6949 research outputs found
Sort by
The persistence transformation; a new methodology of topological data analysis
No full textThe field of Topological Data Analysis (TDA) continues to evolve as a powerful tool for the analysis of complex data. The motivation behind this research lies in the need to extend existing TDA tools to provide more accurate, efficient, and comprehensive analyses of intricate datasets. The primary research problem addressed herein pertains to the limitations of the Persistence Diagram, a fundamental TDA tool that does not inherently incorporate positional information of topological features. The absence of this crucial spatial context can lead to inaccurate results, especially when
analyzing low-dimensional data. To tackle this issue, this dissertation introduces the Persistence Transformation, an innovative extension of the Persistence Diagram. It is designed to capture the positional information of topological peaks, enhancing the robustness and depth of TDA analyses. Key findings of this research include a comprehensive analysis of the properties and stability of the Persistence Tansformation. Furthermore, a real-world application of this method demonstrates its
effectiveness in the classification of MALDI data, highlighting the practical utility of the extension. The originality and contribution of this work are underscored by the extension of the traditional Persistence Diagram. The introduction of the Persistence Transformation empowers mathematicians and data analysts to tackle a broader spectrum of complex problems, fostering more accurate results across diverse application domains. However, it is important to note that the Persistence Transformation generates results of a higher dimensionality when compared to the Persistence Diagram. While this enables a richer analysis of complex data, it may necessitate additional computational resources. Nevertheless, this research not only advances the domain of TDA but also opens the door to a wider array of analytical possibilities for complex datasets, offering valuable insights across various fields
Soziale Beratung in Krisenzeiten: Ist das Hilfesystem am Limit?
No full textLaut Sozialgesetzbuch sind die Behörden zur Beratung verpflichtet, um den hilfesuchenden Bürger:innen den Zugang zu ihren sozialen Rechten zu erleichtern. Ein großer Anteil der Sozial- und Sozialrechtsberatung wird jedoch formal oder informell durch das Hilfesystem der Freien Träger gewährleistet, oftmals als Nebenaufgabe bei der Erbringung sozialer Dienstleistungen vor Ort. Schon vor der Pandemie waren die Arbeitsbedingungen im Hilfesystem nicht optimal. In allen Bereichen – von der Jugendhilfe bis zur Drogenprävention – wirkte die Pandemie wie ein Brennglas: Bereits vorhandene Probleme traten nun umso deutlicher zu Tage.
Am Ende der Pandemie, Im Herbst 2022 wurden sowohl hauptamtlich als auch ehrenamtlich Tätige aus der Sozialberatung in Bremen, die selbst regelmäßig Auskünfte über die Erbringung von Unterstützungsleistung geben, per Online-Befragung nach ihren Arbeitsbedingungen, den besonderen Anforderungen durch die Pandemie, neuen Organisationsformen der Arbeit wie etwa der Nutzung digitaler Instrumente und ihren Wünsche an die Politik gefragt. In die Auswertung flossen 75 vollständig ausgefüllte Fragebögen ein. Unsere Befragung beschreibt die Probleme des lokalen Hilfesystems nun präzise: Die Ergebnisse offenbaren eine starke Belastung im Hilfesystem, die sich durch die Pandemie zusätzlich verschärfte, jedoch auch strukturelle Probleme, die bereits im Regelbetrieb die Hilfeleistung erschweren. Hierzu gehören etwa die zuweilen schwierige Erreichbarkeit der Behörden oder die fehlende ressortübergreifende Zusammenarbeit. Deutlich wurde aber auch, dass die Initiativen in den Einrichtungen der freien Träger an vielen Stellen zu Innovationen in der Alltagsarbeit führten. Vor allem die Digitalisierung birgt ein großes Potential für die Verbesserung des Zugangs zu den sozialen Rechten – vorausgesetzt, die große Expertise der freien Träger findet systematische Berücksichtigung bei der Entwicklung der digitalen Instrumente.Hochschule Breme
Equivariant deep learning for 3D topology optimization
No full textThe 21st century witnessed the ascent of deep learning, paving the way for data-driven topology optimization approaches via neural networks. This paradigm shift holds the promise of significant speed-ups compared to classical methods; however, deep learning still faces challenges related to generalization and the requirement for training data. Moreover, the field suffers from a dramatic lack of research infrastructure, which hampers both progress and comparability of results. Notably, prior to our research, neither a reliable yet flexible code base nor public three-dimensional datasets existed. We tackled these challenges by contributing a new public dataset and developing a Python library for three-dimensional topology optimization utilizing deep learning. Additionally, we demonstrate that incorporating physical information into the training process significantly reduces the reliance on extensive training data and enhances overall generalization capabilities.
This work consists of four publications. The initial publication introduces the SELTO dataset, comprising nearly 10,000 three-dimensional samples, each providing topology optimization problems and corresponding solutions. The second publication presents DL4TO, a pioneering PyTorch-based deep learning library for topology optimization, accompanied by comprehensive documentation and online tutorials. The third publication showcases the efficacy of equivariant neural networks and a physics-inspired data preprocessing strategy, substantially reducing the need for extensive training datasets and enhancing generalization capabilities. Finally, the fourth paper employs neural operators to replace the PDE solver in the widely used SIMP method, which is its primary bottleneck. This publication underscores the efficacy of equivariance in learned methods and emphasizes the crucial role of a gradient-consistent loss function when applied in a gradient-based optimization scheme like SIMP
Evolutionary neural architecture search for the task of road traffic prediction.
No full textThe topic of this dissertation is the application of evolutionary neural architecture search
to find suitable neural networks for predicting speed and flow from road traffic data.
The thesis begins by describing the measurement data and describing the forecasting
problem. Following this, fundamental concepts in the fields of Machine Learning, Deep
Learning, and Neural Architecture Search (NAS), particularly concerning application,
are explained. The last part of this dissertation consists of five articles to which the
author of this thesis has made a significant contribution.
The first two articles provide an overview of the problem of traffic data prediction,
concerning measurement data from the city of Bremen. The machine learning model
k-nearest neighbors is introduced and applied to the measurement data. In addition, we
evaluate data imputation methods to improve models. In the third article, we compare
combined polynomial regression models, a simple machine learning model, with graph
convolutional neural networks. These are neural networks that include special opera-
tions incorporating spatial dependencies between measurement points. Our evolutionary
neural architecture search framework is presented in the fourth article. The outcome of
the genetic algorithm used in our framework depends on the fitness, i.e. performance on
the dataset, of each architecture in the search space. While the choice of validation loss
as fitness is ideal w.r.t. the accuracy, it slows down the algorithm tremendously since
it necessitates training the neural networks until convergence. Hence, to make usage of
our framework viable, in the fifth article, we evaluate zero-cost proxies, which compute
a fitness for architectures based on singular forward or backward passes through the
network. Therefore, evaluating network fitness only takes a few compared to multiple
hours. We show that the naswot zero-cost proxy is robust w.r.t. random initializations
of weights, network sizes and batch sizes and has a high spearman rank correlation with
the validation loss.
My contribution is a neural architecture search framework that finds neural network
architectures that are especially powerful for predicting road traffic data. My NAS
framework finds an architecture for a given dataset that can keep up with or outperform
handcrafted neural networks and neural networks found by other NAS frameworks in
terms of performance and computation time
Look both ways before crossing the street: combined safety and security analysis for autonomous vehicles
No full textWith the advent of autonomous driving, machines are taking over vital functions previously performed by a human driver. Therefore, ensuring their safety and security is paramount.
While safety has always been a primary concern in automotive development, with the increasing use of software-based electronic components, proving it is becoming an exceedingly difficult task.
At the same time, security concerns are rising since more and more interconnected devices are being installed to enhance the automated driving function.
While this risk has not remained unrecognized, security is still not targeted satisfactorily. Furthermore, it is still widely adopted to analyse safety and security separately in practice.
This is a misjudged approach, given the fact that the two properties are intertwined: A security attack on a component endangers its correct operation and thus the system safety. Vice versa, a safety failure of a cryptography module increases the vulnerability of the components relying on it, favouring security attacks.
The present dissertation addresses this issue with the development of a quantitative analysis method that is capable of modelling complex, critical systems and viewing the occurrence of safety failures and security attacks in parallel, as well as in dependence to one another.
Therefore, a graph-based modelling of system level components and their dependencies is developed and a transformation into a Continuous-Time Markov Chain formalized.
In that, the occurrences of single failures and attacks of the individual components are modelled by state changes due to defined probability rates and their consequences to the system's capability of remaining functional are reflected.
The goal is to prepare for a quantitative analysis that yields the system failure probability over a specified period, e.g. the system's lifetime.
The results are meant to support the development and the certification process of new vehicular architectures.
In order to allow for a comfortable modelling and an automated evaluation of complex systems, this method is implemented in a tool called ERIS
Accurate binary-level symbolic execution of embedded firmware
No full textSymbolic execution is an automated software testing technique that has enabled the discovery of numerous bugs in conventional, non-embedded software. Unfortunately, its application to embedded firmware is presently limited due to unique challenges associated with this domain. Central to many of these challenges is the tight integration of hardware and software components. Due to this tight integration, firmware interacts on a low abstraction level with both the processor and the peripherals provided by a hardware platform. In order to support these interactions, a symbolic execution engine needs to implement the reference manuals specifying processor and peripheral behavior. These specifications have an enormous complexity; hence, prior work approximates peripheral behavior and abstracts processor instruction execution. Unfortunately, these approximations may induce inaccuracies, which can result in bugs being missed in the tested firmware. This dissertation accomplishes a more accurate analysis by contributing a binary-level symbolic execution approach that is faithful to both the specification of peripheral behavior and processor instruction execution. This is achieved by facilitating machine-readable formal descriptions of instruction semantics and an established modeling standard for the description of peripheral behavior. Conducted experiments have resulted in the discovery of 16 previously unknown bugs in a popular embedded operating system, thereby illustrating the effectiveness of the proposed approach
Towards sustainable artificial intelligence systems: enhanced system design with machine learning based design techniques
No full textEfficient and timely calculations of Machine Learning (ML) algorithms are crucial for emerging technologies like autonomous driving, the Internet of Things (IoT), and edge computing. One of the primary ML techniques used in such systems is Convolutional Neural Networks (CNNs), which demand high computational resources. This requirement has led to using ML accelerators like General Purpose Graphig Processing Units (GPGPUs) to meet design constraints. However, GPGPUs have high power consumption needs, and thus, selecting the most suitable accelerator involves Design Space Exploration (DSE). This process is usually time-consuming and requires significant manual effort. This thesis presents approaches to improve the DSE process by supporting the identification of the most appropriate GPGPU for CNN inferencing systems. Different techniques are developed to quickly and precisely forecast the power consumption and performance of CNNs during inference. These approaches empower the system designer to estimate power consumption and performance for GPGPUs in the early stages of development without executing the application on real devices. Without the need to execute and profile applications on real devices, the number of prototypes can significantly be reduced.
Besides the system’s power and performance requirements and the ML accelerator selection, the designer has to face the placement problem and decide whether an application is implemented on an IoT device or in the Cloud. The available network, bandwidth, and latency are crucial if the application is implemented in the Cloud. Therefore, this thesis presents a decision-supporting system that is pivotal in helping system designers Make these complex decisions. This system is designed to consider the available network, bandwidth, and latency, and it distinguishes between power or performance optimization needs, thereby empowering the system designer to make informed choices
Benthic carbon cycling on the Antarctic continental shelf
No full textThe sedimentation of pelagic production makes continental shelf sediments important sites for organic matter (OM) remineralization and nutrient regeneration in the ocean. Consequently, shelf sediments play an important role in the bentho-pelagic coupling by providing essential nutrients for algal growth and maintaining the high primary production of shelf areas and the adjacent open ocean. In Antarctica, changes in sea ice cover have a major impact on surface primary production and the subsequent sinking of organic carbon to the seafloor. Recent observations indicate that global warming has led to substantial changes in sea ice cover, with a significant reduction of one million square kilometers in the annual maximum sea ice extent around Antarctica. These changes in sea ice conditions are expected to trigger significant changes in the pelagic ecosystem, with potentially profound effects on the benthic ecosystem. The imprint of climate-sensitive variables such as sea ice cover can be best studied in shelf sediments, where shallow water depths result in increased OM supply and tighter bentho-pelagic coupling. Therefore, a comprehensive understanding of the current carbon cycle on the Antarctic continental shelf is crucial for assessing the vulnerability of the ecosystem to climate change and for predicting the future trajectory of the carbon cycle in Antarctic waters. This thesis aims to quantify benthic carbon remineralization rates on the Antarctic continental shelf and the release of nutrients, particularly iron, that limit primary production in the Southern Ocean. In addition, the thesis aims to identify the main pathways of OM degradation and associated microbial communities, while contextualizing the above variables within the prevailing sea ice conditions.
For this purpose, I first studied the geochemistry of shelf sediments along a gradient of sea ice cover on the eastern shelf of the Antarctic Peninsula (AP) (manuscript 1). The main focus was on carbon and iron fluxes within the sediment and between the water column and the sediment. The results were interpreted in the context of sea ice cover. An increase in carbon remineralization rates was observed as one moved from heavily ice-covered to moderately ice-covered stations where light availability and water column stratification increased. Conversely, the ice-free station displayed lower carbon remineralization rates and was subject to wind-driven mixing of the water column, which can deepen the mixed layer depth below the critical depth, resulting in reduced surface production. In summary, a positive correlation was found between moderate sea ice cover and increased carbon fluxes to the sediment, which followed an exponential increase. The study also revealed significant iron cycling in sediments with increased carbon remineralization, resulting in high dissolved iron fluxes. This finding highlights the importance of sediments underlying the moderate ice cover as a source of limiting nutrients for primary production in this region.
A complementary study of benthic microbial communities along the AP transect was conducted using 16S ribosomal RNA (rRNA) gene sequencing (manuscript 2). The results indicate that sea ice cover and its effect on organic carbon fluxes are the main drivers of changes in benthic microbial communities. As sea ice cover decreases, the benthic microbial community shifts towards anaerobic communities of iron and sulfate reducers. These communities were more abundant at low ice cover stations than at high ice cover stations. Furthermore, an increase in the relative abundance of Sva1033, a Desulfuromonadia clade, with dissolved iron concentration at low ice cover stations suggests a putative role for Sva1033 in dissimilatory iron reduction in surface sediments. In addition to Sva1033, this study successfully identified other taxa that could potentially contribute to dissimilatory iron reduction or have syntrophic partnerships and/or common metabolic preferences with iron reducers.
The focus was extended towards the southern shelf of the Weddell Sea, a region characterized by heavy sea ice cover (manuscript 3). Benthic oxygen uptake rates were measured at stations with different water depths and sediment compositions. Benthic measurements also revealed a dependence of carbon fluxes on sediment grain size and water depth. In general, diffusive oxygen uptake (DOU) rates on the southern Weddell Sea shelf were low. DOU showed a positive correlation with preserved total organic carbon (TOC) at stations with fine-grained sediments, whereas stations with typical of coarse-grained sediments showed a markedly different correlation of DOU with TOC. Common to all stations is that dissolved iron and manganese concentrations in pore water were found only at greater depths, suggesting very limited release of these nutrients back into the water column.
The strong dependence of benthic carbon fluxes on sea ice cover and water depth was then combined to derive a simple empirical model, which was validated by all available DOU measurements reported in the literature for the Antarctic seasonal ice zone (manuscript 4). The model allows extrapolation and budgeting of benthic carbon remineralization for the entire seasonal ice zone (16 million km²), yielding a total of 46 Tg C yr-1. Notably, although the Antarctic continental shelf represents only 15% of the total area, it contributes a significant 71% (33 Tg C yr-1) of the total benthic carbon remineralization. Furthermore, the total organic carbon supply to the sediments and the carbon burial in the sediments were estimated to amount to 52 Tg C yr-1 and 6 Tg C yr-1, respectively.
Overall, the thesis highlights the pivotal role of sea ice cover in controlling the benthic carbon and iron cycling on the Antarctic continental shelf. Through extensive data correlation and empirical modeling, the thesis has provided, for the first time, a quantitative framework for the relationship between sea ice cover and benthic carbon fluxes. It also emphasizes the substantial contribution of Antarctic shelf sediments to the marine carbon remineralization, allowing a better assessment of the carbon cycling and related CO2 sequestration across the Southern Ocean. Furthermore, it has improved our understanding of the main drivers of change in benthic microbial communities and, ultimately, nutrient fluxes across the sediment-water interface. These findings make a significant contribution to our understanding of the complex Antarctic ecosystem, which is necessary to assess the future trajectory of the Southern Ocean and its impact on the global carbon cycle
Qualitative interview study: The role of Social Infrastructures in Reconstruction Processes after Floodwater Events.
No full textDieser Studienbericht beschreibt die qualitative Interviewstudie im Rahmen des vom BMBF geförderten, interdisziplinären Projekts KAHR (Klimaanpassung, Hochwasser und Resilienz), das den Wiederaufbau in den Gebieten in Deutschland begleitet, die von dem Hochwasser im Jahr 2021 betroffen waren. In diesem Projekt wurde eine Kontextanalyse durchgeführt um wesentliche Faktoren zu identifizieren und analysieren, die den Wiederaufbau beeinflussen. Die Kontextanalyse nutzt den konzeptionellen Ansatz der „Hydrosozialen Territorien“ (Boelens et al. 2016) und untersucht die Rolle von sozialen Infrastrukturen im Wiederaufbauprozess. Hierzu wurden in den Gemeinden Stolberg und Eschweiler (Nordrhein-Westfalen) und der Verbandsgemeinde Altenahr (Rheinland-Pfalz) semi-strukturierte Leitfadeninterviews mit Vertreter:innen von sozialen Infrastrukturen durchgeführt. Durch die Interviews soll ermittelt werden, wie soziale Infrastrukturen zur Resilienz von Gemeinschaften beitragen und welche Herausforderungen und Chancen im Wiederaufbauprozess bestehen. Zusätzlich wurde die Perspektive der sozialen Infrastrukturen durch die von Wasserbehörden, Verwaltung und anderen Institutionen ergänzt. Der bei Qualiservice verfügbare Datensatz umfasst 40 Interviewtranskripte. Dieser Bericht beschreibt die theoretische Rahmung und Zielsetzung der Studie, die Kontexte der Datenerhebung sowie Nachnutzungspotenziale der Forschungsdaten und enthält die Interviewleitfäden.Bundesministerium für Bildung und Forschung (BMBF
Last glacial-deglacial variability in sea surface characteristics and ice sheet activities in the Labrador Sea/Baffin Bay
No full textIn the context of global warming, the accelerated melting of the Greenland Ice Sheet (GrIS) and Arctic sea ice in recent years has led to increased meltwater discharge and subsequent sea-level rise. The meltwater discharge may affect sea surface characteristics in the polar and subpolar regions and potentially yield significant influence on ocean circulation patterns. Additionally, the far-reaching implications of ocean forcings on ice sheet instability and sea ice retreat have also raised considerable attention. To enhance our understanding of the influences of meltwater discharge on abrupt climatic changes and interactions among ice sheet instability, sea ice variability, and sea surface characteristics, it is essential to investigate sedimentary records on different timescales beyond modern observation. To achieve these objectives, we conducted detailed paleoenvironmental reconstructions from three sediment cores: one obtained from the eastern Labrador Sea (Core MSM12/2-05-01) and two others from southern Baffin Bay (Core SL170 and SL174).
This study provides new insights into variations in sea surface characteristics and their interactions with ice sheet activities in the Labrador Sea and Baffin Bay during the last glacial-deglacial-Holocene period. Meanwhile, our findings provide more information about the impact of freshwater forcing/ice sheet activities on abrupt climate changes. Furthermore, these findings highlight the significance of oceanic processes within the polar and subpolar regions and their contributions to paleoclimate changes