Open Access
Mejoras a la capacidad de generalización de la inteligencia artificial
(2023) Fumanal Idocin, Javier; Bustince Sola, Humberto; Cordón García, Óscar; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
La fusión de información es un aspecto crucial del análisis moderno de datos y la toma de decisiones. Implica la integración de múltiples fuentes de información para obtener una com prensión más completa y precisa de un tema determinado. Este proceso es especialmente importante en campos como la informática, la ingeniería y las ciencias naturales, donde se generan grandes cantidades de datos procedentes de diversas fuentes que deben sintetizarse para tomar decisiones con conocimiento de causa. La fusión de información también es esencial en el diseño y la implantación de sistemas inteligentes, ya que permite integrar diversos sensores y fuentes de datos para hacer predicciones y recomendaciones más precisas. Desde un punto de vista matemático, una forma de estudiar este problema es a través de la idea de funciones de fusión, que toman como entrada un vector de números y devuelven uno solo, representativo de ellos. Un tipo relevante de funcion de fusión es la familia de funciones de agregación. Estas funciones mantienen dos condiciones de contorno y monotonicidad con respecto a las entradas, que inducen algunas propiedades deseables a la salida de la función. Sin embargo, la fusión de información en los sistemas aplicados comprende algo más que esta noción teórica. A medida que la heterogeneidad, la estructura y el volumen de los datos adquieren mayor relevancia, han surgido otros enfoques para abordar este problema. Por ejemplo, en una estructura de red, las distintas entradas se asocian entre sí según un conjunto preestablecido de relaciones; en las series temporales, los datos presentan dependencias temporales. Cuando se trata de datos no estructurados, como texto, audio e imagen, los enfoques de aprendizaje profundo han tenido mucho exito en la transformación de este tipo de datos en representaciones vectoriales de números reales utilizando series de transformaciones afines. A pesar de los esfuerzos previos en este campo, el problema de combinar eficazmente fuentes de información diversas y heterogéneas, sigue siendo un área de investigación abierta y activa. Esto se debe a los desafíos inherentes a la integracion de múltiples fuentes que pueden estar en diferentes formatos y pueden tener información contradictoria o incompleta. Por ejemplo, el modo en que la información medida se relaciona con otras fuentes de datos y la fiabilidad de esas medidas dependen en gran medida del procedimiento de medición. De hecho, los sistemas que fusionan la información de esas distintas fuentes presentarán también complejidades adicionales al tener en cuenta las particularidades de cada característica considerada. En esta tesis, proponemos un conjunto de funciones y algoritmos para tener en cuenta las posibles interacciones, heterogeneidades e incertidumbres cuando se trabaja con distintas fuentes de información. Lo hacemos mediante la teoría de agregaciones y el análisis de redes sociales, y nos centramos especialmente en aquellos casos en los que los enfoques de aprendizaje profundo no tienen tanto éxito. Aplicamos estos resultados a una amplia gama de problemas, incluyendo la clasificación de se ñales de interfaz cerebro-ordenador, la clasificación de datos tabulares estándar y la detección de anomalías.
Open Access
On the stability of fuzzy classifiers to noise induction
(IEEE, 2023-11-09) Fumanal Idocin, Javier; Bustince Sola, Humberto; Andreu-Pérez, Javier; Hagras, Hani; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
Tabular data classification is one of the most important research problems in the artificial intelligence. One of the most important desired properties of the ideal classifier is that small changes in its input should not result in dramatic changes in its output. However, this might not be the case for many classifiers used in present day. Fuzzy classifiers should be stronger than their crisp counterparts, as they should be able to handle such changes using fuzzy sets and their membership functions. However, this hypothesis has not been empirically tested. Besides, the concept of 'small change' is somewhat imprecise and has not been quantified yet. In this work we propose to use small and progressively bigger changes in test samples to study how different crisp and fuzzy classifiers behave. We also study how to optimize classifiers to be more resistant to such kind of changes. Our results show that different fuzzy sets have different responses to this problem and have a smoother performance response compared to crisp classifiers. We also studied how to improve this and found that resistance to small changes can also result in a worse overall performance.
Open Access
The Krypteia ensemble: designing classifier ensembles using an ancient Spartan military tradition
(Elsevier, 2023) Fumanal Idocin, Javier; Cordón, Óscar; Bustince Sola, Humberto; Institute of Smart Cities - ISC; Universidad Pública de Navarra / Nafarroako Unibertsitate Publikoa
In this work we propose a new algorithm to train and optimize an ensemble of classifiers. We call this algorithm the Krypteia ensemble, based on an ancient Spartan tradition designed to convert their most promising individuals into future leaders of their society. We show how to adapt this ancient custom to optimize classifiers by generating different variations of the same task, each one offering different hardships according to distinct stochastic variables. This is thus applied to induce diversity in the set of individual weak learners. Then, we use a set of agents designed to select those subjects who excel in their assignments, and whose interaction minimizes excessive redundancies in the resulting population. We also study how different Krypteia ensembles can be stacked together, so that more complex classifiers can be built using the same procedure. Besides, we consider a wide range of different aggregation functions in the decision making phase to find the optimal performance for the different Krypteia ensemble variations tested. Finally, we study how different Krypteia ensembles perform for a wide range of classification datasets and we compare them with other state-of-the-art design techniques of classifier ensembles, obtaining favourable results to our proposal.
Open Access
Combinations of affinity functions for different community detection algorithms in social networks
(University of Hawaii Press, 2021) Fumanal Idocin, Javier; Cordón, Óscar; Minárová, María; Alonso Betanzos, Amparo; Bustince Sola, Humberto; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
Social network analysis is a popular discipline among the social and behavioural sciences, in which the relationships between different social entities are modelled as a network. One of the most popular problems in social network analysis is finding communities in its network structure. Usually, a community in a social network is a functional sub-partition of the graph. However, as the definition of community is somewhat imprecise, many algorithms have been proposed to solve this task, each of them focusing on different social characteristics of the actors and the communities. In this work we propose to use novel combinations of affinity functions, which are designed to capture different social mechanics in the network interactions. We use them to extend already existing community detection algorithms in order to combine the capacity of the affinity functions to model different social interactions than those exploited by the original algorithms.
Open Access
A rule-based approach for interpretable intensity-modulated radiation therapy treatment selection
(IEEE, 2024-08-05) González García, Xabier; Fumanal Idocin, Javier; Nunez do Rio, Joan M.; Bustince Sola, Humberto; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
Artificial Intelligence (AI) methods are becoming essential in healthcare. In the context of Intensity-Modulated Radiation Therapy (IMRT), Knowledge-Based Planning (KBP) methodologies have enabled the modification of treatments in real-time to accommodate morphological changes in patients. KBP for IMRT is a data-driven approach that utilises real-time medical imaging to adjust the radiation dose for a patient as needed for the different stages of an illness. In this work we present an interpretable AI model that selects the best IMRT treatment alternatives and determines which is the best. We use an Adaptive Neuforuzzy Adaptive Inference System (ANFIS), which combines the potential of a neural network with the interpretability of a rule based system. We train the model in a supervised manner using the OpenKBP challenge data repository. For this purpose, we also developed a data augmentation method that is supported by Diffusion Probabilistic Models. This approach enables the generation of a wider spectrum of treatment qualities and aids regularisation. The primary advantage of this framework resides in its ability to offer explanations, which is essential in the deployment of medical procedures in real life. Moreover, it serves as a valuable means to test hypotheses concerning the quality of IMRT treatments. Our study reveals that the developed tool has substantial potential to establish itself as a reference in the realm of explainable IMRT treatment selection tools.
Open Access
ARTxAI: explainable artificial intelligence curates deep representation learning for artistic images using fuzzy techniques
(IEEE, 2024) Fumanal Idocin, Javier; Andreu-Pérez, Javier; Cordón, Óscar; Hagras, Hani; Bustince Sola, Humberto; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
Automatic art analysis employs different image processing techniques to classify and categorize works of art. When working with artistic images, we need to take into account further considerations compared to classical image processing. This is because artistic paintings change drastically depending on the author, the scene depicted, and their artistic style. This can result in features that perform very well in a given task but do not grasp the whole of the visual and symbolic information contained in a painting. In this article, we show how the features obtained from different tasks in artistic image classification are suitable to solve other ones of similar nature. We present different methods to improve the generalization capabilities and performance of artistic classification systems. Furthermore, we propose an explainable artificial intelligence method to map known visual traits of an image with the features used by the deep learning model considering fuzzy rules. These rules show the patterns and variables that are relevant to solve each task and how effective is each of the patterns found. Our results show that compared to multitask learning, our proposed context-aware features can achieve up to 19% more accurate results when using the residual network architecture and 3% when using ConvNeXt. We also show that some of the features used by these models can be more clearly correlated to visual traits in the original image than other kinds of features.
Open Access
Fuzzy clustering to encode contextual information in artistic image classification
(Springer, 2022) Fumanal Idocin, Javier; Takáč, Zdenko; Horanská, Lubomíra; Bustince Sola, Humberto; Cordón, Óscar; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
Automatic art analysis comprises of utilizing diverse processing methods to classify and categorize works of art. When working with this kind of pictures, we have to take under consideration different considerations compared to classical picture handling, since works of art alter definitely depending on the creator, the scene delineated or their aesthetic fashion. This extra data improves the visual signals gotten from the images and can lead to better performance. However, this information needs to be modeled and embed alongside the visual features of the image. This is often performed utilizing deep learning models, but they are expensive to train. In this paper we utilize the Fuzzy C-Means algorithm to create a embedding strategy based on fuzzy memberships to extract relevant information from the clusters present in the contextual information. We extend an existing state-of-the-art art classification system utilizing this strategy to get a new version that presents similar results without training additional deep learning models.
Open Access
Clusterig cosmológico: un enfoque del clustering gravitacional clásico inspirado en la estructura y dinámica del cosmos a gran escala
(Universidad de Málaga, 2021) Castillo López, Aitor; Fumanal Idocin, Javier; Fernández Fernández, Francisco Javier; Bustince Sola, Humberto; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
En este trabajo proponemos un nuevo enfoque del algoritmo de clustering gravitacional basado en lo que Einstein considero su 'mayor error': la constante cosmológica. De manera similar al algoritmo de clustering gravitacional, nuestro enfoque está inspirado en principios y leyes del cosmos, y al igual que ocurre con la teoría de la relatividad de Einstein y la teoría de la gravedad de Newton, nuestro enfoque puede considerarse una generalización del agrupamiento gravitacional, donde, el algoritmo de clustering gravitacional se recupera como caso límite. Además, se desarrollan e implementan algunas mejoras que tienen como objetivo optimizar la cantidad de iteraciones finales, y de esta forma, se reduce el tiempo de ejecución tanto para el algoritmo original como para nuestra versión.
Open Access
d-XC integrals: on the generalization of the expanded form of the Choquet integral by restricted dissimilarity functions and their applications
(IEEE, 2022) Wieczynski, Jonata; Fumanal Idocin, Javier; Lucca, Giancarlo; Borges, Eduardo N.; Da Cruz Asmus, Tiago; Emmendorfer, Leonardo R.; Bustince Sola, Humberto; Pereira Dimuro, Graçaliz; Automática y Computación; Automatika eta Konputazioa; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
Restricted dissimilarity functions (RDFs) were introduced to overcome problems resulting from the adoption of the standard difference. Based on those RDFs, Bustince et al. introduced a generalization of the Choquet integral (CI), called d-Choquet integral, where the authors replaced standard differences with RDFs, providing interesting theoretical results. Motivated by such worthy properties, joint with the excellent performance in applications of other generalizations of the CI (using its expanded form, mainly), this paper introduces a generalization of the expanded form of the standard Choquet integral (X-CI) based on RDFs, which we named d-XC integrals. We present not only relevant theoretical results but also two examples of applications. We apply d-XC integrals in two problems in decision making, namely a supplier selection problem (which is a multi-criteria decision making problem) and a classification problem in signal processing, based on motor-imagery brain-computer interface (MI-BCI). We found that two d-XC integrals provided better results when compared to the original CI in the supplier selection problem. Besides that, one of the d-XC integrals performed better than any previous MI-BCI results obtained with this framework in the considered signal processing problem.
Open Access
Análisis de redes sociales basado en las conquistas de César Borgia
(Universidad de Málaga, 2021) Fumanal Idocin, Javier; Cordón, Óscar; Alonso Betanzos, Amparo; Bustince Sola, Humberto; Fernández Fernández, Francisco Javier; Estadística, Informática y Matemáticas; Estatistika, Informatika eta Matematika
En este trabajo presentamos el modelado de redes sociales y detección de comunidades utilizando como base un evento histórico real, las conquistas de César Borgia en el siglo XV. Para ello, proponemos un nuevo conjunto de funciones, llamadas funciones de afinidad, disenadas para capturar la 'naturaleza de las interacciones locales entre cada par de actores en una red. Utilizando estas funciones, desarrollamos un nuevo algoritmo de detección de comunidades, el Borgia Clustering, donde las comunidades surgen naturalmente de un proceso de simulación de interacción de múltiples agentes en la red. También discutimos los efectos del tamaño y la escala de cada comunidad, y como pueden ser tomadas en cuenta en el proceso de simulación. Finalmente, comparamos nuestra detección de comunidades con otros algoritmos representativos, encontrando resultados favorables a nuestra propuesta.