Transfer Learning by Learning Projections from Target to Source

Antoine Cornuéjols; Pierre-Alexandre Murena; Raphaël Olivier

doi:10.1007/978-3-030-44584-3_10

Transfer Learning by Learning Projections from Target to Source

Antoine Cornuéjols^*, Pierre-Alexandre Murena, Raphaël Olivier

^*Corresponding author for this work

Not published at Aalto University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

Abstract

Using transfer learning to help in solving a new classification task where labeled data is scarce is becoming popular. Numerous experiments with deep neural networks, where the representation learned on a source task is transferred to learn a target neural network, have shown the benefits of the approach. This paper, similarly, deals with hypothesis transfer learning. However, it presents a new approach where, instead of transferring a representation, the source hypothesis is kept and this is a translation from the target domain to the source domain that is learned. In a way, a change of representation is learned. We show how this method performs very well on a classification of time series task where the space of time series is changed between source and target.

Original language	English
Title of host publication	Advances in Intelligent Data Analysis XVIII
Subtitle of host publication	18th International Symposium on Intelligent Data Analysis, IDA 2020, Konstanz, Germany, April 27–29, 2020, Proceedings
Pages	119-131
ISBN (Electronic)	978-3-030-44584-3
DOIs	https://doi.org/10.1007/978-3-030-44584-3_10
Publication status	Published - 27 Apr 2020
MoE publication type	A4 Article in a conference publication
Event	International Symposium on Intelligent Data Analysis - Konstanz, Germany Duration: 27 Apr 2020 → 29 Apr 2020 Conference number: 18

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer
Volume	12080
ISSN (Print)	0302-9743

Conference

Conference	International Symposium on Intelligent Data Analysis
Abbreviated title	IDA
Country	Germany
City	Konstanz
Period	27/04/2020 → 29/04/2020

Access to Document

10.1007/978-3-030-44584-3_10Licence: CC BY

Fingerprint Dive into the research topics of 'Transfer Learning by Learning Projections from Target to Source'. Together they form a unique fingerprint.

Cite this

Cornuéjols, A., Murena, P-A., & Olivier, R. (2020). Transfer Learning by Learning Projections from Target to Source. In Advances in Intelligent Data Analysis XVIII: 18th International Symposium on Intelligent Data Analysis, IDA 2020, Konstanz, Germany, April 27–29, 2020, Proceedings (pp. 119-131). (Lecture Notes in Computer Science; Vol. 12080). https://doi.org/10.1007/978-3-030-44584-3_10

@inproceedings{7311340e4b434d46962f751c1b06df0d,

title = "Transfer Learning by Learning Projections from Target to Source",

abstract = "Using transfer learning to help in solving a new classification task where labeled data is scarce is becoming popular. Numerous experiments with deep neural networks, where the representation learned on a source task is transferred to learn a target neural network, have shown the benefits of the approach. This paper, similarly, deals with hypothesis transfer learning. However, it presents a new approach where, instead of transferring a representation, the source hypothesis is kept and this is a translation from the target domain to the source domain that is learned. In a way, a change of representation is learned. We show how this method performs very well on a classification of time series task where the space of time series is changed between source and target.",

author = "Antoine Cornu{\'e}jols and Pierre-Alexandre Murena and Rapha{\"e}l Olivier",

year = "2020",

month = apr,

day = "27",

doi = "10.1007/978-3-030-44584-3_10",

language = "English",

isbn = "978-3-030-44583-6",

series = "Lecture Notes in Computer Science",

publisher = "Springer",

pages = "119--131",

booktitle = "Advances in Intelligent Data Analysis XVIII",

note = "International Symposium on Intelligent Data Analysis, IDA ; Conference date: 27-04-2020 Through 29-04-2020",

}

Cornuéjols, A, Murena, P-A & Olivier, R 2020, Transfer Learning by Learning Projections from Target to Source. in Advances in Intelligent Data Analysis XVIII: 18th International Symposium on Intelligent Data Analysis, IDA 2020, Konstanz, Germany, April 27–29, 2020, Proceedings. Lecture Notes in Computer Science, vol. 12080, pp. 119-131, International Symposium on Intelligent Data Analysis, Konstanz, Germany, 27/04/2020. https://doi.org/10.1007/978-3-030-44584-3_10

Transfer Learning by Learning Projections from Target to Source. / Cornuéjols, Antoine; Murena, Pierre-Alexandre; Olivier, Raphaël.

Advances in Intelligent Data Analysis XVIII: 18th International Symposium on Intelligent Data Analysis, IDA 2020, Konstanz, Germany, April 27–29, 2020, Proceedings. 2020. p. 119-131 (Lecture Notes in Computer Science; Vol. 12080).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Transfer Learning by Learning Projections from Target to Source

AU - Cornuéjols, Antoine

AU - Murena, Pierre-Alexandre

AU - Olivier, Raphaël

N1 - Conference code: 18

PY - 2020/4/27

Y1 - 2020/4/27

N2 - Using transfer learning to help in solving a new classification task where labeled data is scarce is becoming popular. Numerous experiments with deep neural networks, where the representation learned on a source task is transferred to learn a target neural network, have shown the benefits of the approach. This paper, similarly, deals with hypothesis transfer learning. However, it presents a new approach where, instead of transferring a representation, the source hypothesis is kept and this is a translation from the target domain to the source domain that is learned. In a way, a change of representation is learned. We show how this method performs very well on a classification of time series task where the space of time series is changed between source and target.

AB - Using transfer learning to help in solving a new classification task where labeled data is scarce is becoming popular. Numerous experiments with deep neural networks, where the representation learned on a source task is transferred to learn a target neural network, have shown the benefits of the approach. This paper, similarly, deals with hypothesis transfer learning. However, it presents a new approach where, instead of transferring a representation, the source hypothesis is kept and this is a translation from the target domain to the source domain that is learned. In a way, a change of representation is learned. We show how this method performs very well on a classification of time series task where the space of time series is changed between source and target.

U2 - 10.1007/978-3-030-44584-3_10

DO - 10.1007/978-3-030-44584-3_10

M3 - Conference contribution

SN - 978-3-030-44583-6

T3 - Lecture Notes in Computer Science

SP - 119

EP - 131

BT - Advances in Intelligent Data Analysis XVIII

T2 - International Symposium on Intelligent Data Analysis

Y2 - 27 April 2020 through 29 April 2020

ER -

Cornuéjols A, Murena P-A, Olivier R. Transfer Learning by Learning Projections from Target to Source. In Advances in Intelligent Data Analysis XVIII: 18th International Symposium on Intelligent Data Analysis, IDA 2020, Konstanz, Germany, April 27–29, 2020, Proceedings. 2020. p. 119-131. (Lecture Notes in Computer Science). https://doi.org/10.1007/978-3-030-44584-3_10