Esta dissertação apresenta quatro estudos independentes, do tipo quasi-experimental com aplicação de pré-teste e pós-testes e grupo de controle não equivalente. O objetivo dos quatro estudos é investigar como uma ferramenta baseada em mineração de texto e apresentação gráfica multirrepresentacional (SOBEK) pode contribuir com o processo de construção de conceitos científicos. Textos do tipo refutacional oriundos de pesquisas internacionais já publicadas foram utilizados nos estudos, assim como os testes de desempenho correspondentes a cada texto. Os conceitos abordados são energia, evolução Darwiniana e a natureza particulada da matéria. Realizados em uma escola municipal da região metropolitana de Porto Alegre, os estudos contaram com a participação de 73 alunos distribuídos em duas turmas de 8º ano e duas turmas de 9º ano. As análises estatísticas dos escores registrados pelos alunos nos testes identificaram uma melhora significativa no desempenho dos alunos de 9º ano que utilizaram a ferramenta SOBEK como apoio para o estudo do texto. Para os alunos de 8º ano, não houve diferenças estatisticamente significativas entre os alunos que usaram a ferramenta e os que responderam questionários sobre o texto. Estudos de maior profundidade são necessários para identificar a ocorrência e extensão de possíveis mudanças conceituais promovidas pela abordagem proposta. / This work presents four independent studies, in a quasi-experimental design with application of pre-test and post-tests with non-equivalent control groups. The set of four studies aim to investigate how a tool based on text mining and multirrepresentacional graphical presentation (SOBEK) can contribute to the process of constructing scientific concepts. Refutational texts from international published researches were used in the studies, as well as the corresponding performance´s tests about each text. The concepts covered are energy, Darwinian evolution and the particulate nature of matter. Performed in a municipal school in the Porto Alegre´s metropolitan region, in Southern Brazil, the studies included 73 students divided into two classes of 7th grade and two classes of 8th grade. Statistical analysis of the scores performed by students in the tests show a significant improvement in student´s performance in 8th grade that used the SOBEK tool as support for the study of the text. For students in grade 7th, there were no statistically significant differences in the performances of students who used the SOBEK tool and those who answered questionnaires about the text. More detailed studies are needed to identify the occurrence and the extent of possible conceptual changes supported by the proposed approach.

Made available in DSpace on 2014-06-12T15:52:36Z (GMT). No. of bitstreams: 1 license.txt: 1748 bytes, checksum: 8a4605be74aa9ea9d79846c1fba20a33 (MD5) Previous issue date: 2009 / Devido á grande demanda por softwares cada vez mais robustos, complexos e flexíveis, e, sobretudo, pelo curtíssimo tempo de entrega exigido, a engenharia de software tem procurado novos meios de desenvolvimento que supram satisfatoriamente essas demandas. Uma forma de galgar esses novos patamares de produtividade provém do uso de uma metodologia baseada em agentes que se comunicam e com isso, ao invés dos programas serem estritamente programados, o comportamento destes sistemas de software emerge da interação de agentes, robôs, ou subsistemas aut onomos, independentes, além de declarativamente especificados. Isto provê a habilidade para automaticamente configurá -los, otimizá-los, monitorá-los, adaptá-los, diagnosticá-los, repará-los e protegê-los dentro do ambiente. Contudo, um grande problema das linguagens declarativas é a falta de mecanismos que permitem a melhor estruturação de dados, facilitando portanto, o reuso. Portanto, esta dissertação explica o desenvolvimento de nova linguagem lógica declarativa para programar sistemas de raciocínio automático de uma forma modularizada: C2HR∨. A linguagem base escolhida para a extensão com componentes lógicos foi CHR. Os motivos para essa escolha são definidos ao longo da dissertação. Duas abordagens, portanto, são apresentadas: a primeira, conhecida como CHRat, foi desenvolvida numa parceria juntamente com o grupo de pesquisas CONTRAINTES do INRIA/Rocquencourt-Paris, onde o programador ´e o responsável direto por definir os componentes CHR, permitindo o seu reuso por outros componentes; a segunda aplicação, CHRtp, visa atender prioritariamente requisitos de completude e, por isso, se baseia em procedimentos lógicos de inferência como: o raciocínio para frente, o raciocínio para trás, e a resolução/factoring. A dissertação mostra também alguns exemplos práticos, onde uso de componentes facilita radicalmente sua implementação. As contribuições almejadas com essa dissertação são: a definição de uma família bem formalizada de provadores de teoremas automáticos, que podem trabalhar com sentenças especificadas em lógica horn ou em lógica de primeira ordem, a extensão de CHR como uma linguagem modular de propósito geral, a melhor estruturação de bases conhecimentos e até o uso em conjunto de bases heterogêneas, a definição de uma linguagem para a fácil e direta estruturação de dados por meio de componentes, dentre outras

Gradient-based adversarial attacks on neural networks threaten extremely critical applications such as medical diagnosis and biometric authentication. These attacks use the gradient of the neural network to craft imperceptible perturbations to be added to the test data, in an attempt to decrease the accuracy of the network. We propose a defense to combat such attacks, which can be modified to reduce the training time of the network by as much as 71%, and can be further modified to reduce the training time of the defense by as much as 19%. Further, we address the threat of uncertain behavior on the part of the attacker, a threat previously overlooked in the literature that considers mostly white box scenarios. To combat uncertainty on the attacker's part, we train our defense with an ensemble of attacks, each generated with a different attack algorithm, and using gradients of distinct architecture types. Finally, we discuss how we can prevent the attacker from breaking the defense by estimating the gradient of the defense transformation.

Knowledge representation learning (KRL) has been a long-standing and challenging topic in artificial intelligence. Recent years have witnessed the rapidly growing research interest and industrial applications of KRL. However, two important aspects of KRL remains unsatisfactory in the academia and industries, i.e., the quality and the generalization capabilities of the learned representations. This thesis presents a set of methods target at learning high quality distributed knowledge representations and further empowering the learned representations for more general reasoning tasks over knowledge bases. On the one hand, we identify the false negative issue and the data sparsity issue in the knowledge graph completion (KGC) task that can limit the quality of the learned representations. Correspondingly, we design a ranking-based positive-unlabeled learning method along with an adversarial data augmentation strategy for KGC. Then we unify them seamlessly to improve the quality of the learned representations. On the other hand, although recent works expand the supported neural reasoning tasks remarkably by answering multi-hop logical queries, the generalization capabilities are still limited to inductive reasoning tasks that can only provide entity-level answers. In fact, abductive reasoning that provides concept-level answers to queries is also in great need by online users and a wide range of downstream tasks. Therefore, we design a joint abductive and inductive knowledge representation learning and reasoning system by incorporating, representing, and operating on concepts. Extensive experimental results along with case studies demonstrate the effectiveness of our methods in improving the quality and generalization capabilities of the learned distributed knowledge representations.

Indiana University-Purdue University Indianapolis (IUPUI) / Medication non-adherence is a major healthcare problem that negatively impacts the health and productivity of individuals and society as a whole. Reasons for medication non-adherence are multi-faced, with no clear-cut solution. Adherence to medication remains a difficult area to study, due to inconsistencies in representing medicationadherence behavior data that poses a challenge to humans and today’s computer technology related to interpreting and synthesizing such complex information. Developing a consistent conceptual framework to medication adherence is needed to facilitate domain understanding, sharing, and communicating, as well as enabling researchers to formally compare the findings of studies in systematic reviews. The goal of this research is to create a common language that bridges human and computer technology by developing a controlled structured vocabulary of medication adherence behavior—“Medication Adherence Behavior Ontology” (MAB-Ontology) using breast cancer as a case study to inform and evaluate the proposed ontology and demonstrating its application to real-world situation. The intention is for MAB-Ontology to be developed against the background of a philosophical analysis of terms, such as belief, and desire to be human, computer-understandable, and interoperable with other systems that support scientific research. The design process for MAB-Ontology carried out using the METHONTOLOGY method incorporated with the Basic Formal Ontology (BFO) principles of best practice. This approach introduces a novel knowledge acquisition step that guides capturing medication-adherence-related data from different knowledge sources, including adherence assessment, adherence determinants, adherence theories, adherence taxonomies, and tacit knowledge source types. These sources were analyzed using a systematic approach that involved some questions applied to all source types to guide data extraction and inform domain conceptualization. A set of intermediate representations involving tables and graphs was used to allow for domain evaluation before implementation. The resulting ontology included 629 classes, 529 individuals, 51 object property, and 2 data property. The intermediate representation was formalized into OWL using Protégé. The MAB-Ontology was evaluated through competency questions, use-case scenario, face validity and was found to satisfy the requirement specification. This study provides a unified method for developing a computerized-based adherence model that can be applied among various disease groups and different drug categories.

Indiana University-Purdue University Indianapolis (IUPUI) / With the rise of big data and the demands for leveraging artificial intelligence (AI), healthcare requires more knowledge sharing that offers machine-readable semantic formalization. Even though some applications allow shared data interoperability, they still lack formal machine-readable semantics in ICD9/10 and LOINC. With ontology, the further ability to represent the shared conceptualizations is possible, similar to SNOMED-CT. Nevertheless, SNOMED-CT mainly focuses on electronic health record (EHR) documenting and evidence-based practice. Moreover, due to its independence on data quality, the ontology enhances advanced AI technologies, such as machine learning (ML), by providing a reusable knowledge framework. Developing a machine-readable and sharable semantic knowledge model incorporating external evidence and individual practice’s values will create a new revolution for best practice medicine. The purpose of this research is to implement a sharable ontology for the best practice in healthcare, with anterior cruciate ligament reconstruction (ACLR) as a case study. The ontology represents knowledge derived from both evidence-based practice (EBP) and practice-based evidence (PBE). First, the study presents how the domain-specific knowledge model is built using a combination of Toronto Virtual Enterprise (TOVE) and a bottom-up approach. Then, I propose a top-down approach using Open Biological and Biomedical Ontology (OBO) Foundry ontologies that adheres to the Basic Formal Ontology (BFO)’s framework. In this step, the EBP, PBE, and statistic ontologies are developed independently. Next, the study integrates these individual ontologies into the final ACLR Ontology (ACLRO) as a more meaningful model that endorses the reusability and the ease of the model-expansion process since the classes can grow independently from one another. Finally, the study employs a use case and DL queries for model validation. The study's innovation is to present the ontology implementation for best-practice medicine and demonstrate how it can be applied to a real-world setup with semantic information. The ACLRO simultaneously emphasizes knowledge representation in health-intervention, statistics, research design, and external research evidence, while constructing the classes of data-driven and patient-focus processes that allow knowledge sharing explicit of technology. Additionally, the model synthesizes multiple related ontologies, which leads to the successful application of best-practice medicine.

No description available.

<p>  </p> <p>The abundance of heterogeneous data produced and collected each day via multimodal sources may contain hidden events of interest, but in order to extract them the streams of data need to be analyzed with appropriate algorithms, so these events are presented to the end user at the right moment and at the right time. This dissertation proposes a series of algorithms that shape a comprehensive framework for situational knowledge on demand to address this problem. The framework consists of several modules and approaches, each of them is presented in a separate chapter: I begin with video data analysis in streaming video and video at rest for enhanced object detection of real-life surveillance video. For detecting the rare events of interest, I develop a semantic video analysis algorithm which uses an overlay knowledge graph and a semantical network. I show that the usage of the external knowledge for understanding the semantic analysis outperforms other techniques such as transfer learning. </p> <p>The semantical outliers can be used further for improving the algorithm of detecting new objects in the stream of different modalities. I extend the framework with additional modules for natural language data and apply the extended version of the semantic analysis algorithm to define the events of interest from multimodal streaming data. I present a way of combining several feature extractors which can be extended to multiple heterogeneous streams of data in order to efficiently fuse the data based on its semantical similarity, and then show how the serverless architecture of the framework outperforms conventional cloud software architecture. </p> <p>Besides detecting the rare and semantically incompatible events, the semantic analysis can be used for improving the neural networks performance with the data augmentation. The algorithm presented for augmenting the data with the potentially novel objects to circumvent the data drift problem uses the knowledge graph and generative adversarial networks to present the objects to augment the training datasets for supervised learning. I extend the presented framework with a pipeline for generating synthetic novelties to improve the performance of feature extractors and provide the empirical evaluation of the developed method.</p>

No description available.

Fast moving technologies have caused high-tech components to have shortened life cycles, rendering them obsolete quickly. Technology obsolescence creates significant problems for product sectors that use components that are only available for a short period of time for manufacture and maintenance of long field-life systems. Technology obsolescence can make design changes of systems prohibitively expensive, and results in high life cycle costs of systems. While the impact and pervasiveness of obsolescence problems are growing, existing tools and solutions are lacking the needed information and knowledge to do much more than focus on reactively managing obsolescence. Current methods and tools are limited by data conflicts and data inexplicitness, incompleteness, and inconsistency. In response to the drawbacks of current tools, comprehensive knowledge representation that allows information sharing, reuse, and collaboration on obsolescence issues across different organizations is required. Further, decision making tools that can support proactive and strategic obsolescence management are needed. The purpose of this research is to establish an ontology-based knowledge representation scheme for information sharing, reuse, and collaboration on obsolescence issues, and develop decision making models to support proactive and strategic management for overall cost savings in managing obsolescence. Three primary aspects of this research are investigated. First, ontologies for obsolescence knowledge representation are developed in a systematic way with the use of UML diagrams. The generality of the developed ontology is demonstrated with distinct examples. Diminishing Manufacturing Sources and Material Shortages (DMSMS) obsolescence provides the basis for this study. Second, an ontology-based hybrid approach for integrating heterogeneous data resources in existing obsolescence management tools is proposed. Third, decision support models are developed and formalized, and include the obsolescence forecasting method for proactively managing obsolescence, and the mathematical models to determine the optimal design refresh plan to minimize the product life cycle cost for strategic obsolescence management. Finally, the design of the obsolescence management information system is provided along with a system evaluation methodology. Ultimately, the research contributes to the field of knowledge representation as well as design for managing product obsolescence. / Ph. D.