HLS-Based Large Scale Self-Organizing Feature Maps

Porrmann F, Hagemeyer J, Porrmann M (2024)
IEEE Access 12: 142459-142474.

Zeitschriftenaufsatz | Veröffentlicht | Englisch
 
Download
OA 7.33 KB
Abstract / Bemerkung
The Self-Organizing Map (SOM) algorithm is a clustering algorithm used in a wide variety of application domains. Over the last few decades, it has been accelerated using various hardware architectures, including FPGAs, CPUs, and GPUs. This publication presents an High-Level Synthesis-based implementation that utilizes multiple processing elements to realize a high-performance system architecture. An extensive design space exploration was conducted to evaluate the performance range of the architecture. For this, vector dimensions ranging from 8 up to 512 and map sizes from 16 × 16 to 512 × 512 were used. The evaluation was performed using two different AMD/Xilinx UltraScale+ FPGA systems, the VCU128 PCIe-based accelerator card and the ZCU106 stand-alone evaluation kit. From the achieved results, it can be seen that the performance scales nearly linearly for a given vector dimension when the map size is increased. In addition, the energy efficiency for both FPGAs was analyzed, revealing that while the ZCU106 is less powerful in terms of raw compute power, it requires up to 4× less power and, depending on the configuration, can be 2× more energy efficient compared to the VCU128. One of the main reasons for this is that it does not require a dedicated host system but utilizes its internal ARM cores. Finally, a comparison against state- of-the-art SOM implementations was conducted. The proposed design achieves a speed-up of up to 458.7, 1,630.4, and 4.9 compared to other CPU, GPU, and FPGA realizations, respectively.
Stichworte
Field programmable gate array; hardware acceleration; machine learning; reconfigurable architectures; reconfigurable computing; heterogeneous computing; heterogeneous architectures; self- organizing feature maps; optimization; design space exploration
Erscheinungsjahr
2024
Zeitschriftentitel
IEEE Access
Band
12
Seite(n)
142459-142474
eISSN
2169-3536
Finanzierungs-Informationen
Open-Access-Publikationskosten wurden durch die Universität Bielefeld gefördert.
Page URI
https://pub.uni-bielefeld.de/record/2993643

Zitieren

Porrmann F, Hagemeyer J, Porrmann M. HLS-Based Large Scale Self-Organizing Feature Maps. IEEE Access. 2024;12:142459-142474.
Porrmann, F., Hagemeyer, J., & Porrmann, M. (2024). HLS-Based Large Scale Self-Organizing Feature Maps. IEEE Access, 12, 142459-142474. https://doi.org/10.1109/ACCESS.2024.3471471
Porrmann, Florian, Hagemeyer, Jens, and Porrmann, Mario. 2024. “HLS-Based Large Scale Self-Organizing Feature Maps”. IEEE Access 12: 142459-142474.
Porrmann, F., Hagemeyer, J., and Porrmann, M. (2024). HLS-Based Large Scale Self-Organizing Feature Maps. IEEE Access 12, 142459-142474.
Porrmann, F., Hagemeyer, J., & Porrmann, M., 2024. HLS-Based Large Scale Self-Organizing Feature Maps. IEEE Access, 12, p 142459-142474.
F. Porrmann, J. Hagemeyer, and M. Porrmann, “HLS-Based Large Scale Self-Organizing Feature Maps”, IEEE Access, vol. 12, 2024, pp. 142459-142474.
Porrmann, F., Hagemeyer, J., Porrmann, M.: HLS-Based Large Scale Self-Organizing Feature Maps. IEEE Access. 12, 142459-142474 (2024).
Porrmann, Florian, Hagemeyer, Jens, and Porrmann, Mario. “HLS-Based Large Scale Self-Organizing Feature Maps”. IEEE Access 12 (2024): 142459-142474.
Alle Dateien verfügbar unter der/den folgenden Lizenz(en):
Creative Commons Namensnennung 4.0 International Public License (CC-BY 4.0):
Volltext(e)
Access Level
OA Open Access
Zuletzt Hochgeladen
2024-10-21T08:06:27Z
MD5 Prüfsumme
0f8ac5a9bd704638408daa3b5a457ae1


Export

Markieren/ Markierung löschen
Markierte Publikationen

Open Data PUB

Web of Science

Dieser Datensatz im Web of Science®
Suchen in

Google Scholar