The Postgraduate Colloquium Series is an opportunity for Thesis students to present their ongoing research work to the DEI and CISUC research communities, including all Ph.D. and Master's students, academic and research staff. Other interested parties are most welcome to attend, as well. In this page you will find all details regarding the current series. To review past series, please follow the links on the menu. To receive regular updates about the Postgraduate Colloquium Series, join our mailing list here.
DEI Building (Polo II) - Amphitheater B1
“Experimental evaluation of evolutionary algorithms for multiobjective optimisation: methods and algorithms”
Evolutionary Multiobjective Optimisation (EMO) is currently one of the most active areas in Evolutionary Algorithm research, and remains very promising with respect to both theory and applications. On the theoretical front, multiobjective optimisation forces many concepts, such as elitism, solution quality, convergence and algorithm performance, to be generalised. Regarding applications, EMO algorithms offer a great amount of flexibility, both at the objective formulation and at the preference articulation levels, as well as competitive performance with alternative approaches.
This talk will focus on the experimental evaluation of the performance of EMO algorithms. After a general introduction, some of the author's original contributions since the mid-nineties will be presented, including the attainment-function approach (a methodology anchored on results from random-set theory), its relation to the indicator-based approach proposed by others, and the development of computational geometry algorithms to support both methodologies. The results achieved so far and opportunities for further work will be discussed at the end.
|2014-05-02||Carlos Fonseca||ECOS||Experimental evaluation of evolutionary algorithms for multiobjective optimisation: methods and algorithms|
*All sessions start at 14:00 (sharp!)
|2014-04-04||Rafael Alexandre||Universidade Federal de Ouro Preto||Algoritmos Evolucionários Multiobjetivo Aplicados ao Problema de Planejamento Operacional em Minas a Céu Aberto|
|2014-03-28||Filipe Araujo||CISUC-SSE (Senior Member)||A Maximum Independent Set Approach for Collusion Detection in Voting Pools|
|2014-03-21||Roger Immich||LCT||Ensuring QoE in Wireless Networks through Adaptive FEC-based Mechanisms|
|2014-03-14||Renato Panda||CMS||Emotion-based Analysis and Classification of Audio Music|
|2014-03-07||Nuno Lourenço||ECOS||Evolution of Bio-Inspired Algorithms using Hyper-Heuristics|
|2014-02-28||Tiago Cruz||LCT||The CockpitCI project: cyber-security for critical infrastructures|
|2014-02-14||Carlos Eduardo Pedreira||COPPE/Universidade Federal do Rio de Janeiro||Classificação de Padrões e Aplicações em Medicina|
|2013-12-13||Alcides Fonseca||SSE||Towards a Better Automatic Optimization of Parallel Programs|
|2013-12-06||Andreia Guerreiro||ECOS||Towards incremental algorithms for hypervolume-based selection in EMO|
|2013-11-29||Filipe Rodrigues||CMS||Learning from Crowds|
|2013-11-22||Marco Simões||IBILI/ACG||Virtual Reality and EEG for social studies and rehabilitation in autism|
|2013-11-15||Maryam Abbasi||ECOS||On Multi-objective Sequence Alignment|
|2013-11-08||Cidália Fonte||DMUC||Volunteered Geographic Information and Map Validation|
|2013-11-01||José Garzon||ECOS||Visual Evolutionary Framework for Cluster Analysis of DNA Microarray Data|
|2013-10-25||Thanh-Dien Tran||LCT||Wireless Sensor Networks in Noisy and Interference Environments: Problem and Solution|
|2013-10-18||Luís Mendes||ACG||Positron emission tomography|
|2013-10-11||Nuno Vasco Lopes||LCT||Survey of IoT and its Feasibility Analysis for People with Special Needs|
|2013-10-04||Ivano Elia||SSE||A diagnosis based approach to improving intrusion detection technologies|
|2013-09-20||Rui Gomes||CMS||An integrated approach for the design and operation of Demand Responsive Transportation services|
“Algoritmos Evolucionários Multiobjetivo Aplicados ao Problema de Planejamento Operacional em Minas a Céu Aberto”
Universidade Federal de Ouro Preto
O problema de planejamento operacional em uma mina a céu aberto (Open-Pit-Mining Operational Planning Problem – OPMOPP) é amplamente discutido pela comunidade científica. Procura-se atender às necessidades da mineração utilizando os seus recursos (caminhões, máquinas de carga, etc...) da forma mais eficiente possível em busca de competitividade e redução de custos. Neste contexto, estratégias de despacho de equipamentos de transporte em uma mina a céu aberto serão apresentadas bem como as principais características do problema. Um modelo matemático multiobjetivo que envolve estas características será apresentado. Devido à natureza combinatória do problema, algoritmos evolucionários são utilizados para resolver o problema por serem flexíveis, ou seja, aplicados a diversos contextos e permitirem a exploração do espaço de busca a procura de soluções viáveis. Duas propostas de representações de soluções para o problema serão apresentadas assim como operadores de cruzamento e mutação especialistas. Uma estratégia de subdivisão do problema será definida propondo assim, um algoritmo denominado OPMOP-GA. As representações e operadores propostos são utilizados pelos algoritmos NSGAII, SPEA2, OPMOP-GA e o ILS+VND. Com o objetivo de avaliar as soluções geradas pelos algoritmos de otimização, uma ferramenta de simulação de cenários de mina foi construída e suas principais características serão discutidas. É realizado um estudo do espaço de busca do problema para um entendimento do comportamento das funções objetivo definidas pelo modelo matemático proposto. Finalmente são apresentados os resultados dos experimentos computacionais assim como uma análise estatística comprovando diferenças entre os algoritmos propostos no trabalho. Os experimentos indicam que os algoritmos OPMOP-GA, SPEA2, e NSGAII, utilizando as codificações e operadores propostos, tem grande potencial de encontrar soluções eficientes para o problema. Já o algoritmo ILS+VND, não apresenta bons resultados em termos de convergência para soluções presentes na fronteira pareto do problema, contudo, apresenta tempo computacional inferior aos demais.
“A Maximum Independent Set Approach for Collusion Detection in Voting Pools”
CISUC-SSE (Senior Member)
From agreement problems to replicated software execution, we frequently find scenarios with voting pools. Unfortunately, Byzantine adversaries can join and collude to distort the results of an election. We address the problem of detecting these colluders, in scenarios where they repeatedly participate in voting decisions. We investigate different malicious strategies, such as naïve or colluding attacks, with fixed identifiers or in whitewashing attacks. Using a graph-theoretic approach, we frame collusion detection as a problem of identifying maximum independent sets. We then propose several new graph- based methods and show, via analysis and imulations, their effectiveness and practical applicability for collusion detection.
“Ensuring QoE in Wireless Networks through Adaptive FEC-based Mechanisms”
Online video transmissions over wireless networks are rising in popularity and have already become part of our daily life. In the meantime, it is necessary to address a number of challenges ranging from the scarce resources, time-varying, and high error rates, to the fluctuating bandwidth, unveiling the need for an adaptive mechanism to ensure a good video transmission. Adaptive Forward Error Correction (FEC) techniques with Quality of Experience (QoE) assurance are appropriate to deliver QoE-aware video data to wireless users in dynamic and high error rates networks. Our proposal is an adaptive Video-aware FEC and Fuzzy Logic-based mechanism to shield real-time video transmissions against packet loss in wireless networks, improving both user experience and the usage of resources.
“Emotion-based Analysis and Classification of Audio Music”
Over the last decades, technological advances created an explosion of the available digital music, hindering the access to such information. Based on the fact that music’s preeminent functions are social and psychological, research in retrieval indexes that facilitate searching in conformity with this is now gaining attention. Typically, such indexes will focus on stylistic, mood, and similarity information.
In this colloquium, we will discuss the challenge of automatically recognizing emotional content in audio music signals, known as Music Emotion Recognition (MER). To this end, we study the relations between musical characteristics and emotions and how to classify distinct emotions. Furthermore, the limitations of existent approaches are presented, proposing possible solutions to these problems. Therefore, our recent work is focused in improving the existent MER ground truth, studying the combination of additional information sources such as lyrics and MIDI and developing novel ways to extract meaningful information from audio signals. In addition, related work being carried (by MIR@DEI) will be also briefly presented.
“Evolution of Bio-Inspired Algorithms using Hyper-Heuristics”
Evolutionary Algorithms (EAs) are problem solvers inspired by nature. The effectiveness of these methods usually depends on a non-trivial manual crafting and tailoring of their main components and settings for each specific task.
Hyper-Heuristics (HH) is a recent area of research that aims to overcome this limitation by advocating the automation of the optimization algorithm design task. In this colloquium, we will propose a Grammatical Evolution (GE) framework to automatically design evolutionary algorithms. We will present an empirical study on how different training conditions influence the evolved algorithms. Furthermore we present results displaying that the evolved strategies are competitive with the state-of-art algorithms to the problems being tackled.
“The CockpitCI project: cyber-security for critical infrastructures”
Originally isolated by design, Critical Infrastructures (CI) based on Industrial Control Systems (ICS) – such as SCADA (Supervisory Control and Data Acquisition) systems - were born within the scope of industrial process control technologies. Having evolved from proprietary systems, ICS eventually started adopting open architectures and standards, becoming increasingly interconnected with existing corporate networking infrastructures and even the Internet.
However, as these systems overcame their isolation and moved towards interconnected topologies, they also became more exposed to threats that weren’t even remotely conceivable when they were first designed. Particularly, cyber-threats are one of the most significant problems that modern ICS face, as the shortcomings and vulnerabilities of the decade-old ICS technology – some of them known for a long time, but mostly downplayed due to the isolation of such systems – become serious threats that can ultimately compromise human lives.
As the security needs ICT and ICS domains cannot be addressed in the same way, this calls for a domain-specific approach to cyber threat detection, designed from scratch to address its particular needs. It must consider the particular characteristics of each networking context, be it ICS or ICT, in order to provide real-time cyber-security awareness for the security teams operating in the control room – this is one of the most important contributions of the CockpitCI FP7 project (http://CockpitCI.eu), which aims at improving the resilience and dependability of CIs.
In this talk we will present and discuss the CockpitCI cyber-detection and analysis architecture, which is being jointly developed with the Laboratory of Communications and Telematics of the Centre for Informatics and Systems of the UC.
“Classificação de Padrões e Aplicações em Medicina”
Carlos Eduardo Pedreira
COPPE/Universidade Federal do Rio de Janeiro
Nas últimas décadas assistiu-se a introdução de equipamentos fundamentais ao diagnóstico e acompanhamento de doenças como o citômetro de fluxo, ressonância magnética etc. Estes equipamentos, hoje usados rotineiramente, aumentaram de forma exponencial a capacidade de se gerar dados, introduzindo assim um novo desafio que possivelmente norteará boa parcela do progresso em medicina nos próximos anos: ‘como processar, de forma inteligente, toda esta informação que está sendo gerada’. Por outro lado, as dificuldades intrínsecas geradas por problemas advindos da medicina criam uma necessidade permanente de desenvolvimentos de novas ferramentas de processamento e extração de informação de dados. Neste seminário abordaremos algumas questões importantes da área de Classificação Padrões, como por exemplo métodos para projeção e visualização em 2-D e aplicações recentes em medicina, em especial as associadas ao diagnóstico e acompanhamento de leucemias e linfomas.
“Towards a Better Automatic Optimization of Parallel Programs”
Processor clock frequency is no longer increasing every year. Instead, computers now have multicore processors and frequently a co-processor, such as the GPU. In order to take advantage of this hardware to achieve better performance, programs have to be parallelized. Currently, writing a fast parallel program is hard, requires expert knowledge, manual tuning and trial-and-error experiments. As an alternative, there exist parallel languages, constructs, libraries and compilers that extract parallelism automatically, but are unable to optimize the program as well as an expert.
This talk will cover the work being done towards an automatic and effective optimization of parallel programs. It will focus on controlling the granularity of programs, creating the necessary number of tasks to make use of the hardware threads, but not too much that adds unnecessary overhead in context-switching. Another focus of the work is balancing choosing to execute a certain program on the CPU or GPU dynamically according to the input data size. Finally, the talk will also present some ideas for future work on this area.
“Towards incremental algorithms for hypervolume-based selection in EMO”
Indicator-based Evolutionary Algorithms (EAs) are currently the state-of-the-art in Evolutionary Multiobjective Optimization (EMO). The term indicator-based reflects the use of set-quality indicators, which measure population quality as a whole taking individual quality and population diversity into account, to perform selection.
The Hypervolume Indicator is a popular quality indicator used in such EAs. In particular, it is used in SMS-EMOA, a steady-state EMO algorithm where individuals are evaluated based on how much they contribute to the population hypervolume. At each generation, the individual that contributes the least is replaced by a new individual. Although the Hypervolume Indicator has good theoretical properties, its computation becomes computationally expensive as the number of objectives increases, and determining the least hypervolume contributor at each iteration remains an SMS-EMOA performance bottleneck.
The main goal of this work is the development of incremental algorithms to efficiently compute and update hypervolume contributions. Based on an existing O(n log n) algorithm to compute all hypervolume contributions in three dimensions, an algorithm that allows incremental updates to be performed in linear time is developed. Moreover, an O(n^2)-time algorithm to compute all hypervolume contributions in four dimensions is also presented, and the impact of the proposed algorithms on SMS-EMOA performance is evaluated.
“Learning from Crowds”
Learning from multiple annotators and crowds took a valuable step towards modelling data that does not fit the usual single annotator setting. However, multiple annotators sometimes offer varying degrees of expertise. When disagreements arise, the establishment of the correct label through trivial solutions such as majority voting may not be adequate, since without considering heterogeneity in the annotators, we risk generating a flawed model.
This talk will discuss probabilistic approaches for learning from multiple annotators in different problem settings, namely, linear classification, sequence labelling and non-linear classification. Furthermore, we will also discuss an active learning strategy, which allows to increase the model's performance while reducing the annotation cost.
“Virtual Reality and EEG for social studies and rehabilitation in autism”
Autism is a neurodevelopmental disorder affecting around 1% of the population, following an exponential growth for the last two decades. It is related with deficits in communication, social interaction and behavior and interests. Having no cure discovered yet, rehabilitation represents a way of improving quality of life and autonomy for individuals with autism. However, the costs of having a personal therapist dedicated to someone are big and limit the duration and number of rehab sessions, as well as the general access to those treatments by poorer individuals.
Serious games appear as a possible mitigation for this issue. Creating simulations of real-life tasks, individuals can train independently and have their performance followed by the therapists.
Furthermore, electroencephalography (EEG) can be used to study how the brain of individuals with autism differs from the normally developed ones when performing daily activities and interactions. Ultimately, users can try to control their own brain electrical activity, through a neurofeedback approach.
This talk will cover our work and projects on these topics over the last few years.
“On Multi-objective Sequence Alignment”
Sequence alignment is used to identify regions of similarity that may indicate functional, structural and/or evolutionary relationships between two or more biological sequences. Usually, the quality of an alignment is given by a score function that is a weighted sum of several components, such as the number of matches, mismatches, indels and/or gaps in the alignment. However, it is not clear how to set up the weights in advance. Moreover, several choices for the weights may give rise to different alignments.This problem can be solved by multi-objective sequence alignment. In this talk, we will explain the multi-objective pairwise sequence alignment problem and the extensions of dynamic programming algorithms for several problem variants with a novel pruning technique that efficiently reduces the number of states in processing. Moreover, we will present an experimental analysis with real-life data of benchmark Balibase 3.0. In future work, we will discuss multi-objective multiple sequence alignment.
“Volunteered Geographic Information and Map Validation”
The growing availability of Volunteered Geographical Information (VGI), in projects such as the Geo-Wiki project, Degree Confluence Project, OpenStreetMap, Flickr or Panoramio, presents a huge potential that may radically change several aspects of mapping. The European Union recognized the great potential of this source of information and approved in 2012 two COST actions dedicated to the study of several aspects of VGI (actions TD1202 - Mapping and the citizen sensor, and IC1203 - European Network Exploring Research into Geospatial Information Crowdsourcing: software and methodologies for harnessing geographic information from the crowd).
Much information is being collected and research is under development to determine the real potential of VGI. One fundamental aspect to be addresses in relation to VGI is map quality, since to be of value a map must be accurate and up-to-date. This topic includes several aspects that need addressing, namely: 1) the quality of VGI, since this type of information has varied quality and trust levels; 2) the usability and limitations of the use of VGI for map production and 3) the usability and limitations of the use of VGI to validate maps, such as Land Cover Maps. This presentation aims at making an overview of these topics and identify priority areas of research regarding VGI and map validation.
“Visual Evolutionary Framework for Cluster Analysis of DNA Microarray Data”
This research proposes an evolutionary framework where a method of hierarchical clustering represented by an evolutionary model, a set of cluster validation measures and a cluster visualization tool have been fused to create a suitable environment for knowledge discovery from DNA microarray data. On one hand, the clustering evolutionary model of the framework is a novel alternative that attempts to solve some of the problems faced by the existing clustering methods. On the other hand, the alternative of cluster visualization given by a tool couples new visual components, allowing us to validate and analyze clustering results. It also allows a visual checking environment for cluster validation measures.
“Wireless Sensor Networks in Noisy and Interference Environments: Problem and Solution”
Wireless Sensors Network (WSNs) have countless number of applications in almost all of the fields including military, industrial, healthcare, and smart home environments. However, there are several problems that prevent the widespread of sensor networks in real situations. Among them, the reliability of communication especially in noisy industrial environments is difficult to guarantee. Because sensor nodes are usually deployed and kept unattended without human intervention for a long duration, e.g. months or even years, it is a crucial requirement to provide the reliable communication for the WSNs. However, many problems arise during packet transmission and are related to the transmission medium (e.g. signal path-loss, noise and interference). In addition, the effects of these factors might be not the same on different channels of a standard compliant device. In this presentation, we present a study of quality of channels of IEEE 802.15.4 compliant sensor networks at different environments. More importantly, it presents a proposed solution for providing the reliable communication for sensor networks in noisy and interference environments.
“Positron emission tomography”
Positron emission tomography (PET) is a imaging technique that allows to measure the distribution of the radiotracer administered to a subject. Several improvements related to the resolution and sensitivity of the scanners allowed to expand the applications of the Positron Emission Tomography. This technique plays a key role in oncology, neurology, cardiology, preclinical imaging and in drug design and development.
This presentation addresses the problem of PET image formation using recent techniques such as resolution modelling (RM) and the multiscale/multiframe (MS/MF) reconstruction approach.
The presentation begins with a quick overview about the different stages of a PET exam as well as its main applications. We then give an introduction to the main ideas behind the PET data acquisition, correction and reconstruction. Finally, we present a RM reconstruction technique adapted to the geometry of a non-cylindrical scanner and also the novel MS/MF algorithm. The proposed multiscale reconstruction approach aims to generate near real time images in the scale that is most suitable to the data statistics available at a given accumulated time frame.
“Survey of IoT and its Feasibility Analysis for People with Special Needs”
Nuno Vasco Lopes
One of the main purpose of Internet of Things (IoT) is to provide a better life for people, this is particular important for people who need more support due to their disabilities. In people with disability the improvements in their life conditions that will emerge from IoT is even more evident. IoT can help this kind of people with special needs increase their social live giving the assistance they needed in their daily activities.
As we believe that IoT technology can significantly improve the live of people with special needs, giving them the autonomy and independence they never have, in this work we try to make an exhaustive IoT survey for people with special needs. We start by making an IoT taxonomy, then we give thorough review and insightful comments about the current state of the art of the IoT architectures. After that, we analyse IoT requirements for people with special needs. Afterward we stress out same IoT challenges and a specific IoT architecture for this sort of people.
The propose IoT architecture for people with special needs will be aligned with the current IoT standards and specifications. The architecture tries to encompasses healthcare stakeholders perspective views and whole scientific requirements collection that are demanded for this sort of application.
In addition to IoT taxonomy and IoT architecture for people with special need, will be also presented same considerations and challenges that this paticular application domain will face in the future.
“A diagnosis based approach to improving intrusion detection technologies”
In this research work we have analyzed the state of the art of security in two very relevant application domains. Firstly we have evaluated the effectiveness of the intrusion detection technologies for web applications. By exploiting a very powerful attack injection technique we have performed an experimental evaluation of security tools for the detection of SQL Injection attacks, highlighting the very low effectiveness achieved by these tools. Then we have performed a security assessment of a very relevant data collection technology for smart electric grids. Our analysis has shown that vulnerabilities affecting this novel and cutting edge technologies for Critical Infrastructures are in most cases very similar to those affecting classic IT systems, such as web applications. Thus, after having reviewed the limitations of the currently available security solutions we have designed and implemented a novel approach - which we have called Intrusion Detection & Diagnosis System (ID2S) technology - to overcome such limitations. The basic idea is to collect information at several architectural levels, using multiple security probes, which are deployed as a distributed architecture, to perform a sophisticated correlation analysis of intrusion symptoms. This makes it possible to escalate from intrusion symptoms to the adjudged cause of the intrusion, and to assess the damage in individual system components. We also present experimental results showing the improvement of effectiveness that is achieved with this approach. Finally we have proposed another enhancement for Intrusion Detection technologies aimed at improving the protection of Critical Infrastructures. This proposed approach is based on the convergence, in one software solution, of both logical security and physical security events so to create an enhanced situational awareness. In a very preliminary analysis the proposed approach is evaluated against a real Critical Infrastructure scenario: the control system of a dam.
“An integrated approach for the design and operation of Demand Responsive Transportation services”
Providing quality public transportation can be extremely expensive when demand is low, variable and unpredictable. Demand Responsive Transportation (DRT) systems try to address these issues with routes and frequencies that may vary according to observed demand. The design and operation of DRTs involve multiple criteria and have a combinatorial nature that prevents the use of traditional optimization methods. We have developed an innovative Decision Support System (DSS) integrating simulation and optimization, to help design and operate DRT services, minimizing operating costs and maximizing the service quality. Experiments inspired in real problems have shown the potential of this DSS.