Refine
Year of publication
Document Type
- Doctoral Thesis (29)
- Book (4)
- Habilitation (2)
- Report (1)
Has Fulltext
- yes (36)
Is part of the Bibliography
- no (36)
Keywords
- Machine Learning (2)
- Argumentation (1)
- Bewertungskompetenz (1)
- Blaulichtemission (1)
- Case-based Reasoning (1)
- Case-based decision support (1)
- Clustering (1)
- Complexity Metrics (1)
- Data Mining (1)
- Decentralized Model Training (1)
Institute
- Fachbereich IV (36) (remove)
Road accidents are one of the leading causes of death worldwide, particularly among young people. The police and local authorities therefore strive to reduce the risk of accidents through appropriate road safety measures. In order to plan these measures, the relevant types of accidents, i. e., accidents with certain features, must first be recognized. However, the variety of accident features and the amount of resulting feature combinations make it impossible to monitor all accident types manually.
In this thesis, methods are proposed to automatically identify interesting accident types. Here, it is investigated whether combinations of features occur together and how the temporal pattern of the combined occurrence behaves. The change mining approach can then be used to determine whether structural changes in frequency occur during the period under consideration. For example, a feature combination that suddenly appears more frequently or exhibits a change in seasonality should be prioritized for further investigation so that appropriate road safety measures may be initiated for that combination.
The implemented strategic, multi-stage data mining framework based on frequent itemset mining, time series clustering, forecasting methods, and a scoring process is able to detect interesting feature combinations. These are then processed on a map in a web interface suitable for the respective audience in order to support the strategic planning of road safety measures. The framework is applied to several accident data sets from different countries to determine suitable default parameter values for the respective data analysis methods and to carefully align the methods. It is shown that there exist only minor dependencies of the parameter selection on the database to be analyzed.
For operational planning, it is necessary to consider small geographic areas and identify the features that have the greatest impact on accident occurrence there. Therefore, the developed operational framework analyzes and predicts the course of accident time series, taking into account the associated feature-specific time series. On the one hand, this makes it possible to increase the forecast performance, and, on the other hand, to determine which accident features have a significant influence on the course of the accident numbers over time. The insights gained can be used as a basis for short-term measures.
Supervised learning, the standard paradigm in machine learning, only works well if a sufficiently large, diverse, and cleanly-annotated dataset is available. Unfortunately, this is often not the case. In fact, the lack of labeled data is an omnipresent issue in machine learning. The problem is particularly prevalent in computer vision, where unlabeled images or videos can often be acquired at a low cost, whereas labeling them is time-consuming and expensive. To address the issue, this thesis focuses on developing new methods that aim at reducing annotation costs in computer vision by leveraging unlabeled and partially labeled data.
In the first part, we provide an overview of previous research directions and discuss their strengths and weaknesses. Thereby, we identify particularly promising research areas. The subsequent chapters which form the central part of this thesis aim at developing algorithmic improvements in these especially attractive fields. Among them is self-supervised learning, which aims at learning transferable representations given a large number of unlabeled images. We find that existing self supervised methods are optimized for image classification tasks, only compute global per-image feature vectors, and are designed for object-centric datasets like ImageNet. To address these issues, we propose a method that is particularly suited for object detection downstream tasks and works well if multiple objects are present per image like in video data for autonomous driving. Another core downside of self-supervised learning algorithms is that they depend on very large batch sizes with batch norm statistics synchronized across GPUs and also require many epochs of training until convergence. We find that stabilizing the self-supervised training target substantially speeds up convergence and allows for training with much smaller batch sizes. Our method matches ImageNet weights after 25 epochs of training with a batch size of only 32.
Finally, we investigate supervised pretraining. We find that state-of-the-art self-supervised methods match ImageNet weights only in classification or detection but not in both. In addition, we show that more sophisticated supervised training strategies significantly improve upon ImageNet weights.
The second part of the thesis deals with partially labeled data for object detection. We propose to label only large, easy-to-spot objects given a limited budget. We argue that these contain more pixels and therefore usually more information about the underlying object class than small ones. At the same time, they are easier to spot and hence cheaper to label. Because conventional supervised learning algorithms do not work well given this annotation protocol, we develop our own method with does, by combining pseudo-labels, output consistency across scales, and an anchor scale-dependent ignore strategy. Furthermore, many object detection datasets such as MS COCO and CityPersons include group annotations, i.e., bounding boxes that contain multiple objects of a single class. We find that pseudo-labeling instances within a group box is superior to the commonly used training strategies.
In the third part of the thesis, we cover semi-supervised object detection where a subset of the images is fully labeled whereas the remaining ones are unlabeled. We show that existing methods that are almost exclusively developed for Faster R-CNN work much less well if applied to architectures that are sensitive to missing annotations. In the prefinal chapter, we investigate the interaction between data and computer vision algorithms. This is in contrast to the vast majority of research which considers the data to be fixed. We provide computer vision practitioners and researchers with guidelines about what to do in typical situations.
In the final part of the thesis, we discuss the overall findings and investigate if research should put greater weight on acquiring and labeling data. Finally, we discuss options of mimicking human learning with machines, which might eventually result in human-level intelligence. After all, humans are living proof that this kind of learning works, if done properly.
Recent decades have seen exponential growth in data acquisition attributed to advancements in edge device technology. Factory controllers, smart home appliances, mobile devices, medical equipment, and automotive sensors are a few examples of edge devices capable of collecting data. Traditionally, these devices are limited to data collection and transfer functionalities, whereas decision-making capabilities were missing. However, with the advancement in microcontroller and processor technologies, edge devices can perform complex tasks. As a result, it provides avenues for pushing training machine learning models to the edge devices, also known as learning-at-the-edge. Furthermore, these devices operate in a distributed environment that is constrained by high latency, slow connectivity, privacy, and sometimes time-critical applications. The traditional distributed machine learning methods are designed to operate in a centralized manner, assuming data is stored on cloud storage. The operating environment of edge devices is impractical for transferring data to cloud storage, rendering centralized approaches impractical for training machine learning models on edge devices.
Decentralized Machine Learning techniques are designed to enable learning-at-the-edge without requiring data to leave the edge device. The main principle in decentralized learning is to build consensus on a global model among distributed devices while keeping the communication requirements as low as possible. The consensus-building process requires averaging local models to reach a global model agreed upon by all workers. The exact averaging schemes are efficient in quickly reaching global consensus but are communication inefficient. Decentralized approaches employ in-exact averaging schemes that generally reduce communication by communicating in the immediate neighborhood. However, in-exact averaging introduces variance in each worker's local values, requiring extra iterations to reach a global solution.
This thesis addresses the problem of learning-at-the-edge devices, which is generally referred to as decentralized machine learning or Edge Machine Learning. More specifically, we will focus on the Decentralized Parallel Stochastic Gradient Descent (DPSGD) learning algorithm, which can be formulated as a consensus-building process among distributed workers or fast linear iteration for decentralized model averaging. The consensus-building process in decentralized learning depends on the efficacy of in-exact averaging schemes, which have two main factors, i.e., convergence time and communication. Therefore, a good solution should keep communication as low as possible without sacrificing convergence time. An in-exact averaging solution consists of a connectivity structure (topology) between workers and weightage for each link. We formulate an optimization problem with the objective of finding an in-exact averaging solution that can achieve fast consensus (convergence time) among distributed workers keeping the communication cost low. Since direct optimization of the objective function is infeasible, a local search algorithm guided by the objective function is proposed. Extensive empirical evaluations on image classification tasks show that the in-exact averaging solutions constructed through the proposed method outperform state-of-the-art solutions.
Next, we investigate the problem of learning in a decentralized network of edge devices, where a subset of devices are close to each other in that subset but further apart from other devices not in the subset. Closeness specifically refers to geographical proximity or fast communication links.
We proposed a hierarchical two-layer sparse communication topology that localizes dense communication among a subgroup of workers and builds consensus through a sparse inter-subgroup communication scheme. We also provide empirical evidence of the proposed solution scaling better on Machine Learning tasks than competing methods.
Finally, we address scalability issues of a pairwise ranking algorithm that forms an important class of problem in online recommender systems. The existing solutions based on a parallel stochastic gradient descent algorithm define a static model parameter partitioning scheme, creating an imbalance of work distribution among distributed workers. We propose a dynamic block partitioning and exchange strategy for the model parameters resulting in work balance among distributed workers. Empirical evidence on publicly available benchmark datasets indicates that the proposed method scales better than the static block-based methods and outperforms competing state-of-the-art methods.
In der hier vorgelegten Promotionsarbeit wird das Potenzial gruppenbasierter und semistrukturierter Aushandlungsprozesse analysiert. In einer entsprechenden Interventionsstudie mit Pre- und Post-Analysen wurden 146 Schüler:innen einer Gesamtschule in Niedersachsen/Deutschland aufgefordert, Begründungen zu acht selbstentwickelten Argumenten zu einem Thema über den Erhalt der lokalen Biodiversität, einem bioethischen Konflikt im Rahmen von nachhaltiger Entwicklung, vor und nach einer gruppenbasierten Aushandlung zu formulieren und diese zu gewichten. Zu diesem Zweck verwendeten die Schüler:innen in allen Phasen die Zielmat als ein Instrument zur Strukturierung des Bewertungsprozesses. Die Begründungen wurden inhaltsanalytisch hinsichtlich der Nutzung argumentativer Ressourcen analysiert. Darüber hinaus wurde die Richtung der Veränderungen der Begründungen nach der Aushandlung qualitativ verglichen und die Veränderung der Gewichtungen quantitativ berechnet. Bei diesen Analysen wurden individuelle Gewichtungen und Begründungen beider Phasen und die Veränderung der Gewichtungen mit den Gruppengewichtungen verglichen. Die Ergebnisse der Begründungen zeigen, dass die Schüler:innen bereits vor dem Aushandlungsprozess über eine Bandbreite an argumentativen Ressourcen (nämlichen faktenbasierte und normative Ressourcen) verfügen. Die Ergebnisse des Vergleichs der Begründungen von der Pre- zur Post-Phase zeigen, dass etwa ein Drittel aller Begründungen verändert wurden. Die Richtung der Veränderung ist zudem sehr divers, da die Schüler:innen die Begründungen widerlegten, revidierten, aber auch bestätigten und verstärkten. Ebenso wurde etwa ein Drittel aller Gewichtungen in der Post-Phase verändert. Ein Vergleich der Gewichtungsänderung der Pre- zu Post-Phase mit der Gruppengewichtung zeigt, dass diese der Tendenz der Gruppengewichtung entspricht.
Die Ergebnisse dieser Studie machen auf das Potenzial gruppenbasierter Aushandlungsprozesse in bioethischen Konflikten aufmerksam, nämlich die Aktivierung relevanter argumentativer Ressourcen und die Initiierung tiefer und revidierender Denkprozesse. Darüber hinaus zeigen die Daten das besondere Potenzial der in der Studie verwendeten Zielmat, nämlich die Unterstützung komplexer und sonst für Schüler:innen überfordernder kompensatorischer Gewichtungsstrategien.
Finding an available parking spot in city centers can be a cumbersome task for individual drivers and also negatively affects general traffic flow and CO2 emissions.
In the context of smart cities and the internet of things this problem can be mitigated by using available data to monitor and predict parking occupancy in order to guide users to an available parking location near their destination.
With this goal in mind there arise multiple challenges of which we introduce selected ones to propose novel solutions based on machine learning.
The focus of this work is to enable the usage of readily available and inexpensive data sources like parking meter transactions, opposed to expensive technology like in-ground sensors or cameras where the costs prevent a widespread coverage. Our proposed data sources do not directly monitor the actual parking availability but still provide enough signal for our algorithms to infer the real parking situation with high accuracy.
As part of this work we developed a parking availability prediction system based on parking meter transactions that was deployed to 33 german cities.
A main contribution of our work is the proposal of a novel way to generate labels based on the parking transactions and to use semi-supervised-, more specifically positive-unlabeled learning, to leverage the sparse signal in order to require as little data as possible.
Additionally, we utilize and design novel methodologies in the area of transfer learning to learn simultaneously from different cities which leads to the previously seldom explored setting of combining transfer learning with positive-unlabeled learning. We therefore introduce a novel algorithm to tackle this problem type.
We hope that our work enables the deployment of smart parking systems at lower costs and therefore leads towards the goal of smart parking guidance in smart cities.
Human skeletal remains are one of several find categories from archaeological sites. The skeleton constitutes only a small part of a former living organism that was exposed to a variety of environmental factors. As a highly adaptive mineralized tissue, bone – and to a lesser extent also teeth – stores information on an individual’s life. Reading out this information contributes to the historical understanding of individuals and societies from the past. However, obtaining this information can be challenging in many aspects, and, due to the very nature of the archaeological remains, the reconstructed picture of the individual and its society and environment will inevitably always be incomplete. In order to extract as much information from bones and teeth as possible, existing methods must be adapted to specific situations and the diagnostic approaches have to be developed beyond current limitations.
The 17 international publications presented in this thesis are addressing the abovementioned goals. They cover a wide range of topics, mainly from the field of bioarchaeology, augmented by some studies on recent and fossil animal remains. The studies are organized in those dealing with metrical data, including sex determination methods and analysis of measurement error, the analysis of normal and abnormal conditions of bone, including paleopathological diagnostics, the impact of diagenetic agents on bone, and the presentation of cases where bioarchaeological analysis contributed to the understanding of the respective find situations. The methods applied in the studies cover a broad spectrum of approaches to study morphological properties of bones and teeth, including metrics, radiographic, and microscopic techniques.
2. Low-Code-Symposium
(2022)
Zum bundesweiten Digitaltag am 24. Juni 2022 richtete das Zentrum für Digitalen Wandel in Kooperation mit dem Hi-X-DigiHub und der COMPRA GmbH zum zweiten Mal ein digitales Symposium zum Thema der Low-Code-Softwareentwicklung aus.
Mit der Low-Code-Technologie ist die Hoffnung verbunden, mehr Nicht-Informatiker_innen in die Software-Entwicklung einzubinden. Die Technologie ist für verschiedene Problemfelder von besonderer Bedeutung. Es besteht die Möglichkeit, Expert_innen aus Politik, BWL, Recht und anderen Bereichen direkt in die Software-Entwicklung einzubinden, so dass es zu weniger Reibungsverlusten bei der entwickelten Software kommt, da die Fachexpertise direkt in die Entwicklung eingebracht werden kann. Dem bestehenden Fachkräftemangel in der Softwareentwicklung kann insoweit begegnet werden, als dass die Mitwirkung von Expert_innen in der Entwick-lung dazu beitragen kann, dass Systeme entstehen können, bei denen weniger Code und damit weniger Programmier-Expertise erforderlich ist.
Insgesamt verspricht dieses Verfahren Effizienz und Innovation in der Softwareent-wicklung.
Hintergrund: Die Software-Produktlinienentwicklung ermöglicht eine kostengünstige und effiziente Entwicklung von Produktfamilien bei gesteigerter Qualität im Vergleich zur Einzelsystementwicklung. Dieses wird durch die Einführung von Variabilitätsmechanismen ermöglicht, welche eine hohe Anpassbarkeit der Produkte an verschiedene Kundenbedürfnisse ermöglichen. Allerdings erhöhen diese Variabilitätsmechanismen auch die Komplexität, da Entwickler das Zusammenwirken der Komponenten für verschiedene Produktvarianten berücksichtigen müssen. Daher wurden zur Qualitätssicherung von Software-Produktlinien neue Analysemethoden und -strategien entwickelt, darunter auch variabilitätsbasierte Code-Metriken. Ziel dieser Metriken ist es, unnötige Komplexität zu vermeiden und frühzeitig besonders fehleranfälligen Code zu identifizieren, um diesen zusätzlichen Qualitätsmaßnahmen unterziehen zu können. Unsere systematische Literaturstudie zu diesem Thema zeigt jedoch, dass der Nutzen dieser variabilitätsbasierten Code-Metriken nur in wenigen Fällen evaluiert wurde.
Ziel: Diese Arbeit untersucht inwieweit variabilitätsbasierte Code-Metriken zur Qualitätssteigerung von Software-Produktlinien genutzt werden können. Dazu wird betrachtet, ob sich mit Hilfe empirischer Untersuchungen Entwicklungsrichtlinien zur proaktiven Vermeidung von Komplexität und damit verbundenen Fehlern ableiten lassen. Der Fokus liegt auf der Analyse, ob sich die betrachteten Metriken zur Identifikation von potentiell fehleranfälligeren Code nutzen lassen. Dies umfasst sowohl die univariate Verwendung einzelner Metriken als auch den Aufbau von Vorhersagemodellen mit Verfahren des maschinellen Lernens. Dabei wird auch untersucht, ob die speziell für die Software-Produktlinienentwicklung konzipierten variabilitätsbasierten Code-Metriken einen Mehrwert gegenüber etablierten Metriken der Einzelsystementwicklung bieten.
Methodik: Es findet eine empirische Untersuchung von 692 Einzelsystem- und variabilitätsbasierte Code-Metriken auf dem Linux-Kernel statt. Dazu wird zunächst analysiert, inwieweit die Messwerte der Metriken mit Kompilierfehlern und Sicherheitslücken korreliert sind, welche von den Entwicklern übersehen wurden und so erst nach dem Commit bzw. nach dem Release entdeckt wurden. Darüber hinaus werden die Metriken bezüglich der gemessenen Eigenschaften gruppiert und mit vier Verfahren des maschinellen Lernens eine Identifikation der fehleranfälligen Codestellen erprobt, um so den Nutzen verschiedener Codeeigenschaften beurteilen zu können.
Ergebnisse und Schlussfolgerung: Auch wenn für einen Großteil der Metriken ein signifikanter Zusammenhang zwischen den Messwerten und fehleranfälligen Codestellen nachgewiesen werden kann, so zeigt sich, dass univariate Verfahren für die Praxis untauglich sind. Auf Grund der starken Klassenimbalance von nur 1,5% defekten Codefunktionen (Kompilierfehler), bzw. 0,4% Codefunktionen mit nachgewiesenen Sicherheitslücken, lassen sich bei der Verwendung einer einzelnen Metrik nur F1-Werte unterhalb von 0,073 erzielen. Mangels alternativer Implementierungen lassen sich so, entgegen den Aussagen einiger Veröffentlichungen, auch keine Entwicklungsempfehlungen ableiten. Hingegen können variabilitätsbasierte Code-Metriken, insofern sie auch die Variabilität verbundener Artefakte mit berücksichtigen, erfolgreich zur Fehlervorhersage genutzt werden.
So lässt sich beispielsweise bei Verwendung von Random Forest F1-Werte von 0,667 (Kompilierfehler), bzw. 0,711 (Sicherheitslücken), erzielen.
Stream processing is a popular paradigm to process huge amounts of unbounded data, which has gained significant attention in both academia and industry. Typical stream processing applications such as stock trading and network traffic monitoring require continuously analyzed results provided to end-users.
During processing, the characteristics of data streams such as volume or velocity can vary, e.g., peak load or bursty streams can occur at certain points.
In order to cope with such situations, it requires the analytical systems to be able to adapt the execution of stream processing as quickly as possible.
In literature, different approaches adapting data stream processing such as load-shedding and elastic parallelization do exist. However, each of them have their different shortcomings like skewed results (due to the dropped data) or strong limits on the adaptation due to the parallelization overhead. One specific challenge motivating us is to minimize the impact of runtime adaptation on the overall data processing, in particular for real-time data analytics. Moreover, while the need to create adaptive stream processing systems is well known, there is currently no systematic and broad analysis of the solution range of creating adaptation mechanisms for stream processing applications.
In this dissertation, we focus on algorithm switching as a fundamental approach to the construction of adaptive stream processing systems. Algorithm switching is a form of adaptation, where stream processing algorithms, with fundamentally similar input-/output-characteristics but different runtime tradeoffs like resource consumption or precision, are replaced to optimize the processing. As our overall goal, we present a general algorithm switching framework that models a wide range of switching solutions (called switch variants) in a systematic and reusable manner as well as characterizes the switch variants with their quality guarantees.
Concretely, we focus on developing a general model of algorithm switching to systematically capture possible variants of different switching behavior. We also present a theoretical specification to predict the timeliness-related qualities for the switch variants. Moreover, from the practical perspective, we also develop a component-based design to ease the realization effort of the algorithm switching variants. Finally, we provide a validation of the algorithm switching framework against the realized switch variants.
Der überwiegende Teil der in Deutschland erhaltenen Niedermoore weist einen veränderten Wasserhaushalt, ein mooruntypisches Arteninventar und einen stark degradierten Torfkörper auf. Revitalisierungsmaßnahmen werden umgesetzt, um den Zustand dieser überprägten Moorökosysteme zu verbessern. Allerdings existiert bisher kein Erfolgskontrollkonzept, welches gezielt für die Bewertung der Revitalisierungsmaßnahmen in stark degradierten Niedermoorökosystemen Anwendung finden kann.
Am Beispiel des stark degradierten und hydrologisch vielfältig überprägten Bergen-Weißacker Moores (Südbrandenburg) wird in einem Zeitraum von zwei Jahren eine Erfolgskontrolle umgesetzt, um eine Wiedervernässungsmaßnahme zu bewerten. Die Basis der Erfolgskontrolle bildet ein Set aus biotischen und abiotischen Parametern. Um ein grundlegendes Ökosystemverständnis aufzubauen, reicht die Parameteraufnahme über den Maßnahmenbereich hinaus. Für die Einordnung der Aufnahmeergebnisse wird eine Bewertungsbasis aufgebaut.
Die Erfolgskontrolle zeigt, dass auch nach der Wiedervernässung im Maßnahmenbereich des Bergen-Weißacker Moores keine flurnahen Wasserstände vorliegen. Basierend auf den Erkenntnissen werden Grenzen und Potenziale sowie Vorschläge zur Modifizierung des Konzeptes abgeleitet. Weiterhin werden Überlegungen zur Konzeptübertragbarkeit getroffen. Dabei wird herausgestellt, dass Grundvoraussetzungen vorliegen müssen, um das Konzept mit wenigen Adaptationen in anderen stark degradierten minerotrophen Mooren einsetzen zu können. Aus den Monitoringresultaten werden Handlungsempfehlungen abgeleitet, um prospektiv den Zustand des Bergen-Weißacker Moores zu verbessern.
Basierend auf den im Rahmen der vorliegenden Arbeit gewonnenen Erkenntnissen werden Grundsätze für Erfolgskontrollen in stark degradierten Niedermoorökosystemen formuliert. Dazu zählen beispielsweise der Einbezug moorinterner Vergleichsflächen, die Erfassung von Randeffekten und die langfristige Fortführung des Monitorings.
