Principal component analysis hardware acceleration
Since machine learning is getting more attention in various applications, the performance of those applications has become the main concern of its users. To perform machine learning, one of the vital processes is feature extraction which is to reduce the raw data dimension that aims to get rid of no...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/93003/1/NgYeeWeiMSKE2020.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Since machine learning is getting more attention in various applications, the performance of those applications has become the main concern of its users. To perform machine learning, one of the vital processes is feature extraction which is to reduce the raw data dimension that aims to get rid of noise and speed up further data analysis. Principal Component Analysis (PCA) is one of dimension reduction techniques that often used with other complex feature extraction techniques. However, PCA involves heavy computation and plays an important role to determine the speed performance of the application. This project is to propose PCA hardware acceleration to enhance its performance. From software profiling, the most intensive function in the PCA algorithm is the computation of eigenvalues and eigenvectors (eigensolver). Thus, this project has developed an eigensolver hardware accelerator by applying parallel execution through unrolling, pipelining and scheduling techniques in order to improve the performance of PCA. The proposed eigensolver is developed using Vivado HLS 2019.2. The performance of the proposed accelerator is evaluated by comparing it with conventional PCA hardware. The proposed eigensolver hardware accelerator has achieved a speedup of 6.27 compared with its conventional implementation. |
|---|