XoveTIC2022: Papers with Abstracts

Papers
Abstract. Schizophrenia is considered a complex syndrome, with multifactorial neurodevelopmental alterations. Empathy is a complex fundamental component of human emotional experience, which influences one's emotions and behavior. On the Electroencephalogram, an activation of the dorsolateral prefrontal cortex is characterized by a decrease in alpha activity. The main objective of this work is to understand if the immersive tools of Virtual Reality influence the electrical activity of the brain and the heart. The two immersive tools were able to increase empathy, mainly by altering prefrontal brain activity as well as heart rate.
Abstract. The present paper presents a set of experiences with the use of Xiaomi Mi Band (from version 1 to 5) in different populations but with the purpose of monitoring the daily life and the occupational performance of the participants. Monitoring with wearable sensors is an easy and non-intrusive approach to encourage preventive care by occupational therapists or other health providers to facilitate professional practice.
Abstract. Our project is aimed at the creation of SIGTRANS, a tool focused on addressing the need of efficiently storing and analyzing the vast amount of data related to the use of public transport networks. This is a highly relevant research topic given the changes urban mobility is experiencing, including but not limited to those motivated by climate change. We will provide transport authorities and operators with a system, combining the use of GIS technologies, compact data structures and advanced algorithms, to facilitate the exploration and exploitation of the available data. This data refers to both the offer in terms of infrastructure and mobility services available for the citizens, and the demand (that is, the use citizens are expecting out of these services). The analysis of this data will then serve as the foundation for further improvements to public transport services.
Abstract. Optical coherence tomography (OCT) is a non-invasive diagnostic technique that can image ocular structures. Recently, this imaging technique has been used to diagnose and monitor patients with multiple sclerosis (MS), as several clinical studies have linked the development of MS to various changes in the eye. Among the different structures, one of the relevant biomarkers for MS analysis is the choroid. Systems such as Enhanced Depth Imaging (EDI) provide detailed images of the choroid region. However, OCT images are not routinely captured using this technology unless the study is specifically focused on choroidal analysis. In this work we propose a robust approach, based on convolutional neural networks to segment the choroid in non-EDI OCT images. The results obtained show that the proposed network manages to delimit the inferior contour of the choroid in a similar way to the experts.
Abstract. The fear of public speaking is one of the most common social phobias causing anxiety problems in many people.
In this context, this project focuses on developing a tool capable of helping mental health professionals using virtual reality as a controlled environment in expository therapies, in this case applied to Glossophobia.
To enhance the creation of self-control mechanisms in these patients, the use of virtual reality is explored with the aid of Neuro and Biofeedback, allowing the visualization in real time of the physiological response to the stimuli of the virtual environment.
Abstract. An estimator of the probability of default (PD) in credit risk is proposed. It is based on a nonparametric conditional survival function estimator for mixture cure models. Asymp- totic properties of the proposed estimator are proved. A simulation study shows the per- formance of the nonparametric estimator compared with other semiparametric methods. A real data analysis illustrates the practical behaviour.
Abstract. This paper presents a discussion of alternatives for protein structure prediction, showing the advantages and problems of recent alternatives based on deep learning and approaches based on energy optimization of protein solutions. A SARS-CoV-2 protein is included to exemplify the results.
Abstract. As the world’s digital population grows, so does the reach and usage of social media: in 2021, 56% of the global population were social media users [1]. Social networks are now a part of our everyday life and continue to transform the way we interact with others on a global scale The downside is that negative behaviors in social interactions are also increasing their presence. For example, between March 1 and April 30, the OBERAXE (Spanish Observatory of Racism and Xenophobia) has detected a 27% increase in hate speech on social networks with respect to the previous two-month period [2]. In this paper we target the detection and classification of sexist content in social media texts. Two tasks are considered: (i) a binary classification task to decide whether a given text is sexist or not; and (ii) a multiclass classification task according to the type of sexism present in it.
Abstract. The multiple-input multiple-output (MIMO) communications and the intelligent re- flecting surfaces (IRSs) have been envisioned as key technologies for beyond 5G mobile networks. However, the computational complexity of conventional approaches to jointly optimize IRS-assisted MIMO communication systems constitutes a major limitation to their deployment. In this paper, we present an innovative contextual bandit (CB)-based approach for the optimization of the MIMO precoders and the IRS phase-shift matrix en- tries. The proposed optimization framework, termed as deep contextual bandit-oriented deep deterministic policy gradient (DCB-DDPG), considers a CB formulation with con- tinuous state and action spaces. The simulation results show that our proposal performs remarkably better than state-of-the-art heuristic methods in high-interference scenarios.
Abstract. The current BREEAM platform is a software tool that enables building assessors to register projects for which they want to obtain a sustainability certificate, and follow its certification process, which by an ordered exchange of documents between the assessors and the BREEAM technicians, the obtainment of the desired certificate.
One of the main documents needed for the obtainment of a BREEAM sustainability certificate that are currently sent forth and back between the assessors and the BREEAM technician is the evaluation form, which is the main focus of this project.
Abstract. This study aims to analyze the Strategy for Data Cybersecurity in the European Health Data Ecosystem, to be implemented in 2025. The document analysis was carried out to map the different proposals for a regulation of the European Parliament and of the Council, regarding the sharing of the Electronic Health Record in the European space, and the General Data Protection Regulation implemented in the European Union.
After an exhaustive documentary analysis, inconsistencies or flaws were detected that could compromise the rights of citizens enshrined in the General Data Protection Regulation.
Abstract. One of the main problems in mobile robotics is to estimate the global position in complex symmetrical environments. Even when there are many devices or algorithms to achieve that goal, not all of them are useful in all kind of environments. GPS is typically used outdoors whereas algorithms based on Monte Carlo localization (AMCL) are used in- doors. However, they present some disadvantages. Thus, the GPS commercial devices do not work inside the buildings and the AMCL algorithms are limited in symmetrical envi- ronments for the fact that they needs to detect remarkable differences in the environment. Due to the mentioned limitations we propose a global localization approach for symmet- rical indoor environments based on the structure of topological maps. Here, geometrical and semantic information of static objects are considered, respectively, from LIDAR and RGB-D camera. Both sensors provide us, respectively, the information about occupancy areas and the scene perception. The proposed system is divided into four tasks. The first one is the classification of nodes according to their geometrical nature based on the LIDAR signature. The second stage is focused on object detection through a pretrained CNN based on YOLO (You Only Look Once) as model of convolutional neural network, which is able to work in real time. The third task corresponds to tracking and pose es- timation of objects, where is necessary the information from YOLO and the depth data from camera. Finally, the last task consists of estimating the robot’s global pose on the map from the output of object detector, their relative distance and their estimation pose. This algorithm compares the structure of detected nodes and objects with the structure defined on a reference annotated map. In order to match the degree of similarity of both structures we define a evaluation function and the highest value estimates the edge where is located the robot in the topological map. Our main contributions with respect to our previous work are the addition of depth of detected objects and the improvement of the evaluation function.
Abstract. This work presents a web service that allows the entity’s professionals to create digital calendars for each of the people affected by some type of autism disorder and work with them through pictograms. With this tool, both therapists and families will be able to create calendars and complete each day of any month of the year with pictograms that describe the scheduled activities. These calendars will be completely adapted to the needs and preferences of the users to whom it is addressed. They will be able to create several calendars for different or even the same person, with all the data hosted in the cloud and accessible from anywhere via the Internet. In addition, this web service provides an integrated online search tool for pictograms, so you can search for all the pictograms that are needed to complete the calendar without leaving the platform.
Abstract. The interaction among different genes when expressing a particular phenotype is known as epistasis. High-order epistasis, when more than two loci are involved, is an active research area because it could be the cause of many complex traits. The most common abstraction for specifying an epistasis interaction is through a penetrance table, which captures the probability of expressing the studied phenotype given a particular genotype.
Although it is very common for simulators to use penetrance tables, most of them do not allow the user to generate them directly, or present limitations for high-order interac- tions and/or realistic prevalence and heritability values. In this work, we present PyToxo, a Python tool for generating penetrance tables from any-order epistasis models. PyToxo allows to work with more appropriate scenarios than other state-of-the-art tools. Addi- tionally, it also improves in terms of accuracy, speed and ease of use, being available as a library, through a CLI or through a cross-platform GUI.
Abstract. A Software Product Line (SPL) reuses software assets to implement products that share a significant set of their features. When a developer needs to generate a new product, the selection of features determines which components and source code are assembled together as the product. In recent years, the Database Laboratory has been working with SPL tech- nologies in the field of Geographic Information Systems (GIS). Our SPL creates products from a specification that allows, in addition to defining the data model of the application, to customize specific elements of the application such as maps and their associated layers. However, during its use with real projects we detected that this customization was insuf- ficient: since the selected features are included for the whole product, if we need a feature only for a specific element, we need to apply it to all the elements of the same type. In this paper we propose a solution that, using a Domain Specific Language, allows to associate features with specific elements of the generated application in order to achieve a greater customization of the generated GIS and to improve their quality. This way, it is possible to select a feature (e.g., clustering) for a specific element (e.g., map-viewer), thus limiting the functionalities of the application to those parts where they are really necessary.
Abstract. This paper presents an assistive robotic application that enables people with disabili- ties to improve their participation in daily living activities and mobility. Technically, we describe an assistive robot where we implement an online action detection model, which is integrated in an application software that is able to track the progress of users in terms of learning and reinforcement of four dissimilar everyday activities. In this work, we report a first technical validation based on four-week trials with four subjects. There were a variety of neurodevelopmental disorders present in the subjects who participated in the sessions. By considering this aspect, we are able to predict how the system will behave with real end users and the main results of this technical validation are presented. Overall, we can conclude that the artificial intelligence modules for action recognition have performed well, and that the designed user application allows to track the learning and reinforcement of daily life activities performed by the users.
Abstract. State Representation Learning (SRL) is a field in Robotics and Artificial Intelligence that studies how to encode the observations of an environment in a way that facilitates performing specific tasks. A common approach is using autoencoders and learning to reproduce the same state from a low-dimensional representation [1, 2, 3]. Although very task-independent, this method learns to encode features that may not be relevant to the task in which the encoding will be used. An alternative is to use some elements related to the goal to achieve and/or some knowledge about the environment and the problem [1] to produce an appropriate low-dimensional encoding that captures only the relevant knowledge. In this paper, we propose an approach to autonomously obtain latent spaces of the appropriate (low) dimension that permit an efficient representation of the sensorial inputs using information about the environment and the goal. To measure the performance of this methodology, we show the results of a series of simulations of robots performing a task consisting in catching a ball in different environments. In these cases, we have found that the models required for the prediction of the final position of the ball, taking as input the learned encoding, are much simpler than those that would be required using the sensing information directly.
Abstract. Currently, more than 400 million people are visually impaired due to different diseases. To diagnose many of these diseases, it is often fundamental to perform an exhaustive analysis of the retinal vasculature. Notwithstanding, such an analysis is rarely done in clinical practice, since it is arduous. This motivated the proposal of automatic methods. Some of these methods provide valuable results. However, their use is limited, as there are no graphical tools in which to integrate them. In this work, we propose a new web- based tool for the analysis of medical images using automatic methods. The tool already includes methods for analyzing the retinal vasculature. Also, it allows to dynamically add any method that complies with the API. All data is securely stored in the cloud for ubiquitous access. All these features, together with an intuitive interface, make the tool an effective solution for implementing automatic methods in daily clinical practice.
Abstract. The COVID-19 pandemics has shown relevant stochasticity in the spread of viral in- fection. Here, we formulate the dynamics of the epidemics spread as a stochastic SIR model, handling the compartmental model (of the SIR type) as a set of chemical reactions, and applying the Chemical Master Equation that describes the dynamics of the equivalent stochastic chemical reaction system. In this way, the solution (evolution of the probability distribution over time) can be obtained via the classical Stochastic Simulation Algorithm. The proposed methodology has been used to predict the COVID-19 evolution in small and medium size municipalities of Galicia.
Abstract. In several studies, the fear of public speaking acts like pathology. However, giving a lecture or presenting a project are daily tasks that stir the nerves, even the non-phobic person. The article intends to show the pedagogical vision about the influence of virtual environments’ aid in public speaking practice.
Two studies were conducted. A scoping review debated the use of the virtual reality programs for public speaking skills training and has proved, in this context, is still a recent procedure, and so lacks solid evidence. Second, exploratory research detected and compared nine virtual reality applications for public speaking practice.
Abstract. The use of Machine Learning (ML) techniques in the context of Cancer prognosis, di- agnosis and treatment is nowadays a reality. Some types of cancers could greatly benefit from specific techniques that are designed to work in a scarcity of data scenarios, or when obtaining labeled data is a time-consuming and/or costly task. It is the case of the Pan- creatic Adenocarcinoma. We present an experiment where Active Learning (AL) is used as the basis to create a model which performs a classification task where a human expert (in this experiment, a medical doctor) needs to determine whether a pancreatic cancer patient must be treated with chemotherapy, not treated, or he/she is unsure about the therapy. The use of AL techniques allows us to improve the accuracy of the model, and the inclusion of expert opinions may help us in the future to add explanatory capabilities to the system.
Abstract. The current project aims to tackle with Gaia’s BR/RP spectra distortion caused by interstellar dust, called reddening or extinction, which makes data not to be correctly classified. For such, it is proposed a machine learning algorithm that is able to learn how to correct such effect making use of denoising autoencoders. In addition, it was also developed a method to estimate the extinction degree, since for almost any spectra that is going to be corrected, such value was not computed. The previous tasks are going to be resourceful at our research group, since we take part at the Gaia project and deal with outlier spectra. In this way, we will be able to do a finer data-preprocessing prior to their classification.
Abstract. The Epiretinal Membrane (ERM) is an ocular pathology that causes visual distortion. In order to detect and treat the ERM, ophthalmologists visually inspect Optical Coherence Tomography (OCT) images.This is a costly and subjective process. In this work, we present three different fully automatic, end-to-end approaches that make use of multi-task learning to simultaneously screen for and segment ERM symptoms in OCT images. These approaches were implemented into three architectures that capitalise on the way the models share a single architecture for the two complementary tasks.
Abstract. This research analyzes and compares the application of different intelligent supervised classification techniques for detecting anomalies in power cells. For this purpose, a labeled dataset is obtained and generated in which samples of the different charge and discharge cycles of a Lithium Iron Phosphate - LiFePO4 (LFP) battery commonly used in electric vehicles are collected. The final classifiers present successful results.
Abstract. In this work, we introduce a framework that unifies existing implementations for the tasks of constituent and dependency parsing as sequence labeling problems. The system provides a way to encode both formalisms as sequences of one label per word, so they can be used with any existing general-purpose sequence labeling architecture. More particu- larly, we implement three linearizations to encode constituent trees and four linearizations for dependency trees. All encoding functions ensure completeness and injectivity. We will also train a sequence labeling neural system to learn such encodings, and compare their ef- fectiveness on standard constituent (PTB and SPMRL treebanks) and dependency parsing (a subset of treebanks from the UD collection) evaluation frameworks.
Abstract. This paper introduces a novel non-invasive wearable device that can infer whether people are suffering from anxiety or not. The device allows capturing physiological signals such as: electrodermal activity (EDA), skin temperature (ST), electrocardiography (ECG) and photoplethysmography (PPG), and provides an estimation of blood pressure -through the pulse transit time (PTT) technique- and breathing rate (BR) -by analyzing the heart rate variability-. The hardware also includes an SD card to store the signals for offline processing in laboratory tests.
Abstract. Wood is a material with many industrial applications. In order to make it profitable, it usually must be subjected to a prior drying. When dealing with an uncommon wooden topology, the drying time is difficult to predict and so a numerical simulation is the best option. In this work, we present a numerical method for this purpose, with the advantage of an adaptive mesh based on an a posteriori error indicator.
Abstract. Software-Defined Network (SDN) is an emerging architecture which objective is to re- duce the limitations of traditional IP networks by decoupling the network tasks performed on each device in certain planes by controlling and managing the whole network from a centralized location. However, this centralization also introduces new inefficiencies and vulnerabilities, such as those related to southbound and northbound controller interfaces, which often negatively affect security. In the past years, Machine Learning (ML) tech- niques have been implemented in SDN architectures to protect networks and solve security problems but sometimes it is difficult to obtain the right characteristics in real time. In this paper, we introduce a flow-based anomaly detection system in which the controller itself is in charge of receiving, analyzing and classifying the traffic by extracting a group of flow features.
Abstract. Nowadays, the security of companies’ assets and network infrastructure is very impor- tant for the proper development of their commercial activity. Large companies are capable of dealing with existing cyber threats, but small and medium-sized companies do not have the necessary budget. Existing security solutions are aimed at large companies, which can invest a lot of money in cybersecurity and have experts qualified of managing them. For this reason, we have created a network auditing tool capable of evaluating the security status of corporate networks and aimed at small and medium-sized companies. This allows us to obtain real time data through streaming techniques and, in addition, it is low cost, scalable, modular and easy to use, designed so that non-expert personnel can understand the security status of their company.
Abstract. COVID-19 is a disease whose gold standard diagnosis tool, RT-PCR, is unable to provide accurate quantification of its severity in a given patient. Currently, this assessment can be performed with the help of chest X-ray imaging visualization that, however, is a manual, tedious and time-consuming task. In this context, Computer-Aided Diagnosis (CAD) systems are very useful to facilitate the work of clinical specialists in these complex diagnostic tasks, especially in view of recent advances in deep learning techniques in the field of medical image analysis. Despite their great potential, deep learning strategies require a large amount of labelled data, which is often scarce in the context of COVID- 19 pandemic. To mitigate these problems, in this work we propose the use of a image translation paradigm, the Cycle-Consistent Adversarial Networks (CycleGAN) to generate a novel set of synthetic images with the aim to improve an automatic COVID-19 screening system using portable chest X-ray images.
Abstract. Cancellation tests are adequate to detect selective attention in children. These tests are usually performed using paper and pencil, which considerably reduces the capacity to register important parameters and requires some motor skills. In this paper, we present a mobile application that replicates the original Teddy Bear Cancellation Test and adds new features so that therapists can carry out broader studies.
Abstract. X-ray analysis of the lungs was the main method to assess the degree of affliction of SARS-COV-2. Due to the high contagiousness of this pathology, this assessment was conducted using portable X-ray devices. Automatic methodologies were proposed to compensate the image quality of said portable X-ray devices. However, these methodologies were shown to be exploiting external information (such as pacemakers or ventilators present in the images) to determine the severity. For this reason, we present a methodology specially designed to reduce the effect on an automatic methodology of these extraneous artifacts. We extract the lung region and we perform a screening of the presence of the pathology using only the pulmonary region. Finally, to ascertain the performance of the system (and provide explainability to the clinical experts), we generate the corresponding activation maps. The presented methodology has achieved a more than satisfactory performance in all the scenarios and the activation maps clearly indicate that the system is successfully using information from the lung region while excluding elements unrelated to the disease.
Abstract. In a world where technology is becoming more and more a part of our lives, the danger posed by cyberthreats to which we could be exposed is growing. Knowing how to pre- vent and act against them is crucial to avoid becoming the victim of cybercriminals. Good practices and general knowledge on how to act in cyberspace should be more widely dissem- inated to the population, especially to those who do not have sufficient technical knowledge of Information and Communications Technology (ICT). This work aims to provide a tool to raise awareness by providing them with concepts, good practices and practical scenarios to reflect on. For this, a tool was developed consisting of an application aimed at all types of users, and accessible on different devices so that it can be used by as many people as possible.
Abstract. The emergence of the Internet has generated new functionalities in our daily lives: data sharing, people communication or even the creation of a new “digital life” within the virtual world. Nowadays, each user visits many websites and web applications, in which the owner companies stores certain information from the user to be able to use the services offered. As a result, the user’s digital identity is scattered in many websites databases. During the last few years, some solutions have been developed, proposing a new digital identity solution that is known as Self-Sovereign Identity. This solution allows the decentralization of the user personal data and the self-sovereign identity management using blockchain technology. This paper shows a proof of concept of a simplified self-sovereign identity system that allows the user to manage their personal data with different external and compatible web services.
Abstract. Monitoring wireless traffic of Internet of Things (IoT) devices can be interesting for multiple reasons, such as carrying out security audits, assisting in debugging tasks or even checking the correct state of the inventory to detect obsolete or unused equipment.
There are currently several tools available on the market that undertake this sort of functionalities, allowing devices that use these protocols to be debugged or the traffic they emit to be analysed. Some examples of this type of solutions are Ubertooth or CC2531. However, these solutions require human intervention or connection to some other hardware such as a laptop.
In this project, we developed a tool that can analyse Wi-Fi, Bluetooth and ZigBee pro- tocols due to their increased use in IoT environments. Unlike most of the equipment avail- able on the market, the proposed tool can work with multiple protocols and autonomously, including its own storage system and energy supply.
Abstract. Several machine learning (ML) algorithms in combination with natural language pro- cessing (NLP) techniques have been used in recent years in a promising way for the auto- matic classification of software requirements. Nevertheless, several works have focused on the English language. Due to the lack of work in the Spanish language, we performed a con- trolled experiment using ML algorithms in combination with text vectorization techniques to investigate the best combination for Spanish requirements classification. Based on f1- score metrics, we found the combination of SVM with TF-IDF performs better than other combinations, with a value of 0.95 for functional and 0.79 for non-functional classification.
Abstract. Retinal imaging is widely used to diagnose and monitor eye-specific pathologies as well as some systemic diseases. Registering retinal images is crucial to compare the relevant structures within the eye. In this work, we propose a deep learning-based method to register color fundus images using domain-specific landmarks. We employ a deep neural network to detect bifurcations and crossovers of the retinal arterio-venous vessel tree. Then, these keypoints are directly matched using RANSAC. Our method was tested using the public FIRE dataset obtaining a registration score of 0.657, which is comparable to the best state of the art methods although our proposal is comparably much faster. Furthermore, our method improves the results of the state of the art deep learning methods.
Abstract. It is undeniable that computing has settled in the modern world. Many elements have enabled global digitization and products are being developed faster, largely through the use of third-party services. However, the systems developed, whether a proprietary or third party, can have degradation or service outages. Having an automatic system for the generation of incidents would allow acting as soon as possible on specific problems in the services and the development of a system that allows this automation is the main objective of this work.
Abstract. Many research organisations depend on Digital Libraries, Catalogues, or Archives to support their activities, especially in Digital Humanities. These organisations confront the challenge of obtaining adequate financing to develop the necessary software. The funding devoted to software development in the grants available to these research groups is truly insufficient to confront the entire job at once, so they must use several funding rounds to complete the necessary budget, further delaying the development of the library. However, when viewed through the lens of Software Engineering, Digital Libraries have characteristics that place them in the development paradigm whose goal is to automate the creation of code from analysis specifications: Software Product Lines (SPL). Therefore, with the goal of minimising the complexity and expense of developing Digital Libraries, we propose a SPL that allows their software to be generated automatically from the definition of its data model and features, considerably reducing the budget and time necessary for its production. As a result, Digital Humanities organizations may concentrate on their study rather than worrying about software development. During the development of the project, we have followed a methodology created by the authors of this paper and successfully tested also in other domains.
Abstract. This work introduces a new approach in time-series analysis field for automatic co- variates selection in dynamic regression models. Based on [1] and [2] previous study, a forward-selection method is proposed for adding new significant covariates from a given set. This algorithm has been implemented and optimized in R as a package, and it has been applied to multiple simulations to validate its performance. Finally, the obtained results from the IRAS database of Catalonia are presented to analyze the COVID-19 evolution.
Abstract. In the Database Laboratory of the University of A Corun ̃a, we have a software product line for the creation of Geographic Information Systems (GIS) on the web. In the case that the data is available in a data warehouse, the current functionality of the product line is not enough. Therefore, it is necessary to create a component in charge of connecting the data warehouse with the user interface providing the usual functionality (filters and aggregations in the different dimensions of the data warehouse).
Given that integrating the functionality and variability of the exploration of a data warehouse in a software product line is complicated, we have opted for a first approximation to define an exploration component in a GIS of any data warehouse. The component receives, utilizing a domain-specific language, a description of the data model (i.e. the facts and the hierarchy of dimensions, how each of them is related, and the way to retrieve their values) and makes the user interface components independent from the management of the communication with the data warehouse.
Abstract. The goal of this work is to solve a nonlinear parabolic PDE problem that arise in the financial world by means of the so called PINNs methodology. We propose a novel treat- ment of the boundary conditions that allows us to avoid, as far as possible, the heuristic choice of the weights for the contributions of the boundary addends of the loss function that come from the boundary conditions.
Abstract. The aim of this work has been the programming of a low-cost mobile robot that acts as a complementary tool in the therapeutic sessions of gait rehabilitation for people with physical and cognitive disabilities. The robot has an assistive function, being able to guide, accompany and motivate the person during the completion of a route previously designed by a therapist.
Abstract. In the last years, new visualization devices related to technologies like AR, MR and VR have emerged. Such devices have proven to be useful for numerous medical, industrial and manufacturing processes, but the technologies they use tend to be platform dependent. Currently, the Khronos group is making an effort to change this thanks to the OpenXR framework, which provides tools to standardize the development for different XR (Extended Reality) platforms. However, many processes and tasks require multiple users to interact with the same virtual environment simultaneously. To tackle this issue, this paper presents an XR solution that allows for the creation of collaborative applications that can be used at the same time by different platforms, such as computers (i.e., Windows, Mac or Linux PCs) and Mixed Reality smart glasses (e.g., Microsoft HoloLens 2). The proposed architecture is based on Unity and Mirror, a high-level networking tool for the mentioned game engine. The solution allows developers to design just one application that could be compiled and deployed to different platform devices without the need for changing any configuration or for adapting the project to each of the platform requisites. The proposed system also allows for using multiple devices simultaneously, providing a new way for collaborative interaction with the application, showing all the visual components synchronized in the same position and in the same state, thus facilitating communications and awareness of the environment for the developed XR experiences.
Abstract. In recent years Machine Learning (ML) strategies have proven to be useful to automate numerous classification and pattern detection tasks in diverse fields thanks to the increase of computational power in hardware. One of such fields is the Automatic Speech Recog- nition (ASR), which can use ML architectures to transcribe human speech into readable text. The Word Error Rate (WER) obtained with ML strategies can become relatively low while providing quick responses, reaching accuracy levels that approach human tran- scription accuracy. However, one of the main drawbacks in traditional architectures is the high demand of transcribed data to obtain a low WER in training. This kind of data is particularly hard to be achieved due to the high dependency on human processing. Luckily, a new framework proposed in 2020 (wav2vec2), considerably reduces the need for audio labelling thanks to the use of a Convolutional Neural Network (CNN) with self-supervised training on cross-lingual unlabelled audios of multiple languages and the ability to fine- tune the obtained results with labelled audios of a specific language. Thus, the framework can obtain results that outperform previous architectures by using much smaller audio datasets with transcriptions. This paper presents an ASR system based on wav2vec 2.0 that is fine-tuned for Galician, a language which currently only has small audio datasets available. Such a system is evaluated with a spontaneous speech dataset of approximately 1 hour from the Galicia Parliament, showing a relatively low WER (18.61%).
Abstract. Knowing the position of a device such as a smartphone accurately enough to control another device is a complex problem. Specific solutions (such as inertial measurement units) exist, but they are relatively expensive while do not provide accurate positioning over extended periods of time. In this project we explored methods based on the physical properties of sound to achieve quality positioning with affordable and relatively easy to set up equipment. Different options have been tested and we have chosen to use the Doppler effect to determine the speed of movement of the device and calculate its position in both 2D and 3D environments. Multiple tests have been performed demonstrating the advantages of the proposed method and pointing out the most important limitations. Finally, a demonstrator of the proposed technology has been developed to use a smartphone as a video game controller.
Abstract. The wristbands are very popular among the population due to factors such as their low cost, ease of use, designs and the feedback they provide to the consumer. These wearables track activity, heart rate and sleep-wake patterns. However, few studies have analysed the reliability of the data collected by these activity devices.
The validation of sleep data should be carried out by comparing it with Polysomnog- raphy (PSG), which is the standard test for measuring sleep parameters in the clinical setting. Thus, the aim of this project is the validation of the data quality of the Xiaomi Mi Band 5 wristband, when compared with the data by the hospital sleep unit devices during the performance of the polysomnography test.
In order to achieve this objective, an Epoch by Epoch (EBE) analysis will be performed to analyse how similar the results obtained by the two methods are. This analysis will use data from 45 people who underwent a PSG test and wore a Xiaomi Mi Band 5 bracelet for one night in a sleep unit of a hospital in A Corun ̃a. For this analysis, raw data from the PSG device and data from the Xiaomi Mi Band 5 wristband were used. In addition, different sleep variables were determined with the data extracted from both devices and following the guidelines of the American Academy of Sleep Medicine (AASM) manual.
Abstract. There are more and more IoT devices that need to be interconnected with each other to perform compute-intensive tasks due to their limitations in terms of storage, computing power and energy consumption. However, IoT devices encounter the problem of the lack of wireless connectivity in places where they are deployed or where they are traveling through. A solution to this problem consists in the use of opportunistic systems, which provide connectivity and processing resources efficiently by reducing remote communications to the cloud. Opportunistic networks are considered useful both in IoT scenarios where the cloud becomes saturated (e.g., due to an excessive amount of concurrent communications or to Denial-of-Service (DoS) attacks), as well as in those areas where wireless communications coverage is not available, such as it frequently occurs in rural areas or during natural disasters, wars or when other factors cause network outages. This paper presents the design of a novel opportunistic Edge Computing system based on the use of Bluetooth 5 and Single Board Computers (SBCs). To illustrate the performance and feasibility of the proposed system, latency tests are presented. For such latency tests, an experimental testbed was built by communicating two separate IoT networks (each network consisted of an IoT node and an opportunistic Edge Computing gateway). The tests calculated the time of message propagation from one end node to another. The obtained results show that the developed system obtains latencies between 850 and 1200ms, depending on the scenario, which make the solution viable for many application scenarios with low latency requirements.
Abstract. In this work, we build a novel numerical approach to investigate the behaviour of the coating flow in the process of hot-dip galvanization. The latter involves a strong interaction between a liquid film transported by a moving strip and a high speed gas jet. The solver combines a high-fidelity model for the gas jet with a simplified model for the liquid film. The high-fidelity model is a solver that resolves the gas jet flow in OpenFOAM, an open- source CFD software built from C++ libraries. The model for the liquid film is an in-house finite volume solver implemented in Python. The communication between both models is handled with the plug-in coupling software preCICE.
Abstract. Deep learning has demostrated its usefulness in reaching top-level performance. How- ever, inter-database generalization is still a broad of concern due to the aroused differences between local and external datasets’ performances. In this work we explore different deep learning model’s combination strategies applied to a multi-database case of study in the domain of sleep medicine. More specifically, three ensemble combination methods (namely max-voting, output averaging and weighted combination using the Nelder-Mead search) are analyzed in comparison to baseline methods (local models, database assembly approach) in a sleep staging inter-database generalization task.
Abstract. In this work, we analysed 11 imbalance scenarios with female and male COVID-19 patients present in different proportions for the sex analysis, and 6 scenarios where only one specific age range was used for training for the age factor. In each study, 3 different approaches for automatic COVID-19 screening were used: (I) Normal vs COVID-19, (II) Pneumonia vs COVID-19 and (III) Non-COVID-19 vs COVID-19.
The present study was validated using two representative public chest X-ray datasets, allowing a reliable analysis to support the clinical decision-making process.
The results for the sex-related analysis indicate this factor slightly affects the COVID- 19 deep learning-based systems, although the identified differences are not relevant enough to considerably worsen the system. Regarding the age-related analysis, this factor was observed to be influencing the system in a more consistent way than the sex factor, as it was present in all considered scenarios.
Abstract. We live in an era daily inundated with information, and the economy of attention makes us far from the truth. The present study has its core to study the creation, use and sharing of videos originated by artificial intelligence that can make it appear that a person says or does something, although he has never said or done anything of the kind. This content is called deepfake. The problem is the way this content is propagated, which for the untrained eye it can be seen as authentic. Quantitative research was carried out, through an inquiry and a literature review.
Abstract. Virtual Reality, Augmented Reality and Mixed Reality all have useful applications in the field of medicine, but Mixed reality has great potential because it allows for interaction with both real and digital Objects.
Based on a previous literature review, a preliminary Delphi study was performed to obtain the opinion of a panel of 22 experts from several hospitals on the use of Mixed Reality tools, such as the HoLoLens 2, in surgeries.
After data collection, a consensus letter was signed. According to experts, the most useful areas are medical education and surgical planning.