Refine
Document Type
- Doctoral thesis (35)
- Working paper (7)
- Report (5)
- Bachelor thesis (2)
- Habilitation thesis (2)
- Book (1)
- Conference proceeding (1)
- Master thesis (1)
Is part of the Bibliography
- no (54)
Year of publication
- 2021 (54) (remove)
Language
- English (54) (remove)
Keywords
- Nigeria (5)
- Ganzzahlige Optimierung (3)
- Gemischt-ganzzahlige Programmierung (3)
- Kulturerbe (3)
- Maschinelles Lernen (3)
- Tourenplanung (3)
- Altstadt (2)
- Climate change (2)
- Constraint programing (2)
- Constraint-Programmierung (2)
Institute
- FG Ingenieurmathematik und Numerik der Optimierung (7)
- FG Aerodynamik und Strömungslehre (4)
- FG Halbleitermaterialien (3)
- FG Systeme (3)
- FG Anorganische Chemie (2)
- FG Denkmalpflege (2)
- FG Hybride Fertigung (2)
- FG Programmiersprachen und Compilerbau (2)
- FG Thermodynamik / Thermische Verfahrenstechnik (2)
- FG Umweltplanung (2)
Within the last decades, the number of social networks is growing fast. The competition of retaining the customers to grow their platform and increase their profitability is rising. That is why companies need to detect possible churners to retain these. The problem of predicting the users’ lifetime, churning users, and the reasons for churning can be tackled by using machine learning.
The goal of this bachelor thesis is to build machine learning models to predict user churn and the user lifetime within the social network Jodel, a location-based anonymous messaging application for Android and iOS.
To get the best possible prediction results, we have started with extensive literature research, whose approaches we have tested and added to a machine learning pipeline to build predictive models. With these models, we have investigated the performance after different observation time windows and have finally compared the strongest models to detect similarities and understand the insights to learn their behaviour.
The results of this thesis are machine learning models for a selected representative set of communities varying in size within the Kingdom of Saudi Arabia and a country model leveraging all data. These models are used for a regression task by predicting the lifetime of a user and a multi-label classification of a user into six different churn classes. Additionally, we have also given models for a binary classification, where the model will predict if the user will churn within a given time or not. These models have shown general strong predictive power, which is shrinking when limiting the observation time window. Especially the binary classification yielded high accuracy of over 99%.
The best models have been used for predicting user churn within other communities to detect communities with possible similar behaviour. These similarities then have been determined by features’ importance, where the most important features have got fed back into empirics. This has shown statistically significant differences between user groups with a different active time but as of today no clear trends were visible that had led us to define the communities’ behaviours.
Since the competition of social networks is still growing, the retaining of users will stay a core marketing strategy, which will need to be tackled by machine learning and artificial intelligence. The created models could be useful for predicting churning users within the platform Jodel to detect these customers that will churn within a given time.
Researches did not focus much on anonymous and location-based messaging. That is why the results of this thesis on the anonymous messaging application Jodel opens a variety of possible tasks for the future in this context.
Within this work, the growth of out-of-plane Si and SixGe1-x and in-plane Ge nanowire have been investigated. For this purpose, nanowires have been grown ccording to the vapor-liquid-solid mechanism by means of molecular beam epitaxy with Au as metallic solvent on Si(111) and nano-structured Si(001)/SiO2 substrates. Each stage from surface preparation to final nanowire growth has been examined in order to determine relevant parameters which influence the nucleation process nanowire elongation. Particular attention has been given to describe the variation of Au droplet size distribution on Si(111) to evaluate the subsequent nanowire growth and to obtain optimal growth conditions for an site-selective nucleation on the nano-structured substrates.
Due to increased surface diffusion velocity with raising temperature, the droplet diameter distribution shifts from a positive skewed distribution with a high fraction of smaller droplets to a negative skewed distribution with a high fraction of larger droplets. The temperature dependency of the most probable droplet diameter and the number of Au droplets per area has been determined, which can be applied to predict the resulting nanowire diameters.
The out-of-plane Si and SixGe1-x on Si(111) predominantly grow along the surface normal, exhibiting the characteristic sawtooth-like sidewall faceting. During the nanowire formation, the Au surface diffusion velocity increased and causes an increase of the most frequently observed diameter. Furthermore, the total number of droplets/nanowires decreases by a constant factor due to an increase of the contact angle during Si/Ge deposition. The specific diameter range for an possible nanowire formation is reduced by the incorporation of Ge into SixGe1-x nanowires.
To obtain a regular and uniform nanowire growth, a nano-structured substrate consisting of Si(001) terminated pillars surrounded by a SiO2 matrix has been utilized. The initial growth of Ge nanowire starting from Si-Au droplets with SixGe1-x nucleation from ternary alloy is discussed from a thermodynamic point of view and a model based on the Si-Ge-Au ternary phase diagram has been developed to predict the SixGe1-x concentration gradient in the nanowire base. The fully relaxed in-plane Ge nanowires occur within one of the four distinct in-plane ⟨110⟩ directions and nanowires are mainly bounded by two 55° inclined {111} facets and a less pronounced planar (001) top facet.
High-resolution scanning X-ray diffraction microscopy reveals a slightly tilted growth of individual nanowires with respect to each other, causing an abrupt change in the orientation at junction points of interconnected nanowires.
Motivated by computing functionals of high-dimensional, potentially metastable diffusion processes, this thesis studies robustness issues appearing in the numerical approximation of expectation values and their gradients. A major challenge being high variances of corresponding estimators, we investigate importance sampling of stochastic processes for improving statistical properties and provide novel nonasymptotic bounds on the relative error of corresponding estimators depending on deviations from optimality. Numerical strategies that aim to come close to those optimal sampling strategies can be encompassed in the framework of path space measures, and minimizing suitable divergences between those measures suggests a variational formulation that can be addressed in the spirit of machine learning. A key observation is that while several natural choices of divergences have the same unique minimizer, their finite sample properties differ vastly. We provide the novel log-variance divergence, which turns out to have favorable robustness properties that we investigate theoretically and apply in the context of path space measures as well as in the context of densities, for instance offering promising applications in Bayesian variational inference.
Aiming for optimal importance sampling of diffusions is (more or less) equivalent to solving Hamilton-Jacobi- Bellman PDEs and it turns out that our numerical methods can be equally applied for the approximation of rather general high-dimensional semi-linear PDEs. Motivated by stochastic representations of elliptic and parabolic boundary value problems we refine variational methods based on backward SDEs and provide the novel diffusion loss, which can be related to other state-of-the-art attempts, while offering certain numerical advantages.
Wireless communication has become an indispensable part of our life and the demand for achieving higher throughput with lower energy consumption is ever growing. The ambitious throughput of 100 Gb/s and beyond is now becoming a modest goal thanks to comprehensive advances in transmission technologies and protocols. One important aspect of these advances is with regard to channel coding methods and the ability to detect and correct errors at the receiver. Computations needed by such methods become generally more complicated as they become more powerful in their performance. This imposes a great challenge for researchers attempting to devise practical methods for encoding and decoding Forward-error Correction (FEC) techniques tailored for high-throughput scenarios.
In this work we focus on high-throughput Quasi-Cyclic LDPC (QC-LDPC) codes, as they have been selected as one of the main FEC techniques for the two major next generation wireless technologies, namely Wi-Fi 6 (IEEE 802.11ax) and 5G. Our target is to develop complete encoding and decoding design for these codes in order to reach the throughput of 100 Gb/s with affordable power consumption. Toward this goal, we investigate first the appropriate encoder design for these codes which can be used at such high data-rate with reasonably low power consumption. Then we propose several novel ideas for improving the decoding performance and complexity of QC-LDPC codes. The proposed novel ideas collectively facilitate a decoder able to run at 50 Gb/s with less than 12 pJ/b energy consumption for a Latin squares QC-LDPC code. All the proposed methods are practical and implementable and their effectiveness are showcased by either Field Programmable Gate Array (FPGA) or Application-Specific Integrated Circuit (ASIC) synthesis.
In order to integrate necessary interdisciplinary complexity, sustainability, defined targets and endpoints for reclamation progress, the objective of this research is to advance reclamation knowledge on (i) international scale and (ii) national scale, taking Nigeria as an example. On the international scale (i), a systematic reclamation threshold scale for assessing, evaluating, documenting, and monitoring reclamation progress was developed. The threshold stages begin from the initial state I0 down to degraded state D0 (depending on the type of mining). Reclamation starts with soil reconstruction R-2 up to revegetation R-1 (red zones) to reach minimum threshold R0 (amber zone). Beyond R0 are the green zones R1, R2, and R3 representing soil/abiotic condition, biological and improved threshold respectively. Adaptation of the threshold model to degraded ecosystems in national and international laws is recommended. On the national scale with Nigeria (ii), endowed with over 34 solid minerals and abundant crude oil, findings show that almost all mine sites are below the minimum threshold scale R0. The sites are adversely affected by Artisanal and Small Scale Mining (ASM), whereby, contamination and soil lithology disturbances are the state-of-the-art discussions. However, bioremediation (phytoremediation) is the most feasible remediation technique. Therefore, for the crude oil contaminated sites, this study highlights typical cogitation for large in-situ implementation and potential species for remediation. For the mineral resources, gold mining was the typical example. As a practical approach, experimental areas on an abandoned mining site (Site 1), an active mining site (Site 2) and an undisturbed vegetation site (Control) in southwestern Nigeria were mapped out to represent its mining and farming activities. On the sites, (a) the impacts of the ASM activities on the floristic association were assessed. (b) the potential of 2D Electrical Resistivity Tomography (ERT) was explored to characterize the horizontal and vertical landscape homogeneity of the mine sites, and lastly, (c) the study identified native high-accumulating plants growing on the sites that can be employed for phytoremediation. Thus, the results show that (a) the floristic association of the mine sites have been stressed and disturbed. (b) the resistivity structure of the sites shows that ASM has greatly impaired the lithological status of the sites. (c) out of the 9 considered species, Crinum jagus has the highest accumulation factor of 8.71, 37.47, 1.08 and 29.38 for Pb, Cd, Fe and Cu respectively. It is a recommendable novel plant in the tropics. However, future studies must verify its hyperaccumulator potentials, identify other high accumulating species and develop effective approaches and policies to curb ASM activities and enhance remediation in Nigeria.
Reproducibility of simulation experiments is still a significant challenge and has attracted considerable attention in recent years. One cause of this situation is bad habits of the scientific community. Many results are published without data or source code, and only a textual description of the simulation set-up is provided. Other causes are: no complete simulation set-up, no proper output data analysis and inconsistency of published data, which makes it impossible to compare results.
The progress of computational modelling, amount of data and complexity of models requires designing experiments in such a way that ensures reproducibility. A textual description does not provide all the needed details. A computer code is more reliable than a textual description. It is the precise specification that describes a simulation configuration, model, etc. When computer code, data, models and all parameters are provided, the simulation results become reproducible.
The main goal of this thesis is to develop a tool that ensures reproducibility and efficient execution of simulation experiments, often involving many individual simulation runs. The tool should support a wide range of application scenarios, where the typical scenario is simulation of biochemical reaction networks, which are represented as (coloured) Petri nets interpreted in the stochastic, continuous or hybrid paradigm. The model to be simulated can be given in various formats, including SBML.
The result is a command line tool called Spike, which can be used for various scenarios, including benchmarking, simulation of adaptive models and parameter optimization. It builds on a human-readable configuration script SPC, supporting the efficient specification of multiple model configurations as well as multiple simulator configurations in a single configuration file.
Over the last twenty years, Petri nets have been increasingly adopted for modelling and simulating biological systems, as they offer an intuitive and graphical approach for this purpose. Their usability convenience comes from the fact that they offer many types of elements to describe systems in a qualitative and quantitative way. Coloured Petri nets are particularly useful to model systems with repeated components in a compact fashion. Our tool Snoopy for modelling and simulating Petri nets is one of the most well-known tools supporting a family of related Petri net classes comprising stochastic, continuous and hybrid Petri nets, and covering uncoloured and coloured Petri nets alike. However, kinetic information of a biological system, i.e. kinetic parameters may be uncertain, due to many reasons, e.g. environmental factors. Besides, coloured Petri nets as they were previously supported in Snoopy suffered from some inconsistencies. Due to these inconsistencies, exploring the model behaviour using different sizes (scaleability) was not feasible. Both challenges call for a new and more powerful approach integrating the modelling of uncertainties together with modelling features supporting repeated structures in a compact and scalable way.
This thesis comprises two major contributions: Firstly, we introduce the definition and present the simulation algorithm for both uncoloured and coloured fuzzy Petri nets, by extending the existing quantitative uncoloured and coloured Petri nets in Snoopy. This includes discretising the uncertain kinetic parameters to crisp values by using sampling strategies. Secondly, we harmonise coloured Petri nets in Snoopy with their uncoloured counterparts and we extend the Snoopy’s coloured Petri nets by all the features, which are supported by the coloured abstract net description language - an exchange format of coloured Petri nets in our PetriNuts tool family.
By performing fuzzy simulation, one can obtain two kinds of output: fuzzy bands of each output variable and their corresponding timed-membership functions. Each fuzzy band describes the uncertainties associated with the input, whereas membership functions give more accurate information about the associated uncertainties. The most important features that we obtain by harmonising coloured Petri nets are to develop scaleable models, by defining scaling factors as constants and unifying the usage of coloured Petri nets with the other tools in our PetriNuts tool family.
Cultural heritage faces many types of threats such as neglect, conflicts, disasters caused by natural and human induced hazards, unplanned infrastructure projects, climate change, etc. Over time, a number of institutions, organisations, programmes and initiatives to address the issues related to heritage at risk through varied modes of interventions such as advocacy work, conservation efforts, funding, awareness-raising regarding particular sites, etc. have emerged. Despite the stimulus and support of the various institutions, organisations and initiatives, diverse threats continue to impact cultural heritage in present times with some factors amplifying more than before, leading to the subsequent damages to or losses of heritage assets. Though reasons for these issues are varied and complex, one of the important factors to look into is the agency of the diverse ‘Heritage at Risk’ programmes or initiatives. ICOMOS being one of the pioneers in this global discourse of threatened heritage, its role is strategically vital. 20 years after ICOMOS’s first attempts to initiate one of the first ‘Heritage at Risk’ programmes, this research is an opportunity to re-examine and analyse the effectiveness of ICOMOS’ ‘Heritage at Risk’ programme within the current organisational/ institutional landscape of initiatives in the field of heritage at risk, that ICOMOS is embedded in. It is also an opportunity to situate the programme within the diverse views of the heritage at risk framework and the continuing shifts in the foundational paradigms of the international discourse on heritage conservation. The purpose of this analysis is to guide programme improvement for optimising efforts in favour of cultural heritage at risk within and/ or beyond ICOMOS and improve their ability to contribute towards - safeguarding heritage at risk, increasing the resilience of cultural heritage and sustainable development.
Mixed signal integrated circuits (MSICs) contain both analog and digital circuitry, integrated together on the same die. While this integration brings many benefits, it also gives rise to the issue of substrate noise coupling between the noisy digital circuitry and noise-sensitive analog circuitry. In order to counteract this issue, various substrate noise reduction methodologies had been developed so far. This thesis explores a new approach in substrate noise reduction – using a GALS (globally-synchronous, locally-asynchronous) design strategy for the digital part of a MSIC, in order to reduce the noise generation at its source. GALS architectures consist of several locally synchronous modules (LSMs) which communicate asynchronously to each other. By converting an initially synchronous architecture of digital circuitry into a GALS architecture, simultaneous switching noise generated by this circuitry can be reduced. While GALS had already been used for reducing other types of simultaneous switching noise, this is, to the best of the author’s knowledge, the first attempt to develop a GALS-based methodology for substrate noise reduction. In order to be able to theoretically analyze GALS-based methodologies for substrate noise suppression, corresponding models at high abstraction level for substrate noise generation and substrate noise propagation in lightly doped substrates (which is a type of substrate mostly used for MSICs) have been developed. These models have further been used for developing two new GALS-based substrate noise reduction methodologies: harmonic-balanced plesiochronous GALS partitioning (HB) and harmonic-and-area-based plesiochronous GALS partitioning with power domain separation (HAB). A theoretical analysis has shown that HB can reach substrate noise attenuation of up to 20log(M), where M is the number of LSMs of the resulting GALS system. On the other hand, the attenuation achievable by HAB depends on the distribution of switching current harmonics and area among the partitions, as well as from the substrate itself. For each of the two methodologies, a suitable partitioning procedure for a practical application has been developed; these partitionig procedures have been numerically evaluated in MATLAB. HB has further been embedded within the EMIAS CAD tool, where it has been evaluated on a real design example – a wireless sensor node. A special case of HAB for low frequencies has been applied for developing a test chip called SGE (power domain Separation and Galsification Experiment). The measurements on silicon have proved the applicability of the methodology.
Distributive aspects in the design of payments for ecosystem services and agri-environmental schemes
(2021)
Payments for ecosystem services (PES) and agri-environmental schemes (AES) are often introduced with the goal to counteract overexploitation of natural habitats and biodiversity loss in forests and agricultural landscapes. Their ecological effectiveness and cost-effectiveness have been, however, repeatedly questioned in the literature.
One way of increasing the cost-effectiveness of such schemes is spatial differentiation of payments, which reflects regional differences in conservation benefits and costs and thus maximizes the ecological effect achieved with available financial resources.
Another major critique for PES and AES is that they often pursue poverty alleviation and rural development objectives, which can compromise their cost-effectiveness. As the Tinbergen rule in economic theory suggests, each policy goal requires a separate policy instrument.
However, as other policy instruments, PES and AES have distributional effects. Especially in developing countries, where governments are often weak and less able to successfully implement multiple policies, the distributional effects might not be adequately addressed. The distributional implications of PES and AES could even have an adverse effect on the cost-effectiveness of the schemes, if due to low acceptance among the potential ecosystem service (ES) providers, participation rates in these voluntary schemes are lower than optimal or if transaction costs are affected. From an economic perspective, if distributional goals are important for the design of PES and AES, the fairness preferences of potential ES buyers should be considered.
This thesis uses three case studies to address selected aspects of distribution and fairness in PES and AES from different perspectives. It provides insights into the preferences of potential ES buyers in Germany for distribution and other fairness aspects in PES and AES using two choice experiments and analyses possible distributional effects of cost-effective homogeneous and regionalized payments in AES.
The surveys among potential ES buyers suggest preferences for maximin and equal distribution of payments, and preferences for animal welfare and biodiversity conservation (for local as well as for distant biodiversity), which could be directions for improvement of AES and PES.
Regarding distributional impacts of AES, a trade-off between equality and cost-effectiveness is confirmed. Equity generally increases with improved cost-effectiveness of homogeneous payments. Spatially differentiated payments improve the cost-effectiveness substantially, by aligning the payments to the regional costs and the measures to the regional specifics and potential ecological benefits. However, they have substantial redistribution effects, poorer regions are adversely affected. This shows the importance of analyzing the distributional implications of cost-effective AES and PES on different spatial levels and calls for a holistic approach in policy evaluation and implementation.