Uncertainty estimation is critical for the safe clinical deployment of deep learning in medical image segmentation, with aleatoric uncertainty theoretically designed to capture irreducible data ambiguity. However, whether entropy-based measures reflect clinically meaningful ambiguity, i.e. case-level disagreement about whether a pathology is present at all, remains poorly understood. Contrary to most prior work, which focused on pixel-wise boundary disagreement, we systematically evaluate how well aleatoric uncertainty captures presence ambiguity. Our evaluation spans 3D lung nodule segmentation across four architectures with Monte Carlo dropout and deep ensembles, on LIDC-IDRI and an external validation cohort (LNDb). We find that entropy-based uncertainty maps align with boundary noise and minor drawing variation but carry insufficient discriminative signal for presence ambiguity. In contrast, a lightweight supervised ambiguity head trained on frozen segmentation features substantially outperforms all entropy-aggregation-based baselines across architectures, metrics, and both cohorts, and matches or exceeds methods that explicitly model ambiguity under disagreement supervision (Probabilistic U-Net, Annotator-Confusion 3D-UNet). A qualitative feature-space analysis shows that presence ambiguity is already encoded in the frozen encoder features of pixel-wise-trained networks, only to be discarded by the segmentation output and its entropy aggregation. Our findings expose a fundamental mismatch between the theoretical promise of aleatoric uncertainty and its practical behavior, and suggest that practitioners should not rely on entropy-based uncertainty as a proxy for clinical ambiguity in safety-critical applications.

The increasing frequency and intensity of extreme weather events driven by climate change lead to stronger impacts and fatalities, affecting more communities every year. Early warning systems (EWS) are essential adaptation tools designed to provide timely and actionable warnings of natural hazards such as heatwaves, floods, droughts and wildfires. The rapid development of artificial intelligence (AI), including machine learning (ML) and sophisticated deep learning models, together with the availability of large Earth observation data sets, offers a unique opportunity to improve the current capabilities of EWS. This review provides a comprehensive overview of existing AI-based methods applied to some of the most devastating natural hazards worldwide. We identify key gaps in the current landscape, not only geographically, but also in the application to specific natural hazards. While ML has advanced significantly in applications for flood and wildfire detection, its use for drought and heatwave prediction remains comparatively limited and underexplored. Differences among the manifestations of different natural hazards, how we can observe them (and the types of data) and how they are traditionally forecast or monitored, are limiting factors for the integration of AI-based solutions within operational EWS, particularly in regions and for hazard types that are most in need of robust early warning capabilities. Nevertheless, the tremendous effort in improving data-driven models for weather forecasts as well as the development of AI techniques for capturing complex environmental dynamics, creates a promising pathway to strengthening EWS. By addressing current gaps, such as regional coverage, data availability and hazard representation, AI can significantly enhance the precision and timeliness of EWS, ultimately contributing to more effective, inclusive and proactive disaster risk management.

Background/Objectives: Peptide receptor radionuclide therapy (PRRT) is an established treatment for metastatic neuroendocrine tumors (NETs), yet long-term disease control occurs only in a subset of patients. Predicting progression-free survival (PFS) could support individualized treatment planning. This study evaluates laboratory, imaging, and multimodal deep learning models for PFS prediction in PRRT-treated patients. Methods: In this retrospective, single-center study 116 patients with metastatic NETs undergoing [177Lu]Lu-DOTATOC were included. Clinical characteristics, laboratory values, and pretherapeutic somatostatin receptor positron emission tomography/computed tomographies (SR-PET/CTs) were collected. Seven models were trained to classify low- vs. high-PFS groups, including unimodal (laboratory, SR-PET, or CT) and multimodal fusion approaches. Performance was assessed via repeated 3-fold cross-validation with area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC). Explainability was evaluated by feature importance analysis and gradient based saliency maps. Results: Forty-two patients (36%) displayed short PFS (≤1 year) and 74 patients displayed long PFS (>1 year). Groups were similar in most characteristics, except for higher baseline chromogranin A (p = 0.003), elevated γ-GT (p = 0.002), and fewer PRRT cycles (p < 0.001) in short-PFS patients. The Random Forest model trained only on laboratory biomarkers reached an AUROC of 0.59 ± 0.02. Unimodal three-dimensional convolutional neural networks using SR-PET or CT performed worse (AUROC 0.42 ± 0.03 and 0.54 ± 0.01, respectively). A multimodal fusion model integrating laboratory values, SR-PET, and CT-augmented with a pretrained CT branch-achieved the best results (AUROC 0.72 ± 0.01, AUPRC 0.80 ± 0.01). Explainability analyses provided insights into model predictions, with explainability patterns in the fusion model appearing physiologically plausible and predominantly tumor-focused. Conclusions: Multimodal deep learning combining SR-PET, CT, and laboratory biomarkers outperformed unimodal approaches for PFS prediction after PRRT. Upon external validation, such models may support risk-adapted follow-up strategies.

Abstract Renewable energy generation is highly sensitive to both local and large‐scale atmospheric dynamics. Skillful subseasonal to seasonal (S2S) forecasts—ranging from 2 weeks to two months—can provide substantial socioeconomic benefits for the energy sector. In this work, we introduce DiffScale, a diffusion‐based generative model that super‐resolves surface wind speed forecasts by flexibly downscaling coarse‐resolution S2S predictions. DiffScale leverages conditional guidance from low‐resolution weather forecasts to sample high‐resolution realizations, enabling the direct estimation of the conditional density over target wind fields without relying on auto‐regressive or sequence‐based methods. This formulation yields a computationally efficient and scalable framework for continuous downscaling across arbitrary spatial resolutions and forecast lead times. We evaluate DiffScale on synthetic experiments that involve downscaling ECMWF S2S forecasts to the higher‐resolution ERA5 reanalysis data. The key innovation lies in its generalization ability: DiffScale adapts to varying downscaling factors and temporal horizons without retraining, while effectively correcting biases in the original forecasts. Empirical results demonstrate that DiffScale consistently improves both deterministic and probabilistic forecast skill over ECMWF baselines and their quantile‐mapped variants. Moreover, it achieves competitive performance compared to climatology up to lead time week 6, underscoring its potential for enhancing real‐time wind speed forecasts in energy applications.

The “black box” characteristic of AI models has received widespread attention from scientists. This has promoted the development of explainable artificial intelligence (XAI) methods aimed at explaining AI model decisions, e.g., evaluating the importance of input features for the model output. In addition to XAI, another way to open the “black box” is uncertainty estimation of AI models (e.g., output ranges for regression tasks), which provides users with confidence in model decisions. However, little attention has been paid to the explanation of model uncertainty, e.g., which input features contribute to model uncertainty? This study takes SHapley Additive exPlanations (SHAP) as an example of perturbation-based XAI methods and utilizes it to explain AI model uncertainty by changing the model input to observe the variation of the estimated uncertainty. The experimental results in three cases (i.e., housing price prediction, remote sensing image classification, and electrocardiogram classification) show that SHAP can effectively explain the uncertainty of AI models. Moreover, further analysis reveals that the uncertainty of AI models is affected by the frequency of occurrence of input features in the training data. Specifically, if a certain feature of the sample rarely appears in the training data, it will lead to greater uncertainty. This conclusion aligns with common sense and supports the validity of the uncertainty explanation.

Machine learning in high-stakes domains such as healthcare requires not only strong predictive performance but also reliable uncertainty quantification (UQ) to support human oversight. Multi-label text classification (MLTC) is a central task in this domain, yet remains challenging due to label imbalances, dependencies, and combinatorial complexity. Existing MLTC benchmarks are increasingly saturated and may be affected by training data contamination, making it difficult to distinguish genuine reasoning capabilities from memorization. We introduce MADE, a living MLTC benchmark derived from medical device adverse event reports and continuously updated with newly published reports to prevent contamination. MADE features a long-tailed distribution of hierarchical labels and enables reproducible evaluation with strict temporal splits. We establish baselines across more than 20 encoder- and decoder-only models under fine-tuning and few-shot settings (instruction-tuned/reasoning variants, local/API-accessible). We systematically assess entropy-/consistency-based and self-verbalized UQ methods. Results show clear trade-offs: smaller discriminatively fine-tuned decoders achieve the strongest head-to-tail accuracy while maintaining competitive UQ; generative fine-tuning delivers the most reliable UQ; large reasoning models improve performance on rare labels yet exhibit surprisingly weak UQ; and self-verbalized confidence is not a reliable proxy for uncertainty. Our work is publicly available at https://hhi.fraunhofer.de/aml-demonstrator/made-benchmark.

Abstract AI documentation frameworks remain poorly designed for point-of-care use, leaving clinicians without actionable information on how to use clinical AI models when they need it most. We developed the Clinician Model Card, an interactive, clinician-centered documentation tool, and evaluated it in a sequential exploratory mixed-methods study: interviews with 12 physicians informed iterative co-design, evaluated in a national survey of 129 physicians across Germany. The tool was well-received: 84% agreed it should be routinely available, and 66% considered its content relevant to clinical decision-making. Yet comprehensibility of statistical performance metrics remained poor despite targeted interventions: only 32% understood the Validation & Performance section well, and fewer than 54% correctly interpreted AUROC or PPV, with AI literacy as strong predictor of comprehension (ρ = 0.59). We propose empirically derived design principles for clinician-centered AI documentation. Effective AI transparency requires not only clinician-friendly design and workflow integration, but sustained investment in AI literacy.

The explainable AI (XAI) research community has proposed numerous technical methods, yet deploying explainability as systems remains challenging: Interactive explanation systems require both suitable algorithms and system capabilities that maintain explanation usability across repeated queries, evolving models and data, and governance constraints. We argue that operationalizing XAI requires treating explainability as an information systems problem where user interaction demands induce specific system requirements. We introduce X-SYS, a reference architecture for interactive explanation systems, that guides (X)AI researchers, developers and practitioners in connecting interactive explanation user interfaces (XUI) with system capabilities. X-SYS organizes around four quality attributes named STAR (scalability, traceability, responsiveness, and adaptability), and specifies a five-component decomposition (XUI Services, Explanation Services, Model Services, Data Services, Orchestration and Governance). It maps interaction patterns to system capabilities to decouple user interface evolution from backend computation. We implement X-SYS through SemanticLens, a system for semantic search and activation steering in vision-language models. SemanticLens demonstrates how contract-based service boundaries enable independent evolution, offline/online separation ensures responsiveness, and persistent state management supports traceability. Together, this work provides a reusable blueprint and concrete instantiation for interactive explanation systems supporting end-to-end design under operational constraints.

In the evaluation of attribution quality, the quantitative assessment of explanation legibility is particularly difficult, as it is influenced by varying shapes and internal organization of attributions not captured by simple statistics. To address this issue, we introduce Minimum Spanning Tree Compactness (MST-C), a graph-based structural metric that captures higher-order geometric properties of attributions, such as spread and cohesion. These components are combined into a single score that evaluates compactness, favoring attributions with salient points spread across a small area and spatially organized into few but cohesive clusters. We show that MST-C reliably distinguishes between explanation methods, exposes fundamental structural differences between models, and provides a robust, self-contained diagnostic for explanation compactness that complements existing notions of attribution complexity.

​AI-based systems that use Earth observation (EO) and satellite communication (SatCom) services can support disaster risk management. However, successful deployment in the field requires ongoing stakeholder involvement throughout system development. Human-Centered AI (HCAI) provides a suitable framework for this by explicitly addressing human needs and requirements. In this work, we present an HCAI-inspired approach that connects stakeholder participation to the design of a demonstrator. We conducted participatory online workshops with stakeholders to understand their roles, goals, tasks, and requirements. These insights guided the development of an AI-supported prototype integrated into a multi-domain marketplace demonstrator that combines EO and SatCom services. The first prototype iteration was evaluated in two use cases: burned area detection and flood modeling, indicating differing access to EO and SatCom services across stakeholder groups but broadly similar goals and strong interest in integrated, multi-modal services, alongside concerns about real-time data and user training.

Neural Controlled Differential Equations (NCDE) provide a powerful continuous-time framework for forecasting time series, but standard graph-based extensions typically learn spatial structure purely from data, even in settings where a directed graph structure is known a priori. We introduce Informed Neural controlled Differential EQuationS (INDEQS), a graph-based NCDE forecasting method that incorporates prior knowledge of a directed graph at distinct architectural positions. INDEQS separates inner mixing of hidden states across graph nodes from outer mixing between vector field and control, and offers both a lightweight graph-constrained variant and a more expressive variant, learning additional graph connections from data via adaptive graph convolutions. To systematically study when graph informedness is beneficial in forecasting, we devise a continuous advection simulation on directed graphs, yielding synthetic spatio-temporal datasets with known ground-truth flow structure. We then evaluate INDEQS on two real-world tasks: river discharge forecasting on a hydrological network and traffic flow prediction on PeMS08. Across these synthetic and real-world benchmarks, outer informedness consistently improves mean absolute error over an uninformed NCDE with comparable parameter count, particularly on larger graphs, while inner informedness offers a more parameter-efficient alternative when strict adherence to a known adjacency is desired. A comparison of discrete convolutional and continuous-time decoders further shows that continuous decoders yield better accuracy and greater temporal flexibility on real-world tasks. An implementation of INDEQS and the advection simulation is available at https://github.com/Mitchi1/indeqs.

End-to-end weather forecasting systems produce skillful global gridded and station forecasts directly from raw Earth observations, replacing the numerical weather prediction pipeline, including data assimilation, at a fraction of its cost. These systems are deterministic and issue no uncertainty. Here we render the Aardvark Weather model probabilistic by attaching one stochastic mechanism to each component: learned, input-dependent noise at the observation encoder, capturing aleatoric uncertainty inherited from the observing system, and Monte Carlo dropout in the processor, capturing epistemic uncertainty in the learned dynamics. The resulting nested ensemble attributes forecast spread to the two sources through a law-of-total-variance decomposition, cross-checked by withholding observation streams. Probabilistic finetuning significantly improves the mean forecast, by 4.2

The use of artificial intelligence (AI) in medicine has great potential to improve quality and efficiency. At the same time, its use is associated with risks. In order to prevent harm, experts from research and politics are developing requirements, testing tools, and guidelines to identify and control these risks.This article first describes the requirements for AI in medicine. It then follows with a presentation of how concrete quality criteria and testing tools can be derived from those requirements. A deeper demonstration of "explainability" as a key criterion is presented as an example. A testing framework for the use of AI in medicine is also presented. Finally, the article provides an outlook on standardization of quality criteria.The development of AI, especially in medicine, is progressing quickly. Suitable testing procedures can help to minimize risks and strengthen trust in the technology. However, for the successful implementation of such testing procedures, continuous research efforts are necessary.

Renewable resources are strongly dependent on local and large-scale weather situations. Skillful subseasonal to seasonal (S2S) forecasts -beyond two weeks and up to two months- can offer significant socioeconomic advantages to the energy sector. In particular, accurate wind speed forecasts result in optimized generation of wind-based electric power. This study aims to enhance wind speed predictions using a diffusion model with classifier-free guidance to downscale S2S forecasts of surface wind speed. We propose DiffScale, a diffusion model that super-resolves spatial information for continuous downscaling factors and lead times. Leveraging weather priors as guidance for the generative process of diffusion models, we adopt the perspective of conditional probabilities on sampling super-resolved S2S forecasts. We aim to directly estimate the density, associated with the target S2S forecasts at different spatial resolutions and lead times without auto-regression or sequence prediction, resulting in an efficient and flexible model. Synthetic experiments were designed to super-resolve wind speed S2S forecasts from the European Center for Medium-Range Weather Forecast (ECMWF) from a coarse resolution to a finer resolution of data from ERA5, which serves as a high-resolution target, derived from reanalysis data. We achieve a significant increase in the quality of predictions, utilizing the proposed diffusion model for continuous downscaling and bias correction of the ECMWF forecasts.

Abstract Accurate forecasts on subseasonal (S2S) timescales are essential for the preparation and mitigation of the impacts of high‐impact events, such as flash droughts. To improve the accuracy of soil moisture forecasts—a critical factor in identifying flash droughts—we present a hybrid modeling framework that combines dynamical forecasts from the European Centre for Medium‐Range Weather Forecasts with deep learning (DL) models. This approach not only corrects biases in numerical weather prediction models but also improves spatial resolution, increasing the accuracy of S2S forecasts. By using deterministic inputs, such as the ensemble mean and spread, we further assess the uncertainty of forecasts through dropout neural networks via Monte Carlo sampling. Our results demonstrate that the DL models outperform baseline methods, offering skillful S2S forecasts of soil moisture. This advanced hybrid framework provides more accurate soil moisture predictions, ultimately supporting improved strategies for managing and mitigating the impacts of flash droughts.

Uptake of explainable artificial intelligence (XAI) methods in geoscience is currently limited. We argue that such methods that reveal the decision processes of AI models can foster trust in their results and facilitate the broader adoption of AI.

Deploying deep learning models in clinical practice often requires leveraging multiple data modalities, such as images, text, and structured data, to achieve robust and trustworthy decisions. However, not all modalities are always available at inference time. In this work, we propose multimodal privileged knowledge distillation (MMPKD), a training strategy that utilizes additional modalities available solely during training to guide a unimodal vision model. Specifically, we used a text-based teacher model for chest radiographs (MIMIC-CXR) and a tabular metadata-based teacher model for mammography (CBIS-DDSM) to distill knowledge into a vision transformer student model. We show that MMPKD can improve the resulting attention maps' zero-shot capabilities of localizing ROI in input images, while this effect does not generalize across domains, as contrarily suggested by prior research.

Methods from artificial intelligence (AI) and, in particular, machine learning and deep learning, have advanced rapidly in recent years and have been applied to multiple fields including geospatial analysis. Due to the spatial heterogeneity and the fact that conventional methods can not mine large data, geospatial studies typically model homogeneous regions locally within the entire study area. However, AI models can process large amounts of data, and, theoretically, the more diverse the train data, the more robust a well-trained model will be. In this paper, we study a typical machine learning method XGBoost, with the question: Is it better to build a single global or multiple local models for XGBoost in geospatial studies? To compare the global and local modeling, XGBoost is first studied on simulated data and then also studied to forecast daily infection cases of COVID-19 in Germany. The results indicate that if the data under different relationships between independent and dependent variables are balanced and the corresponding value ranges are similar, i.e., low spatial variation, global modeling of XGBoost is better for most cases; otherwise, local modeling of XGBoost is more stable and better, especially for the secondary data. Besides, local modeling has the potential of using parallel computing because each sub-model is trained independently, but the spatial partition of local modeling requires extra attention and can affect results.

Renewable resources are strongly dependent on local and large-scale weather situations. Skillful subseasonal to seasonal (S2S) forecasts – beyond two weeks and up to two months – can offer significant socioeconomic advantages to the energy sector. This study aims to enhance wind speed predictions using a diffusion model with classifier-free guidance to downscale S2S forecasts of surface wind speed. We propose DiffScale, a diffusion model that super-resolves spatial information for continuous downscaling factors and lead times. Leveraging weather priors as guidance for the generative process of diffusion models, we adopt the perspective of conditional probabilities on sampling super-resolved S2S forecasts. We aim to directly estimate the density associated with the target S2S forecasts at different spatial resolutions and lead times without auto-regression or sequence prediction, resulting in an efficient and flexible model. Synthetic experiments were designed to super-resolve wind speed S2S forecasts from the European Center for Medium-Range Weather Forecast (ECMWF) from a coarse resolution to a finer resolution of ERA5 reanalysis data, which serves as a high-resolution target. The innovative aspect of DiffScale lies in its flexibility to downscale arbitrary scaling factors, enabling it to generalize across various grid resolutions and lead times -without retraining the model- while correcting model errors, making it a versatile tool for improving S2S wind speed forecasts. We achieve a significant improvement in prediction quality, outperforming baselines up to week 3.

Accurate prediction of extreme weather events remains a major challenge for artificial intelligence based weather prediction systems. While deterministic models such as FuXi, GraphCast, and SFNO have achieved competitive forecast skill relative to numerical weather prediction, their ability to represent uncertainty and capture extremes is still limited. This study investigates how state of the art deterministic artificial intelligence based models respond to initial-condition perturbations and evaluates the resulting ensembles in forecasting extremes. Using three perturbation strategies (Gaussian noise, Hemispheric Centered Bred Vectors, and Huge Ensembles), we generate 50 member ensembles for two major events in August 2022: the Pakistan floods and the China heatwave. Ensemble skill is assessed against ERA5 and compared with IFS ENS and the probabilistic AIFSENS model using deterministic and probabilistic metrics. Results show that flow dependent perturbations produce the most realistic ensemble spread and highest probabilistic skill, narrowing but not closing the performance gap with numerical weather prediction ensembles. Across variables, artificial intelligence based weather models capture temperature extremes more effectively than precipitation. These findings demonstrate that input perturbations can extend deterministic models toward probabilistic forecasting, paving the way for approaches that combine flow dependent perturbations with generative or latent-space uncertainty modeling for reliable artificial intelligence-driven early warning systems.