ICEx UFMG - Teses
Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Dissertação:

SCHEDULING IN HETEROGENEOUS ARCHITECTURES VIA MULTIVARIATE LINEAR REGRESSION ON FUNCTION INPUTS

Junio Cezar Ribeiro da Silva

Heterogeneous multicore systems, such as the ARM big.LITTLE, feature a single instruction set with different types of processors to conciliate high performance with low energy consumption. An important question concerning such systems is how to determine the best hardware configuration for a particular program execution. Current solutions are either completely dynamic, e.g., based on in-vivo profiling, or completely static, based on supervised machine learning approaches. Whereas the former solution might bring unwanted runtime overhead, the latter fails to account for the diversity in program inputs. In this dissertation, we show how to circumvent this last shortcoming. To this end, we provide a suite of code transformation techniques that perform numeric regression on function arguments, which can have either scalar or aggregate types, so as to match parameters with ideal hardware configurations at runtime.

 

Comissão Examinadora:

Prof. Fernando Magno Quintão Pereira - Orientador (DCC - UFMG)

Prof. Daniel Fernandes Macedo (DCC - UFMG)

Prof. Rupesh Nasre (CSE - Indian Institute of Technology Madras)

 

2 de Outubro de 2019

09:00h

 

Sala 2077 do ICEX

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Projeto de Tese:

Meta-aprendizado para determinação da qualidade de materiais biológicos por espectroscopia no infravermelho

Habib Asseiss Neto

Na indústria de laticínios, o FTIR é uma técnica interessante para análises de composição do leite, podendo analisar diferentes características de qualidade do produto, como a contagem de gorduras, lactose e outras proteínas. Essa técnica produz uma coleção de espectros que, se exploradas computacionalmente, podem oferecer conhecimentos importantes. Neste trabalho, aplicam-se métodos de aprendizado de máquina para a análise de amostras do leite bovino com o objetivo de determinar a qualidade das amostras e verificar a existência de adulterantes. Uma arquitetura de rede neural convolucional é proposta para análise dos espectros e oferece uma taxa de acerto de até 98,76% para a caracterização de adulterantes nas amostras. Além disso, uma metodologia de meta-aprendizado é proposta, que utiliza conceitos da Teoria de Resposta ao Item e permite determinar as capacidades de diferentes modelos de acordo com a determinação da dificuldade de instâncias.

Comissão Examinadora:

Prof. Sérgio Vale Aguiar Campos - Orientador (DCC - UFMG)

Prof. Ronnie Cley de Oliveira Alves (ITV - Instituto Tecnológico Vale)

Prof. Leorges Moraes da Fonseca (DTIPOA - UFMG)

Prof. Adriano César Machado Pereira (DCC - UFMG)

 

25 de Setembro de 2019

13:00h

 

Sala 2077 do ICEX

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Projeto de Tese:

 

CHARACTERIZING MULTIPLE INTERACTIONS IN DYNAMIC ATTRIBUTED NETWORKS BASED ON SOCIAL CONCEPTS

Thiago Henrique Pereira Silva

 

How to classify the dynamic interactions in social networks? We address this task by exploring the behavior and the dynamics of the actors in a social network. Specifically, we reinforce the importance of social concepts to capture the social meaning of relationships. Then, we propose a new method to classify nodes and dynamic edges based on node-attribute relationships. As a result, we unveil the differences of social relationships in different academic social network scenarios. Our method differs from an existing one by defining more elaborated classes in terms of social concepts and performing 7.7 times faster than it. We also validate our social definitions in terms of the best positioned nodes according to network metrics and investigate the robustness of our classification method by performing a sensitivity analysis of it. Finally, we apply our method to a ranking task in order to measure the social influence of the actors involved.

 

Comissão Examinadora:

Prof. Alberto Henrique Frade Laender - Orientador (DCC - UFMG)

Prof. Pedro Olmo Stancioli Vaz de Melo (DCC - UFMG)

Prof. Daniel Ratton Figueiredo (COPPE - UFRJ)

Prof. Wagner Meira Júnior (DCC - UFMG)

 

13 de Setembro de 2019

13:00h

 

Sala 2077 do ICEX

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem

a satisfação de convidá-lo para a Defesa de Tese:

 A Framework for Gamification of Project-based Software Engineering Education

 

Maurício Ronny de Almeida Souza

 

Balancing theory and practice is a recurring challenge in software engineering (SE) education. Project and game based approaches have been largely used to address this issue. The goal of this thesis is the design and evaluation of a framework to support educators in the adoption of PBL and gamification in the context of practical SE education. To achieve this goal, we evaluated the use of PBL and gamification in a series of empirical studies, and mapped a set of lessons and best practices. These lessons and practices were used as inputs for the design of the GaPSEE framework, which includes guidelines and a process for the setup of practical assignments in SE education. To evaluate the framework, we conducted five case studies in SE related courses from three federal universities. The evaluation consisted in the interview of four SE lecturers and responses from 76 students in a survey study. The results show a positive acceptance from both students and lecturers.

Comissão Examinadora:

Prof. Eduardo Magno Lages Figueiredo - Orientador (DCC- UFMG)

Profa. Claudia Maria Lima Werner (COPPE - UFRJ)

Prof. Marco Túlio de Oliveira Valente (DCC - UFMG)

Profa. Raquel Oliveira Prates (DCC - UFMG)

 

Profa. Christiane Anneliese Gresse von Wangenheim (INE - UFSC )

Publicado em: Teses
Avalie este item
(1 Voto)

A coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a defesa de projeto de tese:

TRE: Typing REST-based APIs by Example

Gleison Brito Batista

 

REST is “the facto” standard for implementing Web APIs. However, a key problem with the REST is that clients receive more data they actually required from servers. GraphQL is a novel technology that handles this problem by mans of a query language. In this thesis proposal, we first evaluated the gains achieved by GraphQL in practice. For example, we found that  GraphQL can reduce the size of the JSON documents returned by REST in 99%. We also conducted a controlled experiment to investigate the effort to implement REST and GraphQL queries. To conclude the PhD work, we are proposing a novel technique --- called TRE --- to generate type declarations for REST APIs by using as input only examples of endpoints. TRE has the potential to bring to REST the same benefits reached with GraphQL but without requiring a major reimplementation of API services.

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Tese:

Identifying and Characterizing Unmaintained Projects in GitHub
Jailton Junior de Sousa Coelho

Open source projects are key components in modern software development. Due to the emergence of novel platforms (e.g., GitHub and GitLab) for developing public code, developers are creating open source software at speeds and scales never seen before. As a consequence, a significant number of open source projects is also becoming unmaintained. To contribute to extend the lifetime of such projects, in this thesis we first survey the owners of unmaintained open source projects in order to reveal the reasons that motivate them to abandon the maintenance of their projects. Then, we conducted a second survey with developers who recently become core contributors of popular GitHub projects. Finally, we propose and validate a quantitative model to identify GitHub projects that are not actively maintained. This model presented a precision of 80% and a recall of 96%, when used in large sample of projects.

Comissão Examinadora:

Prof. Marco Túlio de Oliveira Valente - Orientador (DCC - UFMG)
Prof. Eduardo Magno Lages Figueiredo (DCC - UFMG)
Prof. André Cavalcante Hora (FACOM - UFMS)
Prof. Ricardo Terra Nunes Bueno Villela (DCC - UFLA)
Prof. Igor Fabio Steinmacher (CM - UTFPR)

30 de Agosto de 2019
13:30h
Sala 6321 do ICEX

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Projeto de Tese:

OPTIMIZATION PROBLEMS WITH INTERACTION COSTS
Dilson Almeida Guimarães

In this work, we investigate some combinatorial optimization problems with cost interactions between variables. First, we investigate two Lagrangian Relaxation approaches for computing Semidefinite Programmig (SDP) lower bounds for the Quadratic Minimum Spanning Tree Problem (QMSTP). We developed two QMSTP Branch-and-bound algorithms based on them. One of these algorithms stands out as the new best exact solution approach to the problem. Second, we investigate the Minimum Area Spanning Tree Problem (MASTP). We apply discrete geometry techniques to explore its cost structure and propose its first formulation. We also develop a MASTP Branch-and-bound algorithm based on this formulation and evaluated it experimentally.

Comissão Examinadora:

Prof. Alexandre Salles da Cunha - Orientador (DCC - UFMG)
Prof. Nelson Maculan Filho (PESC - UFRJ)
Prof. Geraldo Robson Mateus (DCC - UFMG)
Prof. Gabriel de Morais Coutinho (DCC - UFMG)
Prof. Dilson Lucas Pereira (DCC - UFLA)

21 de Agosto de 2019
15:00h
Sala 2077 do ICEX

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Projeto de Tese:

OPTIMIZATION PROBLEMS WITH INTERACTION COSTS
Dilson Almeida Guimarães

In this work, we investigate some combinatorial optimization problems with cost interactions between variables. First, we investigate two Lagrangian Relaxation approaches for computing Semidefinite Programmig (SDP) lower bounds for the Quadratic Minimum Spanning Tree Problem (QMSTP). We developed two QMSTP Branch-and-bound algorithms based on them. One of these algorithms stands out as the new best exact solution approach to the problem. Second, we investigate the Minimum Area Quadratic Minimum Spanning Tree Problem (MASTP). We apply discrete geometry techniques to explore its cost structure and propose its first formulation. We also develop a MASTP Branch-and-bound algorithm based on this formulation and evaluated it experimentally.

Comissão Examinadora:

Prof. Alexandre Salles da Cunha - Orientador (DCC - UFMG)
Prof. Nelson Maculan Filho (PESC - UFRJ)
Prof. Geraldo Robson Mateus (DCC - UFMG)
Prof. Gabriel de Morais Coutinho (DCC - UFMG)
Prof. Dilson Lucas Pereira (DCC - UFLA)

21 de Agosto de 2019
13:00h
Sala 2077 do ICEX

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Projeto de Tese:

Partial Least Squares: A Deep Space Odyssey
Artur Jordao Lima Correia

Deep convolutional networks have achieved remarkable results in pattern recognition, but they are computationally expensive. In addition, features that emerge in different layers are interrupted in deep architectures. To handle these problems, we propose two strategies. The first combines discriminative features distributed across the layers of the network, enhancing data representation. The second eliminates unimportant neurons of the convolutional network, reducing computational cost and memory requirements. Both strategies are based on Partial Least Squares (PLS), a discriminative dimensionality reduction technique, which is infeasible for large datasets due to memory constraints. To address this problem, we propose a novel PLS that learns a low-dimensional space by using a single sample at a time, being computationally efficient and enabling PLS-based approaches on large datasets. We assess the proposed methods on activity recognition, face verification and image classification.

Comissão Examinadora:

Prof. William Robson Schwartz - Orientador (DCC - UFMG)
Prof. Hélio Pedrini (IC - UNICAMP)
Prof. Jefersson Alex dos Santos (DCC - UFMG)
Prof. Luiz Eduardo Soares de Oliveira (DI - UFPR)

16 de Agosto de 2019
13:30h
Sala 2077 do ICEx

Publicado em: Teses
Avalie este item
(0 votos)

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem 
a satisfação de convidá-lo para a Defesa de Dissertação:

Misinformation, Radicalization and Hate Through the Lens of Users
Manoel Horta Ribeiro

The popularization of Online Social Networks has changed the dynamics of content creation and consumption. In this setting, society has witnessed an amplification in phenomena such as misinformation and hate speech. In this dissertation, we study these issues by analyzing user activity in social media. Through the lens of users, it is possible to approach possibly fake or hateful content with surrounding context: political orientation, activity patterns, connections, etc. Furthermore, we are able to study more complex phenomena, such as radicalization. In three case studies on social networks, we: (i) provide insight on how the perception of what is misinformation is altered by political opinion; (ii) propose a methodology to study hate speech on a user-level; (iii) characterize user radicalization in far-right channels on YouTube through time. Altogether, we advance the framework of understanding ill-defined social phenomena within the context of users

Comissão Examinadora:

Prof. Wagner Meira Júnior - Orientador (DCC - UFMG)
Prof. Virgílio Augusto Fernandes Almeida - Coorientador (DCC - UFMG)
Prof. Luis da Cunha Lamb (INF - UFRGS)
Prof. Pedro Olmo Stancioli Vaz de Melo (DCC - UFMG)
Profa. Jussara Marques de Almeida Gonçalves (DCC - UFMG)

13 de Agosto de 2019
08:30h
Sala 2077 do ICEX

Página 2 de 3