|
Title:
|
EVALUATION OF GPU-CENTRIC EMBEDDED ARCHITECTURES IN THE FIELD OF NEUROMUSCULAR EMG-BASED AI |
|
Author(s):
|
Simon Pfenning, Raul C. Simpetru, Niklas Pollak, Alessandro Del Vecchio and Dietmar Fey |
|
ISBN:
|
978-989-8704-53-5 |
|
Editors:
|
Paula Miranda and Pedro Isaías |
|
Year:
|
2023 |
|
Edition:
|
Single |
|
Keywords:
|
EMG, CNN, Deep Learning, GPU, Embedded Hardware |
|
Type:
|
Full |
|
First Page:
|
39 |
|
Last Page:
|
46 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
The recent progress in the design of deep neural networks has led a huge step forward in inference accuracy and opened new opportunities also in the design of AI aided medical hardware. However deep neural models have a high demand in both computational as well as memory resources. To make these systems applicable in the field, it is necessary to find a suitable hardware-software ecosystem, which can meet these demands at a sufficiently low energy consumption. For this purpose, we evaluated a modern heterogeneous embedded platform with a deep convolutional network for hand position recognition, based on the collection of electromyography signals. The goal of our evaluation was to find out, how much effort must be spent for optimization, if the architecture itself would be sufficient for the usage as human wearable device and which of the available accelerators on the embedded platform are best suited for the task. |
|
|
|
|
|
|
