Search

The introduction of low-cost sensors (LCSs) in air quality Internet of Things (IoT) monitoring platforms presents the challenge of improving the quality of the data that these sensors provide. In this article, we propose two algorithms to perform denoising and calibration for LCSs used in IoT monitoring platforms. Sensors are first calibrated in situ using linear or nonlinear machine learning models that only take into account instantaneous measurements. The best calibration model is used to estimate the values measured by the sensor during the sensor deployment. To improve the values of the estimates produced by the in situ calibration model, we propose to take into account the temporal patterns present in signals, such as temperature or tropospheric ozone that have regular patterns, e.g., daily. The first method, which we call temporal pattern-based denoising (TPB-D), performs signal denoising by projecting the daily signals of the in situ calibrated LCS onto a subspace generated by the daily signals stored in a database taken by reference instruments. The second method, which we call temporal pattern-based calibration (TPB-C), considers that if we also have a reference instrument colocated to the LCSs over a period of time, we can correct with a linear mapping with regularization the daily LCS signals projected in the subspace produced by the reference database to be as similar as possible to the projected signals of the colocated reference instrument. The results show that the TPB-D improves the estimates made by in situ calibration by up to 10%–20%, while the TPB-C improves the estimates made by in situ calibration by up to 20%–40%. This work was supported in part by the National Spanish Funding under Grant PID2019-107910RB-I00 and in part by the Regional Project under Grant 2021 SGR 01059.

To effectively guide the exploration of the code transform space for automated code evolution techniques, we present in this paper the first approach for structurally predicting code transforms at the level of AST nodes using conditional random fields (CRFs). Our approach first learns offline a probabilistic model that captures how certain code transforms are applied to certain AST nodes, and then uses the learned model to predict transforms for arbitrary new, unseen code snippets. Our approach involves a novel representation of both programs and code transforms. Specifically, we introduce the formal framework for defining the so-called AST-level code transforms and we demonstrate how the CRF model can be accordingly designed, learned, and used for prediction. We instantiate our approach in the context of repair transform prediction for Java programs. Our instantiation contains a set of carefully designed code features, deals with the training data imbalance issue, and comprises transform constraints that are specific to code. We conduct a large-scale experimental evaluation based on a dataset of bug fixing commits from real-world Java projects. The results show that when the popular evaluation metric top-3 is used, our approach predicts the code transforms with an accuracy varying from 41% to 53% depending on the transforms. Our model outperforms two baselines based on history probability and neural machine translation (NMT), suggesting the importance of considering code structure in achieving good prediction accuracy. In addition, a proof-of-concept synthesizer is implemented to concretize some repair transforms to get the final patches. The evaluation of the synthesizer on the Defects4j benchmark confirms the usefulness of the predicted AST-level repair transforms in producing high-quality patches. This work was partially supported by National Natural Science Foundation of China (Grant No. 62102233), Shandong Province Overseas Outstanding Youth Fund (Grant No. 2022HWYQ-043), Qilu Young Scholar Program of Shandong University, and Wallenberg Artificial Intelligence, the Wallenberg Autonomous Systems and Software Program (WASP) funded by Knut and Alice Wallenberg Foundation. Some experiments were performed on resources provided by the Swedish National Infrastructure for Computing.

Automated metadata annotation is only as good as training dataset, or rules that are available for the domain. It's important to learn what type of data content a pre-trained machine learning algorithm has been trained on to understand its limitations and potential biases. Consider what type of content is readily available to train an algorithm—what's popular and what's available. However, scholarly and historical content is often not available in consumable, homogenized, and interoperable formats at the large volume that is required for machine learning. There are exceptions such as science and medicine, where large, well documented collections are available. This paper presents the current state of automated metadata annotation in cultural heritage and research data, discusses challenges identified from use cases, and proposes solutions.

Weakly supervised learning is aimed to learn predictive models from partially supervised data, an easy-to-collect alternative to the costly standard full supervision. During the last decade, the research community has striven to show that learning reliable models in specific weakly supervised problems is possible. We present an empirical study that analyzes the value of weak information of supervision throughout its entire spectrum, from none to full supervision. Its contribution is assessed under the realistic assumption that a small subset of fully supervised data is available. Particularized in the problem of learning with candidate sets, we adapt Cozman and Cohen [1] key study to learning from weakly supervised data. Standard learning techniques are used to infer generative models from this type of supervision with both synthetic and real data. Empirical results suggest that weakly labeled data is helpful in realistic scenarios, where fully labeled data is scarce, and its contribution is directly related to both the amount of information of supervision and how meaningful this information is.

The degradation of critical components inside large industrial assets, such as ball-bearings, has a negative impact on production facilities, reducing the availability of assets due to an unexpectedly high failure rate. Machine learning- based monitoring systems can estimate the remaining useful life (RUL) of ball-bearings, reducing the downtime by early failure detection. However, traditional approaches for predictive systems require run-to-failure (RTF) data as training data, which in real scenarios can be scarce and expensive to obtain as the expected useful life could be measured in years. Therefore, to overcome the need of RTF, we propose a new methodology based on online novelty detection and asymmetrical hidden Markov models (As-HMM) to work out the health assessment. This new methodology does not require previous RTF data and can adapt to natural degradation of mechanical components over time in data-stream and online environments. As the system is designed to work online within the electrical cabinet of machines it has to be deployed using embedded electronics. Therefore, a performance analysis of As-HMM is presented to detect the strengths and critical points of the algorithm. To validate our approach, we use real life ball-bearing data-sets and compare our methodology with other methodologies where no RTF data is needed and check the advantages in RUL prediction and health monitoring. As a result, we showcase a complete end-to-end solution from the sensor to actionable insights regarding RUL estimation towards maintenance application in real industrial environments. This study was supported partially by the Spanish Ministry of Economy and Competitiveness through the PID2019-109247GB-I00 project and by the Spanish Ministry of Science and Innovation through the RTC2019-006871-7 (DSTREAMS project). Also, by the H2020 IoTwins project (Distributed Digital Twins for industrial SMEs: a big-data platform) funded by the EU under the call ICT-11-2018- 2019, Grant Agreement No. 857191.

The use of crowdsourcing for annotating data has become a popular and cheap alternative to expert labelling. As a consequence, an aggregation task is required to combine the different labels provided and agree on a single one per example. Most aggregation techniques, including the simple and robust majority voting—to select the label with the largest number of votes—disregard the descriptive information provided by the explanatory variable. In this paper, we propose domain-aware voting, an extension of majority voting which incorporates the descriptive variable and the rest of the instances of the dataset for aggregating the label of every instance. The experimental results with simulated and real-world crowdsourced data suggest that domain-aware voting is a competitive alternative to majority voting, especially when a part of the dataset is unlabelled. We elaborate on practical criteria for the use of domain-aware voting.

Glioma is the most common type of tumor in humans originating in the brain. According to the World Health Organization, gliomas can be graded on a four-stage scale, ranging from the most benign to the most malignant. The grading of these tumors from image information is a far from trivial task for radiologists and one in which they could be assisted by machine-learning-based decision support. However, the machine learning analytical pipeline is also fraught with perils stemming from different sources, such as inadvertent data leakage, adequacy of 2D image sampling, or classifier assessment biases. In this paper, we analyze a glioma database sourced from multiple datasets using a simple classifier, aiming to obtain a reliable tumor grading and, on the way, we provide a few guidelines to ensure such reliability. Our results reveal that by focusing on the tumor region of interest and using data augmentation techniques we significantly enhanced the accuracy and confidence in tumor classifications. Evaluation on an independent test set resulted in an AUC-ROC of 0.932 in the discrimination of low-grade gliomas from high-grade gliomas, and an AUC-ROC of 0.893 in the classification of grades 2, 3, and 4. The study also highlights the importance of providing, beyond generic classification performance, measures of how reliable and trustworthy the model’s output is, thus assessing the model’s certainty and robustness. Carla Pitarch is a fellow of Eurecat’s “Vicente López” PhD grant program.

Treballs Finals de Grau d'Enginyeria Biomèdica. Facultat de Medicina i Ciències de la Salut. Universitat de Barcelona. Curs: 2022-2023. Tutor/Director: Gavara i Casas, Núria, Fibroblasts undergo significant morphological and functional changes in response to specific environmental cues and functional changes. In wound healing processes and cancerous environments, fibroblasts undergo transformative activation adopting novel phenotype. The aim of this project was to understand the changes in cell morphology and cytoskeletal reorganization that occur for both normal-associated fibroblasts (NAFs) and cancer-associated fibroblasts (CAFs) activation. To achieve this, we used an innovative approach using biophysical biomarkers derived from epifluorescence imaging of the cell's cytoskeleton, employing CSKmorphometrics. Clustering algorithms have been applied to do a preliminarily identification of CAFs subpopulations. Our findings confirm that cytoskeletal reorganization occurs during both physiological and pathological activation. Non-tumoral fibroblasts experience larger morphological changes characterized by an increase in area and cell convexity, as well as changes in the total fluorescence of F-actin fibers during the 24 hours posterior to the administration of TGF-β. On the contrary CAFs exhibited sustained larger areas throughout the process regardless of TGF-β administration. They underwent most drastic changes in fiber length and showed a significant increase in nuclear volume. The application of logistic regression algorithms has allowed for a classification with 81% accuracy to differentiate between CAFs and NAFs, highlighting the differences in the cytoskeleton of these cell types in both study contexts. On the other hand, the intragroup analysis provided by clustering has enabled the identification of 5 clusters for non-activated CAFs, which converge at 72 hours into two larger clusters with significant differences. This study enhances understanding of the changes occurring in CAFs and NAFs during activation from a cytoskeletal point of view and remarks the need for a study of fibroblasts subpopulations as well as the need for novel biomarkers.

Treballs Finals de Grau d'Enginyeria Biomèdica. Facultat de Medicina i Ciències de la Salut. Universitat de Barcelona. Curs: 2022-2023. Tutor/Director: Gambús Cerrillo, Pedro Luis, Several studies address the process of loss of consciousness during the induction of general anaesthesia, but few of them discuss or study the process of recovery of consciousness once the of general anaesthesia has been administered successfully. The main objective of this project is to study and develop a predictive model of the duration of this process of consciousness recovery based on Machine Learning (ML) and the analysis of electroencephalographic (EEG) signals. A dataset comprising 143 patients from the 4th operating room of the Hospital Clínic of Barcelona was analysed. The project involved data pre-processing, including the segmentation of EEG signals during the recovery process, feature extraction, and correlation analysis. Five ML regression algorithms, namely Linear, Lasso, and Ridge Regression, Support Vector Regression (SVR), and Random Forest (RF), were evaluated using a Cross-Validation pipeline. Model performance, feature selection, and hyperparameter optimization were assessed using the R-squared score criterion. The best performing algorithm was the regularized linear regression model, Lasso, achieving an R-squared score of 0.74 ± 0.032 (mean and standard deviation). Through the correlation analysis and the feature selection performed by the algorithm, high predictive capabilities of consciousness recovery time were obtained for alpha and beta relative averaged band power in the first minute before stopping general anaesthesia administration. The findings demonstrate that EEG signals contain valuable information regarding the process of consciousness recovery, enabling the construction of ML predictive models. However, further studies are required to enhance our understanding of the consciousness recovery process and to validate the predictive model in a clinical setting. Future investigations should focus on increasing data variability, addressing biases in validation techniques, exploring additional EEG channels to capture global brain activity, and considering regulatory considerations for Artificial Intelligence algorithms.

Cada cop més, en la indústria del futbol, tant entrenadors, com clubs, com productores, etc. busquen l’automatització de diferents processos per tal de no malgastar recursos humans en aquestes tasques. Una tasca és la detecció d’accions en un partit, la qual busca evitar que un humà supervisi tot el partit complet i hagi d’anotar els diferents temps en què tenen lloc determinades accions. En aquest treball es presenta un mètode innovador per detectar diferents accions que tenen lloc en un partit de futbol a partir de la gravació del partit complet. El mètode que es presenta utilitza el vídeo per treure metadades, concretament la posició i l’equip dels jugadors, que posteriorment s’utilitzen de vector d’entrada a una xarxa neuronal que retorna la probabilitat associada a cada acció. Per tal de recollir aquestes metadades s’utilitza la tècnica de stitching per tal d’obtenir la panoràmica del camp i tot seguit amb el detector YOLO es detecta la posició que ocupa cada jugador dins del terreny de joc i es classifica en un dels dos equips. En aquest treball es presenta diferents tècniques per processar aquestes dades, algunes més tradicionals i altres innovadores, elaborades pròpiament en el desenvolupament del treball. També es presenten i es comparen diferents arquitectures per tal de classificar d’identificar correctament si les accions en qüestió succeeixen o no. L’arquitectura final que ofereix millors resultats està composta per una xarxa LSTM que s’alimenta de dades processades en forma de comptatge dividint el camp de futbol en 7x5 (35 quadrants). L’algorisme que es descriu en aquest treball presenta bons resultats a l’hora de classificar accions i a més computacionalment és molt eficient i la seva implementació/utilització no requereix de màquines sofisticades. A més a més, tal com es descriu en el treball la seva aplicació potser interessant en diferents camps de la indústria esportiva.

The mental health landscape worldwide has been profoundly affected by the COVID-19 pandemic, with an escalating rise in stress, anxiety, and depression. As a result, the urgency to create and apply efficacious methods for averting and addressing these mental health issues has never been greater. In this context COVICAT is a research project that is being carried out in Catalonia, Spain. The project is led by the Germans Trias i Pujol Research Institute (IGTP) and the Barcelona Institute for Global Health (ISGlobal). The goal of COVICAT is to evaluate the impact of the COVID-19 pandemic on the Catalan population. This master's thesis aims to leverage machine learning methodologies for psychological risk assessment using the COVICAT dataset (N = 9,398), contributing to healthcare practices by facilitating the initial classification of patients requiring psychological intervention. The study focuses on predicting four outcome variables reflecting patients' psychological wellbeing, one of which encapsulates a patient's self-perceived mental state. Initial exploratory data analysis and data engineering were performed to prepare the dataset for machine learning applications, including bagging, boosting, and stacking models. Each model underwent an analogous pipeline with minor variations, dictated by the distinct nature of the respective model. Post-training, these models were evaluated on unseen data, generating several performance metrics. The F1-score, designated as the critical criterion by the subject matter experts, was employed to gauge the models' efficacy. Based on these evaluations, the optimal predictive model was selected for each target. The resulting models delivered performance surpassing the adequate standard established initially by the experts (F1 = 0.6). These models aim to augment, not supplant, professional healthcare diagnoses by enabling efficient patient triage for psychological assistance.

Aquest projecte té la intenció d’avançar en àmbit de la rehabilitació de pacients els quals, d’alguna forma o altre, necessitin assistència per fer exercicis. Principalment, es tractarà d’entendre el moviment del braç des d’un punt de vista electrònic, on s’han d’extreure uns senyals amb un sensor especialitzat per la tasca, tenir un exoesquelet per donar l’assistència al moviment necessària i amb ajuda d’una intel·ligència artificial crear models personalitzats per poder proporcionar aquesta ajuda. També tracta d’aprendre sobre aquest nou fenomen que són les intel·ligències artificials i treure profit, per dur a terme tasques complicades, com seria el fet de crear un model molt complex. Finalment, per acabar si dona temps, provar el model amb pacients reals i veure si a part de l’assistència de l’exoesquelet, generar una electroestimulació per ajudar en la rehabilitació

Este trabajo se centra en la implementación de una propuesta de software y hardware para su uso en diferentes aplicaciones industriales en los que se requiere de un sistema capaz de identificar y clasificar ciertos elementos como por ejemplo en un proceso de ensamblado de piezas o en etapas de empaquetado. Para ello, se desarrollará un sistema de visión por computador para su uso junto a brazos robot. El sistema estará compuesto por diferentes elementos que interactúan entre sí para aportar una solución autónoma. Durante este proceso, el sistema será capaz de tomar una imagen del espacio de trabajo en el que se localizan las piezas a manipular a través de un sensor digital y realizará un proceso de extracción de varias propiedades para cada uno de los elementos analizados. En concreto, las propiedades a extraer serán la posición y la orientación de cada elemento en el espacio de trabajo, así como la región que ocupan en dicho espacio. Además se dotará al sistema de la capacidad de identificar la clase de cada pieza para poder realizar su clasificación de forma correcta. Esto se realizará a través del entrenamiento de un modelo predictivo. Toda esta información está formateada para facilitar su transmisión al brazo robot. Esta memoria detalla el proceso de desarrollo de la solución a través de la configuración y puesta en marcha del hardware de adquisición de imagen, así como la programación de la algoritmia necesaria para que los elementos de software interactúen entre sí y realizen el procesado digital necesario

In the last decades, we have witnessed many changes regarding the improvement of physical models and processes with positive impact on humanity, which we call Industrial Revolutions. In this paper, a description of the fundamental concepts of the last of these revolutions present until today is made, referring to the Fourth Industrial Revolution, called "Industry 4.0". In this work, a description of the State of the Art of the enabling technologies of this Industry 4.0 is presented, but because this topic is very broad, the description is limited to Artificial Intelligence and Cyber-Physical Systems applied to the industrial sector. Likewise, a case study is presented on how to carry out the design of an application and the analytics associated with a typical service in Industry 4.0, based on a cyber-physical system structure and supported by machine learning procedures. The methodology carried out to address this work consists of two stages. Firstly, an exploratory stage based on document search and management, mainly through academic research articles to identify, organize and describe the concepts and components involved in the application of artificial intelligence and cyber-physical systems in the industrial context. Secondly, a design and implementation stage of the logical part of an industrial maintenance application based on data processing in the Matlab platform. Finally, from the case study developed in the second stage, the results are presented using graphs and statistics to summarize the data obtained. And a conclussion section where the interpretation of the results and recomendations for future directions of the case of study are presented. En las últimas décadas, hemos presenciado muchos cambios respecto a la mejora de modelos y procesos físicos con repercusión positiva en la humanidad, a los que llamamos Revoluciones Industriales. En este trabajo, se realiza una descripción sobre conceptos fundamentales de la última de estas revoluciones presente hasta el día de hoy, refiriéndonos a la Cuarta Revolución Industrial, denominada “Industria 4.0”. Se presenta, en este trabajo, una descripción del Estado de Arte de las tecnologías habilitadoras de dicha Industria 4.0, pero debido a que este tema es muy amplio, se limita la descripción, a la Inteligencia Artificial y a los Sistemas Ciber-físicos aplicados al sector industrial. Asimismo, se presenta un caso de estudio sobre llevar a cabo el diseño de una aplicación y la analítica asociada a un servicio tipo en industria 4.0, basado en una estructura de sistema ciber-físico y soportado por procedimientos de machine learning. La metodología llevada a cabo para afrontar este trabajo, consta de dos etapas. En primer lugar, una etapa exploratoria a base de búsqueda y gestión documental, principalmente a través de artículos de investigación académica para identificar, organizar y describir los conceptos y componentes que intervienen en la aplicación de inteligencia artificial y sistemas ciber-físicos en el contexto industrial. En segundo lugar, una etapa de diseño e implementación de la parte lógica de una aplicación de mantenimiento industrial basada en el tratamiento de datos bajo plataforma Matlab. Finalmente, a partir del caso práctico desarrollado en la segunda etapa, se presentan los resultados mediante gráficos y estadísticas para resumir los datos obtenidos. Y una sección de conclusiones donde se presenta la interpretación de los resultados y recomendaciones para las direcciones futuras del caso de estudio. Incoming

Until today, a recurring person was in charge of manually reviewing the Excel files of the entity incidents. Then a software was proposed to analyze and identify incidental patterns and classify them. However, the creation of this software requires constant maintenance, since there are different processes that need to be automated. In addition, we must migrate the software to the cloud and also incorporate artificial intelligence modules to improve its performance. Fins a dia d'avui un recurrent s'encarregava de revisar manualment els arxius Excel de les incidències de l'entitat. Aleshores es va proposar un programari per analitzar i identificar els patrons d'incidental i classificar-los. No obstant això, la creació daquest programari requereix un manteniment constant, ja que hi ha diferents processos que han de ser automatitzats. A més, hem de migrar el programari al núvol i també incorporar mòduls d'intel·ligència artificial per millorar-ne el rendiment.

Brain-age can be inferred from structural neuroimaging and compared to chronological age (brain-age delta) as a marker of biological brain aging. Accelerated aging has been found in neurodegenerative disorders like Alzheimer's disease (AD), but its validation against markers of neurodegeneration and AD is lacking. Here, imaging-derived measures from the UK Biobank dataset (N=22,661) were used to predict brain-age in 2,314 cognitively unimpaired (CU) individuals at higher risk of AD and mild cognitive impaired (MCI) patients from four independent cohorts with available biomarker data: ALFA+, ADNI, EPAD and OASIS. Brain-age delta was associated with abnormal amyloid-b, more advanced stages (AT) of AD pathology and APOE-e4 status. Brain-age delta was positively associated with plasma neurofilament light, a marker of neurodegeneration, and sex differences in the brain effects of this marker were found. These results validate brain-age delta as a non-invasive marker of biological brain aging in non-demented individuals with abnormal levels of biomarkers of AD and axonal injury. Marc Suárez-Calvet: Horizon 2020 - Research and Innovation Framework Programme 948677, Instituto de Salud Carlos III PI19/00155, La Caixa Foundation 100010434, Horizon 2020 - Research and Innovation Framework Programme 847648 / Juan Domingo Gispert: EU/EFPIA Innovative Medicines Initiative Joint Undertaking AMYPAD 115952 / Marc Suárez-Calvet: La Caixa Foundation 100010434; LCF/PR/ GN17/50300004 / Irene Cumplido-Mayoral: TriBEKa Imaging Platform project TriBEKa-17-519007, Universities and Research Secretariat, Ministry of Business and Knowledge of the Catalan Government 2017-SGR-892 / Juan Domingo Gispert: EIT Digital Grant 2021, Spanish Ministry of Science and Innovation (MCIN)/Spanish Research Agency (AEI) MCIN/AEI /10.13039/501100011033 RTI2018-102261, cofunded by the European Regional Development Fund (FEDER) / Marc Suárez-Calvet: NIHR biomedical research center at UCLH Frederik Barkhof Instituto de Salud Carlos III PI22/00456 / Raffaele Cacciaglia: Spanish Ministry of Science and Innovation (MCIN)/Spanish Research Agency (AEI) MCIN/ AEI/10.13039/501100011033 PID2021-125433OA-100 / Verónica Vilaplana: Spanish Research Agency (AEI) PID2020-116907RB-I00 of the call MCIN/ AEI /10.13039/501100011033 Peer Reviewed Article signat per 27 autors/es: Irene Cumplido-Mayoral, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Marina García-Prat, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Grégory Operto, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Carles Falcon, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Mahnaz Shekari, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Raffaele Cacciaglia, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Marta Milà-Alomà, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Luigi Lorenzini, Department of Radiology and Nuclear Medicine, Amsterdam University Medical Centers, Amsterdam, Netherlands; Silvia Ingala, Department of Radiology and Nuclear Medicine, Amsterdam University Medical Centers, Amsterdam, Netherlands; Alle Meije Wink, Department of Radiology and Nuclear Medicine, Amsterdam University Medical Centers, Amsterdam, Netherlands; Henk JMM Mutsaerts, Department of Radiology and Nuclear Medicine, Amsterdam University Medical Centers, Amsterdam, Netherlands; Carolina Minguillón, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Karine Fauria, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; José Luis Molinuevo, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Sven Haller, CIRD Centre d'Imagerie Rive Droite, Geneva, Switzerland; Gael Chetelat, Normandie Univ, UNICAEN, INSERM, U1237, Caen, France; Adam Waldman, Centre for Dementia Prevention, University of Edinburgh, Edinburgh, United Kingdom; Adam J Schwarz, Takeda Pharmaceutical Company Ltd, Cambridge, United States; Frederik Barkhof, Department of Radiology and Nuclear Medicine, Amsterdam University Medical Centers, Amsterdam, Netherlands; Ivonne Suridjan, Roche Diagnostics International Ltd, Rotkreuz, Switzerland; Gwendlyn Kollmorgen, Roche Diagnostics GmbH, Penzberg, Germany; Anna Bayfield, Roche Diagnostics GmbH, Penzberg, Germany; Henrik Zetterberg, Institute of Neuroscience and Physiology, University of Gothenburg, Gothenburg, Sweden; Kaj Blennow, Institute of Neuroscience and Physiology, University of Gothenburg, Mölndal, Sweden; Marc Suárez-Calvet, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain; Verónica Vilaplana, Department of Signal Theory and Communications, Universitat Politècnica de Catalunya, Barcelona, Spain; Juan Domingo Gispert López, Barcelonaβeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain

The version of record is available online at: http://dx.doi.org/10.1007/s00607-023-01179-5 In the high-performance computing (HPC) domain, federated learning has gained immense popularity. Especially in emotional and physical health analytics and experimental facilities. Federated learning is one of the most promising distributed machine learning frameworks because it supports data privacy and security by not sharing the clients' data but instead sharing their local models. In federated learning, many clients explicitly train their Machine Learning/Deep Learning models (local training) before aggregating them as a global model at the global server. However, the FL framework is difficult to build and deploy across multiple distributed clients due to its heterogeneous nature. We developed Docker-enabled Federated Learning (DFL) by utilizing client-agnostic technologies like Docker containers to simplify the deployment of FL frameworks for data stream processing on the heterogeneous client. In the DFL, the clients and global servers are written using TensorFlow and lightweight Message Queuing Telemetry Transport (MQTT) protocol to communicate between clients and global servers in the IoT environment. Furthermore, the DFL's effectiveness, efficiency, and scalability are evaluated in the test case scenario where real-time emotion state classification is done from distributed multi-modal physiological data streams under various practical configurations. Arijit Nandi is a fellow of Eurecat’s "Vicente López" PhD grant program. This study has been partially funded by ACCIÓ, Spain (Pla d’Actuació de Centres Tecnológics 2021) under the project TutorIA.

En un context de popularització de les tècniques d’Intel·ligència Artificial, aquest treball es proposa aplicar la versió més coneguda d’aquestes, les xarxes neuronals, per tractar comprendre la dinàmica del tipus de cuc C.elegans. Xarxes denses, convolucionals i recurrents van ser provades per tractar de predir moments de flexió a través de posicions i validar l’actual teoria sobre el moviment d’aquests cucs. La capacitat de les Xarxes Neuronals de predir moments a través de posicions sintètiques i experimentals demostrarà ser molt bona, tot i que caldria més investigació per aconseguir una resposta definitiva. En un contexto de popularización de las técnicas de Inteligencia Artificial, este trabajo se propone aplicar la versión más conocida de estas, las redes neuronales, para tratar de comprender la dinámica del tipo de gusano C.elegans. Redes densas, convolucionales y recurrentes serán probadas para tratar de predecir momentos de flexión a través de posiciones y validar la actual teoría sobre el movimiento de estos gusanos. La capacidad de las Redes Neuronales de predecir momentos a través de posiciones sintéticas demostrará ser muy buena, aunque haría falta más investigación para llegar a una respuesta definitiva. In a context of popularization of Artificial Intelligence techniques, this project apllies its most known implementation, neural networks, to try to understand the dynamics underlying the motion of the nematode C.elegans. Dense, convolutional and recurrent networks will be tried to predict moments in base of positions and validate the current theory of these worm’s motion. The capacity of Neural Networks to predict muscle activity, i.e. bending moments, from both synthetic and experimental positions will prove to be very good. However, more investigation would be needed in order to reach a definitive answer.

Atesa la naturalesa difícil dels problemes d'optimització combinatoris, molts algoritmes es basen en heurístiques dissenyades manualment per prendre decisions que poden ser massa costoses de calcular o matemàticament no ben definides. Així, l'aprenentatge automàtic (ML) sembla un candidat factible per prendre aquestes decisions d'una manera més òptima. En aquest projecte, explorarem la relació entre el ML i els algoritmes d'optimització per alguns casos d'ús de xarxes de telecomunicacions. Dada la naturaleza difícil de los problemas de optimización combinatoria, muchos algoritmos se basan en heurísticas diseñadas a mano para tomar decisiones que de otro modo serían demasiado costosas de calcular o matemáticamente no están bien definidas. Así, el aprendizaje automático (ML) parece un candidato natural para tomar esas decisiones de una manera más óptima. En este proyecto, exploraremos la relación entre el ML y los algoritmos de optimización para algunos casos de uso de redes de telecomunicaciones. Given the hard nature of combinatorial optimization problems, many state-of-the-art algorithms rely on handcrafted heuristics for making decisions that are otherwise too expensive to compute or mathematically not well defined. Thus, machine learning (ML) looks like a natural candidate to make such decisions in a more principled and optimized way. In this project, we will explore the relationship between ML and optimization algorithms for telecommunication network use cases.

The last few years have seen an increasing wave of attacks with serious economic and privacy damages, which evinces the need for accurate Network Intrusion Detection Systems (NIDS). Recent works propose the use of Machine Learning (ML) techniques for building such systems (e.g., decision trees, neural networks). However, existing ML-based NIDS are barely robust to common adversarial attacks, which limits their applicability to real networks. A fundamental problem of these solutions is that they treat and classify flows independently. In contrast, in this paper we argue the importance of focusing on the structural patterns of attacks, by capturing not only the individual flow features, but also the relations between different flows (e.g., the source/destination hosts they share). To this end, we use a graph representation that keeps flow records and their relationships, and propose a novel Graph Neural Network (GNN) model tailored to process and learn from such graph-structured information. In our evaluation, we first show that the proposed GNN model achieves state-of-the-art results in the well-known CIC-IDS2017 dataset. Moreover, we assess the robustness of our solution under two common adversarial attacks, that intentionally modify the packet size and inter-arrival times to avoid detection. The results show that our model is able to maintain the same level of accuracy as in previous experiments, while state-of-the-art ML techniques degrade up to 50% their accuracy (F1-score) under these attacks. This unprecedented level of robustness is mainly induced by the capability of our GNN model to learn flow patterns of attacks structured as graphs., Comment: 7 pages, 4 figures, 1 table

The goal of the presented thesis is to find a classifier able to recognize cycles in different vehicle models from a company. The problem faced is an unsupervised type, so there is no prior information on how many cycles exist. To deal with it, two contrastive methodologies were used. Once the model is trained and tuned, the generated encoder is used for downstream tasks, in this case, the space generated is used to apply a KMeans model to find clusters and therefore, the cycles within. With this cluster, a classifier is trained to have a final model able to be used in real-time. Previous work has been done by the company without applying any contrastive methodology and without any automation. Therefore, automation of the processes is required by the company. Silhouette score is used to check the performance and compare the results. Contrastive learning methodologies outperform the current solution of the company when it is applied to a specific vehicle model

Anàlisis de textos (posts i comentaris) per la predicció de migracions d'usuaris entre xarxes socials basades en la blockchain després d'un hard fork. Cas d'estudi sobre el fork Steem-Hive de Març del 2020.

Many sweat-based wearable monitoring systems have been recently proposed, but the data provided by those systems often lack a reliable and meaningful relation to standardized blood values. One clear example is lactate, a relevant biomarker for both sports and health sectors, with a complex sweat–blood bioequivalence. This limitation decreases its individual significance as a sweat-based biomarker. Taking into account the insights of previous studies, a multiparametric methodology has been proposed to predict blood lactate from non-invasive independent sensors: sweat lactate, sweat rate, and heart rate. The bioequivalence study was performed with a large set of volunteers (>30 subjects) in collaboration with sports institutions (Institut Nacional d’Educació Física de Catalunya, INEFC, and Centre d’Alt Rendiment, CAR, located in Spain). A neural network algorithm was used to predict blood lactate values from the sensor data and subject metadata. The developed methodology reliably and accurately predicted blood lactate absolute values, only adding 0.3 mM of accumulated error when compared to portable blood lactate meters, the current gold standard for sports clinicians. The approach proposed in this work, along with an integrated platform for sweat monitoring, will have a strong impact on the sports and health fields as an autonomous, real-time, and continuous monitoring tool The authors kindly acknowledge the support from the Spanish Ministerio de Indústria, Energía y Turismo (AEI Clusters Program, AEI2009L1CA011), Ministerio de Ciencia y Innovació n (PID2020-114070RB-I00), Agencia Estatal de Investigación (RED2018-102829-T and CPP2021-009021), and AGAUR (2019 DI 18 and 2021PROD00064)

The manufacturing domain is regarded as one of the most important engineering areas. Recently, smart manufacturing merges the use of sensors, intelligent controls, and software to manage each stage in the manufacturing lifecycle. Additionally, the increasing use of point clouds to model real products and machining tools in a virtual space facilitates the more accurate monitoring of the end-to-end production lifecycle. Thus, the conjunction of both, intelligent methods and more accurate 3D models allows the prediction of uncertainties and anomalies in the manufacturing process as well as reduces the final production costs. However, the high complexity of the geometrical structures defined by point clouds and the high accuracy required by the Quality Assurance/Quality control parameters during the process, pave the way for continuous improvements in smart manufacturing methods. This paper addresses a comprehensive analysis of machining tool identification utilizing temporal point cloud data. Specifically, we deal with the identification of machining tools from temporal 3D point clouds. To do that, we propose a process to construct and train intelligent models utilizing such data. Moreover, in our case study, we provide the research community with two labeled temporal 3D point cloud datasets, and we experiment with the pioneering PointNet neural network and three of its variants demonstrating an accuracy of 95% in the identification of the utilized machining tools in a machining process. Finally, we provide a prototype end-to-end intelligent system of machining tool identification.

MRI; Machine learning; Multiple sclerosis Ressonància magnètica; Aprenentatge automàtic; Esclerosi múltiple Resonancia magnética; Aprendizaje automático; Esclerosis múltiple Introduction: Conventional MRI is routinely used for the characterization of pathological changes in multiple sclerosis (MS), but due to its lack of specificity is unable to provide accurate prognoses, explain disease heterogeneity and reconcile the gap between observed clinical symptoms and radiological evidence. Quantitative MRI provides measures of physiological abnormalities, otherwise invisible to conventional MRI, that correlate with MS severity. Analyzing quantitative MRI measures through machine learning techniques has been shown to improve the understanding of the underlying disease by better delineating its alteration patterns. Methods: In this retrospective study, a cohort of healthy controls (HC) and MS patients with different subtypes, followed up 15 years from clinically isolated syndrome (CIS), was analyzed to produce a multi-modal set of quantitative MRI features encompassing relaxometry, microstructure, sodium ion concentration, and tissue volumetry. Random forest classifiers were used to train a model able to discriminate between HC, CIS, relapsing remitting (RR) and secondary progressive (SP) MS patients based on these features and, for each classification task, to identify the relative contribution of each MRI-derived tissue property to the classification task itself. Results and discussion: Average classification accuracy scores of 99 and 95% were obtained when discriminating HC and CIS vs. SP, respectively; 82 and 83% for HC and CIS vs. RR; 76% for RR vs. SP, and 79% for HC vs. CIS. Different patterns of alterations were observed for each classification task, offering key insights in the understanding of MS phenotypes pathophysiology: atrophy and relaxometry emerged particularly in the classification of HC and CIS vs. MS, relaxometry within lesions in RR vs. SP, sodium ion concentration in HC vs. CIS, and microstructural alterations were involved across all tasks. This project has received funding under the European Union's Horizon 2020 Research and Innovation Programme under grant agreement No. 634541. FG received the support of a fellowship from “la Caixa” Foundation (ID 100010434). The fellowship code is “LCF/BQ/PR22/11920010”. FG has also received support from the Beatriu de Pinós (2020 BP 00117) programme, funded by the Secretary of Universities and Research (Government of Catalonia). BK, FP, and OC are supported by the National Institute of Health Research Biomedical Research Centre at UCL and UCLH. EPSRC grants EP/M020533/1 and EP/J020990/01, MRC MR/T046422/1 and MR/T046473/1, Wellcome Trust award 221915/Z/20/Z, and the NIHR UCLH BRC support DCA's work in this area. CGWK also receives funding from Horizon 2020 [Research and Innovation Action Grants Human Brain Project 945539 (SGA3)], BRC (#BRC704/CAP/CGW), MRC (#MR/S026088/1), and Ataxia UK.

Machine learning has received increasing attention in computer science in recent years and many types of methods have been proposed. In computer networks, little attention has been paid to the use of ML for fault detection, the main reason being the lack of datasets. This is motivated by the reluctance of network operators to share data about their infrastructure and network failures. In this paper, we attempt to fill this gap using anomaly detection techniques to discern hardware failure events in wireless community networks. For this purpose we use 4 unsupervised machine learning, ML, approaches based on different principles. We have built a dataset from a production wireless community network, gathering traffic and non-traffic features, e.g. CPU and memory. For the numerical analysis we investigated the ability of the different ML approaches to detect an unprovoked gateway failure that occurred during data collection. Our numerical results show that all the tested approaches improve to detect the gateway failure when non-traffic features are also considered. We see that, when properly tuned, all ML methods are effective to detect the failure. Nonetheless, using decision boundaries and other analysis techniques we observe significant different behavior among the ML methods. This work has received funding through the DiPET CHIST-ERA under grant agreement PCI2019-111850-2; Spanish grant PID2019-106774RB-C21; Romanian DIPET (62652/15.11.2019) project funded via PN 124/2020; and has been partially supported by the EU research project SERRANO (101017168) and hardware resources courtesy of the Romanian Ministry of Research and Innovation UEFISCDI COCO research project PN III-P4-ID-PCE-2020-0407.

Novelty detection is a statistical method that verifies new or unknown data, determines whether these data are inliers (within the norm) or outliers (outside the norm), and can be used, for example, in developing classification strategies in machine learning systems for industrial applications. To this end, two types of energy that have evolved over time are solar photovoltaic and wind power generation. Some organizations around the world have developed energy quality standards to avoid known electric disturbances; however, their detection is still a challenge. In this work, several techniques for novelty detection are implemented to detect different electric anomalies (disturbances), which are k-nearest neighbors, Gaussian mixture models, one-class support vector machines, self-organizing maps, stacked autoencoders, and isolation forests. These techniques are applied to signals from real power quality environments of renewable energy systems such as solar photovoltaic and wind power generation. The power disturbances that will be analyzed are considered in the standard IEEE-1159, such as sag, oscillatory transient, flicker, and a condition outside the standard attributed to meteorological conditions. The contribution of the work consists of the development of a methodology based on six techniques for novelty detection of power disturbances, under known and unknown conditions, over real signals in the power quality assessment. The merit of the methodology is a set of techniques that allow to obtain the best performance of each one under different conditions, which constitutes an important contribution to the renewable energy systems.

The project presented in this paper aims to formulate a recommendation framework that consolidates the higher education students’ particulars such as their academic background, current study and student activity records, their attended higher education institution’s expectations of graduate attributes and self-assessment of their own generic competencies. The gap between the higher education students’ generic competence development and their current statuses such as their academic performance and their student activity involvement was incorporated into the framework to come up with a recommendation for the student activities that lead to their generic competence development. For the formulation of the recommendation framework, the data mining tool Orange with some programming in Python and machine learning models was applied on 14,556 students’ activity and academic records in the case higher education institution to find out three major types of patterns between the students’ participation of the student activities and (1) their academic performance change, (2) their programmes of studies, and (3) their English results in the public examination. These findings are also discussed in this paper.

[cat] En l’actual transició digital, on es generen milions de dades cada dia, l’ús de tècniques remotes digitals i els processats amb tècniques de machine learning és cada cop més present i necessari, sense que l’àmbit de les geociències en sigui una excepció. En aquest context, el present doctorat industrial, se centra en l’anàlisi d’afloraments fracturats utilitzant les tècniques remotes i machine learning, amb dos objectius específics; (1) l’avaluació de la representativitat de tècniques remotes per a identificar i caracteritzar las fractures i posteriorment representar-les en models discrets de xarxes de fractures (DFN), i (2) l’ús de models classificadors amb machine learning en processos associats als medis fracturats, com són les caigudes de blocs. METODOLOGIA DE TREBALL: En el primer cas d’estudi, s’avaluen les tècniques remotes; Terrestrial Laser Scanner (TLS) i Fotogrametria (SfM). L’experiment es desenvolupa en la pedrera activa de La Fou a Barcelona. Els afloraments fracturats s’han reconstruït a partir d’un alt nombre de dades capturades remotament (Santana et al., 2012; Geyer et al., 2015) i s’han agrupat en famílies per caracteritzar geomètricament la xarxa de fractures (García-Sellés et al., 2018). Amb aquests resultats, i els recollits amb tècniques directes amb els scan-lines, es construeixen models tridimensionals (DFN). Es construeix un model amb cada tècnica de captura (DFNTLS, DFNSfM i DFNscan-line), on quedaran representades les principals propietats geomètriques de les fractures. Finalment, se generen els mapes de traces de fractures per calcular la porositat i permeabilitat de la xarxa que hi representen. El segon cas té com objectiu automatitzar la classificació de caigudes de blocs. L’àrea d’estudi se situa en els escarpaments històricament inestables de Degotalls, prop de l’accés al Monestir de Montserrat. La monitorització d’aquests escarpaments mitjançant els escanejos periòdics amb TLS, es va iniciar l’any 2007, com a conseqüència dels episodis de caigudes de blocs iniciats l'any 2001, que van afectar a l’aparcament del monestir (limitant l’accés al Santuari en diverses ocasions). La metodologia consisteix en alinear dos escanejos o núvols de punts del mateix aflorament capturats en dos moments diferents, amb la intenció de detectar les agrupacions de diferències existents entre ambdós escanejos. Posteriorment, es revisen visualment les agrupacions de punts per identificar i classificar les caigudes de blocs. La metodologia desenvolupada en aquesta tesi, incorporant algorismes de machine learning proposa automatitzar aquesta última etapa d’identificació i classificació de blocs caiguts. RESULTATS: En el primer cas d’estudi a La Fou les principals propietats geomètriques de cada fractura capturada i processada són la posició, orientació, alçada i longitud. Aquestes s’utilitzen per caracteritzar les estadístiques de les fractures, com l’espaiat, intensitat, P10, P20, P21 i P32, entre d’altres. Del conjunt de fractures s’identifica la Família I WNW- ESE, Família II NNW/SSE, Família III, N-S i Família IV NE-SW. La tècnica amb major quantitat de fractures caracteritzades ha estat el TLS amb més de 52.000, després la tècnica de fotogrametria amb 1.094 i finalment els scan- lines amb 398 fractures mesurades. L’alçada i longitud de les fractures són molt variables, amb valors mitjos que oscil·len entre els pocs centímetres als 18m. El valors obtinguts amb TLS no arriben al metre d’alçada i longitud. Els espaiats entre les fractures de la Família I presenten mides centimètriques i la resta de famílies valors mitjos més variables des de centimètrica fins a 6m. La fotogrametria obté valors que en ocasions dupliquen o tripliquen la resta de tècniques. Amb la introducció de les obertures de les fractures en el model s’han generat els valors de porositat i permeabilitat de la xarxa de fractures. En els tres models DFN, el valor mig de porositat obtingut està entre 5,4 i 5,7% i els valors de permeabilitat mitja entre 4x105 mD i 6,2x105mD. En el segon cas d’estudi, els resultats generats a partir de la combinació de 11 algorismes classificadors amb 15 mètodes de remostreig, permeten identificar la millor combinació predictiva para la classificació de blocs caiguts, segons el període i l’escarpament estudiat. A Degotalls-E, a partir d’aproximadament 5.800 agrupaments de diferències identificats per cada monitoratge, s’assoleix una reducció del 98% del nombre d’agrupaments a validar, entre Falsos positius (FP) i Veritables Positius (VP), entenent (VP) com les caigudes reals de blocs. A Degotalls-N la reducció ha estat del 80,16% per a una identificació completa, mentre que per identificar el 96% de les caigudes de blocs reals, la reducció ha estat del 90%, a partir d'una població d’uns 3.700 agrupaments inicials. És important també assenyalar, que la validació de la tasca mitjançant la comparació d'imatges d'alta resolució, ha permès augmentar notablement la fiabilitat dels resultats obtinguts. CONCLUSIONS I PERSPECTIVA D’AVANÇ: En l’avaluació de la representativitat de l’ús de tècniques remotes per a la construcció de models DFN, els resultats suggereixen que les dades capturades i tractades amb tècniques remotes es representen en els models DFN amb resultats molt acceptables. La tècnica remota que més s’aproxima a l’aflorament depèn de la propietat geomètrica que s’ha capturat i representat. La disposició de l’aflorament i les orientacions de les fractures també influencien en la captura de les dades. Al analitzar els resultats de la porositat i permeabilitat de la xarxa de fractures s’obté valors similars en els tres models DFN. En conseqüència les diferències geomètriques representades per cada tècnica en conjunt, no han condicionat en gran mesura l’estimació d’aquests valors, mostrant models de fractures equivalents en el cas de la Fou. Els models DFN que actualment s’obtenen mitjançant sensors remots, permeten mesurar i reduir incerteses en la caracterització dels models regionals tridimensionals. Això serà objecte de noves investigacions tant acadèmiques com industrials. Finalment, els treballs duts a terme a Montserrat demostren que l’aplicació de tècniques remotes en el camp de la digitalització i la identificació de diferències, durant el seguiment d’afloraments, és precisa i d’alta resolució. Aquesta nova metodologia que integra d’algorismes de machine learning per a la identificació i classificació de blocs caiguts demostra, especialment a Degotalls- E, que l’aplicació d’aquest sistema genera resultats acceptables en termes de Vertader Positiu (VP), Fals Positiu (FP) i Fals Negatiu (FN). Aquesta automatització, permet apropar-nos un pas més cap a implementar un sistema d’alerta de caiguda de blocs, amb un dispositiu TLS integrat, que permeti i faciliti la detecció de deformacions prèvies a caigudes, comportant una aplicació de la tècnica utilitzada durant el desenvolupament d’aquesta tesi al món industrial., [eng] he ongoing digital transition results in millions of data points gathered every day. Within the context of geosciences, the use of digital remote sensing techniques is increasing. Machine learning techniques are interesting alternative to support the processing and interpretation of large amounts of collected data. The present industrial PhD focuses on the analysis of fractured outcrops using remote sensing techniques and machine learning with two specific objectives; (1) the evaluation of how representative are remote sensing techniques to identify fractures and subsequently represent them in discrete fracture network (DFN) models; and (2) the use of classification machine learning models in processes associated with fractured environments, such as block falls. METHODOLOGY: The first case study, evaluates remote techniques for data gathering (Terrestrial Laser Scanner (TLS) and Photogrammetry (SfM)). The experiment was carried out in active La Fou quarry in Barcelona. Fractures surfaces were reconstructed from a large number of remotely captured data (Santana et al., 2012; Geyer et al., 2015), the fractures network is characterized by multiple families (García-Sellés et al., 2018). With these results, and those collected with manual scanline techniques, three-dimensional (DFN) models are constructed. One model is built using data from each capture technique (DFNTLS, DFNSfM and DFNscan-line), where the main geometric properties of the fractures will be represented. Finally, the fracture trace maps are generated to calculate the porosity and permeability using aperture data for each of the scenarios. The second case study is aimed at automating the classification of block falls. The study area is located on the historically unstable Degotalls escarpments, close to the access road of the Monastery of Montserrat. Monitoring of these escarpments using periodic TLS scanning began in 2007, as a result of multiple falling block episodes starting back in 2001, which affected the Monastery car park (limiting access to the Sanctuary on several occasions). The methodology consists of aligning two scans or point clouds of the same outcrop, captured at two different times, aiming to detect clusters of differences between the scans. Subsequently, each cluster of differences is visually reviewed to identify and classify block falls. The methodology developed in this Thesis, incorporates machine learning algorithms to automate this last stage of identification and classification of fallen blocks. RESULTS: In the first case study, the main geometric properties for each fracture captured and processed are position, orientation, height, and length. These can be used to characterize fracture statistics such as spacing, intensity, P10, P20, P21 and P32, among others. For individual families are identified Family I WNW-ESE, Family II NNW/SSE, Family III, N-S and Family IV NE-SW. The technique with the highest number of characterised fractures was TLS with more than 52,000, followed by photogrammetry with 1,094 and finally scan-lines with 398 fractures measured. Fracture height and length are very variable, average values range from a few centimetres to 18m. TLS average lengths are less than one metre. The spacings between fractures in Family I have centimetric sizes and the rest of the families have more variable spacing from centimetric to 6m. Photogrammetry obtains values that sometimes double or triple the rest of the techniques. Furthermore, with the input of the fracture openings into the model, trace maps have been generated and porosity and permeability value of the fracture network has been obtained. In the three DFN models, the average porosity value obtained is between 5.4-5.7% and the average permeability values between 4x105 mD and 6.2x105mD. In the second case study, the results generated from the combination of 11 classifier algorithms and 15 resampling methods, allow us to identify the best predictive combination for the classification of fallen blocks, according to the temporal period and steepness. In Degotalls-E, from approximately 5,800 clusters identified by each monitoring, a 98% reduction is achieved in the number of clusters to be validated, between False Positives (FP) and True Positives (VP), understanding (VP) as the real block falls. In Degotalls-N the reduction would be 80.16% for a complete identification, while for identifying 96% of the real block drops, the reduction would be 90%, from a population of about 3,700 initial clusters. It is also important to note that the validation of the task by comparing high-resolution images has significantly increased the reliability of the results obtained. CONCLUSIONS AND THE WAY FORWARD: In the evaluation of the representativeness of using remote sensing techniques for the construction of DFN models, the results suggest that data captured and processed with remote techniques are represented in DFN models with very acceptable results. The remote sensing technique that more accurately approximates the outcrop depends on the geometric property that need to be captured and represented. The layout of the outcrop and fracture orientations also influence data capture. When analysing the results for permeability and porosity of the fracture network, similar values are obtained for all three DFN models. Consequently, the geometrical differences of the models by each technique do not greatly influence the estimation of these values, showing equivalent fracture models in the case of the Fou. The DFN models, which are currently obtained using remote sensors allow the uncertainties generated in the three- dimensional regional models to be characterized measured and reduced. This will be the obhect of further research both academic and industrial. Finally, the work carried out at Montserrat shows that the application of remote techniques in the field of digitisation and the identification of differences during the monitoring of outcrops is precise and of high resolution. This new methodology integrating machine learning algorithms for the identification and classification of fallen blocks demonstrates, especially in Degotalls-E, that applying this system generates acceptable results in terms of True Positive (VP), False Positive (FP) and False Negative (FN). This automation will eventually allow to implement a block drop alert system, with an integrated TLS device, which allows and facilitates the detection of precursors, involving an application of the technique used during the development of this thesis in the industrial world.

El treball en qüestió tracta de l’estudi i desenvolupament d’un model predictiu mitjançant ciència de dades basat en un conjunt de perfils de clients enregistrats per una campanya de màrqueting telefònic associada a una entitat bancària portuguesa. Per portar a terme el projecte s’ha aplicat la metodologia CRISP-DM i s’han implementat les eines necessàries mitjançant llibreries de Python com Pandas o scikit-learn. S’han desenvolupat models amb diferents enfocaments per tal de satisfer potencials escenaris d’interès de l’empresa encarregada de la campanya. La capacitat predictiva dels mateixos ha estat avaluada mitjançant diverses mètriques i s’han extret conclusions sobre el seu rendiment tenint en compte el grau de complexitat que presentaven. Els resultats obtinguts mostren com si bé els models basats en algorismes més complexos tendeixen a oferir una millor capacitat predictiva, depenent dels objectius de l’empresa un model més senzill pot resultar suficient i aportar unes prestacions similars El trabajo en cuestión se basa en el estudio y desarrollo de un modelo predictivo mediante ciencia de datos a partir de un conjunto de perfiles de clientes provenientes de una campaña de marketing telefónico asociada a una entidad bancaria portuguesa. Para llevar a cabo el proyecto se ha aplicado la metodología CRISP-DM y se han implementado las herramientas necesarias a través del uso de librerías de Python como Pandas o scikit-learn. Se han desarrollado modelos con enfoques distintos para tal de satisfacer potenciales escenarios de interés de la empresa encargada de la campaña. La capacidad predictiva de los mismos ha sido evaluada mediante diversas métricas y se han extraído conclusiones de su rendimiento teniendo en cuenta el grado de complejidad que presentasen. Los resultados muestran como si bien en líneas generales, modelos basados en algoritmos de mayor complejidad tienden a ofrecer una mejor capacidad predictiva, dependiendo de los objetivos de la empresa un modelo más sencillo puede resultar suficiente y aportar unas prestaciones similares The project bases itself in the study and development of a predictive model powered by data science and built from a data set of client profiles registered from a marketing telephonic campaign associated with a Portuguese banking institution. The methodology applied throughout the project has been CRISP-DM and the procedure required the implementation of specific tools that have been accessed through Python libraries such as Pandas or scikit- learn. Prediction models have been developed based on different approaches to satisfy potential scenarios of interest for the company behind the marketing campaign. The predictive capabilities of the models have been evaluated using different metrics and conclusions of their performance were extracted considering the degree of complexity they presented. The results obtained show how models based on more complex algorithms tend to come with better predictive capabilities. However, depending on the objectives set by the company, a simpler model may be sufficient and showcase similar outcomes

The diagnosis and treatment of epilepsy depend on accurate seizure detection. In clinical practice, the evaluation of seizures is done by visual inspection of an electroencephalogram (EEG). it is very time­consuming and requires trained experts. Automatic seizure detection is important. Machine learning approaches are intensely being applied to this problem due to their ability to classify seizure conditions from a large amount of data, and provide pre­screened results for neurologists. This work proposes a variety of experiments with different machine­learning architectures (support vector machine SVM, K nearest neighbour KNN, random forest RF, feef forward neural network FFNN and convolutional neural network CNN) for the detection of epileptic seizures using multichannel EEG signals from the CHBT­MIT Scalp EEG Database. The best model built in this work contains a combination of a feed­forward neural network (FFNN) and a convolutional neural network (CNN). CNN input images are constructed by applying short­time Fourier transform (STFT) to electroencephalography (EEG) signals and then merged with statistical metrics into a FFNN. The best model of this project showed an outstanding performance of 98.615% accuracy, 98.737% sensitivity and 98.425% specificity. This work also includes a discussion of other exciting ideas that could lead to future research investigations.

L’infart de miocardi és una de les causes principals de la mort d’homes i dones arreu del món. Es dona lloc quan les artèries coronàries s’estrenyen, a causa d’un coàgul de sang i l’aterosclerosis, llavors s’obstrueix totalment el pas de la sang al llarg de l’artèria amb la subsegüent manca de reg i mort d’una zona del cor. Aproximadament un 30 % dels pacients que pateixen un infart moren abans d’arribar a l’hospital, mentre que la mortalitat dins de l’hospital és menor, tot i que també és elevada. Un dels aspectes més importants quan un pacient ingressa amb una IM és la decisió direc‐ ta sobre quin tractament és necessari realitzar‐li, ja que té un efecte directe en les probabilitats de supervivència del pacient. Aquest projecte utilitzarà una base de dades disponible sobre pacients amb diagnòstic d’infart de miocardi, la qual ja ha estat imputada i equilibrada anteriorment per a diagnosticar la mortalitat als pacients. L’objectiu final serà emprar aquestes dades imputades y balancejada, a les quals se’ls aplicaran el model d’aprenentatge automàtic escollit, per així poder estudiar, analitzar i interpretar el comportament de les variables utilitzades dins del model i relacionar la predicció amb les compli‐ cacions de l’infart de miocardi. Aquesta tesi es pot considerar com el tancament a l’estudi de la predicció del IM, ja que en tre‐ balls anteriors [2] i [3] va ser estudiada la millor tècnica d’imputació i de balanceig per a aquestes dades, les quals es troben detallades posteriorment. A més a més, es realitza una interpretació de les variables més importants per a la predicció del IM. L’estudi de les variables es realitza a partir d’una base de dades obtinguda per S.E. Golovenkin, A.N. Gorban, E.M.Mirkes, V.A. Shulman, D.A. Rossiev, P.A. Shesternya, S.Yu. Nikulina, Yu.V. Orlova, and M.G. Dorrer [1], on es descriuen els paràmetres clínics més característics de 1700 pacients du‐ rant el primer dia d’entrada a l’UCI, i un cop transcorregudes les primeres 24, 48 i 72 hores d’admissió a l’hospital. Un resultat precís i correcte d’aquest estudi serà molt crucial, ja que permetrà ajudar els metges a poder predir un IM en un pacient una vegada s’hagi dut a terme l’estudi previ a l’entrada a la unitat de cures intensives. El infarto de miocardio es una de las principales causas de la muerte de hombres y mujeres en todo el mundo. Éste ocurre cuando las arterias coronarias se estrechan, debido a un coágulo de san‐ gre, la aterosclerosis, la cual obstruye totalmente el paso de sangre a lo largo de la arteria con la subsiguiente falta de riego y muerte de una zona del corazón. Aproximadamente un 30 % de los pa‐ cientes con infarto fallece antes de llegar al hospital, mientras que la mortalidad dentro del hospital es menor, pero también elevada. Uno de los aspectos más importantes cuando un paciente ingresa con un infarto de miocardio es la decisión directa sobre qué tratamiento realizar, ya que va relacio‐ nado con la probabilidad de supervivencia del paciente. Este proyecto utilizará una base de datos disponibles sobre pacientes con diagnóstico de Infarto de miocardio, la cual ya ha sido imputada y equilibrada anteriormente. El objetivo final será em‐ plear estos datos imputados y balanceados a los que se le aplicarán el mejor modelo clasificador de aprendizaje automático, para así estudiar, analizar e interpretar el comportamiento de las variables utilizadas dentro del modelo de predicción y relacionar la predicción con las complicaciones del in‐ farto de miocardio. Esta tesis se puede considerar como el cierre al estudio de la predicción del IM (“Infarto de mio‐ cardio”), puesto que en trabajos anteriores [2] y [3] se estudió la mejor técnica de imputación y de balanceado para estos datos, los cuales se detallarán más adelante, y en esta además se realiza el estudio del mejor modelo de predicción y la interpretación de las variables más importantes para la predicción del IM. El estudio de las variables se realizará a partir de una base de datos obtenida por S.E. Golo‐ venkin, A.N. Gorban, E.M.Mirkes, V.A. Shulman, D.A. Rossiev, P.A. Shesternya, S.Yu. Nikulina, Yu.V. Orlova, and M.G. Dorrer [1], donde se describen los parámetros clínicos más característicos de 1700 pacientes, durante el primer día de entrada en la UCI y una vez transcurridas las primeras 24, 48 y 72 horas de admisión en el hospital. Un resultado preciso y concreto de este estudio será muy importante ya que podrá ayudar a los médicos a poder predecir un IM en un paciente una vez hayan hecho el estudio previo de entrada a la unidad de cuidados intensivos. Myocardial infarction is one of the leading causes of death for men and women worldwide. It occurs when the coronary arteries become narrowed, due to a blood clot and atherosclerosis, which completely blocks the passage of blood along the artery with subsequent lack of blood supply and death of an area of the heart. Approximately 30 percent of heart attack patients die before reaching hospital, while in‐hospital mortality is lower, but also high. One of the most important aspects when a patient is admitted with an MI is the direct decision on what treatment to give, as it is related to the patient’s chance of survival. This project will use an available database of patients with a diagnosis of myocardial infarction, which has been previously imputed and balanced to diagnose patient mortality. The final objective will be to use this imputed data to which the models used will be applied, in order to study, analyse and interpret the behaviour of the variables used within the models and to relate the prediction to the complications of myocardial infarction in an autonomous learning model. This thesis can be considered as the closure to the study of MI prediction, since in previous works [2] and [3] the best imputation and equilibrium technique for these data was studied, which will be detailed later, and in this thesis an interpretation of the most important variables for the prediction of MI is also carried out. The study of the variables will be carried out on the basis of a database obtained by S.E. Golo‐ venkin, A.N. Gorban, E.M.Mirkes, V.A. Shulman, D.A. Rossiev, P.A. Shesternya, S.Yu. Nikulina, Yu.V. Orlova, and M.G. Dorrer [1], describing the most characteristic clinical parameters of 1700 patients, during the first day of admission to the ICU and after the first 24, 48 and 72 hours of admission to the hospital. A good result of this study will be very important as it may help clinicians to predict MI in a pa‐ tient once they have done a pre‐intensive care unit admission study.

The project consist of the study of the popularity of online news, with given data, the objective is to analyze the popularity index of the news by training Machine Learning models and evaluating their performance. El proyecto consiste en el estudio de la popularidad de las noticias online, con datos dados, el objetivo es analizar el índice de popularidad de las noticias entrenando modelos de Machine Learning y evaluando su desempeño. El projecte consisteix en l'estudi de la popularitat de les notícies en línia, amb dades donades, l'objectiu és analitzar l'índex de popularitat de les notícies formant models d'aprenentatge automàtic i avaluant-ne el rendiment.

This project is focused on the development of an audio-denoising system to improve the robustness of audio fingerprinting algorithms. The state of the art on deep learning has advanced in the last years, specifically in multimedia processing, this opens a door to using this kind of tool for the described task. The goal is to develop software that learns to clean noisy audio samples so the cleaned samples are better matched. First, a review of the previous work on different topics is made. Then, different ways to create a dataset are presented as well as a deep neural network to clean the audio, also an evaluation step has been developed to compare fairly the different models trained. Results suggest that the neural network is learning the task to perform, despite this, there is a large margin to improve the evaluation results. Furthermore, the limitations in terms of memory usage and inference speed are described. Finally, a discussion introduces the necessity of a larger and more robust dataset and the idea of tailoring the dataset to the collection of songs to be denoised. Este proyecto se centra en el desarrollo de un sistema de eliminación de ruido de audio para mejorar la robustez de los algoritmos de detección basados en la extracción de huellas digitales acústicas. El estado del arte en el aprendizaje profundo ha avanzado en los últimos años, específicamente en el procesamiento multimedia, lo que abre una puerta para usar este tipo de herramienta para la tarea descrita. El objetivo es desarrollar un software que aprenda a limpiar muestras de audio ruidosas para que las muestras limpias se identifiquen mejor. Primeramente, realizamos una revisión del trabajo previo sobre diferentes temas. Luego, presentamos diferentes formas de crear un conjunto de datos, así como una red neuronal profunda para limpiar el audio, también hemos desarrollado un módulo de evaluación para comparar de manera justa los diferentes modelos entrenados. Los resultados sugieren que la red neuronal está aprendiendo la tarea a realizar, a pesar de esto, hay un gran margen para mejorar los resultados de evaluación. Además, describimos las limitaciones en términos de uso de memoria y velocidad de inferencia. Finalmente, se presenta la discusión de la necesidad de un conjunto de datos más grande y robusto y la idea de adaptar el conjunto de datos a la colección de canciones que queremos detectar. Aquest projecte se centra en el desenvolupament d'un sistema d'eliminació de soroll d'àudio per millorar la robustesa dels algoritmes d'empremtes digitals d'àudio. L'estat de l'art en l'aprenentatge profund ha avançat en els darrers anys, específicament en el processament multimèdia, això obre una porta per utilitzar aquest tipus d'eina per a la tasca descrita. L'objectiu és desenvolupar un programari que aprèn a netejar mostres d'àudio amb soroll perquè les mostres netes es detectin millor. En primer lloc, realitzem una revisió del treball previ sobre diferents temes. Després, presentem diferents formes de crear un conjunt de dades, així com una xarxa neuronal profunda per netejar l'àudio, també hem desenvolupat un pas d'avaluació per comparar de manera justa els diferents models entrenats. Els resultats suggereixen que la xarxa neuronal està aprenent la tasca a fer, malgrat això, hi ha un gran marge per millorar els resultats d'avaluació. A més, presentem les limitacions en termes d'ús de memòria i velocitat d'inferència. Finalment, presentem la discussió de la necessitat d'un conjunt de dades més gran i robust i la idea de personalitzar el conjunt de dades per a la col·lecció de cançons que es vulguin detectar.

The number of nano-sensors connected to wireless electromagnetic nano-network generates different traffic volumes that have increased dramatically, enabling various applications of the Internet of nano-things. Nano-network traffic classification is more challenging nowadays to analyze different types of flows and study the overall performance of a nano-network that connects to the Internet through micro/nanogateways. There are traditional techniques to classify traffic, such as port-based technique and load-based technique, however the most promising technique used recently is machine learning. As machine learning models have a great impact on traffic classification and network performance evaluation in general, it is difficult to declare which is the best or the most suitable model to address the analysis of large volumes of traffic collected in operational nano-networks. In this paper, we study the classification problem of nano-network traffic captured by micro/nano-gateway, and then five supervised machine learning algorithms are used to analyze and classify the nano-network traffic from traditional traffic. Experimental analysis of the proposed models is evaluated and compared to show the most adequate classifier for nano-network traffic that gives very good accuracy and performance score to other classifiers. This work was supported in part by the ‘‘Agencia Estatal de Investigación’’ of ‘‘Ministerio de Ciencia e Innovación’’ of Spain under Project PID2019-108713RB-C51/MCIN/AEI/10.13039/501100011033, and in part by the ‘‘Agència de Gestió d’Ajuts Universitaris i de Recerca’’ (AGAUR) of the ‘‘Generalitat de Catalunya’’ under Grant 2021FI_B2 00091.

Causal Queries are usually estimated by means of an estimand, a formula consisting of observational terms that can be computed using passive data. Each query results in a different formula, which makes estimand-based methods extremely ad-hoc. In this work, we propose an estimand-agnostic framework capable of computing any identifiable causal query on an arbitrary Causal Graph (even in the presence of latent confounders) with only one general model. We provide multiple implementations of this general framework that leverage the expressive power of Neural Networks and Normalizing Flows to model complex distributions, and we derive estimation procedures for all kinds of observational, interventional and counterfactual queries, valid for any kind of graph for which the query is identifiable. Finally, we test our techniques in a modelling setting and an estimation benchmark to show how, despite being a query-agnostic framework, it can compete with query-specific models. Our proposal includes an open-source library that allows easy application and extension of our techniques for researchers and practitioners alike.

The emergence of deep learning has considerably advanced the state-of-the-art in cardiac magnetic resonance (CMR) segmentation. Many techniques have been proposed over the last few years, bringing the accuracy of automated segmentation close to human performance. However, these models have been all too often trained and validated using cardiac imaging samples from single clinical centres or homogeneous imaging protocols. This has prevented the development and validation of models that are generalizable across different clinical centres, imaging conditions or scanner vendors. To promote further research and scientific benchmarking in the field of generalizable deep learning for cardiac segmentation, this paper presents the results of the Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation (M&Ms) Challenge, which was recently organized as part of the MICCAI 2020 Conference. A total of 14 teams submitted different solutions to the problem, combining various baseline models, data augmentation strategies, and domain adaptation techniques. The obtained results indicate the importance of intensity-driven data augmentation, as well as the need for further research to improve generalizability towards unseen scanner vendors or new imaging protocols. Furthermore, we present a new resource of 375 heterogeneous CMR datasets acquired by using four different scanner vendors in six hospitals and three different countries (Spain, Canada and Germany), which we provide as open-access for the community to enable future research in the field.

In this study, we address the problem of classification of carrot fruit in order to manage and control their waste using improved deep neural networks. In this work, we perform a deep study of the problem of carrot classification and show that convolutional neural networks are a straightforward approach to solve the problem. Additionally, we improve the convolutional neural network (CNN) based on learning a pooling function by combining average pooling and max pooling. We experimentally show that the merging operation used increases the accuracy of the carrot classification compared to other merging methods. For this purpose, images of 878 carrot samples in various shapes (regular and irregular) were taken and after the preprocessing operation, they were classified by the improved deep CNN. To compare this method with the other methods, image features were extracted using Histograms of Oriented Gradients (HOG) and Local Binary Pattern (LBP) methods and they were classified by Multi-Layer Perceptron (MLP), Gradient Boosting Tree (GBT), and K-Nearest Neighbors (KNN) algorithms. Finally, the method proposed based on the improved CNN algorithm, was compared with other classification algorithms. The results showed 99.43% of accuracy for grading carrot through the CNN by configuring the proposed Batch Normalization (BN)-CNN method based on mixed pooling. Therefore, CNN can be effective in increasing marketability, controlling waste and improving traditional methods used for grading carrot fruit.

El Machine Learning disposa de la tecnologia necessària per aplicar tècniques estadístiques i algoritmes a grans volums de dades, per tal d’entendre i respondre d’una manera automàtica a la informació. Aquesta tecnologia està essent cada cop més utilitzada per diferents empreses i sectors en aplicacions com el reconeixement facial o de la parla, l’anàlisi del comportament del consum o la productivitat i en la predicció del clima o el trànsit. En aquest projecte es mostra la relació de necessitat que manté la intel·ligència artificial amb l’ús de tècniques estadístiques, mitjançant la realització de dos estudis de classificació i un de regressió. En ells s’usen quatre algoritmes d’aprenentatge supervisat que requereixen tècniques estadístiques per obtenir classificacions i prediccions, i són: kernel-Nearest Neighbors, Naive Bayes, Decision Trees i Support Vector Machines. En els estudis, es determina que no tots els algoritmes presenten la mateixa precisió a l’hora de classificar i predir, i que aquests depenen de l’estructura i el format de les dades per adaptar-se i ajustar-se correctament al model. A més, es conclou que per aconseguir bons resultats, és necessari que el conjunt de dades disposi del nombre d’observacions necessàries en cadascuna de les classes perquè l’algoritme pugui entrenar-les profundament, i també, es percep que hi han classes que es classifiquen amb més facilitat que altres. El Machine Learning dispone de la tecnología necesaria para aplicar técnicas estadísticas y algoritmos a grandes volúmenes de datos con tal de entender y responder de una manera automática a la información. Esta tecnología está siendo cada vez más usada por diferentes empresas y sectores en aplicaciones como el reconocimiento facial o del habla, el análisis del comportamiento del consumo o la productividad y en la predicción del clima o del tráfico. En este proyecto se muestra la relación de necesidad que mantiene la inteligencia artificial con el uso de métodos estadísticos, mediante la realización de dos estudios de clasificación y uno de regresión. En ellos se utilizan cuatro algoritmos de aprendizaje supervisado que requieren técnicas estadísticas para obtener clasificaciones y predicciones, y son: kernel-Nearest Neighbors, Naive Bayes, Decision Trees y Support Vector Machines. En los estudios, se determina que no todos los algoritmos presentan la misma precisión a la hora de clasificar y predecir, y que estos dependen de la estructura y el formato de los datos para adaptarse y ajustarse correctamente al modelo. Además, se concluye que para obtener buenos resultados, es necesario que el conjunto de datos disponga del número de observaciones necesarias en cada clase para que el algoritmo pueda entrenarlas profundamente, y también, se observa que hay clases que se clasifican con más facilidad que otras. Machine Learning has the necessary technology to apply statistical techniques and algorithms to large volumes of data in order to understand and automatically respond to information. This technology is being increasingly used by different companies and sectors in applications such as facial recognition or speech, analysis of consumer behaviour or productivity, and weather or traffic prediction. In this project the relationship between artificial intelligence and the use of statistical techniques is shown through the execution of two classification studies and one regression study. Four supervised learning algorithms that require statistical techniques to obtain classifications and predictions are used: kernel-Nearest Neighbors, Naive Bayes, Decision Trees and Support Vector Machines. In the studies, it is determined that not all algorithms have the same accuracy in classifying and predicting, and that these depend on the structure and format of the data to adapt and fit correctly to the model. In addition, it is concluded that in order to obtain good results, the data set must have the necessary number of observations in each class so that the algorithm can train them deeply, and it is also perceived that there are classes that are classified more easily than others.

The goal of this work is to optimize the sedimentation stage of a water purification plant, by real-time adjusting the use of coagulant in relation to the quality of the incoming water, in order to reduce costs and prevent the consequences of unexpected changes. Coagulants are used to aggregate particles and suspended impurities in the water, allowing them to settle. To achieve this goal, a predictive model has been build using the Extra Trees Regressor algorithm, which is capable of calculating the coagulant setpoint based on the quality parameters of the river, the dosage of dioxide in the water, and the turbidity in the sedimentation stage. The model has been developed using a set of 57,481 samples collected at the ETAP of Sant Joan Despí (data cleaning was carried out and 15,878 samples were eliminated, 10% has been used for validation). Finally, a validation of the model was performed which gave an R2 of 97.8% and a MAPE of 0.05%, which demonstrates its viability as a useful tool for the improve, efficiency and quality of the water treatment process. El presente trabajo tiene como objetivo optimizar la etapa de decantación de una planta de potabilización de agua, ajustando a tiempo real el uso del coagulante en relación a la calidad del agua de entrada, con el fin de reducir costos y prevenir las consecuencias de cambios inesperados. Los coagulantes son utilizados para aglomerar partículas e impurezas suspendidas en el agua, permitiendo que estas sedimenten. Para llevar a cabo el objetivo se ha construido un modelo predictivo mediante el algoritmo Extra Trees Regressor, capaz de calcular la consigna de coagulante en función de los parámetros de calidad del río, la dosificación de dióxido en el agua y la turbidez en la etapa de decantación. El modelo ha sido desarrollado mediante un conjunto de 57.481 muestras recolectadas en la ETAP de Sant Joan Despí (se ha realizado una limpieza de datos con la que se han eliminado 15.878 muestras y se ha usado un 10% para la validación). Finalmente se ha realizado una validación del modelo que ha dado un R2 del 97.8% y un MAPE del 0.05%, que demuestran su viabilidad como herramienta útil para la mejora, la eficiencia y calidad del proceso de tratamiento del agua. Aquest treball té com a objectiu optimitzar l'etapa de decantació d'una planta de potabilització d'aigua, ajustant en temps real l'ús del coagulant en relació a la qualitat de l'aigua d'entrada, amb la finalitat de reduir costos i prevenir les conseqüències dels canvis inesperats. Els coagulants s'utilitzen per aglomerar partícules i impureses suspeses en l'aigua, permetent que aquestes sedimentin. Per portar a terme l'objectiu s'ha construït un model predictiu mitjançant l'algoritme Extra Trees Regressor, capaç de calcular la consigna de coagulant en funció dels paràmetres de qualitat del riu, la dosificació de diòxid en l'aigua i la terbolesa en l'etapa de decantació. El model s'ha desenvolupat mitjançant un conjunt de 57.481 mostres recollides en l'ETAP de Sant Joan Despí (s'ha realitzat una neteja de dades amb la qual s'han eliminat 15.878 mostres i s'ha utilitzat un 10% per a la validació). Finalment s'ha realitzat una validació del model que ha donat un R2 del 97.8% i un MAPE del 0.05%, que demostren la seva viabilitat com a eina útil per a la millora, l'eficiència i qualitat del procés de tractament de l'aigua.

This report presents SIEVA, an AI-powered software that utilizes internet traffic logs to provide a taxonomy for classifying cyberattacks and intrusions using the MITRE framework. The software is divided into three parts: a database for storing logs, an AI engine for classifying and mapping the logs, and a graphical user interface for visualizing the results. The AI engine will implement a natural language processing machine learning classifier to classify the logs, and experimentation with name entity recognition techniques will be conducted to gain a better understanding of the logs. The ultimate goal of the project is to provide a user-friendly and efficient way for visualizing potential threats to a network using internet traffic logs.

Recent advancements in vehicle-to-everything (V2X) communication have notably improved existing transport systems by enabling increased connectivity and driving autonomy levels. The remarkable benefits of V2X connectivity come inadvertently with challenges which involve security vulnerabilities and breaches. Addressing security concerns is essential for seamless and safe operation of mission-critical V2X use cases. This paper surveys current literature on V2X security and provides a systematic and comprehensive review of the most relevant security enhancements to date. An in-depth classification of V2X attacks is first performed according to key security and privacy requirements. Our methodology resumes with a taxonomy of security mechanisms based on their proactive/reactive defensive approach, which helps identify strengths and limitations of state-of-the-art countermeasures for V2X attacks. In addition, this paper delves into the potential of emerging security approaches leveraging artificial intelligence tools to meet security objectives. Promising data-driven solutions tailored to tackle security, privacy and trust issues are thoroughly discussed along with new threat vectors introduced inevitably by these enablers. The lessons learned from the detailed review of existing works are also compiled and highlighted. We conclude this survey with a structured synthesis of open challenges and future research directions to foster contributions in this prominent field. This work is supported by the H2020-INSPIRE-5Gplus project (under Grant agreement No. 871808), the ”Ministerio de Asuntos Económicos y Transformacion Digital” and the European Union-NextGenerationEU in the frameworks of the ”Plan de Recuperación, Transformación y Resiliencia” and of the ”Mecanismo de Recuperación y Resiliencia” under references TSI-063000-2021-39/40/41, and the CHIST-ERA-17-BDSI-003 FIREMAN project funded by the Spanish National Foundation (Grant PCI2019-103780).

There is no doubt that Transformers have revolutionized the field of Natural Language Processing (NLP). In 2017, by replacing convolutions and recurrence with attention mechanisms, they achieved State of the art in a number of NLP tasks. One of these tasks is Natural Language Inference (NLI), which tests whether a hypothesis is true (entailment), false (contradiction), or undetermined (neutral) given a premise. Zero-shot text classification tasks (0shot TC), i.e. classifying text without any single labeled instance, have greatly benefited from the fact that Transformers excel in NLI. A clear example is this master's thesis, which will provide a use case for 0shot TC in a business environment. In particular, this work has been developed while working at Fever as a Data Scientist, and its main objective has been to label Fever's products using 0shot TC. The main contribution of this work is creating a ML model flexible enough to quickly adapt to business needs. In other words, developing a model where labels can be added and removed at ease and it can be extended to other areas (cross domain). Furthermore, Large Language Models will prove that generalization is possible even in a low resource setting. Lastly, it should be noted that this thesis is presented as a self-contained document and it can be understood by a reader with no prior knowledge of Machine Learning and Fever's business. The bulk of the thesis consists of an exhaustive study of the application of 0shot TC in this particular business problem, as well as the posterior optimization of the models used. In addition, semi supervision will be applied to improve performance in an environment without tagged data.

El mercado de activos y criptomonedas ha crecido significativamente en los últimos años al igual que el número de jóvenes interesados en invertir y ganar dinero "fácil". Este aumento, sin embargo, no se ha acompañado de un incremento en formación, resultando en un gran número de errores evitables y la pérdida de grandes cantidades de dinero. Una metodología comúnmente utilizada para minimizar dichos riesgos consiste en consultar las redes sociales en busca de expertos que orienten o de identificar tendencias. El uso de la inteligencia artificial puede ser muy beneficioso para automatizar los pasos seguidos en esta metodología, en combinación con métricas como la cantidad de menciones o el sentimiento asociado a los comentarios de las redes sociales. En este proyecto se pretende desarrollar una aplicación capaz de obtener y procesar datos de redes sociales en tiempo real con la que fomentar la inversión consciente y no especulativa, valorando también el rendimiento de la base de datos utilizada y creando una página web donde los usuarios puedan consultar las distintas métricas generadas. The asset market has grown significantly in the last few years as well as the number of young people interested in investing and making "easy" money. This increase in demand, however, has not been met with an increase in formation or education, resulting in lots of avoidable mistakes and the loss of considerable amounts of money. A common approach on minimizing losses consists of using social media to look for the guidance of specialists or try to identify the latest trends. The use of artificial intelligence can be highly beneficial to automate the steps followed by this approach along with the use of metrics such as the number of mentions or the sentiment of the comments. The purpose of this project is to develop an app capable of obtaining and processing data from social media in real time, automating the process of identifying trends and promoting conscious investment against speculation. Besides the AI engine behind the app, this project also focuses on the efficiency and performance of the chosen database and the implementation of a webpage where users can see the processed data and generated metrics.

Les empreses d'avui en dia necessiten una gran quantitat d'informació per portar a terme qualsevol dels seus processos interns. El creixent volum de la documentació i el temps dedicat a la cerca d'informació en documents per part dels treballadors fa imprescindible la implementació de sistemes de gestió documental, per tal d'optimitzar totes aquestes tasques intermèdies. Un gran nombre d'entitats financeres es veuen amb la necessitat d'automatitzar tots aquells tràmits documentals essencials per conèixer certa informació dels seus clients. A l'hora de concedir o denegar qualsevol préstec és primordial disposar de tota la informació personal i econòmica respectiva a aquell client en particular. Aquestes dades es troben encapsulades en documents com per exemple: nòmines, declaracions de la renda, factures, etcètera. Tot el procés respectiu a l'extracció manual de dades comporta un cost temporal alt, i en conseqüència, una major espera en la concessió o denegació del préstec. Per aquest motiu, moltes d'aquestes empreses busquen solucions enfocades en l'automatització d'aquests processos documentals. Aquest projecte té com a objectiu suplir aquestes necessitats, construint un sistema automàtic capaç de portar a terme la classificació de certes tipologies documentals i l'extracció de camps específics. Aquest mòdul ha de ser capaç de retornar totes aquelles dades que necessitin els equips de gestió de riscos i scoring de l'entitat financera per concedir o denegar els préstecs sol·licitats. Estudiant les diferents tècniques i mètodes existents, i tenint en compte totes les opcions possibles, s'opta per la implementació d'un sistema de processament intel·ligent particular adaptat al problema a resoldre. S'apliquen tècniques d'Aprenentatge Automàtic (CNN) per classificar els diferents tipus documentals considerats. A més a més, s'empren mètodes d'extracció textual seguint estratègies específiques, amb la finalitat de maximitzar l'extracció de dades a partir del procés automàtic. Today's companies need a lot of information to carry out any of their internal processes. The increasing volume of documentation and the time spent searching for information in documents by workers makes it essential to implement document management systems, in order to optimize all these intermediate tasks. Numerous financial institutions are faced with the need to automate all those essential documentary processes in order to know certain information of their customers. When granting or denying any loan, it is essential to have all the personal and economic information regarding that particular client. These data are encapsulated in documents such as: payrolls, income statements, invoices, etc. The whole process of extracting data manually involves a high time cost, and consequently, a longer wait in the granting or refusal of the loan. For this reason, many companies are looking for solutions focused on the automation of these documentary processes. This project aims to meet these needs, building an automatic system capable of performing the classification of certain documentary typologies and the extraction of specific fields. This module must be able to return all the data needed by the financial institution's risk management and scoring teams to grant or deny the requested loans. By studying the different techniques and methods available, and taking into account all possible options, we opt for the implementation of a particular intelligent processing system adapted to the problem to be solved. Machine Learning (CNN) techniques are applied to classify the different documentary types considered. In addition, textual extraction methods are used following specific strategies, with the aim of maximizing the extraction of data from the automatic process.

Aplicar pruning en una xarxa neuronal disminueix la complexitat de la representació interna, permetent així un augment de la velocitat d'entrenament i inferència. A més, en ser un model més senzill, pot millorar la seva capacitat de generalització, per això l'aplicació de tècniques de pruning en xarxes preentrenades pot ajudar a facilitar l'aplicació de transfer learning i aconseguir millors models. En reduir la complexitat del model, també reduïm la quantitat de paràmetres necessaris per entrenar en la nova tasca i, per tant, reduïm la capacitat computacional necessària, facilitant l'aplicació de transfer learning. El nostre treball es basarà a aplicar múltiples mètodes de pruning sobre models preentrenats per tal de conèixer com afecta posteriorment a l'aplicació de transfer learning. Applying pruning on a neural network decreases the complexity of the internal representation, allowing an increase in the speed of training and inference. In addition, producing a simpler model, it can improve its generalization capacity, which is why applying pruning techniques on pre-trained neural networks can help improve the application of transfer learning and achieve better models. By reducing the complexity of the model, we also reduce the number of parameters needed to train on the new task, therefore reducing the computational capacity required, and improving the application of transfer learning. Our work will be based on applying multiple pruning methods on pre-trained models in order to know how it affects the application of transfer learning.

L'aprenentatge automàtic, un tipus particular d'intel·ligència artificial, s'utilitza cada vegada més en els sistemes de software modern. A més de requerir molta energia, el seu ús augmenta les emissions de CO2, té un elevat cost financer i consumeix molts recursos. Encara que existeixen nombrosos estudis que analitzen les possibilitats del ML, molt pocs d'ells se centren en les seves característiques ecològiques, com l'ús d'energia o la seva petjada de carboni. Aquest estudi tracta d'augmentar la conscienciació dels costos associats al desplegament d'un sistema de ML en algunes de les plataformes d'allotjament en el núvol més avançades. Per a això, duem a terme un ampli estudi empíric per a avaluar i contrastar les emissions i la latencia de quatre proveïdors de núvol diferents, Amazon Web Services, Azure, Heroku i Railway Machine learning, a particular type of artificial intelligence, is increasingly being used in modern software systems. In addition to requiring a lot of energy, its use raises CO2 emissions, has a high financial cost, and consumes many resources. Even while there are numerous studies that look at the possibilities of ML, very few of them concentrate on its green features, like energy consumption or its carbon footprint. This bachelor thesis tries to increase understanding of the costs associated with deploying an ML system into some of the state-of-the-art cloud hosting platforms. In order to do this, we conduct an extensive empirical study to assess and contrast the emissions and latency of four different cloud providers: Amazon Web Services, Azure, Heroku, and Railway.

Treballs Finals de Grau d'Enginyeria Informàtica, Facultat de Matemàtiques, Universitat de Barcelona, Any: 2023, Director: Petia Radeva i Estefanía Talavera, [en] The field of machine learning is being applied to all aspects of our daily lives. From chatbots that speak like a human to artificial intelligence able to generate art, there are many neural networks capable of doing a better job than humans. In 2022, the Ego4D dataset was published, a set of large-scale first-person annotated videos as never seen before, which opened the door to new branches of research on video analysis. The publication of the dataset was accompanied by several challenges, in particular, one of them is the one we will face in this memory: the Moment Queries, which consists of the temporal localization of concrete actions in a video. This is a highly complex problem that needs very powerful image analysis techniques! Transformers are a type of neural network based on the concept of attention that revolutionized the field of Deep Learning. Since their appearance in 2017, they have proven their usefulness in various artificial intelligence applications such as natural text processing or computer vision, being able to outperform all previous results of Neural networks. In this work, we have made an in-depth study of Transformers and we have tested their performance in the field of image classification. With the knowledge obtained, we have analyzed how they can be applied to the Moment Queries problem. We have developed our proposal - a model that uses a pyramid of attention mechanisms to refine the data and provide the prediction modules with the best possible information. With this implementation, we have managed to locate and catalog 28% of the actions with a tIoU value better than 0.5.

Forecasting electricity consumption plays an important role in the planning and operation of the electricity market. The main problem is that prediction is a difficult and complex task, due to a large number of factors involved, including the non-linearity, non-seasonality, and volatility of the price over time. Electricity providers base their purchases and sales on forecasting and it is essential for them to have the best approximation of the real purchases of consumers since under-estimation will provoke a deficit and over-estimation will result in the penalty of the government, "Red Eléctrica". HolaLuz is one of the relevant electricity providers in Spain with almost 400,000 users dedicated to the commercialization of electrical energy of 100\% renewable origin and gas. The company classifies its consumers into predefined segments and uses a forecasting algorithm for the electricity consumption prediction. The algorithms which is Holaluz is currently using are: an XGBoost model and an Heuristic approach. The current implementations have their weaknesses which have a significant effect on its precision. The model is making a prediction based on hourly data such as previous "total\_consumptions", temporal features, weather data , and customer's information. The algorithm is executed every day to predict the clients' hourly consumption for the next 2 days. However, it takes about 7 to 10 days to get the total clients' consumption, "total\_consumption", which means that at the "execution\_time" only a fraction of the total information is available. An improvement of current HolaLuz's forecasting models of hourly electricity consumption is achieved by using advanced Deep Learning techniques. Several models are presented to improve the current models' performance. There are two Machine Learning models: Prophet and XGBoost and two Deep Learning models: DeepAR and Time Fusion Transformer. Temporal Fusion Transformer is the best model and outperforms current HolaLuz's models. It is using a dataset configuration which included features such as labor days, weather, CUPS metadata and time cyclic encodings. Finally the proposed model has been tested as well to predict on a forecast horizon of 14 days which still outperforms HolaLuz's models.