Presentation 2016-11-17
Transportation aspect of infinitely deep denoising autoencoder
Sho Sonoda, Noboru Murata,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The hidden layers of a deep neural network extract feature of the input data. We propose to analyze the feature map of a neural network by regarding it as a transportation map. The transportation map of denoising autoencoders (DAEs) is derived, which can even characterize the infinitely deep architecture. It is revealed that the composition of DAEs and the continuous DAE transport data points to reduce the entropy of the input data. The stacked DAE, which is a product of a complicated heuristics, can be identified as a composition of DAEs from the view point of a transportation map.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) deep learning / infinitely deep neural network / denoising autoencoder / transportation / backward heat equation
Paper # IBISML2016-88
Date of Issue 2016-11-09 (IBISML)

Conference Information
Committee IBISML
Conference Date 2016/11/16(3days)
Place (in Japanese) (See Japanese page)
Place (in English) Kyoto Univ.
Topics (in Japanese) (See Japanese page)
Topics (in English) Information-Based Induction Science Workshop (IBIS2016)
Chair Kenji Fukumizu(ISM)
Vice Chair Masashi Sugiyama(Univ. of Tokyo) / Hisashi Kashima(Kyoto Univ.)
Secretary Masashi Sugiyama(Univ. of Tokyo) / Hisashi Kashima(Nagoya Inst. of Tech.)
Assistant Toshihiro Kamishima(AIST) / Tomoharu Iwata(NTT)

Paper Information
Registration To Technical Committee on Infomation-Based Induction Sciences and Machine Learning
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Transportation aspect of infinitely deep denoising autoencoder
Sub Title (in English)
Keyword(1) deep learning
Keyword(2) infinitely deep neural network
Keyword(3) denoising autoencoder
Keyword(4) transportation
Keyword(5) backward heat equation
1st Author's Name Sho Sonoda
1st Author's Affiliation Waseda University(Waseda Univ.)
2nd Author's Name Noboru Murata
2nd Author's Affiliation Waseda University(Waseda Univ.)
Date 2016-11-17
Paper # IBISML2016-88
Volume (vol) vol.116
Number (no) IBISML-300
Page pp.pp.297-304(IBISML),
#Pages 8
Date of Issue 2016-11-09 (IBISML)