Submitted by Daniella Santos (daniella.santos@ufma.br) on 2017-11-23T14:52:48Z No. of bitstreams: 1 DanielGomesJúnior.pdf: 8247020 bytes, checksum: ccc1a51b78a0a368f50ca5393fa8d97c (MD5) / Made available in DSpace on 2017-11-23T14:52:48Z (GMT). No. of bitstreams: 1 DanielGomesJúnior.pdf: 8247020 bytes, checksum: ccc1a51b78a0a368f50ca5393fa8d97c (MD5) Previous issue date: 2017-08-25 / This research proposes a methodology for development of Virttual Reality (VR) and Augmented Reality (AR) aplications, using natural markers for industrial scenarios. The proposed methodology uses the object annotation concept and visualization proposals are presented both for development of VR as for AR environments. In VR environments, the methodology is applied for object detection step of the semi-automatic environment development. On the other hand, in AR environments, is presented the concept of georreferenced natural markers, which use the georreferenced data integrated with object detection process using image processing techniques. The energy substations scenarios were used as case study for both approaches. Architectures are presented for construction and data visualization in industrial environments. Both for VR as for AR approaches, this work proposes using 3D natural markers based in Haar-like features for object training and detection process. The results enable the equipment detection at different points of view, within the operating scenario. Besides that, in AR, it enables the pose estimation in real-time using ORB features, while in VR it enables the semi-automatic object detection, which are used as information points for inclusion of virtual information. Several industrial scenarios, and especially the energy sector, has a high degree of complexity in the information processing and visualization. In this sense, beyond the 3D natural markers methodology, this work presents new visualization applications for industrial scenario visualization in VR and AR approaches. / Esta pesquisa propõe uma metodologia para construção de aplicações de Realidade Virtual (RV) e Realidade Aumentada (RA) com uso de marcadores naturais em cenários industriais. A metodologia usa o conceito de anotação de objetos e são apresentadas propostas de visualização para ambientes industriais tanto em formato de RV quanto de RA. Nos ambientes de RV, a metodologia é aplicada através da detecção de objetos no processo de construção semiautomática dos ambientes. Por outro lado, nos ambientes de RA, apresenta-se o conceito de marcadores naturais georreferenciados, que associam dados georreferenciados ao processo de detecção de objetos com técnicas de processamento de imagens. O cenário de subestações de energia elétrica foi utilizado como estudo de caso para as duas abordagens. São apresentadas arquiteturas para construção e visualização de dados em ambientes industriais. Tanto sob a forma de RV quanto de RA, este trabalho propõe o uso de marcadores naturais 3D baseados em Haar-like features para o processo de treinamento e detecção de objetos. Os resultados permitem a detecção de equipamentos a partir de diferentes pontos de vista no cenário de operação. Além disso, em RA, esta abordagem permite a estimativa de pose em tempo real com uso de ORB features e permite, em RV, a detecção semiautomática de objetos que são utilizados como pontos de informação para adição de informações virtuais. Diversos cenários industriais, principalmente o setor elétrico, possuem grau elevado de complexidade no tratamento e visualização das informações. Nesse sentido, além da metodologia de marcadores naturais 3D, este trabalho apresenta novas aplicações de visualização no cenário industrial com abordagens em RV e RA.

No Aprendizado de Máquina Supervisionado - AM - é apresentado ao algoritmo de indução um conjunto de instâncias de treinamento, no qual cada instância é um vetor de features rotulado com a classe. O algoritmo de indução tem como tarefa induzir um classificador que será utilizado para classificar novas instâncias. Algoritmos de indução convencionais baseam-se nos dados fornecidos pelo usuário para construir as descrições dos conceitos. Uma representação inadequada do espaço de busca ou da linguagem de descrição do conjunto de instâncias, bem como erros nos exemplos de treinamento, podem tornar os problemas de aprendizado difícies. Um dos problemas centrais em AM é a Seleção de um Subconjunto de Features - SSF - na qual o objetivo é tentar diminuir o número de features que serão fornecidas ao algoritmo de indução. São várias as razões para a realização de SSF. A primeira é que a maioria dos algoritmos de AM, computacionalmente viáveis, não trabalham bem na presença de muitas features, isto é a precisão dos classificadores gerados pode ser melhorada com a aplicação de SSF. Ainda, com um número menor de features, a compreensibilidade do conceito induzido pode ser melhorada. Uma terceira razão é o alto custo para coletar e processar grande quantidade de dados. Existem, basicamente, três abordagens para a SSF: embedded, filtro e wrapper. Por outro lado, se as features utilizadas para descrever os exemplos de treinamento são inadequadas, os algoritmos de aprendizado estão propensos a criar descrições excessivamente complexas e imprecisas. Porém, essas features, individualmente inadequadas, podem algumas vezes serem, convenientemente, combinadas gerando novas features que podem mostrar-se altamente representativas para a descrição de um conceito. O processo de construção de novas features é conhecido como Construção de Features ou Indução Construtiva - IC. Neste trabalho são enfocadas as abordagens filtro e wrapper para a realização de SSF, bem como a IC guiada pelo conhecimento. É descrita uma série de experimentos usando SSF e IC utilizando quatro conjuntos de dados naturais e diversos algoritmos simbólicos de indução. Para cada conjunto de dados e cada indutor, são realizadas várias medidas, tais como, precisão, tempo de execução do indutor e número de features selecionadas pelo indutor. São descritos também diversos experimentos realizados utilizando três conjuntos de dados do mundo real. O foco desses experimentos não está somente na avaliação da performance dos algoritmos de indução, mas também na avaliação do conhecimento extraído. Durante a extração de conhecimento, os resultados foram apresentados aos especialistas para que fossem feitas sugestões para experimentos futuros. Uma parte do conhecimento extraído desses três estudos de casos foram considerados muito interessantes pelos especialistas. Isso mostra que a interação de diferentes áreas de conhecimento, neste caso específico, áreas médica e computacional, pode produzir resultados interessantes. Assim, para que a aplicação do Aprendizado de Máquina possa gerar frutos é necessário que dois grupos de pesquisadores sejam unidos: aqueles que conhecem os métodos de AM existentes e aqueles com o conhecimento no domínio da aplicação para o fornecimento de dados e a avaliação do conhecimento adquirido. / In supervised Machine Learning - ML - an induction algorithm is typically presented with a set of training instances, where each instance is described by a vector of feature values and a class label. The task of the induction algorithm (inducer) is to induce a classifier that will be useful in classifying new cases. Conventional inductive-learning algorithms rely on existing (user) provided data to build their descriptions. Inadequate representation space or description language as well as errors in training examples can make learning problems be difficult. One of the main problems in ML is the Feature Subset Selection - FSS - problem, i.e. the learning algorithm is faced with the problem of selecting some subset of features upon which to focus its attention, while ignoring the rest. There are a variety of reasons that justify doing FSS. The first reason that can be pointed out is that most of the ML algorithms, that are computationally feasible, do not work well in the presence of a very large number of features. This means that FSS can improve the accuracy of the classifiers generated by these algorithms. Another reason to use FSS is that it can improve comprehensibility, i.e. the human ability of understanding the data and the rules generated by symbolic ML algorithms. A third reason for doing FSS is the high cost in some domains for collecting data. Finally, FSS can reduce the cost of processing huge quantities of data. Basically, there are three approaches in Machine Learning for FSS: embedded, filter and wrapper approaches. On the other hand, if the provided features for describing the training examples are inadequate, the learning algorithms are likely to create excessively complex and inaccurate descriptions. These individually inadequate features can sometimes be combined conveniently, generating new features which can turn out to be highly representative to the description of the concept. The process of constructing new features is called Constructive Induction - CI. Is this work we focus on the filter and wrapper approaches for FSS as well as Knowledge-driven CI. We describe a series of experiments for FSS and CI, performed on four natural datasets using several symbolic ML algorithms. For each dataset, various measures are taken to compare the inducers performance, for example accuracy, time taken to run the inducers and number of selected features by each evaluated induction algorithm. Several experiments using three real world datasets are also described. The focus of these three case studies is not only comparing the induction algorithms performance, but also the evaluation of the extracted knowledge. During the knowledge extraction step results were presented to the specialist, who gave many suggestions for the development of further experiments. Some of the knowledge extracted from these three real world datasets were found very interesting by the specialist. This shows that the interaction between different areas, in this case, medical and computational areas, may produce interesting results. Thus, two groups of researchers need to be put together if the application of ML is to bear fruit: those that are acquainted with the existing ML methods, and those with expertise in the given application domain to provide training data.

<p>The present thesis investigates environmental factors impact on office employees. More specifically, it investigates: 1) perception and experience of office environments, 2) satisfaction with office environments, and 3) health status and job satisfaction in connection to office environment. It is based on an empirical study with 491 office employees from twenty-six companies and divisions in larger companies. Each one respectively represents one of seven identified office-types in office design: cell-office, sharedroom office, small open plan office, medium open plan office, large open plan office, flex-office and combi-office. This study takes its basis in architecture, although an interdisciplinary approach from organizational and management theory, environmental psychology, and social and stress medicine has been used. Qualitative and quantitative methods were used.</p><p> In Article I a review of the different research fields that investigate environmental influences are presented with a focus on office environments. Different perspectives on the environmental impact on office employees are investigated.</p><p>In Article II an analysis of office environment based on the employee’s perception and experience of the architecture is done based on in-depth interviews using a method originally developed by Kevin Lynch (1960). The method measures the "imagebility" of a space, rated by the users with following elements: landmark, node, path, edge and district. The result showed that the method, based on employees’ perception and use of space, is a possible tool in the design process to get a better understanding of where the elements that reinforce "imageability" most likely will appear in an office environment. The method thus gives a better idea of the future "imageability" of a space and could be useful as guidance in the design process of how the architectural design will be received by the users in the end.</p><p>In Article III employees’ satisfaction with the office environment in different office-types is investigated. The article focuses on three domains: 1) Ambient factors, 2) Noise and Privacy and 3) Designrelated factors. The statistical analysis was done using a logistic regression model with multivariate analysis. Adjustment was done for: age, gender, job rank, job satisfaction and market division. The results show differences in satisfaction with the office environment between employees in different office-types, many of which were statistically significant. When differences persist in the multivariate analysis they can possibly be ascribed to the office-type. Results show that employees in cell-offices are prominently most satisfied followed by those in flex-offices. Cell-offices rate only low on social aspects of Design-related factors. A major finding is internal differences between different office-types where employees share workspace and facilities. The medium and large open plan offices could be described as high-risk officetypes.</p><p>In Article IV differences between employees in different office-types with regard to health, wellbeing and job satisfaction are analyzed. A multivariate analysis of the data was done with adjustment for the confounders: age, gender, job rank and market division. The results show that there are risks of ill health and poor well-being in medium and small open plan offices. Employees in these office-types show significantly higher risks compared with those in other office-types. In medium open plan and combioffices the employees show the highest prevalence of low job satisfaction. The best chance for good health status and job satisfaction is among employees in cell-offices and flex-offices; there are, however, internal differences in distribution on different outcome variables for job satisfaction. The major finding of these studies is that there are significant differences with regard to satisfaction with office environments as well as health status and job satisfaction between employees in different office-types; differences that can possibly can be ascribed to the office-types as they persist after adjustment for important confounders.</p>

In order to generate substantial market impact, mobile applications must be able to run on multiple platforms. Hence, software engineers face a multitude of technologies and system versions resulting in static variability. Furthermore, due to the dependence on sensors and connectivity, mobile software has to adapt its behavior accordingly at runtime resulting in dynamic variability. However, software engineers need to assure quality of a mobile application even with this large amount of variability—in our approach by the use of model-based testing (i.e., the generation of test cases from models). Recent concepts of test metamodels cannot efficiently handle dynamic variability. To overcome this problem, we propose a process for creating black-box test models based on dynamic feature Petri nets, which allow the description of configuration-dependent behavior and reconfiguration. We use feature models to define variability in the system under test. Furthermore, we illustrate our approach by introducing an example translator application.

Magistro darbo tema „Nužudymą kvalifikuojantys požymiai pagal Lietuvos ir kitų valstybių baudžiamuosius įstatymus“, todėl pagrindinis darbo tikslas yra atskleisti nužudymą kvalifikuojančius požymius, numatytus Lietuvos Respublikos bei kitų valstybių baudžiamuosiuose įstatymuose, pasitelkiant lyginamąjį metodą. Siekiant išsamiai atskleisti kvalifikuoto nužudymo požymius, darbe nagrinėjamas nužudymo nusikaltimo teisinis reglamentavimas – pirmoje darbo dalyje aptarta nužudymo samprata, toliau išanalizuojami teisės šaltiniai, kurie turėjo įtakos formuojant dabartinę kvalifikuoto nužudymo sudėties konstrukciją Lietuvos, Prancūzijos, Vokietijos bei Rusijos Federacijos galiojančiuose baudžiamuosiuose įstatymuose. Kadangi nužudymo kvalifikuota sudėtis apima pagrindinę sudėtį, trečioje darbo dalyje nagrinėjama pagrindinės nužudymo sudėties objektyvieji ir subjektyvieji požymiai. Nužudymą kvalifikuojančių požymių analizei skiriama didžioji darbo dalis, kurioje, remiantis galiojančiais baudžiamaisiais įstatymais, moksline literatūra bei teismų praktika, nagrinėjamas kiekvienas nužudymą kvalifikuojantis požymis, lyginant su pasirinktų užsienio valstybių baudžiamaisiais įstatymais. / The title of the master‘s thesis is „The qualifying features of murder according to criminal laws of Lithuania and other countries“, therefore the core aim of this work is to disclose the qualifying features of murder provided in criminal codes of the Republic of Lithuania and other countries, invoking comparative method. In order to thoroughly disclose the features of the qualified murder, the juridical reglamentation of this crime is examined in the paper - the conception of murder is discussed in the first part of the paper, afterwards, legal sources that had influence on the development of the present qualified murder content construction in valid criminal laws of the Republic of Lithuania, France, Federation of Germany and Russia are analyzed. As the qualified content of murder includes the main content of murder, the general content of this crime is described in the third part of the paper analyzing objective and subjective features. The analysis of the qualifying features is presented in the greater part of the work where according to the valid criminal laws of the chosen countries, scholarly literature and the practice of courts, every qualifying feature of murder is examined comparing it with the criminal laws of the chosen countries.

The main results of the PhD thesis are described in the summary. The main objective of the thesis is to define the acquisition of lexical and morphosyntactic features of Lithuanian adjective. The analysis is based on the large quantity of linguistic data (~ 400 000 words), which is the corpus of four children compiled according the method of longitudinal observation. The study has interchangeably applied several methods of analysis: longitudinal observation, corpus linguistics, error analysis and comparative method. Statements to be Defended 1. Lexical features of adjectives are acquired by children with difficulties: they often confuse adjectives of a single semantic group and make many errors. 2. Children easily acquire morphosyntactic features of adjectives: they do not make many errors. Some errors may be observed only in cases, when adjective paradigms do not coincide with noun paradigms. 3. Lexical and morphosyntactic features of adjectives are acquired simultaneously. The meaning of adjective is realised only when multi-word combinations and morphosyntactic features have been acquired. 4. The acquisition of adjectives correlates with the individual language development. Bigger lexical variety of adjectives is more characteristic to children with early and late language development, if compared to children with typical language development. Morphosyntactic features of adjectives do not pose problems to children with late language development. 5. The acquisition of... [to full text] / Santraukoje pateikiami pagrindiniai disertacinio tyrimo metu gauti rezultatai. Aprašomo tyrimo tikslas – išanalizuoti leksines ir morfosintaksines lietuvių kalbos būdvardžio įsisavinimo ypatybes. Tyrimui pasirinktas didelės apimties šaltinis – keturių vaikų kalbos tekstynas (~ 400 000 žodžių), sukauptas ilgalaikio stebėjimo metodu. Tyrimas atliktas taikant ir derinant kelis metodus: ilgalaikio stebėjimo, tekstynų lingvistikos, klaidų analizės ir lyginamąjį. Atlikus tyrimą darbe pagrindžiami šie ginamieji teiginiai: 1. Vaikams sudėtinga įsisavinti būdvardžio leksines ypatybes: dažnai painiojami vienos leksinės semantinės grupės būdvardžiai, daroma daug klaidų. 2. Vaikai lengvai įsisavina morfosintaksines būdvardžio ypatybes: klaidų nėra daug, dažniau klystama tik retų paradigmų būdvardžius derinant su daiktavardžiais. 3. Būdvardžio leksinės ir morfosintaksinės ypatybės įsisavinamos drauge. Būdvardžio reikšmė suvokiama tik pradėjus vartoti keliažodžius pasakymus, įsisavinus gramatines ypatybes. 4. Būdvardžio įsisavinimas susijęs su individualia vaiko kalbos raida. Ankstyvajai ir vėlyvajai kalbos raidai būdinga didesnė būdvardžių leksinė įvairovė nei įprastai kalbos raidai, o vaikams, kurių kalbos raida vėlyva, nesudėtinga įsisavinti būdvardžio morfosintaksinius požymius. 5. Būdvardžio įsisavinimą veikia ne tik aplinkos veiksniai (vaikiškoji kalba), bet ir kalbos sistema. Kaip ir įsisavinant kitas kalbos dalis, vaikų kalboje vyrauja vaikiškojoje kalboje dažniausiai vartojamos... [toliau žr. visą tekstą]

Disertacijoje aprašomo tyrimo tikslas – išanalizuoti leksines ir morfosintaksines lietuvių kalbos būdvardžio įsisavinimo ypatybes. Tyrimui pasirinktas didelės apimties šaltinis – keturių vaikų kalbos tekstynas (~ 400 000 žodžių), sukauptas ilgalaikio stebėjimo metodu. Tyrimas atliktas taikant ir derinant kelis metodus: ilgalaikio stebėjimo, tekstynų lingvistikos, klaidų analizės ir lyginamąjį. Atlikus tyrimą darbe pagrindžiami šie ginamieji teiginiai: 1. Vaikams sudėtinga įsisavinti būdvardžio leksines ypatybes: dažnai painiojami vienos leksinės semantinės grupės būdvardžiai, daroma daug klaidų. 2. Vaikai lengvai įsisavina morfosintaksines būdvardžio ypatybes: klaidų nėra daug, dažniau klystama tik retų paradigmų būdvardžius derinant su daiktavardžiais. 3. Būdvardžio leksinės ir morfosintaksinės ypatybės įsisavinamos drauge. Būdvardžio reikšmė suvokiama tik pradėjus vartoti keliažodžius pasakymus, įsisavinus gramatines ypatybes. 4. Būdvardžio įsisavinimas susijęs su individualia vaiko kalbos raida. Ankstyvajai ir vėlyvajai kalbos raidai būdinga didesnė būdvardžių leksinė įvairovė nei įprastai kalbos raidai, o vaikams, kurių kalbos raida vėlyva, nesudėtinga įsisavinti būdvardžio morfosintaksinius požymius. 5. Būdvardžio įsisavinimą veikia ne tik aplinkos veiksniai (vaikiškoji kalba), bet ir kalbos sistema. Kaip ir įsisavinant kitas kalbos dalis, vaikų kalboje vyrauja vaikiškojoje kalboje dažniausiai vartojamos būdvardžio leksemos ir gramatinės kategorijos. Kalbos sistema... [toliau žr. visą tekstą] / The main objective of the study is to define the acquisition of lexical and morphosyntactic features of Lithuanian adjective. The analysis is based on the large quantity of linguistic data (~ 400 000 words), which is the corpus of four children compiled according the method of longitudinal observation. The study has interchangeably applied several methods of analysis: longitudinal observation, corpus linguistics, error analysis and comparative method. Statements to be Defended 1. Lexical features of adjectives are acquired by children with difficulties: they often confuse adjectives of a single semantic group and make many errors. 2. Children easily acquire morphosyntactic features of adjectives: they do not make many errors. Some errors may be observed only in cases, when adjective paradigms do not coincide with noun paradigms. 3. Lexical and morphosyntactic features of adjectives are acquired simultaneously. The meaning of adjective is realised only when multi-word combinations and morphosyntactic features have been acquired. 4. The acquisition of adjectives correlates with the individual language development. Bigger lexical variety of adjectives is more characteristic to children with early and late language development, if compared to children with typical language development. Morphosyntactic features of adjectives do not pose problems to children with late language development. 5. The acquisition of adjective is influenced by the input, as well as by the linguistic... [to full text]

The goal of this thesis is to advance the use of recurrent neural network language models (RNNLMs) for large vocabulary continuous speech recognition (LVCSR). RNNLMs are currently state-of-the-art and shown to consistently reduce the word error rates (WERs) of LVCSR tasks when compared to other language models. In this thesis we propose various advances to RNNLMs. The advances are: improved learning procedures for RNNLMs, enhancing the context, and adaptation of RNNLMs. We learned better parameters by a novel pre-training approach and enhanced the context using prosody and syntactic features. We present a pre-training method for RNNLMs, in which the output weights of a feed-forward neural network language model (NNLM) are shared with the RNNLM. This is accomplished by first fine-tuning the weights of the NNLM, which are then used to initialise the output weights of an RNNLM with the same number of hidden units. To investigate the effectiveness of the proposed pre-training method, we have carried out text-based experiments on the Penn Treebank Wall Street Journal data, and ASR experiments on the TED lectures data. Across the experiments, we observe small but significant improvements in perplexity (PPL) and ASR WER. Next, we present unsupervised adaptation of RNNLMs. We adapted the RNNLMs to a target domain (topic or genre or television programme (show)) at test time using ASR transcripts from first pass recognition. We investigated two approaches to adapt the RNNLMs. In the first approach the forward propagating hidden activations are scaled - learning hidden unit contributions (LHUC). In the second approach we adapt all parameters of RNNLM.We evaluated the adapted RNNLMs by showing the WERs on multi genre broadcast speech data. We observe small (on an average 0.1% absolute) but significant improvements in WER compared to a strong unadapted RNNLM model. Finally, we present the context-enhancement of RNNLMs using prosody and syntactic features. The prosody features were computed from the acoustics of the context words and the syntactic features were from the surface form of the words in the context. We trained the RNNLMs with word duration, pause duration, final phone duration, syllable duration, syllable F0, part-of-speech tag and Combinatory Categorial Grammar (CCG) supertag features. The proposed context-enhanced RNNLMs were evaluated by reporting PPL and WER on two speech recognition tasks, Switchboard and TED lectures. We observed substantial improvements in PPL (5% to 15% relative) and small but significant improvements in WER (0.1% to 0.5% absolute).

Made available in DSpace on 2015-05-14T12:36:52Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 13394167 bytes, checksum: 812505c41510e689271080958a7ad1cc (MD5) Previous issue date: 2012-12-18 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / In the last decades, the cultural heritage has broadened its concept and several knowledge areas are being restructured to ensure the total protection of the heritage. Anthropology, Law, History, Sociology are just a few examples of areas that are adapting to this broadest concept of Cultural Heritage. The Computer Science, although can contribute in a more helpful way to the immaterial heritage protection, it still doesn't have many specific tools and means to lead how to guarantee the cultural legacy, through its several tools. This work is attentive to the need of preservation and protection of culture and intends to contribute to this goal using a features oriented approach, explaining how and why to preserve digitally the several aspects of the immaterial heritage. The evaluation of the approach proposed in this work was made by de development of PAMIN. This experimentation space enabled the validation of the Immaterial Cultural Heritage representation and management through its interface, functionalities and its several aspects of digital surrogates. / O patrimônio cultural nas últimas décadas vem ampliando seu conceito e as mais diversas áreas de conhecimento vêm sendo reestruturadas para garantir a proteção do patrimônio em sua totalidade. A Antropologia, o Direito, a História e a Sociologia são apenas alguns exemplos dessas áreas que vêm se moldando a ampliação do conceito de Patrimônio Cultural. A Área de Ciências da Computação, embora seja uma vertente que pode contribuir de uma maneira mais completa para a proteção do bem imaterial, ainda não possui muitos instrumentos e meios que moldem ou ditem como garantir o legado cultural através de suas variadas ferramentas. O presente trabalho está atento à necessidade de preservação e proteção da cultura e visa contribuir com esta finalidade através de uma abordagem orientada a features, explicando como e porque conservar os variados aspectos do bem imaterial em meio digital. A avaliação da abordagem proposta neste trabalho foi feita através do desenvolvimento do PAMIN. Este espaço de experimentação possibilitou a validação da representação e gerenciamento do Patrimônio Cultural Imaterial através de sua interface, de suas funcionalidades e seus variados aspectos de representação em meio digital.

No Aprendizado de Máquina Supervisionado - AM - é apresentado ao algoritmo de indução um conjunto de instâncias de treinamento, no qual cada instância é um vetor de features rotulado com a classe. O algoritmo de indução tem como tarefa induzir um classificador que será utilizado para classificar novas instâncias. Algoritmos de indução convencionais baseam-se nos dados fornecidos pelo usuário para construir as descrições dos conceitos. Uma representação inadequada do espaço de busca ou da linguagem de descrição do conjunto de instâncias, bem como erros nos exemplos de treinamento, podem tornar os problemas de aprendizado difícies. Um dos problemas centrais em AM é a Seleção de um Subconjunto de Features - SSF - na qual o objetivo é tentar diminuir o número de features que serão fornecidas ao algoritmo de indução. São várias as razões para a realização de SSF. A primeira é que a maioria dos algoritmos de AM, computacionalmente viáveis, não trabalham bem na presença de muitas features, isto é a precisão dos classificadores gerados pode ser melhorada com a aplicação de SSF. Ainda, com um número menor de features, a compreensibilidade do conceito induzido pode ser melhorada. Uma terceira razão é o alto custo para coletar e processar grande quantidade de dados. Existem, basicamente, três abordagens para a SSF: embedded, filtro e wrapper. Por outro lado, se as features utilizadas para descrever os exemplos de treinamento são inadequadas, os algoritmos de aprendizado estão propensos a criar descrições excessivamente complexas e imprecisas. Porém, essas features, individualmente inadequadas, podem algumas vezes serem, convenientemente, combinadas gerando novas features que podem mostrar-se altamente representativas para a descrição de um conceito. O processo de construção de novas features é conhecido como Construção de Features ou Indução Construtiva - IC. Neste trabalho são enfocadas as abordagens filtro e wrapper para a realização de SSF, bem como a IC guiada pelo conhecimento. É descrita uma série de experimentos usando SSF e IC utilizando quatro conjuntos de dados naturais e diversos algoritmos simbólicos de indução. Para cada conjunto de dados e cada indutor, são realizadas várias medidas, tais como, precisão, tempo de execução do indutor e número de features selecionadas pelo indutor. São descritos também diversos experimentos realizados utilizando três conjuntos de dados do mundo real. O foco desses experimentos não está somente na avaliação da performance dos algoritmos de indução, mas também na avaliação do conhecimento extraído. Durante a extração de conhecimento, os resultados foram apresentados aos especialistas para que fossem feitas sugestões para experimentos futuros. Uma parte do conhecimento extraído desses três estudos de casos foram considerados muito interessantes pelos especialistas. Isso mostra que a interação de diferentes áreas de conhecimento, neste caso específico, áreas médica e computacional, pode produzir resultados interessantes. Assim, para que a aplicação do Aprendizado de Máquina possa gerar frutos é necessário que dois grupos de pesquisadores sejam unidos: aqueles que conhecem os métodos de AM existentes e aqueles com o conhecimento no domínio da aplicação para o fornecimento de dados e a avaliação do conhecimento adquirido. / In supervised Machine Learning - ML - an induction algorithm is typically presented with a set of training instances, where each instance is described by a vector of feature values and a class label. The task of the induction algorithm (inducer) is to induce a classifier that will be useful in classifying new cases. Conventional inductive-learning algorithms rely on existing (user) provided data to build their descriptions. Inadequate representation space or description language as well as errors in training examples can make learning problems be difficult. One of the main problems in ML is the Feature Subset Selection - FSS - problem, i.e. the learning algorithm is faced with the problem of selecting some subset of features upon which to focus its attention, while ignoring the rest. There are a variety of reasons that justify doing FSS. The first reason that can be pointed out is that most of the ML algorithms, that are computationally feasible, do not work well in the presence of a very large number of features. This means that FSS can improve the accuracy of the classifiers generated by these algorithms. Another reason to use FSS is that it can improve comprehensibility, i.e. the human ability of understanding the data and the rules generated by symbolic ML algorithms. A third reason for doing FSS is the high cost in some domains for collecting data. Finally, FSS can reduce the cost of processing huge quantities of data. Basically, there are three approaches in Machine Learning for FSS: embedded, filter and wrapper approaches. On the other hand, if the provided features for describing the training examples are inadequate, the learning algorithms are likely to create excessively complex and inaccurate descriptions. These individually inadequate features can sometimes be combined conveniently, generating new features which can turn out to be highly representative to the description of the concept. The process of constructing new features is called Constructive Induction - CI. Is this work we focus on the filter and wrapper approaches for FSS as well as Knowledge-driven CI. We describe a series of experiments for FSS and CI, performed on four natural datasets using several symbolic ML algorithms. For each dataset, various measures are taken to compare the inducers performance, for example accuracy, time taken to run the inducers and number of selected features by each evaluated induction algorithm. Several experiments using three real world datasets are also described. The focus of these three case studies is not only comparing the induction algorithms performance, but also the evaluation of the extracted knowledge. During the knowledge extraction step results were presented to the specialist, who gave many suggestions for the development of further experiments. Some of the knowledge extracted from these three real world datasets were found very interesting by the specialist. This shows that the interaction between different areas, in this case, medical and computational areas, may produce interesting results. Thus, two groups of researchers need to be put together if the application of ML is to bear fruit: those that are acquainted with the existing ML methods, and those with expertise in the given application domain to provide training data.