Presentation | 2019-12-06 Prevention of redundant representations and of the black box in stacked autoencoders Masumi Ishikawa, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Recent progress in deep learning (DL) is remarkable and its recognition capability is said to surpass that of humans. The acquired features, however, don’t seem to be sufficiently clear. Furthermore, the basis of judgement by DL is mostly unknown. In 1990s the author proposed L1-norm regularization and relevant terms to create understandable neural networks mainly with discrete valued inputs and outputs. Because DL is inherently nonlinear, L1-norm alone might not be enough to understand the resulting models due to frequently emerging redundant representations. The present paper extends the previous proposal to deep learning models with special emphasis on redundant representations. Training with L1-norm and selective L1-norm to connection weights, decomposition of a model, and suppressing redundant representations are proposed. The paper clarifies the effectiveness of the proposal by applying it to stacked autoencoders using red wine quality data. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Stacked autoencoder / Sparse modeling / Deep learning / Redundant representation / Black box |
Paper # | MBE2019-56,NC2019-47 |
Date of Issue | 2019-11-29 (MBE, NC) |
Conference Information | |
Committee | NC / MBE |
---|---|
Conference Date | 2019/12/6(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Toyohashi Tech |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Hayaru Shouno(UEC) / Taishin Nomura(Osaka Univ.) |
Vice Chair | Kazuyuki Samejima(Tamagawa Univ) / Takashi Watanabe(Tohoku Univ.) |
Secretary | Kazuyuki Samejima(NAIST) / Takashi Watanabe(NTT) |
Assistant | Takashi Shinozaki(NICT) / Ken Takiyama(TUAT) / Yasuyuki Suzuki(Osaka Univ.) / Akihiro Karashima(Tohoku Inst. of Tech.) |
Paper Information | |
Registration To | Technical Committee on Neurocomputing / Technical Committee on ME and Bio Cybernetics |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Prevention of redundant representations and of the black box in stacked autoencoders |
Sub Title (in English) | |
Keyword(1) | Stacked autoencoder |
Keyword(2) | Sparse modeling |
Keyword(3) | Deep learning |
Keyword(4) | Redundant representation |
Keyword(5) | Black box |
1st Author's Name | Masumi Ishikawa |
1st Author's Affiliation | Kyushu Institute of Technology(Kyutech) |
Date | 2019-12-06 |
Paper # | MBE2019-56,NC2019-47 |
Volume (vol) | vol.119 |
Number (no) | MBE-327,NC-328 |
Page | pp.pp.67-72(MBE), pp.67-72(NC), |
#Pages | 6 |
Date of Issue | 2019-11-29 (MBE, NC) |