Ciclo de Seminários
O Ciclo de Seminários PESC tem como objetivo trazer palestras acessíveis a um público mais amplo
ministradas por pesquisadores e professores mais experientes. As palestras ao longo do ano terão
tema e foco variados podendo ser mais específicas (ex. avanços no contexto de um problema específico)
ou mais abrangentes (ex. desafios de uma área). A apresentação e discussão de ideias novas e antigas
de diferentes temas contribui de maneira fundamental para a formação e pesquisa desenvolvida por
alunos e professores.
As palestras do Ciclo de Seminários PESC ocorrem em geral uma vez ao mês, mas sempre
quarta-feira às 11h na sala H-324B. Segue a programação confirmada para 2019. Não percam!
NetMicroscope: Passive Measurements of Residential Internet Performance
Renata Cruz Teixeira, Senior Researcher, Inria Paris, França
27 de fevereiro (quarta), 11h
As access network speeds increase, the access link is less and less often the performance bottleneck for home users. In this context, performance tests a la "speed test" are rapidly becoming inadequate. First, they measure capacity to a dedicated server, not to real application servers. Second, they are becoming too disruptive as they must send enough probes to fill up the link. Instead of focusing on active measurements of access performance, our goal is to develop a mostly passive measurement system to monitor the performance of user applications. This talk will discuss our current system which measures video streaming quality completely passively. We will also discuss the lessons learned from deploying this system in 50 homes in the US and 10 homes in France.
Renata Teixeira is a senior researcher at Inria Paris and visiting scholar at Stanford University. She received her Ph.D. degree in computer science from the University of California, San Diego, in 2005. During her Ph.D. studies, she worked on Internet routing at the AT&T Research. She was a researcher with the Centre National de la Recherche Scientifique (CNRS) at LIP6, UPMC Sorbonne Universites, Paris, France from 2006 to 2013. She was a visiting scholar at UC Berkeyley/ICSI in 2011. Her research interests are in measurement, analysis, and management of data networks. Renata is co-auhtor of the MOOC "Internet Measurements: A Hands-on Introduction". She appeared in the 2017 list of "N2Women: Stars in Computer Networking and Communications". Renata is an ACM distinguished member. She was vice-chair of ACM SIGCOMM and member of the steering committee of the ACM Internet Measurement Conference. She has been active in the program committees of ACM SIGCOMM, ACM IMC, ACM CoNEXT, IEEE INFOCOM, among others.
Open Science - Challenges and Opportunities
Claudia Bauzer Medeiros, Professor Titular, UNICAMP
13 de maio (segunda), 11h
The term "Open Science" refers to the dissemination of all the material associated with scientific discovery, in a broad spectrum, with the goal of accelerating the advancement of science and increasing interdisciplinary collaboration. It is part of the European Commission's core strategies for financing projects, and one of the key issues in its long-term research agenda. The digital material to be disseminated refers to a variety of content types, which are broadly divided into three categories: Open Access (papers), Open Processes (mostly software), and Open Data (basically, any kind of digital artifact that does not fall within the two other categories). The talk will give an overview of the Open Science movement around the world, concentrating on the "Open Data" aspects, in which many interesting research, ethical, political - and even cultural - challenges have to be faced. Data sharing requires long term planning of data management, covering the entire lifecycle of all data items - from their collection, through cleaning, storage and preservation. Not only are there technical barriers (e.g., due to data heterogeneity or interoperability across repositories and cyber-infrastructures), but distinct norms, regulations and standards often hamper effective sharing. At the end, the talk will present a broad ongoing Open Data initiative in Brazil, involving 48 campi and over 20,000 researchers across the state of São Paulo.
Claudia Bauzer Medeiros is full professor of databases at the Institute of Computing, University of Campinas (Unicamp), Brazil. She holds a degree in Electrical Engineering (1976) and an MSc degree in Computer Science (1979) from PUC-Rio, Brazil and a PhD in Computer Science from the University of Waterloo, Canada (1985). For the past 20 years, she has been working as a visiting professor at the University Paris-Dauphine, France. She has received Brazilian and international awards for research, teaching, and also for her work in fostering the participation of women in IT-related activities. Her research is centered on the design and development of scientific databases. Her main interests lie in facing the challenges posed by large, real world applications, which require handling distributed and very heterogeneous data sources. In particular, she has coordinated large multidisciplinary projects, in Brazil, involving applications in agro-environmental planning and biodiversity. Data to be handled include, among others, sensor data streams, satellite images, photos, videos, sound and all kinds of textual sources. She has also coordinated research projects in scientific data management, workflow systems and geographic information, in cooperation with universities and research labs in Brazil, Germany and France. She is a Commander of the Brazilian Order of Scientific Merit, Dr. Honoris Causa from the University Antenor Orrego, Peru (2007), and Dr. Honoris Causa from the University Paris-Dauphine, France (2015). In 2018, elected for the Council of the Research Data Alliance, and as Member at Large of the Council of the ACM. In 2018, elected as a Member of the Brazilian Academy of Sciences.
Graph Representation Learning: Where Probability Theory, Data Mining, and Neural Networks Meet
Bruno Ribeiro, Assistant Professor, Purdue University, USA
17 de maio (sexta), 10h
My talk starts by turning back the clock to 1979-1983, introducing the ideas that culminated with the fundamental representation theorem of graphs (the Aldous-Hoover theorem). I will then show how these ideas connect to a probabilistic interpretation of matrix factorization methods, explaining why matrix factorization is fundamentally not as expressive as it could be to describe finite graphs. I will then turn to early machine learning attempts to represent graphs and how these attempts connect to graph mining algorithms. I will introduce the concept of representation learning with graph neural networks (GNNs) and explain its connections to statistical graph models and the Weisfeiler-Lehman isomorphism test. Finally, I will introduce a newly proposed general framework for graph representation learning using deep neural networks, which is directly rooted in the ideas that gave us the Aldous-Hoover representation theorem. This new representation framework points to novel graph models, new approaches to make existing methods scalable, and provides a unifying approach connecting matrix factorization, graph mining algorithms, and graph neural networks. I will end my talk with a few open problems.
This talk is in part based on joint work with Ryan Murphy, Balasubramanian Srinivasan, and Vinayak Rao.
Bruno Ribeiro is an Assistant Professor in the Department of Computer Science at Purdue University. He obtained his Ph.D. at the University of Massachusetts Amherst and did his postdoctoral studies at Carnegie Mellon University from 2013-2015. His research interests are in deep learning and data mining, with a focus on sampling and modeling relational and temporal data.
Integer Linear Programming Tricks with DEA (Data Envelopment Analysis) Applications
Mehdi Toloo, Full Professor, Technical University of Ostrava, Czech Republic
19 de junho (quarta), 11h
The fastest and most powerful solution methods are those for linear programming models. It is often advisable to use this format instead of solving a nonlinear model where possible. It is interesting to note that several practical problems can be transformed into linear integer programs. For example, integer variables can be introduced so that a nonlinear function can be approximated by a "piecewise linear" function. We first introduce some tricks in order to deal with nonlinearity issues in the following cases:
1. When a variable taking discontinuous values
2. Fixed costs
3. Either-or constraints
4. Conditional constraints
5. Elimination of products of variables
Moreover, we illustrate that how practically these tricks help us to address some problems in performance evaluation context.
Nowadays, it is necessary to evaluate efficiency (doing things right), effectiveness (doing the right things) and economy (doing things at a low price) in an organization. However, it is difficult to do this when there are multiple inputs and multiple outputs to the system. Data Envelopment Analysis (DEA) is a powerful nonparametric quantitative method in operations research and economics for evaluating the relative efficiency score of a set of Decision Making Units (DMUs), such as universities, car makers, hospitals, banks and so on.
Mehdi Toloo, B.Sc. (Pure Mathematics), M.Sc. (Applied Mathematics), Ph.D. (Operations Research), professor in the Department of Systems Engineering and Informatics, Technical University of Ostrava, Czech Republic. Areas of interest include Operations Research, Optimization, Linear Programming, Data Envelopment Analysis, Multi-Objective Programming, and Network Flows. He acts as an area editor in Computers & Industrial Engineering at ELSEVIER and Rairo - Operations Research. He has written fourteen books and his research has been published in top-tier journals including OMEGA, Energy, European Journal of Operational Research, Computers & Industrial Engineering, Computers & Operations Research, International Journal of Production Research, Journal of the Operational Research Society, Annals of Operations Research, Applied Mathematics and Computers, Applied Mathematical Modelling, Expert Systems with Applications, International Journal of Advanced Manufacturing Technology, Computers and Mathematics with Applications and Measurement.
Competition in randomly growing processes
Alexandre Stauffer, Associate Professor, Università Roma Tre (Italy) and Reader, University of Bath (England)
17 de julho (quarta), 11h
We consider 2-type random growth processes that compete for space over time. This is by now a classical topic in probability theory. The standard behavior expected from such processes is that, when the two types have different speeds of growth, then one of the types (usually the faster one) "wins" against the other. This means that the winning type grows indefinitely, whereas the other type stops growing after a finite amount of time. It is quite rare to find natural models showing coexistence, which refers to the situation when both types grow indefinitely. In this talk I will discuss a random growth model, which we introduced as a tool to analyze a well-known model of dendritic growth from physics. This growth model can also be regarded as a model for blocking the spread of fake news in a network. We will discuss the behavior of this process, its phase transition and the occurrence of coexistence.
This talk is based on joint works with Elisabetta Candellero, Tom Finn and Vladas Sidoravicius.
Alexandre Stauffer is an Associate Professor at the Dipartimento di Matematica e Fisica of the Università Roma Tre, and also a Reader at the Department of Mathematical Sciences of the University of Bath. He currently holds a EPSRC Early Career Fellowship, a major grant from the UK Research Council. He obtained his Ph.D. in 2011 from the University of California, Berkeley, and held a post-doctoral position at Microsoft Research.
The Future Quantum Internet: Research Challenges
Don Towsley, Distinguished University Professor, University of Massachusetts Amherst, USA
21 de agosto (quarta), 11h, sala G-122
Quantum information processing is at the cusp of having significant impacts on technology and society in the form of providing unbreakable security, ultra-high-precision distributed sensing with applications to metrology and science discovery (e.g., LIGO), much higher-rate deep space optical communications than possible with conventional systems, and polynomial speeds up on graphical search with implications to big data. Most of these applications are enabled by high-rate distributed shared entanglement between pairs and groups of users. A critical missing component that prevents crossing this threshold is a distributed infrastructure in the form of a world-wide "Quantum Internet" to enable this. This motivates our study of quantum networks, namely what the right architecture is and how to operate it, i.e., route multiple quantum information flows, and allocate resources fairly and dynamically.
In this talk we review a specific entanglement-based quantum network architecture and present opportunities and challenges related to resource sharing among multiple parties of users. In particular, we focus on issues related to resource allocation based on global/local state information and the benefits of path diversity. Last, we evaluate the performance of an entanglement-based quantum switch.
Don Towsley holds a B.A. in Physics (1971) and a Ph.D. in Computer Science (1975) from University of Texas. He is currently a Distinguished Professor at the University of Massachusetts in the College of Information & Computer Sciences. He has held visiting positions at numerous universities and research labs. His research interests include network science, performance evaluation, and quantum networking.
He was co-founder and Co-Editor-in-Chief of the ACM Transactions on Modeling and Performance Evaluation of Computing Systems (TOMPECS), and has served as Editor-in-Chief of IEEE/ACM Transactions on Networking and on numerous editorial boards. He has served as Program Co-Program Chair of several conferences including INFOCOM 2009.
He is a corresponding member of the Brazilian Academy of Sciences and has received several achievement awards including the 2007 IEEE Koji Kobayashi Award and the 2011 INFOCOM Achievement Award. He has received numerous Test of Time Awards. He also received the 1998 IEEE Communications Society William Bennett Best Paper Award. Last, he has been elected Fellow of both the ACM and IEEE.
Modelling and Optimization of Non-linear Complex Systems
Elizabeth Wanner, Senior Lecturer, Aston University, England
28 de agosto (quarta), 11h
The Aston Lab for Intelligent Collectives Engineering (Alice) is Aston University's Research Group for intelligent systems. Our research expertise includes various forms of intelligent, social and collective computing systems. Alice's work on modelling and optimization of non-linear complex systems is being applied to a wide range of domain including sensor networks, non-linear dynamic systems modelling techniques, statistical-based comparison methodology for evaluating algorithms, the mathematical foundation of optimization algorithms, and adaptation law synthesis for a self-adaptative evolution strategy. This talk will discuss some active research projects such as Aedes aegypti control, phoneme aware speech recognition through evolutionary optimization, vehicle fleet optimization using Ant Colony Optimization and Lyapunov design of success-based step-size adaptation rules.
Elizabeth Wanner is a Senior Lecturer at Aston University, Birmingham, UK. She received her PhD degree in Electrical Engineering from UFMG, Brazil in 2006. and her MSc degree in Mathematics from UFMG, Brazil, 2002. During her PhD, she worked with the Rolls-Royce Optimization Group at University of Sheffield, Sheffield, UK. Elizabeth's research interests are in population-based multiobjective optimization and matheuristics, multi-criteria decision analysis and mathematical and statistical aspects of optimization theory.
Sistemas Brasileiros de Ensino Superior, Ciência e Tecnologia: fatos e experiências pessoais
Cláudia Linhares Sales, Professora Titular, Universidade Federal do Ceará
23 de outubro (quarta), 13h
Nessa palestra, à luz da organização e estrutura do sistema de educação superior, ciência e tecnologia do Brasil, comentaremos os impasses atuais, tais como financiamento, novas políticas das agências de fomento, avaliação e medidas de qualidade, entre outros. Esses comentários serão suportados pelas experiências da palestrante como coordenadora de curso de pós-graduação na UFC, diretora científica de fundação de amparo a pesquisa FUNCAP, pró-reitora adjunta de pós-graduação e pesquisa da UFC, membro de comitês de avaliação quadrienal da CAPES e comitê de assessoramento do CNPq, e participante ativa das sociedades científicas SBC e SBPC. Os comentários finais serão dedicados às formas de resistência e organização da comunidade científica para o enfrentamento dos atuais desafios.
Cláudia Linhares Sales é professora titular da Universidade Federal do Ceará (UFC) e ocupou o cargo de Diretora Científica da FUNCAP (Fundação Cearense de Apoio ao Desenvolvimento Científico e Tecnológico) em 2010-2011 e 2012-2014. Possui mestrado em Engenharia de Sistemas e Computação pela Universidade Federal do Rio de Janeiro (1990) e doutorado em Informatique - Recherche Operationnelle - Université de Grenoble I (Scientifique Et Medicale - Joseph Fourier) (1996), sob a supervisão de Frédéric Maffray. Fez pós-doutorado no INRIA/Sophia-Antipolis, França, entre 2006 e 2007, e na Simon Fraser University, Canadá, entre 2015 e 2016. Trabalha com Teoria dos Grafos e Algoritmos, atuando principalmente nos temas de coloração e decomposição de grafos. Ela fundou e coordena o grupo de pesquisa ParGO (Paralelismo, Grafos e Otimização), do Departamento de Computação da UFC.
Relato do Processo no USPTO da Patente: "Hyperbolic Smoothing Clustering and Minimum Distance Methods"
Adilson Elias Xavier, Professor Colaborador, PESC/COPPE/UFRJ
11 de dezembro, 11h
Hyperbolic Smoothing Clustering Methods (HSCM), muito mais do que novos algoritmos, apresenta um diferenciado enfoque completamente diferenciável, absolutamente inovador no tópico Clustering Analysis. Graças à essa característica, o HSCM oferece a vantagem singular de poder resolver uma amplitude de formulações de clustering, definidas por minimização de soma de avaliações de uma arbitrária função monótona crescente de distâncias, as de cada observação ao centroide mais próximo, calculadas segundo diferentes métricas, tais como as mais utilizadas: Euclidiana, Manhattan, Minkowski e Chebychev. O outro componente básico, a estratégia de particionamento Boundary Band Zone and Gravitational Regions Partition, possibilita uma simplificação expressiva do trabalho computacional, pois reduz o conjunto de observações inicial a um conjunto efetivo com somente um pequeno percentual delas. Para problemas de grandes dimensões, essa redução do trabalho computacional chega à ordem de 99,9% do total de observações. Ademais, quanto maior o problema, maior o fator de redução. A inovadora articulação da suavização hiperbólica com a estratégia de partição acima descrita, aplicada à especificação mais usual minimum sum-of-squares clustering problem produz resultados computacionais superiores ao k-means segundo quatro diferentes critérios: de acurácia, de velocidade, de consistência e de escalabilidade. Igualmente, outras especificações alternativas, como a "minimum sum-of-distances clustering problem", são tratadas com igual sucesso. Finalmente, a metodologia HSCM tem a diferenciada capacidade de tanto poder produzir resultados segundo o enfoque tradicional hard, como também segundo o fuzzy, fato igualmente inaudito na literatura e nos softwares oferecidos no mercado. A palestra apresentará toda a série de peripécias ocorridas até a aprovação da patente, após nove anos de tramitação no United States Patent and Trademark Office.
Adilson Elias Xavier obteve a graduação em Engenharia Mecânica pela UFMG e os títulos de M.Sc. e D.Sc. pelo PESC/COPPE/UFRJ. Trabalhou no PESC na linha de Otimização como Professor até sua aposentadoria em novembro de 2013 e onde, desde então, continua trabalhando como Professor Colaborador. Seu projeto central de pesquisa tem o nome: Penalização Hiperbólica, Lagrangeano Hiperbólico e Suavização Hiperbólica.
Understanding and Modelling Consciousness
David Gamez, Lecturer at Middlesex University, England
16 de dezembro (segunda), 11h
The first part of my talk will explain how our modern concept of consciousness emerged. In the 17th Century many people used the properties of invisible atoms to explain regularities in the world. This led to a distinction between primary qualities, such as size, which were properties of the atoms, and secondary qualities, such as colour, which appeared when atoms interacted with our senses. Secondary qualities were real non-physical properties that had to be accommodated somewhere. Galileo's and Locke's solution was to locate secondary qualities in consciousness. While modern science has more elaborate physical descriptions, consciousness continues to be a placeholder for experiences that cannot be captured by a physical description.
Philosophers often study the relationship between consciousness and the physical world using thought experiments. For example, in the hard problem of consciousness, philosophers imagine a red object, imagine activity in a grey neuron, and try (and fail) to imagine the relationship between the two. A more promising methodology is to measure consciousness, measure the physical world and look for correlations between the two sets of data. Using this scientific approach, we might be able to discover precise mathematical relationships between measurements of consciousness and measurements of the physical world, which could be used to make accurate predictions about conscious states.
The last part of my talk will discuss how scientific research on consciousness connects with models of consciousness and conscious machines. Many people have built neural and cognitive models of the correlates of consciousness, which have been used to improve our understanding of consciousness in the brain and to build intelligent machines. Models of conscious experience have also been used to control robots. There are no grounds for believing that systems based on these types of models have real conscious experiences. However, if science could discover mathematical relationships between consciousness and the physical world, then it would become possible to build machines with specific states of consciousness and to make accurate predictions about the consciousness of machines.
Holding two PhD, one in Computing and Electronic Systems and the other in Philosophy, both from Essex University, David Gamez is currently a lecturer at the Department of Computer Science, Middlesex University. From 2012-2015 he was supported by a JTF Turing Research Fellowship at the Sackler Centre for Consciousness Science, where he worked on a three year project on the scientific study of natural and artificial minds. Between 2010 and 2012 he worked with Murray Shanahan at the Department of Computing at Imperial College London. As part of the EPSRC project 'Modular Neural Simulation with Reconfigurable Hardware' I integrated the NeMo CUDA accelerated neural simulator with his SpikeStream neural simulator and contributed to the development of the iSpike sensory interface for the iCub robot. From 2009-2010 he worked with Igor Aleksander on a new technique for analyzing neural networks for information integration. In 2008, his PhD on machine consciousness, supervised by Professor Owen Holland at the University of Essex, UK, was carried out as part of the EPSRC-funded CRONOS project to build a conscious robot. His contribution to this project included the development of new techniques for analyzing systems for signs of consciousness and hebuilt a spiking neural simulator that he used to model a neural network that controlled the eye movements of a virtual robot. In his previous job on the IST Safeguard project he worked on agents, anomaly-detection and GOFAI. His most recent book, Human and Machine Consciousness, explains how we can neutralise the traditional philosophical problems with consciousness and develop a science of consciousness that can make accurate predictions about the consciousness of humans, animals and machines. His first book, What We Can Never Know, explores the limits of philosophy and science through studies of perception, time, madness and knowledge.
Entre em contato e envie seus comentários e sugestões, inclusive potenciais palestrantes.
Organizado por Daniel R. Figueiredo.