There is great interest in developing virtual and augmented reality applications for neuromotor rehabilitation treatments, with which patients can interact through body tracking devices. These applications provide the benefits of traditional rehabilitation without becoming overwhelming, as they allow the user to immerse themselves in various scenarios and situations. Furthermore, the way these applications interact makes it possible to retrieve and analyze the data generated by the user's movements during treatment. However, it is challenging to find body movement datasets, and those that exist do not include movements specific to the rehabilitation context. Thus, we present a framework that provides a standardized storage system for data from motor and neuromotor rehabilitation sessions, the ReBase. It was designed to allow the gathering of data generated by any application, regardless of the form of body tracking employed, and to support Virtual and Augmented Reality applications. From the data stored in the ReBase database, it is possible to create specific movement datasets over which machine learning or even data mining techniques can be applied. This paper presents several improvements to be implemented in the existing ReBase infrastructure and demonstrates an application capable of recognizing the Libras alphabet through machine learning developed using the framework CLibras.

Topic Modeling (TM) is a popular approach to extracting and organizing information from large amounts of textual data by discovering and representing semantic topics from documents. In this work, we investigate an important challenge in the TM context, namely Topic evaluation, responsible for driving the advances in the field and assessing the overall quality of the topic generation process. Traditional TM metrics capture the quality of topics by strictly evaluating the words that built the topics syntactically (i.e., NPMI, TF-IDF Coherence) or semantically (i.e., WEP). In here, we investigate whether we are approaching the limits of what the current evaluation metrics can assess regarding topic quality for TM. We performed a comprehensive experiment, considering three data collections widely used in automatic classification, for which each document's topic (class) is known (i.e., ACM, 20News and WOS). We contrast the quality of topics generated by four of the main TM techniques (i.e., LDA, NMF, CluWords and BerTopic) with the previous topic structure of each collection. Our results show that, despite the importance of the current metrics, they could not capture some important idiosyncratic aspects of TM, indicating the need to propose new metrics that consider, for example, the structure and organization of the documents that comprise the topics. In order to mitigate this limitation, we propose to adapt metrics commonly used to evaluate clustering algorithms since there are significant similarities between the TM and clustering strategies. Both have an unsupervised nature and the purpose of grouping similar elements. We evaluate three distinct metrics (Silhouette Score, Calinski-Harabasz and BetaCV) in the same previous scenarios and the results highlight the effectiveness of clustering metrics in distinguishing the results of MT algorithms and ground truth. However, this implies expanding the analysis space by including a new set of metrics. Therefore, we propose consolidating the various metrics, which consider both the quality of the words that make up the topics and the organizational structure of the documents, into a unified result, using Multiattribute Utility Theory (MAUT). Our results demonstrated that this approach allowed us to classify more precisely the different Topic Modeling, showing that the semantic advances generated by the use of word embeddings present in some MT strategies, as well as the solidity and consistency in the construction of topics through matrix factorization strategies.

Student dropout is the abandonment of studies by formally enrolled students. This phenomenon is influenced by different factors, such as loss of interest, financial difficulties, lack of investment, and appropriate policies, among others. The impacts of evasion reach different levels, from the student's interrupted training to the waste of resources and even society, which loses a professional future. Although the analysis of evasion is a complex task, the advancement of computational techniques such as artificial intelligence has enabled analysis and inference from data, aiming to understand and even predict this phenomenon. This work aims to predict dropout rates in the context of higher education at the Federal University of São João del-Rei (UFSJ), considering periods before and during the COVID-19 pandemic, through the application of machine learning techniques. Furthermore, evaluate the main variables related to the evasion phenomenon in both periods in search of differences. Data from students who graduated and dropped out between 2018 and 2021 from in-person undergraduate courses at UFSJ are included in the analysis and creation of prediction models. Sociodemographic variables, information related to the course, and student performance are selected as possible predictors. The initial data analysis made it possible to evaluate the quality of the available data and the frequency distribution of the variables and to carry out their preparation. After preparation, the database has 7.305 records and 23 variables in addition to the target class. The CRISP-DM methodology commanded understanding the business and data, preparing data, creating models, and evaluating. Based on similar works, the use of Decision Tree and Random Forest algorithms presented accuracy, precision, AUC, and F1-score rates close to or above 90%, revealing Random Forest as the holder of the best results. Analyzing the relevance of the variables in the results, attributes of academic performance and social/financial assistance were expected to be important and confirmed the expectation in both periods analyzed. Other attributes have already shown movements of gain or loss in the transition or even low significance between periods. In short, the results of this study reveal a promising performance of the model proposed in the analysis of student dropout at UFSJ. Furthermore, the insights gained about student characteristics provide a basis for developing preventive and student support strategies.

Cardiovascular diseases are the main causes of death in the world. Many of these dis-
eases require a deep and detailed understanding of electrophysiological changes for
to the study of new drugs and clinical devices to aid in the treatment. Consequently,
numerical simulations emerge as a relevant tool in the investigation of these electro-
physiological changes in heart disease. However, the complexity of applying mathe-
matical models of cardiac electrophysiology falls into systems of ordinary differential
equations (ODE) with a high number of unknowns, demanding great computational
effort. The work developed here consisted of the implementation of a distributed
memory parallelization and multi-GPU (Graphics Processing Unit) version of an
existing cardiac electrophysiology simulator with the aim of accelerate the solution
of models of this class. The monodomain cell membrane model and Bondarenko cell
dynamics model were adopted, together with adaptive time steps. The simulator
was run in a benchmark domain. Graphics cards of different performance were used
to carry out the tests. The combination of two lower performance graphics cards
allowed acceleration of more than 1.5 times for cases where the most refined mesh
was used. Load balancing between graphics cards of different performance provided
the best results. The results showed that the proposed implementation was able to
accelerate the solution of ODEs in different scenarios, proving to be an important
tool for numerical simulations of complex cardiac electrophysiology problems.

Access to technology as a fundamental human necessity; and sharing and distribution can be not only as a product, but also as thought and liberated code adaptable for each need. Access to source code is directly related to collaboration, considering that distributed and collaborative efforts depend directly on the sharing of source code. This work addresses collaboration mediated by technology based on open-source software, discussing traditional art and digital art; the values and freedoms of open-source software; and the sharing of software artifacts at different levels of granularity. A descriptive-analytical study is conducted in the light of the foundations of Computer-Supported Cooperative Work (CSCW) and Groupware and Low-Code platforms. The study yields requirements and features of Groupware that support collaboration in creative processes present in software development, and it conducts a case study of the implementation of the defined features, with the tool Mosaicode as the object of the study. This tool is a visual programming environment, source code generator, and open-source software, with a focus on generating software applications for the domain of digital arts. It was developed at the Federal University of São João del-Rei (UFSJ), in the ALICE laboratory — Arts Lab in Interfaces, Computers, and Everything Else. The work contributes to understanding the possibilities and challenges of collaborative development in this context, enabling the way for a more inclusive, cooperative, and open environment, allowing digital artists and non-programmers (citizen developers) to explore creativity and collaboration in new technological dimensions.

School dropout is a phenomenon characterized by being influenced by several variables, which makes the study to identify which factors contribute to the dropout of a student from their academic institution complex. In the last decade there has been a considerable expansion in the offer of higher education courses in Federal Education Institutions, especially due to public policies that have fostered improvements in the physical infrastructure and personnel of educational units, allowing individuals with the most varied profiles to start their studies and do make the task of understanding school dropout more complex for managers. Parallel to this scenario, the Machine Learning area also expanded its application possibilities to the most diverse areas, including education, providing different ways of analyzing and understanding the data that are generated in the environment of each institution/organization. This Dissertation aimed to use Machine Learning techniques to predict the risk of school dropout in undergraduate courses at the Federal Institute of Science, Education and Technology of Minas Gerais (IFMG), as well as to identify which attributes are most associated with this phenomenon in institution. The structuring and organization of the activities foreseen in this work was supported by the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology. Three phases of experiments were conducted, the first dealing with the balancing of the dataset, the second using Feature Selection techniques and the third applying a semi-supervised learning strategy to improve the performance metrics collected. As a main result, we obtained a model capable of classifying dropout with 90% accuracy and 86% F1, indicating a considerable possibility of complementing institutional action with regard to actions aimed at controlling dropout levels at the IFMG.

Nowadays, most e-commerce and entertainment services have adopted interactive Recommender Systems (RS) to guide the entire journey of users into the system. This task has been addressed as a Multi-Armed Bandit problem where systems must continuously learn and recommend at each iteration. However, despite the recent advances, there is still a lack of consensus on the  best practices to evaluate such bandit solutions. Several variables might affect the evaluation process, but most of the works have  only been concerned about the accuracy of each method. Thus, this work proposes an interactive RS framework named iRec. It covers the whole experimentation process by following the main RS guidelines. The iRec provides three modules to prepare the dataset, create new recommendation agents, and simulate the interactive scenario. Moreover, it also contains several state-of-the-art algorithms, a hyperparameter tuning module, distinct evaluation metrics, different ways of visualizing the results, and statistical validation.

A key challenge in education is devising novel strategies to motivate and engage students throughout the learning process. Gamification has been used extensively in variety of settings to support the learning process and keep students engaged. Essentially, gamification is centered on employing game-based mechanics and gamelike features to non-game problems. Recently, researchers have started looking into how gamification can be explored to improve the performance of students while trying to master challenging topics. We set out to evaluate the extent to which gamification can be a helpful in teaching graph-based software testing criteria. To this end, we designed and implemented Gamifying Graph Coverage Criteria (GGCC), which is a tool that presents instructional information on graph-based criteria in a gamified fashion. We also carried out an experiment involving 20 participants to examine the effectiveness of gamification in the context of teaching graph-based software testing criteria. We found that participants exposed to the content through GGCC and its gamelike quizzes performed better than participants exposed to the content using traditional classroom and pen-and-paper-based exams. A later survey confirmed this positive attitude from the participants towards GGCC.

Reinforcement learning algorithms allow agents to learn from experience, without the need for prior knowledge. For this reason, they have been widely used and the use of low and medium complexity digital games as benchmark environments has become a common practice. In 2013, a new algorithm, called DQN (Deep Q Network), caused a great impact in the academic environment by obtaining human-level results in several Atari 2600 games, using artificial neural networks. Consequently, new lines of research emerged and new derived algorithms were proposed. Among these, the FQF (Fully Parameterized Quantile Function) stands out, an algorithm that has become the state of the art among the non-distributional algorithms in the Atari 2600 domain. However, the FQF has not yet achieved results obtained by a human expert in all evaluated games. Considering the ability of artificial intelligence to detect patterns imperceptible by humans, this led us to believe that better results than the current ones could still be obtained. Therefore, in this work, a search for related works was carried out and three improvements that brought success in algorithms proposed before the FQF were chosen to be combined and evaluated together with the FQF, thus seeking to improve the algorithm. The improvements applied to the FQF are: the use of three steps in temporal difference, the application of the Munchausen approach and the use of prioritized experience replay. The combination of the three improvements made it possible to analyze eight algorithms, which were evaluated in five MinAtar games. According to the analyzed metrics, the version of the FQF that makes use of the three improvements was better than the original FQF in all experiments carried out, thus making a more promising version of the algorithm available to the scientific community.

provisório

The interest in the dynamics and understanding of the characteristics of infectious diseases spread dates back more than two centuries. Several perspectives can be considered in these studies, where there is a special appreciation for immunization. The assertive choice of groups of individuals to be immunized can directly impact the dynamics of contamination, both in protecting parts of the population and in minimizing the speed of contagion. Such a task is important and challenging in epidemic scenarios with reduced number of doses available, as it involves numerous combinations of choices.

The objective of this work is to propose an improvement in a methodology for choosing individuals to be immunized based on a genetic algorithm, which uses the SIR epidemiological model to evaluate the different sets proposed. Its effectiveness will be verified through a comparative study with criteria that use metrics of centrality in networks and a random choice of individuals. The results obtained by experiments that simulate epidemics with vaccination by each of the criteria will be evaluated from the perspectives that consider the number of individuals affected, the largest number of infected at the same time and the estimated contamination rate. The results obtained indicate that the systematic choice of individuals is a correct decision for this type of problem and that the results obtained by the optimization methodology are equivalent to the results delivered by the main centrality metrics used in complex networks.

The National Immunization Program (PNI), created in 1973 by the Brazilian Ministry of Health, is responsible for regulating the vaccination room, including operational procedures regarding its use. The nurse is the professional responsible for supervising the vaccination room and the team's permanent education process, being necessary to carry out periodic training. However, training usually does not occur systematically, being focused mainly on updating the vaccination schedule. In this work, we propose using innovative approaches, such as virtual worlds, digital games, and active methodologies to contribute to an environment for vaccination training. The research is outlined according to the Design Science Research (DSR) methodology, with the guidance of a multidisciplinary team. A three-dimensional virtual environment was modeled and developed as a desktop application using the Unity game engine to support a virtual vaccination room training. A system for the creation and education of scenarios for the simulation was also modeled and implemented.  Two evaluations were performed, one with expert judges to validate the artifact built through the Delphi method, and another with apprentices, to validate the usability and acceptance of technology through the SUS (System Usability Scale) and UTAUT2 (Unified Theory of Acceptance and Use of Technology) methods. The results showed that the proposed simulation was well accepted and had good usability according to the SUS method (81.38 scores). The analysis promoted by UTAUT2 also allowed us to identify that performance expectancy was the factor that most positively influenced apprentices' behavioral intentions. It is also noteworthy that both audiences that participated in the evaluations emphasized the relevance of the proposed simulation and pointed out improvements concerning the representation of some items in the virtual room. Finally, we conclude that the proposed simulation is a viable and desirable alternative for teaching vaccination.

As redes sociais refletem os relacionamentos e interações entre indivíduos e têm desempenhado um papel muito importante na difusão de informação, em que a comunicação de ideias e compartilhamento de opiniões acontecem a todo momento. São diversos os exemplos de como as redes sociais podem afetar o comportamento dos indivíduos, tais como o marketing viral, a difusão de memes e a propagação de fake news. Essa dinâmica de difusão de informação tem motivado o estudo de diversas abordagens para identificar os principais influentes em uma rede. 
O problema de Maximização de Influência consiste em encontrar um subconjunto S, chamado de conjunto de sementes, de no máximo k elementos, de modo que a propagação máxima (esperada) seja alcançada por meio de um modelo de difusão, tendo S como os influenciadores iniciais em uma rede. Já foi demonstrado que o o problema de maximização de influência é um problema de otimização NP-difícil, portanto, devido à sua complexidade, é inviável encontrar o subconjunto S que garanta a difusão mais abrangente. 
A abordagem mais comum para este problema é a utilização de algoritmos aproximados, destacando-se o Cost-Effective Lazy Forward (CELF), cerca de 700 vezes mais rápido do que a estratégia gulosa proposta por Kemp et al., e o CELF++, que apresenta um ganho em tempo de execução entre 35 a 55% sobre o CELF.
Este trabalho apresenta uma modificação dos dois algoritmos estado-da-arte supracitados, CELF e CELF++, substituindo as simulações de Monte Carlo por estimativas de difusão (metamodelos), para a seleção de conjunto de sementes. Buscou-se sintetizar e, com um foco especial, mostrar que: (1) os metamodelos podem ser usados para estimar qualitativamente a difusão de conjuntos de sementes; (2) o uso de métodos já conhecidos na literatura em conjunto com metamodelos é capaz de identificar, ordens de grandeza mais rápido, indivíduos mais influentes e, em alguns casos, até superar o resultado desses métodos em propagação.

Online social networks, such as WhatsApp, Twitter and Facebook are used daily by thousands of people as a means of entertainment, communication, access to information and claims. In this environment, anyone has the power to generate and propagate news, resulting in new opportunities and dilemmas, such as the misinformation epidemic. Within this ecosystem, bots and strategies like Reflexive Control (RC) can be used by different agents in order to confuse, manipulate and distort public opinion on matters of interest. Therefore, approaches that help to characterize this environment with a focus on the user become essential. To this end, we present SARA (Automated System with Complex Networks and Analytics), an approach that allows the characterization of large-scale events centered on users in online social networks, especially on Twitter. Combining complex networks, text mining, machine learning with an communities approach to SARA, allows the identification of automated accounts (bots), mapping of antagonistic views on a given subject, extraction of topics of interest, identification of propagation patterns information and user interaction in a semi-automated way. The present work also presents its application in the analysis of three real issues in Brazil. About 3 million tweets about STF, anxiety and vaccination were analyzed.

The Internet is a network of computers that interconnect hundreds of millions of devices around the world. Not too long ago, these devices were mainly desktop computers, Linux workstations and so-called servers that store and transmit information, but nowadays, more and more modern Internet end systems like TVs, game consoles, cell phones, webcams, automobiles, environmental detection devices, frames and internal electrical and security systems are all part of the network. Following this shift that is taking place on the Internet (and also on private computer networks), the term “Internet of Things” (IoT – Internet of Things) is becoming more and more pervasive. This work aims to implement a working prototype for the Internet of Things using Arduino and LoRaWAN communication technology. Analyzes were carried out indoors and outdoors.

The Database Knowledge Discovery (KDD) process applied to structured data is a challenge where several approaches have achieved relative success for some time. Many of these approaches are based on traditional Machine Learning and its paradigms. The processing of unstructured data, on the other hand, had its popularity only recently, with the emergence of Deep Learning, an approach based on Artificial Neural Networks (ANNs). Although ANNs are not a method new, the evolution of computational power in an accessible way, especially through GPUs, together with the emergence of increasingly efficient optimization techniques, boosted its applications in an extraordinary way in domains such as image processing, audio, natural language, among others. This work seeks to investigate how these new techniques can be applied in order to improve the results in models generated on structured data. From this investigation, OlapNet's proposal arose, a Convolutional Neural Network (CNN) architecture that is based on implicit data cubes as input. Formally it was identified that this architecture is capable of surpassing the results of a specific group of transformations carried out on the data, allowing in part the automation of the data transformation stage in the KDD process. In order to verify the proposal empirically, a sample of data from a real database was used containing anonymized data on the historical indebtedness of clients of a financial institution. Based on this data, a predictive classification problem was modeled to estimate the likelihood that any customer will contract new credits in the next three months. Thus, traditional learning methods were used Machine and CNN variations, including the proposal of this work. The results showed that in almost all cases the CNNs outperform traditional methods, indicating that the Feature Maps generated from the convolutional kernels learned by the network are capable of extracting relevant characteristics. These kernels not only allow you to extract characteristics, they also reduce the complexity of the network by delimiting a neighborhood for each pixel and decrease the propensity to occur overfitting. Among the CNNs tested, OlapNet outperforms all other methods, indicating that the proposed architecture is very promising.

The design and development of autonomous software systems has been extensively investigated mainly based on the manifest disseminated by IBM concomitant with the advent of Autonomic Computing. Autonomous (or autonomous) software systems are those capable of adapting existing behaviors to the demand for new requirements and of resolving incidents with minimal human intervention. However, investigations on how to incorporate the capabilities of adaptation and autonomous incident resolution in existing (legacy) systems, although desired, little has been investigated in recent years. Allowing legacy software systems to adapt requires the identification and incorporation of actuators responsible for adapting the behavior and sensors that measure the variation in the state of the monitored behaviors. Ideally, the incorporation of actuators and sensors should be carried out in a flexible way, that is, without the software system having to be designed and implemented again. To this end, in this article, we propose a new approach for autonomous incident management in legacy software systems. Our approach is based on two fundamental concepts: (i) transversal composition of the sensors with the code exists through invasive composition techniques; and (ii) adapting behavior through virtualization technologies. With this, it is possible to incorporate new autonomous behaviors in legacy systems quickly and flexibly. We also present an assessment of the applicability of our approach in Moodle, a legacy system widely used in practice for conducting distance courses.

Music plays a fundamental and cultural role in influencing society and individuals, in which the relationship between music and sounds inspires the way listeners connect and deal in a musical environment, which is always changing due to its richness in orchestrating musical training models and procedures in environments offered by traditional music theory.

This includes low-latency communication infrastructures and protocols, embedded and specialized audio IoT hardware, APIs (dedicated application programming interfaces), software based on ontological and semantic audio processes and the design of new dedicated devices that provide interaction between users and musical devices and things, in addition to offering resources for the production or consumption of musical content.

In this work, after analyzing the concepts and surveys of the musical Internet of Things, listing some difficulties found in connecting and developing Internet of Things music solutions, we seek to present a protocol and scenarios based on the need in which the user can connect from his device in a practical and interactive way to the components and musical objects, capturing and sending information, making it possible to place more animation and musical and sound effects in the environment according to the musical rhythms.

Nowadays, Recommender Systems (RSs) have been used to help users tobdiscover relevant Points Of Interest (POI) in Location Based Social Network (LBSN),bsuch as Yelp and FourSquare. In face to the main data sparity challenges and different sources of information, such as geographical influence, in this scenario, most of works about POI recommendations has only focused on improving the system’s accuracy. However, there is a consensus that just it is not enough to assess the practical effectiveness. In real scenarios, categorical and geographic diversities have been identified as key dimensions of user satisfaction and recommendation utility. The few existing works are concentrated on just one of these concepts, singly. In this work, we propose a novel postprocessing strategy to combine these concepts in order to improve the usefulness of recommended POIs. In this context, postprocessing involves reordering a POI list from a Base Recommender (RB) to maximize categorical and geographic diversities among the first items on the list. Our experimental results in Yelp datasets show that our strategy can improve the usefulness of the list of recommended user POIs, considering different RSs and multiple diversification metrics. Our method is able to improve the diversity up to 120% without major accuracy losses.

The modernization of computational platforms has led to the emergence of new features in information systems that are little explored, and when they are, for the most part, they do not present standardization regarding their use. In this scenario, one of the points to be explored is the sharing of resources created by the most diverse types of computer system users. However, it is possible to verify that the sharing of resources between end users of computer systems usually does not have a standardization to control their packaging, distribution and versioning. Considering the particularities involved in each information system, it is necessary to define an abstraction layer that allows this sharing to occur transparently between users of the most diverse systems. With this, this work brings UAISharing - a Universal Interface for Access to Shared Resources, a proposal that aims to present a model of packaging and distribution of resources produced by end users.