Summary: In different scenarios, the architectural style in microservices has gained notoriety as an architectural reference to be followed in modern applications, based on computational cloud, which seek scalability as a fundamental requirement. Although this architecture promotes a series of benefits, it also poses challenges such as managing its evolution and operational complexity. In this context, the use of frameworks based on messaging systems (such as Kafka, for example) has become a standard in the industry, being a reliable and efficient method for establishing asynchronous dependencies between services. Thus, in a distributed architecture such as microservices, an increase in complexity is expected and, consequently, to lose reference to how the system components communicate with each other, causing services with many dependencies. In this work, we aim to explore the issue of dependency management, developing a conceptual approach and a practical implementation of this to map asynchronous dependencies via Kafka in the case of the SiteWhere project

After analyzing the repository (source code) of the project, we were able to develop an algorithm, in addition to carrying out its implementation and application in the project architecture and, finally, we compared the results obtained with the relations obtained through manual inspection of the project and from analysis of your documentation. The algorithm was able to correctly map 12 of the 16 relationships provided for by manual inspection and documentation. Also presented are limitations and opportunities for improving the algorithm, as well as an analysis of the SiteWhere coupling itself from the results obtained.

Summary: Ensuring high availability and high reliability of the network infrastructure is extremely important for several companies, especially for those whose business is dependent on information systems. More than that, having a quick response in the event of a drop in equipment connectivity is crucial to maintaining continuity of operation. In this scenario, this work proposes to simulate a network of telecommunications equipment, introduce connectivity failures in some equipment, evaluate the impact of these failures on the other equipment of that network and, based on these impacts and the network topology, perform the analysis of root cause.

The root cause analysis proposes to establish, in decreasing order of probability, the equipment causing the loss of network connectivity in situations in which several equipment present this symptom.

Summary: With the entry into force of the General Data Protection Law (LGPD) in Brazil, several organizations began a process of readjusting their information systems. In this context, systems that manipulate patient health data contain extremely sensitive information. This project aims to improve the DuploJ Controle health mobile application based on the suitability of the system to the LGPD, while expanding its functionality to support other use cases. For that, non-essential data fields were removed and an authentication mechanism was implemented, both via external API and biometrics. The final product also allows adding and viewing a catheter of a different type from DuploJ.

Summary: Fuzz testing is a testing method that generates and inserts random entries at a very high speed into the target program in order to generate and document crashes. Its high volume of inputs and its automated approach have the consequence of finding bugs that escape the programmer and tester.

The aim of this work was to analyze the use of fuzzers and their application in four case studies written in C language, finding two bugs after campaigns of up to 24 hours with the AFL and honggfuzz tools.

Summary: Congestion control algorithms are a crucial part of the performance of the TCP protocol. These algorithms aim to balance and increase the sending of packages in the face of adverse conditions, so that the exchange of packages is always as close as possible to the optimal scenario. Over time, several different solutions to this problem have been proposed, which have been analyzed under different metrification approaches. This report aims to evaluate two variants of CCA (Congestion Control Algorithm) known as BBR and CUBIC, from the point of view of how they are impacted by independent competing traffic and also how they influence competing performance.

Abstract: Congestion control algorithms (CCAs) are crucial for TCP performance. Those algorithms are responsible for balancing the exchange of packets according to link conditions as to use it as efficiently as possible. Many CCAs have been proposed and this work evaluates the performance of two of them (BBR and CUBIC) according to different metrics when competing with each other and sharing the link with unrelated background traffic.

Summary: Numerous surveys are based on qualitative data analysis, a process that consists of analyzing non-numerical data sources and extracting information from them by listing the concepts identified in them. Qualitative analysis techniques can be simplified when using digital tools, by the ease of managing the sources and information extracted and better data visualization, but an effective, accessible, free, and reliable tool is not yet available.

Thus, the objective of this project is to develop an integral part of an open source platform for qualitative data analysis, a graphic and visual analysis module of qualitative data.

The methods used to develop this work began with the understanding of the main methods of qualitative analysis, followed by the collection of data and requirements through interviews with potential users and inspiration on existing platforms. Finally, an iterative development was adopted with feedback collection for each functionality developed.

At the end of the development cycle, it was found that the objectives of functionality and reliability are achievable, considering that the implemented features support not only the base method used as a reference in this project, the Grounded Theory (Grounded Theory), but also several others, given the agnostic development in relation to the method used in the analysis.

Summary: The analysis of qualitative data is an essential step for research involving mostly descriptive data, and can be supported by a range of technological solutions available on the market. However, stable and efficient products are scarce and, when they exist, have impeding costs for researchers and students. In this project, functionalities were developed for a new open-source web application for qualitative data analysis that include the procedures performed during the open coding phase of a research.

Summary: The Unicamp campus is a diverse place, where different experiences can be lived. Each person who passes through the campus makes their own path, and has their own experiences.

Based on this theme, the Mapa Afetivo application was developed in order to allow people to record their paths, their routes and, mainly, describe the experiences lived during their journeys. Other people can, thus, discover new paths and experience feelings never before experienced.

During the requirements gathering of the application, the decision was made that the best way to put the idea into practice would be to develop an Android application. Naturally, an effective application needs an efficient architecture to support it, guaranteeing its good performance and providing an adequate development and maintenance process. This work deals with the proposal and development of the architecture of the developed application, which was based on the MVP (Model-View-Presenter) architectural style and adopted principles of Clean Architecture. Based on this, a hybrid architecture was developed in order to support the application development, considering an evolutionary scenario. The architecture in question was created in such a way that it is not tightly coupled to the concepts and functionalities of the Affective Map - its characteristics and structures can be used in other software projects, academic or commercial.

Summary: The walk can be seen as something beyond the displacement between one place and another, it can involve affective and creative experiences in a person's life. Sharing these experiences would help the Unicamp community to get to know its space better and connect people from completely different contexts. Observing this, this project proposes the creation of an application capable of recording routes using GPS with the addition of route points with images and media so that the creator can express his experiences during the journey.

The application was developed for the Android platform with the basic usage requirements for the Unicamp community, which in the future can be expanded to other regions as well. After implementation, some future improvements for the application were noted.

Summary: The rapid spread of the COVID-19 pandemic, in 2020, made early and accurate diagnosis of the disease essential, as a way to control the spread and mitigate sequelae in the population. Therefore, a methodology and a tool applying fundamentals of image processing, computer vision and machine learning were proposed and implemented in order to perform the detection of COVID-19 by means of lung images, quickly and accurately. Thus, a study was carried out on neural network architectures for segmentation of biomedical images, among which were adopted U-Net, U-Net ++ and MultiResUNet to solve the problem in question, in addition to the convolutional neural network Inception- ResNet-v2. Although U-Net ++ showed an accuracy of 84,34%, the segmentation architecture that had the best effectiveness was MultiResUNet, which, with a 99,10% accuracy, surpassed six of the seven architectures experienced by the creators of the database SARS-CoV-2 CT-scan []. However, the best result was obtained by Inception-ResNet-v2, which surpassed the segmentation networks with an accuracy of 99,40%.

Summary: This work aims to study the systems of OMR (optical musical recognition or optical musical recognition in English) and the development of a score reader in digital images. These systems seek to convert an image containing a score into a musical file format that can be easily read by a computer. In this project, experiments were carried out with some methods frequently used in OMR systems and a system was also developed capable of converting images containing simple monophonic scores to a MIDI file, using some of the existing methods in the literature. The project is available in a public repository.

Abstract: In Linked Data, data sets should include links to others by interlinking new information with existing resources. This must improve the “discoverability” of new and existing data in LOD (Linked Open Data) cloud. However, this is somewhat overlooked because finding and maintaining links between different data sets are a costly and time consuming task. In this work, we designed and implemented an interactive system to support quality inspection of links. This must help and encourage LOD players to find and maintain links. In our approach, evaluating a link is based on two graphical representations: a radar chart and a network graph. The radar chart is used to display similarity values ​​between the subject and the object of the link; and the network graph is used to display the link's subject and object. We present to which extent our proposed interaction mechanisms help users to evaluate the correctness of a link, which is an essential task for preserving data quality in LOD data sets.

Summary: Currently, there is an increasing number of sales on e-commerces. Customers often ask questions about a product before purchasing it. Online stores can improve the user experience, satisfaction and also the sales rate by answering these questions instantly. Many questions are particularly difficult to be answered automatically because they demand specific knowledge. More specifically, there are a lot of questions about product compatibility in the E-commerce. This work aims to study how to model compatibility relationships about products in the construction of a Knowledge Graph (GC) structured based on ontologies. To that end, we need to determine an ontology that models the concepts involved in the domain and the relationships between them. In particular, we approach possible modeling and compare two different ontologies, showing examples of their use. The final version of the ontology is explored in a GC built in the company gobots¹. GC is a solution that helps the company to automatically answer questions in natural language on e-commerces.

Summary: With the expansion and popularization of the Internet of Things concept, Fog Computing's approach starts to stand out as a technological solution to meet this new demand. However, the mobility of devices creates difficulties for the development of this technology, mainly related to migration decisions and policies. Given this scenario, the use of MobFogSim, a mobility simulator in Fog Computing, can be of extreme help in choosing the migration configuration. The objective of this work is to explore the use of this simulator, with different input parameters, to evaluate how they influence the results of the simulation.

Summary: This work aims to test latency time, that is, the delay between sending a message and receiving a response to it, and its influence on the user experience in the case of Cloud Gaming systems, that is, systems that allow a user to remotely interact with a game running on a cloud service. The gaming industry has a lot to benefit from advances in cloud computing technology, however, large-scale Cloud Gaming services are not yet viable. The technology itself exists and is functional, however, the negative effect of latency on the user experience makes it impractical for extensive use like other popular streaming services, and there is still no concrete and logistically feasible solution to this problem, since that it is impossible to have servers close enough to all users on a global scale system.

Summary: This project aims to develop an infrastructure that makes it possible to perform a benchmark of resource utilization of the infrastructure provider in the cloud of Amazon Web Services, in addition to evaluating certain scenarios of instances with varying computational power. Through a container architecture, we create all the infrastructure available for testing using a generic application that allows you to perform operations that cause a load on CPU memory resources in the most diverse types of instances, in addition to collecting all the necessary metrics and make them available in a data visualization application. We were able to observe some aspects of their behaviors, advantages and disadvantages of a more vertical or horizontal solution, from the point of view of scalability, and we discussed the development process.

Summary: A study was carried out to define what microservices are, to know when to apply, to know their definitions and their negative points. And with a negative point related to the communication time between microservices, analyzes were made to be able to create microservices that have adequate response time and security and authentication according to the system requirements. Some analyzes of time, stress and performance were made to see what are the best settings to be able to serve a customer according to their requirements.

Summary: The objective of this project is to compare the performance and scalability of two web services that use languages ​​run on the Java Virtual Machine: Java and Kotlin. We will focus on a situation in which the two systems will perform a parallel execution algorithm when receiving an HTTP request. The system made in Java will use threads, already the system in Kotlin Coroutines. To perform the comparison, we initially performed a theoretical analysis to understand how applications should behave in high-risk situations. throughput. We concluded that the service that uses Coroutines should present better results due to the fact that its parallel execution algorithm consumes less memory and processing. Then we put the two systems under load tests using several different configurations, and compare the results to assess the behavior in practice. The result was that the Kotlin application performed substantially better, proving what we had understood in the theoretical analysis.

Summary: In the last year, COMVEST (Unicamp's Permanent Commission for Vestibular) introduced facial recognition to Unicamp's Vestibular candidates to increase the reliability of the selection process. The capture of the images of the faces of each individual was made through the use of the camera of mobile devices. For this purpose, such devices were handled by human inspectors during the application of the exam, which accessed the camera through third-party software, developed by a partner company. As the partner software does not meet all requirements imposed by the commission, the focus of the project was to develop a new application, called Smart Face, to replace it permanently. The Smart Face was developed during the entire project and, currently, it is ready for use, meeting all the imposed requirements, besides allowing to meet future requirements that may arise, through the installation of updates.

Summary: This work aims to carry out studies in offloading in mobile applications. Offloading consists of transferring intensive computations to another processor, either because of device limitations or to improve performance, battery consumption or memory. To carry out the studies, an Android application was implemented that computes a filter on an image using CPU and GPU. This computation was performed locally and on Google Cloud servers. From this, experiments can be carried out to analyze the cases in which offloading would be advantageous or disadvantageous when comparing latency, CPU and battery consumption.

Summary: In robotics and computer vision, Visual Odometry (VO) is the process of determining the position and orientation of a robot from the analysis of images obtained by its cameras. Their techniques have been widely used in applications related to robotics, a famous example being the robotic vehicles used by the North American Space Agency (NASA) in the initial exploration of the territory of Mars. Classical techniques to solve VO problems usually involve the kinematics of robots, so it is common to fail in very complicated situations, such as when there are many changes in direction or the environment is very challenging, for example, poor lighting. For this reason, new methods of VO have been based on bringing new techniques, such as neural networks or attentional models. In this work, we seek to evaluate these new models of visual odometry and to study ways to implement them from images obtained through simulators, such as Tartan Air.

Abstract: In robotics and computer vision, Visual Odometry is the process of determining the position and orientation of a robot by analyzing the associated camera images. It has been used in a wide variety of robotic applications, such as on the Mars Exploration Rovers. The classical approaches for solving the VO problems are usually based on the kinematics of the robot, so they can be easily induced to fail when either the motion of the robot or the environment is too challenging. In this way, novel VO methods use deep neural networks or attention models to improve or even replace the entire algorithm pipeline. In this project, we aim to evaluate these algorithms and study the fit of these models under simulated images obtained from the Tartan Air dataset.

Summary In robotics and computer vision, Visual Odometry (VO) is the process of determining the position and orientation of a robot from the analysis of images obtained by its cameras. Its techniques have been widely used in applications related to robotics, being a famous example of robotic vehicles used by the North American Space Agency (NASA) in the initial exploration of the territory of Mars. The classic techniques to solve problems of VO just involve the kinematics of robots, so it is common in very complicated situations, as when there are many changes in direction around the environment, for example, because of the lighting. Therefore, new methods of VO have been based on bringing new techniques, such as neuronal networks or attentional models. In this work we seek to evaluate these new models of visual odometry and to study ways to implement them from images obtained through simulators, such as Tartan Air.

Abstract: Computer-aided diagnosis has the potential to alleviate the burden on medical doctors and decrease misdiagnosis, but building a successful method for automatic classification is challenging due to insufficient labeled data. In this work, we investigate the usage of convolutional neural networks to diagnose musculoskeletal abnormalities using radiographs (X-rays) of the upper limb and measure the impact of several techniques in our model. While these techniques are overall well-established, some did not generalized to out setting. We achieved the best results by using an ensemble model that employs a support vector machine to combine different models, resulting in an overall AUC ROC of 0.8791 and Kappa of 0.6724 when evaluated using an independent test set.

Summary: Unicamp's smart parking project, developed within the context of the Smart Campus initiative, uses machine learning to identify the number of spaces available from photos taken from the parking lots. The system includes a totem pole, installed at the entrance to the parking pocket, to inform drivers of the number of available spaces. In order to improve and expand the availability of this information to the campus community, this work presents the design and implementation of a mobile application capable of informing, in real time and with safety for the driver, the number of spaces available in the parking pockets closest to the user's destination.

Summary: This project aims to assist the maintainers of the air conditioners of the Data Center of the Institute of Computing (IC) of the State University of Campinas to identify problems in their functioning quickly and visually through graphics and alarms generated by a monitoring system based on in IoT (Internet of Things). The motivation for the development of this work came from a failure - not detected quickly due to the lack of an appropriate notification system - that occurred in the air conditioning of the CI. Considering that a defect in these devices would affect the ability to keep the room cool, this could cause damage to the data center equipment with the progressive heating of the machines, and could burn server processors if it is not corrected quickly, which would cost the university for the replace. The solution proposes to add temperature and humidity sensors, providing assistance in detecting anomalous behaviors in the cooling of the monitored environment, in order to send alarms via e-mail in case of measurements outside a previously defined interval. It is also expected that, with time and an increasingly complete measurement database - acquired in atypical situations and different climates - it will be possible to develop projects capable of predicting system failures and even diagnosing their causes.

Summary: This work proposed to build a mobile computer system, which allows to add, control and analyze the data of patients submitted to treatments that make use of the Double J type catheter. The final product is a mobile application, based on the iOS platform, from Apple, which allows the addition of data related to the insertion of the catheter in patients, the updating of this data when the patient removes the catheter and the control of patients who are within the expected time for removal or have already exceeded the time limit allowed with the catheter. In addition, the application generates several statistical analyzes on the sample space of patients in the database.

Summary: Web pages that mimic banking services, monetary transactions and other financial transactions that require authentication or registration of sensitive information are called phishing websites. They are developed by attackers in order to obtain user credentials. In this project, a phishing website detector was developed from a trained classifier with several characteristics of a page, in order to predict whether it is phishing or not. It is believed that the use of this tool in browsers, firewalls and e-mail services can drastically reduce the number of victims of this type of fraud. The final model produced showed 95% accuracy and 95% F1 measurement.

Abstract: This work explores the impacts of three microservices design patterns on software evolution: API Gateway, Service Discovery, and Externalized Configuration. The main goal was to comprehend the counter effects microservices design patterns have on software evolution. For this, we present an analysis of the impact on software evolution through technical discussions on the counter effects of the design patterns on a couple of evolution scenarios. Results show important aspects of software evolution are impacted by microservices design patterns and emphasize the relevance of comprehending counter effects of microservices design patterns to orchestrate software evolution effectively.

Summary: The city of Campinas is full of stories and culture that, in general, go unnoticed not only by tourists, but also by its own long-time residents. Although local historical knowledge has been recorded and compiled, there is no way for it to be accessed or distributed. In view of the scenario described, a free software project (open source) was developed in partnership with the Campinas.com.br portal, in order to provide a platform for historical and cultural itineraries to be made available free of charge for the local population and tourists. Although initially planned for use in the city of Campinas, the service can be adapted and implemented in any location, by introducing content by organizations linked to culture and tourism. Based on this initial proposal, we developed (1) a mobile application that allows end users to travel tourist routes with access to multimedia content on historical and cultural elements, and (2) a Web application for platform management, following an agile methodology . Usability tests applied in the first show that, despite the short development period, the usability indicators collected are mostly positive, aligned with the purpose of the project and demonstrate that the services can in fact be deployed, if subjected to more development cycles, testing and improvement.

Summary: The architectural style in microservices, is one of the most used styles in the context of continuous delivery and that, due to its recent creation, it is necessary to define ways to identify, automatically, patterns that represent bad design practices and, therefore, should not be followed. This paper presents a study on antipatterns in microservice architecture. Based on the literature study, we have developed an automated way to identify the two most frequent anti-patterns that are megaservice e nanoservice, analyzing only the source code. We apply analysis to a project open source called SiteWhere, which uses the microservice architecture, and we were able to identify six microservices, among the 11 existing in the project, which presented the characteristics of the anti-patterns. This selection would allow project developers to analyze these microservices and take the actions they deem necessary to remedy the problem.

Abstract: This paper describes an interactive and student-friendly framework for teaching digital circuits and computer architecture topics. It aims to improve students learning process by providing a visual drag-and-drop circuit design editor, interactive simulation, signal monitoring and testbench tools - all integrated in a widely accessible application that runs in the browser. Circuitly does so in a programmatic way, to help students better understand the Hardware Description Languages ​​they will encounter in the future.

Abstract: Knowledge graphs (KGs) defines facts expressed as triples considering subject, predicate and object in the representation of knowledge. Usually, several knowledge graphs are published in a given domain. It is relevant to create alignments both for classes that model concepts and between instances of those classes defined in different knowledge graphs. In this work, we study techniques for aligning entities expressed in KGs. Our solution explores supervised ranking aggregation method in the alignment based on similarity values. Our experiments rely on the dataset from the Ontology Alignment Evaluation Initiative to evaluate the proposed method in experimental analyzes.

Summary: Networks of different natures, such as electrical, telephone and transport networks, are extremely common in our daily lives. The use of these networks usually consists of transporting something from one point to another in the most efficient way possible, thus ensuring the optimization of the use of resources in the network in question. Many of the problems that arise in the context of these networks can be modeled mathematically and solved using concepts and algorithms of Network Theory of Flows.

This project falls within the area of ​​Graph Theory and deals specifically with the Theory of Flows in Networks. We present a brief review of fundamental concepts of Network Theory of Flows, as well as the Theory of Pairing, as presented by Lovász and Plummer. In addition, concepts from the Theory of Flows in Networks are applied in the demonstration of some basic results of the Theory of Matching.