05 Fakultät Informatik, Elektrotechnik und Informationstechnik
Permanent URI for this collectionhttps://elib.uni-stuttgart.de/handle/11682/6
Browse
295 results
Search Results
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 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 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.Item Open Access Situated interactive guidance and assistance in extended reality using multimodal large language models(2025) Zhao, JiahaoThis paper addresses the need for "providing actionable guidance for process tasks in real-world environments” by proposing an interactive XR system that combines a multimodal large language model with Apple Vision Pro. The system employs a three-stage architecture with cloud-end decoupling: Vision Pro handles acquisition and spatial projection, a Python bridge server manages sessions and state machines, and the LLM backend uses an open-source vision model for object localization and generates voice/text instructions. We also compare local inference models with AWS cloud-based inference models: local deployment offers lower latency, while cloud-based deployment is more scalable but subject to greater EBS throughput and initialization overhead. Finally, a study with eight participants shows that the AR-based SUS had higher usability and lower overall NASA-TLX load. The difference in completion time was not significant but trended faster. Participants also perceived AR guidance as more accurate. The paper discusses trade-offs between local GPU and cloud-based deployment, current technical bottlenecks, and potential future research directions.Item Open Access KGGLDM : Knowledge Graph Guided Diffusion Models for advanced learning(2024) Gupta, AkshatThis thesis explores a novel approach by bridging the gap of diffusion modeling and knowledge graphs, unveiling a potentially groundbreaking direction that serves as the central theme of this work. We propose incorporating knowledge graph guidance into LDM models to augment precise control over sample generation using domain conceptual knowledge.Item Open Access Enhancing safety through human factor monitoring in virtual reality(2025) Muzaffari, Feda HussainHuman-Robot Collaboration (HRC) is a foundational element of modern Industry 4.0 paradigms, creating synergistic manufacturing environments that combine robotic precision with human flexibility. However, the close proximity and dynamic nature of these interactions introduce significant challenges to worker safety and performance, necessitating advanced training and monitoring solutions. This thesis addresses these challenges through the design, development, and evaluation of an immersive Virtual Reality (VR) training system for HRC tasks. The system’s core innovation lies in its capacity for real-time human factor monitoring, which enables a closed-loop adaptive feedback mechanism to enhance operator safety and proficiency. The developed system integrates a sophisticated suite of non-invasive sensors within a distributed, real-time data architecture. A full-body motion capture suit provides continuous kinematic data, which is processed using a stream processing engine (Apache Flink) to perform a real-time ergonomic risk assessment. This assessment is informed by biomechanical principles derived from established methodologies to identify high-risk postures of the neck, back, and shoulders. Concurrently, an integrated eye-tracker within the VR headset monitors the practitioner’s visual attention, identifying periods of inattention or distraction from critical task areas. For post-session analysis, surface Electromyography (sEMG) data is collected from key upper-limb muscles. Due to hardware sampling rate limitations precluding frequency-domain analysis, this research evaluates trends in muscle load using robust time-domain features, such as Root Mean Square (RMS), which are effective for differentiating muscle states [VNR14]. Based on the real-time analysis of ergonomic and attentional data, the system provides immediate, multimodal feedback to the practitioner through visual and auditory cues within the VR environment. Crucially, it also generates adaptive commands that dynamically alter the behavior of the simulated collaborating robot-a KUKA KR210-by modulating its speed or pausing its actions to mitigate identified risks. This research details the complete system architecture, from the Unity-based VR simulation and ROS 2 robot control interface to the Kafka and Flink backend for stream processing and data persistence. The primary contribution of this work is the creation and planned evaluation of this integrated, adaptive system, which aims to demonstrate the feasibility and potential benefits of real-time, data-driven feedback for creating safer, more effective, and human-aware HRC training environments. Keywords: Human-Robot Collaboration, Virtual Reality, Ergonomics, RULA, Fatigue Monitoring, Eye Tracking, Motion Capture, EMG, Real-Time Feedback, Apache Flink, Kafka, MQTT, Industrial Safety.Item Open Access Visual analysis of image data with generative models(2025) Chaudhari, AishwaryaThe increasing ubiquity of high-resolution digital imagery across domains such as professional photography, curatorial practice, and documentation necessitates advanced tools for effective semantic organization and retrieval. This thesis presents the design, implementation, and evaluation of an interactive system that enables semantic filtering and visual exploration of large-scale image datasets using generative AI models, with particular emphasis on usability, transparency, and creative flexibility. Leveraging the capabilities of the GPT-4 language model, the system interprets natural language prompts to filter, cluster, and rank images based on semantic similarity. Central to the interaction paradigm is a visual semantic logic tree that externalizes user intent, facilitates prompt reuse, and supports iterative exploration. In addition, a projection-based clustering view allows for context-aware navigation through thematic image groupings, thereby enhancing discovery and interpretability. The system was developed using a modular architecture comprising a Python-based backend and an interactive D3.js frontend. The framework supports multiple query modes-including binary, rating-based, and projection-driven workflows-offering users differentiated levels of semantic control. Evaluation through domain-specific case studies (e.g., wedding photography, travel documentation, art exhibition curation) demonstrates that the system significantly reduces manual effort while improving cognitive alignment, traceability, and creative agency. This thesis contributes to the broader field of human-AI collaboration by showcasing how large-scale language models can be integrated into visual analytics interfaces that prioritize explainability, adaptability, and domain relevance. The concluding discussion addresses current limitations and outlines future research directions, including domain-specific fine-tuning, multimodal interaction support, and reinforcement-based adaptive learning mechanisms.Item Open Access Exploration of structured data using AI-enabled magic lenses(2025) Grote, MarcelIn the era of increasingly complex and high-dimensional datasets, enabling intuitive and effective data exploration remains a significant challenge. This thesis presents a novel AI-enabled magic lens system designed to support rapid and insightful exploration of structured data through natural language interaction and dynamic visualization. The system projects the provided data onto a 2D canvas based on the prior dimensionality reduction. It leverages large language models (LLMs) to interpret user queries and generate contextually appropriate visualizations or textual responses. Beyond visual exploration, the system supports iterative dialogue with the AI, allowing users to refine visualizations and request explanations. Furthermore, users can document their findings through a notes panel. These notes can be evaluated using an LLM to provide feedback, contextual insights, and suggestions for further exploration. Additional key features include the ability to annotate clusters using LLM-generated labels, compare data subsets using multiple adjustable lenses, and view global visualizations for comparative analysis. To evaluate the system, an expert evaluation was conducted. Results indicate strong support for the system’s effectiveness and usability, highlighting the potential of LLM-enhanced data exploration tools. This work proves that the concept of integrating natural language interfaces and generative AI into interactive visualization systems can significantly enhance data analysis workflows, paving the way for future advancements in human-AI collaborative data exploration.Item Open Access OpenID for Verifiable Credentials : formal security analysis using the Web Infrastructure Model(2023) Hauck, FabianIn our increasingly connected world, digital identities play a fundamental role in delivering secure online services around the globe. To enable the seamless exchange of identification data among various entities, the adoption of standardized protocols is essential. The protocol family OpenID for Verifiable Credentials (OID4VC) is ideally suited for exchanging identities. The two most important protocols in this family are OpenID for Verifiable Credential Issuance (OID4VCI) and OpenID for Verifiable Presentations (OID4VP) with a wide range of applications in e-government as well as in the private sector. A prominent example is the European Digital Identity Framework, which includes these two protocols, among others. This means that any future wallet in the European Union will implement OID4VCI and OID4VP. Therefore, it is extremely important to guarantee their security. This thesis performs a rigorous formal security analysis of both the OpenID for Verifiable Credential Issuance and the OpenID for Verifiable Presentations protocols. In particular, we focus on analyzing the security of both protocols when they interact in an ecosystem. It is not sufficient to consider the two protocols separately, because the interaction between them may introduce new vulnerabilities. Therefore, the formal model created in this thesis models both protocols simultaneously. The model is based on the Web Infrastructure Model (WIM), which closely follows existing web technologies. To describe what security means in this context, we define an authentication security property and a session integrity security property for OID4VCI and OID4VP. We prove that the model is secure with respect to the security properties under the assumption of a vigilant user. If this assumption is violated, we have discovered a number of attacks. This work makes several contributions to the protocol specifications: First, the discovered vulnera- bilities were brought to the attention of the working group. Second, several issues were filed to improve the quality and security of the specifications. Lastly, we engaged in ongoing discussions on related issues.Item Open Access Knowledge graph-based retrieval-augmented generation on enterprise knowledge graphs(2026) Chaturvedi, PrateekRetrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by incorporating external knowledge sources at inference time to improve factual accuracy and reduce hallucination. Recent work has explored Knowledge Graph-based RAG (KG-RAG) to enable structured retrieval and complex reasoning. However, most existing methods are evaluated on open-domain Knowledge Graphs (KGs) and rely on assumptions-such as sparse connectivity and clean schema-that do not hold in industrial settings. In dense, schema-driven enterprise KGs, these assumptions break down, causing uncontrolled search expansion and unreliable reasoning behavior that severely degrades answer quality, thereby limiting the reliability and deployability of existing KG-RAG systems in real-world industrial applications. This gap is critical because enterprise question answering often supports operational, maintenance, and diagnostic decisions, where unreliable reasoning can lead to incorrect conclusions and costly downstream consequences. This thesis investigates KG-RAG in the context of a real-world enterprise KG derived from Bosch’s Configuration Management Database (CMDB) as a representative industrial case study, while targeting challenges common across enterprise KGs. We construct an evaluation benchmark based on business-relevant questions drawn from real enterprise practice, ensuring that the queries reflect actual information needs rather than synthetic or open-domain benchmarks. Our empirical analysis reveals substantial performance degradation compared to open-domain benchmarks, which can be partially attributed to dense graph topology, schema-driven structure, and the prevalence of high-degree attribute nodes that induce uncontrolled search expansion. To enable KG-RAG systems to operate reliably in real enterprise and business settings, we propose a Schema-Aware and Question-Guided Agentic KG-RAG method. It enforces a guided reasoning process that respects enterprise schema constraints, follows the logical structure of business questions, and actively controls search expansion. The proposed approach remains training-free and preserves interpretability, while explicitly incorporating schema and structural constraints into the reasoning process. Experimental results show that the improved framework outperforms all baselines on a diverse set of queries, and ablation studies confirm that these gains arise from explicit agentic control mechanisms tailored to enterprise KG characteristics. Overall, this work shows that existing KG-RAG methods fail to generalize from simplified benchmarks to real enterprise use cases, introduces a benchmark grounded in real-world business queries, and demonstrates an Agentic KG-RAG approach that can significantly improve complex question answering over industrial KGs.