05 Fakultät Informatik, Elektrotechnik und Informationstechnik
Permanent URI for this collectionhttps://elib.uni-stuttgart.de/handle/11682/6
Browse
970 results
Search Results
Item Open Access Classifying physical exercises and counting repetitions using three-dimensional pose estimation(2023) Wallmann, JonasResistance training is known to increase physical and mental health but requires a lot of knowledge and experience to be done effectively and safely. Personal trainers and physiotherapists provide their knowledge to athletes but their profession requires a lot of learning and experience, thus making their services often not affordable to the general public. Automating certain aspects of their work will make their services more available to the general population and therefore lead to more safe and more effective athletes. The first steps of automating personal training lie in observing a subject train and understanding their performed workout. This provides the basics for future work of automating providing feedback on exercise execution and improving their training regimes. In order to do so, we developed a proof-of-concept program, that uses a two-dimensional camera video as an input to classify what exercise a user performs and automatically counts the number of performed repetitions, in real-time. It should work without imposing requirements in the camera perspective or needing to know what exercise will be performed in advance. This is achieved by using a three-dimensional pose estimation model and defining a rule-based algorithm, that considers the position and angle of joints that characterize the performed exercises We evaluate our proof-of-concept program using videos of subjects performing squats and push-ups in order to understand the accuracy in a real-world scenario. Our program achieved an overall accuracy of 95.57% for the squats and 93.69% for the push-up evaluation.Item Open Access Development of an Euler-Lagrangian framework for point-particle tracking to enable efficient multiscale simulations of complex flows(2023) Kschidock, HelenaIn this work, we implement, test, and validate an Euler-Lagrangian point-particle tracking framework for the commercial aerodynamics and aeroacoustics simulation tool ultraFluidX, which is based on the Lattice Boltzmann Method and optimized for GPUs. Our framework successfully simulates one-way and two-way coupled particle-laden flows based on drag forces and gravitation. Trilinear interpolation is used for determining the fluid's macroscopic properties at the particle position. Object and domain boundary conditions are implemented using a planar surface approximation. The whole particle framework is run within three dedicated GPU kernels, and data is only copied back to the CPU upon output. We show validation for the velocity interpolation, gravitational acceleration, back-coupling forces and boundary conditions, and test runtimes and memory requirements. We also propose the next steps required to make the particle framework ready for use in engineering applications.Item Open Access Locking-enabled security analysis of cryptographic circuits(2024) Upadhyaya, Devanshi; Gay, Maël; Polian, IliaHardware implementations of cryptographic primitives require protection against physical attacks and supply chain threats. This raises the question of secure composability of different attack countermeasures, i.e., whether protecting a circuit against one threat can make it more vulnerable against a different threat. In this article, we study the consequences of applying logic locking, a popular design-for-trust solution against intellectual property piracy and overproduction, to cryptographic circuits. We show that the ability to unlock the circuit incorrectly gives the adversary new powerful attack options. We introduce LEDFA (locking-enabled differential fault analysis) and demonstrate for several ciphers and families of locking schemes that fault attacks become possible (or consistently easier) for incorrectly unlocked circuits. In several cases, logic locking has made circuit implementations prone to classical algebraic attacks with no fault injection needed altogether. We refer to this “zero-fault” version of LEDFA by the term LEDA, investigate its success factors in-depth and propose a countermeasure to protect the logic-locked implementations against LEDA. We also perform test vector leakage assessment (TVLA) of incorrectly unlocked AES implementations to show the effects of logic locking regarding side-channel leakage. Our results indicate that logic locking is not safe to use in cryptographic circuits, making them less rather than more secure.Item Open Access The impact of domain models on energy consumption of classical planners(2025) Tekin, SerhatThe increasing integration of artificial intelligence into real-world systems has intensified concerns about the ecological footprint of computational processes. As the capabilities of AI expand and their applications spread into diverse areas of society, questions of efficiency are no longer confined to algorithmic performance alone but extend to the broader impact of computation on energy usage. Within this context classical planning provides a particularly relevant case since it is a core technique in automated planning. Research in this field has traditionally emphasized runtime efficiency, plan quality and algorithmic design while the energetic dimension has remained largely neglected. This thesis examines that omission by shifting the focus from planners to the domain models that constitute their input. Through systematic modifications of syntactic, semantic and solvability related features it demonstrates that modeling decisions can exert a measurable influence on energy consumption. Rather than viewing energy use as an inherent property of planners, the study shows it to emerge from the interaction between algorithmic behavior and representational form. The work introduces a replicable framework that combines controlled domain transformations with fine grained energy measurements, thereby enabling systematic evaluations of energy usage in symbolic AI. The empirical analysis indicates that syntactic variations usually result in only minor fluctuations, whereas modeling inefficiencies can increase energy demand, with operator arity standing out as a recurring factor. The most pronounced effects arise from solvability constraints which, depending on the planner and the domain, can lead to substantial increases in energy usage or in some cases reductions. Taken together the results highlight that domain modeling is not only a matter of syntactic correctness or semantic adequacy but also of energetic efficiency. The contribution of this thesis is twofold. It establishes a framework for investigating the energy implications of domain features and provides empirical evidence that modeling choices shape the energy profile of planning systems. These findings offer a foundation for further research and provide practical guidance for approaching domain modeling with energy consumption in mind.Item Open Access Prompt-based personality profiling : filtering social media posts using reinforcement learning(2024) Hofmann, JanAuthor profiling is the task of inferring characteristics about individuals by analyzing content they share. To date, systems that perform this task automatically, predominantly use supervised machine learning approaches and borrow from advances in the field of natural language understanding. However, while for many language understanding tasks immense progress has been made in recent years (e.g. by using pre-train then fine-tune paradigm), such progress most often does not transfer to automated profiling systems directly. One reason for this is that author profiling is inherently different from typical text inference tasks due to the possibly large amounts of content associated with an author. Therefore, this work proposes a new method for profiling that tries to select the most relevant parts of content shared by an author before inferring a characteristic. Here, instead of relying on ground-truth labels, this work uses the feedback from the zero-shot capabilities of a large language model to learn such a selection model via reinforcement learning, and evaluates this approach for personality profiling in social media. In experiments predicting big five personality traits, this work finds that prediction quality of such a system is comparable yet slightly worse to using all content associated to a profile in a zero-shot setting, while prediction time is reduced significantly due to the limited amount of content used for inferring personality in the proposed method. In addition, this work finds that simply selecting content arbitrarily leads to performance degradation for most traits, and therefore, this work concludes that, to some extent, the proposed method is able to distinguish between relevant and irrelevant content. Further, this work compares the proposed approach to existing supervised approaches and finds that such methods outperform the proposed method substantially. Still, since the ability of the proposed system to distinguish between relevant and irrelevant content of authors is closely tied to the capabilities of large language models, it can be expected that, with advances of such models, prediction quality of the proposed approach will increase in the future.Item Open Access Improving usability of gaze and voice based text entry systems(2023) Sengupta, Korok; Staab, Steffen (Prof. Dr.)Item Open Access Enhancing HTN planning with deep reinforcement learning for method selection(2025) Bahrami, SepidehAutomated planning is a central area within Artificial Intelligence (AI), enabling intelligent behavior in domains such as cloud computing, autonomous systems, context-aware activity recognition, and smart environments. Hierarchical Task Network (HTN) planning, which decomposes complex tasks into simpler subtasks using predefined methods, has proven effective in such structured domains. However, its performance is often constrained by static method selection strategies that lack adaptability to varying planning contexts. To address this limitation, this thesis proposes a neuro-symbolic framework that integrates HTN planning with Deep Reinforcement Learning (DRL), combining the strengths of symbolic reasoning and data-driven learning. Among the available DRL algorithms, Deep Q-Learning (DQL) is particularly suitable due to its off-policy nature, batch-efficient learning, and robust generalization across symbolic planning states. These characteristics align well with deterministic and hierarchical planners, enabling offline learning from curated datasets without requiring interactive exploration. The proposed integration introduces a learning-based decision layer that improves adaptability while preserving the reproducibility and determinism of the underlying planner. The effectiveness of this approach is demonstrated through a comprehensive evaluation across planning efficiency, memory consumption, and plan quality. Results highlight the potential of reinforcement learning to enhance classical HTN systems and support intelligent decision-making in complex, structured environments.Item Open Access Enhancing automotive safety through an ADAS violation dashboard(2024) Senger, TobiasAutonomous Driving (AD) is an active area of research in which Advanved Driver Assistance Systems (ADAS) play an important role. Ensuring the safety of ADAS systems is critical. However, most ADAS systems nowadays make use of Deep Learning or other types of Machine Learning. Formally verifying these systems to ensure their safety is hardly possible. For this reason, Radic explored the use of Runtime Monitoring (RM) to ensure the safety of ADAS systems by detecting violations of several specified Safety Requirements (SR) at runtime. After performing a test run with the system, she manually analyzed the causes of each series of violations in the extracted Violations Report. As this was laborious and time-consuming, this thesis should explore available approaches and techniques to automatically derive the root causes of violation series. To do this, we first perform an exploratory literature search. This allows us to identify that the most suitable approach to address our problem is Root Cause Analysis (RCA) using Language Models (LMs), Large Language Models (LLMs), Knowledge Graphs (KGs), or a combination of them. We perform a Rapid Review (RR) to find concrete techniques for this approach. We then conduct a narrative data synthesis to explore the techniques retrieved with our RR. This allows us to derive a plan to automatically analyze the causes of SR violations in a Violations Report. Our solution is then incorporated into a web-based safety dashboard application. This application enables our safety engineers to configure ADAS use cases, test tracks, and test runs. Then, the safety engineer can select a test run to display an interactive view of the test run. The safety engineer can then select individual violation series and analyze their root causes using our automated RCA solution based on LLMs. To evaluate the effectiveness of our system, we conduct a simple experiment. This experiment shows that our system already achieves comparable performance to a human baseline provided by Radic. Our system, therefore, represents a valuable tool for safety engineers to identify and repair safety-critical problems in ADAS systems in the context of AD. We also propose modified variants of our system that allow researchers to improve our automated RCA system in the future, e.g., by incorporating a KG.Item Open Access Stationary vehicle classification based on scene understanding(2024) Wang, WeitianNavigating through dense traffic situations like merging onto highways and making unprotected left turns remains a challenge for the existing autonomous driving system. Classifying vehicles into parked, stopped, and moving vehicles can benefit the decision-making system in this case because they play different roles during the vehicle-to-vehicle negotiation process. Existing works in vehicle classification focused on trivial cases and used methods that are not generalized enough. To fill this gap, after analyzing this problem and summarizing the necessary information needed for this problem, we propose a multi-modal model that can leverage information from lidar, radar, camera, and high-definition maps. To meet the complexity of our task and the needs of our model, we collect the dataset in real driving scenario and then preprocess and label it. By utilizing a pretrained vision encoder for fine-grained visual feature extraction and vision foundation model (CLIP) for scene understanding, our model achieves a 97.63% test accuracy on our dataset. Through visualization methods, experiments, and quantitative analyses, we investigate the effectiveness and importance of different encoders used in our model. We interpret and explain the successes and failures of our model to give a better understanding of how different latent features contribute to the final result. In the end, the limitations of our model and potential improvements are discussed.Item Open Access Klassifizieren von Bewegungsabläufen mit Machine-Learning-Verfahren unter Verwendung der Movella-Dot-Sensoren(2025) Volpe, AnnaDiese Arbeit untersucht den Einfluss verschiedener Modellarchitekturen und Optimierungsverfahren auf die Klassifikation von Human Activity Recognition (HAR) anhand von Sensordaten der Movella DOT Sensoren. Im Zentrum stehen Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Residual Neural Network (ResNet) sowie klassische Verfahren wie Support Vector Machine (SVM) und Random Forest (RF). Neben dem Vergleich der Optimierer Adaptive Moment Estimation (Adam), Root Mean Squared Propagation (RMSprop) und Stochastic Gradient Descent (SGD) mit verschiedenen Lernraten, wird analysiert, wie sich unterschiedliche Sensorpositionen und deren Kombinationen auf die Klassifikationsgenauigkeit auswirken. Hierzu werden Sliding-Window-Techniken zur Segmentierung der Zeitreihen eingesetzt und die Ergebnisse mit Konfusionsmatrizen, Receiver-operating characteristic (ROC)-Kurven sowie Metriken bewertet. Zur Durchführung der Experimente wurde eine Graphical User Interface (GUI) implementiert, die die Auswahl von Modellen, Hyperparametern und Optimierungsverfahren ermöglicht und die Ergebnisse visualisiert. Die implementierten Modelle wurden auf verschiedenen Bewegungsabläufen trainiert und getestet, darunter Gehen, Joggen, Treppensteigen, Fahrradfahren, Liegestütze, vereinfachte Liegestütze, Bizeps-Curls, Hampelmann, Liegen und Autofahren. Zusätzlich wurden insgesamt acht Handrotationen erfasst, jeweils vier mit der linken und vier mit der rechten Hand. Dabei wurde zwischen Faust- und offener Handhaltung sowie zwischen Links- und Rechtsrotation unterschieden. Es konnte gezeigt werden, dass insbesondere CNNs und ResNets stabile und hohe Genauigkeiten erzielten, während LSTMs stark von der Wahl des Optimierers abhängig waren und klassische Verfahren wie RF solide Ergebnisse bei geringer Modellkomplexität lieferten. Fußsensoren erwiesen sich für geh-und laufbezogene Aktivitäten als besonders zuverlässig, während Handsensoren für armbezogene Bewegungen den größten Mehrwert lieferten. Damit konnte bestätigt werden, dass die gezielte Auswahl weniger Sensoren mit optimaler Platzierung für robuste Klassifikationen ausreicht. Ziel ist es, die Bedeutung der Sensorpositionen und der Hyperparameterauswahl für robuste Modelle zu charakterisieren und so Empfehlungen für die Auswahl geeigneter Modell- und Sensorkonfigurationen aufzustellen.