TY - CHAP A1 - Wagner, Robin A1 - Kitzelmann, Emanuel A1 - Boersch, Ingo T1 - Mitigating Hallucination by Integrating Knowledge Graphs into LLM Inference – a Systematic Literature Review T2 - Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) N2 - Large Language Models (LLMs) demonstrate strong performance on different language tasks, but tend to hallucinate – generate plausible but factually incorrect outputs. Recently, several approaches to integrate Knowledge Graphs (KGs) into LLM inference were published to reduce hallucinations. This paper presents a systematic literature review (SLR) of such approaches. Following established SLR methodology, we identified relevant work by systematically search in different academic online libraries and applying a selection process. Nine publications were chosen for indepth analysis. Our synthesis reveals differences and similarities of how the KG is accessed, traversed, and how the context is finally assembled. KG integration can significantly improve LLM performance on benchmark datasets and additionally to mitigate hallucination enhance reasoning capabilities, explainability, and access to domain-specific knowledge. We also point out current limitations and outline directions for future work. KW - LLMs KW - hallucination KW - knowledge graphs KW - inference KW - literature review Y1 - 2025 UR - https://aclanthology.org/2025.acl-srw.53.pdf U6 - https://doi.org/10.18653/v1/2025.acl-srw.53 SP - 795 EP - 805 PB - Association for Computational Linguistics CY - Vienna ER - TY - CHAP A1 - Vielhauer, Claus A1 - Loewe, Fabian A1 - Pilgermann, Michael T1 - Towards Modeling Hidden & Steganographic Malware Communication based on Images T2 - IH&MMSEC '25: ACM Workshop on Information Hiding and Multimedia Security, San Jose, CA, USA, 2025 N2 - Recently, an increasing number of IT security incidents involving malware, which makes use of hidden and steganographic channels for malicious communication (a.k.a. as "stegomalware"), can be observed in the wild. Especially the use of images to hide malicious code is rising. In consideration of this shift, a new model is proposed in this paper, which aims to help security professionals to identify and analyze incidents revolving around steganographic malware in the future. The model focuses on practical aspects of steganalysis of communication data to elaborate linking properties to previous code analysis knowledge. The model features two distinct roles that interact with a knowledge base which stores malware features and helps building a context for the incident. For evaluation, two image steganography malware types are chosen from popular databases (malpedia and MITRE ATT&CK®), which are analyzed in multiple steps including steganalysis and code analysis. It is conceptually shown, how the extracted features can be stored in a knowledge base for later use to identify stegomalware from communication data without the need of a thorough code analysis. This allows to uncover previously hidden meta-information about the examined malicious programs, enrich the incident’s forensic context traces and thus allows for thorough forensic insights, including attribution and improved preventive security measures in the future. Y1 - 2025 UR - https://dl.acm.org/doi/pdf/10.1145/3733102.3733152 U6 - https://doi.org/10.1145/3733102.3733152 SP - 52 EP - 63 ER - TY - CHAP A1 - Trancón y Widemann, Baltasar A1 - Lepper, Markus ED - Ertl, M. Anton ED - Kirsch, Christoph M. T1 - Imagine There’s No Source Code: Replay Diagnostic Location Information in Dynamic EDSL Meta-programming T2 - MPLR 2024: Proceedings of the 21st ACM SIGPLAN International Conference on Managed Programming Languages and Runtimes N2 - Programs in embedded domain-specific languages are realized as graphs of objects of the host language rather than as static input texts. This property enables dynamic meta-programming, but also makes it harder to attach location information to diagnostic messages that arise at a later stage, after the program graph construction. Thus, EDSL-generating expressions and algorithms can be difficult to debug. Here, we present a technique for transparently capturing and replaying location information about the origin of EDSL program objects. It has been implemented in the context of the LLJava-live EDSL-to-bytecode compiler framework on the JVM. The basic idea can be generalized to other contexts, and to any managed runtime environment with reified stack traces Y1 - 2024 U6 - https://doi.org/10.1145/3679007.3685061 SP - 1 EP - 11 PB - ACM CY - New York ER - TY - CHAP A1 - Tetzlaff, Laura A1 - Purohit, Anne-Maria A1 - Spallek, Jacob A1 - Holmberg, Christine A1 - Schrader, Thomas T1 - Evaluating Interoperability in German Critical Incident Reporting Systems T2 - German Medical Data Sciences 2023 – Science. Close to People. Proceedings of the 68th Annual Meeting of the German Association of Medical Informatics, Biometry, and Epidemiology e.V. (gmds) 2023, Heilbronn, Germany N2 - In industrialised countries, one in ten patients suffers harm during hospitalization. Critical Incident Reporting Systems (CIRS) aim to minimize this by learning from errors and identifying potential risks. However, a lack of interoperability among the 16 CIRS in Germany hampers their effectiveness. Y1 - 2023 U6 - https://doi.org/10.3233/SHTI230722 SP - 249 EP - 257 PB - IOS Press ER - TY - CHAP A1 - Simmering, Veronika A1 - Pfeiffer, Thies A1 - Kindsmüller, Martin Christof T1 - The Impact of Embodiment and Output Modality on Learning with Conversational Agents in VR T2 - GI Workshop 2025 on Virtual and Augmented Reality : VR/AR 2025, Lecture Notes in Informatics (LNI), Gesellschaft für Informatik, Bonn 2025 N2 - In Virtual Reality (VR) learning applications, the integration of pedagogical agents is particularly promising, as they function as a virtual social support. With the recent developments in Large Language Models (LLMs), conversational language models are now available that enable natural language interaction in a VR environment. We evaluated four possible designs for LLM-based agents (푛 = 21) in an aircraft-engine training. Focusing on the primary questions when bringing LLM-based agents into VR: whether to use text or speech and whether to embody the conversational agent in 3D or not. Pre- and post-tests were used to measure retention and questionnaires were used to measure the User Experience (UX) of different design variants. Contrary to our hypotheses, retention was higher when using a non-embodied design compared to an embodied design. The output modality did not have a significant impact on the learning success, but it did have an impact on the UX. Y1 - 2025 U6 - https://doi.org/10.18420/vrar2025_12 SN - 2944-7682 SP - 235 EP - 254 PB - Gesellschaft für Informatik ER - TY - CHAP A1 - Querner, Erich A1 - Ewerling, Philipp A1 - Kindsmüller, Martin Christof T1 - Touch gets it done, tangibles make it fun? Balancing objective accuracy and user experience T2 - Mensch und Computer 2025 - Workshopband : MCI-WS11: 18th Workshop “Be-greifbare Interaktion”. Chemnitz. 31. August - 03. September 2025 N2 - This study compares Tangible User Interfaces (TUIs) with conventional touch interfaces (CTIs) on multi-touch tables for exploratory multi-user map applications. We developed a prototype that enables multiple users to interact with map content simultaneously and independently. Four interaction methods were implemented and evaluated: gesture-based touch control, widget-based touch joystick control, and two tangible interaction variants (a joystick and car-steering metaphor). A user study with 15 participants was conducted in which users navigated predefined routes with varying difficulty. We collected both objective performance measurements and subjective user assessments. While touch-based methods yielded higher accuracy in objective metrics, TUI-based interactions were rated significantly better by participants in terms of user experience. Notably, the car-steering tangible control was particularly well-received, highlighting how physical, playful interaction can enhance usability despite lower accuracy. These findings contribute to our understanding of how different interaction paradigms support collaborative exploration on multi-touch surfaces. Y1 - 2025 U6 - https://doi.org/https://doi.org/10.18420/muc2025-mci-ws11-164 SP - 1 EP - 6 PB - Gesellschaft für Informatik ER - TY - CHAP A1 - Pepe, Alberto A1 - Lasenby, Joan A1 - Buchholz, Sven T1 - CGAPoseNet+GCAN: A Geometric Clifford Algebra Network for Geometry-aware Camera Pose Regression T2 - Winter Conference on Applications of Computer Vision, Waikoloa, Hawaii, US, 2024 N2 - We introduce CGAPoseNet+ GCAN, which enhances CGAPoseNet, an architecture for camera pose regression, with a Geometric Clifford Algebra Network (GCAN). With the addition of the GCAN we obtain a geometry-aware pipeline for camera pose regression from RGB images only. CGAPoseNet employs Clifford Geometric Algebra to unify quaternions and translation vectors into a single mathematical object, the motor, which can be used to uniquely describe camera poses. CGAPoseNet solves the issue of balancing rotation and translation components in the loss function, and can obtain comparable results to other approaches without the need of expensive tuning of the loss function or additional information about the scene, such as 3D point clouds, which might not always be available. CGAPoseNet, however, like several approaches in the literature, only learns to predict motor coefficients, and it is unaware of the mathematical space in which predictions sit in and of their geometrical meaning. By leveraging recent advances in Geometric Deep Learning, we modify CGAPoseNet with a GCAN: proposals of possible motor coefficients associated with a camera frame are obtained from the InceptionV3 backbone, and the GCAN downsamples them to a single motor through a sequence of layers that work in G_ 4, 0. The network is hence geometry-aware, has multivector-valued inputs, weights and biases and preserves the grade of the objects that it receives in input. CGAPoseNet+ GCAN has almost 4 million fewer trainable parameters, it reduces the average rotation error by 41% and the average translation error by 8.8% compared to CGAPoseNet. Similarly, it reduces rotation and translation errors by 32.6% and 19.9%, respectively, compared to the best performing PoseNet strategy. CGAPoseNet+ GCAN reaches the state-of-the-art results on 13 commonly employed datasets. To the best of our knowledge, it is the first experiment in GCANs applied to the problem of camera pose regression. Y1 - 2024 U6 - https://doi.org/10.1109/WACV57701.2024.00646 SP - 6593 EP - 6603 PB - IEEE ER - TY - CHAP A1 - Pepe, Alberto A1 - Lasenby, Joan A1 - Buchholz, Sven T1 - GA-ReLU: an activation function for Geometric Algebra Networks applied to 2D Navier-Stokes PDEs T2 - ICLR 2024 Workshop on AI4DifferentialEquations In Science, Wien, Österreich N2 - Many differential equations describing physical phenomena are intrinsically geometric in nature. It has been demonstrated how this geometric structure of data can be captured effectively through networks sitting in Geometric Algebra (GA) that work with multivectors, making them suitable candidates to solve differential equations. GA networks however, are still mostly uncharted territory. In this paper we focus on non-linearities, since applying them to multivectors is not a trivial task: they are generally applied in a point-wise fashion over each real-valued component of a multivector. This approach discards interactions between different elements of the multivector input and compromises the geometric nature of GA networks. To bridge this gap, we propose GA-ReLU, a GA approach to the rectified linear unit (ReLU), and show how it can improve the solution of Navier-Stokes PDEs. Y1 - 2024 UR - https://openreview.net/pdf?id=T54Hn1e98P SP - 1 EP - 6 ER - TY - CHAP A1 - Pepe, Alberto A1 - Buchholz, Sven A1 - Lasenby, Joan T1 - Clifford Group Equivariant Neural Network Layers for Protein Structure Prediction T2 - Northern Lights Deep Learning Conference, VOL 233, 09-11 Januar 2024, Tromso, Norwegen N2 - We employ Clifford Group Equivariant Neural Network (CGENN) layers to predict protein coordinates in a Protein Structure Prediction (PSP) pipeline. PSP is the estimation of the 3D structure of a protein, generally through deep learning architectures. Information about the geometry of the protein chain has been proven to be crucial for accurate predictions of 3D structures. However, this information is usually flattened as machine learning features that are not representative of the geometric nature of the problem. Leveraging recent advances in geometric deep learning, we redesign the 3D projector part of a PSP architecture with the addition of CGENN layers . CGENNs can achieve better generalization and robustness when dealing with data that show rotational or translational invariance such as protein coordinates, which are independent of the chosen reference frame. CGENNs inputs, outputs, weights and biases are objects in the Geometric Algebra of 3D Euclidean space, i.e. G3,0,0, and hence are interpretable from a geometrical perspective. We test 6 approaches to PSP and show that CGENN layers increase the accuracy in term of GDT scores by up to 2.1\%, with fewer trainable parameters compared to linear layers and give a clear geometric interpretation of their outputs. Y1 - 2024 UR - https://proceedings.mlr.press/v233/pepe24a/pepe24a.pdf SP - 1 EP - 7 PB - PMLR ER - TY - CHAP A1 - Orlowski, Maximilian A1 - Knauff, Emilia A1 - Marquardt, Florian T1 - Towards Personal Assistants for Energy Processes Based on Locally Deployed LLMs T2 - Proceedings of the 17th International Conference on Agents and Artificial Intelligence (ICAART 2025) , 2025, Porto, Portugal N2 - This paper presents a coaching assistant for network operator processes based on a Retrieval-Augmented Gen- eration (RAG) system leveraging open-source Large Language Models (LLMs) as well as Embedding Models. The system addresses challenges in employee onboarding and training, particularly in the context of increased customer contact due to more complex and extensive processes. Our approach incorporates domain-specific knowledge bases to generate precise, context-aware recommendations while mitigating LLM hallucination. We introduce our systems architecture to run all components on-premise in an our own datacenter, ensuring data security and process knowledge control. We also describe requirements for underlying knowledge doc- uments and their impact on assistant answer quality. Our system aims to improve onboarding accuracy and speed while reducing senior employee workload. The results of our study show that realizing a coaching assistant for German network operators is reasonable, when addressing performance, correctness, integration and locality. However current results regarding accu- racy do not yet meet the requirements for productive use. Y1 - 2025 U6 - https://doi.org/10.5220/0013175600003890 SP - 695 EP - 706 PB - SCITEPRESS ER -