GPU Implementation of Sparse Matrice on Artificial Intelligence
| Autor: | Hsu, Chun-Kai, 許俊凱 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 In recent years, artificial intelligence AI (Artificial Intelligence) has become one of the most popular research, it refers to computer software through statistical analysis and rule reasoning of datasets. Machine Learning ML (Machine Learning) is one of the branches of AI development. The ML classifies the message according to the information and category rules contained in the neurons in the neural network, and uses the probability of calculating the data to determine the prediction or Decision. Support vector machine (SVM), which belongs to supervised learning in ML, is one of the learning methods in AI today. With the evolution of hardware technology, graphics processing unit GPU (Graphics Processing Unit) and CUDA (Compute Unified Device Architecture) technology have also emerged. Foreign scholars have proposed a parallelization method for SVM called OHD-SVM (Optimized Hierarchical Decomposition SVM). In order to improve the computational efficiency, a sparse matrix format conversion is designed for the difference of the characteristics of the data set. The paper's research on OHD-SVM found that the sparse matrix format used by it has room for improvement on the CUDA platform. Therefore, for this problem, we use a sparse matrix format that is more suitable for the CUDA platform to achieve the performance improvement of OHD-SVM. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|