Evaluación del rendimiento de redes neuronales en dispositivos edge mediante el framework NC-SDK
In the world of Internet of Things (IoT), edge computing has increased its popularity in recent years. This study describes the Neural Compute Software Development Kit (NCSDK) package, which contains a set of development tools in order to be able to adapt, compile and run inferences from different d...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2024 |
| País: | España |
| Institución: | Universidad Complutense de Madrid (UCM) |
| Repositorio: | Docta Complutense |
| Idioma: | inglés |
| OAI Identifier: | oai:docta.ucm.es:20.500.14352/109073 |
| Acceso en línea: | https://hdl.handle.net/20.500.14352/109073 |
| Access Level: | acceso abierto |
| Palabra clave: | 004(043.3) NC-SDK PowerVR Redes neuronales Rendimiento Inferencia Neural networks Performance Inference Informática (Informática) 33 Ciencias Tecnológicas |
| Sumario: | In the world of Internet of Things (IoT), edge computing has increased its popularity in recent years. This study describes the Neural Compute Software Development Kit (NCSDK) package, which contains a set of development tools in order to be able to adapt, compile and run inferences from different deep learning models on certain specific devices. This project demonstrates the workflow with the NC-SDK package, highlighting the processes of adapting deep learning models from various implementations, compiling neural networks for execution on the deployment platform, and executing distinct models using the PowerVR GPU. Finally, it is conducted a performance benchmark on different ML categories, such as image classification, object detection and segmentation, and language processing. The obtained results will be compared and analyzed with CPU usage by other frameworks. |
|---|