Дисертації з теми "Interpretable methods"

Thesis: Ph. D., Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, May, 2020
Cataloged from the official PDF of thesis.
Includes bibliographical references (pages 131-142).
With data becoming increasingly available in recent years, black-box algorithms like boosting methods or neural networks play more important roles in the real world. However, interpretability is a severe need for several areas of applications, like health care or business. Doctors or managers often need to understand how models make predictions, in order to make their final decisions. In this thesis, we improve and propose some interpretable machine learning methods by using modern optimization. We also use two examples to illustrate how interpretable machine learning methods help to solve problems in health care. The first part of this thesis is about interpretable machine learning methods using modern optimization. In Chapter 2, we illustrate how to use robust optimization to improve the performance of SVM, Logistic Regression, and Classification Trees for imbalanced datasets. In Chapter 3, we discuss how to find optimal clusters for prediction. we use real-world datasets to illustrate this is a fast and scalable method with high accuracy. In Chapter 4, we deal with optimal regression trees with polynomial function in leaf nodes and demonstrate this method improves the out-of-sample performance. The second part of this thesis is about how interpretable machine learning methods improve the current health care system. In Chapter 5, we illustrate how we use Optimal Trees to predict the risk mortality for candidates awaiting liver transplantation. Then we develop a transplantation policy called Optimized Prediction of Mortality (OPOM), which reduces mortality significantly in simulation analysis and also improves fairness. In Chapter 6, we propose a new method based on Optimal Trees which perform better than original rules in identifying children at very low risk of clinically important traumatic brain injury (ciTBI). If this method is implemented in the electronic health record, the new rules may reduce unnecessary computed tomographies (CT).
by Yuchen Wang.
Ph. D.
Ph.D. Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center

This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Thesis: S.M., Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2019
Cataloged from student-submitted PDF version of thesis.
Includes bibliographical references (pages 75-80).
Black box machine learning methods have allowed researchers to design accurate models using large amounts of data at the cost of interpretability. Model interpretability not only improves user buy-in, but in many cases provides users with important information. Especially in the case of the classification problems addressed in this thesis, the ideal model should not only provide accurate predictions, but should also inform users of how features affect the results. My research goal is to solve real-world problems and compare how different classification models affect the outcomes and interpretability. To this end, this thesis is divided into two parts: food safety risk analysis and human trafficking detection. The first half analyzes the characteristics of supermarket suppliers in China that indicate a high risk of food safety violations. Contrary to expectations, supply chain dispersion, internal inspections, and quality certification systems are not found to be predictive of food safety risk in our data. The second half focuses on identifying human trafficking, specifically sex trafficking, advertisements hidden amongst online classified escort service advertisements. We propose a novel but interpretable keyword detection and modeling pipeline that is more accurate and actionable than current neural network approaches. The algorithms and applications presented in this thesis succeed in providing users with not just classifications but also the characteristics that indicate food safety risk and human trafficking ads.
by Jessica H. Zhu.
S.M.
S.M. Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center

Malignant tumours of the brain represent one of the most difficult to treat types of cancer due to the sensitive organ they affect. Clinical management of the pathology becomes even more intricate as the tumour mass increases due to proliferation, suggesting that an early and accurate diagnosis is vital for preventing it from its normal course of development. The standard clinical practise for diagnosis includes invasive techniques that might be harmful for the patient, a fact that has fostered intensive research towards the discovery of alternative non-invasive brain tissue measurement methods, such as nuclear magnetic resonance. One of its variants, magnetic resonance imaging, is already used in a regular basis to locate and bound the brain tumour; but a complementary variant, magnetic resonance spectroscopy, despite its higher spatial resolution and its capability to identify biochemical metabolites that might become biomarkers of tumour within a delimited area, lags behind in terms of clinical use, mainly due to its difficult interpretability. The interpretation of magnetic resonance spectra corresponding to brain tissue thus becomes an interesting field of research for automated methods of knowledge extraction such as machine learning, always understanding its secondary role behind human expert medical decision making. The current thesis aims at contributing to the state of the art in this domain by providing novel techniques for assistance of radiology experts, focusing on complex problems and delivering interpretable solutions. In this respect, an ensemble learning technique to accurately discriminate amongst the most aggressive brain tumours, namely glioblastomas and metastases, has been designed; moreover, a strategy to increase the stability of biomarker identification in the spectra by means of instance weighting is provided. From a different analytical perspective, a tool based on signal source separation, guided by tumour type-specific information has been developed to assess the existence of different tissues in the tumoural mass, quantifying their influence in the vicinity of tumoural areas. This development has led to the derivation of a probabilistic interpretation of some source separation techniques, which provide support for uncertainty handling and strategies for the estimation of the most accurate number of differentiated tissues within the analysed tumour volumes. The provided strategies should assist human experts through the use of automated decision support tools and by tackling interpretability and accuracy from different angles
Els tumors cerebrals malignes representen un dels tipus de càncer més difícils de tractar degut a la sensibilitat de l’òrgan que afecten. La gestió clínica de la patologia esdevé encara més complexa quan la massa tumoral s'incrementa degut a la proliferació incontrolada de cèl·lules; suggerint que una diagnosis precoç i acurada és vital per prevenir el curs natural de desenvolupament. La pràctica clínica estàndard per a la diagnosis inclou la utilització de tècniques invasives que poden arribar a ser molt perjudicials per al pacient, factor que ha fomentat la recerca intensiva cap al descobriment de mètodes alternatius de mesurament dels teixits del cervell, tals com la ressonància magnètica nuclear. Una de les seves variants, la imatge de ressonància magnètica, ja s'està actualment utilitzant de forma regular per localitzar i delimitar el tumor. Així mateix, una variant complementària, la espectroscòpia de ressonància magnètica, malgrat la seva alta resolució espacial i la seva capacitat d'identificar metabòlits bioquímics que poden esdevenir biomarcadors de tumor en una àrea delimitada, està molt per darrera en termes d'ús clínic, principalment per la seva difícil interpretació. Per aquest motiu, la interpretació dels espectres de ressonància magnètica corresponents a teixits del cervell esdevé un interessant camp de recerca en mètodes automàtics d'extracció de coneixement tals com l'aprenentatge automàtic, sempre entesos com a una eina d'ajuda per a la presa de decisions per part d'un metge expert humà. La tesis actual té com a propòsit la contribució a l'estat de l'art en aquest camp mitjançant l'aportació de noves tècniques per a l'assistència d'experts radiòlegs, centrades en problemes complexes i proporcionant solucions interpretables. En aquest sentit, s'ha dissenyat una tècnica basada en comitè d'experts per a una discriminació acurada dels diferents tipus de tumors cerebrals agressius, anomenats glioblastomes i metàstasis; a més, es proporciona una estratègia per a incrementar l'estabilitat en la identificació de biomarcadors presents en un espectre mitjançant una ponderació d'instàncies. Des d'una perspectiva analítica diferent, s'ha desenvolupat una eina basada en la separació de fonts, guiada per informació específica de tipus de tumor per a avaluar l'existència de diferents tipus de teixits existents en una massa tumoral, quantificant-ne la seva influència a les regions tumorals veïnes. Aquest desenvolupament ha portat cap a la derivació d'una interpretació probabilística d'algunes d'aquestes tècniques de separació de fonts, proporcionant suport per a la gestió de la incertesa i estratègies d'estimació del nombre més acurat de teixits diferenciats en cada un dels volums tumorals analitzats. Les estratègies proporcionades haurien d'assistir els experts humans en l'ús d'eines automatitzades de suport a la decisió, donada la interpretabilitat i precisió que presenten des de diferents angles.

Today, companies like Volvo GTO experience a vast increase in data and the ability toprocess it. This makes it possible to utilize machine learning models to construct a rootcause analysis system in order to predict, explain and prevent defects. However, thereexists a trade-off between model performance and explanation capability, both of whichare essential to such system.This thesis aims to, with the use of machine learning models, inspect the relationshipbetween sensor data from the painting process and the texture defectorange peel. Theaim is also to evaluate the consistency of different explanation methods.After the data was preprocessed, and new features were engineered, e.g. adjustments,three machine learning models were trained and tested. In order to explain a linearmodel, one can use its coefficients. In the case of a tree-based model, MDI is a commonglobal explanation method. SHAP is a state-of-the-art model-independent method thatcan explain a model globally and locally. These three methods were compared in orderto evaluate the consistency of their explanations. If SHAP would be consistent with theothers on a global level, it can be argued that SHAP can be used locally in an root causeanalysis.The study showed that the coefficients and MDI were consistent with SHAP as theoverall correlation between them were high and because they tended to weight thefeatures in a similar way. From this conclusion, a root cause analysis algorithm wasdeveloped with SHAP as a local explanation method. Finally, it cannot be concludedthat there is a relationship between the sensor data andorange peel, as the adjustments ofthe process were the most impactful features.
Idag upplever företag som Volvo GTO en stor ökning av data och en förbättrad förmågaatt bearbeta den. Detta gör det möjligt att, med hjälp av maskininlärningsmodeller,skapa ett rotorsaksanalyssystem för att förutspå, förklara och förebygga defekter. Detfinns dock en balans mellan modellprestanda och förklaringskapacitet, där båda ärväsentliga för ett sådant system.Detta examensarbete har som mål att, med hjälp av maskininlärningsmodeller, under-söka förhållandet mellan sensordata från målningsprocessen och strukturdefektenorangepeel. Målet är även att utvärdera hur konsekventa olika förklaringsmetoder är.Efter att datat förarbetats och nya variabler skapats, t.ex. förändringar som gjorts, trä-nades och testades tre maskinlärningsmodeller. En linjär modell kan tolkas genomdess koefficienter. En vanlig metod för att globalt förklara trädbaserade modeller ärMDI. SHAP är en modern modelloberoende metod som kan förklara modeller bådeglobalt och lokalt. Dessa tre förklaringsmetoder jämfördes sedan för att utvärdera hurkonsekventa de var i sina förklaringar. Om SHAP skulle vara konsekvent med de andrapå en global nivå, kan det argumenteras för att SHAP kan användas lokalt i en rotorsak-analys.Studien visade att koefficienterna och MDI var konsekventa med SHAP då den över-gripande korrelationen mellan dem var hög samt att metoderna tenderade att viktavariablerna på ett liknande sätt. Genom denna slutsats utvecklades en rotorsakanalysal-goritm med SHAP som lokal förklaringsmetod. Slutligen går det inte att dra någonslutsats om att det finns ett samband mellan sensordatat ochorange peel, eftersom förän-dringarna i processen var de mest betydande variablerna.

Cette thèse explore de nouvelles approches d'apprentissage profond pour l'analyse des données 3D du monde réel. Le traitement des données 3D est utile pour de nombreuses applications telles que la conduite autonome, la gestion du territoire, la surveillance des installations industrielles, l'inventaire forestier et la mesure de biomasse. Cependant, l'annotation et l'analyse des données 3D peuvent être exigeantes. En particulier, il est souvent difficile de respecter des contraintes liées à l'utilisation des ressources de calcul ou à l'efficacité de l'annotation. La difficulté d'interpréter et de comprendre le fonctionnement interne des modèles d'apprentissage profond peut également limiter leur adoption.Des efforts considérables ont été déployés pour concevoir des méthodes d'analyse des données 3D, afin d'effectuer des tâches telles que la classification des formes ou la segmentation et la décomposition de scènes. Les premières analyses automatisées s'appuyaient sur des descripteurs créés à la main et incorporaient des connaissances préalables sur les acquisitions du monde réel. Les techniques modernes d'apprentissage profond ont de meilleures performances, mais, sont souvent coûteuses en calcul, dépendent de grands ensembles de données annotées, et sont peu interprétables. Les contributions de cette thèse répondent à ces limitations.La première contribution est une architecture d'apprentissage profond pour l’analyse efficace de séquences LiDAR en temps réel. Notre approche prend en compte la géométrie d'acquisition des capteurs LiDAR rotatifs, que de nombreuses pipelines de conduite autonome utilisent. Par rapport aux travaux antérieurs, qui considèrent les rotations complètes des capteurs LiDAR individuellement, notre modèle traite l'acquisition par petits incréments. L'architecture que nous proposons à une performance comparable à celle des meilleures méthodes, tout en réduisant le temps de traitement de plus de cinq fois, et la taille du modèle de plus de cinquante fois.La deuxième contribution est une méthode d'apprentissage profond permettant de résumer de vastes collections de formes 3D à l'aide d'un petit ensemble de formes 3D. Nous apprenons un faible nombre de formes prototypiques 3D qui sont alignées et déformées pour reconstruire les nuages de points d'entrée. Notre représentation compacte et interprétable des collections de formes 3D permet d'obtenir des résultats à l'état de l'art de la segmentation sémantique avec peu d'exemples annotés.La troisième contribution développe l'analyse non supervisée pour la décomposition de scans 3D du monde réel en parties interprétables. Nous introduisons un modèle de reconstruction probabiliste permettant de décomposer un nuage de points 3D à l'aide d'un petit ensemble de formes prototypiques apprises. Nous surpassons les méthodes non supervisées les plus récentes en termes de précision de décomposition, tout en produisant des représentations visuellement interprétables. Nous offrons des avantages significatifs par rapport aux approches existantes car notre modèle ne nécessite pas d'annotations lors de l'entraînement.Cette thèse présente également deux jeux de données annotés du monde réel en accès libre, HelixNet et Earth Parser Dataset, acquis respectivement avec des LiDAR terrestres et aériens. HelixNet est le plus grand jeu de données LiDAR de conduite autonome avec des annotations denses, et fournit les métadonnées du capteur pour chaque points, cruciales pour mesurer précisément la latence des méthodes de segmentation sémantique. Le Earth Parser Dataset se compose de sept scènes LiDAR aériennes, qui peuvent être utilisées pour évaluer les performances des techniques de traitement 3D dans divers environnements.Nous espérons que ces jeux de données, et ces méthodes fiables tenant compte des spécificités des acquisitions dans le monde réel, encourageront la poursuite de la recherche vers des modèles plus efficaces et plus interprétables
This thesis explores new deep-learning approaches for modeling and analyzing real-world 3D data. 3D data processing is helpful for numerous high-impact applications such as autonomous driving, territory management, industry facilities monitoring, forest inventory, and biomass measurement. However, annotating and analyzing 3D data can be demanding. Specifically, matching constraints regarding computing resources or annotation efficiency is often challenging. The difficulty of interpreting and understanding the inner workings of deep learning models can also limit their adoption.The computer vision community has made significant efforts to design methods to analyze 3D data, to perform tasks such as shape classification, scene segmentation, and scene decomposition. Early automated analysis relied on hand-crafted descriptors and incorporated prior knowledge about real-world acquisitions. Modern deep learning techniques demonstrate the best performances but are often computationally expensive, rely on large annotated datasets, and have low interpretability. In this thesis, we propose contributions that address these limitations.The first contribution of this thesis is an efficient deep-learning architecture for analyzing LiDAR sequences in real time. Our approach explicitly considers the acquisition geometry of rotating LiDAR sensors, which many autonomous driving perception pipelines use. Compared to previous work, which considers complete LiDAR rotations individually, our model processes the acquisition in smaller increments. Our proposed architecture achieves accuracy on par with the best methods while reducing processing time by more than five times and model size by more than fifty times.The second contribution is a deep learning method to summarize extensive 3D shape collections with a small set of 3D template shapes. We learn end-to-end a small number of 3D prototypical shapes that are aligned and deformed to reconstruct input point clouds. The main advantage of our approach is that its representations are in the 3D space and can be viewed and manipulated. They constitute a compact and interpretable representation of 3D shape collections and facilitate annotation, leading to emph{state-of-the-art} results for few-shot semantic segmentation.The third contribution further expands unsupervised analysis for parsing large real-world 3D scans into interpretable parts. We introduce a probabilistic reconstruction model to decompose an input 3D point cloud using a small set of learned prototypical shapes. Our network determines the number of prototypes to use to reconstruct each scene. We outperform emph{state-of-the-art} unsupervised methods in terms of decomposition accuracy while remaining visually interpretable. We offer significant advantages over existing approaches as our model does not require manual annotations.This thesis also introduces two open-access annotated real-world datasets, HelixNet and the Earth Parser Dataset, acquired with terrestrial and aerial LiDARs, respectively. HelixNet is the largest LiDAR autonomous driving dataset with dense annotations and provides point-level sensor metadata crucial for precisely measuring the latency of semantic segmentation methods. The Earth Parser Dataset consists of seven aerial LiDAR scenes, which can be used to evaluate 3D processing techniques' performances in diverse environments.We hope that these datasets and reliable methods considering the specificities of real-world acquisitions will encourage further research toward more efficient and interpretable models

Diploma thesis describes the measurement and evaluation of displacements and deformations of the wooden roof structure of the aquapark hall in Brno – Kohoutovice. Part of the work is devoted to the preparation and testing of used devices and tools. The main part of the work consists of performing one stage of measurement using the polar method and the laser scanning method. The polar method measurement is compared with the results of the polar method of the previous stage. The next part of the work deals with the comparison of the polar method and the laser scanning method measured in the last stage. The results achieved are interpreted in the final part of the work.

The thesis has been written on the basis of main architectural concepts and their application in real life within a historical context investigation. Main architectural concepts are represented in a harmony of architectural composition that deals with relations among form, structure and space in the unique art work. Genius loci play very important role in the architectural creation - it represents a special and extra-ordinary urban locality and its architectural value in the historical, geographical and cultural context. The origin of Ostravice village within the historical frame concept of Moravia domain is the topic of this thesis. Other historical documents gave information about Bruneswerde as the Brno Castle. That means Brno Castle had to be founded not in Brno (as it has been consid-ered for many years) but it was located in Beskydy at Ostravice. The sacred city – Civitas Dei had been located in Bruneswerde region in early ages as the centre of European culture, education and spiritual life. Civitas Dei – divine Jerusalem in the transcription of St. Augustine's book De civitate Dei – is not only glorification of God and religious fantasy. It is also one of significant clues for recognition of historical architecture. The depiction of unknown settlement from the book of unde-fined origin is a superb testimony about extinct architectural works. They are legible from cadastral maps. Brno Castle - residence of nobility and power served shelter to St. Vojtěch, St. Prokop, St. Václav and St. Ludmila as it is obvious for the mentioned picture. Three major temples, three com-position axes symbolized by three towers on coins are in analogical relation to Brno triangle of four saints - the Saint family of Brno temples. Powerful ambitions of Brno City principals and clergy are inscribed into the urban structure in the way of composition relations which are legible to them who devoted themselves to the mystery of harmony. The absence of historical continuity affects personal attitudes and also identity of the whole community. The architecture truly reflects the past state of polis and it is eloquent even after its death. The architectural composition relations influence the natural environment for long time, longer than the architectural work existence. The geometric order of Renaissance Brno existence has not been in attention of architects so far. The features were discovered in characteristic paintings by Albrecht Durer. They are evidently secret works of the genius. A meaningful collaborator and follower in the extensive project was also his friend Jan Čert from Brno and lately from Vienna. His noble genealogy played a significant role in history of Silesia and Moravia for centuries. It is tendency to consider him as Austrian or even German architect. It is because of the fact that the genealogy tree of his noble family had roots in Moravia. It is supposed that Durer with Jan Čert´s support created the extraordinarily monumental architectural and urban works in Brno. A remarkable consensus in the urban composition of two squares and transition of traditional urban structure of Ostravice Civitas Dei into the modern Brno is also confirmed due to the identification of noble creators and owners, who were at the foundation, transformation and extinction of elements of the Moravian urban structure.

Decision support systems based on machine learning (ML) techniques are excelling in most artificial intelligence (AI) fields, over-performing other AI methods, as well as humans. However, challenges still exist that do not favour the dominance of AI in some applications. This proposal focuses on a critical one: lack of transparency and explainability, reducing trust and accountability of an AI system. The fact that most AI methods still operate as complex black boxes, makes the inner processes which sustain their predictions still unattainable. The awareness around these observations foster the need to regulate many sensitive domains where AI has been applied in order to interpret, explain and audit the reliability of the ML based systems. Although modern-day biometric recognition (BR) systems are already benefiting from the performance gains achieved with AI (which can account for and learn subtle changes in the person to be authenticated or statistical mismatches between samples), it is still in the dark ages of black box models, without reaping the benefits of the mismatches between samples), it is still in the dark ages of black box models, without reaping the benefits of the XAI field. This work will focus on studying AI explainability in the field of biometrics focusing in particular use cases in BR, such as verification/ identification of individuals and liveness detection (LD) (aka, antispoofing). The main goals of this work are: i) to become acquainted with the state-of-the-art in explainability and biometric recognition and PAD methods; ii) to develop an experimental work xxxxx Tasks 1st semester (1) Study of the state of the art- bibliography review on state of the art for presentation attack detection (2) Get acquainted with the previous work of the group in the topic (3) Data preparation and data pre-processing (3) Define the experimental protocol, including performance metrics (4) Perform baseline experiments (5) Write monography Tasks 2nd semester (1) Update on the state of the art (2) Data preparation and data pre-processing (3) Propose and implement a methodology for interpretability in biometrics (4) Evaluation of the performance and comparison with baseline and state of the art approaches (5) Dissertation writing Referências bibliográficas principais: (*) [Doshi17] B. Kim and F. Doshi-Velez, "Interpretable machine learning: The fuss, the concrete and the questions," 2017 [Mol19] Christoph Molnar. Interpretable Machine Learning. 2019 [Sei18] C. Seibold, W. Samek, A. Hilsmann, and P. Eisert, "Accurate and robust neural networks for security related applications exampled by face morphing attacks," arXiv preprint arXiv:1806.04265, 2018 [Seq20] Sequeira, Ana F., João T. Pinto, Wilson Silva, Tiago Gonçalves and Cardoso, Jaime S., "Interpretable Biometrics: Should We Rethink How Presentation Attack Detection is Evaluated?", 8th IWBF2020 [Wilson18] W. Silva, K. Fernandes, M. J. Cardoso, and J. S. Cardoso, "Towards complementary explanations using deep neural networks," in Understanding and Interpreting Machine Learning in MICA. Springer, 2018 [Wilson19] W. Silva, K. Fernandes, and J. S. Cardoso, "How to produce complementary explanations using an Ensemble Model," in IJCNN. 2019 [Wilson19A] W. Silva, M. J. Cardoso, and J. S. Cardoso, "Image captioning as a proxy for Explainable Decisions" in Understanding and Interpreting Machine Learning in MICA, 2019 (Submitted)

Machine Learning (ML) has been increasingly used to aid humans making high-stakes decisions in a wide range of areas, from public policy to criminal justice, education, healthcare, or financial services. However, it is very hard for humans to grasp the rationale behind every ML model’s prediction, hindering trust in the system. The field of Explainable Artificial Intelligence (XAI) emerged to tackle this problem, aiming to research and develop methods to make those “black-boxes” more interpretable, but there is still no major breakthrough. Additionally, the most popular explanation methods — LIME and SHAP — produce very low-level feature attribution explanations, being of limited usefulness to personas without any ML knowledge. This work was developed at Feedzai, a fintech company that uses ML to prevent financial crime. One of the main Feedzai products is a case management application used by fraud analysts to review suspicious financial transactions flagged by the ML models. Fraud analysts are domain experts trained to look for suspicious evidence in transactions but they do not have ML knowledge, and consequently, current XAI methods do not suit their information needs. To address this, we present JOEL, a neural network-based framework to jointly learn a decision-making task and associated domain knowledge explanations. JOEL is tailored to human-in-the-loop domain experts that lack deep technical ML knowledge, providing high-level insights about the model’s predictions that very much resemble the experts’ own reasoning. Moreover, by collecting the domain feedback from a pool of certified experts (human teaching), we promote seamless and better quality explanations. Lastly, we resort to semantic mappings between legacy expert systems and domain taxonomies to automatically annotate a bootstrap training set, overcoming the absence of concept-based human annotations. We validate JOEL empirically on a real-world fraud detection dataset, at Feedzai. We show that JOEL can generalize the explanations from the bootstrap dataset. Furthermore, obtained results indicate that human teaching is able to further improve the explanations prediction quality.
A Aprendizagem de Máquina (AM) tem sido cada vez mais utilizada para ajudar os humanos a tomar decisões de alto risco numa vasta gama de áreas, desde política até à justiça criminal, educação, saúde e serviços financeiros. Porém, é muito difícil para os humanos perceber a razão da decisão do modelo de AM, prejudicando assim a confiança no sistema. O campo da Inteligência Artificial Explicável (IAE) surgiu para enfrentar este problema, visando desenvolver métodos para tornar as “caixas-pretas” mais interpretáveis, embora ainda sem grande avanço. Além disso, os métodos de explicação mais populares — LIME and SHAP — produzem explicações de muito baixo nível, sendo de utilidade limitada para pessoas sem conhecimento de AM. Este trabalho foi desenvolvido na Feedzai, a fintech que usa a AM para prevenir crimes financeiros. Um dos produtos da Feedzai é uma aplicação de gestão de casos, usada por analistas de fraude. Estes são especialistas no domínio treinados para procurar evidências suspeitas em transações financeiras, contudo não tendo o conhecimento em AM, os métodos de IAE atuais não satisfazem as suas necessidades de informação. Para resolver isso, apresentamos JOEL, a framework baseada em rede neuronal para aprender conjuntamente a tarefa de tomada de decisão e as explicações associadas. A JOEL é orientada a especialistas de domínio que não têm conhecimento técnico profundo de AM, fornecendo informações de alto nível sobre as previsões do modelo, que muito se assemelham ao raciocínio dos próprios especialistas. Ademais, ao recolher o feedback de especialistas certificados (ensino humano), promovemos explicações contínuas e de melhor qualidade. Por último, recorremos a mapeamentos semânticos entre sistemas legados e taxonomias de domínio para anotar automaticamente um conjunto de dados, superando a ausência de anotações humanas baseadas em conceitos. Validamos a JOEL empiricamente em um conjunto de dados de detecção de fraude do mundo real, na Feedzai. Mostramos que a JOEL pode generalizar as explicações aprendidas no conjunto de dados inicial e que o ensino humano é capaz de melhorar a qualidade da previsão das explicações.

As the complexity of our data increased exponentially in the last decades, so has our need for interpretable features. This thesis revolves around two paradigms to approach this quest for insights. In the first part we focus on parametric models, where the problem of interpretability can be seen as a “parametrization selection”. We introduce a quantile-centric parametrization and we show the advantages of our proposal in the context of regression, where it allows to bridge the gap between classical generalized linear (mixed) models and increasingly popular quantile methods. The second part of the thesis, concerned with topological learning, tackles the problem from a non-parametric perspective. As topology can be thought of as a way of characterizing data in terms of their connectivity structure, it allows to represent complex and possibly high dimensional through few features, such as the number of connected components, loops and voids. We illustrate how the emerging branch of statistics devoted to recovering topological structures in the data, Topological Data Analysis, can be exploited both for exploratory and inferential purposes with a special emphasis on kernels that preserve the topological information in the data. Finally, we show with an application how these two approaches can borrow strength from one another in the identification and description of brain activity through fMRI data from the ABIDE project.

碩士
國立臺灣大學
統計碩士學位學程
105
Despite the abundance of the available variables, ground truth is privy to knowledge about the problem seldom revealed in practice. By discovering important features, researchers can further conduct a more targeted follow-up experiment on the selected features tailored for understanding the scientific phenomenon. A natural requirement is that we wish to discover as many relevant variables as possible and make as few mistakes as possible at the same time. We propose a modified RuleFit with FDR control by knockoff procedure and with alpha control by Neyman-Pearson method.

3 Abstract Gravimetry is an enormously expanding geophysical exploration method during the last decade. Unfortunately the gravity inversion ambiguity problem introduces the necessity to constrain gravity model parameters by other independent data. Overview of gravity data processing and interpretation is presented. All stages of the project lifecycle are discussed with emphasis to the methodological aspects and new challenges introduced by common use of modern digital zero-length spring gravimeters. Special attention is drawn on approaches used in the presented case histories projects. Description of the interpretation stage of the project concentrates on constraining of gravity model parameters with a complex of geophysical and geological data in order to reduce ambiguity of the inversion process in gravimetry. In addition, approaches used for delineation of tectonic structures, an important aspect of the interpretation stage, are described. Three case histories examples are presented to demonstrate methodological aspects of the interpretation of gravity data. They are focused on environmental problems and tectonic structure assessment. The first project represents an attempt to use micro-gravity for determination of internal structure of an undocumented sealed landfill. The second project demonstrates...

- Handling the problem of deconstruction of the law in perspective of the interpretation of the law methodology The target of this thesis is firstly to describe the concept of the deconstruction of the law - the phenomenon of the last and probably also future decades. Secondly, the thesis describes the influence of the deconstruction of the law on the interpretation of the law. The thesis is dived into the three main parts and the conclusion. The concept of the deconstruction of the law and its causes are described in the first part. The deconstruction of the law is a process changing quality of legal order. As a consequence of that interpretational difficulties come in specific situations (or more precisely - it is not sufficient to infer a legal rule from regulatory legislation just by standard interpretative methods). This process has its several causes. Decomposition of religion and ethics (as the two significant social regulatory systems) and perception of the law as a technical instrument guiding the society, they both caused the extension of the law. If the law is now understood as the technical instrument guiding the society, then an acceleration of social development results in the acceleration of changes in legal rules. The extension and dynamics of the law they both negatively influence...

The objective of the PhD-thesis is analytical and comparative description of the main theoretical and philosophical approaches to the issue of legal interpretation, focusing on practical application. Initially, attention is paid to explanation of the concept of legal interpretation itself, focusing also on the practical meaning of differentiation between each designatum of the concept of legal interpretation as well as the reflection thereof in the applicable law. Other parts of the thesis include an introduction to the general typology of the theories of legal interpretation followed by a more detailed analysis of the three basic contemporary theories (philosophies) of legal interpretation, namely analytical approach (both the so-called horse-shoe analysis and soft-shoe analysis), hermeneutic approach (attention is paid to the traditional as well as modern hermeneutics, both the phenomenological and methodological branch), and discursive approach (represented by the rhetorical-topical view of discourse as well as the procedural concept of discourse). The PhD-thesis results in the pronunciation of a syncretic opinion that the different theories of interpretation of law should be explained and taught in mutual context because this dialogic approach provides a guarantee that their conclusions and...

The aim of this diploma thesis entitled "Didactic interpretation of Čapek's works in literary education for the 2nd stage of primary education" is to chart how the works of Karel Čapek are represented in each series of reading books for the 2nd stage of primary education. It will also further evaluate the choice of works represented and introduce the fundamental concepts in Čapek's work. The thesis is divided into many parts. The first of these is concerned with the current state of literary education, the question of literacy, and other individuals who have an effect on the process of literary education. Furthermore, recorded methods that are used in literacy lessons are outlined. The methods that focus principally on the development of literacy are described in more detail. On the basis of classic literature a didactic interpretation is described, its situation, function, determinants and policies. The thesis concisely describes Čapek's works. A key part of the thesis is made up of the analysis of reading books from many different publishers Alter, Fraus, Nová škola, Prodos and SPN. The thesis focuses not only on how many snippets of the work of Karel Čapek are found in any given reading book, but also takes into account the representation of the genre and the ideas presented alongside the texts. After evaluation, the actual concept that the entirety of Čapek's work attempts to cover is introduced.

The diploma thesis is focused on the development of interpretation of violin playing at the advanced level at music schools. The aim of the thesis is to go through different periods of music development and to deepen the quality of means of reproduction and develop basic principles of interpretation in advanced students on the basis of several selected compositions. Another objective is to handle the practising process of each composition creatively. The work is oriented practically and may become useful for violin teachers and advanced students. The problems connected with violin playing are treated from the point of view of performance. The teaching process goes from imagination to realization where the technical skills are means for the realization of imagination and, simultaneously, the final performance is a great deal influenced by the player's technical skills. Also, the possibilities how to develop musical imagination and accomplish successful realization of the process mentioned above are discussed. The central themes of the thesis are the development of interpretation of violin playing and the problems connected with it, both theoretical and practical. A scheme for the teaching process is defined and the possible ways of application of the scheme during practice are shown on examples of...

This diploma thesis aims to analyze the issue of one of the most significant congressional powers found in Article I., Section 8 of the U.S. Constitution. Congressional power to regulate interstate commerce has been instrumental to the federal government's legislative efforts in many areas of law. This constitutional provision enabled the Congress to react to the changing conditions and new problems the country has been facing, especially in the area of working conditions, civil rights, criminal justice or even environmental law and many others. The expansion of power of the federal government, however, was not always greeted with enthusiasm, especially in the first three decades of the 20th century, before the Supreme Court began to read the commerce power much more broadly, to the point that it ceased to be a factual limitation of its powers. This trend was meant to be stopped by the New Federalism movement and the five new conservative justices who issued rulings that limited the scope of the Commerce Clause. However, this group of justices proved to be very inconsistent in its own approach toward this constitutional provision and eventually fell apart, which rendered Rehnquist's attempted constitutional revolution with respect to state's rights partly a failure. As the new Court membership under...

Submitted Ph.D. thesis is concerning the application and interpretation of seismic surface waves in a broad range of frequencies and scales. Using surface waves as a supplement to the methods dealing with body waves seems to be worth the effort. Surface wave interpretation can be used to obtain new information about the studied medium and simultaneously it can overcome, in some cases, the limitations of other seismic techniques. Moreover, surface waves are usually present on measured records and hence for its usage it is not necessary to modify the standard measuring procedures. One of the results of this thesis is an original algorithm for dispersive waveform calculation. The program works in an arbitrary range of frequencies and scales. The input parameter for the calculation is the dispersion curve. In this point the algorithm differs from all other approaches used so far. Algorithm is based on a summation of frequency components with shifts corresponding to the velocity dispersion and distance. The resulting waveform only contains an individual dispersive wave of the selected mode, thus being particularly suitable for testing of methodologies for dispersive wave analysis. The algorithm was implemented into the program DISECA. Furthermore, a new procedure was designed to calculate the dispersion...

The dissertation is be concerned with the comparison of two Czech translations of novel "The Castle" by Franz Kafka from the linguistic viewpoint. The dissertation compares text translation by Vladimír Kafka of 1969 and Jana Zoubková of 2014. The dissertation focuses on the methodes of translation, demands, that are placed on a literary translations. The dissertation focuses on a lifetime period of translations and their critism and last but not least, some problems of modern text translation. After closer exploration and analyses of problems of translation our dissertation will evaluate the strengths and weakness of both our translations. Key words: Translation, language, interpretation, mistake, method, critism, equivalent, idiom

This diploma thesis describes historical evolution of calculation of sphere's volume and surface and provides an analysis of textbooks for secondary and primary schools. It is made with the intention to inspire high school teachers with various approaches of teaching the volume and surface of solid bodies. It can help teachers with motivation of students as well as with selection of textbook and teaching methods for the issue. This thesis is meant to inspire high school students interested in history of mathematics, too. It includes analysis of preserved exercises on the topic from ancient Egypt and Mesopotamia as well as findings from Archimedes' works, which were devoted to this topic. Moreover it describes contribution of enlighteners on the subject and shows exact procedures of derivation of formulas using integral calculus.

In the history of exegesis 1 Corinthians 14:34-35 and 1 Timothy 2:11-15 have functioned as the heavy artillery against women in the ministries. From the 4th century BC, when the Church really started to develop from a dynamic underground movement of believers to a state Church organised in the image of the Roman Empire and so became the Catholic Church, women were more and more suppressed under the influence of deeply rooted Hellenistic anthropological ideas that were read into these passages. Only in the second half of the Twentieth Century, under the influence of changes in society after the sexual and feminist revolutions, changes set in that sparked the discussion about the role of women in the Church. This discussion is still continuing – in the Reformed tradition at least. These two passages, however, seem to oppose an overwhelming number of biblical themes and data that at least bring a strong nuance to the picture the two passages seem to portray. The creation of man and woman in the image of God and the protection for women against the arbitrariness of men clearly picture an original and principal equality of men and women. In the circle of disciples around Jesus Christ this becomes even more manifest. This attitude is also visible in the earliest churches. Paul expresses this in Galatians 3:28: In Christ there is no … male nor female. The passages that seem to limit the rights of women in the Church do not actually oppose this picture, but show that for the sake of the proclamation of the gospel not everything is (immediately) allowed. This dissertation attempted to interpret these two passages with the help of the grammatical-historic method. In 1 Corinthians 14:34-35 Paul personally intervenes in the apparent chaotic meetings of the cosmopolitan and charismatic congregation. He requires from different categories of participants – among them the married women that have a Christian husband – to not burden the fellowship with – in the case of the aforementioned women – their (otherwise rightful) participation in the discussion of the prophetic message during the worship service. In 1 Timothy 2:11-15 he gives his friend and student Timothy, the young pastor and teacher of the congregation in Ephesus, tools to call upon the members of the congregation not to start a revolution but to conquer the world for Christ by living an exemplary life. The context of this directive is a heresy that was particularly influencing some women that developed a prominent and domineering attitude in the Church. None of these directives or instructions of Paul talk about special ministries in the Church. It is all about attitude. For the sake of the steady progress of the gospel this attitude is to be determined by discipline and humbleness. However, the form this discipline and humbleness have largely depends on the context of the believers.
MTh (New Testament), North-West University, Potchefstroom Campus, 2014

Presented master thesis tackles common piano practicing issues among students at the first stage of piano learning. The main purpose of the study is to investigate the difficulties that block students' progress, and analyze proper approaches that improve students' musical capabilities. The practicing issues are scientifically categorized into three categories: technical, interpretative and general practicing and performing issues, and are further analyzed from physical, psychological, and most importantly, educational perspective. In addition, contemporary piano teaching methods from various countries are explored and various methodologies and teaching philosophies are investigated and presented. Effective approaches, suitable strategies and practicing suggestions are provided in the study for both teachers and learners. Chinese piano educators can use this study to enrich their teaching methodologies and to develop new piano teaching methods.

Matouš Holeka's dissertation examines various interpretation methods of the Scriptures by theologians in the Roman-Catholic, Orthodox and Protestant traditions. These interpretation methods can be most clearly identified in the relationship between the Scriptures and tradition, in historical-critical methodology, and in political-cultural readings of the biblical texts. In this dissertation, these themes are shown as representing different ways of common critical reading of the sacred texts. As well as the historical-critical approach, there are also pre-critical and ideological readings. The dissertation starts by looking at this diversity in documents of the World Council of Churches. It shows that the differences in understanding the Scriptures are no longer connected primarily to the specific views of the respective traditions, but rather that similarities or differences in readings across churches. Christians within these traditions refer to their own sources: in the case of the Roman-Catholics to Church documents, in the case of the Orthodox to the Church Fathers and as far as Protestants are concerned, to the Confessions. However, their reception in each of these traditions only confirms the diversity of readings. This is also true for the theological interpretation of the Scriptures, which...

The work confronts prosaic pieces with noticeable autobiographical features. It deals namely with works of Edgar Dutka, Eliška Vlasáková and Antonín Bajaja. It was not a coincidence to choose right these three Czech authors. They have quite a few features in common. They all belong to the same generation. They are born in the 40's of the 20th century, so they have spent an essential part of their lives (childhood, adolescence and substantial part of their working age) in the second half of the 20th century - in the time of political convulsions and social changes. A childhood spent during this period has become their common literary theme. They all make their memories a literary piece. Each of them does it in a different way and with a different amount of accuracy. Their inspiration comes from their real life, from their memories. The next common feature of these three authors is that they all wrote some of their texts without a vision to make them public. They wrote them for themselves. Book editions were carried out after the year 2000. Last but not least mutual fact is that all of these three authors have been nominated for a prestigious literary price. Edgar Dutka's works show the interdigitation of fictional worlds and also the diffusion of these worlds in the real life of the author. Eliška...

This thesis considers two questions, "How children aged 12 to 15 years rethinking of biblical texts? And how can using the Philosophy for Children stimulate thinking about faith of children aged 12 to 15 years?" To answer these questions, I chose the qualitative research, which consisted in various discussions in selected focus groups. This dialogue was guided by the principles of philosophy for children. The theoretical part of my thesis deals with the definition of important concepts, such as religious education, catechesis from the perspective of various documents of the Church and other educational programs, and specifying the key differences between these two concepts, define the structure and principles of philosophy for children, but also the historical development of this program. The practical part deals with the qualitative research in focus groups. This research includes discussion of selected focus groups, where individual statements are encoded and then included in the table of frequencies. According to these tables discussions is evaluated in terms of statements frequencies. Subsequently the content of discussions is also evaluated. The codes are divided into categories from which discussions are evaluated in the final summary with answers to the key questions.