Discovering class-wise trends of max-pooling in subspace

Yuchen Zheng, Brian Kenji Iwana, Seiichi Uchida

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

1 Citation (Scopus)

Abstract

The traditional max-pooling operation in Convolutional Neural Networks (CNNs) only obtains the maximal value from a pooling window. However, it discards the information about the precise position of the maximal value. In this paper, we extract the location of the maximal value in a pooling window and transform it into 'displacement feature'. We analyze and discover the class-wise trend of the displacement features in many ways. The experimental results and discussion demonstrate that the displacement features have beneficial behaviors for solving the problems in max-pooling.

Original languageEnglish
Title of host publicationProceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages98-103
Number of pages6
ISBN (Electronic)9781538658758
DOIs
Publication statusPublished - Dec 5 2018
Event16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018 - Niagara Falls, United States
Duration: Aug 5 2018Aug 8 2018

Publication series

NameProceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR
Volume2018-August
ISSN (Print)2167-6445
ISSN (Electronic)2167-6453

Other

Other16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018
CountryUnited States
CityNiagara Falls
Period8/5/188/8/18

Fingerprint

Neural networks

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Computer Vision and Pattern Recognition

Cite this

Zheng, Y., Iwana, B. K., & Uchida, S. (2018). Discovering class-wise trends of max-pooling in subspace. In Proceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018 (pp. 98-103). [8563233] (Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR; Vol. 2018-August). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICFHR-2018.2018.00026

Discovering class-wise trends of max-pooling in subspace. / Zheng, Yuchen; Iwana, Brian Kenji; Uchida, Seiichi.

Proceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018. Institute of Electrical and Electronics Engineers Inc., 2018. p. 98-103 8563233 (Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR; Vol. 2018-August).

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

Zheng, Y, Iwana, BK & Uchida, S 2018, Discovering class-wise trends of max-pooling in subspace. in Proceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018., 8563233, Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR, vol. 2018-August, Institute of Electrical and Electronics Engineers Inc., pp. 98-103, 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018, Niagara Falls, United States, 8/5/18. https://doi.org/10.1109/ICFHR-2018.2018.00026
Zheng Y, Iwana BK, Uchida S. Discovering class-wise trends of max-pooling in subspace. In Proceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018. Institute of Electrical and Electronics Engineers Inc. 2018. p. 98-103. 8563233. (Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR). https://doi.org/10.1109/ICFHR-2018.2018.00026
Zheng, Yuchen ; Iwana, Brian Kenji ; Uchida, Seiichi. / Discovering class-wise trends of max-pooling in subspace. Proceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018. Institute of Electrical and Electronics Engineers Inc., 2018. pp. 98-103 (Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR).
@inproceedings{4a9de35c5c7f43bbbce09d20c0dc9e11,
title = "Discovering class-wise trends of max-pooling in subspace",
abstract = "The traditional max-pooling operation in Convolutional Neural Networks (CNNs) only obtains the maximal value from a pooling window. However, it discards the information about the precise position of the maximal value. In this paper, we extract the location of the maximal value in a pooling window and transform it into 'displacement feature'. We analyze and discover the class-wise trend of the displacement features in many ways. The experimental results and discussion demonstrate that the displacement features have beneficial behaviors for solving the problems in max-pooling.",
author = "Yuchen Zheng and Iwana, {Brian Kenji} and Seiichi Uchida",
year = "2018",
month = "12",
day = "5",
doi = "10.1109/ICFHR-2018.2018.00026",
language = "English",
series = "Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "98--103",
booktitle = "Proceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018",
address = "United States",

}

TY - GEN

T1 - Discovering class-wise trends of max-pooling in subspace

AU - Zheng, Yuchen

AU - Iwana, Brian Kenji

AU - Uchida, Seiichi

PY - 2018/12/5

Y1 - 2018/12/5

N2 - The traditional max-pooling operation in Convolutional Neural Networks (CNNs) only obtains the maximal value from a pooling window. However, it discards the information about the precise position of the maximal value. In this paper, we extract the location of the maximal value in a pooling window and transform it into 'displacement feature'. We analyze and discover the class-wise trend of the displacement features in many ways. The experimental results and discussion demonstrate that the displacement features have beneficial behaviors for solving the problems in max-pooling.

AB - The traditional max-pooling operation in Convolutional Neural Networks (CNNs) only obtains the maximal value from a pooling window. However, it discards the information about the precise position of the maximal value. In this paper, we extract the location of the maximal value in a pooling window and transform it into 'displacement feature'. We analyze and discover the class-wise trend of the displacement features in many ways. The experimental results and discussion demonstrate that the displacement features have beneficial behaviors for solving the problems in max-pooling.

UR - http://www.scopus.com/inward/record.url?scp=85060010396&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85060010396&partnerID=8YFLogxK

U2 - 10.1109/ICFHR-2018.2018.00026

DO - 10.1109/ICFHR-2018.2018.00026

M3 - Conference contribution

AN - SCOPUS:85060010396

T3 - Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR

SP - 98

EP - 103

BT - Proceedings - 2018 16th International Conference on Frontiers in Handwriting Recognition, ICFHR 2018

PB - Institute of Electrical and Electronics Engineers Inc.

ER -