Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning
Deep Metric Learning (DML) is a group of techniques that aim to measure the similarity between objects through the neural network. Although the number of DML methods has rapidly increased in recent years, most previous studies cannot effectively handle noisy data, which commonly exists in practical...
Published in: | Proceedings of the AAAI Conference on Artificial Intelligence |
---|---|
Main Authors: | , , |
Format: | Article in Journal/Newspaper |
Language: | English |
Published: |
Association for the Advancement of Artificial Intelligence
2023
|
Subjects: | |
Online Access: | https://ojs.aaai.org/index.php/AAAI/article/view/26324 https://doi.org/10.1609/aaai.v37i9.26324 |
id |
ftjaaai:oai:ojs.aaai.org:article/26324 |
---|---|
record_format |
openpolar |
spelling |
ftjaaai:oai:ojs.aaai.org:article/26324 2023-07-23T04:19:01+02:00 Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning Zhang, Chenkang Luo, Lei Gu, Bin 2023-06-26 application/pdf https://ojs.aaai.org/index.php/AAAI/article/view/26324 https://doi.org/10.1609/aaai.v37i9.26324 eng eng Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI/article/view/26324/26096 https://ojs.aaai.org/index.php/AAAI/article/view/26324 doi:10.1609/aaai.v37i9.26324 Copyright (c) 2023 Association for the Advancement of Artificial Intelligence Proceedings of the AAAI Conference on Artificial Intelligence; Vol. 37 No. 9: AAAI-23 Technical Tracks 9; 11183-11191 2374-3468 2159-5399 ML: Representation Learning ML: Deep Neural Network Algorithms info:eu-repo/semantics/article info:eu-repo/semantics/publishedVersion 2023 ftjaaai https://doi.org/10.1609/aaai.v37i9.26324 2023-07-01T22:51:31Z Deep Metric Learning (DML) is a group of techniques that aim to measure the similarity between objects through the neural network. Although the number of DML methods has rapidly increased in recent years, most previous studies cannot effectively handle noisy data, which commonly exists in practical applications and often leads to serious performance deterioration. To overcome this limitation, in this paper, we build a connection between noisy samples and hard samples in the framework of self-paced learning, and propose a Balanced Self-Paced Metric Learning (BSPML) algorithm with a denoising multi-similarity formulation, where noisy samples are treated as extremely hard samples and adaptively excluded from the model training by sample weighting. Especially, due to the pairwise relationship and a new balance regularization term, the sub-problem w.r.t. sample weights is a nonconvex quadratic function. To efficiently solve this nonconvex quadratic problem, we propose a doubly stochastic projection coordinate gradient algorithm. Importantly, we theoretically prove the convergence not only for the doubly stochastic projection coordinate gradient algorithm, but also for our BSPML algorithm. Experimental results on several standard data sets demonstrate that our BSPML algorithm has better generalization ability and robustness than the state-of-the-art robust DML approaches. Article in Journal/Newspaper DML AAAI Publications (Association for the Advancement of Artificial Intelligence) Proceedings of the AAAI Conference on Artificial Intelligence 37 9 11183 11191 |
institution |
Open Polar |
collection |
AAAI Publications (Association for the Advancement of Artificial Intelligence) |
op_collection_id |
ftjaaai |
language |
English |
topic |
ML: Representation Learning ML: Deep Neural Network Algorithms |
spellingShingle |
ML: Representation Learning ML: Deep Neural Network Algorithms Zhang, Chenkang Luo, Lei Gu, Bin Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning |
topic_facet |
ML: Representation Learning ML: Deep Neural Network Algorithms |
description |
Deep Metric Learning (DML) is a group of techniques that aim to measure the similarity between objects through the neural network. Although the number of DML methods has rapidly increased in recent years, most previous studies cannot effectively handle noisy data, which commonly exists in practical applications and often leads to serious performance deterioration. To overcome this limitation, in this paper, we build a connection between noisy samples and hard samples in the framework of self-paced learning, and propose a Balanced Self-Paced Metric Learning (BSPML) algorithm with a denoising multi-similarity formulation, where noisy samples are treated as extremely hard samples and adaptively excluded from the model training by sample weighting. Especially, due to the pairwise relationship and a new balance regularization term, the sub-problem w.r.t. sample weights is a nonconvex quadratic function. To efficiently solve this nonconvex quadratic problem, we propose a doubly stochastic projection coordinate gradient algorithm. Importantly, we theoretically prove the convergence not only for the doubly stochastic projection coordinate gradient algorithm, but also for our BSPML algorithm. Experimental results on several standard data sets demonstrate that our BSPML algorithm has better generalization ability and robustness than the state-of-the-art robust DML approaches. |
format |
Article in Journal/Newspaper |
author |
Zhang, Chenkang Luo, Lei Gu, Bin |
author_facet |
Zhang, Chenkang Luo, Lei Gu, Bin |
author_sort |
Zhang, Chenkang |
title |
Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning |
title_short |
Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning |
title_full |
Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning |
title_fullStr |
Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning |
title_full_unstemmed |
Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning |
title_sort |
denoising multi-similarity formulation: a self-paced curriculum-driven approach for robust metric learning |
publisher |
Association for the Advancement of Artificial Intelligence |
publishDate |
2023 |
url |
https://ojs.aaai.org/index.php/AAAI/article/view/26324 https://doi.org/10.1609/aaai.v37i9.26324 |
genre |
DML |
genre_facet |
DML |
op_source |
Proceedings of the AAAI Conference on Artificial Intelligence; Vol. 37 No. 9: AAAI-23 Technical Tracks 9; 11183-11191 2374-3468 2159-5399 |
op_relation |
https://ojs.aaai.org/index.php/AAAI/article/view/26324/26096 https://ojs.aaai.org/index.php/AAAI/article/view/26324 doi:10.1609/aaai.v37i9.26324 |
op_rights |
Copyright (c) 2023 Association for the Advancement of Artificial Intelligence |
op_doi |
https://doi.org/10.1609/aaai.v37i9.26324 |
container_title |
Proceedings of the AAAI Conference on Artificial Intelligence |
container_volume |
37 |
container_issue |
9 |
container_start_page |
11183 |
op_container_end_page |
11191 |
_version_ |
1772181764431675392 |