Producción científica

URI permanente para esta comunidad

hdl:123456789/1

Examinar

Mostrando 1 - 6 de 6

Enhancing Location Entity Recognition in Spanish Medical Texts by Leveraging Domain Language Models and Data Augmentation
(Sociedad Española para el Procesamiento del Lenguaje Natural, 2026-03-01) Garitano López de Uralde, Irati; Martínez Unanue, Raquel; Ministerio de Ciencia, Innovación y Universidades (España)
This work focuses on the automatic recognition of location entities in Spanish clinical reports, using the MEDDOPLACE challenge (IberLEF 2023) as the experimental framework. We evaluated both general-domain pre-trained models and biomedical-specific models. Furthermore, we explored data augmentation techniques via back-translation and LLM-based paraphrase generation. Our results outperform previous state-of-the-art approaches, demonstrating the effectiveness of combining these data augmentation strategies with pre-trained clinical domain models.
Explainable AI for predicting household demand flexibility: Insights from smart meter data and price-based programs
(Elsevier, 2026-01-27) Bañales, Santiago; Dormido Canto, Raquel; Duro Carralero, Natividad
Unlocking Demand‐Side Flexibility (DSF) at scale is essential for integrating variable renewables and electrified end-uses. We develop a scalable, explainable-AI framework to assess the predictability and drivers of household responsiveness to price-based programs using only data typically available to utilities (smart meters, basic weather, limited socio-economic tags). Using the public Low Carbon London Time-of-Use (ToU) pilot, we first estimate responsiveness with Least Absolute Shrinkage and Selection Operator (LASSO) at both aggregated and household levels—overall and by hour—to quantify effect sizes and heterogeneity. We then train Gradient-Boosting (GB) models and apply SHapley Additive exPlanations (SHAP) to assess the hierarchy and direction of drivers of flexibility. Results show statistically significant but moderate average responses with wide dispersion across households and time-of-day, including a significant percentage of counter-intuitive reactions to price. Features capturing unexplained variability in hourly and daily load (e.g., dispersion measures of residual components) are the strongest positive predictors of flexibility, whereas seasonality/predictability indicators (autocorrelation and seasonal strength) are neutral or negative. SHAP dependence plots reveal clear thresholds, breakpoints, and saturation effects, underscoring the nonlinearity of behavioral response. Because the feature set is derived from routinely collected data, the approach is replicable and operationally practical. The findings enable data-driven targeting of high-potential households and support the design of digital orchestration platforms for near-time demand response, informing tariff design, aggregator strategies, and regulatory guidance for market-based DSF.
Low-Cost IoT and LoRaWAN-Based System for Laying Hen Identification in Family Poultry Farms
(MDPI, 2025-04-27) Finistrosa, Roberto; Mañoso Hierro, María Carolina; Pérez de Madrid y Pablo, Ángel; Romero Hortelano, Miguel
In medium- and large-scale poultry farms, automated systems optimize key processes, from egg production and grading to environmental control, reducing manual labor and ensuring an optimal environment for the birds. However, these technologies remain largely inaccessible to small family farms due to high implementation costs. In particular, the selection of laying hens, an essential process for productivity, is still performed manually and requires considerable time and effort. This study presents the development of a modular, low-cost, and minimally invasive IoT system for the automatic detection of laying hens in family-run poultry farms. Additionally, the system enables environmental monitoring and utilizes LoRaWAN networks for efficient long-range data transmission. The collected data are stored on a centralized platform and integrated with web, mobile, and messaging applications to provide real-time access to information. The modular system architecture, developed using open-source software, ensures replicability, scalability, and adaptability to different production environments. The feasibility of the system has been validated through field trials in a real-world environment, demonstrating effective performance, low implementation costs, and high farmer satisfaction, with the user highlighting its positive impact on poultry farm management.
A model-based smart meters time series decomposition approach for demand flexibility characterization of SMEs and households
(Elsevier, 2026-03-10) Bañales, Santiago; Dormido Canto, Raquel; Duro Carralero, Natividad
Demand flexibility is a fundamental service in the current energy system transformation contributing to balance intermittent renewable generation with increasing electrified demand. This paper proposes an innovative model-based methodology for smart meters time series decomposition to characterize and segment customers for demand flexibility programs. First, the daily energy times are split into normalized daily and hourly profiles, and periods of structurally low energy use are identified. The series is then decomposed using a dynamic regression and ARIMA-GARCH approach, resulting in residuals that follow a non-Gaussian white noise distribution. Next, complexity reduction for normalized hourly energy use is achieved via a regression-based feature engineering model, which feeds into a k-means clustering procedure to determine similar hourly energy profile patterns for working week and weekend days. The decomposition yields a cohesive set of data-driven, baseline-free, and explainable metrics that quantify and characterize the demand flexibility potential of each customer and cluster. These metrics capture key dimensions such as flexibility quantity, variability, reversion speed, calendar effects, and predictability. The methodology is validated on a real publicly available dataset of Irish households and SMEs customers. The results highlight the robustness and replicability of the approach while providing actionable insights and comparison between the two customer segments. This approach enables energy companies, engaged citizens and other stakeholders to design and deploy effective demand flexibility strategies in the energy industry.
On the optimal selection of Mel-Frequency Cepstral Coefficients for voice deepfake detection
(Wiley, 2026-03-24) Falcón López, Sergio A.; Tobarra Abad, María de los Llanos; Robles Gómez, Antonio; Pastor Vargas, Rafael
The continuous evolution of techniques for generating manipulated audio, known as voice deepfakes, and the widespread availability of tools that produce convincing forgeries have created an urgent need for reliable detection methods. This work considers the dimensionality of Mel-Frequency Cepstral Coefficients (MFCCs) as a core design variable for practical, deployable systems. The aim is to identify the smallest number of coefficients that preserves detection performance across heterogeneous models while reducing computational cost, a critical factor for mobile and edge deployment. This study evaluates a hybrid setting on the ASVspoof 2019 Logical Access dataset, in which the same feature family serves as input to five traditional machine learning algorithms (Random Forest, k-Nearest Neighbors, Linear Support Vector Classification, Extreme Gradient Boosting and Support Vector Machine with radial basis function kernel) and five deep learning models (Convolutional Neural Network, Recurrent Neural Network, Convolutional Recurrent Neural Network, Xception and ResNet). Results indicate that deep models reach near-peak performance with a small number of coefficients, whereas classical methods require a larger number to achieve stable performance (except Linear Support Vector Classification, which consistently underperforms). Accordingly, 32 coefficients are considered an effective operating point for hybrid deployments. Overall, the results provide evidence to guide the selection of the number of MFCC coefficients in voice deepfake detection, aiming for efficient, reproducible and explainable systems.
Towards the Integration of Remote Laboratories in Massive Open Online Courses: Insights from Usage Logs and Student Feedback
(Wiley, 2026-03-05) Gómez, Manuel J.; Albaladejo González, Mariano; Ruipérez Valiente, José A.; Rejón Gómez, Carlos; García Loro, Félix; Robles Gómez, Antonio; Martín Gutiérrez, Sergio; Agencia Estatal de Investigación (España)
Massive Open Online Courses (MOOCs) have established a new paradigm in education, enabling asynchronous, remote learning. Although MOOCs offer diverse educational content to students, comprehensive and realistic education requires hands-on training. Therefore, we present our integration of remote laboratories within a MOOC focused on Industry 4.0, together with a mixed-methods analysis combining demographic data, questionnaire responses, platform logs, and remote laboratory server records. The course incorporated seven distinct laboratories in two categories: remote Arduino-based laboratories, which connected physical devices to the MOOC, and a virtual infrastructure based on JupyterHub to support practical activities. Our findings indicate that the majority of participants in our MOOC were actively employed, with a mean age of 46.9 years, and males representing 88%. The students were mainly motivated by personal and professional growth, and 66% of these learners had no prior remote laboratory experience. Learners reported generally positive perceptions of the laboratory experience, including high levels of interest, usefulness, and self-efficacy, although engagement varied substantially across participants. Behavioral analyses revealed that students who ultimately pursued certification showed markedly higher participation and involvement in the remote laboratories than those who did not. At the same time, the course exhibited the substantial dropout patterns commonly reported in MOOCs, indicating that the inclusion of remote laboratories does not by itself eliminate persistence challenges. Our work provides an illustrative example of integrating remote laboratories into a MOOC and offers an analysis of enrolled students, delivering valuable findings for future researchers.

Examinar

Examinando Producción científica por Centro "Escuela Técnica Superior de Ingeniería Informática"

Resultados por página

Opciones de ordenación