Presentation 2017-10-12
An autoencoder reversing abnormal inputs
Akihiro Suzuki, Hakaru Tamukoh,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Usages of autoencoders (AEs) are not only a dimension reducer, but a generative model using reconstruction. AEs are used as the generative model in this work and applied to an abnormal detection. In usual abnormal detectors without label data, they require normal value to generate a model which generate normal values. By contrast, our AE as abnormal detector requires both of normal and abnormal data to generate a model. Proposed AE reconstructs a normal input regularly and an abnormal input irregularly, respectively. By proposed methods, reconstruction errors by abnormal input come to increase and it is easy to detect as abnormal input. This paper reports experiments about an abnormal detection with MNIST dataset by proposed AE.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Deep Learning / Autoencoder / Abnormal detection
Paper # SIS2017-26
Date of Issue 2017-10-05 (SIS)

Conference Information
Committee SIS / IPSJ-AVM
Conference Date 2017/10/12(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Todaiji Culture Center
Topics (in Japanese) (See Japanese page)
Topics (in English) Intelligent Multimedia Systems, Applied Enbedded Systems, etc.
Chair Hirokazu Tanaka(Hiroshima City Univ.) / Hideaki Kimata(NTT)
Vice Chair Takayuki Nakachi(NTT) / Noriaki Suetake(Yamaguchi Univ.)
Secretary Takayuki Nakachi(Kanagawa Inst. of Tech.) / Noriaki Suetake(Kyushu Inst. of Tech.) / (NTT)
Assistant Masaaki Fujiyoshi(Tokyo Metropolitan Univ.) / Takanori Koga(TCT)

Paper Information
Registration To Technical Committee on Smart Info-Media Systems / Special Interest Group on Audio Visual and Multimedia Information Processing
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) An autoencoder reversing abnormal inputs
Sub Title (in English)
Keyword(1) Deep Learning
Keyword(2) Autoencoder
Keyword(3) Abnormal detection
1st Author's Name Akihiro Suzuki
1st Author's Affiliation Kyushu Institute of Technology(Kyushu Inst. of Tech.)
2nd Author's Name Hakaru Tamukoh
2nd Author's Affiliation Kyushu Institute of Technology(Kyushu Inst. of Tech.)
Date 2017-10-12
Paper # SIS2017-26
Volume (vol) vol.117
Number (no) SIS-235
Page pp.pp.29-34(SIS),
#Pages 6
Date of Issue 2017-10-05 (SIS)