| アイテムタイプ |
学術雑誌論文 / Journal Article(1) |
| 公開日 |
2023-12-27 |
| タイトル |
|
|
タイトル |
Automated Quantization and Retraining for Neural Network Models Without Labeled Data |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Automated machine learning |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
model compression |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
model retraining |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
multi-objective optimization |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
edge machine learning |
| 資源タイプ |
|
|
資源タイプ |
journal article |
| アクセス権 |
|
|
アクセス権 |
open access |
| 著者 |
Thonglek, Kundjanasith
Takahashi, Keichi
市川, 昊平
Nakasan, Chawanat
Nakada, Hidemoto
Takano, Ryousei
Leelaprute, Pattara
飯田, 元
|
| 抄録 |
|
|
内容記述タイプ |
Abstract |
|
内容記述 |
Deploying neural network models to edge devices is becoming increasingly popular because such deployment decreases the response time and ensures better data privacy of services. However, running large models on edge devices poses challenges because of limited computing resources and storage space. Researchers have therefore proposed various model compression methods to reduce the model size. To balance the trade-off between model size and accuracy, conventional model compression methods require manual effort to find the optimal configuration that reduces the model size without significant degradation of accuracy. In this article, we propose a method to automatically find the optimal configurations for quantization. The proposed method suggests multiple compression configurations that produce models with different size and accuracy, from which users can select the configurations that suit their use cases. Additionally, we propose a retraining method that does not require any labeled datasets for retraining. We evaluated the proposed method using various neural network models for classification, regression and semantic similarity tasks, and demonstrated that the proposed method reduced the size of models by at least 30% while maintaining less than 1% loss of accuracy. We compared the proposed method with state-of-the-art automated compression methods, and showed that it can provide better compression configurations than existing methods. |
| 書誌情報 |
en : IEEE Access
巻 10,
p. 73818-73834,
発行日 2022-07-13
|
| 出版者 |
|
|
出版者 |
IEEE |
| ISSN |
|
|
収録物識別子タイプ |
EISSN |
|
収録物識別子 |
2169-3536 |
| 出版者版DOI |
|
|
関連タイプ |
isReplacedBy |
|
|
識別子タイプ |
DOI |
|
|
関連識別子 |
https://doi.org/10.1109/ACCESS.2022.3190627 |
| 出版者版URI |
|
|
関連タイプ |
isReplacedBy |
|
|
識別子タイプ |
URI |
|
|
関連識別子 |
https://ieeexplore.ieee.org/document/9828404 |
| 権利 |
|
|
権利情報Resource |
https://creativecommons.org/licenses/by/4.0/ |
|
権利情報 |
IEEE is not the copyright holder of this material. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ |
| 著者版フラグ |
|
|
出版タイプ |
NA |
