning (DML) Dictionaries Measurement Meta-learning Neural networks Sample weighting Self-supervised learning Task analysis Training Jiang, Xiruo Liu, Sheng Dai, Xili Hu, Guosheng Huang, Xingguo Yao, Yazhou Xie, Guo-Sen Shao, Ling Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information |
topic_facet |
Adaptation models Deep metric learning (DML) Dictionaries Measurement Meta-learning Neural networks Sample weighting Self-supervised learning Task analysis Training |
description |
Recently, deep metric learning (DML) has achieved great success. Some existing DML methods propose adaptive sample mining strategies, which learn to weight the samples, leading to interesting performance. However, these methods suffer from a small memory (e.g., one training batch), limiting their efficacy. In this work, we introduce a data-driven method, meta-mining strategy with semiglobal information (MMSI), to apply meta-learning to learn to weight samples during the whole training, leading to an adaptive mining strategy. To introduce richer information than one training batch only, we elaborately take advantage of the validation set of meta-learning by implicitly adding additional validation sample information to training. Furthermore, motivated by the latest self-supervised learning, we introduce a dictionary (memory) that maintains very large and diverse information. Together with the validation set, this dictionary presents much richer information to the training, leading to promising performance. In addition, we propose a new theoretical framework that can formulate pairwise and tripletwise metric learning loss functions in a unified framework. This framework brings new insights to society and facilitates us to generalize our MMSI to many existing DML methods. We conduct extensive experiments on three public datasets, CUB200-2011, Cars-196, and Stanford Online Products (SOP). Results show that our method can achieve the state of the art or very competitive performance. Our source codes have been made available at https://github.com/NUST-Machine-Intelligence-Laboratory/MMSI. IEEE |
format |
Article in Journal/Newspaper |
author |
Jiang, Xiruo Liu, Sheng Dai, Xili Hu, Guosheng Huang, Xingguo Yao, Yazhou Xie, Guo-Sen Shao, Ling |
author_facet |
Jiang, Xiruo Liu, Sheng Dai, Xili Hu, Guosheng Huang, Xingguo Yao, Yazhou Xie, Guo-Sen Shao, Ling |
author_sort |
Jiang, Xiruo |
title |
Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information |
title_short |
Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information |
title_full |
Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information |
title_fullStr |
Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information |
title_full_unstemmed |
Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information |
title_sort |
deep metric learning based on meta-mining strategy with semiglobal information |
publisher |
Institute of Electrical and Electronics Engineers Inc. |
publishDate |
2022 |
url |
https://repository.hkust.edu.hk/ir/Record/1783.1-121522 https://doi.org/10.1109/TNNLS.2022.3202571 http://lbdiscover.ust.hk/uresolver?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rfr_id=info:sid/HKUST:SPI&rft.genre=article&rft.issn=2162-237X&rft.volume=&rft.issue=&rft.date=2022&rft.spage=1&rft.aulast=Jiang&rft.aufirst=&rft.atitle=Deep+Metric+Learning+Based+on+Meta-Mining+Strategy+With+Semiglobal+Information&rft.title=IEEE+Transactions+on+Neural+Networks+and+Learning+Systems http://www.scopus.com/record/display.url?eid=2-s2.0-85139440557&origin=inward http://gateway.isiknowledge.com/gateway/Gateway.cgi?GWVersion=2&SrcAuth=LinksAMR&SrcApp=PARTNER_APP&DestLinkType=FullRecord&DestApp=WOS&KeyUT=000859867000001 |
genre |
DML |
genre_facet |
DML |
op_relation |
https://repository.hkust.edu.hk/ir/Record/1783.1-121522 IEEE Transactions on Neural Networks and Learning Systems, 15 September 2022, article number 9893740, p. 1-14 2162-237X 2162-2388 https://doi.org/10.1109/TNNLS.2022.3202571 http://lbdiscover.ust.hk/uresolver?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rfr_id=info:sid/HKUST:SPI&rft.genre=article&rft.issn=2162-237X&rft.volume=&rft.issue=&rft.date=2022&rft.spage=1&rft.aulast=Jiang&rft.aufirst=&rft.atitle=Deep+Metric+Learning+Based+on+Meta-Mining+Strategy+With+Semiglobal+Information&rft.title=IEEE+Transactions+on+Neural+Networks+and+Learning+Systems http://www.scopus.com/record/display.url?eid=2-s2.0-85139440557&origin=inward http://gateway.isiknowledge.com/gateway/Gateway.cgi?GWVersion=2&SrcAuth=LinksAMR&SrcApp=PARTNER_APP&DestLinkType=FullRecord&DestApp=WOS&KeyUT=000859867000001 |
op_doi |
https://doi.org/10.1109/TNNLS.2022.3202571 |
container_title |
IEEE Transactions on Neural Networks and Learning Systems |
container_start_page |
1 |
op_container_end_page |
14 |
_version_ |
1766397303300554752 |