Presentation 2018-11-23
Hypernetwork-based Implicit Posterior Estimation of CNN
Kenya Ukai, Takashi Matsubara, Kuniaki Uehara,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Deep neural networks have a rich ability to learn complex representations and achieved remarkable results in various tasks. However, they are prone to overfitting due to the limited number of training samples; regularizing the learning process of neural networks is critical. In this paper, we propose a novel regularization method, which estimates parameters of a large convolutional neural network as implicit probabilistic distributions generated by a hypernetwork. Also, we can perform model averaging to improve the network performance. Experimental results demonstrate our regularization method outperformed the commonly-used maximum a posterior (MAP) estimation.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) hypernetwork / Bayesian estimation / convolutional neural network / image recognition
Paper # CCS2018-45
Date of Issue 2018-11-15 (CCS)

Conference Information
Committee CCS
Conference Date 2018/11/22(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 Mikio Hasegawa(Tokyo Univ. of Science)
Vice Chair Makoto Naruse(NICT) / Shigeki Shiokawa(Kanagawa Inst. of Tech.)
Secretary Makoto Naruse(Tokyo City Univ.) / Shigeki Shiokawa(Hiroshima City Univ.)
Assistant Yusuke Kawakita(Kanagawa Inst. of Tech.) / Hiroyasu Ando(Tsukuba Univ.) / Takashi Matsubara(Kobe Univ.) / Ryo Takahashi(AUT)

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) Hypernetwork-based Implicit Posterior Estimation of CNN
Sub Title (in English)
Keyword(1) hypernetwork
Keyword(2) Bayesian estimation
Keyword(3) convolutional neural network
Keyword(4) image recognition
1st Author's Name Kenya Ukai
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 2018-11-23
Paper # CCS2018-45
Volume (vol) vol.118
Number (no) CCS-316
Page pp.pp.67-72(CCS),
#Pages 6
Date of Issue 2018-11-15 (CCS)