Controlling Imbalanced Error in Deep Learning with the Log Bilinear Loss.

Yehezkel S. Resheff; Amit Mandelbom; Daphna Weinshall

Controlling Imbalanced Error in Deep Learning with the Log Bilinear Loss.

Yehezkel S. Resheff, Amit Mandelbom, Daphna Weinshall

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

Abstract

Deep learning has become the method of choice for many machine learning tasks in recent years, and especially for multi-class classification. The most common loss function used in this context is the cross-entropy loss. While this function is insensitive to the identity of the assigned class in the case of misclassification, in practice it very common to have imbalanced sensitivity to error, meaning some wrong assignments are much worse than others. Here we present the bilinear-loss (and related log-bilinear-loss) which differentially penalizes the different wrong assignments of the model. We thoroughly test the proposed method using standard models and benchmark image datasets.

Original language	English
Title of host publication	Proceedings of the First International Workshop on Learning with Imbalanced Domains
Subtitle of host publication	Theory and Applications
Publisher	PMLR
Pages	141-151
Number of pages	11
State	Published - 2017
Event	International Workshop on Learning with Imbalanced Domains: Theory and Applications - Skopje, Macedonia, The Former Yugoslav Republic of Duration: 22 Sep 2017 → 22 Sep 2017 Conference number: 1 https://proceedings.mlr.press/v74

Publication series

Name	Proceedings of Machine Learning Research
Volume	74
ISSN (Electronic)	2640-3498

Conference

Conference	International Workshop on Learning with Imbalanced Domains
Abbreviated title	LIDTA 2017
Country/Territory	Macedonia, The Former Yugoslav Republic of
City	Skopje
Period	22/09/17 → 22/09/17
Internet address	https://proceedings.mlr.press/v74

Access to Document

http://proceedings.mlr.press/v74/resheff17a.html

Cite this

@inproceedings{342e7309bac34246b283d969b92ed4c5,

title = "Controlling Imbalanced Error in Deep Learning with the Log Bilinear Loss.",

abstract = "Deep learning has become the method of choice for many machine learning tasks in recent years, and especially for multi-class classification. The most common loss function used in this context is the cross-entropy loss. While this function is insensitive to the identity of the assigned class in the case of misclassification, in practice it very common to have imbalanced sensitivity to error, meaning some wrong assignments are much worse than others. Here we present the bilinear-loss (and related log-bilinear-loss) which differentially penalizes the different wrong assignments of the model. We thoroughly test the proposed method using standard models and benchmark image datasets.",

author = "Resheff, {Yehezkel S.} and Amit Mandelbom and Daphna Weinshall",

year = "2017",

language = "אנגלית",

series = "Proceedings of Machine Learning Research",

publisher = "PMLR",

pages = "141--151",

booktitle = "Proceedings of the First International Workshop on Learning with Imbalanced Domains",

note = "International Workshop on Learning with Imbalanced Domains ; Conference date: 22-09-2017 Through 22-09-2017",

url = "https://proceedings.mlr.press/v74",

}

Resheff, YS, Mandelbom, A & Weinshall, D 2017, Controlling Imbalanced Error in Deep Learning with the Log Bilinear Loss. in Proceedings of the First International Workshop on Learning with Imbalanced Domains: Theory and Applications. Proceedings of Machine Learning Research, vol. 74, PMLR, pp. 141-151, International Workshop on Learning with Imbalanced Domains, Skopje, Macedonia, The Former Yugoslav Republic of, 22/09/17. <http://proceedings.mlr.press/v74/resheff17a.html>

Controlling Imbalanced Error in Deep Learning with the Log Bilinear Loss. / Resheff, Yehezkel S.; Mandelbom, Amit; Weinshall, Daphna.
Proceedings of the First International Workshop on Learning with Imbalanced Domains: Theory and Applications. PMLR, 2017. p. 141-151 (Proceedings of Machine Learning Research; Vol. 74).

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

TY - GEN

T1 - Controlling Imbalanced Error in Deep Learning with the Log Bilinear Loss.

AU - Resheff, Yehezkel S.

AU - Mandelbom, Amit

AU - Weinshall, Daphna

N1 - Conference code: 1

PY - 2017

Y1 - 2017

N2 - Deep learning has become the method of choice for many machine learning tasks in recent years, and especially for multi-class classification. The most common loss function used in this context is the cross-entropy loss. While this function is insensitive to the identity of the assigned class in the case of misclassification, in practice it very common to have imbalanced sensitivity to error, meaning some wrong assignments are much worse than others. Here we present the bilinear-loss (and related log-bilinear-loss) which differentially penalizes the different wrong assignments of the model. We thoroughly test the proposed method using standard models and benchmark image datasets.

AB - Deep learning has become the method of choice for many machine learning tasks in recent years, and especially for multi-class classification. The most common loss function used in this context is the cross-entropy loss. While this function is insensitive to the identity of the assigned class in the case of misclassification, in practice it very common to have imbalanced sensitivity to error, meaning some wrong assignments are much worse than others. Here we present the bilinear-loss (and related log-bilinear-loss) which differentially penalizes the different wrong assignments of the model. We thoroughly test the proposed method using standard models and benchmark image datasets.

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

T3 - Proceedings of Machine Learning Research

SP - 141

EP - 151

BT - Proceedings of the First International Workshop on Learning with Imbalanced Domains

PB - PMLR

T2 - International Workshop on Learning with Imbalanced Domains

Y2 - 22 September 2017 through 22 September 2017

ER -

Controlling Imbalanced Error in Deep Learning with the Log Bilinear Loss.

Abstract

Publication series

Conference

Access to Document

Fingerprint

Cite this