Presentation 2021-03-03
Visualization of CNNs using Preferred Stimulus in Receptive Fields
Genta Kobayashi, Hayaru Shouno,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Convolutional neural networks have shown high performance at image processing task, and they are interpreted by various methods. For instance, there are visualization methods to estimate an input image by maximizing the activation of a neuron. These methods have the problem that shows various solutions by optimization methods. Here, we propose a visualization method using the preferred stimulus in the receptive field. We apply our method to ResNet and PlainNet without skip connection trained on ImageNet. Consequently, we show the effectiveness and confirm the properties of skip connection.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Convolutional Neural Network / Receptive Field / Visualization / Preferred Stimulus / Residual Network
Paper # NC2020-47
Date of Issue 2021-02-24 (NC)

Conference Information
Committee NC / MBE
Conference Date 2021/3/3(3days)
Place (in Japanese) (See Japanese page)
Place (in English) Online
Topics (in Japanese) (See Japanese page)
Topics (in English) Neuro Computing, Medical Engineering, etc.
Chair Kazuyuki Samejima(Tamagawa Univ) / Takashi Watanabe(Tohoku Univ.)
Vice Chair Rieko Osu(Waseda Univ.) / Ryuhei Okuno(Setsunan Univ.)
Secretary Rieko Osu(NTT) / Ryuhei Okuno(ATR)
Assistant Ken Takiyama(TUAT) / Nobuhiko Wagatsuma(Toho Univ.) / Akihiro Karashima(Tohoku Inst. of Tech.) / Jun Akazawa(Meiji Univ. of Integrative Medicine)

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) Visualization of CNNs using Preferred Stimulus in Receptive Fields
Sub Title (in English)
Keyword(1) Convolutional Neural Network
Keyword(2) Receptive Field
Keyword(3) Visualization
Keyword(4) Preferred Stimulus
Keyword(5) Residual Network
1st Author's Name Genta Kobayashi
1st Author's Affiliation The University of Electro-Communications(UEC)
2nd Author's Name Hayaru Shouno
2nd Author's Affiliation The University of Electro-Communications(UEC)
Date 2021-03-03
Paper # NC2020-47
Volume (vol) vol.120
Number (no) NC-403
Page pp.pp.25-30(NC),
#Pages 6
Date of Issue 2021-02-24 (NC)