Generalizable Embeddings with Cross-batch Metric Learning ...
IEEE ICIP 2023, Hybrid Event, 8-11 October 2023, Kuala Lumpur, Malaysia ... : Global average pooling (GAP) is a popular component in deep metric learning (DML) for aggregating features. Its effectiveness is often attributed to treating each feature vector as a distinct semantic entity and GAP as a c...
Main Authors: | , |
---|---|
Format: | Article in Journal/Newspaper |
Language: | unknown |
Published: |
IEEE
2023
|
Subjects: | |
Online Access: | https://dx.doi.org/10.17023/72c9-wg50 https://rc.signalprocessingsociety.org/conferences/icip-2023/spsicip23vid0152 |
id |
ftdatacite:10.17023/72c9-wg50 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.17023/72c9-wg50 2024-03-31T07:52:28+00:00 Generalizable Embeddings with Cross-batch Metric Learning ... Yeti Z. Gurbuz A. Aydın Alatan 2023 https://dx.doi.org/10.17023/72c9-wg50 https://rc.signalprocessingsociety.org/conferences/icip-2023/spsicip23vid0152 unknown IEEE Audiovisual article MediaObject 2023 ftdatacite https://doi.org/10.17023/72c9-wg50 2024-03-04T13:42:16Z IEEE ICIP 2023, Hybrid Event, 8-11 October 2023, Kuala Lumpur, Malaysia ... : Global average pooling (GAP) is a popular component in deep metric learning (DML) for aggregating features. Its effectiveness is often attributed to treating each feature vector as a distinct semantic entity and GAP as a combination of them. Albeit substantiated, such an explanation's algorithmic implications to learn generalizable entities to represent unseen classes, a crucial DML goal, remain unclear. To address this, we formulate GAP as a convex combination of learnable prototypes. We then show that the prototype learning can be expressed as a recursive process fitting a linear predictor to a batch of samples. Building on that perspective, we consider two batches of disjoint classes at each iteration and regularize the learning by expressing the samples of a batch with the prototypes that are fitted to the other batch. We validate our approach on 4 popular DML benchmarks. ... Article in Journal/Newspaper DML DataCite Metadata Store (German National Library of Science and Technology) |
institution |
Open Polar |
collection |
DataCite Metadata Store (German National Library of Science and Technology) |
op_collection_id |
ftdatacite |
language |
unknown |
description |
IEEE ICIP 2023, Hybrid Event, 8-11 October 2023, Kuala Lumpur, Malaysia ... : Global average pooling (GAP) is a popular component in deep metric learning (DML) for aggregating features. Its effectiveness is often attributed to treating each feature vector as a distinct semantic entity and GAP as a combination of them. Albeit substantiated, such an explanation's algorithmic implications to learn generalizable entities to represent unseen classes, a crucial DML goal, remain unclear. To address this, we formulate GAP as a convex combination of learnable prototypes. We then show that the prototype learning can be expressed as a recursive process fitting a linear predictor to a batch of samples. Building on that perspective, we consider two batches of disjoint classes at each iteration and regularize the learning by expressing the samples of a batch with the prototypes that are fitted to the other batch. We validate our approach on 4 popular DML benchmarks. ... |
format |
Article in Journal/Newspaper |
author |
Yeti Z. Gurbuz A. Aydın Alatan |
spellingShingle |
Yeti Z. Gurbuz A. Aydın Alatan Generalizable Embeddings with Cross-batch Metric Learning ... |
author_facet |
Yeti Z. Gurbuz A. Aydın Alatan |
author_sort |
Yeti Z. Gurbuz |
title |
Generalizable Embeddings with Cross-batch Metric Learning ... |
title_short |
Generalizable Embeddings with Cross-batch Metric Learning ... |
title_full |
Generalizable Embeddings with Cross-batch Metric Learning ... |
title_fullStr |
Generalizable Embeddings with Cross-batch Metric Learning ... |
title_full_unstemmed |
Generalizable Embeddings with Cross-batch Metric Learning ... |
title_sort |
generalizable embeddings with cross-batch metric learning ... |
publisher |
IEEE |
publishDate |
2023 |
url |
https://dx.doi.org/10.17023/72c9-wg50 https://rc.signalprocessingsociety.org/conferences/icip-2023/spsicip23vid0152 |
genre |
DML |
genre_facet |
DML |
op_doi |
https://doi.org/10.17023/72c9-wg50 |
_version_ |
1795031595090968576 |