39936 research outputs found
Sort by
Applied Machine Learning Techniques in Maritime and Aerial Surveillance Systems
Full text linkMención Internacional en el título de doctor.Tesis por compendio de publicaciones.In recent years, advancements in data collection technologies and the proliferation of autonomous and remotely controlled systems have revolutionized the fields of surveillance, transportation, and mobility analysis. Vehicles operating across aerial, maritime, and terrestrial domains generate extensive trajectory data that encapsulates spatial, temporal, and contextual information. However, extracting meaningful insights from these trajectories presents significant challenges due to data variability, sensor inconsistencies, and the diverse environments in which the vehicles operate.
The difference present due to data variability between different trajectories and data sources, the different characteristics of the tracked vehicles, and the problematic existing on real-world scenarios (data corruption, misses, unbalance of instances, etc) all make essential the use of strong preprocessing techniques that are able to prepare the data for the information extraction algorithms. Also, strong analysis is needed to infer useful information from all the process.
This thesis explores the application of machine learning techniques to address key challenges in trajectory analysis, with a focus on enhancing surveillance capabilities and knowledge discovery. The study is structured around three core aspects: first, the data acquisition and preprocessing, secondly the trajectory modelling, and finally the extraction of useful insights through artificial intelligence techniques. Techniques such as data fusion estimation filters, trajectory segmentation, data balancing, trajectory reconstruction, clustering, and classification are employed to identify the detected target, uncover movements patterns, detect anomalies, and predict behaviours. Special attention is given to the coordination of multi-observer systems, optimizing their ability to track and analyse vehicle movements collaboratively.
Case studies across different application areas, including traffic management, maritime monitoring, and UAV-based surveillance demonstrate the proposed methods' versatility and practical utility. The research findings underline the potential of machine learning techniques to transform raw trajectory data into actionable knowledge, enabling improved decision-making and operational efficiency in complex surveillance environments.
The thesis is formed through a collection of studies related with the proposed approach. First, a data preparation process to manage real-world kinematic data and to achieve the detection of fishing vessels. The selected data for the experimentation are characterized by classic data mining problems that deal with real-world data such as noise, inconsistencies and data unbalance. The proposal is to use a feature extraction approach to model ship behaviour from which the information can be extracted through a classification approach. Second, a context information extraction process over build over the same data, building a system with the capability to extract representative points of a trajectory cluster and form a representative trajectory that can be used as context information for future maritime surveillance problems.
Finally, a swarm-based Unmanned Aerial Vehicle (UAV) system designed for surveillance tasks, specifically the system is designed for detecting and tracking ground vehicles. The proposal is to assess how a system consisting of multiple cooperating UAVs can enhance performance by utilizing fast detection algorithms. Instead of using high accuracy but slow algorithms the proposed system uses faster algorithms with lower accuracy that are improved due to the multiple sources approach of the swarm, allowing for a high accuracy real-time surveillance system.En los últimos años, los avances en tecnologías de recolección de datos y la proliferación de sistemas autónomos y controlados remotamente han revolucionado los campos de la vigilancia, el transporte y el análisis de movilidad. Los vehículos que operan en dominios aéreos, marítimos y terrestres generan extensos datos de trayectoria que encapsulan información espacial, temporal y contextual. Sin embargo, extraer información significativa de estas trayectorias presenta desafíos significativos debido a la variabilidad de los datos, inconsistencias de los sensores y los diversos entornos en los que operan los vehículos.
Las diferencias presentes debido a la variabilidad de los datos entre diferentes trayectorias y fuentes de datos, las características diversas de los vehículos rastreados y los problemas inherentes a los escenarios del mundo real (corrupción de datos, pérdidas, desequilibrios de instancias, etc.) hacen esencial el uso de técnicas de preprocesamiento robustas que puedan preparar los datos para los algoritmos de extracción de información. Además, es necesario un análisis sólido para inferir información útil de todo el proceso.
Esta tesis explora la aplicación de técnicas de aprendizaje automático para abordar los principales desafíos en el análisis de trayectorias, con un enfoque en mejorar las capacidades de vigilancia y el descubrimiento de conocimiento. El estudio se estructura en torno a tres aspectos fundamentales: adquisición y preprocesamiento de datos, modelado de trayectorias y extracción de información útil a través de técnicas de inteligencia artificial. Se emplean técnicas como filtros de estimación para fusión de datos, segmentación de trayectorias, balanceo de datos, reconstrucción de trayectorias, agrupamiento y clasificación para identificar objetivos detectados, descubrir patrones de movimiento, detectar anomalías y predecir comportamientos. Se presta especial atención a la coordinación de sistemas con múltiples observadores, optimizando su capacidad para rastrear y analizar los movimientos de los vehículos de manera colaborativa.
Estudios de caso en diferentes áreas de aplicación, incluyendo la gestión del tráfico, el monitoreo marítimo y la vigilancia basada en vehículos aéreos no tripulados (UAV), demuestran la versatilidad y utilidad práctica de los métodos propuestos. Los hallazgos de la investigación subrayan el potencial del aprendizaje automático para transformar datos de trayectorias en bruto en conocimiento accionable, permitiendo una mejor toma de decisiones y una mayor eficiencia operativa en entornos de vigilancia complejos.
La tesis se compone de una colección de estudios relacionados con el enfoque propuesto. En primer lugar, un proceso de preparación de datos para manejar datos cinemáticos del mundo real y lograr la detección de embarcaciones pesqueras. Los datos seleccionados para la experimentación se caracterizan por problemas clásicos de minería de datos que incluyen ruido, inconsistencias y desequilibrio de datos. La propuesta es utilizar un enfoque de extracción de características para modelar el comportamiento de los barcos, a partir del cual se pueda extraer información mediante un enfoque de clasificación. En segundo lugar, un proceso de extracción de información contextual construido sobre los mismos datos, desarrollando un sistema con la capacidad de extraer puntos representativos de un clúster de trayectorias y formar una trayectoria representativa que pueda utilizarse como información contextual para futuros enfoques.
Finalmente, un sistema basado en un enjambre de vehículos aéreos no tripulados (UAV) diseñado para tareas de vigilancia, específicamente para la detección y seguimiento de vehículos terrestres. La propuesta evalúa cómo un sistema compuesto por múltiples UAV cooperantes puede mejorar el rendimiento utilizando algoritmos de detección rápidos. En lugar de emplear algoritmos de alta precisión pero lentos, el sistema propuesto utiliza algoritmos más rápidos con menor precisión que se ven mejorados gracias al enfoque de múltiples fuentes del enjambre, permitiendo un sistema de vigilancia en tiempo real de alta precisión.I would also like to thank all the funding provided for the different projects of this thesis to the Ministry of Science and Innovation, the Spanish Ministry of Economy and Competitivity and the Madrid Government with the Multiannual Agreement with UC3M in the line of Excellence of University Professors. With references PID2020-118249RB-C22 and PDC2021-121567-C22 - AEI/10.13039/501100011033, PID2023-151605OB-C22, TEC2017-88048-C2-2-R, EPUC3M17. And the project under the call PEICTI 2021-2023 with identifier TED2021-131520B-C22.Programa de Doctorado en Ciencia y Tecnología Informática por la Universidad Carlos III de MadridPresidente: Javier Bajo Pérez. - Vocal: Daniel Arias Medina. - Secretario: Miguel Ángel Patricio Guisado
The role of aging in the microstructure and mechanical properties of two multi-principal element alloys
Full text linkHigh-entropy alloys or multi-principal element alloys (MPEAs) exhibit promising corrosion resistance in aggressive environments, making them viable candidates for high-temperature applications. However, in addition to the study of high-temperature corrosion behavior, understanding microstructural stability with respect to temperature and time is crucial, as microstructural modifications can influence the final properties of the samples. This study investigates the microstructural stability and mechanical properties of two MPEAs after aging treatments at 560 degrees C and 780 degrees C for over 1000 h. The first alloy, with an FCC-BCC eutectic microstructure (EMPEA), and the second, primarily BCC with a secondary B2 phase (MPEA6), were analyzed in as-cast and aged conditions. Upon aging at 560 degrees C, the hardness of EMPEA increased, likely due to the formation of nanosized coherent precipitates. However, at 780 degrees C, the microstructure coarsened, and the formation of a needle-like phase within the FCC matrix was observed, accompanied by a slight reduction in hardness, while the elastic modulus remained consistently high. MPEA6 exhibited an increase in both hardness and elastic modulus after aging at 560 degrees C, associated with the redistribution and slight alignment of the B2 precipitates. At 780 degrees C, these precipitates coalesced into elongated structures, particularly near grain boundaries, forming mesh-like patterns. This microstructural evolution led to a broader hardness distribution, with an overall decrease compared to the 560 degrees C treated sample, while the modulus showed an increase, suggesting a stiffening effect due to the precipitate's coalescence. These findings underscore the impact of thermal treatment on the microstructural evolution and mechanical properties of these MPEAs.This work has been supported by the Madrid Government (Comunidad de Madrid-Spain) under the Multiannual Agreement with UC3M (“Fostering Young Doctors Research”, HEATextreme-CM-UC3M), and in the context of the V PRICIT (Research and Technological Innovation Regional Program
Generation of a three-dimensional skin model based on a fibrin gel in a dynamic skin-on-a-chip platform
No full textThe use of animals as preclinical models has been an obligation before clinical trials in humans, serving to assess the efficacy and toxicity of specific molecules. This requirement also applies to other industries, such as cosmetics and textiles, which must also test their products on the skin to evaluate their safety. However, the prohibition of animal testing for cosmetic purposes, as well as the commercialisation of animal-tested products in the EU, and the application of this ban by other countries such as Canada or India, has created an urgent need for new in vitro testing models that closely replicate the structure and physiology of native tissues.
In the field of skin, which is of particular relevance for the cosmetic and textile industries, the development of human skin equivalents (HSEs) has advanced significantly over the last few years. These models include epidermal and full-thickness equivalents, as well as other more sophisticated models including the hypodermis, immune system components, pigmentation, or even disease models such as psoriasis, dermatitis or melanoma. However, current models face important limitations, such as the absence of a dynamic flow for nutrient delivery and waste removal and the poor mechanical properties of the dermo-epidermal models derived from the composition of the matrices used to mimic the dermal compartment.
Recent technological advances, including microfabrication and microfluidics, have driven the development of skin-on-a-chip systems, miniaturised dynamic models that better replicate the tissue microenvironment. Most of these on-chip skin models are based on cellular monolayers, although advances are being produced to generate the dermo-epidermal 3D structure of the skin within them. The systems including both layers of the skin are non-miniaturised systems that include microchannels for medium perfusion but rely on open structures in the chip for tissue generation rather than using the channels of the device.
This thesis proposes a 3D skin-on-a-chip system with three channels, incorporating a plasma-derived fibrin matrix supplemented with commercial fibrinogen as the dermal compartment. The matrix was confined between two porous membranes, with the epidermal layer constructed on top of one of them. A particle image velocimetry was performed to select the fluid flow rate to be applied on the chip. The system demonstrated stability during culture and influence of the dynamic system over cell proliferation. Nevertheless, limitations remain in culture in situ monitoring and tissue characterisation. In particular, biological improvements in the differentiation of the epidermis must be assessed to enhance the utility of the system.
The stability of the gel was significantly improved by the supplementation of the plasma-derived fibrin gels with commercial fibrinogen. A chapter of this thesis presents a detailed characterisation of these supplemented matrices. Results showed a marked reduction in gel contraction compared to only plasma gels, along with a slight decrease in the embedded fibroblast proliferation, while maintaining their viability. A final fibrinogen concentration of 3.5 mg/mL was finally selected as optimal for use in the chip.
Skin-on-a-chip systems are promising candidates for chemical testing. Adapting the OECD-standardised tests for skin irritation and corrosion, based on the MTT assay, to these systems would be of great usefulness. This thesis includes the validation of an alternative method to measure the results of the MTS assay (a similar functioning test) for cell viability assessment using electrochemistry instead of the traditional colourimetry readings. The results showed the feasibility of this approach showing a high correlation between the two measurement methods. Moreover, a chemical testing experiment demonstrated the capacity of the electrochemical approximation to calculate the viability of the cell cultures with accuracy. The future integration of an electrochemical sensor in the skin-on-a-chip platform would help to use this system for testing purposes.Programa de Doctorado en Ciencia y Tecnología Biomédica por la Universidad Carlos III de MadridPresidente: Pablo Acedo Gallardo.- Secretaria: María del Carmen de Arriba Pérez.- Vocal: Alberto Gallardo Rui
Inteligência Artificial Generativa na Produção de Média. O Papel Emergente do Artista de Inteligência Artificial em Espanha
Full text linkArtificial intelligence (AI) technologies have advanced exponentially in recent years, particularly in machine learning, including convolutional neural networks and generative adversarial networks. Their implementation in the creative industries has rapidly evolved from information analysis and data compression to the use of generative AI tools for media production. This exploratory study analyses the emerging role of the AI artist in applying generative AI techniques to the audio-visual post-production processes of the television series La Mesías (The Messiah; Movistar+, 2023) and the music video Pesadillas (Nightmares; Martina Hache, 2024), which were implemented by Alejandra G. López. The characteristics of the visual style resulting from their implementation will be studied. The methodological design combines approaches from media industry studies and organisational sociology, utilising a systematic hemerographic and bibliographic review, an in-depth interview and a technical analysis of the sequences involved. The workflow phases where AI was used are identified and classified according to the categories proposed by Anantrasirichai and Bull (2022): content creation, information analysis, content and workflow improvement, and information extraction. The results show that generative AI has a particularly significant impact on visual effects and 2D/3D compositing, creating a style that enhances realism with dreamlike atmospheres. The analysis also shows that these techniques, implemented with generative AI, require specialised profiles in the field and will be integrated into the audio-visual post-production workflow alongside other classic digital compositing and visual effects procedures.As tecnologias de inteligência artificial (IA) avançaram exponencialmente nos últimos anos, particularmente na aprendizagem automática, incluindo redes neurais convolucionais e redes adversárias generativas. A sua implementação nas indústrias criativas evoluiu rapidamente da análise de informações e compressão de dados para o uso de ferramentas de IA generativa para a produção de média. Este estudo exploratório analisa o papel emergente do artista de IA na aplicação de técnicas de IA generativa aos processos de pós-produção audiovisual da série de televisão La Mesías (A Messias; Movistar+, 2023) e do videoclipe Pesadillas (Pesadelos; Martina Hache, 2024), que foram implementados por Alejandra G. López. Serão estudadas as características do estilo visual resultante da sua implementação. O desenho metodológico combina abordagens dos estudos da indústria dos média e da sociologia organizacional, utilizando uma revisão hemerográfica e bibliográfica sistemática, uma entrevista aprofundada e uma análise técnica das sequências envolvidas. As fases do fluxo de trabalho onde foi utilizada a IA são identificadas e classificadas conforme as categorias propostas por Anantrasirichai e Bull (2022): criação de conteúdo, análise de informação, melhoria de conteúdo e fluxo de trabalho e extração de informação. Os resultados demonstram que a IA generativa tem um impacto particularmente significativo nos efeitos visuais e na composição 2D/3D, criando um estilo que aumenta o realismo com atmosferas oníricas. A análise também evidencia que estas técnicas, implementadas com IA generativa, requerem perfis especializados na área e serão integradas no fluxo de trabalho de pós-produção audiovisual, com outros procedimentos clássicos de composição digital e efeitos visuais
The technological revolution on financial markets: disintermediation or re-intermediation? Implications in financial regulation and supervision
No full textMención Internacional en el título de doctorPrograma de Doctorado en Derecho por la Universidad Carlos III de MadridPresidente: Matteo Gargantini.- Secretario: Manuel Alba Fernández.- Vocal: Jorge Armando Corredor Higuer
Optimal Distribution Planning of Solar Plants and Storage in a Power Grid with High Penetration of Renewables
No full textIntegrating variable renewable energy sources such as solar power into existing power grids presents major planning and reliability challenges. This study introduces an approach to optimize the placement of solar plants and allocation of storage in grids with high share of these variable energy sources by using a simulation framework that captures system-wide emergent behaviors. Unlike traditional engineering models focused on detailed component-level dynamics, a modified ORNL-PSERC-Alaska model based on self-organized criticality is used to reproduce the statistical features of blackouts, including cascading failures and long-range correlations. A distinctive feature of this approach is the explicit inclusion of key ingredients that shape these statistics, such as the transmission grid structure, generation and consumer buses, power flow balance, periodic dispatches, system failures, secular demand growth, demand fluctuations, and variability of renewable energy sources. When applied to the Balearic Islands grid, this method identifies generation and storage layouts that minimize storage requirements while maintaining reliability levels comparable to conventional power systems. The results offer a complementary systems-level perspective for planning resilient and efficient renewable energy integration.This research was funded by MICIU/AEI/10.13039/501100011033 and FEDER, EU grant number PID2021-122256NB-C22 (APASOS), and by MICIU/AEI/10.13039/501100011033 grant number CEX2021-001164-M under the Maria de Maeztu Program for units of Excellence in R&D. In addition, B.A.C. and J.M.R.-B. acknowledge access to Uranus, a supercomputer cluster located at Universidad Carlos III de Madrid (Spain) funded jointly by EU FEDER funds and by the Spanish Government via the National Research Project Nos. UNC313-4E-2361, ENE2009-12213-C03-03, ENE2012-33219, and ENE2012-31753
Determination of the Condition of Railway Rolling Stock Using Automatic Classifiers
No full textEfficient maintenance is paramount for rail transport systems to avoid catastrophic accidents. Therefore, a method that enables the early detection of defects in critical components is crucial for increasing the availability of rolling stock and reducing maintenance costs. This work's main contribution is the proposal of a methodology for analyzing vibration signals. The vibration signals, obtained from a bogie axle on a test bench, are decomposed into intrinsic functions, to which classical signal processing techniques are then applied. Finally, decision trees are employed to characterize the axle's state, yielding excellent results.The research work described in this paper was supported by the Spanish State Research Agency through the projects TED2021-131372A-I00 (AEI) and MCIN/AEI/10.13039/501100011033 (grant numbers MC4.0 PID2020-116984RB-C21-C22) and the project MEMRIAAP-CM-UC3M, supported by Comunidad de Madrid
Reassessing the great compression among top earners: The overlooked role of taxation and self-employment
Full text linkThis paper provides new estimates of wage inequality in the United States from 1918 to 1949, leveraging a novel top-income methodology that integrates both tax records and census data. Our analysis reveals no sustained decline in wage inequality before the Second World War but a marked decrease during the war years. This decline was driven primarily by stagnation among the top 1 % of earners and significant wage growth at the lower end of the income distribution. However, the relative underperformance of the top earners was largely influenced by a major compositional shift triggered by unprecedented increases in corporate and personal income tax rates. These tax changes led to a shift in business preferences toward partnerships, resulting in a substantial transition from salaried employment to self-employment. This shift, previously overlooked in inequality studies, resulted in a 30 % overestimation of wage compression, significantly altering the wage distribution dynamics of the 1940s.The authors acknowledge funding provided by Fundación BBVA [Beca Leonardo, Year 2020], Fundación Ramón Areces [Ayudas a la Investigación en Economía] and Funding for APC: Universidad Carlos III de Madrid [Agreement CRUE-Madroño 2024]
Privacy-Preserving Data Obfuscation for Credit Scoring
No full textProceedings of: SAC '25: 40th ACM/SIGAPP Symposium on Applied Computing Catania International Airport. Catania (Italy) 31 March 2025- 4 April 2025In this work, we present a privacy-preserving framework for credit scoring systems deployed on Machine Learning as a Service (MLaaS) platforms. Our approach integrates an obfuscator-classifier model that enhances privacy while maintaining high accuracy for loan default prediction tasks. The obfuscator transforms sensitive financial data into a privacy-protected representation, minimizing the risk of privacy leakage and input reconstruction during inference. By employing a combination of center loss and noise addition, our model ensures a robust balance between privacy and utility.Through extensive experiments, we demonstrate the effectiveness of our solution in reducing information leakage. For instance, our method achieves a 95.05% reduction in the average R2 score of reconstruction attacks, from 0.921 to 0.045. At the same time, we maintain high prediction accuracy, with only a negligible loss of 1.06% in public task accuracy, despite the added noise. These results highlight the scalability and adaptability of our framework for financial MLaaS applications, providing strong privacy protection without significantly compromising model performance.The research efforts of UC3M and NEC have received support from the Spanish Ministry of Economic Affairs and Digital Transformation as part of the UNICO 5G I+D initiative, specifically the 6G-RIEMANN project. Additionally, NEC’s work has been partially funded by the European Commission under the Horizon Europe program through the LICORICE project (Grant Agreement No. 101168311)
Environmental Effects on Child Development: Spatial Inequality and Socio-Economic Disparities in Early Life Outcomes
Full text linkMención Internacional en el título de doctorPrograma de Doctorado en Ciencias Sociales/ Social Sciences por la Universidad Carlos III de MadridPresidente: Jan Paul Heisig.- Secretaria: Leire Salazar.- Vocal: Tobias Rüttenaue