Examinando por Autor "Moguerza, Javier M."

Mostrando 1 - 5 de 5

Adapting support vector optimisation algorithms to textual gender classification
(Springer, 2024-04-13) Gómez, Javier; Alfaro, Cesar; Ortega, Felipe; Moguerza, Javier M.; Algar, Maria Jesus; Moreno, Raul
In this paper, we focus on the problem of determining the gender of the person described in a biographical text. Since support vector machine classifiers are well suited for text classification tasks, we present a new stopping criterion for support vector optimisation algorithms tailored to this problem. This new approach exploits the geometric properties of the vector representation of such content. An experiment on a set of English and Spanish biographical articles retrieved from Wikipedia illustrates this approach and compares it to other machine learning classification algorithms. The proposed method allows real-time classification algorithm training. Moreover, these results confirm the advantage of leveraging additional gender information in strongly inflected languages, like Spanish, for this task
Building robust morphing attacks for face recognition systems
(Institute Of Electrical And Electronics Engineers Inc., 2023) Gallardo Cava, Roberto; Ortega del Campo, David; Palacios Alonso, Daniel; Conde, Cristina; Moguerza, Javier M.; Cabello, Enrique
En este artículo se presenta un método para construir un ataque de morphing robusto a un sistema de verificación facial. El método propuesto ha sido desarrollado para investigar la robustez y el impacto de los ataques de morphing en los sistemas de reconocimiento facial. En este tipo de ataque, un impostor accede a un sistema de reconocimiento facial (FRS) que compara su imagen en tiempo real con una imagen morphed almacenada, construida con el impostor y un usuario legítimo. El ataque tiene éxito cuando el FRS acepta al impostor y accede al sistema. El enfoque actual ofrece un método para construir un ataque robusto al FRS, en el sentido de que la imagen morphed estará más cerca del umbral de decisión. Los ataques de morphing suelen evaluarse solo con imágenes en las que ambos sujetos contribuyen de la misma manera a las imágenes morphed. La base de datos de imágenes considerada, la Base de Datos FRAV, estaba compuesta por 200 imágenes. Asimismo, se llevaron a cabo dos etapas. La primera etapa fue diseñada para construir una referencia de línea base: se probó un sistema FRS (entrenado solo con usuarios legítimos) con imágenes morphed. Una contribución de este artículo es que esta prueba, que generalmente solo considera una fusión del 50% entre dos imágenes, se ha enriquecido y se han considerado algunas contribuciones de fusión. Se realizaron pruebas con contribuciones del 20%, 40%, 50%, 60% y 80% de cada imagen a la imagen morphed. La comparación de la tasa de error igual (EER) lograda mostrará qué contribución define el mejor ataque plausible. Es importante destacar que el ataque que logra las mejores tasas con la mínima perturbación de las imágenes. La segunda etapa consistió en el refuerzo del FRS, entrenándolo con el conjunto de contribuciones definido en la etapa anterior. Los resultados obtenidos lograron mejoras del 3% en las puntuaciones de EER.
General Performance Score for classification problems
(Springer, 2021) Martín De Diego, Isaac; Redondo, Ana R.; Fernández, Rubén R.; Navarro, Jorge; Moguerza, Javier M.
Several performance metrics are currently available to evaluate the performance of Machine Learning (ML) models in classifcation problems. ML models are usually assessed using a single measure because it facilitates the comparison between several models. However, there is no silver bullet since each performance metric emphasizes a diferent aspect of the classifcation. Thus, the choice depends on the particular requirements and characteristics of the problem. An additional problem arises in multi-class classifcation problems, since most of the well-known metrics are only directly applicable to binary classifcation problems. In this paper, we propose the General Performance Score (GPS), a methodological approach to build performance metrics for binary and multi-class classifcation problems. The basic idea behind GPS is to combine a set of individual metrics, penalising low values in any of them. Thus, users can combine several performance metrics that are relevant in the particular problem based on their preferences obtaining a conservative combination. Diferent GPS-based performance metrics are compared with alternatives in classifcation problems using real and simulated datasets. The metrics built using the proposed method improve the stability and explainability of the usual performance metrics. Finally, the GPS brings benefts in both new research lines and practical usage, where performance metrics tailored for each particular problem are considered.
Hostility measure for multi-level study of data complexity
(Springer, 2022) Lancho, Carmen; Martín De Diego, Isaac; Cuesta, Marina; Aceña, Víctor; Moguerza, Javier M.
Complexity measures aim to characterize the underlying complexity of supervised data. These measures tackle factors hindering the performance of Machine Learning (ML) classifiers like overlap, density, linearity, etc. The state-of-the-art has mainly focused on the dataset perspective of complexity, i.e., offering an estimation of the complexity of the whole dataset. Recently, the instance perspective has also been addressed. In this paper, the hostility measure, a complexity measure offering a multi-level (instance, class, and dataset) perspective of data complexity is proposed. The proposal is built by estimating the novel notion of hostility: the difficulty of correctly classifying a point, a class, or a whole dataset given their corresponding neighborhoods. The proposed measure is estimated at the instance level by applying the k-means algorithm in a recursive and hierarchical way, which allows to analyze how points from different classes are naturally grouped together across partitions. The instance information is aggregated to provide complexity knowledge at the class and the dataset levels. The validity of the proposal is evaluated through a variety of experiments dealing with the three perspectives and the corresponding comparative with the state-of-the-art measures. Throughout the experiments, the hostility measure has shown promising results and to be competitive, stable, and robust.
Stimulating children’s engagement with an educational serious videogame using Lean UX co-design
(Elsevier, 2021) Ramos-Vega, Maria C.; Palma-Morales, Victor M.; Pérez-Marín, Diana; Moguerza, Javier M.
The motivation to stimulate children’s learning engagement could be found in the fact that learning is not always motivational in itself. This is particularly true when learning is obligatory and based upon material that has not been chosen by the children themselves. A Lean UX approach to the co-design of an educational serious videogame (MOBI) is proposed in this paper. The core idea is that children's natural interest in playing can be stimulated by engendering the feeling that they are participating in the creation of something. The hypothesis is that this approach can increase the children's level of engagement and can facilitate their awareness of their learning perception. With the aim of testing this hypothesis, this paper describes an experience with 50 children with ages between 10 and 12 years old. The results indicate that the children’s satisfaction grew significantly during the process, with an important reduction in the requests for changes and that 60% had the perception of having learned. It can be concluded that the co-design based upon a Lean UX methodology, of a children’s educational serious videogame increases their level of product engagement and facilitates their awareness of their learning perception.