Made available in DSpace on 2016-03-15T19:37:32Z (GMT). No. of bitstreams: 1 Samuel Fontes Lima.pdf: 2071921 bytes, checksum: 2d66276b3ab526cbf68f0f727c2e1f3d (MD5) Previous issue date: 2010-08-18 / Nowadays, object oriented languages are among the most used ones. Recent studies with beginners have shown that the object oriented programming language (OOP) concepts are not so easy to learn. Students have difficulties not only in understanding those OOP concepts, but also in applying them to solving problems. Therefore, an Intelligent Tutoring System (ITS) is suggested in order to help in the learning process of these concepts. One of the most important challenge for ITS development is the individual treatment of student, which is mainly obtained by system adaptativity to the characteristics of each student. This adaptativity to student is a complex issue and the focus of several researches. It encloses several areas of personality such as: learning style, intelligence, previous knowledge, student background and emotions. Thus, the student model is extremely important because all individual information of student is in it. It is here where Artificial Intelligence (AI) techniques have been employed. This paper approaches the adaptativity issue of tutoring system to student knowledge through a student model based on Bayesian Network (BN). Taking into consideration that the student knowledge level is an inaccurate information, and even professors deal with this uncertainty, the Bayesian Networks (BNs), also known as Belief Network, have been chosen. They are considered proper for uncertain spheres for employing the probabilistic reasoning, which allows to identify a certain belief degree of student knowledge level. Therefore, the BNs have been employed in ITS in the inferences concerning student behavior as well as in the decision-making process concerning tutor actions. / Atualmente as linguagens orientadas a objeto encontram-se entre as mais utilizadas. Estudos recentes com alunos iniciantes demonstram que os conceitos de programação orientada a objetos (POO) não são tão fáceis de assimilar. Os estudantes encontram dificuldades não somente no entendimento dos conceitos de POO, mas também na aplicação deles na resolução de problemas. Diante disso, propõe-se o emprego de um Sistema Tutor Inteligente (STI) para auxiliar no processo de aprendizagem desses conceitos. Um dos principais desafios para o desenvolvimento de um STI é o tratamento individualizado do estudante, que é obtido principalmente por meio da adaptatividade do sistema às características de cada aprendiz. A adaptatividade ao aprendiz é uma questão complexa, foco de várias pesquisas, abrange várias características da personalidade: estilo de aprendizagem, inteligência, conhecimento anterior, histórico do aprendiz e as emoções. Dessa forma, o modelo do estudante é de fundamental importância, pois contém as informações individuais do aprendiz. É nesse ponto que as técnicas de Inteligência Artificial (IA) têm sido empregadas. Este trabalho aborda a questão da adaptatividade do sistema tutor ao conhecimento do estudante por meio de um modelo de aprendiz baseado em Rede Bayesiana. Considerando-se que o nível de conhecimento do aprendiz é uma informação imprecisa, e que até mesmo professores lidam com essa incerteza, optou-se pela utilização de Redes Bayesianas (RBs), também chamadas de Redes de Crença, que são consideradas adequadas para ambientes sob incerteza pois empregam o raciocínio probabilístico, o qual permite identicar um certo grau de crença sobre o nível de conhecimento do aprendiz. Devido a isso, RBs têm sido empregadas em STIs nas inferências sobre o comportamento do aprendiz e nas tomadas de decisões sobre as ações do tutor.

The international maritime industry is growing fast due to an increasing number of transportations over sea. In pace with this development, the maritime surveillance capacity must be expanded as well, in order to be able to handle the increasing numbers of hazardous cargo transports, attacks, piracy etc. In order to detect such events, anomaly detection methods and techniques can be used. Moreover, since surveillance systems process huge amounts of sensor data, anomaly detection techniques can be used to filter out or highlight interesting objects or situations to an operator. Making decisions upon large amounts of sensor data can be a challenging and demanding activity for the operator, not only due to the quantity of the data, but factors such as time pressure, high stress and uncertain information further aggravate the task. Bayesian networks can be used in order to detect anomalies in data and have, in contrast to many other opaque machine learning techniques, some important advantages. One of these advantages is the fact that it is possible for a user to understand and interpret the model, due to its graphical nature. This thesis aims to investigate how the output from a Bayesian network can be explained to a user by first reviewing and presenting which methods exist and second, by making experiments. The experiments aim to investigate if two explanation methods can be used in order to give an explanation to the inferences made by a Bayesian network in order to support the operator’s situation awareness and decision making process when deployed in an anomaly detection problem in the maritime domain.

Agent-oriented software engineering is a new software engineering paradigm that is ideally suited to the analysis and design of complex systems. Open distributed environments place a growing demand on complex systems to be adaptive as well. Complex systems that can learn from and adapt to dynamically changing environments are called complex adaptive systems. These systems are characterized by emergent behaviour caused by interactions between system components and the environment. Agent-oriented software engineering methodologies attempt to control emergence during analysis and design by engineering the complex system in such a way that the correct emergent behaviour results during run-time. In a complex adaptive system however, emergent behaviour cannot be predicted during analysis and design, as it evolves only after implementation. By restricting emergent behaviour, as is done in most agent-oriented software engineering approaches, a complex system cannot be fully adaptive as well. We propose the BaBe methodology that will enable a complex system to be adaptive by learning from its environment and modifying its behaviour during run-time. This methodology adds a run-time emergence model consisting of distributed Bayesian behaviour networks to the agent-oriented software engineering lifecycle. These networks are initialised by the human software engineer during analysis and design and deployed by Bayesian agencies (also complex adaptive systems). The Bayesian agents are simple, and collectively they implement distributed Bayesian behaviour networks. These networks, being specialized Bayesian networks, enable the Bayesian agents to collectively mine relationships between emergent behaviours and the interactions that caused them to emerge, in order to adapt the behaviour of the system. The agents are organized into heterarchies of agencies, where each agency activates one or more component behaviour depending on the inference in the underlying Bayesian behaviour network. These agencies assist the human software engineer to bridge the gap between the implementation and the understanding of emergent behaviour in complex adaptive systems. Due to the simplicity of the agents and the minimal communication amongst them, they can be implemented using a commercially available component architecture. We describe a prototype implementation of the Bayesian agencies using Sun’s Enterprise JavaBeans™ component architecture. / Thesis (PhD (Computer Science))--University of Pretoria, 2005. / Computer Science / unrestricted

A segurança no tráfego de carros é um assunto em foco nos dias de hoje e, dentro dele, podem-se citar os sistemas de auxílio ao motorista que vêm sendo desenvolvidos com a finalidade de reduzir o grande número de fatalidades em acidentes de trânsito. Tais sistemas de auxílio buscam mitigar falhas humanas como falta de atenção e imprudência. Visto isso, o projeto SENA, desenvolvido pelo Laboratório de Robótica Móvel da Escola de Engenharia de São Carlos, busca contribuir com a evolução dessa assistência ao motorista. O presente trabalho realiza um estudo sobre uma técnica de inteligência artificial chamada de Redes Bayesianas. Essa técnica merece atenção em virtude de sua capacidade de tratar dados incertos em forma de probabilidades. A rede desenvolvida por esse trabalho utiliza, como dados de entrada, os classificadores em desenvolvimento no projeto SENA e tem como resposta um comportamento que o veículo deve executar, por um ser humano ou por um planejador de trajetórias. Em função da alta dimensionalidade do problema abordado, foram realizados dois experimentos em ambiente simulado de duas situações distintas. A primeira, um teste de frenagem próximo a um ponto de intersecção e a segunda, um cenário de entroncamento. Os testes feitos com a rede indicam que classificadores pouco discriminantes deixam o sistema mais propenso a erros e que erros na localização do ego-veículo afetam mais o sistema se comparado a erros na localização dos outros veículos. Os experimentos realizados mostram a necessidade de um sistema de tempo real e um hardware mais adequado para tratar as informações mais rapidamente / The safety of cars in traffic scenarios is being addressed on the past few years. One of its topics is the Advanced Driver-Assistance Systems which have been developed to reduce the fatality numbers of traffic accidents. These systems try to decrease human failures, such as imprudence and lack of attention while driving. For these reasons, the SENA project, in progress on the Mobile Robotics Laboratory at the Sao Carlos School of Engineering (EESC), aims to contribute for the evolution of these assistance systems. This work studies an artificial intelligence technique called Bayesian Networks. It deserves our attention due to its capability of handling uncertainties with probability distributions. The network developed in this Masters Thesis has, as input, the result of the classifiers used on SENA project and has, as output, a behavior which has to be performed by the vehicle with a driver or autonomously by the means of a path planner. Due to the high dimensionality of this issue, two different tests have been carried out. The first one was a braking experiment near a intersection point and the other one was a T-junction scenario. The tests made indicate that weak classifiers leaves the system more instable and error-prone and localization errors of the ego-vehicle have a stronger effect than just localization errors of other traffic participants. The experiments have shown that there is a necessity for a real-time system and a hardware more suitable to deal quickly with the information

Este trabalho teve como objetivo principal modelar a Rede de Regulação Gênica do organismo modelo Halobacterium salinarum NRC-1, estabelecendo interações entre as entidades da rede por intermédio de experimentos inéditos de interação física: ChIP- *, RIP-* e dRNA-seq. Em contraponto com as abordagens clássicas de construção de redes, que estimam interações através de medições de expressão gênica, este trabalho as estabeleceu exclusivamente de interações físicas, permitindo que a estrutura final seja uma representação mais fiel ao fenômeno físico de regulação gênica, baseando-se nos fundamentos da Biologia Sistêmica. Em vista da abundância de dados públicos de expressão gênica para o organismo e do objetivo primário, um objetivo secundário foi traçado: identificar, computacionalmente, genes de fato controlados pelas interações fornecidas pela nova rede. Para isso, a estrutura estabelecida foi transformada numa Rede Bayesiana, e a identificação de genes foi efetuada através da análise de suas Tabelas de Probabilidade Condicionais. Finalmente, como os resultados obtidos para o objetivo secundário foram desfavoráveis a utilização de Redes Bayesianas, os resultados efetivos deste trabalho foram a criação de uma nova Rede de Regulação Gênica para a H. salinarum e uma análise em torno da efetividade de Redes Bayesianas neste contexto. / The main goal of this work was modeling the gene regulatory network of the model organism Halobacterium salinarum NRC-1, establishing new interactions between networks entities through unpublished physical interaction experiments: ChIP-*, RIP-* e dRNA-seq. Instead of using classical approaches to build network structures that estimates interactions using gene expression data, this work established them exclusively from physical interactions. Therefore, the final structure is a more reliable representation of the physical phenomenon of gene expression, built using the principles of systems biology. Considering the amount of public available gene expression data and the primary goal, another objective was proposed: a computational analysis to detect genes actually controlled by the interactions of the new network. To achieve this goal the established network was transformed in a Bayesian network, detecting genes through the analysis of their conditional probability tables. Lastly, as the results of the secondary goal went against the use of Bayesian networks, the effective results of this thesis were the creation of a new genetic regulatory network for H. salinarum and an analysis around Bayesian networks in this context.

Traditional forecasting models have been widely used for decision-making in production, finance and energy. Such is the case of the ARIMA models, developed in the 1970s by George Box and Gwilym Jenkins [1], which incorporate characteristics of the past models of the same series, according to their autocorrelation. This work compares advanced statistical methods for determining the demand for electricity in Colombia, including the SARIMA, econometric and Bayesian methods.

Das Treffen komplexer medizinischer Entscheidungen wird durch die stetig steigende Menge an zu berücksichtigenden Informationen zunehmend komplexer. Dieser Umstand ist vor allem auf die Verfügbarkeit von immer präziseren diagnostischen Methoden zur Charakterisierung der Patienten zurückzuführen (z.B. genetische oder molekulare Faktoren). Hiermit einher geht die Entwicklung neuartiger Behandlungsstrategien und Wirkstoffe sowie die damit verbundenen Evidenzen aus klinischen Studien und Leitlinien. Dieser Umstand stellt die behandelnden Ärztinnen und Ärzte vor neuartige Herausforderungen im Hinblick auf die Berücksichtigung aller relevanten Faktoren im Kontext der klinischen Entscheidungsfindung. Moderne IT-Systeme können einen wesentlichen Beitrag leisten, um die klinischen Experten weitreichend zu unterstützen. Diese Assistenz reicht dabei von Anwendungen zur Vorverarbeitung von Daten für eine Reduktion der damit verbundenen Komplexität bis hin zur systemgestützten Evaluation aller notwendigen Patientendaten für eine therapeutischen Entscheidungsunterstützung. Möglich werden diese Funktionen durch die formale Abbildung von medizinischem Fachwissen in Form einer komplexen Wissensbasis, welche die kognitiven Prozesse im Entscheidungsprozess adaptiert. Entsprechend werden an den Prozess der IT-konformen Wissensabbildung erhöhte Anforderungen bezüglich der Validität und Signifikanz der enthaltenen Informationen gestellt. In den ersten beiden Kapiteln dieser Arbeit wurden zunächst wichtige methodische Grundlagen im Kontext der strukturierten Abbildung von Wissen sowie dessen Nutzung für die klinische Entscheidungsunterstützung erläutert. Hierbei wurden die inhaltlichen Kernthemen weiterhin im Rahmen eines State of the Art mit bestehenden Ansätzen abgeglichen, um den neuartigen Charakter der vorgestellten Lösungen herauszustellen. Als innovativer Kern wurde zunächst die Konzeption und Umsetzung eines neuartigen Ansatzes zur Fusion von fragmentierten Wissensbausteinen auf der formalen Grundlage von Bayes-Netzen vorgestellt. Hierfür wurde eine neuartige Datenstruktur unter Verwendung des JSON Graph Formats erarbeitet. Durch die Entwicklung von qualifizierten Methoden zum Umgang mit den formalen Kriterien eines Bayes-Netz wurden weiterhin Lösungen aufgezeigt, welche einen automatischen Fusionsprozess durch einen eigens hierfür entwickelten Algorithmus ermöglichen. Eine prototypische und funktionale Plattform zur strukturierten und assistierten Integration von Wissen sowie zur Erzeugung valider Bayes-Netze als Resultat der Fusion wurde unter Verwendung eines Blockchain Datenspeichers implementiert und in einer Nutzerstudie gemäß ISONORM 9241/110-S evaluiert. Aufbauend auf dieser technologischen Plattform wurden im Anschluss zwei eigenständige Entscheidungsunterstützungssysteme vorgestellt, welche relevante Anwendungsfälle im Kontext der HNO-Onkologie adressieren. Dies ist zum einen ein System zur personalisierten Bewertung von klinischen Laborwerten im Kontext einer Radiochemotherapie und zum anderen ein in Form eines Dashboard implementiertes Systems zur effektiveren Informationskommunikation innerhalb des Tumor Board. Beide Konzepte wurden hierbei zunächst im Rahmen einer initialen Nutzerstudie auf Relevanz geprüft, um eine nutzerzentrische Umsetzung zu gewährleisten. Aufgrund des zentralen Fokus dieser Arbeit auf den Bereich der klinischen Entscheidungsunterstützung, werden an zahlreichen Stellen sowohl kritische als auch optimistische Aspekte der damit verbundenen praktischen Lösungen diskutiert.:1 Introduction 1.1 Motivation and Clinical Setting 1.2 Objectives 1.3 Thesis Outline 2 State of the Art 2.1 Medical Knowledge Modeling 2.2 Knowledge Fusion 2.3 Clinical Decision Support Systems 2.4 Clinical Information Access 3 Fundamentals 3.1 Evidence-Based Medicine 3.1.1 Literature-Based Evidence 3.1.2 Practice-Based Evidence 3.1.3 Patient-Directed Evidence 3.2 Knowledge Representation Formats 3.2.1 Logic-Based Representation 3.2.2 Procedural Representation 3.2.3 Network or Graph-Based Representation 3.3 Knowledge-Based Clinical Decision Support 3.4 Conditional Probability and Bayesian Networks 3.5 Clinical Reasoning 3.5.1 Deterministic Reasoning 3.5.2 Probabilistic Reasoning 3.6 Knowledge Fusion of Bayesian Networks 4 Block-Based Collaborative Knowledge Modeling 4.1 Data Model 4.1.1 Belief Structure 4.1.2 Conditional Probabilities 4.1.3 Metadata 4.2 Constraint-Based Automatic Knowledge Fusion 4.2.1 Fusion of the Bayesian Network Structures 4.2.2 Fusion of the Conditional Probability Tables 4.3 Blockchain-Based Belief Storage and Retrieval 4.3.1 Blockchain Characteristics 4.3.2 Relevance for Belief Management 5 Selected CDS Applications for Clinical Practice 5.1 Distributed Knowledge Modeling Platform 5.1.1 Requirement Analysis 5.1.2 System Architecture 5.1.3 System Evaluation 5.1.4 Limitations of the Proposed Solution 5.2 Personalization of Laboratory Findings 5.2.1 Requirement Analysis 5.2.2 System Architecture 5.2.3 Limitations of the Proposed Solution 5.3 Dashboard for Collaborative Decision-Making in the Tumor Board 5.3.1 Requirement Analysis 5.3.2 System Architecture 5.3.3 Limitations of the Proposed Solution 6 Discussion 6.1 Goal Achievements 6.2 Contributions and Conclusion 7 Bibliography

L'inférence de la causalité est une problématique récurrente pour un large éventail de domaines où les méthodes d'interventions ou d'acquisition de données temporelles sont inapplicables. Toutefois, établir des relations de causalité uniquement à partir de données d'observation peut se révéler être une tâche complexe. Je présente ici une méthode d'apprentissage de réseaux qui combine les avantages des méthodes d'inférence par identification de contraintes structurales et par optimisation de scores bayésiens pour reconstruire de manière robuste des réseaux causaux malgré le bruit d'échantillonnage inhérent aux données d'observation. Cette méthode repose sur l'identification de v-structures à l'aide de l'information (conditionnelle) à trois variables, une mesure issue de la théorie de l'information, qui est négative quand elle est associée à un collider et positive sinon. Cette approche soustrait itérativement l'information conditionnelle à trois variables la plus forte à l'information conditionnelle à deux variables entre chaque paire de noeuds. Les indépendences conditionnelles sont progressivement calculées en collectant les contributions les plus fortes. Le squelette est ensuite partiellement orienté et ces orientations sont propagées aux liens non orientés selon le signe et la force de l'interaction dans les triplets ouverts. Cette approche obtient de meilleurs résultats que les méthodes par contraintes ou optimisation de score sur un ensemble de réseaux benchmark et fournit des prédictions prometteuses pour des systèmes biologiques complexes, tels que les réseaux neuronaux du poisson zèbre ou l'inférence des cascades de mutations dans les tumeurs. / The inference of causality is an everyday life question that spans a broad range of domains for which interventions or time-series acquisition may be impracticable if not unethical. Yet, elucidating causal relationships in real-life complex systems can be convoluted when relying solely on observational data. I report here a novel network reconstruction method, which combines constraint-based and Bayesian frameworks to reliably reconstruct networks despite inherent sampling noise in finite observational datasets. The approach is based on an information theory result tracing back the existence of colliders in graphical models to negative conditional 3-point information between observed variables. This enables to confidently ascertain structural independencies in causal graphs, based on the ranking of their most likely contributing nodes with (significantly) positive conditional 3-point information. Dispensible edges from a complete undirected graph are progressively pruned by iteratively taking off the most likely positive conditional 3-point information from the 2-point (mutual) information between each pair of nodes. The resulting skeleton is then partially directed by orienting and propagating edge directions based on the sign and magnitude of the conditional 3-point information of unshielded triples. This new approach outperforms constraint-based and Bayesian inference methods on a range of benchmark networks and provides promising predictions when applied to the reconstruction of complex biological systems, such as hematopoietic regulatory subnetworks, zebrafish neural networks, mutational pathways or the interplay of genomic properties on the evolution of vertebrates.

Despite tremendous progress in the understanding of breast cancer (BC), gaps remain in our knowledge of the molecular basis underlying the aggressiveness of BC and BC disparities. Nuclear respiratory factor 1 (NRF1) is a transcription factor (TF) known to control breast cancer cell cycle progression. DNA response elements bound by NRF1 positively correlate with the progression of malignant breast cancer. Mechanistic aspects by which NRF1 contributes to susceptibility to different breast tumor subtypes are still not fully understood. Therefore, the primary objective of this dissertation was to decipher mechanisms by which NRF1 coordinates changes in the transcriptional and chromatin landscape affecting development and progression of invasive breast cancer. Our hypothesis was that NRF1 reprogramming the transcription of tumor initiating gene(s) and tumor suppressor gene(s) contribute in the development and progression of invasive breast cancer. To test this hypothesis, we proposed three specific aims: (a) Decipher regulatory landscape of NRF1 networks in breast cancer. (b) Determine the role of NRF1 gene networks in different subtypes of breast cancer. (c) Determine differential NRF1 gene network sensitivity contributing to breast cancer disparities. Our approach to test these aims consisted of a systematic integration of ChIP DNA-seq, RNA-Seq, NRF1 protein-DNA motif binding, signal pathway analysis, and Bayesian machine learning. We uncovered a novel oncogenic role for NRF1. This discovery strongly supported the supposition that NRF1 overexpression is sufficient to derive breast tumorigenesis. We also observed new roles for NRF1 in the acquisition of breast tumor initiating cells, regulation of epithelial to mesenchymal transition (EMT), and invasiveness of BC stem cells. Furthermore, through the use of Bayesian network structure learning we found that the NRF1 motif was enriched in 14 associated with HER2 amplified breast cancer. Three genes—GSK3B, E2F3, and PIK3CA—were able to predict HER2 breast tumor status with 96% to100% confidence. The findings of this study also showed the roles of NRF1 sensitivity to development of lobular A, Her2+, and TNBC in different racial/ethnic groups of breast cancer patients. In summary, our study revealed for the first time the role of NRF1 in the pathogenesis of invasive BC and BC disparities.

No description available.