PG 613: ACC++ See

Autonomous Driving Using Neural Networks See

Bioinformatics from genetic variants to methylation See

An important research topic in bioinformatics is the analysis of DNA, the molecule that encodes the genetic information of all organisms. The basis for this is sequencing, a procedure in which the sequence of DNA bases is determined. In addition to the identification of variations in the base sequence itself, advances in sequencing methods and a steady reduction in sequencing costs open up new fields of research: the analysis of functionally relevant non-base-related changes, so-called ...

Uncertainty handling in surrogate assisted optimisation of games See

In this thesis entitled Uncertainty handling in surrogate assisted optimisation of games, we started out with the goal to investigate the uncertainty in game optimisation problems, as well as to identify or develop suitable optimisation algorithms. In order to approach this problem systematically, we first created a benchmark consisting of suitable game optimisation functions (GBEA). The suitability of these functions was determined using a taxonomy that was created based on the results of a ...

Contributions to computer-aided analysis of cuneiform tablet fragments See

This thesis presents methods for computer-aided three-dimensional analysis of digitized cuneiform tablets, an ancient type of writing documents. Since cuneiform script is predominantly conserved in the form of fractured clay tablet fragments, identifying matching fragments is a central task of manuscript reconstruction. This goal can benefit from the increasing 3D digitization of cuneiform fragments, which offers access to highly accurate cuneiform representations. The main contribution...

Learning attribute representations with deep convolutional neural networks for word spotting See

Understanding the contents of handwritten texts from document images has long been a traditional field of research in computer science. The ultimate goal is to automatically transcribe the text in the images into an electronic format. This would make the documents from which the images were generated much easier to access and would also allow for a fast extraction of information. Especially for historical documents a possibility to easily sift through large document image collections would...

Surrogate models for discrete optimization problems See

Surrogate models are crucial tools for many real-world optimization problems. An optimization algorithm can evaluate a data-driven surrogate model instead of an expensive objective function. While surrogate models are well-established in the continuous optimization domain, they are less frequently applied to more complex search spaces with discrete or combinatorial solution representations. The main goal of this thesis is to fill this gap. We develop and improve methods for data-driven su...

K-best enumeration - theory and application See

Where an optimal solution does not contain sufficient information about a given problem instance, enumerating good solutions is a common coping strategy. In combinatorial optimization, k-best enumeration, or ranking, has been studied and applied extensively. The k shortest simple path problem in directed, weighted graphs (kSSP), introduced in 1963 by Clarke, Krikorian and Rausen, is particularly well known. Efficient existing algorithms are based on Yen's algorithm for this problem; they all ...

Output-sensitive complexity of multiobjective combinatorial optimization with an application to the multiobjective shortest path problem See

In this thesis, we are concerned with multiobjective combinatorial optimization (MOCO) problems. We attempt to fill the gap between theory and practice, which comes from the lack of a deep complexity theory for MOCO problems. In a first part, we consider a complexity notion for multiobjective optimization problems which derives from output-sensitive complexity. The primary goal of such a complexity notion is to classify problems into easy and hard problems. We show that the multiobjectiv...

Partially supervised learning of models for visual scene and object recognition See

When creating a visual recognition system for a novel task, one of the main burdens is the collection and annotation of data. Often several thousand samples need to be manually reviewed and labeled so that the recognition system achieves the desired accuracy. The goal of this thesis is to provide methods that lower the annotation effort for visual scene and object recognition. These methods are applicable to traditional pattern recognition approaches as well as methods from the field of deep ...

On large-scale probabilistic and statistical data analysis See

In this manuscript we develop and apply modern algorithmic data reduction techniques to tackle scalability issues and enable statistical data analysis of massive data sets. Our algorithms follow a general scheme, where a reduction technique is applied to the large-scale data to obtain a small summary of sublinear size to which a classical algorithm is applied. The techniques for obtaining these summaries depend on the problem that we want to solve. The size of the summaries is usually paramet...

Exploring regular structures in strings See

This thesis is dedicated to string processing algorithms and to combinatorics on words. With respect to the former, we devise Lempel-Ziv (LZ) factorization and sparse suffix sorting algorithms. With respect to the latter, we search for all distinct squares, all maximal α-gapped repeats, and all maximal α-gapped palindromes. The topics of the presented approaches are related: for instance, our results for the LZ factorization have applications for finding all distinct squares, whereas our spar...

Markov decision processes with uncertain parameters See

Markov decision processes model stochastic uncertainty in systems and allow one to construct strategies which optimize the behaviour of a system with respect to some reward function. However, the parameters for this uncertainty, that is, the probabilities inside a Markov decision model, are derived from empirical or expert knowledge and are themselves subject to uncertainties such as measurement errors or limited expertise. This work considers second-order uncertainty models for Markov decisi...

Methods for efficient resource utilization in statistical machine learning algorithms See

In recent years, statistical machine learning has emerged as a key technique for tackling problems that elude a classic algorithmic approach. One such problem, with a major impact on human life, is the analysis of complex biomedical data. Solving this problem in a fast and efficient manner is of major importance, as it enables, e.g., the prediction of the efficacy of different drugs for therapy selection. While achieving the highest possible prediction quality appears desirable, doing so is o...

Efficient implementation of resource-constrained cyber-physical systems using multi-core parallelism See

The quest for more performance of applications and systems became more challenging in the recent years. Especially in the cyber-physical and mobile domain, the performance requirements increased significantly. Applications, previously found in the high-performance domain, emerge in the area of resource-constrained domain. Modern heterogeneous high-performance MPSoCs provide a solid foundation to satisfy the high demand. Such systems combine general processors with specialized accelerators ran...

Exponential families on resource-constrained systems See

This work is about the estimation of exponential family models on resource-constrained systems. Our main goal is learning probabilistic models on devices with highly restricted storage, arithmetic, and computational capabilities—so called, ultra-low-power devices. Enhancing the learning capabilities of such devices opens up opportunities for intelligent ubiquitous systems in all areas of life, from medicine, over robotics, to home automation—to mention just a few. We investigate the inhe...

Advances in next-track music recommendation See

Technological advances in the music industry have dramatically changed how people access and listen to music. Today, online music stores and streaming services offer easy and immediate means to buy or listen to a huge number of songs. One traditional way to find interesting items in such cases when a vast amount of choices are available is to ask others for recommendations. Music providers utilize correspondingly music recommender systems as a software solution to the problem of music overloa...

Acoustic sensor network geometry calibration and applications See

In the modern world, we are increasingly surrounded by computation devices with communication links and one or more microphones. Such devices are, for example, smartphones, tablets, laptops or hearing aids. These devices can work together as nodes in an acoustic sensor network (ASN). Such networks are a growing platform that opens the possibility for many practical applications. ASN based speech enhancement, source localization, and event detection can be applied for teleconferencing, camer...

Ein skalierbares agglomeratives Clusterverfahren basierend auf erste Nachbarschaftsbeziehungen See

Eine der revolutionären Veränderungen des 20. und 21. Jahrhunderts der Menschheit ist die Computerisierung und Digitalisierung der Welt. Tagtäglich werden große Menge von Daten erzeugt. Es ist unmöglich für uns Menschen, ohne anspruchsvolle Verfahren und Algorithmen, aus diesen enorm großen Datenmengen versteckte Strukturen (Wissen) herauszufiltern. Datamining, Machine Learning und Pattern Recognition sind ähnliche Disziplinen, mit denen man versuchen will, diese Herausforderung zu bewältige...