| Presentation | 2018-08-21 On Structural Constraint for Convolutional Layers in CNN Shota Kayamori, Shogo Muramatsu, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | This work investigates Convolutional Neural Network (CNN) constituted by convolution layers imposed structural constraint. A specific convolutional structure is constructed based on NonSeparable Oversampled Lapped Transforms (NSOLT) and convolutional layers satisfying the linear phase property and Parseval tight frame property is constructed. NSOLT, which the authors proposed in previous works, achieves sparse representation performance equivalent or superior to Sparse K-SVD of Rubinstein et al. Moreover, because of the structural constraint, the number of optimization parameters can be significantly reduced from normal CNN, and convergence acceleration of learning can be expected. In this research, the performance of proposed CNN and usual CNN are compared by using CIFAR-10 data set and the significance of the proposed method is verified. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | CNN / Convolutional layers / Structural constraint / NSOLT |
| Paper # | SIP2018-63 |
| Date of Issue | 2018-08-13 (SIP) |
| Conference Information | |
| Committee | SIP |
|---|---|
| Conference Date | 2018/8/20(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Takushoku Univ. Bunkyo Campus. |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Shogo Muramatsu(Niigata Univ.) |
| Vice Chair | Naoyuki Aikawa(TUS) / Kazunori Hayashi(Osaka City Univ) |
| Secretary | Naoyuki Aikawa(Takushoku Univ.) / Kazunori Hayashi(Hiroshima Univ.) |
| Assistant | Katsumi Konishi(Hosei Univ.) / hyihsin(Takushoku Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Signal Processing |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | On Structural Constraint for Convolutional Layers in CNN |
| Sub Title (in English) | |
| Keyword(1) | CNN |
| Keyword(2) | Convolutional layers |
| Keyword(3) | Structural constraint |
| Keyword(4) | NSOLT |
| 1st Author's Name | Shota Kayamori |
| 1st Author's Affiliation | Niigata University(Niigata Univ.) |
| 2nd Author's Name | Shogo Muramatsu |
| 2nd Author's Affiliation | Niigata University(Niigata Univ.) |
| Date | 2018-08-21 |
| Paper # | SIP2018-63 |
| Volume (vol) | vol.118 |
| Number (no) | SIP-182 |
| Page | pp.pp.33-36(SIP), |
| #Pages | 4 |
| Date of Issue | 2018-08-13 (SIP) |