Presentation | 2022-12-15 A DNN compression method based on output error of activation functions Koji Kamma, Toshikazu Wada, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Deep Neural Networks (DNNs) are dominant in the field of machine learning. However, because DNN models have large computational complexity, implementation of DNN models on resource-limited equipment is challenging. Therefore, techniques for compressing DNN models without degrading their accuracy is desired. Pruning is one such technique that re- moves redundant neurons (or channels). In this paper, we present Pruning with Output Error Minimization (POEM), a method that performs not only pruning but also reconstruction to compensate the error caused by pruning. The strength of POEM lies in its reconstruction to minimize the output error of the activation function, whereas the previous methods minimize the error before the activation function. The experiments with well-known DNN models (VGG-16, ResNet-18, MobileNet) and image recognition datasets (ImageNet, CUB-200-2011) were conducted. The results show that POEM significantly outperformed the previous methods in maintaining the accuracy of the compressed models. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | pruning / reconstruction / activation function |
Paper # | PRMU2022-38 |
Date of Issue | 2022-12-08 (PRMU) |
Conference Information | |
Committee | PRMU |
---|---|
Conference Date | 2022/12/15(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Toyama International Conference Center |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Seiichi Uchida(Kyushu Univ.) |
Vice Chair | Takuya Funatomi(NAIST) / Mitsuru Anpai(Denso IT Lab.) |
Secretary | Takuya Funatomi(CyberAgent) / Mitsuru Anpai(Univ. of Tokyo) |
Assistant | Nakamasa Inoue(Tokyo Inst. of Tech.) / Yasutomo Kawanishi(Riken) |
Paper Information | |
Registration To | Technical Committee on Pattern Recognition and Media Understanding |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A DNN compression method based on output error of activation functions |
Sub Title (in English) | |
Keyword(1) | pruning |
Keyword(2) | reconstruction |
Keyword(3) | activation function |
1st Author's Name | Koji Kamma |
1st Author's Affiliation | Wakayama University(Wakayama Univ.) |
2nd Author's Name | Toshikazu Wada |
2nd Author's Affiliation | Wakayama University(Wakayama Univ.) |
Date | 2022-12-15 |
Paper # | PRMU2022-38 |
Volume (vol) | vol.122 |
Number (no) | PRMU-314 |
Page | pp.pp.34-39(PRMU), |
#Pages | 6 |
Date of Issue | 2022-12-08 (PRMU) |