Learning to Detect Open Classes for Universal Domain Adaptation View Full Text


Ontology type: schema:Chapter     


Chapter Info

DATE

2020-11-16

AUTHORS

Bo Fu , Zhangjie Cao , Mingsheng Long , Jianmin Wang

ABSTRACT

Universal domain adaptation (UniDA) transfers knowledge between domains without any constraint on the label sets, extending the applicability of domain adaptation in the wild. In UniDA, both the source and target label sets may hold individual labels not shared by the other domain. A de facto challenge of UniDA is to classify the target examples in the shared classes against the domain shift. A more prominent challenge of UniDA is to mark the target examples in the target-individual label set (open classes) as “unknown”. These two entangled challenges make UniDA a highly under-explored problem. Previous work on UniDA focuses on the classification of data in the shared classes and uses per-class accuracy as the evaluation metric, which is badly biased to the accuracy of shared classes. However, accurately detecting open classes is the mission-critical task to enable real universal domain adaptation. It further turns UniDA problem into a well-established close-set domain adaptation problem. Towards accurate open class detection, we propose Calibrated Multiple Uncertainties (CMU) with a novel transferability measure estimated by a mixture of uncertainty quantities in complementation: entropy, confidence and consistency, defined on conditional probabilities calibrated by a multi-classifier ensemble model. The new transferability measure accurately quantifies the inclination of a target example to the open classes. We also propose a novel evaluation metric called H-score, which emphasizes the importance of both accuracies of the shared classes and the “unknown” class. Empirical results under the UniDA setting show that CMU outperforms the state-of-the-art domain adaptation methods on all the evaluation metrics, especially by a large margin on the H-score. More... »

PAGES

567-583

Book

TITLE

Computer Vision – ECCV 2020

ISBN

978-3-030-58554-9
978-3-030-58555-6

Identifiers

URI

http://scigraph.springernature.com/pub.10.1007/978-3-030-58555-6_34

DOI

http://dx.doi.org/10.1007/978-3-030-58555-6_34

DIMENSIONS

https://app.dimensions.ai/details/publication/pub.1132655784


Indexing Status Check whether this publication has been indexed by Scopus and Web Of Science using the SN Indexing Status Tool
Incoming Citations Browse incoming citations for this publication using opencitations.net

JSON-LD is the canonical representation for SciGraph data.

TIP: You can open this SciGraph record using an external JSON-LD service: JSON-LD Playground Google SDTT

[
  {
    "@context": "https://springernature.github.io/scigraph/jsonld/sgcontext.json", 
    "about": [
      {
        "id": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/08", 
        "inDefinedTermSet": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/", 
        "name": "Information and Computing Sciences", 
        "type": "DefinedTerm"
      }, 
      {
        "id": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/0801", 
        "inDefinedTermSet": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/", 
        "name": "Artificial Intelligence and Image Processing", 
        "type": "DefinedTerm"
      }
    ], 
    "author": [
      {
        "affiliation": {
          "alternateName": "Research Center for Big Data, Tsinghua University, Beijing, China", 
          "id": "http://www.grid.ac/institutes/grid.12527.33", 
          "name": [
            "School of Software, BNRist, Tsinghua University, Beijing, China", 
            "Research Center for Big Data, Tsinghua University, Beijing, China"
          ], 
          "type": "Organization"
        }, 
        "familyName": "Fu", 
        "givenName": "Bo", 
        "id": "sg:person.016233366155.30", 
        "sameAs": [
          "https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.016233366155.30"
        ], 
        "type": "Person"
      }, 
      {
        "affiliation": {
          "alternateName": "Research Center for Big Data, Tsinghua University, Beijing, China", 
          "id": "http://www.grid.ac/institutes/grid.12527.33", 
          "name": [
            "School of Software, BNRist, Tsinghua University, Beijing, China", 
            "Research Center for Big Data, Tsinghua University, Beijing, China"
          ], 
          "type": "Organization"
        }, 
        "familyName": "Cao", 
        "givenName": "Zhangjie", 
        "id": "sg:person.013022415410.38", 
        "sameAs": [
          "https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.013022415410.38"
        ], 
        "type": "Person"
      }, 
      {
        "affiliation": {
          "alternateName": "Research Center for Big Data, Tsinghua University, Beijing, China", 
          "id": "http://www.grid.ac/institutes/grid.12527.33", 
          "name": [
            "School of Software, BNRist, Tsinghua University, Beijing, China", 
            "Research Center for Big Data, Tsinghua University, Beijing, China"
          ], 
          "type": "Organization"
        }, 
        "familyName": "Long", 
        "givenName": "Mingsheng", 
        "id": "sg:person.013417115303.81", 
        "sameAs": [
          "https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.013417115303.81"
        ], 
        "type": "Person"
      }, 
      {
        "affiliation": {
          "alternateName": "Research Center for Big Data, Tsinghua University, Beijing, China", 
          "id": "http://www.grid.ac/institutes/grid.12527.33", 
          "name": [
            "School of Software, BNRist, Tsinghua University, Beijing, China", 
            "Research Center for Big Data, Tsinghua University, Beijing, China"
          ], 
          "type": "Organization"
        }, 
        "familyName": "Wang", 
        "givenName": "Jianmin", 
        "id": "sg:person.012303351315.43", 
        "sameAs": [
          "https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.012303351315.43"
        ], 
        "type": "Person"
      }
    ], 
    "datePublished": "2020-11-16", 
    "datePublishedReg": "2020-11-16", 
    "description": "Universal domain adaptation (UniDA) transfers knowledge between domains without any constraint on the label sets, extending the applicability of domain adaptation in the wild. In UniDA, both the source and target label sets may hold individual labels not shared by the other domain. A de facto challenge of UniDA is to classify the target examples in the shared classes against the domain shift. A more prominent challenge of UniDA is to mark the target examples in the target-individual label set (open classes) as \u201cunknown\u201d. These two entangled challenges make UniDA a highly under-explored problem. Previous work on UniDA focuses on the classification of data in the shared classes and uses per-class accuracy as the evaluation metric, which is badly biased to the accuracy of shared classes. However, accurately detecting open classes is the mission-critical task to enable real universal domain adaptation. It further turns UniDA problem into a well-established close-set domain adaptation problem. Towards accurate open class detection, we propose Calibrated Multiple Uncertainties (CMU) with a novel transferability measure estimated by a mixture of uncertainty quantities in complementation: entropy, confidence and consistency, defined on conditional probabilities calibrated by a multi-classifier ensemble model. The new transferability measure accurately quantifies the inclination of a target example to the open classes. We also propose a novel evaluation metric called H-score, which emphasizes the importance of both accuracies of the shared classes and the \u201cunknown\u201d class. Empirical results under the UniDA setting show that CMU outperforms the state-of-the-art domain adaptation methods on all the evaluation metrics, especially by a large margin on the H-score.", 
    "editor": [
      {
        "familyName": "Vedaldi", 
        "givenName": "Andrea", 
        "type": "Person"
      }, 
      {
        "familyName": "Bischof", 
        "givenName": "Horst", 
        "type": "Person"
      }, 
      {
        "familyName": "Brox", 
        "givenName": "Thomas", 
        "type": "Person"
      }, 
      {
        "familyName": "Frahm", 
        "givenName": "Jan-Michael", 
        "type": "Person"
      }
    ], 
    "genre": "chapter", 
    "id": "sg:pub.10.1007/978-3-030-58555-6_34", 
    "inLanguage": "en", 
    "isAccessibleForFree": false, 
    "isPartOf": {
      "isbn": [
        "978-3-030-58554-9", 
        "978-3-030-58555-6"
      ], 
      "name": "Computer Vision \u2013 ECCV 2020", 
      "type": "Book"
    }, 
    "keywords": [
      "domain adaptation", 
      "label set", 
      "evaluation metrics", 
      "target examples", 
      "art domain adaptation methods", 
      "mission-critical tasks", 
      "domain adaptation problem", 
      "domain adaptation methods", 
      "classification of data", 
      "novel evaluation metric", 
      "domain shift", 
      "class detection", 
      "adaptation method", 
      "ensemble model", 
      "adaptation problem", 
      "large margin", 
      "class accuracy", 
      "individual labels", 
      "uncertainty quantity", 
      "transfer knowledge", 
      "prominent challenges", 
      "metrics", 
      "open class", 
      "previous work", 
      "accuracy", 
      "set", 
      "setting show", 
      "empirical results", 
      "conditional probability", 
      "challenges", 
      "multiple uncertainties", 
      "task", 
      "domain", 
      "example", 
      "classification", 
      "labels", 
      "constraints", 
      "class", 
      "adaptation", 
      "detection", 
      "applicability", 
      "work", 
      "consistency", 
      "knowledge", 
      "uncertainty", 
      "wild", 
      "model", 
      "data", 
      "method", 
      "confidence", 
      "probability", 
      "show", 
      "measures", 
      "results", 
      "state", 
      "source", 
      "importance", 
      "quantity", 
      "scores", 
      "margin", 
      "shift", 
      "inclination", 
      "mixture", 
      "complementation", 
      "problem", 
      "Universal domain adaptation (UniDA) transfers knowledge", 
      "domain adaptation (UniDA) transfers knowledge", 
      "adaptation (UniDA) transfers knowledge", 
      "UniDA", 
      "challenge of UniDA", 
      "target-individual label set", 
      "entangled challenges", 
      "real universal domain adaptation", 
      "universal domain adaptation", 
      "UniDA problem", 
      "close-set domain adaptation problem", 
      "accurate open class detection", 
      "open class detection", 
      "CMU", 
      "novel transferability measure", 
      "transferability measure", 
      "multi-classifier ensemble model", 
      "new transferability measure", 
      "UniDA setting show"
    ], 
    "name": "Learning to Detect Open Classes for Universal Domain Adaptation", 
    "pagination": "567-583", 
    "productId": [
      {
        "name": "dimensions_id", 
        "type": "PropertyValue", 
        "value": [
          "pub.1132655784"
        ]
      }, 
      {
        "name": "doi", 
        "type": "PropertyValue", 
        "value": [
          "10.1007/978-3-030-58555-6_34"
        ]
      }
    ], 
    "publisher": {
      "name": "Springer Nature", 
      "type": "Organisation"
    }, 
    "sameAs": [
      "https://doi.org/10.1007/978-3-030-58555-6_34", 
      "https://app.dimensions.ai/details/publication/pub.1132655784"
    ], 
    "sdDataset": "chapters", 
    "sdDatePublished": "2022-01-01T19:14", 
    "sdLicense": "https://scigraph.springernature.com/explorer/license/", 
    "sdPublisher": {
      "name": "Springer Nature - SN SciGraph project", 
      "type": "Organization"
    }, 
    "sdSource": "s3://com-springernature-scigraph/baseset/20220101/entities/gbq_results/chapter/chapter_244.jsonl", 
    "type": "Chapter", 
    "url": "https://doi.org/10.1007/978-3-030-58555-6_34"
  }
]
 

Download the RDF metadata as:  json-ld nt turtle xml License info

HOW TO GET THIS DATA PROGRAMMATICALLY:

JSON-LD is a popular format for linked data which is fully compatible with JSON.

curl -H 'Accept: application/ld+json' 'https://scigraph.springernature.com/pub.10.1007/978-3-030-58555-6_34'

N-Triples is a line-based linked data format ideal for batch operations.

curl -H 'Accept: application/n-triples' 'https://scigraph.springernature.com/pub.10.1007/978-3-030-58555-6_34'

Turtle is a human-readable linked data format.

curl -H 'Accept: text/turtle' 'https://scigraph.springernature.com/pub.10.1007/978-3-030-58555-6_34'

RDF/XML is a standard XML format for linked data.

curl -H 'Accept: application/rdf+xml' 'https://scigraph.springernature.com/pub.10.1007/978-3-030-58555-6_34'


 

This table displays all metadata directly associated to this object as RDF triples.

181 TRIPLES      23 PREDICATES      109 URIs      102 LITERALS      7 BLANK NODES

Subject Predicate Object
1 sg:pub.10.1007/978-3-030-58555-6_34 schema:about anzsrc-for:08
2 anzsrc-for:0801
3 schema:author N664faf57737842bc945296d848994be3
4 schema:datePublished 2020-11-16
5 schema:datePublishedReg 2020-11-16
6 schema:description Universal domain adaptation (UniDA) transfers knowledge between domains without any constraint on the label sets, extending the applicability of domain adaptation in the wild. In UniDA, both the source and target label sets may hold individual labels not shared by the other domain. A de facto challenge of UniDA is to classify the target examples in the shared classes against the domain shift. A more prominent challenge of UniDA is to mark the target examples in the target-individual label set (open classes) as “unknown”. These two entangled challenges make UniDA a highly under-explored problem. Previous work on UniDA focuses on the classification of data in the shared classes and uses per-class accuracy as the evaluation metric, which is badly biased to the accuracy of shared classes. However, accurately detecting open classes is the mission-critical task to enable real universal domain adaptation. It further turns UniDA problem into a well-established close-set domain adaptation problem. Towards accurate open class detection, we propose Calibrated Multiple Uncertainties (CMU) with a novel transferability measure estimated by a mixture of uncertainty quantities in complementation: entropy, confidence and consistency, defined on conditional probabilities calibrated by a multi-classifier ensemble model. The new transferability measure accurately quantifies the inclination of a target example to the open classes. We also propose a novel evaluation metric called H-score, which emphasizes the importance of both accuracies of the shared classes and the “unknown” class. Empirical results under the UniDA setting show that CMU outperforms the state-of-the-art domain adaptation methods on all the evaluation metrics, especially by a large margin on the H-score.
7 schema:editor Nd16488412299445ca6e7746f3f98e3b3
8 schema:genre chapter
9 schema:inLanguage en
10 schema:isAccessibleForFree false
11 schema:isPartOf N0d59038ed87e4550ab5d2fb86232ee0c
12 schema:keywords CMU
13 UniDA
14 UniDA problem
15 UniDA setting show
16 Universal domain adaptation (UniDA) transfers knowledge
17 accuracy
18 accurate open class detection
19 adaptation
20 adaptation (UniDA) transfers knowledge
21 adaptation method
22 adaptation problem
23 applicability
24 art domain adaptation methods
25 challenge of UniDA
26 challenges
27 class
28 class accuracy
29 class detection
30 classification
31 classification of data
32 close-set domain adaptation problem
33 complementation
34 conditional probability
35 confidence
36 consistency
37 constraints
38 data
39 detection
40 domain
41 domain adaptation
42 domain adaptation (UniDA) transfers knowledge
43 domain adaptation methods
44 domain adaptation problem
45 domain shift
46 empirical results
47 ensemble model
48 entangled challenges
49 evaluation metrics
50 example
51 importance
52 inclination
53 individual labels
54 knowledge
55 label set
56 labels
57 large margin
58 margin
59 measures
60 method
61 metrics
62 mission-critical tasks
63 mixture
64 model
65 multi-classifier ensemble model
66 multiple uncertainties
67 new transferability measure
68 novel evaluation metric
69 novel transferability measure
70 open class
71 open class detection
72 previous work
73 probability
74 problem
75 prominent challenges
76 quantity
77 real universal domain adaptation
78 results
79 scores
80 set
81 setting show
82 shift
83 show
84 source
85 state
86 target examples
87 target-individual label set
88 task
89 transfer knowledge
90 transferability measure
91 uncertainty
92 uncertainty quantity
93 universal domain adaptation
94 wild
95 work
96 schema:name Learning to Detect Open Classes for Universal Domain Adaptation
97 schema:pagination 567-583
98 schema:productId N79a2c4f86f594bed90fe497c442eead9
99 N92f96157b90c47dcb6bf27a8953cf195
100 schema:publisher Nb0203082131048048757c64851b6c82d
101 schema:sameAs https://app.dimensions.ai/details/publication/pub.1132655784
102 https://doi.org/10.1007/978-3-030-58555-6_34
103 schema:sdDatePublished 2022-01-01T19:14
104 schema:sdLicense https://scigraph.springernature.com/explorer/license/
105 schema:sdPublisher Ne9a6b0a61d3146adbb9ac486f1a1046a
106 schema:url https://doi.org/10.1007/978-3-030-58555-6_34
107 sgo:license sg:explorer/license/
108 sgo:sdDataset chapters
109 rdf:type schema:Chapter
110 N05d93e7aa35b45c7b1e2725633c44e77 schema:familyName Brox
111 schema:givenName Thomas
112 rdf:type schema:Person
113 N0d59038ed87e4550ab5d2fb86232ee0c schema:isbn 978-3-030-58554-9
114 978-3-030-58555-6
115 schema:name Computer Vision – ECCV 2020
116 rdf:type schema:Book
117 N2eb487a297994bc58184021fe383ebd9 rdf:first Nb7ee003fa95f404fa7d4505c4ad18285
118 rdf:rest rdf:nil
119 N410169f89b494cdd861c59cd608d4d51 rdf:first N05d93e7aa35b45c7b1e2725633c44e77
120 rdf:rest N2eb487a297994bc58184021fe383ebd9
121 N414003f47e774c698a3ed08e7e064833 rdf:first sg:person.012303351315.43
122 rdf:rest rdf:nil
123 N62e838df71724fee806934b18b59e232 schema:familyName Bischof
124 schema:givenName Horst
125 rdf:type schema:Person
126 N664faf57737842bc945296d848994be3 rdf:first sg:person.016233366155.30
127 rdf:rest Nd77a66cb30fc49e8930534260402d42c
128 N79a2c4f86f594bed90fe497c442eead9 schema:name dimensions_id
129 schema:value pub.1132655784
130 rdf:type schema:PropertyValue
131 N92f96157b90c47dcb6bf27a8953cf195 schema:name doi
132 schema:value 10.1007/978-3-030-58555-6_34
133 rdf:type schema:PropertyValue
134 Nb0203082131048048757c64851b6c82d schema:name Springer Nature
135 rdf:type schema:Organisation
136 Nb7ee003fa95f404fa7d4505c4ad18285 schema:familyName Frahm
137 schema:givenName Jan-Michael
138 rdf:type schema:Person
139 Nbb16ba62b203491dbf0583a0934aa388 rdf:first sg:person.013417115303.81
140 rdf:rest N414003f47e774c698a3ed08e7e064833
141 Nd16488412299445ca6e7746f3f98e3b3 rdf:first Nfd246d8b54ba48c18f52025937392f40
142 rdf:rest Ne659db3e972245dc816c81796c7d46ff
143 Nd77a66cb30fc49e8930534260402d42c rdf:first sg:person.013022415410.38
144 rdf:rest Nbb16ba62b203491dbf0583a0934aa388
145 Ne659db3e972245dc816c81796c7d46ff rdf:first N62e838df71724fee806934b18b59e232
146 rdf:rest N410169f89b494cdd861c59cd608d4d51
147 Ne9a6b0a61d3146adbb9ac486f1a1046a schema:name Springer Nature - SN SciGraph project
148 rdf:type schema:Organization
149 Nfd246d8b54ba48c18f52025937392f40 schema:familyName Vedaldi
150 schema:givenName Andrea
151 rdf:type schema:Person
152 anzsrc-for:08 schema:inDefinedTermSet anzsrc-for:
153 schema:name Information and Computing Sciences
154 rdf:type schema:DefinedTerm
155 anzsrc-for:0801 schema:inDefinedTermSet anzsrc-for:
156 schema:name Artificial Intelligence and Image Processing
157 rdf:type schema:DefinedTerm
158 sg:person.012303351315.43 schema:affiliation grid-institutes:grid.12527.33
159 schema:familyName Wang
160 schema:givenName Jianmin
161 schema:sameAs https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.012303351315.43
162 rdf:type schema:Person
163 sg:person.013022415410.38 schema:affiliation grid-institutes:grid.12527.33
164 schema:familyName Cao
165 schema:givenName Zhangjie
166 schema:sameAs https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.013022415410.38
167 rdf:type schema:Person
168 sg:person.013417115303.81 schema:affiliation grid-institutes:grid.12527.33
169 schema:familyName Long
170 schema:givenName Mingsheng
171 schema:sameAs https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.013417115303.81
172 rdf:type schema:Person
173 sg:person.016233366155.30 schema:affiliation grid-institutes:grid.12527.33
174 schema:familyName Fu
175 schema:givenName Bo
176 schema:sameAs https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.016233366155.30
177 rdf:type schema:Person
178 grid-institutes:grid.12527.33 schema:alternateName Research Center for Big Data, Tsinghua University, Beijing, China
179 schema:name Research Center for Big Data, Tsinghua University, Beijing, China
180 School of Software, BNRist, Tsinghua University, Beijing, China
181 rdf:type schema:Organization
 




Preview window. Press ESC to close (or click here)


...