Explainable artificial intelligence for the analysis of deep learning models applied to outdoor images
Loading...
Download
Official URL
Full text at PDC
Publication date
2026
Defense date
17/06/2025
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Complutense de Madrid
Citation
Abstract
El aprendizaje profundo, en particular las redes neuronales convolucionales, se ha convertido en una herramienta indispensable en el campo de la visión por computador. Dado que este último campo abarca multitud de áreas de investigación y aplicación, es vital adaptar y mejorar los modelos de aprendizaje profundo utilizados. Para poder sacar el máximo partido a las redes neuronales, y lograr la mayor eficacia y operatividad, las modificaciones realizadas han de estar fundamentadas. Para ello, y debido a la alta complejidad y opacidad de estos modelos, es necesario analizar su comportamiento y comprender su proceso de aprendizaje. El campo de la inteligencia artificial explicativa se encarga de esta tarea, desarrollando estrategias para explicar, interpretar y entender los modelos de la caja negra.Esta tesis pretende ir más allá de las redes neuronales convolucionales, profundizando en su funcionamiento interno y aprovechando los conocimientos adquiridos para adaptarlas según las necesidades. Centrado en la difícil tarea del procesamiento de imágenes de exterior, este trabajo analiza el proceso de razonamiento de las redes neuronales con el fin de modificarlas, validarlas y mejorarlas, considerando su despliegue en dispositivos con recursos de cómputo y memoria limitados...
Deep learning algorithms, specially Convolutional Neural Networks (CNNs), have become an indispensable tool in computer vision. Given that this field encompasses multiple areas of research and application, it is vital to adapt and improve deep learning models. To make the most of convolutional neural networks, and achieve their greatest efficiency and operability, informed modifications should be performed. To this end, and due to the high complexity and opaque nature of deep learning models, analyses of their behaviour and comprehension of their learning process are necessary. The field of eXplainable Artificial Intelligence (XAI) is in charge of this task, developing strategies to explain, interpret, and understand black-box models.This research thesis aims to go beyond CNNs, delving into their inner workings and leveraging the insghts gained to adapt them as needed. Focusing on the challenging task of outdoor image processing, this work analyzes the reasoning process of convolutional neural networks in order to modify, validate, and improve them, considering their deployment on devices with limited computing power and memory...
Deep learning algorithms, specially Convolutional Neural Networks (CNNs), have become an indispensable tool in computer vision. Given that this field encompasses multiple areas of research and application, it is vital to adapt and improve deep learning models. To make the most of convolutional neural networks, and achieve their greatest efficiency and operability, informed modifications should be performed. To this end, and due to the high complexity and opaque nature of deep learning models, analyses of their behaviour and comprehension of their learning process are necessary. The field of eXplainable Artificial Intelligence (XAI) is in charge of this task, developing strategies to explain, interpret, and understand black-box models.This research thesis aims to go beyond CNNs, delving into their inner workings and leveraging the insghts gained to adapt them as needed. Focusing on the challenging task of outdoor image processing, this work analyzes the reasoning process of convolutional neural networks in order to modify, validate, and improve them, considering their deployment on devices with limited computing power and memory...
Description
Tesis inédita de la Universidad Complutense de Madrid, Facultad de Informática, Departamento de Ingeniería del Software e Inteligencia Artificial, leída el 17 de junio de 2025.













