Presentation 2019-11-14
Calibration of Confidence in Deep Learning under Dataset Shift
Kazuki Yoshida, Takashi Matsubara, Kuniaki Uehara,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Calibration of confidence is important when you want to obtain not only the prediction class of unknown data but also the reliability of prediction in deep learning. However, calibration is often inadequate if the confidence distribution is highly skewed. In this study, we calibrated a convolutional neural network and analyzed the effects of model differences and the presence or absence of data set shift on calibration. In addition, we introduce a new statistic of calibration that does not depend on a distribution of a dataset, and discuss problems of the existing methods.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) convolutional neural network / confidence / calibration / dataset shift
Paper # CCS2019-25
Date of Issue 2019-11-07 (CCS)

Conference Information
Committee CCS
Conference Date 2019/11/14(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Kobe Univ.
Topics (in Japanese) (See Japanese page)
Topics (in English) Interaction and Communication, etc.
Chair Makoto Naruse(NICT)
Vice Chair Shigeki Shiokawa(Kanagawa Inst. of Tech.) / Tetsuya Asai(Hokkaido Univ.)
Secretary Shigeki Shiokawa(Hiroshima City Univ.) / Tetsuya Asai(Kanagawa Inst. of Tech.)
Assistant Hidehiro Nakano(Tokyo City Univ.) / Kazuki Nakada(Tsukuba Univ. of Tech.) / Hiroyasu Ando(Tsukuba Univ.) / Takashi Matsubara(Kobe Univ.)

Paper Information
Registration To Technical Committee on Complex Communication Sciences
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Calibration of Confidence in Deep Learning under Dataset Shift
Sub Title (in English)
Keyword(1) convolutional neural network
Keyword(2) confidence
Keyword(3) calibration
Keyword(4) dataset shift
1st Author's Name Kazuki Yoshida
1st Author's Affiliation Kobe University(Kobe Univ.)
2nd Author's Name Takashi Matsubara
2nd Author's Affiliation Kobe University(Kobe Univ.)
3rd Author's Name Kuniaki Uehara
3rd Author's Affiliation Kobe University(Kobe Univ.)
Date 2019-11-14
Paper # CCS2019-25
Volume (vol) vol.119
Number (no) CCS-289
Page pp.pp.5-8(CCS),
#Pages 4
Date of Issue 2019-11-07 (CCS)