Presentation 2005/3/23
Blind Source Deconvolution by Auto-Encoder
Takashi NAKABE, Syozo YASUI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Previously, the use of an auto-encoder is shown to be a new method for Blind Source Separation (BSS). This approach which needs no pre-processing and has high adaptability/robustness is extended in this paper to deal with the Blind Source Deconvolution (BSD) problem. The de-mixing filter matrix is determined as a result of the sensor-signal identity mapping due to the auto-encoder so that the blind signals are recovered at nonlinear hidden units. Also, the external mixing filter matrix is reconstructed in the de-coder part as an internal model of the mixing. Furthermore, as in the case of BSS, a pruning algorithm is incorporated to the auto-encoder, to deal with cases where the number of sources is unknown, i.e., many hidden units are made available initially and ones that survive the pruning will be the source separators. Some simulation results are presented to demonstrate the feasibility of the present approach to BSD.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Blind Source Deconvolution / Auto-Encoder / Puruning
Paper # NC2004-209
Date of Issue

Conference Information
Committee NC
Conference Date 2005/3/23(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Blind Source Deconvolution by Auto-Encoder
Sub Title (in English)
Keyword(1) Blind Source Deconvolution
Keyword(2) Auto-Encoder
Keyword(3) Puruning
1st Author's Name Takashi NAKABE
1st Author's Affiliation Graduate School of Life Science and System Engineering, Kyushu Institute of Technology()
2nd Author's Name Syozo YASUI
2nd Author's Affiliation Graduate School of Life Science and System Engineering, Kyushu Institute of Technology
Date 2005/3/23
Paper # NC2004-209
Volume (vol) vol.104
Number (no) 760
Page pp.pp.-
#Pages 6
Date of Issue