Adding new classes without access to the original training data with applications to language identification

Hagai Taitelbaum; Ehud Ben-Reuven; Jacob Goldberger

doi:10.21437/interspeech.2018-1342

Adding new classes without access to the original training data with applications to language identification

Hagai Taitelbaum, Ehud Ben-Reuven, Jacob Goldberger

Bar-Ilan University

Research output: Contribution to journal › Conference article › peer-review

3 Scopus citations

Abstract

In this study we address the problem of adding new classes to an existing neural network classifier. We assume that new training data with the new classes is available. In many applications, dataset used to train machine learning algorithms contain confidential information that cannot be accessed during the process of extending the class set. We propose a method for training an extended class-set classifier using only examples with labels from the new classes while avoiding the problem of forgetting the original classes. This incremental training method is applied to the problem of language identification. We report results on the 50 languages NIST 2015 dataset where we were able to classify all the languages even though only part of the classes was available during the first training phase and the other languages were only available during the second phase.

Original language	English
Pages (from-to)	1808-1812
Number of pages	5
Journal	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Volume	2018-September
DOIs	https://doi.org/10.21437/interspeech.2018-1342
State	Published - 2018
Event	19th Annual Conference of the International Speech Communication, INTERSPEECH 2018 - Hyderabad, India Duration: 2 Sep 2018 → 6 Sep 2018

Bibliographical note

Publisher Copyright:
© 2018 International Speech Communication Association. All rights reserved.

Funding

This research is partially supported by the BIU Center for Research in Applied Cryptography and Cyber Security in conjunction with the Israel National Cyber Directorate in the Prime Minister's Office.

Funders	Funder number
Canadian Centre for Applied Research in Cancer Control

Keywords

Adding new classes
Catastrophic forgetting
Language identification
Learning privacy

Access to Document

10.21437/interspeech.2018-1342

Cite this

@article{b875754d27864ba8845144ed128d859a,

title = "Adding new classes without access to the original training data with applications to language identification",

abstract = "In this study we address the problem of adding new classes to an existing neural network classifier. We assume that new training data with the new classes is available. In many applications, dataset used to train machine learning algorithms contain confidential information that cannot be accessed during the process of extending the class set. We propose a method for training an extended class-set classifier using only examples with labels from the new classes while avoiding the problem of forgetting the original classes. This incremental training method is applied to the problem of language identification. We report results on the 50 languages NIST 2015 dataset where we were able to classify all the languages even though only part of the classes was available during the first training phase and the other languages were only available during the second phase.",

keywords = "Adding new classes, Catastrophic forgetting, Language identification, Learning privacy",

author = "Hagai Taitelbaum and Ehud Ben-Reuven and Jacob Goldberger",

note = "Publisher Copyright: {\textcopyright} 2018 International Speech Communication Association. All rights reserved.; 19th Annual Conference of the International Speech Communication, INTERSPEECH 2018 ; Conference date: 02-09-2018 Through 06-09-2018",

year = "2018",

doi = "10.21437/interspeech.2018-1342",

language = "אנגלית",

volume = "2018-September",

pages = "1808--1812",

journal = "Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH",

issn = "2308-457X",

}

Adding new classes without access to the original training data with applications to language identification. / Taitelbaum, Hagai; Ben-Reuven, Ehud; Goldberger, Jacob.
In: Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH, Vol. 2018-September, 2018, p. 1808-1812.

Research output: Contribution to journal › Conference article › peer-review

TY - JOUR

T1 - Adding new classes without access to the original training data with applications to language identification

AU - Taitelbaum, Hagai

AU - Ben-Reuven, Ehud

AU - Goldberger, Jacob

PY - 2018

Y1 - 2018

N2 - In this study we address the problem of adding new classes to an existing neural network classifier. We assume that new training data with the new classes is available. In many applications, dataset used to train machine learning algorithms contain confidential information that cannot be accessed during the process of extending the class set. We propose a method for training an extended class-set classifier using only examples with labels from the new classes while avoiding the problem of forgetting the original classes. This incremental training method is applied to the problem of language identification. We report results on the 50 languages NIST 2015 dataset where we were able to classify all the languages even though only part of the classes was available during the first training phase and the other languages were only available during the second phase.

AB - In this study we address the problem of adding new classes to an existing neural network classifier. We assume that new training data with the new classes is available. In many applications, dataset used to train machine learning algorithms contain confidential information that cannot be accessed during the process of extending the class set. We propose a method for training an extended class-set classifier using only examples with labels from the new classes while avoiding the problem of forgetting the original classes. This incremental training method is applied to the problem of language identification. We report results on the 50 languages NIST 2015 dataset where we were able to classify all the languages even though only part of the classes was available during the first training phase and the other languages were only available during the second phase.

KW - Adding new classes

KW - Catastrophic forgetting

KW - Language identification

KW - Learning privacy

UR - http://www.scopus.com/inward/record.url?scp=85054971775&partnerID=8YFLogxK

U2 - 10.21437/interspeech.2018-1342

DO - 10.21437/interspeech.2018-1342

M3 - ???researchoutput.researchoutputtypes.contributiontojournal.conferencearticle???

AN - SCOPUS:85054971775

SN - 2308-457X

VL - 2018-September

SP - 1808

EP - 1812

JO - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

JF - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

T2 - 19th Annual Conference of the International Speech Communication, INTERSPEECH 2018

Y2 - 2 September 2018 through 6 September 2018

ER -

Adding new classes without access to the original training data with applications to language identification

Abstract

Bibliographical note

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this