Center for Artificial Intelligence (CAIRO)
Refine
Document Type
- Article (20)
- Conference Proceeding (20)
- Book (7)
- Preprint (2)
Is part of the Bibliography
- yes (49)
Keywords
- LLM (2)
- Ethical AI (1)
- European AI Act (1)
- Green AI (1)
- Human Oversight (1)
- Human Supervision (1)
- LLM for language reconstruction (1)
- LLM pretraining (1)
- Machine learning (1)
- Process optimisation (1)
Autonomous race cars, such as in Formula Student Driverless, operate close to their physical handling limits. The resulting highly nonlinear vehicle behavior increases the path tracking complexity, especially on narrow tracks. Model Predictive Control (MPC) is commonly used to address this issue, a method whose performance is closely tied to the accuracy of the underlying prediction model. This paper presents a novel, real-time capable prediction model for autonomous race cars that adjusts to changing conditions by combining information from past runs and the current driving situation. Our model is divided into three consecutive submodels: a nominal Kinematic Bicycle Model, an offline Bayesian Linear Regression (BLR) model, and an online Sparse Gaussian Process Regression (SGPR) model. The proposed approach enables efficient integration of all available data without significantly increasing computational cost, ensuring high prediction accuracy and a quantitative uncertainty assessment right from the start of the run. Compared to existing approaches, an improvement in prediction accuracy of up to 57 % was achieved. Further, we successfully demonstrated the practical applicability of the model within an MPC-based path tracking controller on a real Formula Student race car.
In autonomous racing, especially in competitions such as Formula Student Driverless, precise planning of the target velocity of a race car is crucial for competitive lap times and stable driving behavior. Especially at high speeds, Velocity Planning (VP) is a significant challenge as it has to be performed in real time, taking into account track layouts, environmental influences, mechanical tolerances, and the resulting control inaccuracies. In this paper, we present a novel approach to VP that dynamically adapts to such changing conditions. Instead of estimating the physical Tire-Road Friction Coefficient (TRFC), a continuous scaling factor is inferred indirectly from vehicle stability. This factor not only reflects the effective tire-road interaction but also captures effects of control inaccuracies. From this, we generate a continuous friction map, which serves as a robust, adaptive basis for computing the optimal target speed, accounting for both vehicle and environmental limits. Our proposed approach was evaluated on a real Formula Student race car, showing a lap time improvement of 35% over ten laps and an average increase of 8% compared to a non-adaptive approach.
The qualification of additive manufacturing processes is necessary to ensure print quality and material properties, but it is cost intensive. To predict suitable process parameters, this study compares classical statistical methods of experimental design with modern machine learning (ML) methods. The vat polymerisation process with a highly filled resin is investigated through combining data collection (response surface methodology, Sobol, particle swarm optimisation) and regression methods (polynomial regression, various ML algorithms). The results show that ML methods deliver more robust models on average than classical regression, although no universally optimal combination could be identified. Individual combinations such as particle swarm optimisation and support vector regression, CatBoost and random forest delivered the best results, while polynomial regression showed major shortcomings with the available data. For a higher tensile strength, high exposure intensity, short washing time and long tempering time proved to be decisive parameters.
Large Language Models (LLMs) often exhibit highly agreeable and reinforcing conversational styles, also known as AI-sycophancy. Although this pattern arises from training objectives that reward user satisfaction over accuracy, it may become problematic when interacting with user prompts that reflect negative social tendencies. Such responses risk amplifying harmful behavior rather than mitigating it. In this study, we examine how LLMs respond to user prompts expressing varying degrees of Dark Triad traits (Machiavellianism, Narcissism, and Psychopathy) using a curated dataset. Our analysis reveals differences across models, whereby all models predominantly exhibit corrective behavior, while showing reinforcing output in certain cases. Model behavior also depends on the severity level and differs in the sentiment of the response. Our findings raise implications for designing safer conversational systems that can detect and respond appropriately when users escalate from benign to harmful requests.
Human oversight, as one of several ethical principles required for the safe use of AI systems, plays a prominent role in the European Union’s Artificial Intelligence Act (AIA). This position paper argues that, while human oversight remains a necessary safeguard to mitigate the risks posed by AI systems aimed at directly interacting with humans across diverse domains, its current conceptualization rests on problematic assumptions about human judgment, epistemic access, and institutional responsibilities. Drawing on empirical findings from psychology and
human–AI interaction research, we show that human oversight, as presently envisaged, runs risk to remaining an illusion of control if the conditions for meaningful oversight are not made explicit. To address these limitations, we propose a conceptual reorientation of human oversight away from reactive approval and toward structurally embedded control, by suggesting incentive-based ethical stress testing, the establishment of normative clarity,
and the systematic accounting for the limits of human oversight. We conclude that only by grounding oversight in realistic assumptions about human cognition and institutional incentives can it fulfill its intended role in ensuring trustworthy AI.
This paper introduces a multi-turn large language model (LLM) jailbreaking attack called Defamiliarization, in which malicious queries are embedded within ostensibly harmless narratives. By reframing requests in “unmarked” contexts, LLMs can be coerced into producing undesirable outputs. A range of scenarios is documented, from planning ethically dubious actions to selectively overlooking critical events in literary texts, thereby exposing the limitations of alignment strategies predicated on detecting trigger words or semantic cues. Rather than substituting vocabulary, defamiliarization manipulates context and presentation, highlighting vulnerabilities that cannot be addressed by token-level fixes alone. Beyond demonstrating the effectiveness of defamiliarization as an attack strategy, evidence is presented of a systematic relationship between model scale and susceptibility. Experiments reveal that smaller-parameter models are significantly easier to manipulate using defamiliarized prompts. This finding raises important concerns regarding the growing popularity of lightweight, locally hosted LLMs, which are favored for their lower computational requirements but may lack alignment safeguards. A more holistic approach to LLM safety is advocated—one that incorporates insights from literary theory, ethics, and user experience—treating these models as interpretive agents. By doing so, defenses against covert manipulations can be strengthened and AI systems can remain aligned with human values.
Interpolation of Position Estimates for Radio Fingerprinting using Gaussian Process Regression
(2025)
Language models can greatly benefit from efficient tokenization. However, they still mostly utilize the classical Byte-Pair Encoding (BPE) algorithm, a simple and reliable method. BPE has been shown to cause such issues as under-trained tokens and sub-optimal compression that may affect the downstream performance. We introduce PickyBPE, a modified BPE algorithm that carries out vocabulary refinement during tokenizer training by removing merges that leave intermediate “junk” tokens. Our method improves vocabulary efficiency, eliminates under-trained tokens, and does not compromise text compression. Our experiments show that this method either improves downstream performance or does not harm it.
This article explores the zero-shot performance of state-ofthe-art large language models (LLMs) on one of the most challenging tasks in authorship analysis: sentence-level style change detection. Benchmarking four LLMs on the official PAN 2024 and 2025 “MultiAuthor Writing Style Analysis” datasets, we present several observations. First, state-of-the-art generative models are sensitive to variations in writing style—even at the granular level of individual sentences. Second, their accuracy establishes a challenging baseline for the task, outperforming suggested baselines of the PAN competition. Finally, we explore the influence of semantics on model predictions and present evidence suggesting that the latest generation of LLMs may be more sensitive to content-independent and purely stylistic signals than previously reported.
The environmental impact of Artificial Intelligence (AI)-enabled systems is increasing rapidly, and software engineering plays a critical role in developing sustainable solutions. The ''Greening AI with Software Engineering'' workshop,1 funded by the Centre Europ´een de Calcul Atomique et Mol´eculaire (CECAM) and the Lorentz Center, provided an interdisciplinary forum for 29 participants, from practitioners to academics, to share knowledge, ideas, practices, and current results dedicated to advancing green software and AI research. The workshop was held February 3-7, 2025, in Lausanne, Switzerland. Through keynotes, flash talks, and collaborative discussions, participants identified and prioritized key challenges for the field. These included energy assessment and standardization, benchmarking practices, sustainability-aware architectures, runtime adaptation, empirical methodologies, and education. This report presents a research agenda emerging from the workshop, outlining open research directions and practical recommendations to guide the development of environmentally sustainable AI-enabled systems rooted in software engineering principles.
Russenorsk, a pidgin language historically used in trade interactions between Russian and Norwegian speakers, represents a unique linguistic phenomenon. In this paper, we attempt to analyze its lexicon using modern large language models (LLMs), based on surviving literary sources. We construct a structured dictionary of the language, grouped by synonyms and word origins. Subsequently, we use this dictionary to formulate hypotheses about the core principles of word formation and grammatical structure in Russenorsk and show which hypotheses generated by large language models correspond to the hypotheses previously proposed ones in the academic literature. We also develop a “reconstruction” translation agent that generates hypothetical Russenorsk renderings of contemporary Russian and Norwegian texts.
This work explores transfer learning from several synthetic languages to English. We investigate the structure of the embeddings in the finetuned models, the information they contain, and the capabilities of the finetuned models on simple linguistic tasks. We also introduce a new synthetic language that leads to better transfer to English than the languages used in previous research. Finally, we introduce Tiny-Cloze Benchmark — a new synthetic benchmark for natural language understanding that is more informative for less powerful models. We use Tiny-Cloze Benchmark to evaluate fine-tuned models in several domains demonstrating that finetuning on a new synthetic language allows for better performance on a variety of tasks.
We propose a modeling approach for position estimation based on the observed radio propagation in an environment. The approach is purely similarity-based and therefore free of explicit physical assumptions. What distinguishes it from classical related methods are probabilistic position estimates. Instead of just providing a point estimate for a given signal sequence, our model returns the distribution of possible positions as continuous probability density function, which allows for appropriate integration into recursive state estimation systems. The estimation procedure starts by using a kernel to compare incoming data with reference recordings from known positions. Based on the obtained similarities, weights are assigned to the reference positions. An arbitrarily chosen density estimation method is then applied given this assignment. Thus, a continuous representation of the distribution of possible positions in the environment is provided. We apply the solution in a Particle Filter (PF) system for smartphone-based indoor localization. The approach is tested both with radio signal strength (RSS) measurements (Wi-Fi and Bluetooth Low Energy RSSI) and round-trip time (RTT) measurements, given by Wi-Fi Fine Timing Measurement. Compared to distance-based models, which are dedicated to the specific physical properties of each measurement type, our similarity-based model achieved overall higher accuracy at tracking pedestrians under realistic conditions. Since it does not explicitly consider the physics of radio propagation, the proposed model has also been shown to work flexibly with either RSS or RTT observations.
Diffusion models have emerged as powerful generative models, inspiring extensive research into their underlying mechanisms. One of the key questions in this area is the loss functions these models shall train with. Multiple formulations have been introduced in the literature over the past several years with some links and some critical differences stemming from various initial considerations. In this paper, we explore the different target objectives and corresponding loss functions in detail. We present a systematic overview of their relationships, unifying them under the framework of the variational lower bound objective. We complement this theoretical analysis with an empirical study providing insights into the conditions under which these objectives diverge in performance and the underlying factors contributing to such deviations. Additionally, we evaluate how the choice of objective impacts the model ability to achieve specific goals, such as generating high-quality samples or accurately estimating likelihoods. This study offers a unified understanding of loss functions in diffusion models, contributing to more efficient and goal-oriented model designs in future research.
Classifier guidance is intended to steer a diffusion process such that a given classifier reliably recognizes the generated data point as a certain class. However, most classifier guidance approaches are restricted to robust classifiers, which were specifically trained on the noise of the diffusion forward process. We extend classifier guidance to work with general, non-robust, classifiers that were trained without noise. We analyze the sensitivity of both non-robust and robust classifiers to noise of the diffusion process on the standard CelebA data set, the specialized SportBalls data set and the high-dimensional real-world CelebA-HQ data set. Our findings reveal that non-robust classifiers exhibit significant accuracy degradation under noisy conditions, leading to unstable guidance gradients. To mitigate these issues, we propose a method that utilizes one-step denoised image predictions and implements stabilization techniques inspired by stochastic optimization methods, such as exponential moving averages. Experimental results demonstrate that our approach improves the stability of classifier guidance while maintaining sample diversity and visual quality. This work contributes to advancing conditional sampling techniques in generative models, enabling a broader range of classifiers to be used as guidance classifiers
