HES-SO Valais-Wallis

The rapid expansion of crypto-asset markets and the introduction of the Markets in Crypto-Assets Regulation (MiCAR) pose novel supervisory challenges. Existing blockchain intelligence platforms focus predominantly on on-chain surveillance, leaving gaps in off-chain documentary due diligence automation. This paper presents a Multi-Agent System (MAS) integrating Large Language Model (LLM) capabilities with rule-based compliance frameworks. The architecture comprises seven specialized agents: a Coordinator Agent for orchestration; data acquisition agents (Searcher, Crawler); three parallel analytical agents—Heuristic Agent (LLM-powered qualitative risk assessment), Compliance Agent (hybrid-AI MiCAR asset classification and regulatory requirement verification), and On-Chain Agent (machine learning-based fraud detection); and a Reconciliator Agent synthesizing findings into unified alerts. Component-level empirical validation on 150 projects indicates 95% output reproducibility (identical alert tier and score deviation ≤0.05 across five reruns) and 210 s mean latency, providing proof-of-concept evidence for the integrated pipeline. A pilot user evaluation (six researchers/master students and two experts from regulatory authorities) provides preliminary usability evidence and surfaces domain-specific feedback from regulatory-authority experts. The architecture advances proactive regulatory technology by enabling scalable analysis combining off-chain documentary evidence with on-chain forensics.

This paper explores the organized relationships of parts and wholes through Harvey Sacks’s work, taking it as an inspiration for empirical studies. This topic is often conceived in terms of gestalts, where each part appears through its functional significations. I propose a complementary examination of parts and wholes as members’ phenomena. In an early lecture, Sacks [1992: 89] notes that “we need ... a notion that what Members see is decomposable by them”, and that they “seem to do this [by] treating something that they see as a combination of parts, some of which have names”. To develop this direction of inquiry, I offer the notion of structured objects: entities with recognizable features such as “left” and “right” sides (e.g., photographs and faces) or “beginnings” and “ends” (e.g., stories and streets). Two illustrations – covering directly and retrospectively available structuredness – elucidate how structured objects are constituted and maintained. Importantly, such a decomposition is not imposed by professional analysts. The structuredness is rather established in practical manipulations with the relevant “parts”, the relevant structures of social objects being produced or discovered in and as these operations.

BackgroundComputed tomography (CT) is widely used in clinical practice due to its ability to provide detailed anatomical information. However, variations in radiation dose can affect image quality, potentially compromising the performance and reliability of artificial intelligence (AI) models applied to these images.PurposeTo evaluate the robustness of radiomics‐based and deep learning‐based models to variations in CT dose levels using a standardized dataset obtained from a 3D‐printed anthropomorphic phantom simulating liver tissue with anomalies, as well as in the publicly available dataset CT‐ORG with real patient data for organ classification. This study is in an early experimental stage, tested only on retrospective data.MethodsA total of 1378 image series from 649 scans were acquired across 13 scanners from four manufacturers at five dose levels. Features were extracted from six regions of interest (ROIs), representing four liver tissue types (normal, cyst, hemangioma, metastasis), using four methods: PyRadiomics, a shallow convolutional neural network (CNN), SwinUNETR, and a CT foundation model (CT‐FM). Feature stability was assessed using the Intraclass Correlation Coefficient (ICC), while Uniform Manifold Approximation and Projection (UMAP) was employed to evaluate tissue types separability and the influence of scanner variations. Generalizability was tested by training liver tissue classifiers on one dose level and testing on others, alongside a dose classification task (10‐fold cross‐validation) to determine the sensitivity of each method to dose variations. In addition, we compared the four methods in addressing the task of organ classification (10‐fold cross‐validation) with the CT‐ORG dataset containing 140 CT scans acquired with varying dose levels.ResultsRadiomic features showed limited robustness to dose variations, leading to reduced performance in liver tissue classification and the lowest ICC among methods (ICC: 0.8355 ± 0.1705). SwinUNETR and CT‐FM exhibited the highest stability (SwinUNETR ICC: 0.9528 ± 0.0272; CT‐FM ICC: 0.9347 ± 0.0420), clearly above the Shallow CNN (ICC: 0.8416 ± 0.2018). CT‐FM also showed strong generalization across dose levels: its features effectively distinguished between liver tissue types and dose levels simultaneously, without compromising performance in either task. Consistent with these trends in dose sensitivity, CT‐FM obtained the highest dose‐classification accuracy (0.6517 ± 0.0179), whereas SwinUNETR showed the lowest (0.3796 ± 0.0250). These trends were confirmed in the context of organ classification with real patient data on the CT‐ORG dataset, where CT‐FM achieved the highest accuracy (0.965).ConclusionsThe study highlights the limited robustness of traditional radiomics and deep models to CT dose variation and underscores the potential of foundation models like CT‐FM to enable robust clinical applications by mitigating dose‐related variability. This enhanced performance is likely due to the model's pretraining on large and diverse datasets, allowing it to learn robust and generalizable representations across varying acquisition conditions.

In recent years, explainable methods for artificial intelligence (XAI) have tried to reveal and describe models' decision mechanisms in the case of classification and even for segmentation. However, XAI methods for semantic segmentation and in particular for single specific instances (e.g. one given lesion among others of the same class in medical imaging) have yet to be developed to understand what drove the detection and contouring of the latter, which is crucial for all multi-lesional diseases. We proposed instance-level explanation maps for semantic segmentation extending both SmoothGrad and Grad-CAM++ methods and yielding quantitative instance saliency for the former. The instance-level methods were applied to the segmentation of white matter lesions (WML), a magnetic resonance imaging (MRI) biomarker in multiple sclerosis (MS). 687 patients diagnosed with MS for a total of 4023 FLAIR and MPRAGE MRI scans were collected at the University Hospital of Basel, Switzerland. WM lesion masks were annotated by four expert clinicians on baseline and follow-up imaging. Three deep learning networks-a 3D U-Net, nnU-Net, and Swin UNETR-were trained and tested on these data (test normalized Dice score, respectively of 0.71, 0.78, 0.80; true positive rate of 79%, 78%, and 85%; false discovery rate of 37%, 38%, and 36%; false negative rate of 20%, 22%, and 14%), then saliency maps were computed. Consistent with clinical practice, the proposed instance saliency maps revealed that the model relied more on FLAIR than MPRAGE to segment WMLs, with positive saliency values inside a lesion and negative in its neighborhood. FLAIR hyperintensity combined with healthy WM around the lesion border was required for their detection. Beyond the aforementioned sanity checks, we observed that peak values of the generated saliency maps presented distributions that differ significantly between TP, FN, FP and TN predictions, suggesting that the quantitative nature of the proposed saliency could be used to identify errors. In conclusion, we introduced two XAI methods to generate quantitative instance-level explanations in semantic segmentation. The proposed XAI maps can be applied to any architecture and could serve as a basis to (i) improve model performance (e.g. reducing FPs), (ii) optimize their internal architecture (e.g. patch size), and (iii) justify the model's decisions to the end users, which are contextualized to a specific lesion instance of interest.

We applied flow cytometry (FCM) to monitor microbial dynamics during grape must fermentation at the winery scale. Experiments were performed on Pinot Noir grapes using three distinct winemaking protocols: inoculation with active dried yeast, Pied-de-Cuve, and spontaneous fermentation. FCM enabled the assessment of yeast viability and metabolic activity, as well as the detection and monitoring of viable bacterial populations during alcoholic fermentation. Amplicon-based DNA sequencing was performed to characterize the associated microbial communities and evaluate protocol-specific effects. Trends identified by amplicon sequencing were partially mirrored by patterns observed in unsupervised FCM analysis. Overall, our results indicate that FCM is a practical tool for monitoring microbial dynamics during fermentation, providing near-real-time information that can support monitoring strategies and risk management in winemaking.

Balance is referred to as a state of equilibrium where forces of agonist and antagonistic muscles are equal. This is particularly relevant in the practice of Goju-Ryu Karate-Do, a martial art style with combinations of hard and soft techniques. Particularly, karate stances not only depend on technical aspects but also on the ability to achieve a centered posture. In this narrative review, we aim to integrate the existing knowledge about alignment parameters of the spine to various stances in Goju-Ryu Karate-Do to improve biomechanical understanding, allow technical modifications, and consequently enhance positive training experience. Balance is constantly challenged during the interplay of accelerated movements and subsequent controlled postures (Japanese: “Kamae”). This requires continuous neuromuscular coordination to maintain the body’s equilibrium. In particular, the body’s center of gravity, which is located around the second sacral vertebra in a standing position, needs to be kept within Dubousset’s “efficiency cone” for low energy consumption and minimal fatigue. This state is primarily maintained by aligning the spine, the pelvis, and the lower extremities, which is a result of complex biomechanical interactions of various spinopelvic parameters. Applying these concepts of Dubousset to stances in Goju-Ryu Karate-Do helps to explain why during the aging process or natural degeneration, technical modifications are needed to guarantee an optimal training experience in senior or disabled practitioners of Karate-Do. Biomechanical parameters of the spinopelvic axis are crucial in mastering the art of Goju-Ryu Karate-Do. Only with a balanced stance and an optimally situated center of gravity, a block or attack may be successful and movement strategies effective. However, technical modifications of stances must be considered in aged or disabled karateka to allow a sustained training experience.

Generally, formamidinium (FA)-based halide perovskite thin films are fabricated using halide precursors, such as PbI2 and FAI, but this approach often leads to stoichiometric distortions, resulting in perovskite films with structural defects and reduced crystallinities. These problems can adversely influence power conversion efficiencies and open-circuit potentials of perovskite solar cells. In this study, we propose a microcrystalline perovskite powder (MCP) synthesized by controlling the stoichiometry of the FAI precursor. We optimize the synthesis of α-FAPbI3 powder using 1.1 equiv of FAI. Interestingly, the synthesized black powder produced an excellent crystallinity and phase stability for up to six months. Remarkably, the MCP forms large colloids in solutions that are stably cohesive, promoting spontaneous nucleation and enabling the fabrication of low-defect thin films. Consequently, perovskite solar cells fabricated using the MCP display significantly improved efficiencies of 23.12% compared to those (19.64%) of the cells fabricated using the conventional PbI2 and FAI precursors. This approach highlights the potential of MCPs for use in enhancing the performances and stabilities of perovskite-based devices.

BackgroundNon-invasive spectroscopic methods are increasingly valued in life sciences, where preserving the native state of biomolecules is essential for accurate interpretation. Traditional analyses of microbial compounds typically involve solvent-based extraction and chromatographic separation processes, which are time consuming, damaging to samples, and can alter biomolecular structures of complexes. To overcome these limitations, we developed a novel spectroscopic workflow for direct metabolite monitoring in microbial cells.ResultsIn this study, we established a combined spectroscopic methodology that allows direct pigment and polyhydroxyalkanoates (PHAs) analysis in complex biological samples without requiring chemical extraction procedures. The UV-Vis spectroscopy technique using an integrating sphere enables direct monitoring of pigments even in turbid whole cell suspensions, providing detailed fingerprints of bacteriochlorophyll a and carotenoids in their natural environment. Together, these techniques provide consistent information about cellular composition. Using the photosynthetic bacterium Rhodospirillum rubrum as a model organism, we demonstrate that our combined spectroscopic approach can resolve pigment states, reveal intracellular PHA content and crystallinity, and measure carotenoids and bacteriochlorophylls directly in native whole cell suspensions. Furthermore, advanced data processing provided an improved interpretation of pigment and PHA states in different cellular forms.ConclusionsThis innovative combination of spectroscopic techniques reduces sample manipulation, preserves cellular integrity and provides rapid, precise, and environmentally friendly analysis of microbial metabolites in their natural physiological conditions. The demonstrated workflow is broadly applicable to biological samples where maintaining biomolecular integrity is crucial, and it has strong potential for applications in process analytical technology and industrial biotechnology.Graphical

This paper investigates the potential of storing photovoltaic (PV) electricity in electric vehicle (EV) batteries and reinjecting the stored electricity into the grid (V2G) during high electricity demand periods. We developed a geographical methodology to quantify the mobility and EV charging needs across Europe. Scenarios of charging behaviour are then used to distribute the charging needs among locations corresponding to residential areas, workplaces, or points of interest (e.g., shops, restaurants, etc.). Resulting load curves and maps of the charging demand, with a resolution of one hectare, are then compared to a potential PV production obtained from typical meteorological data of irradiance, solar panel performance, and building coverage. This comparison is conducted hourly over one year, monitoring the evolution of the daily storage capacity of the EV fleet. Case studies were conducted on three cities in Switzerland: Sion, Neuchâtel, and Zürich. Three scenarios are used to compare home charging behaviour, balanced home and work charging behaviour, and charging at points of interest. Results highlight the impact of charging behaviour on the flexibility potential, with higher potential when charging at work and point of interest is privileged. The methodology is implemented in the online platformcitiwatts[1].