Final Graduation Projects published in 2017

  • IC-PFG-17-24 pdf bib
    Study and Implementation of the Lateral Channel Attack in the OpenSSL ECDSA.
    Felipe Rodrigues Novaes, Diego de Freitas Aranha, and Yuval Yarom.
    December 2017. In Portuguese, 29 pages.

    Summary: The ECDSA elliptic curve digital signature algorithm (Elliptic Curve Digital Signature Algorithm) has an efficient and secure implementation in the software OpenSSL. However, this work shows that this implementation for curves over $ \ mathbb {F} _ {2 ^ m} $ in the current version it is susceptible to timing side channel attacks to be able to recover the most significant bit of the scalar nuncio during the multiplication phase.

    To demonstrate the leak, this work implements an attack using the technique of FLUSH + RELOAD together with performance degradation in signatures made with the private key generated by the sect163r1 koblitz curve. The test in question generated 10.000 signatures applying the attack to recover the most significant bit, taking around 8 hours. The result obtained was a precision of $ 97,3 \% $ e $ 89,6 \% $ recall for cases with bit 1 and $ 90,3 \% $ e $ 90,3 \% $ for bit 0.

    Future work can use this bit to break the ECDSA by finding the private key through Bleichenbacher's modified attack. In addition, future work may investigate methods of speeding up the bit recovery attack by decreasing the effect of performance degradation by removing it from the rest of the signature, which would make the attack more practical and efficient.

  • IC-PFG-17-23 pdf bib
    Algorithm for the Graph Isomorphism Test.
    Pedro Tadahiro Furtado Kaneko and Eduardo C. Xavier.
    December 2017. In Portuguese, 16 pages.

    Summary: Graph isomorphism detection is a known problem in Graph Theory which consists of verifying if two graphs have the same morphological structure, that is, if there is a bijection between the sets of vertices with edge preservation. The definition of which complexity class the problem belongs to is still open, as a polynomial solution was not found, as well as the NP-Difficult problem class.

    For this study, an exact algorithm of backtracking proposed by Mittal, to implement, execute and compare results with other algorithms. The comparison was made in relation to the results found by Mattos for the Weisfeiler and Lehman algorithm.

    The results showed that the algorithm is superior to the "Brute Force" method, in which all vertex permutations are tested, and is more efficient than the Weisfeiler and Lehman algorithm for graphs of the type regular mesh.

    Even though the algorithm implemented has the worst case factor complexity, all test cases were resolved within one timeout 5 min, except for difficult instances. For the vast majority of tests, the result was generated quickly in less than 1 min.

  • IC-PFG-17-22 pdf bib
    Application of chatbots in the development of health games.
    Rogério O. Bernardo and André Santanchè.
    December 2017. In Portuguese, 11 pages.

    Summary: The use of digital tools to aid learning has a number of advantages, the automation of part of the pedagogical process and the ease of sharing knowledge being the most notorious. With the popularization of smartphones, many pedagogical tools have been migrated to this context in the form of applications, whose need for installation is often a limitation for devices with low memory, or responsive web pages, whose navigability is not always simple on small screens. In this work, we developed a platform for health training, taking advantage of the popular messaging apps on these devices.

  • IC-PFG-17-20 pdf bib
    Probabilistic location in robotics.
    João Guilherme Daros Fidelis and Esther Luna Colombini.
    December 2017. In Portuguese, 17 pages.

    Summary: The problem of auto location is fundamental for several types of autonomous robots. Robots depend on their local position to make many of their decisions about what their next actions will be. In this paper, we discuss the location problem facing humanoid robots that participate in RoboCup robot football. We studied the Monte Carlo particle filter probabilistic algorithm and implemented it using a Webots simulator with a virtual model of the Aldebaran NAO robot. We concluded that it is possible to obtain good position estimates, depending mainly on your movement model and observation model. We also show that increasing the number of particles in the filter decreases the error of the estimates.

  • IC-PFG-17-19 pdf bib
    Visual Odometry for robots.
    William Nobuaki Ikedo and Esther Luna Colombini.
    December 2017. In Portuguese, 11 pages.

    Summary: For a robot to act autonomously, it is essential that it is able to locate itself in relation to the environment, and thus determine the path it must follow to perform its task. In this context, visual odometry presents itself as one of the most used methods, and therefore the development and evaluation of precise and robust algorithms for this purpose are of great relevance. Thus, this work investigates the method Direct Sparse Odometry (DSO) applied to the context of mobile robots. For this, tests were performed using a model of the differential robot robotino in the simulation environment V-Rep, using different speeds for the robot and textures applied to the environment. The results obtained in this study call into question the precision and robustness shown in the original publication, and reinforce the need for further investigation on the limitations of the method.

  • IC-PFG-17-18 pdf bib
    Atari-playing Robot.
    Renato Landim Vargas and Esther Luna Colombini.
    December 2017. In English, 14 pages.

    Abstract: Great results achieved by Deep Q-Networks on Atari games gathered a lot of attention from researchers in the past year. Despite this, not many research has been conducted on using these networks on real-world robots learning through high-dimensional sensory input. This work proposes ALE-Robot, which is an architecture that uses ALE (Arcade Learning Environment) and the V-REP environment to simulate a robot learning to play Atari directly from its camera and a game controller through Reinforcement Learning. Results have demonstrated that, when applied to real robotics architectures, longer training periods and re-shaping rewards functions are required. Moreover, more time and computational power is necessary to test the feasibility of the proposed approach over different hyperparameters.

  • IC-PFG-17-17 pdf bib
    Study of realistic simulations in LTE / LTE-A mobile access networks.
    Luiz Rodolfo Felet Sekijima and Nelson Luis Saldanha da Fonseca.
    January 2018. In Portuguese, 28 pages.

    Summary: This is a study of mobile access networks Long Term Evolution (LTE) and LTE-Advanced (LTE-A). A module for the LTE-Sim simulator was implemented in c ++ in order to study the LTE random access procedure, traffic models, communication Machine to Machine, Device to Device and energy consumption on mobile devices. The module contains implementations of new proposed procedures or the implementation of the standard that were missing from the simulator. Problems with LTE-Sim made it impossible to use it for D2D communication. Several scenarios were created to validate the module and the results found are consistent with the standard.

  • IC-PFG-17-16 pdf bib
    Improvement of the RISC-V processor model.
    Renan Camargo de Castro and Rodolfo Azevedo.
    December 2017. In Portuguese, 18 pages.

    Summary: This work was developed as a final graduation project and sought to understand and evolve the computational model of the RISC-V processor developed with the ArchC platform. The use of the model to generate a simulator presented several challenges, such as: execution problems, instructions with bugs, and others. The work sought to solve them, having been successful in this regard, including perfecting the tool used to test processor models.

    As a result of the research and development carried out in this project, ArchC has a minimally tested model of RISC-V, as well as a minimum set of tools necessary for future research.

  • IC-PFG-17-15 pdf bib
    Software Defined Networks.
    Lucas Calzolari, Daniel Ricci, Bleno Claus, and Luiz Fernando Bittencourt.
    December 2017. In Portuguese, 11 pages.

    Summary: This work is related to a sub-project of the INCT of the Internet of the Future that aims at the development of a prototype of unified, open and free network infrastructure, based on the technology of virtualization of network functions (Network Functions Virtualization - NFV), which allows us to offer, in an agile way and with low operating cost, an innovative communication, processing and storage service. The project adopts the open source platform OPNFV and as a result, a cloud environment, accessed and interconnected via the Internet, fully transparent of the location of physical machines and the offer of virtual network management and security functions, such as firewall, is expected. , traffic characterization and intrusion detection system. In this work we set up all the infrastructure and environment necessary for the next stages of the project.

  • IC-PFG-17-14 pdf bib
    Solutions for Cloud Vendor Lock-In.
    Marcos Massayuki Kobuchi and Luiz Fernando Bittencourt.
    December 2017. In Portuguese, 20 pages.

    Summary: This work is the report of a project aimed at building a single interface for accessing storage services, in which AWS S3, Azure Blob Storage and OpenStack Swift are the best known. Using an existing framework developed in.NET, implemented for AWS S3, Azure Blob Storage and Google APIs, this work seeks to initiate the studies necessary for the implementation of yet another service, OpenStack Swift. Based on these studies, we found the existence of an unofficial framework that allows interaction with the OpenStack API, however, we found that it was out of date, with an update branch already in progress. To work around this problem, we have provisionally studied the interactions with the API through HTTP requests.

  • IC-PFG-17-13 pdf bib
    Heuristic and Exact Algorithms for the 2D Knapsack Problem with Relations Between Items.
    K. Rollmann, W. Cardoso, V. L. Lima, and F. K. Miyazawa.
    December 2017. In English, 27 pages.

    Abstract: This work deals with a variation of the 0-1 two-dimensional knapsack problem, where each item has a value and there is also a value between each pair of items. The value of adding one item into the packing depends not only on the item's value, but also on its relation values ​​with other items that are in the packing. We adapted some inputs commonly found on the literature to our problem and we propose some exact formulations to solve the problem. Our formulations use integer programming models that work in two phases: the first finds a packing considering relaxed constraints, and the second checks if the packing is feasible. Three approaches to check the packing feasibility were compared. Furthermore, we propose four decoders to the BRKGA meta-heuristic and compare their quality with the solutions found using exact models. Computational results are also provided to show the quality of our methods.

  • IC-PFG-17-11 pdf bib
    Comparison of Measurements for Blur Detection in Images.
    Flávio Marcos Helder Moura and Helio Pedrini.
    December 2017. In Portuguese, 10 pages.

    Summary: Blurring is one of the most common unwanted effects in photography. There are several different causes of blur, from poor choices for determining the focus range to long shutter release times or movements of objects in the scene. Instantaneous blur detection in photographs allows cameras to capture photographs in sequence or execute correction algorithms. In this work, different methods for detecting blur are compared. The methods are applied to images with controlled levels of blur generated by averaging filters over sharp original images.

  • IC-PFG-17-09 pdf bib
    System for managing events.
    André Nogueira Brandão - Guilherme Costa Zanelato - Matheus Pinheiro dos Santos - Diego Silva de Carvalho - Breno Bernard Nicolau de França.
    December 2017. In Portuguese, 8 pages.

    Summary: The problem of managing events is of great relevance to the academic community since there are several types of activities and lectures with specific needs that occur every year in this medium. In this context, this work aims to develop a web application that is capable of effectively managing such events for both users and their administrators. For this project, modern agile development techniques and practices were used with validation by the main stakeholders.

  • IC-PFG-17-08 pdf bib
    Object Detection in Robot Football.
    Gabriel Borges Magalhães and Esther Luna Colombini.
    July 2017. In English, 21 pages.

    Summary: The problem of tracking the position of a robot in a scene is extremely relevant for autonomous robotics and the quality of this process is instrumentally related to the quality of the extraction of characteristics and the estimation of the distance of the robot to the elements of interest in a known map. In this context, this work aims to implement an object detection system - relevant to the field of robot football - capable of identifying elements of interest in images captured by a camera and estimating the location of these objects in relation to the robot. For the project, the Webots smilator was used as well as the virtual model of the humanoid robot NAO. Techniques for detecting low-level objects based on the images collected were studied and implemented, and subsequently the highest level analysis was performed for the classification or disposal of candidates and the representation of objects.

  • IC-PFG-17-07 pdf bib
    Humanoid Robot Walking Optimization using Genetic Algorithms.
    Luiz Fernando Cirigliano Villela - Esther Luna Colombini.
    July 2017. In English, 15 pages.

    Summary: The problem of creating fast and stable walking for humanoid robots is a very complex one due to the large degree of freedom and the external variables involved. This work tackles this problem by using a model free approach where each joint is represented by a sinusoidal function of time. The parameters of the functions for all actuated joints are optimized using a genetic algorithm. Experiments were performed with a NAO robot in a simulated environment under V-REP. The optimized robot was able to walk at a speed of $ 54 cm / s $ in a straight line and for up to 200 meters without falling. Experiments were also carried out to evaluate the individuals capacity to adapt to different scenarios, such as walking up and down ramps. Results showed different movement patterns, a slower pace and more upright positions for the robot walking uphill.

  • IC-PFG-17-06 pdf bib
    RoboCup Soccer Ball Depth Detection using Convolutional Neural Networks.
    Gabriel Militão Vinhas Lopes and Esther Colombini.
    July 2017. In English, 08 pages.

    Abstract: The RoboCup, a soccer competition played by autonomous robots, raises a variety of hard problems in different fields such as Artificial Intelligence and Dynamics. In particular, vision is the essential input to take actions based on this highly dynamic environment. Although traditional techniques might be cheaper for object detection, new rules such as multicolored balls require more robust detection methods and allow a significant computing power improvement. Given these recent changes, this work presents a solution for ball localization using convolutional neural networks (CNNs), a largely applied machine learning technique holding the state of the art technology for several imagery problems. To perform the object detection, we train the CNN with robot's camera images as input and the ball's coordinates relative to the robot as output.

  • IC-PFG-17-05 pdf bib
    SimPoints applied to multiple entries.
    Luis Fernando Antonioli and Rodolfo Azevedo.
    July 2017. In English, 17 pages.

    Summary: Understanding the cycle-level behavior of a processor running a program is vital for modern computer architecture research. In order to obtain this information, detailed simulators are generally used. The complete simulation of a benchmarking standard can take weeks or even months to complete. To address these problems, statistical techniques such as the methodology simpoint have been proposed. The methodology simpoint it tries to identify repetitive phases and to find a small set of samples of the execution of the program that represents the majority of the execution of the program, that is, it seeks to predict some property of the architecture based on the individual execution of samples of the phases of the program. Architectures can be compared by simulating their behavior in the code samples selected by simpoint to quickly determine which architecture performs best. The methodology simpoint performs the analysis of the phases of each program-entry pair separately. In this work we study the methodology simpoint, we propose and implement an extension of it to allow the analysis of phases of a program for several inputs, aiming to try to reduce the total number of SimPoints necessary to simulate a benchmarking all.

  • IC-PFG-17-04 pdf bib
    Time synchronization under temperature and distance variations.
    Matheus Susin and Lucas Francisco Wanner.
    July 2017. In English, 18 pages.

    Abstract: The Trustful Space-Time Protocol (TSTP) allows for time synchronization to be performed upon receiving any message from another node in a sensor network, removing the need for explicit messages. Previous work has shown that TSTP performs well under controlled experimental environments. In this work, we analyze how the quality of synchronization in TSTP is affected when nodes are communicating over varying distances, and across a temperature gradient. The results show that, while distance and packet loss has only a small effect on the quality of the synchronization, high temperatures of 80 Celsius and up can negatively affect it. Snippets of the results are presented, along with charts, statistics, and an analysis. A repository that contains all results, as well as the tools to analyze them, is available at https://gitlab.com/mathy/PFG.

  • IC-PFG-17-03 pdf bib
    Machine Learning Applied to Sorting Permutations by Reversals and Transpositions.
    Flavio Altinier Maximiano da Silva, Andre Rodrigues Oliveira, and Zanoni Dias.
    July 2017. In English, 20 pages.

    Abstract: The problem of determining relationship trees between genomes is fundamental when studying life evolution on the planet. As mutations are rare, it is believed that when one genome transforms into another, it probably used the fewest operations possible. If we represent genomes as numeric permutations, we reduce that problem to the one of sorting permutations using specific operations. In this work we use two of the most common genome mutations: reversals and transpositions. We propose a machine learning approach where a classifier is trained on a set of features of small permutations and then used to sort bigger permutations. Results show that this method is competitive when compared to others in literature, especially when dealing with small permutations or considering the others' maximum approximation factors.

  • IC-PFG-17-02 pdf bib
    Infrastructure Analysis for Scientific Experimentation.
    Thiago de Oliveira Favero and Breno Bernard Nicolau de França.
    December 2017. In Portuguese, 25 pages.

    Summary: In a world where competition between companies that provide products or services based on software it is getting bigger and bigger, it becomes more and more important not only to make consumers and users loyal by creating features that meet their needs, but also to reduce unnecessary expenditure of time and money by companies. With these objectives in mind, several models and methodologies for software emerged in recent years, one of the most recent being Continuous Experimentation.

    In this work we study Continuous Experimentation by implementing some stages of its most complete and widespread theoretical model so far. Thus, we seek tools available on the market that allow us to collect and analyze the usage and user data of an application, and to carry out experiments in an attempt to validate or refute hypotheses raised, aiming to improve the users experience and achieve a pre-defined objective .

    With this tool, we selected an application in production on which we collected data and, despite some limitations, we carried out a successful experiment, managing to achieve results that confirmed our initial hypotheses.

  • IC-PFG-17-01 pdf bib
    Performance of binary classifiers in multi-class data sets.
    Pedro De Nigris Vasconcellos and Jacques Wainer.
    July 2017. In English, 14 pages.

    Summary: Through this work, we seek to better understand the problem of multi-class data classification. We seek to understand the performance, both in terms of accuracy and time, of inherently binary algorithms, exploring new approaches such as the internal search for hyper parameters. We focus on exploring the differences between internal and external hyperparameter search for modeling One vs. one e One vs. Rest, which use binary algorithms for classification. We are pleased with the results obtained, since they allowed a better understanding of this problem and other possible ways that we can solve it. Comparisons were made involving the execution time and accuracy of the classifiers and we can conclude that there are significant differences between the internal and external search for hyper parameters, as well as differences between the binary algorithms used in the classifiers. This e ovR.


  • Instituto de Computação :: State University of Campinas
    Av. Albert Einstein, 1251 - Cidade Universitária Zeferino Vaz • 13083-852 Campinas, SP - Brazil • Phone: [19] 3521-5838