Optimización de algoritmos de IA aplicando técnicas enfocadas al cómputo de alto rendimiento
Loading...
Download
Official URL
Full text at PDC
Publication date
2024
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
El trabajo que se presenta se enfoca en la optimización de algoritmos de Inteligencia Artificial (IA) mediante el uso de MPI (Message Passing Interface), una biblioteca estándar desarrollada para el cómputo de alto rendimiento. El objetivo principal consiste en reducir el tiempo de ejecución de los algoritmos, explotando el paralelismo de los recursos de cómputo y la memoria distribuida. Esta tarea es especialmente relevante debido al alto coste computacional y de recursos que implica entrenar o ejecutar estos algoritmos.
Este proyecto incluye una descripción de los fundamentos teóricos de los algoritmos que se van a implementar, así como el funcionamiento de la biblioteca MPI. Una vez puesto en contexto, se desarrollan en profundidad las estrategias propuestas para mejorar los algoritmos. Además, se ha realizado un exhaustivo estudio empírico para analizar las estrategias desarrolladas, las cuales han sido ejecutadas en un ordenador personal y en un sistema distribuido que consta de 128 núcleos de CPU y 256 GB de RAM.
The work presented focuses on the optimization of Arti cial Intelligence (AI) algorithms using MPI (Message Passing Interface), a standard library developed for high-performance computing. The main objective consists in reducing the execution time of the algorithms, by exploiting the parallelism of computing resources and distributed memory. This task is especially relevant due to the high computational and resource cost involved in training or running these algorithms This project includes a description of the theoretical foundations of the algorithms that will be implemented. Moreover, functioning of the MPI library is also presented. Once put in context, the strategies employed to enhance the algorithms are described in detail. In addition, an exhaustive empirical study has been carried out to analyze the developed strategies, which have been executed on a personal computer and in a high distributed system consisting of 128 CPU cores and 256 GB of RAM.
The work presented focuses on the optimization of Arti cial Intelligence (AI) algorithms using MPI (Message Passing Interface), a standard library developed for high-performance computing. The main objective consists in reducing the execution time of the algorithms, by exploiting the parallelism of computing resources and distributed memory. This task is especially relevant due to the high computational and resource cost involved in training or running these algorithms This project includes a description of the theoretical foundations of the algorithms that will be implemented. Moreover, functioning of the MPI library is also presented. Once put in context, the strategies employed to enhance the algorithms are described in detail. In addition, an exhaustive empirical study has been carried out to analyze the developed strategies, which have been executed on a personal computer and in a high distributed system consisting of 128 CPU cores and 256 GB of RAM.
Description
Trabajo de Fin de Grado en Ingeniería Informática, Facultad de Informática UCM, Departamento de Sistemas Informáticos y Computación, Curso 2023/2024