Learning factorized representations for open-set domain adaptation

Mahsa Baktashmotlagh, Masoud Faraki, Tom Drummond, Mathieu Salzmann

Research output: Chapter in Book/Report/Conference proceedingConference PaperOther

Abstract

Domain adaptation for visual recognition has undergone great progress in the past few years. Nevertheless, most existing methods work in the so-called closed-set scenario, assuming that the classes depicted by the target images are exactly the same as those of the source domain. In this paper, we tackle the more challenging, yet more realistic case of open-set domain adaptation, where new, unknown classes can be present in the target data. While, in the unsupervised scenario, one cannot expect to be able to identify each specific new class, we aim to automatically detect which samples belong to these new classes and discard them from the recognition process. To this end, we rely on the intuition that the source and target samples depicting the known classes can be generated by a shared subspace, whereas the target samples from unknown classes come from a different, private subspace. We therefore introduce a framework that factorizes the data into shared and private parts, while encouraging the shared representation to be discriminative. Our experiments on standard benchmarks evidence that our approach
outperforms the state of the art in open-set domain adaptation.
Original languageEnglish
Title of host publicationInternational Conference on Learning Representations 2019
Place of PublicationNew Orleans LA USA
PublisherInternation Conference on Learning Representation (ICLR)
Pages1-11
Number of pages11
Publication statusPublished - 2019
EventInternational Conference on Learning Representations 2019 - New Orleans, United States of America
Duration: 6 May 20199 May 2019
https://iclr.cc/

Conference

ConferenceInternational Conference on Learning Representations 2019
Abbreviated titleICLR 2019
CountryUnited States of America
CityNew Orleans
Period6/05/199/05/19
Internet address

Cite this

Baktashmotlagh, M., Faraki, M., Drummond, T., & Salzmann, M. (2019). Learning factorized representations for open-set domain adaptation. In International Conference on Learning Representations 2019 (pp. 1-11). New Orleans LA USA: Internation Conference on Learning Representation (ICLR).
Baktashmotlagh, Mahsa ; Faraki, Masoud ; Drummond, Tom ; Salzmann, Mathieu. / Learning factorized representations for open-set domain adaptation. International Conference on Learning Representations 2019. New Orleans LA USA : Internation Conference on Learning Representation (ICLR), 2019. pp. 1-11
@inproceedings{bf173695859e439586db578439303540,
title = "Learning factorized representations for open-set domain adaptation",
abstract = "Domain adaptation for visual recognition has undergone great progress in the past few years. Nevertheless, most existing methods work in the so-called closed-set scenario, assuming that the classes depicted by the target images are exactly the same as those of the source domain. In this paper, we tackle the more challenging, yet more realistic case of open-set domain adaptation, where new, unknown classes can be present in the target data. While, in the unsupervised scenario, one cannot expect to be able to identify each specific new class, we aim to automatically detect which samples belong to these new classes and discard them from the recognition process. To this end, we rely on the intuition that the source and target samples depicting the known classes can be generated by a shared subspace, whereas the target samples from unknown classes come from a different, private subspace. We therefore introduce a framework that factorizes the data into shared and private parts, while encouraging the shared representation to be discriminative. Our experiments on standard benchmarks evidence that our approachoutperforms the state of the art in open-set domain adaptation.",
author = "Mahsa Baktashmotlagh and Masoud Faraki and Tom Drummond and Mathieu Salzmann",
year = "2019",
language = "English",
pages = "1--11",
booktitle = "International Conference on Learning Representations 2019",
publisher = "Internation Conference on Learning Representation (ICLR)",

}

Baktashmotlagh, M, Faraki, M, Drummond, T & Salzmann, M 2019, Learning factorized representations for open-set domain adaptation. in International Conference on Learning Representations 2019. Internation Conference on Learning Representation (ICLR), New Orleans LA USA, pp. 1-11, International Conference on Learning Representations 2019, New Orleans, United States of America, 6/05/19.

Learning factorized representations for open-set domain adaptation. / Baktashmotlagh, Mahsa; Faraki, Masoud; Drummond, Tom; Salzmann, Mathieu.

International Conference on Learning Representations 2019. New Orleans LA USA : Internation Conference on Learning Representation (ICLR), 2019. p. 1-11.

Research output: Chapter in Book/Report/Conference proceedingConference PaperOther

TY - GEN

T1 - Learning factorized representations for open-set domain adaptation

AU - Baktashmotlagh, Mahsa

AU - Faraki, Masoud

AU - Drummond, Tom

AU - Salzmann, Mathieu

PY - 2019

Y1 - 2019

N2 - Domain adaptation for visual recognition has undergone great progress in the past few years. Nevertheless, most existing methods work in the so-called closed-set scenario, assuming that the classes depicted by the target images are exactly the same as those of the source domain. In this paper, we tackle the more challenging, yet more realistic case of open-set domain adaptation, where new, unknown classes can be present in the target data. While, in the unsupervised scenario, one cannot expect to be able to identify each specific new class, we aim to automatically detect which samples belong to these new classes and discard them from the recognition process. To this end, we rely on the intuition that the source and target samples depicting the known classes can be generated by a shared subspace, whereas the target samples from unknown classes come from a different, private subspace. We therefore introduce a framework that factorizes the data into shared and private parts, while encouraging the shared representation to be discriminative. Our experiments on standard benchmarks evidence that our approachoutperforms the state of the art in open-set domain adaptation.

AB - Domain adaptation for visual recognition has undergone great progress in the past few years. Nevertheless, most existing methods work in the so-called closed-set scenario, assuming that the classes depicted by the target images are exactly the same as those of the source domain. In this paper, we tackle the more challenging, yet more realistic case of open-set domain adaptation, where new, unknown classes can be present in the target data. While, in the unsupervised scenario, one cannot expect to be able to identify each specific new class, we aim to automatically detect which samples belong to these new classes and discard them from the recognition process. To this end, we rely on the intuition that the source and target samples depicting the known classes can be generated by a shared subspace, whereas the target samples from unknown classes come from a different, private subspace. We therefore introduce a framework that factorizes the data into shared and private parts, while encouraging the shared representation to be discriminative. Our experiments on standard benchmarks evidence that our approachoutperforms the state of the art in open-set domain adaptation.

M3 - Conference Paper

SP - 1

EP - 11

BT - International Conference on Learning Representations 2019

PB - Internation Conference on Learning Representation (ICLR)

CY - New Orleans LA USA

ER -

Baktashmotlagh M, Faraki M, Drummond T, Salzmann M. Learning factorized representations for open-set domain adaptation. In International Conference on Learning Representations 2019. New Orleans LA USA: Internation Conference on Learning Representation (ICLR). 2019. p. 1-11