ABSTRACT. AI is becoming more integrated into our daily lives, raising questions about potential threats within these systems and their outputs. In this talk, we will examine the security challenges and threats associated with generative AI. This includes the deception of humans with generated media and the deception of machine learning systems.

In the first part, we look at threat scenarios where generative models produce content that is impossible to distinguish from human-generated content and is used for fraudulent and manipulative purposes; as these models evolve, attacks become easier to automate and harder to detect. This talk will provide an overview of our current challenges in detecting fake media in human and machine interactions and the effects on consumers’ trust. The second part covers exploits of LLMs and agentic AI to disrupt alignment or steal sensitive information and, from an alternative perspective, how prompt obfuscation can help protect intellectual property while maintaining similar utility with minimal overhead, highlighting that defenses in foundation models may require fundamentally different approaches.

ABSTRACT. Large language models fail systematically on modal and con ditional reasoning tasks, yet whether such failures are foreseeable re mains unexplored. Selective prediction, the ability to abstain when a prediction is likely wrong, has been studied for factual NLP tasks but not for structured logical reasoning. We investigate whether LLM fail ures on seven modal and conditional inference patterns are predictable from the structural properties of reasoning problems, without relying on model outputs. Using 3,776 inference instances across nine models and three prompting strategies, we train an external classifier to pre dict per-instance failure from input-side features. Leave-one-pattern-out cross-validation reveals substantial variation in predictability: AUCROC ranges from 0.69 to 0.93, with patterns involving “must” operators being most predictable. We further show that targeted prompting (LogiCue), while reducing overall failures substantially, makes remaining failures structurally harder to predict. These findings suggest that input-side and output-side reliability mechanisms serve complementary roles in LLM reasoning pipelines.

ABSTRACT. A promising application for the potential revolu tionization offered by quantum computing is the field of machine learning through the application of quantum machine learning. Yet a critical chal lenge remains: no definitive evidence demonstrat ing a systematic performance advantage of quan tum machine learning models over their classical approaches. Researchers have attempted to address this uncertainty by employing focused benchmark ing experiments that explore particular model ar chitectures and particular computational tasks and datasets with different complexities. In this work, we introduce the first quantum implementation of the temporal convolutional network (QTCN), in which quantum operations are integrated into the dilated convolutional components that constitute the core mechanism of temporal convolutional net works (TCNs)for modeling temporal dependencies with a specific form of quantumencodingthat guar antees an effective and regular representation of se quential data. Experimental results shows that the QTCN outperforms classical models on complex, richly structured datasets like those from the mu sic domain. However, on simpler benchmarks such as the Adding Problem and Sequential MNIST, it shows performance on par with or below classical methods. Furthermore, by showing exactly how en tanglement improves the model’s performance in complex datasets, we provide a clear path for fu ture improvements.

ABSTRACT. Robots are increasingly employed in highly dynamic human environments such as hospitals, factories, or homes. These environments require a robot to be able to adapt and learn, which can be facilitated with imitation learning from human demonstration. Imitation learning from human demonstration often employs tools such as marker-based motion capture to facilitate learning from observation. The utilization of such tools restricts its application to controlled environments. Video-based motion capture presents an alternative, which decreases the recording effort. Our goal is to conduct a comparison of marker- and video-based modalities for motion capture in the context of imitation learning. The strengths and weaknesses of three state-of-the-art video-based 3D pose estimation models are documented and a conclusion on if and to what extent they can replace marker-based motion capture is drawn. We develop a novel evaluation framework that describes a video-based modality in comparison to the marker-based ground truth. It is separated into four categories, pose estimation quality, trajectory distance, behaviour separability, and grasp characterization. We conduct a study of a pick-and-place scenario with parallel recording of the marker- and video-based modalities. The results of the study show that pose estimation using a depth map as an external reference is a viable alternative to marker-based motion capture. The external depth reference gave it a substantial advantage over the models that utilized only on RGB data. But all modality only had limited success when characterizing the human grasp behaviour. We conclude that video-based pose estimation utilizing a depth map is a viable alternative to marker-based pose estimation since the actions and trajectories are well captured, and an imitation learning agent could therefore learn the shown behaviours, but high fidelity movement, such as grasping an object, is beyond the current capabilities of video-based pose estimation.

ABSTRACT. The evaluation of professional conduct requires the identification of applicable obligations when multiple provisions of the same professional code generate competing requirements. In prior work we defined a nine-component case representation for professional ethics analysis. This paper presents the ontological implementation of that representation, grounded in the Basic Formal Ontology (BFO) and classified through the Information Artifact Ontology (IAO), with type-level disjointness verified by the Pellet OWL-DL reasoner. Three object properties (\texttt{competesWith}, \texttt{prevailsOver}, \texttt{defeasibleUnder}) expose the competition structure between obligations as SPARQL-queryable relationships between typed individuals. OntServe, an ontology server, instantiates the representation as a knowledge graph of case ontologies extracted from a professional board of ethics archive.

ABSTRACT. Recent advances have introduced speech-to-speech (S2S) conversational assistants capable of producing natural-sounding interactions, including non-verbal cues like tonality and mood. In the automotive domain, this enables intuitive and humanlike in-car dialogue experiences. However, integrating these end-to-end assistants limits architectural options for programmable domain-specific safeguards. This paper discusses two implementation approaches for S2S guardrails: transcript-based and tool-based. Through an empirical evaluation, we demonstrate that both strategies are insufficient for industrial deployment due to prohibitive latency (delaying each answer by 0 to 1.4 seconds even for computationally cheap checks) and technical impediments (like potentially non-deterministic tool call behavior). Finally, we outline open challenges for S2S guardrails in the automotive context.

ABSTRACT. This paper introduces Warehouse Logistics 2D (WaLo2D), an open-source environment designed for testing multi-agent reinforcement learning algorithms in automated warehouse logistics. WaLo2D is a configurable, low-compute simulation that models simple 2D warehouse tasks. This enables researchers and developers to quickly evaluate cooperative strategies for package storage, pick-up, and redistribution centre management. While more realistic 3D simulations are essential for final deployment, they often slow down early policy training due to the time-consuming manual crafting of details and high hardware requirements. In contrast, overly simplified models, such as discrete grid-based navigation, may not generalise well. WaLo2D bridges this gap by providing an accessible testbed that accelerates the initial evaluation of algorithms and tuning of reward functions, which is a critical advantage given the high number of repetitions required in multi-agent reinforcement learning experiments. It supports logistics scenarios, such as package routing, storage management and truck loading. The project is available at https://github.com/KIForAN/WaLo2D.

ABSTRACT. Effective planning and scheduling of industrial processes are vital across numerous industries, including the daily operations of coal mining logistics. The SPP aims to address these tasks in stockyards efficiently and cost-effectively. This research examines a realistic scenario involving an opencast mine linked to a stockyard, where bulk materials of varying quality are stored on stockpiles and transported via trains. We approach the SPP in this setting using constraint programming and develop a model that accounts for the time-dependent nature of the problem: materials can only be stored on stockpiles for a limited duration before export, and transitions in machine operating states introduce blocking periods. The primary objective of the SPP in this scenario is to facilitate seamless operations of both opencast mining and stockyard activities. It is crucial to ensure continuous material flow over conveyor belts to prevent stoppages in mining activities while simultaneously meeting export requirements.

ABSTRACT. Multi-agent path finding (MAPF) is the problem of finding collision-free paths for a team of agents. Conflict-based search (CBS) is a state-of-the-art algorithm for solving MAPF optimally that repeatedly picks a collision (known as conflict) between two agents and resolves it by constraining the movement of the agents. One of the earliest runtime enhancement of CBS categorized conflicts into three groups (cardinal, semi-cardinal, and non-cardinal) and resolved them in that priority order. When multiple collisions share the same type, CBS typically breaks ties by selecting one at random. In this work, we show that CBS with disjoint splitting (which is another CBS' runtime enhancement) is further improved by introducing a new tie-breaking strategy when multiple cardinal (or semi-cardinal) conflicts are present. More specifically, we break ties in favor of conflicts that involve agents that participate in multiple cardinal (or semi-cardinal) conflicts. Our experiments show that the new tie-breaking strategy reduces both the number of node expansions and the runtime of CBS with disjoint splitting for many instances from the standard MAPF benchmark suite.

ABSTRACT. Insertion heuristics such as Nearest Insertion, Farthest Inser-tion, and Max-Regret are well-established constructive methods for the Traveling Salesperson Problem (TSP). However, they are typically applied as standalone algorithms and are not structurally embedded within exact optimization models. In this paper, we propose a novel approach that transfers the decision principles of insertion heuristics directly into a constraint programming (CP) model. Starting from the classical Miller-Tucker-Zemlin (MTZ) formulation, we extend the model by introducing additional positional variables, a dedicated value selection strategy, and a custom constraint equipped with a specialized propagator linking insertion positions to edge-selection variables. This integration enables the systematic emulation of insertion heuristics within the search process of the CP solver. We instantiate the framework using four heuristics: Nearest Insertion, Farthest Insertion, Farthest-Nearest Insertion, and Max-Regret Insertion. An experimental evaluation on 64 benchmark instances with up to 500 nodes demonstrates that all proposed variants signi?cantly outperform the classical MTZ model and a Savings-based adaptation. The results highlight the potential of structurally embedding heuristic knowledge into constraint models to improve robustness and scalability.

ABSTRACT. Deep neural networks (DNNs) have been demonstrated to enable autonomous decision-making across a range of domains, including finance, industrial control, robotics and medicine. Nevertheless, the opaque nature of these systems renders it challenging to verify behaviour, anticipate edge cases, or define operational boundaries. To address this issue, we propose the use of explainability-based training methods for reinforcement learning (RL) agents, with the aim of improving both model performance and robustness. Using concept bottleneck models (CBMs), sparse autoencoders (SAEs) and the Remove and Retrain (ROAR) approach during training has been shown to improve model performance. Random and guided occlusion methods have been evaluated using models that make decisions based on visual input. The results show that xAI-based training improves agent behaviour in benign environments and enhances robustness in scenarios where key information is missing. We can therefore demonstrate the potential of xAI-based training as a practical pathway towards more robust and reliable autonomous systems.

ABSTRACT. Large language models are increasingly used for knowledge graph question answering (KGQA), but can fail to correctly ground answers in the underlying graph. Current approaches to LLM-based KGQA either rely on full semantic parsing into executable queries such as SPARQL, which is brittle in practice due to complex schemas or incompleteness of real-world KGs, or on LLM-reasoning and answer generation over KGs, which can be more robust but lacks formal guarantees.

In this work, we study a complementary setting in which candidate answers are generated by an LLM-based system and subsequently verified using lightweight symbolic constraints derived from the question. We introduce Constrained Entity Selection under Partial Knowledge (CES-PK), a problem formulation that focuses on eliminating invalid answers and providing symbolic support for valid ones without requiring construction of executable logical forms. To account for incomplete KGs, we employ a three-valued constraint semantics (satisfied, violated, unknown) that avoids incorrect rejections under open-world assumptions.

To demonstrate the effects of our method, we instantiate this framework over the Hetionet biomedical knowledge graph and evaluate the impact of type, relation, and exclusion constraints. Experiments show that precision improves by filtering invalid candidates, while recall is preserved due to retaining candidates whose constraints are not explicitly violated. Satisfied constraints provide additional positive symbolic evidence to rank remaining candidates.

ABSTRACT. This article presents an experimental methodology for analyzing human–agent interaction from two complementary perspectives. Formal mechanism design is used to model interactions as dynamic human-in-the-loop mechanisms, while social mechanism theory from analytical sociology are employed to assess their causal, generative, normative, and collective properties. The article argues that evaluating AI systems requires moving beyond the Turing test, which focuses on human-like intelligence rather than social usefulness. As a core contribution, a Star–Durkheim-inspired test is proposed that assesses an AI system’s contribution to social mechanisms by examining collective cohesion and potential risks of social disintegration. By integrating economic mechanism design with sociological theory, the article outlines a foundation for social mechanism design and supports the development of conceptually grounded AI regulation.

ABSTRACT. LLMs deployed in high-stakes domains face fundamental reliability challenges: hallucinations, inconsistencies, and privacy vulnerabilities introduce unacceptable risks where errors carry legal, financial, or safety consequences. This paper presents a hybrid verification architecture combining formal symbolic methods with neural semantic analysis to provide complementary guarantees for LLM-generated content. This architecture employs logical reasoning for input verification, leveraging completeness properties to provide decidable guarantees on structured requirements. For output validation, embedding-based semantic similarity detects contextual hallucinations where formal methods lack expressiveness. This separation is realized in a parallel, actor-based pipeline, addressing limitations of prompt-based self-verification approaches, which inherit the distributional biases that produce hallucinations. The proposed architecture and type-aware verification method are validated with HAIMEDA, a real-world medical device damage assessment reporting system developed through Action Design Research. Evaluation shows hallucination detection rates of over 83% for structured entities and 72% for semantic fabrications, with a 30% reduction in report creation time, demonstrating that neuro-symbolic architectures can provide principled safeguards for LLM deployment in data-sensitive domains.

ABSTRACT. Efficiently finding safe and feasible trajectories for mobile objects is a critical field in robotics and computer science. In this paper, we propose SIL-RRT*, a novel learning-based motion planning algorithm that extends the RRT* algorithm by using a deep neural network to predict a sampling distribution at each iteration. We evaluated SIL-RRT* on various 2D and 3D environments and established that it can efficiently solve high-dimensional motion planning problems with fewer samples than traditional sampling-based algorithms. Moreover, SIL-RRT* is able to scale to more complex environments, making it a promising approach for solving challenging robotic motion planning problems.

ABSTRACT. Linear recurrent neural networks (LRNNs) provide a simple yet powerful framework for modeling time series. Unlike traditional recurrent neural networks, LRNNs use linear activation functions and can be trained without backpropagation by solving linear equation systems. This paper summarizes their theoretical properties, learning procedure, and ability to approximate arbitrary time-dependent functions. A key contribution is a method for reducing network size using eigenvalue analysis, enabling both efficient learning and architecture optimization, which is nowadays an important topic. We demonstrate the properties and advantages of LRNNs by several case studies, among them multiple superimposed oscillators (MSO), robotic soccer (RoboCup), and stock price prediction. LRNNs outperform the previous state-of-the-art for the MSO task with a minimal number of units.