Abstract
Unsupervised deep metric learning (UDML) focuses on learning a semantic representation space using only unlabeled data. This challenging problem requires accurately estimating the similarity between data points, which is used to supervise a deep network. For this purpose, we propose to model the high-dimensional data manifold using a piecewise-linear approximation, with each low-dimensional linear piece approximating the data manifold in a small neighborhood of a point. These neighborhoods are used to estimate similarity between data points. We empirically show that this similarity estimate correlates better with the ground truth than the similarity estimates of current state-of-the-art techniques. We also show that proxies, commonly used in supervised metric learning, can be used to model the piecewise-linear manifold in an unsupervised setting, helping improve performance. Our method outperforms existing unsupervised metric learning approaches on standard zeroshot image retrieval benchmarks.
Original language | English (US) |
---|---|
Pages (from-to) | 269-281 |
Number of pages | 13 |
Journal | Proceedings of Machine Learning Research |
Volume | 234 |
State | Published - 2024 |
Event | 1st Conference on Parsimony and Learning, CPAL 2024 - Hongkong, China Duration: Jan 3 2024 → Jan 6 2024 |
ASJC Scopus subject areas
- Artificial Intelligence
- Software
- Control and Systems Engineering
- Statistics and Probability
Library availability
Related links
Fingerprint
Dive into the research topics of 'Piecewise-Linear Manifolds for Deep Metric Learning'. Together they form a unique fingerprint.
View full fingerprint
Cite this
- APA
- Standard
- Harvard
- Vancouver
- Author
- BIBTEX
- RIS
Bhatnagar, S. (2024). Piecewise-Linear Manifolds for Deep Metric Learning. Proceedings of Machine Learning Research, 234, 269-281.
Piecewise-Linear Manifolds for Deep Metric Learning. / Bhatnagar, Shubhang; Ahuja, Narendra.
In: Proceedings of Machine Learning Research, Vol. 234, 2024, p. 269-281.
Research output: Contribution to journal › Conference article › peer-review
Bhatnagar, S 2024, 'Piecewise-Linear Manifolds for Deep Metric Learning', Proceedings of Machine Learning Research, vol. 234, pp. 269-281.
Bhatnagar S, Ahuja N. Piecewise-Linear Manifolds for Deep Metric Learning. Proceedings of Machine Learning Research. 2024;234:269-281.
Bhatnagar, Shubhang ; Ahuja, Narendra. / Piecewise-Linear Manifolds for Deep Metric Learning. In: Proceedings of Machine Learning Research. 2024 ; Vol. 234. pp. 269-281.
@article{f103560bf1c646b8a7ca2e7daf9d2b2a,
title = "Piecewise-Linear Manifolds for Deep Metric Learning",
abstract = "Unsupervised deep metric learning (UDML) focuses on learning a semantic representation space using only unlabeled data. This challenging problem requires accurately estimating the similarity between data points, which is used to supervise a deep network. For this purpose, we propose to model the high-dimensional data manifold using a piecewise-linear approximation, with each low-dimensional linear piece approximating the data manifold in a small neighborhood of a point. These neighborhoods are used to estimate similarity between data points. We empirically show that this similarity estimate correlates better with the ground truth than the similarity estimates of current state-of-the-art techniques. We also show that proxies, commonly used in supervised metric learning, can be used to model the piecewise-linear manifold in an unsupervised setting, helping improve performance. Our method outperforms existing unsupervised metric learning approaches on standard zeroshot image retrieval benchmarks.",
author = "Shubhang Bhatnagar and Narendra Ahuja",
note = "Publisher Copyright: {\textcopyright} 2024 Proceedings of Machine Learning Research; 1st Conference on Parsimony and Learning, CPAL 2024 ; Conference date: 03-01-2024 Through 06-01-2024",
year = "2024",
language = "English (US)",
volume = "234",
pages = "269--281",
journal = "Proceedings of Machine Learning Research",
issn = "2640-3498",
}
TY - JOUR
T1 - Piecewise-Linear Manifolds for Deep Metric Learning
AU - Bhatnagar, Shubhang
AU - Ahuja, Narendra
N1 - Publisher Copyright:© 2024 Proceedings of Machine Learning Research
PY - 2024
Y1 - 2024
N2 - Unsupervised deep metric learning (UDML) focuses on learning a semantic representation space using only unlabeled data. This challenging problem requires accurately estimating the similarity between data points, which is used to supervise a deep network. For this purpose, we propose to model the high-dimensional data manifold using a piecewise-linear approximation, with each low-dimensional linear piece approximating the data manifold in a small neighborhood of a point. These neighborhoods are used to estimate similarity between data points. We empirically show that this similarity estimate correlates better with the ground truth than the similarity estimates of current state-of-the-art techniques. We also show that proxies, commonly used in supervised metric learning, can be used to model the piecewise-linear manifold in an unsupervised setting, helping improve performance. Our method outperforms existing unsupervised metric learning approaches on standard zeroshot image retrieval benchmarks.
AB - Unsupervised deep metric learning (UDML) focuses on learning a semantic representation space using only unlabeled data. This challenging problem requires accurately estimating the similarity between data points, which is used to supervise a deep network. For this purpose, we propose to model the high-dimensional data manifold using a piecewise-linear approximation, with each low-dimensional linear piece approximating the data manifold in a small neighborhood of a point. These neighborhoods are used to estimate similarity between data points. We empirically show that this similarity estimate correlates better with the ground truth than the similarity estimates of current state-of-the-art techniques. We also show that proxies, commonly used in supervised metric learning, can be used to model the piecewise-linear manifold in an unsupervised setting, helping improve performance. Our method outperforms existing unsupervised metric learning approaches on standard zeroshot image retrieval benchmarks.
UR - http://www.scopus.com/inward/record.url?scp=85183840131&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85183840131&partnerID=8YFLogxK
M3 - Conference article
AN - SCOPUS:85183840131
SN - 2640-3498
VL - 234
SP - 269
EP - 281
JO - Proceedings of Machine Learning Research
JF - Proceedings of Machine Learning Research
T2 - 1st Conference on Parsimony and Learning, CPAL 2024
Y2 - 3 January 2024 through 6 January 2024
ER -