GPU accelerated singular value thresholding
Matrix completion (MC) is widely used in machine learning and signal processing to fill in the missing data of an incomplete observation matrix. Singular value thresholding (SVT) is one of the most popular algorithms among numerous MC methods. A Python-based GPU-accelerated SVT software is presented...
| 主要な著者: | , , |
|---|---|
| フォーマット: | Journal article |
| 言語: | English |
| 出版事項: |
Elsevier
2023
|
| 要約: | Matrix completion (MC) is widely used in machine learning and signal processing to fill in the missing data of an incomplete observation matrix. Singular value thresholding (SVT) is one of the most popular algorithms among numerous MC methods. A Python-based GPU-accelerated SVT software is presented in this paper. It is a user-friendly software package to minimise the nuclear norm with high accuracy and high computational efficiency. Its architecture and functionalities are illustrated, followed by a demonstration on how to use this software. Two examples, image inpainting and traffic sensing, are shown to illustrate potential applications of this software. Its impact on scientific and wider audiences is also analysed. |
|---|