05 Fakultät Informatik, Elektrotechnik und Informationstechnik
Permanent URI for this collectionhttps://elib.uni-stuttgart.de/handle/11682/6
Browse
513 results
Search Results
Item Open Access Generating code for distributed deployments of cyber-physical systems using the MechatronicUML(2022) Stürner, DavidModels are applied in engineering disciplines to describe systems from a higher level of abstraction. In Model-Driven Software Engineering (MDSE), formal models are used to design and verify software systems and to infer platform-specific models and implementations. The MechatronicUML is an MDSE method specifically designed for distributed cyber-physical systems (CPS). This thesis explores how the MechatronicUML may be used for generating code. The exact state of previous code generation approaches is not precisely known. The objective of this thesis is to design and implement a MechatronicUML-based code generator for distributed deployments of CPS. Previous code generation approaches are analyzed for this purpose and one approach is selected and extended to support a particular type of robot car as a target platform. A taxonomy for model-based code generation is proposed to structure the analysis of the previous approaches. Based on the selected previous approach, a code generator is presented and implemented. Additionally, an automotive application scenario is used as a case study for evaluating the concept and the implementation of the presented code generator. This code generator supports modeling the distributed deployment of a CPS with the MechatronicUML and generates platform-specific source code which can be successfully compiled and deployed on the Arduino-based robot cars. Ultimately, the thesis presents a proof of concept to generate the code for a distributed CPS based on the MechatronicUML.Item Open Access Towards a neuro-symbolic approach for occupant activity recognition : combining temporal HTN planning with hidden Markov models(2025) Hösch, PeterThe problem of occupant activity recognition has gained in relevance due to demographic shifts and growing environmental concerns where context-sensitive applications promise to help. The prevalent approach to this problem is based around the use of supervised machine learning, which faces challenges due to its requirement for large amounts of annotated training data and its tendency to overfit. Using preexisting common sense or expert knowledge, usually in the form of ontologies, presents another option, but carries its own set of shortcomings. Recently, the usage of hierarchical task network planning as an alternative to this ontological approach has been proposed. Hybrid systems that utilize both machine learning and preexisting knowledge promise to preserve the strength of both approaches while alleviating their drawbacks. We propose a new hybrid occupant activity system using hierarchical task network planning to support the training of a Hidden Markov Model, which, to the best of our knowledge, has not been done before. In addition, we evaluate the system on real sensor data in order to find out how much merits this new design has. Hereby we attempt and compare multiple approaches to the problem. Although not all methods improve the performance, the results show that the basic idea is sound and can generate measurable improvements.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 Ein Ansatz für IoT-Sicherheitstests basierend auf dem MQTT-Protokoll(2021) Chen, KaiDas Internet der Dinge (IoT) besteht aus einer stark wachsenden Anzahl an vernetzten Geräten und gewinnt immer mehr an Bedeutung. Aufgrund der Komplexität und Heterogenität der verwendeten Technologien existieren im IoT-Bereich viele Sicherheitsprobleme. MQTT ist das meist verwendete IoT-spezifische Protokoll für die Kommunikation, wodurch es einen attraktiven Angriffspunkt darstellt. Daher muss die Sicherheit bei MQTT-Systemen gewährleistet sein. Durch eine Literaturrecherche wurden als Hauptprobleme im Zusammenhang mit der Sicherheit von MQTT die unsichere Standardkonfiguration der Broker, sowie Schwachstellen im Umgang mit fehlerhaften Paketen identifiziert. Das Ziel dieser Arbeit ist, einen Testansatz zu entwerfen, der die Sicherheitsprobleme von MQTT-Broker-Implementierungen mittels automatisierten Sicherheitstests untersucht. Der Ansatz, genannt MQTT-AIO, besteht aus drei Testkomponenten und ist in der Lage, die Konfiguration des Brokers zu analysieren, Angriffe basierend auf Angriffsmustern auszuführen und weitere Schwachstellen mithilfe von Fuzzing zu finden. Eine weitere Komponente überwacht das System während des Testprozesses und zeichnet relevante Daten auf. Die Ergebnisse der Testdurchläufe werden als Bericht ausgegeben und können weiter analysiert werden. Der Testansatz MQTT-AIO wird im Rahmen dieser Masterarbeit prototypisch implementiert und anhand einer Fallstudie validiert.Item Open Access Quantenunterstütztes Clustering mit hybriden neuronalen Netzen(2021) Wundrack, PhilippMaschinelles Lernen und Quantencomputer sind zwei aktuelle Forschungsthemen, die großes Potenzial haben. Aktuell wird erforscht, wie diese beiden Gebiete kombiniert werden können, um voneinander zu profitieren. In diesen Bereich fällt die vorliegende Arbeit. In dieser Arbeit wird untersucht, ob hybride neuronale Netze genutzt werden können, um die Ergebnisse von Clustering-Algorithmen zu verbessern. Hierzu wird auf den Daten Dimensionsreduktion mit hybriden Autoencodern durchgeführt, bevor die Daten den Clustering-Algorithmen übergeben werden. Als Ergebnis konnte festgestellt werden, dass für bestimmte Datensätze Clustering-Algorithmen bessere Cluster erstellen können, wenn Dimensionsreduktion mit hybriden Autoencodern durchgeführt wurde, anstatt mit klassischen Autoencodern oder PCA.Item Open Access Explainability of operating systems(2021) Huschle, TobiasWith the recent rise of machine learning and artificial intelligence, the explainability of software has found its way into the focus of research activities. Black box-like approaches that take critical decisions must be enabled to justify its actions in a comprehensible manner. This thesis takes these considerations and applies them to the area of operating systems and problem analysis thereof. To do so, a user study, conducted among professionals, is presented that shows that simplifying the generation of explanations of the operating system behavior can bring additional value. Furthermore, already available tools will be discussed based on their capabilities with regard to explanation generation. Subsequently, a new approach is proposed that allows to visualize decisions taken by the operating system in a decision graph. These graphs allow to examine how and why a certain value was set by the operating system in a convenient and efficient way. Finally, this approach is evaluated in another user study, which is again conducted among professionals. The final conclusion of this thesis then yields, that an increased focus on explainability capabilities in the context of operating system problem analysis would bring additional value to people working in this area. There is a wide range of other publications that focus on either problem analysis or explainable software, but not on the combination thereof. The proposed approach aims to connect the two areas by providing assistance in deriving explanations and justifications for the internal reasoning processes of operating systems in a convenient way. The potential value is successfully confirmed with an evaluation study conducted among professionals.Item Open Access Towards a framework for algorithm learning : cross-domain induction and the principled limits of algorithmic abstraction(2025) Kunz, PhilippComputation graphs over homogeneous algebras allow for the learning of algorithms from data that sequentially reflect individual computation steps- so-called memory traces - to be framed as a classical machine learning problem, specifically as a classification task. However, previous work revealed two central limitations: (1) Changes in the input data distribution lead to a significant drop in model accuracy, limiting transferability. (2) The concept of abstraction in the context of computation graphs over homogeneous algebras remained vague and lacked formal definition. We evaluate the effectiveness of domain generalization methods in handling shifts in input distributions. Based on a systematic literature review, we select two established approaches and compare their performance with our domain-specific method, DIBE. We conduct a case study using classical comparison-based sorting algorithms and empirically demonstrate that some of these methods improve model robustness to domain shifts. Furthermore, we extend the theory of computation graphs over homogeneous algebras by introducing the notion of a quotient algebra. Quotient algebras are used to formally define two forms of abstraction: structural abstraction, as the compression of linear computation sequences independent of the underlying carrier set, and modular abstraction, as black-box operations. Finally, we prove that modular abstraction constitutes an undecidable decision problem, indicating that this type of abstraction is partially inaccessible to algorithmic approaches.Item Open Access Learning free-surface flow with physics-informed neural networks(2021) Hurler, MarcelThis thesis examines the application of physics-informed neural networks to solve free-surface flow problems modeled with the shallow water equations. Physics-informed neural networks allow training of a surrogate model that resembles the latent solution of an underlying partial differential equation, without using any training data sampled from experiments or numerical simulations. The shallow water equations are an approximation of the Navier stokes equations and serve as a model to many environmental flow problems including dam-breaks, floods, and tsunami propagation. The equations form a non-linear system of hyperbolic partial differential equations that describe the evolution of a fluid's depth and momentum through time. Contrary to other models for free-surface flow, where the exact location of the free surface is only given implicitly as an isosurface and needs reconstruction, here, the depth directly yields its location. One characteristic of the shallow water equations is the formation of steep wavefronts and discontinuities. The thesis examines four state-of-the-art techniques to improve accuracy and training speed and discusses their behavior on three initial value problems. These include the famous idealized dam-break and two depth perturbations, one above a flat and one above varying bathymetry. For each of the scenarios, an inspection of suitable network architectures was considered. Additionally, three different formulations of the physics-informed neural network are presented and tested, where one approach implicitly fulfills the mass conservation and thus eliminates one equation of the system. The results show, that it is possible to train a surrogate model with a relative L^2 error of less than 10^(-4) compared to a solution computed by a high-resolution numerical solver in case of a moderate steepening of wavefronts. A relative error close to 10^(-3) can be achieved for the dam break problem, where the initial conditions are discontinuous, and the solution contains shocks that propagate over time. Additionally, it shows that training with bathymetry is possible and the learned depth approximates the varying underground without any noticeable difference.Item Open Access Plug-and-play domain adaptation for neural machine translation(2023) Kadiķis, EmīlsNeural machine translation has emerged as a powerful tool, yet its performance heavily relies on training data. In a fast-changing world, dealing with out-of-domain data remains a challenge, prompting the need for adaptable translation systems. While fine-tuning is a proven effective adaptation method, it is not always feasible due to data availability, memory, and computational constraints. This thesis introduces a dynamic plug-and-play method inspired by controllable text generation to enhance machine translation across various domains without fine-tuning. This method, called Plug-and-Play Neural Machine Translation (PPNMT), uses a mono-lingual domain-specific bag-of-words to push the hidden state of the decoder through backrpopogation, making the output more in-domain. The method is tested on two types of domains: formality, gender (where the source language does not make a distinction between these aspects, but the target language does), and fine-grained technical domains (which are more based on topic inherent in the text on both the source and target sides). The method performs reasonably well for adapting the translation to different formality levels and, to a lesser extent, grammatical genders, even with an incredibly simple bag-of-words. However, it struggles with adapting the model to technical domains, and a fine-tuning baseline outperforms the proposed method in anything but very low few-shot settings in all tried domains. Despite that, the method shows some interesting behaviour, adapting to the formality on a level that goes beyond just using formal pronouns.Item Open Access Supervised semantic proximity noise and disagreement detection(2024) Choppa, TejaswiThe quality and reliability of annotated data are crucial for the development of Machine Learning models. In this work, we particularly focus on word sense annotation in context (a.k.a. Word-in-Context, WiC). WiC datasets in real-world contexts often exhibit significant disagreement. As a result, information is lost when instances are discarded during the creation of the gold label by adjudicating the annotations through majority or median judgment. Recent advancements have sought to address this issue by incorporating disagreement data through novel label aggregation methods (Uma et al., 2022). Modeling this disagreement is important because, in a real-world scenario, we often do not have clean data. We need to predict on samples where high disagreement is expected and which are inherently difficult to categorize. Predicting disagreement can help detect or filter highly complex samples. Through this thesis, we aim to build machine learning models that predict human disagreement in annotated text instances. Moreover, we focus on data with noise instances where annotators cannot confidently assign a label or the data does not fit predefined categories. We aim to measure both disagreement and noise, as they both stem from a common source: ambiguity. By modeling these aspects, we aim to design modeling approaches that predict not only the semantic proximity label but also the annotator disagreement, as well as data noisiness.