Institutes
Refine
Year of publication
Document Type
- Doctoral Thesis (91)
- Article (59)
- Bachelor Thesis (18)
- Book (16)
- Master's Thesis (10)
- Conference Proceeding (4)
- Contribution to a Periodical (4)
- Habilitation (2)
- Preprint (2)
- Diploma Thesis (1)
Has Fulltext
- yes (207)
Is part of the Bibliography
- no (207)
Keywords
- Machine Learning (5)
- NLP (5)
- ALICE (3)
- Annotation (3)
- Machine learning (3)
- Text2Scene (3)
- TextAnnotator (3)
- Virtual Reality (3)
- mathematics education (3)
- Artificial intelligence (2)
Institute
- Informatik und Mathematik (207) (remove)
Diese Bachelorarbeit befasst sich mit der Themenklassifikation von unstrukturiertem Text. Aufgrund der stetig steigenden Menge von textbasierten Daten werden automatisierte Klassifikationsmethoden in vielen Disziplinen benötigt und erforscht. Aufbauend auf dem text2ddc-Klassifikator, der am Text Technology Lab der Goethe-Universität Frankfurt am Main entwickelt wurde, werden die Auswirkungen der Vergrößerung des Trainingskorpus mittels unterschiedlicher Methoden untersucht. text2ddc nutzt die Dewey Decimal Classification (DDC) als Zielklassifikation und wird trainiert auf Artikeln der Wikipedia. Nach einer Einführung, in der Grundlagen beschrieben werden, wird das Klassifikationsmodell von text2ddc vorgestellt, sowie die Probleme und daraus resultierenden Aufgaben betrachtet. Danach wird die Aktualisierung der bisherigen Daten beschrieben, gefolgt von der Vorstellung der verschiedenen Methoden, das Trainingskorpus zu erweitern. Mit insgesamt elf Sprachen wird experimentiert. Die Evaluation zeigt abschließend die Verbesserungen der Qualität der Klassifikation mit text2ddc auf, diskutiert die problematischen Fälle und gibt Anregungen für weitere zukünftige Arbeiten.
Aufgrund der §§20, 44 Abs. 1 Nr. 1 des Hessischen Hochschulgesetzes in der Fassung vom 14. Dezember 2009 (GVBl. I, S. 666), zuletzt geändert durch Art. 2 des Gesetzes vom 18. Dezember 2017 (GVBl. I, S. 284), hat der Fachbereichsrat des Fachbereichs Informatik und Mathematik der Johann Wolfgang Goethe-Universität Frankfurt am Main am 25. Mai 2020 die folgende Ordnung für den Bachelorstudiengang Mathematik beschlossen. Diese Ordnung hat das Präsidium der Goethe-Universität gemäß §37 Abs. 5 Hessisches Hochschulgesetz am 30. Juni 2020 genehmigt. Sie wird hiermit bekannt gemacht.
Das Ziel dieser Arbeit ist die realitätsgetreue Entwicklung eines interaktiven 3D-Stadtmodells, welches auf den ÖPNV zugeschnitten ist. Dabei soll das Programm anhand von Benutzereingaben und mit Hilfe einer Datenquelle, automatisch eine dreidimensionale Visualisierung der Gebäude erzeugen und den lokalen ÖPNV mitintegrieren. Als Beispiel der Ausarbeitung diente das ÖPNV-Netz der Stadt Frankfurt. Hierbei wurde auf die Problematik der Erhebung von Geoinformationen und der Verarbeitung von solchen komplexen Daten eingegangen. Es wurde ermittelt, welche Nutzergruppen einen Mehrwert durch eine derartige 3D Visualisierung haben und welche neuen Erweiterungs- und Nutzungspotenziale das Modell bietet.
Dem Leser soll insbesondere ein Einblick in die Generierung von interaktiven 3D-Modellen aus reinen Rohdaten verschafft werden. Dazu wurde als Entwicklungsumgebung die Spiele-Engine Unity eingesetzt, welche sich als sehr fähiges und modernes Entwicklungswerkzeug bei der Erstellung von funktionalen 3D-Visualisierungen herausgestellt hat. Als Datenquelle wurde das OpenStreetMap Projekt benutzt und im Rahmen dieser Arbeit behandelt. Anschließend wurde zur Evaluation, das Modell verschiedenen Nutzern bereitgestellt und anhand eines Fragebogens evaluiert.
The World Wide Web is increasing the number of freely accessible textual data, which has led to an increasing interest in research in the field of computational linguistics (CL). This area of research addresses theoretical research to answer the question of how language and knowledge must be represented in order to understand and produce language. For this purpose, mathematical models are being developed to capture the phenomena at various levels in human languages. Another field of research experiencing an increase in interest that is closely related to CL is Natural Language Processing (NLP), which is primarily concerned with developing effective and efficient data structures and algorithms that implement the mathematical models of CL.
With increasing interest in these areas, NLP tools are rapidly and frequently being developed incorporating different CL models to handle different levels of language. The open source trend has benefited all those in the scientific community who develop and use these tools. Due to yet undefined I/O standards for NLP, however, the rapid growth leads to a heterogeneous NLP landscape in which the specializations of the tools cannot benefit from each other because of interface incompatibility. In addition, the constantly growing amount of freely accessible text data requires a high-performance processing solution. This performance can be achieved by horizontal and vertical scaling of hardware and software. For these reasons the first part of this thesis deals with the homogenization of the NLP tool landscape, which is achieved by a standardized framework called TextImager. It is a cloud computing based multi-service, multi-server, multi-pipeline, multi-database, multi-user, multi-representation and multi-visual framework that already provides a variety of tools for various languages to process various levels of linguistic complexity. This makes it possible to answer research questions that require the processing of a large amount of data at several linguistic levels.
The integrated tools and the homogenized I/O data streams of the TextImager make it possible to combine the built-in tools in two dimensions: (1) the horizontal dimension to achieve NLP task-specific improvement (2) the orthogonal dimension to implement CL models that are based on multiple linguistic levels and thus rely on a combination of different NLP tools. The second part of this thesis therefore deals with the creation of models for the horizontal combination of tools in order to show the possibilities for improvement using the example of the NLP task of Named Entity Recognition (NER). The TextImager offers several tools for each NLP task, most of which have been trained on the same training basis, but can produce different results. This means that each of the tools processes a subset of the data correctly and at the same time makes errors in another subset. In order to process as large a subset of the data as possible correctly, a horizontal combination of tools is therefore required. Machine learning-based voting mechanisms called LSTMVoter and CRFVoter were developed for this purpose, which allow a combination of the outputs of individual NLP tools so that better partial data results can be achieved. In this thesis the benefit of Voter is shown using the example of the NER task, whose results flow
back into the TextImager tool landscape.
The third part of this thesis deals with the orthogonal combination of TextImager tools to accomplish the verb sense disambiguation (VSD). The CL question is investigated, how verb senses should be modelled in order to disambiguate them computatively. Verbsenses have a syntagmatic-paradigmatic relationship with surrounding words. Therefore, preprocessing on several linguistic levels and consequently an orthogonal combination of NLP tools is required to disambiguate verbs on a computational level. With TextImager’s integrated NLP landscape, it is now possible to perform these preprocessing steps to induce the information needed for the VSD. The newly developed NLP tool for the VSD has been integrated into the TextImager tool landscape, enabling the analysis of a further linguistic level.
This thesis presents a framework that homogenizes the NLP tool landscape in a cluster-based way. Methods for combining the integrated tools are implemented to improve the analysis of a specific linguistic level or to develop tools that open up new linguistic levels.
Generic tasks for algorithms
(2020)
Due to its links to computer science (CS), teaching computational thinking (CT) often involves the handling of algorithms in activities, such as their implementation or analysis. Although there already exists a wide variety of different tasks for various learning environments in the area of computer science, there is less material available for CT. In this article, we propose so-called Generic Tasks for algorithms inspired by common programming tasks from CS education. Generic Tasks can be seen as a family of tasks with a common underlying structure, format, and aim, and can serve as best-practice examples. They thus bring many advantages, such as facilitating the process of creating new content and supporting asynchronous teaching formats. The Generic Tasks that we propose were evaluated by 14 experts in the field of Science, Technology, Engineering, and Mathematics (STEM) education. Apart from a general estimation in regard to the meaningfulness of the proposed tasks, the experts also rated which and how strongly six core CT skills are addressed by the tasks. We conclude that, even though the experts consider the tasks to be meaningful, not all CT-related skills can be specifically addressed. It is thus important to define additional tasks for CT that are detached from algorithms and programming.
Density visualization pipeline: a tool for cellular and network density visualization and analysis
(2020)
Neuron classification is an important component in analyzing network structure and quantifying the effect of neuron topology on signal processing. Current quantification and classification approaches rely on morphology projection onto lower-dimensional spaces. In this paper a 3D visualization and quantification tool is presented. The Density Visualization Pipeline (DVP) computes, visualizes and quantifies the density distribution, i.e., the “mass” of interneurons. We use the DVP to characterize and classify a set of GABAergic interneurons. Classification of GABAergic interneurons is of crucial importance to understand on the one hand their various functions and on the other hand their ubiquitous appearance in the neocortex. 3D density map visualization and projection to the one-dimensional x, y, z subspaces show a clear distinction between the studied cells, based on these metrics. The DVP can be coupled to computational studies of the behavior of neurons and networks, in which network topology information is derived from DVP information. The DVP reads common neuromorphological file formats, e.g., Neurolucida XML files, NeuroMorpho.org SWC files and plain ASCII files. Full 3D visualization and projections of the density to 1D and 2D manifolds are supported by the DVP. All routines are embedded within the visual programming IDE VRL-Studio for Java which allows the definition and rapid modification of analysis workflows.
Due to the resurrection of data-hungry models (such as deep convolutional neural nets), there is an increasing demand for large-scale labeled datasets and benchmarks in the computer vision fields (CV). However, collecting real data across diverse scene contexts along with high-quality annotations is often expensive and time-consuming, especially for detailed pixel-level label prediction tasks such as semantic segmentation, etc. To address the scarcity of real-world training sets, recent works have proposed the use of computer graphics (CG) generated data to train and/or characterize performance of modern CV systems. CG based virtual worlds provide easy access to ground truth annotations and control over scene states. Most of these works utilized training data simulated from video games and pre-designed virtual environments and demonstrated promising results. However, little effort has been devoted to the systematic generation of massive quantities of sufficiently complex synthetic scenes for training scene understanding algorithms. In this work, we develop a full pipeline for simulating large-scale datasets along with per-pixel ground truth information. Our simulation pipeline constitutes of mainly two components: (a) a stochastic scene generative model that automatically synthesizes traffic scene layouts by using marked point processes coupled with 3D CAD objects and factor potentials, (b) an annotated-image rendering tool that renders the sampled 3D scene as RGB image with a chosen rendering method along with pixel-level annotations such as semantic labels, depth, surface normals etc. This pipeline is capable of automatically generating and rendering a potentially infinite variety of outdoor traffic scenes that can be used to train convolutional neural nets (CNN).
However, several recent works, including our own initial experiments demonstrated that the CV models that are trained naively on simulated data lack generalization capabilities to real-world scenes. This opens up several fundamental questions about what is it lacking in simulated data compared to real data and how to use it effectively. Furthermore, there has been a long debate since 1980’s on the usefulness of CG generated data for tuning CV systems. Primarily, the impact of modeling errors and computational rendering approximations, due to various choices in the rendering pipeline, on trained CV systems generalization performance is still not clear. In this thesis, we take a case study in the context of traffic scenarios to empirically analyze the performance degradations when CV systems trained with virtual data are transferred to real data. We first explore system performance tradeoffs due to the choice of the rendering engine (e.g., Lambertian shader (LS), ray-tracing (RT), and Monte-Carlo path tracing (MCPT)) and their parameters. A CNN architecture, DeepLab, that performs semantic segmentation, is chosen as the CV system being evaluated. In our case study, involving traffic scenes, a CNN trained with CG data samples generated with photorealistic rendering methods (such as RT or MCPT), shows already a reasonably good performance on real-world testing data from CityScapes benchmark. Use of samples from an elementary rendering method, i.e., LS, degraded the performance of CNN by nearly 20%. This result conveys that training data must be photorealistic enough for better generalizability of the trained CNN models. Furthermore, the use of physics-based MCPT rendering improved the performance by 6% but at the cost of more than three times the rendering time. This MCPT generated dataset when augmented with just 10% of real-world training data from CityScapes dataset, the performance levels achieved are comparable to that of training CNN with the complete CityScapes dataset.
The next aspect we study in the thesis involves the impact of choice of parameter settings of scene generation model on the generalization performance of CNN models trained with the generated data. Towards this end, we first propose an algorithm to estimate our scene generation model parameters given an unlabeled real world dataset from the target domain. This unsupervised tuning approach utilizes the concept of generative adversarial training, which aims at adapting the generative model by measuring the discrepancy between generated and real data in terms of their separability in the space of a deep discriminatively-trained classifier. Our method involves an iterative estimation of the posterior density of prior distributions for the generative graphical model used in the simulation. Initially, we assume uniform distributions as priors over parameters of a scene described by our generative graphical model. As iterations proceed the uniform prior distributions are updated sequentially to distributions for the simulation model parameters that leads to simulated data with statistics that are closer to the distributions of the unlabeled target data.
...
Zielsetzung dieser Arbeit ist es Nutzern, ohne Programmierkenntnisse oder Fachwissen im Bereich der Informatik, Zugang zu der automatischen Verarbeitung von Texten zu gewährleisten. Speziell soll es um Geotagging, also das Referenzieren verschiedener Objekte auf einer Karte, gehen. Als Basis soll ein ontologisches Modell dienen, mit Hilfe dessen Struktur die Objekte in Klassen eingeteilt werden. Zur Verarbeitung des Textes werden NaturalLanguage Processing Werkzeuge verwendet. Natural Language Processing beschreibt Methoden zur maschinellen Verarbeitung natürlicher Sprache. Sie ermöglichen es, die in Texten enthaltenen unstrukturierten Informationen in eine strukturierte Form zu bringen. Die so erhaltenen Informationen können für weitere maschinelle Verarbeitungsschritte verwendet oder einem Nutzer direkt bereitgestellt werden. Sollten sie direkt bereitgestellt werden, ist es ausschlaggebend, sie in einer Form zu präsentieren, die auch ohne Fachkenntnisse oder Vorwissen verständlich ist. Im Bereich der Geographie wird oft der Ansatz befolgt, die erhaltenen Informationen auf Basis verschiedener Karten, also visuell zu verarbeiten. Visualisierungen dienen hierbei der Veranschaulichung von Informationen. Durch sie werden die relevanten Aspekte dem Nutzer verdeutlicht und so die Komplexität der Informationen reduziert. Es bietet sich also an, die durch das Natural Language Processing gesammelten Informationen in Form einer Visualisierung für den Nutzer zugänglich zu machen. Im Rahmen dieser Arbeit über Geotagging und Ontologie-basierte Visualisierung für das TextImaging wird ein Tool entwickelt, das diese Brücke schlägt. Die Texte werden auf einer Karte visualisiert und bieten so eine Möglichkeit, beschriebene geographische Zusammenhänge auf einen Blick zu erfassen. Durch die Kombination der Visualisierung auf einer Karte und der Markierung der entsprechenden Entitäten im Text kann eine zuverlässige und nutzerfreundliche Visualisierung erzeugt werden. Bei einer abschließenden Evaluation hat sich gezeigt das mit dem Tool der Zeitaufwand und die Anzahl der fehlerhaften Annotationen reduziert werden konnte.Die von dem Tool gebotenen Funktionen machen dieses auch für weiterführende Arbeiten interessant. Eine Möglichkeit ist die entwickelten Annotatoren zu verwenden um ein ontology matching auf Basis bestimmter Texte auszuführen. Im Bereich der Visualisierung bieten sich Projekte wie die Visualisierung historischer Texte auf Basis automatisch ermittelter, zeitgerechter Karten an.
Neuropsychiatric disorders are complex, highly heritable but incompletely understood disorders. The clinical and genetic heterogeneity of these disorders poses a significant challenge to the identification of disorder related biomarkers. Besides significant progress in unveiling the genetic basis of these disorders, the underlying causes and biological mechanisms remain obscure. With the advancement in the array, sequencing, and big data technologies, a huge amount of data is generated from individuals across different platforms and in various data structures. But there is a paucity of bioinformatics tools that can integrate this plethora of data. Therefore, there is a need to develop an integrative bioinformatics data analysis tool that combines biological and clinical data from different data types to better understand the underlying genetics.
This thesis presents a bioinformatics pipeline implementing data from different platforms to provide a thorough understanding of the genetic etiology of a neuropsychiatric quantitative as well as a qualitative trait of interest. Throughout the thesis, we present two aspects: one is the development and architecture of the bioinformatics pipeline named MApping the Genetics of neuropsychiatric traits to the molecular NETworks of the human brain (MAGNET). The other part demonstrates the implementation and usefulness of MAGNET analysing large Autism Spectrum Disorder (ASD) cohorts.
MAGNET is a freely available command-line tool available on GitHub (https://github.com/SheenYo/MAGNET). It is implemented within one framework using data integration approaches based on state-of-the-art algorithms and software to ultimately identify the genes and pathways genetically associated with a trait of interest. MAGNET provides an edge over the existing tools since it performs a comprehensive analysis taking care of the data handling and parsing steps necessary to communicate between the different APIs (Application Program Interface). Thus, this avoids the in-between data handling steps required by researchers to provide output from one analysis to the next. Moreover, depending on the size of the dataset users can deduce important information regarding their trait of interest within a time frame of a few days. Besides gaining insights into genetic associations, one of the central features is the mapping of the associated genes onto developing human brain implementing transcriptome data of 16 different brain regions starting from the 5th post-conceptional week to over 40 years of age.
In the second part as proof of concept, we implemented MAGNET on two ASD cohorts. ASD is a group of psychiatric disorders. Clinically, ASD is characterized by the following psychopathology: A) limitations in social interaction and communication, and B) restricted, repetitive behavior. The etiology of this disorder is extremely complex due to its heterogeneous clinical traits and genetics. Therefore, to date, no reliable biomarkers are identified. Here, the aim is to characterize the genetic architecture of ASD taking into account the two aforementioned ASD diagnostic domains. As well as to investigate if these domains are genetically linked or independent of each other. Moreover, we addressed the question if these traits share genetic risk with the categorical diagnosis of ASD and how much of the phenotypic variance of these traits can be explained by the underlying genetics.
We included affected individuals from two ASD cohorts, i.e. the Autism Genome Project (AGP) and a German cohort consisting of 2,735 and 705 families respectively. MAGNET was applied to each of the ASD subdomains as a quantitative dependent variable. MAGNET is divided into five main sections i.e. (1) quality check of the genotype data, (2) imputation of missing genotype data, (3) association analysis of genotype and trait data, (4) gene-based analysis, and (5) enrichment analysis using gene expression data from the human brain.
MAGNET was applied to each of the individual traits in each cohort to perform quality control of the genetic data and imputed the missing data in an automated fashion. MAGNET identified 292 known and new ASD risk genes. These genes were subsequently assigned to biological signaling pathways and gene ontologies via MAGNET. The underlying biological mechanisms converged with respect to neuronal transmission and development processes. By reconciling these genes with the transcriptome of the developing human brain, MAGNET was able to identify that the significant genes associated with the subdomains are expressed at specific time points in brain areas such as the hippocampus, amygdala, and cortical regions. Further, we found that ASD subdomains related to domain A but not
to domain B have a shared genetic etiology.