Deep Distributional Sequence Embeddings Based on a Wasserstein Loss
Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an...
Ausführliche Beschreibung
Autor*in: |
Abdelwahab, Ahmed [verfasserIn] |
---|
Format: |
Artikel |
---|---|
Sprache: |
Englisch |
Erschienen: |
2022 |
---|
Schlagwörter: |
---|
Anmerkung: |
© The Author(s) 2022 |
---|
Übergeordnetes Werk: |
Enthalten in: Neural processing letters - Springer US, 1994, 54(2022), 5 vom: 18. März, Seite 3749-3769 |
---|---|
Übergeordnetes Werk: |
volume:54 ; year:2022 ; number:5 ; day:18 ; month:03 ; pages:3749-3769 |
Links: |
---|
DOI / URN: |
10.1007/s11063-022-10784-y |
---|
Katalog-ID: |
OLC207973489X |
---|
LEADER | 01000caa a22002652 4500 | ||
---|---|---|---|
001 | OLC207973489X | ||
003 | DE-627 | ||
005 | 20230506074941.0 | ||
007 | tu | ||
008 | 221221s2022 xx ||||| 00| ||eng c | ||
024 | 7 | |a 10.1007/s11063-022-10784-y |2 doi | |
035 | |a (DE-627)OLC207973489X | ||
035 | |a (DE-He213)s11063-022-10784-y-p | ||
040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
041 | |a eng | ||
082 | 0 | 4 | |a 000 |q VZ |
100 | 1 | |a Abdelwahab, Ahmed |e verfasserin |4 aut | |
245 | 1 | 0 | |a Deep Distributional Sequence Embeddings Based on a Wasserstein Loss |
264 | 1 | |c 2022 | |
336 | |a Text |b txt |2 rdacontent | ||
337 | |a ohne Hilfsmittel zu benutzen |b n |2 rdamedia | ||
338 | |a Band |b nc |2 rdacarrier | ||
500 | |a © The Author(s) 2022 | ||
520 | |a Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. | ||
650 | 4 | |a Metric learning | |
650 | 4 | |a Sequence embeddings | |
650 | 4 | |a Deep learning | |
700 | 1 | |a Landwehr, Niels |4 aut | |
773 | 0 | 8 | |i Enthalten in |t Neural processing letters |d Springer US, 1994 |g 54(2022), 5 vom: 18. März, Seite 3749-3769 |w (DE-627)198692617 |w (DE-600)1316823-X |w (DE-576)052842762 |x 1370-4621 |7 nnns |
773 | 1 | 8 | |g volume:54 |g year:2022 |g number:5 |g day:18 |g month:03 |g pages:3749-3769 |
856 | 4 | 1 | |u https://doi.org/10.1007/s11063-022-10784-y |z lizenzpflichtig |3 Volltext |
912 | |a GBV_USEFLAG_A | ||
912 | |a SYSFLAG_A | ||
912 | |a GBV_OLC | ||
912 | |a SSG-OLC-PSY | ||
912 | |a SSG-OLC-MAT | ||
951 | |a AR | ||
952 | |d 54 |j 2022 |e 5 |b 18 |c 03 |h 3749-3769 |
author_variant |
a a aa n l nl |
---|---|
matchkey_str |
article:13704621:2022----::epitiuinleunemednsaeo |
hierarchy_sort_str |
2022 |
publishDate |
2022 |
allfields |
10.1007/s11063-022-10784-y doi (DE-627)OLC207973489X (DE-He213)s11063-022-10784-y-p DE-627 ger DE-627 rakwb eng 000 VZ Abdelwahab, Ahmed verfasserin aut Deep Distributional Sequence Embeddings Based on a Wasserstein Loss 2022 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2022 Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. Metric learning Sequence embeddings Deep learning Landwehr, Niels aut Enthalten in Neural processing letters Springer US, 1994 54(2022), 5 vom: 18. März, Seite 3749-3769 (DE-627)198692617 (DE-600)1316823-X (DE-576)052842762 1370-4621 nnns volume:54 year:2022 number:5 day:18 month:03 pages:3749-3769 https://doi.org/10.1007/s11063-022-10784-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-PSY SSG-OLC-MAT AR 54 2022 5 18 03 3749-3769 |
spelling |
10.1007/s11063-022-10784-y doi (DE-627)OLC207973489X (DE-He213)s11063-022-10784-y-p DE-627 ger DE-627 rakwb eng 000 VZ Abdelwahab, Ahmed verfasserin aut Deep Distributional Sequence Embeddings Based on a Wasserstein Loss 2022 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2022 Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. Metric learning Sequence embeddings Deep learning Landwehr, Niels aut Enthalten in Neural processing letters Springer US, 1994 54(2022), 5 vom: 18. März, Seite 3749-3769 (DE-627)198692617 (DE-600)1316823-X (DE-576)052842762 1370-4621 nnns volume:54 year:2022 number:5 day:18 month:03 pages:3749-3769 https://doi.org/10.1007/s11063-022-10784-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-PSY SSG-OLC-MAT AR 54 2022 5 18 03 3749-3769 |
allfields_unstemmed |
10.1007/s11063-022-10784-y doi (DE-627)OLC207973489X (DE-He213)s11063-022-10784-y-p DE-627 ger DE-627 rakwb eng 000 VZ Abdelwahab, Ahmed verfasserin aut Deep Distributional Sequence Embeddings Based on a Wasserstein Loss 2022 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2022 Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. Metric learning Sequence embeddings Deep learning Landwehr, Niels aut Enthalten in Neural processing letters Springer US, 1994 54(2022), 5 vom: 18. März, Seite 3749-3769 (DE-627)198692617 (DE-600)1316823-X (DE-576)052842762 1370-4621 nnns volume:54 year:2022 number:5 day:18 month:03 pages:3749-3769 https://doi.org/10.1007/s11063-022-10784-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-PSY SSG-OLC-MAT AR 54 2022 5 18 03 3749-3769 |
allfieldsGer |
10.1007/s11063-022-10784-y doi (DE-627)OLC207973489X (DE-He213)s11063-022-10784-y-p DE-627 ger DE-627 rakwb eng 000 VZ Abdelwahab, Ahmed verfasserin aut Deep Distributional Sequence Embeddings Based on a Wasserstein Loss 2022 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2022 Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. Metric learning Sequence embeddings Deep learning Landwehr, Niels aut Enthalten in Neural processing letters Springer US, 1994 54(2022), 5 vom: 18. März, Seite 3749-3769 (DE-627)198692617 (DE-600)1316823-X (DE-576)052842762 1370-4621 nnns volume:54 year:2022 number:5 day:18 month:03 pages:3749-3769 https://doi.org/10.1007/s11063-022-10784-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-PSY SSG-OLC-MAT AR 54 2022 5 18 03 3749-3769 |
allfieldsSound |
10.1007/s11063-022-10784-y doi (DE-627)OLC207973489X (DE-He213)s11063-022-10784-y-p DE-627 ger DE-627 rakwb eng 000 VZ Abdelwahab, Ahmed verfasserin aut Deep Distributional Sequence Embeddings Based on a Wasserstein Loss 2022 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2022 Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. Metric learning Sequence embeddings Deep learning Landwehr, Niels aut Enthalten in Neural processing letters Springer US, 1994 54(2022), 5 vom: 18. März, Seite 3749-3769 (DE-627)198692617 (DE-600)1316823-X (DE-576)052842762 1370-4621 nnns volume:54 year:2022 number:5 day:18 month:03 pages:3749-3769 https://doi.org/10.1007/s11063-022-10784-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-PSY SSG-OLC-MAT AR 54 2022 5 18 03 3749-3769 |
language |
English |
source |
Enthalten in Neural processing letters 54(2022), 5 vom: 18. März, Seite 3749-3769 volume:54 year:2022 number:5 day:18 month:03 pages:3749-3769 |
sourceStr |
Enthalten in Neural processing letters 54(2022), 5 vom: 18. März, Seite 3749-3769 volume:54 year:2022 number:5 day:18 month:03 pages:3749-3769 |
format_phy_str_mv |
Article |
institution |
findex.gbv.de |
topic_facet |
Metric learning Sequence embeddings Deep learning |
dewey-raw |
000 |
isfreeaccess_bool |
false |
container_title |
Neural processing letters |
authorswithroles_txt_mv |
Abdelwahab, Ahmed @@aut@@ Landwehr, Niels @@aut@@ |
publishDateDaySort_date |
2022-03-18T00:00:00Z |
hierarchy_top_id |
198692617 |
dewey-sort |
0 |
id |
OLC207973489X |
language_de |
englisch |
fullrecord |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">OLC207973489X</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230506074941.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">221221s2022 xx ||||| 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s11063-022-10784-y</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC207973489X</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s11063-022-10784-y-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">000</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Abdelwahab, Ahmed</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Deep Distributional Sequence Embeddings Based on a Wasserstein Loss</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2022</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© The Author(s) 2022</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Metric learning</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Sequence embeddings</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Deep learning</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Landwehr, Niels</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">Neural processing letters</subfield><subfield code="d">Springer US, 1994</subfield><subfield code="g">54(2022), 5 vom: 18. März, Seite 3749-3769</subfield><subfield code="w">(DE-627)198692617</subfield><subfield code="w">(DE-600)1316823-X</subfield><subfield code="w">(DE-576)052842762</subfield><subfield code="x">1370-4621</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:54</subfield><subfield code="g">year:2022</subfield><subfield code="g">number:5</subfield><subfield code="g">day:18</subfield><subfield code="g">month:03</subfield><subfield code="g">pages:3749-3769</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s11063-022-10784-y</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-PSY</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">54</subfield><subfield code="j">2022</subfield><subfield code="e">5</subfield><subfield code="b">18</subfield><subfield code="c">03</subfield><subfield code="h">3749-3769</subfield></datafield></record></collection>
|
author |
Abdelwahab, Ahmed |
spellingShingle |
Abdelwahab, Ahmed ddc 000 misc Metric learning misc Sequence embeddings misc Deep learning Deep Distributional Sequence Embeddings Based on a Wasserstein Loss |
authorStr |
Abdelwahab, Ahmed |
ppnlink_with_tag_str_mv |
@@773@@(DE-627)198692617 |
format |
Article |
dewey-ones |
000 - Computer science, information & general works |
delete_txt_mv |
keep |
author_role |
aut aut |
collection |
OLC |
remote_str |
false |
illustrated |
Not Illustrated |
issn |
1370-4621 |
topic_title |
000 VZ Deep Distributional Sequence Embeddings Based on a Wasserstein Loss Metric learning Sequence embeddings Deep learning |
topic |
ddc 000 misc Metric learning misc Sequence embeddings misc Deep learning |
topic_unstemmed |
ddc 000 misc Metric learning misc Sequence embeddings misc Deep learning |
topic_browse |
ddc 000 misc Metric learning misc Sequence embeddings misc Deep learning |
format_facet |
Aufsätze Gedruckte Aufsätze |
format_main_str_mv |
Text Zeitschrift/Artikel |
carriertype_str_mv |
nc |
hierarchy_parent_title |
Neural processing letters |
hierarchy_parent_id |
198692617 |
dewey-tens |
000 - Computer science, knowledge & systems |
hierarchy_top_title |
Neural processing letters |
isfreeaccess_txt |
false |
familylinks_str_mv |
(DE-627)198692617 (DE-600)1316823-X (DE-576)052842762 |
title |
Deep Distributional Sequence Embeddings Based on a Wasserstein Loss |
ctrlnum |
(DE-627)OLC207973489X (DE-He213)s11063-022-10784-y-p |
title_full |
Deep Distributional Sequence Embeddings Based on a Wasserstein Loss |
author_sort |
Abdelwahab, Ahmed |
journal |
Neural processing letters |
journalStr |
Neural processing letters |
lang_code |
eng |
isOA_bool |
false |
dewey-hundreds |
000 - Computer science, information & general works |
recordtype |
marc |
publishDateSort |
2022 |
contenttype_str_mv |
txt |
container_start_page |
3749 |
author_browse |
Abdelwahab, Ahmed Landwehr, Niels |
container_volume |
54 |
class |
000 VZ |
format_se |
Aufsätze |
author-letter |
Abdelwahab, Ahmed |
doi_str_mv |
10.1007/s11063-022-10784-y |
dewey-full |
000 |
title_sort |
deep distributional sequence embeddings based on a wasserstein loss |
title_auth |
Deep Distributional Sequence Embeddings Based on a Wasserstein Loss |
abstract |
Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. © The Author(s) 2022 |
abstractGer |
Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. © The Author(s) 2022 |
abstract_unstemmed |
Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions. © The Author(s) 2022 |
collection_details |
GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-PSY SSG-OLC-MAT |
container_issue |
5 |
title_short |
Deep Distributional Sequence Embeddings Based on a Wasserstein Loss |
url |
https://doi.org/10.1007/s11063-022-10784-y |
remote_bool |
false |
author2 |
Landwehr, Niels |
author2Str |
Landwehr, Niels |
ppnlink |
198692617 |
mediatype_str_mv |
n |
isOA_txt |
false |
hochschulschrift_bool |
false |
doi_str |
10.1007/s11063-022-10784-y |
up_date |
2024-07-04T01:56:18.087Z |
_version_ |
1803611724145754112 |
fullrecord_marcxml |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">OLC207973489X</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230506074941.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">221221s2022 xx ||||| 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s11063-022-10784-y</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC207973489X</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s11063-022-10784-y-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">000</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Abdelwahab, Ahmed</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Deep Distributional Sequence Embeddings Based on a Wasserstein Loss</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2022</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© The Author(s) 2022</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a feature vector produced by a deep neural network and Euclidean distance or cosine similarity defines distances between these vectors. This paper studies deep distributional embeddings of sequences, where the embedding of a sequence is given by the distribution of learned deep features across the sequence. The motivation for this is to better capture statistical information about the distribution of patterns within the sequence in the embedding. When embeddings are distributions rather than vectors, measuring distances between embeddings involves comparing their respective distributions. The paper therefore proposes a distance metric based on Wasserstein distances between the distributions and a corresponding loss function for metric learning, which leads to a novel end-to-end trainable embedding model. We empirically observe that distributional embeddings outperform standard vector embeddings and that training with the proposed Wasserstein metric outperforms training with other distance functions.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Metric learning</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Sequence embeddings</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Deep learning</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Landwehr, Niels</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">Neural processing letters</subfield><subfield code="d">Springer US, 1994</subfield><subfield code="g">54(2022), 5 vom: 18. März, Seite 3749-3769</subfield><subfield code="w">(DE-627)198692617</subfield><subfield code="w">(DE-600)1316823-X</subfield><subfield code="w">(DE-576)052842762</subfield><subfield code="x">1370-4621</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:54</subfield><subfield code="g">year:2022</subfield><subfield code="g">number:5</subfield><subfield code="g">day:18</subfield><subfield code="g">month:03</subfield><subfield code="g">pages:3749-3769</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s11063-022-10784-y</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-PSY</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">54</subfield><subfield code="j">2022</subfield><subfield code="e">5</subfield><subfield code="b">18</subfield><subfield code="c">03</subfield><subfield code="h">3749-3769</subfield></datafield></record></collection>
|
score |
7.401272 |