Konferenzbeitrag: h5-Index > 30
Refine
Document Type
Language
- English (20)
Keywords
- Action Design Research (2)
- Artificial Intelligence (2)
- Attention behaviour (1)
- BITA (1)
- Business-IT Communication (1)
- CBT-I (1)
- COBIT (1)
- Citizen Development (2)
- Computer vision (1)
- Data-set (1)
Neural network representations of simple models, such as linear regression, are being studied increasingly to better understand the underlying principles of deep learning algorithms. However, neural representations of distributional regression models, such as the Cox model, have received little attention so far.We close this gap by proposing a framework for distributional regression using inverse flow transformations (DRIFT), which includes neural representations of the aforementioned models. We empirically demonstrate that the neural representations of models in DRIFT can serve as a substitute for their classical statistical counterparts in several applications involving continuous, ordered, timeseries, and survival outcomes. We confirm that models in DRIFT empirically match the performance of several statistical methods in terms of estimation of partial effects, prediction, and aleatoric uncertainty quantification. DRIFT covers both interpretable statistical models and flexible neural networks opening up new avenues in both statistical modeling and deep learning.
Transfer of Logistics Optimizations to Material Flow Resource Optimizations using Quantum Computing
(2024)
The complexity of industrial logistics and manufacturing processes increases constantly. As a key enabling technology of the upcoming decades, quantum computing is expected to play a crucial role in solving arising combinatorial optimization problems superior to traditional approaches. This study analyzes the current progress of quantum optimization applications in the logistics sector and aims to transfer an existing vehicle routing use case to a newly conceptualized matrix production use case regarding resource-efficient material flows. The simulation of the originating simple model is executed on a local circuit-based quantum simulator that emulates the behavior of real quantum hardware. Using a QAOA algorithm for problem-solving, optimal results have been achieved for all simulated scenarios. The theoretical material flow model is based on multiple assumptions and was created for testing reasons exclusively. For a realistic practical application, the model must therefore first be adapted and extended to include additional features.
The Operationalization of Strategic Alignment: Driving IT Value through Business-IT Dialogues
(2025)
Information technology (IT) as a strategic asset urges organizations to establish strategic alignment between business and IT stakeholders to maximize the contribution of IT. Yet, they still struggle to effectively communicate the strategic perspective and the business value of IT. In three organizational settings, we conduct action design research to effectively communicate IT business value. We develop a structured end-to-end process with four steps: initialization, stakeholder identification, strategic dialogues, and consolidation & analysis. We derive eight design principles to design and implement this business-IT dialogue format that strengthens mutual understanding and strategic perspective. Hence, this paper provides guidance for organizations to establish strategic conversations between business and IT stakeholders and advances research on operationalizing business-IT alignment.
This paper presents an approach to control the temperature of an industrial heating process where a workpiece is transported through a heating unit in order to reach a certain temperature. To control this process is challenging due to the variable and possibly abruptly changing feed rate while requiring the final workpiece temperature to be maintained at a constant value. The speed itself cannot be influenced by the control system. However, its future trajectory is known to a certain extent at runtime. A model predictive controller is presented, which is characterized by the fact that the prediction horizon is dynamically adjusted to the transportation speed, that the traveled distance of the workpiece within the horizon is constant. Moreover, the step size of the integrator is also adapted, yielding a constant number of integration steps despite large variations in velocity. Because of the special structure of the used model, the system state cannot be reused from one time step to the next and needs to be reconstructed in a receeding horizon fashion, even if the model was perfect. This is also a distinctive feature of the presented control scheme. Simulation and measurement results are provided, showing that the presented controller achieves promising results and outperforms the existing lookup-table-based controller currently in use.
Efficient and safe autonomous control of surface
vessels is seminal for the future of maritime transport systems.
In this paper, we use an iterative learning–based nonlinear
model predictive control scheme leveraging past experiences of
the motion of vessels in a current field to reach optimal behavior.
We define an optimal control problem including a detailed vessel
model but only a roughly estimated current model. This current
model is improved from trial to trial. The learned controller is
compared to a linear track controller, a zero–offset nonlinear
model predictive controller without current information, and a
nonlinear model predictive controller including a perfect model
of the current field. The results of this comparison show that by
including experiences from previous trials, the controller can
improve its performance significantly.We believe that numerical
optimal control has the potential to disrupt the future control
design of maritime systems.
Semi-structured regression models enable the joint modeling of interpretable structured and complex unstructured feature effects. The structured model part is inspired by statistical models and can be used to infer the input-output relationship for features of particular importance. The complex unstructured part defines an arbitrary deep neural network and thereby provides enough flexibility to achieve competitive prediction performance. While these models can also account for aleatoric uncertainty, there is still a lack of work on accounting for epistemic uncertainty. In this paper, we address this problem by presenting a Bayesian approximation for semi-structured regression models using subspace inference. To this end, we extend subspace inference for joint posterior sampling from a full parameter space for structured effects and a subspace for unstructured effects. Apart from this hybrid sampling scheme, our method allows for tunable complexity of the subspace and can capture multiple minima in the loss landscape. Numerical experiments validate our approach’s efficacy in recovering structured effect parameter posteriors in semi-structured models and approaching the full-space posterior distribution of MCMC for increasing subspace dimension. Further, our approach exhibits competitive predictive performance across simulated and real-world datasets
Driving behaviour is a critical factor in accidents today. Physiological factors have a significant impact on driving behaviour. A potential solution lies in vehicle services that benefit from sensing environmental conditions to improve road safety, such as collision avoidance routines in driver assistance systems. Stress, assessed subjectively or physiologically, influences decision making and behaviour, with implications for individuals and the economy. In this paper we present a novel approach to formulate a risk index by combining data from subjective self-reports and objective physiological measures (in particular heart rate). The model identifies stress tendencies in driving behaviour by monitoring behavioural and physiological markers. We present our evaluation results and explore potential ways to implement the model in vehicle systems and its implications for improving road safety. We discuss potential enhancements to improve driving safety and enable timely responses in situations with an increased risk of accidents due to stress or drowsiness.
Development of a digital CBT-I tool for user-friendly treatment and observation of insomnia patients
(2024)
Sufficient and regular sleep is essential for people’s physical and psychological health and a good life quality. Insomnia is one of the most frequent sleep disorders worldwide. Patients with a significant leak of sleep can also have a higher risk for depression. The main goal of this research was to create the concept of a software solution based on “Cognitive behavior therapy for insomnia” or short CBT-I, to support patients to enhance their sleep quality. But also, to make the monitoring of multiple patients more efficient and accessible, for the therapist’s side. Methods that can be quantified and are subsequently easy to automate were transferred, form CBT-I. One of the most essential components of this concept is the sleep diary, being used for the collection of the data for further processing. The PSQI questionnaire with an evaluation function was also included because it is an important tool for the therapist side. To guarantee access to the data and results for both sides, the data are stored on an online database. An internal test with 3 users confirmed a good user experience and has shown that the implementation of a user-friendly CBT-I based software solution for the simultaneous use is realizable and can provide benefits for patients and therapists. Like better monitoring and the automation of numerous proven CBT-I methods, e.g. sleep restriction or the transfer of CBT-I knowledge, making a considerable part of the process more efficient. This increase in efficiency can enable therapists to treat more patients simultaneously, while maintaining the same level of quality.
The classification of sleep and wake states is of paramount importance in the context of sleep disorders. In order to detect and monitor disorders such as obstructive sleep apnea (OSA), it is essential to obtain the total sleep time (TST) so as to assess the severity of the patient’s sleep apnea. With the advent of new technologies for detecting events associated with sleep disorders, it is not always straightforward to calculate the sleep/wakefulness state. Consequently, this work presents the development of a deep learning model (a variant of U-Net) for the detection of sleep/wakefulness states. For this purpose, an engineering approach using Keras Tuner and the use of three signals with minimal processing was employed. The three signals, oxygen saturation (SpO2), heart rate (HR) and abdominal respiratory effort (AbdRes), were selected to ensure both patient comfort during signal collection and the possibility of using portable monitors. The models were trained and tested on data from polysomnography studies, namely the Sleep Heart Health Study (SHHS) and the Multiethnic Study of Atherosclerosis (MESA). The best performing model achieved results with 88% binary precision, 88% recall, 89% precision, 89% f1-score and Cohen’s Kappa of 0.74 for the SHHS test set. The model obtained 82% binary accuracy, 82% recall, 84% precision, 82% f1-score and 0.62 Cohen’s kappa for the MESA data set.
Deployment of Artificial Intelligence Models for Sleep Apnea Recognition in the Sleep Laboratory
(2024)
There are a large number of scientific publications that focus on the development and evaluation of artificial intelligence (AI) models for the detection of various pathologies in the field of sleep medicine. However, most of these publications do not show the process or methodology to be followed for the final deployment of these models in a complete diagnostic system (in terms of software and hardware). This is a major drawback when translating from the development or research environment to the real clinical setting. This work focuses on a methodology for deploying an AI model for sleep apnea detection with the end user in mind: the clinician. For the deployment, the transmission of data between the device, the cloud platform and the machine learning server, as well as the protocols used, were considered. In addition, the storage and visualization of the data has been taken into account so that it can be analyzed accurately by experts.
