05 Fakultät Informatik, Elektrotechnik und Informationstechnik
Permanent URI for this collectionhttps://elib.uni-stuttgart.de/handle/11682/6
Browse
214 results
Search Results
Item Open Access Distributed Deep Reinforcement Learning for Learn-to-optimize(2023) Mayer, PaulIn the context of increasingly complex applications, e.g., robust performance tuning in Integrated Circuit Design, conventional optimization methods have difficulties in achieving satisfactory results while keeping to a limited time budget. Therefore, learning optimization algorithms becomes more and more interesting, replacing the established way of hand-crafting or tweaking algorithms. Learned algorithms reduce the amount of assumptions and expert knowledge necessary to create state-of-the-art solvers by decreasing the need of hand-crafting heuristics and hyper-parameter tuning. First advancements using Reinforcement Learning have shown great success in outperforming typical zeroth- and first-order optimization algorithms, especially with respect to generalization capabilities. However, training still is very time consuming. Especially challenging is training models on functions with free parameters. Changing these parameters (that could represent, e.g., conditions in a real world example) affects the underlying objective function. Robust solutions therefore depend on thorough sampling, which tends to be the bottleneck considering time consumption. In this thesis we identified the runtime bottleneck of the Reinforcement Learning Algorithm and were able to decrease runtime drastically by distributing data collection. Additionally, we studied the effects of combining sampling strategies in regards to generalization capabilities of the learned algorithm.Item Open Access Scheduling with uncertainty for Time-Sensitive Networking using robust optimization techniques and integer linear programming(2024) Bauer, FlorianApplication services depend on the network to guarantee reliability, which is critical for safety and correct operation. Time-Sensitive Networking is a technology for reliable real-time communication of time-sensitive applications. While many schedulers exist that provide reliability for wired Time-Sensitive Networks (TSN) with the assumption of deterministic packet delays, scheduling for wireless TSN with uncertain packet delays has received significantly less attention. This work leverages the methodology of Robust Optimization (RO) to propose a robust scheduling approach that ensures provable reliability for both wired and wireless TSN. An uncertainty set defines the range of possible values, ensuring that the schedule remains feasible under all possible realizations within this set. As uncertainty sets are a key component in RO, we introduce methods to compute boxed and polytope uncertainty sets containing possible packet delays based on a set of given reliability requirements. A scheduler is deemed robust if it satisfies the given reliability constraints for all possible packet delays within the computed uncertainty set. Although robustness can be achieved through strict isolation and conservative filtering of packets, we demonstrate that several limitations prevent known robust schedulers from fully exploiting arbitrary uncertainty set shapes. As certain problem instances are unsolvable using simple boxed uncertainty sets, we indicate the need for schedulers that can utilize complex shapes of uncertainty sets rather than boxes. In response to this challenge, we introduce Uncertain No-Wait Packet Scheduling (UNWPS), a scheduler capable of computing robust schedules, and prove that UNWPS is robust against arbitrary upper-bounded boxed and polytope uncertainty sets. We assess the influence of uncertainty sets on the quality of the resulting UNWPS schedules, compare their performances to the performance of other robust scheduling approaches across various exemplary TSN networks and message stream configurations and carry out simulations conducted using the DetCom simulation framework to validate the robustness of UNWPS empirically.Item Open Access Second-order projection-based mapping methods for coupled multi-physics simulations(2022) Ariguib, BoshraData mapping describes the exchange of variables between different, usually non-matching grids for storing data. As different physics require different physical constraints, so do different simulation require different mesh properties. This makes data mapping a crucial part when coupling single-physics simulations into a multi-physics simulation. However, the tradeoff for the available computationally efficient methods is usually low accuracy order. Such a method is the nearest-neighbor mapping method, which relies on a computationally inexpensive mapping algorithm and shows a first-order accuracy, as it is based on a constant interpolation. A second-order projection-based mapping method nearest-neighbor-gradient aims to improve the accuracy order of the nearest-neighbor mapping, while preserving the low computational costs. This is achieved through the extension of the existing method by considering additional gradient data information and applying a Hermite interpolation, in order to balance out both the computational efficiency and the accuracy of the mapping. In this thesis, we implemented this method by extending the coupling library preCICE, which uses state-of-the-art algorithms for coupling partitioned multi-physics simulations in a black-box manner. We confirmed the theoretical observations of the expected second-order accuracy of the method and we found that the method shows the best convergence order by contrast with the existing mapping methods, including radial basis function mappings. It also performs just as well as the existing projection-based method in terms of computational cost and outperforms the radial basis function mapping in respect of runtime costs.Item Open Access A deep learning approach for large-scale groundwater heat pump temperature prediction(2022) Scheurer, StefaniaHeating and cooling buildings is one of the most energy-intensive aspects of modern life. To minimize the impact on global warming and decelerate climate change, more efficient and carbon emission-mitigating technologies such as openloop groundwater heat pumps (GWHP) for heating and cooling buildings are being used and quickly adopted. Nowadays, in order to guarantee their optimal use and prevent negative interactions, city planners need to optimize their placement in the urban landscape. This optimization process requires fast models that simulate the effect of a GWHP on the groundwater temperature. Considering a large domain with multiple GWHPs, this work introduces a framework for the groundwater temperature prediction. While using a learned local surrogate model, a convolutional neural network, to predict the local temperature field around every single GWHP, a physics-informed neural network (PINN) is employed afterwards to correct the global initial solution of stitched together local predictions. As the violations of the physical laws described by the underlying partial differential equation(s) are spatially unevenly distributed, two different methods for drawing sampling points, on the basis of which the training of the PINN to correct the global initial solution takes place, are investigated and compared. This work shows that it is possible for a PINN to correct the global initial solution of stitched together local predictions in a domain with multiple GWHPs. However, there are still opportunities to improve the quality and decrease the computational time of the presented framework. The best method for drawing sampling points depends on the scenario and the placement of the GWHPs. Thus, no general statement can be made, which of the two methods is more suitable. This work provides a good basis for further investigation of the presented framework.Item Open Access A data plane interface for resource-constrained microcontrollers in time-sensitive networking(2026) Kupka, BastianTime-Sensitive Networking (TSN) enables deterministic Ethernet communication through coordinated transmission control. While there are dedicated TSN switches and network interface cards, they are rarely available on resource-constrained microcontrollers. In particular, transmit-time-based scheduling approaches such as the Earliest TxTime First (ETF) queuing discipline are typically not supported on embedded platforms. This thesis investigates the feasibility of implementing ETF-like transmit-time scheduling on a microcontroller using Zephyr RTOS. The implementation targets the NXP i.MX RT1062, which provides a precise PTP hardware clock and compare interrupt mechanism but lacks dedicated hardware support for traffic shaping. The proposed solution integrates transmit-time scheduling into the Zephyr networking stack by combining hardware-triggered interrupts with software-based buffer management inside the network driver. Scheduled frames are prepared in advance and transmitted using a PTP compare event to trigger the transmission routine close to the target time, followed by a short busy-wait phase to improve precision. Best-effort traffic is handled through a driver-level guard band to avoid interference with scheduled transmissions. Experimental evaluation shows that, for periodic traffic with fixed inter-packet gaps, the implementation achieves a bounded transmission window of approximately 3 µs. For variable inter-packet gaps, a larger timing spread is observed. The results demonstrate that ETF-like transmit-time scheduling can be realized on a low-cost microcontroller for certain traffic patterns by leveraging existing PTP hardware features.Item Open Access Investigation of self-learned zeroth-order optimization algorithms(2022) Schüttler, KilianDesigning optimization algorithms manually is a laborious process. In Addition, many optimization algorithms rely on hand-crafted heuristics and perform poorly in applications for which they are not specifically designed. Thus, automating the algorithm design process is very appealing. Moreover, learned algorithms minimize the amount of a priori assumptions and do not rely on hyperparameters after training. Several works exist that present methods to learn an optimization algorithm. In this project, we focus on the reinforcement learning perspective. Therefore, any particular optimization algorithm is represented as a policy. Evaluation of the existing methods shows, learned algorithms outperform existing algorithms in terms of convergence speed and final objective value on particular training tasks. However, the inner mechanisms of learned algorithms largely remain a mystery. A first work has discovered that learned first-order algorithms show a set of intuitive mechanisms that are tuned to the training task. We aim to explore the inner workings of learned zeroth-order algorithms and compare our discoveries to previous works. To address this issue, we study properties of learned zeroth-order algorithms to understand the relationship between what is learned and the quantitative and qualitative properties, e.g., curvature or convexity of the objective function. Furthermore, we study the generalization in relation to these properties. Moreover, we explore the feasibility of finetuning a learned zeroth-order optimization algorithm to a related objective function. Finally we provide guidelines for training and application of learned zeroth-order optimization algorithms.Item Open Access Ansätze für flexible und fehlertolerante modellgetriebene IoT-Anwendungen in dynamischen Umgebungen(2024) Del Gaudio, Daniel; Mitschang, Bernhard (Prof. Dr.-Ing. habil.)Item Open Access Implementing a Cholesky decomposition using SYCL(2025) Bloch, MichalPLSSVM, an LS-SVM implementation, now only uses the Conjugate Gradient algorithm for solving a set of linear equations. However, for an ill-conditioned matrix, it especially gets into trouble, as the converged solution drifts away from the actual solution due to rounding errors. Therefore, this thesis implements a different solver, e.g., the Cholesky Decomposition, which will be implemented in SYCL. We will implement multiple variations of the Cholesky Decomposition algorithm, including a blocked version, and utilize many different features of SYCL. The focus will primarily be on the fastest implementations. In the end, the fastest implementation will be integrated into PLSSVM alongside a Forward and Backward Substitution implementation for solving the set of linear equations. We will conclude with a runtime comparison between the implementations, a comparison of our best Cholesky Decomposition with the Conjugate Gradient using a dataset and a small discussion about numerical errors.Item Open Access Federated reinforcement learning for the edge(2024) Werner, PeterWith the growing demand to perform increasingly complex computations such as image, audio or video processing or machine learning on mobile devices, energy consumption becomes a crucial factor for the lifetime and usability of these devices. A common way to reduce the amount of consumed energy is to offload the computation to a cloud server with a stable power supply. However, the decision of whether to offload a computation or to perform it locally is not always unambiguous. Using cellular or wireless network to send data to, and receive data from an offloading site can require a non-negligible amount of energy surpassing the energy required for local computation. In this thesis linear contextual bandits are employed to make informed offloading decisions based on contextual information of the task to be executed. To reduce the load on the central server federated, individually acting contextual bandits, able to share collected and aggregated data via synchronization, are considered as well. An increasing privacy consciousness among users further motivates the use of a differentially private mechanism to protect the shared data. While more work on privacy mechanisms that do not compromise prediction performance is needed, result on three real-world datasets show that the linear contextual bandit can improve energy saving when compared supervised solutions.Item Open Access Scalable traffic engineering heuristics for time-triggered communication in real-time networks(2026) Geppert, Heiko; Rothermel, Kurt (Prof. Dr. rer. nat. Dr. h. c.)Distributed safety-critical cyber-physical systems require real-time behavior. This means they must respond not just quickly, but in time, to new situations considering both, the task processing and network communication time. From a networking perspective, meticulous, time-driven traffic planning performed at the frame level is necessary to guarantee low end-to-end delay bounds and low latency. This involves carefully planning transmission operations along each time-critical frame's network path are carefully planned, including precise timing, to limit or even eliminate interference from cross-traffic and ensure timely delivery. Since modern real-time systems can consist of hundreds or thousands of devices - for example, large manufacturing plants or continental-sized power grids - the traffic planning must be highly scalable. Although there are many traffic planning approaches in the literature, there is a lack of very fast heuristics that can handle very large stream sets and networks quickly. This thesis investigates traffic planning heuristics and optimization techniques, focusing on different aspects of the traffic planning domain. The traffic planning consists of novel methods for conflict-graph-based scheduling and new heuristics for very large instances of traffic planning problem. The optimizations include multicast partitioning, which combines the benefits of multicast and unicast traffic plans, and load-balanced stream placement, which generates traffic plans that can accommodate additional streams joining the system later. We created prototype implementations and analyzed their performance in solving the traffic planning problem. Our traffic plans yielded a higher accumulated network throughput or admitted more streams while maintaining computation times ranging from sub-seconds to minutes, even for extremely large-scale problem instances. The traffic planning methods and optimization techniques presented in this thesis can be applied to modern real-time networking technologies, such as Time-Sensitive Networking and TTEthernet.