Interpretable-by-design architectures, such as Concept-Bottleneck Models (CBMs), are essential for trustworthy AI under regulations like the EU AI Act. While classical CBMs rely on costly expert labels, recent automated methods use vision–language models for concept discovery. However, whether automation preserves interpretability remains an open question. We systematically evaluate four automated methods (LF-CBMs, LaBo, PCBMs, and VLG-CBMs) against an expert-supervised CBM using fine-grained binary classification (Amanita muscaria vs. Boletus edulis) from the FungiTastic dataset. Although automated methods match expert accuracy, they systematically fail with regard to two key interpretability requirements: concept atomicity and instance-level grounding. Specifically, automated concepts conflate multiple morphological properties, yield scores ungrounded in visual evidence, and produce biologically implausible class-concept associations; a specific failure mode, cross-class contamination, is one that the expert-supervised vocabulary avoids by construction, though expert supervision remains subject to its own annotation and validation risks. We characterize four distinct failure modes across these architectures, concluding that automated concept discovery cannot yet substitute domain expert supervision in safety-critical tasks where explanation fidelity is a functional requirement alongside predictive accuracy.

Deep learning classifiers have achieved high accuracy on plant disease recognition tasks, but their decision-making processes remain opaque. Counterfactual explanations (CFs), minimally modified inputs flipping a classifier’s prediction, reveal which input changes are sufficient to alter the decision. While diffusion-based counterfactual generation has been studied primarily on controlled face datasets (CelebA), its systematic evaluation for fine-grained plant disease classification, particularly under in-the-wild conditions, remains limited. In this work, we apply for the first time DiME (Diffusion Models for CFs) to plant disease classification, evaluating its behavior on both controlled (PlantVillage) and in-the-wild (PlantWild) data. We adapt the pipeline using Stable Diffusion with LoRA fine-tuning as the generative backbone, and we propose Plant Verification Accuracy (PVA), a domain-adapted variant of the Face Verification Accuracy (FVA) metric, to measure species identity preservation in the generated counterfactuals. We compare DiME against three established baselines: Wachter (pixel-space gradient), xGEM+, and DiVE (both based on Variational Autoencoders). On PlantVillage, DiME achieves a 7.1 percentage-point PVA drop versus 60–65 pp for Variational Autoencoders baselines while preserving the target flip rate. Extension to PlantWild yields larger PVA drops (16.6–18.6 pp) and qualitative degradation on in-the-wild imagery, identifying current limitations of the approach when applied outside controlled settings.

Sensorized assets are increasingly used in smart and connected settings to support safe and effective operations.However, human operators often struggle to interpret monitoring outputs, especially under continuous monitoring.This paper presents a privacy-preserving LLM-based reporting framework that converts structured monitoring outputs into insights to support operators’ decision-making. The framework combines overlap-aware chunking with hierarchical aggregation to address finite context-window constraints.The results support the feasibility of on-premise agentic AI as a compliant and interpretable decision-support layer for privacy-sensitive IoT monitoring contexts.

This study presents an artificial intelligence-assisted visual inspection procedure and a preliminary resilience assessment technique for the post-earthquake evaluation of masonry structures affected by major Italian earthquakes since 1980. A dataset of 250 images, collected during official surveys conducted by the Italian Civil Protection Department, was analyzed to automatically identify earthquake-induced damage patterns in spatial masonry components. The images, acquired both inside and outside damaged buildings and domed structures, were divided into training, validation, and test sets. The proposed methodology, although still at a preliminary stage due to the limited size of the dataset employed, aims to advance the use of artificial intelligence as a decision-support tool for enhancing structural resilience in post-earthquake scenarios, with particular attention to historic masonry constructions. After training, the AI model achieved a strong ability to correctly identify earthquake-induced damage patterns in masonry structures. The model was also deployed for inference on previously unseen test images, where the predicted bounding boxes qualitatively confirmed its effectiveness in detecting damage patterns. From a mechanical standpoint, the proposed approach supports the formulation of discrete no-tension models for masonry walls and domes affected by seismic events, based on the damage predictions provided by the AI-assisted detection procedure, which are subsequently translated into mechanical representations through engineering-driven post-processing operations. A recently developed strut-and-net approach is then employed to verify the existence of a network of compressed masonry struts capable of sustaining the vertical and horizontal loads acting on the examined structural systems.

Kolmogorov–Arnold Networks employ learnable univariate activation functions on edges rather than fixed node nonlinearities. Standard B-spline implementations require O(3KW) parameters per layer (K basis functions, W connections). We introduce shared Gaussian radial basis functions with learnable centers μk(l) and widths σk(l) maintained globally per layer, reducing parameter complexity to O(KW+2LK) for L layers—a threefold reduction, while preserving Sobolev convergence rates O(hs−Ω). Width clamping at σmin=10−6 and tripartite regularization ensure numerical stability. On MNIST with architecture [784,128,10] and K=5, RBF-KAN achieves 87.8% test accuracy versus 89.1% for B-spline KAN with 1.4× speedup and 33% memory reduction, though generalization gap increases from 1.1% to 2.7% due to global Gaussian support. Physics-informed neural networks demonstrate substantial improvements on partial differential equations: elliptic problems exhibit a 45× reduction in PDE residual and maximum pointwise error, decreasing from 1.32 to 0.18; parabolic problems achieve a 2.1× accuracy gain; hyperbolic wave equations show a 19.3× improvement in maximum error and a 6.25× reduction in L2 norm. Superior hyperbolic performance derives from infinite differentiability of Gaussian bases, enabling accurate high-order derivatives without polynomial dissipation. Ablation studies confirm that coefficient regularization reduces mean error by 40%, while center diversity prevents basis collapse. Optimal basis count K∈[3,5] balances expressiveness and overfitting. The architecture establishes Gaussian RBFs as efficient alternatives to B-splines for learnable activation networks with advantages in scientific computing.

The deployment of interpretable deep learning models is usually constrained across several domains due to a fundamental scarcity of appropriately annotated data. Developing advanced interpretable models, particularly those designed for fine-grained classification, requires going beyond simple class labels to include rich metadata, such as attributes and part locations. This necessity is further amplified by growing international regulations that mandate the use of trustworthy and eXplainable Artificial Intelligence (XAI). Existing publicly available datasets often lack this crucial fine-grained information, necessitating custom annotation, often performed manually, for practical application. To overcome these challenges, we introduce the Concept Annotation Tool (CAT), a general-purpose, platform-independent annotation application. CAT is specifically designed to facilitate rapid and versatile enrichment of any image dataset. The application enables users to annotate each image by adding attributes, perform part annotation, and apply standard annotations, such as bounding boxes and image cropping. We demonstrate the practical utility and the general applicability of our tool by enriching two datasets, one from the agrifood sector related to mushroom classification, and another from healthcare making reference to pigmented skin lesion classification. Through these case studies, we demonstrate that using CAT significantly accelerates the assignment of fine-grained attributes for large datasets compared to manual methods.

Artificial Intelligence is increasingly reshaping geomorphological research by enabling scalable, data-driven analyses of complex Earth surface processes. In this contribution, we present a supervised machine-learning framework for reconstructing Late-Quaternary coastal paleo-landscapes, applied to the rocky coasts of the Cilento Promontory (southern Tyrrhenian Sea), a tectonically quasi-stable sector preserving well-constrained sea-level indicators.We trained a Random Forest classifier on an expert-labelled geomorphological dataset integrating DEM-derived morphometric parameters, lithology, distance from the coastline, and field-validated paleo-environmental markers. The model was developed within a fully reproducible workflow and validated against independent geomorphological mapping and sea-level proxy datasets.Results demonstrate high classification performance and the ability to automatically discriminate between Last Interglacial paleo-sea cliffs and polycyclic, currently active coastal cliffs across different lithological contexts. The AI-based approach overcomes key limitations of traditional “bathtub” methods, allowing the detection of relict and partially buried landforms and extending paleo-landscape reconstructions into areas lacking direct field evidence.Beyond the specific case study, this work illustrates how machine-learning approaches can be effectively integrated with geomorphological knowledge to reconstruct complex coastal paleo-landscapes. The proposed framework allows the identification of inherited and partially obscured landforms that are difficult to detect through traditional methods alone, offering a transferable tool for investigating long-term coastal evolution. This integration of AI and geomorphology provides new insights into the geomorphic response of rocky coasts to Quaternary sea-level fluctuations and climatic forcing.

Standard feedforward nets hang their nonlinearity on the nodes and keep the edges linear. Kolmogorov–Arnold Networks swap the two: each edge is now a learnable one-variable map, each node a plain summation. A small architectural move, but with a payoff we want to test here, namely, that the trained network becomes partly readable, edge by edge, without extra tooling. Our test bed is the shared-basis Gaussian variant of the KAN, recently published, and the task is regression on California Housing. We train a three-layer network with five Gaussian kernels per layer and then look at it from four angles. The first angle is the tensor of first-layer coefficients, summarized into a per-feature importance number. The second is KernelSHAP run globally; the third is LIME, pointwise; the fourth, and the one that only makes sense on a KAN, is the direct drawing of the learned edge activations as curves. Intrinsic and SHAP rankings land on the same two features at the top, MedInc and Latitude, and the same feature at the bottom (Population). They shuffle the middle, and that shuffle is diagnostic: coefficient size reports on encoding strength in layer zero, SHAP on end-to-end marginal effect. The plotted edge activations confirm what domain knowledge would suggest: income acts roughly monotonically on price, geography acts locally, crowding has a threshold effect. Taken together, the four views make the model easier to audit than a comparable black box would be.

In the literature, several studies have shown that state-of-the-art image similarity metrics are not perceptual metrics; moreover, they have difficulty evaluating images, especially when texture distortion is also present. In this work, we propose a new perceptual metric composed of two terms. The first term evaluates the dissimilarity between the textures of two images using Earth Mover's Distance. The second term evaluates the chromatic dissimilarity between two images in the Oklab perceptual color space. We evaluated the performance of our metric on a non-traditional dataset, called Berkeley-Adobe Perceptual Patch Similarity, which contains a wide range of complex distortions in shapes and colors. We have shown that our metric outperforms the state of the art, especially when images contain shape distortions, confirming also its greater perceptiveness. Furthermore, although deep black-box metrics could be very accurate, they only provide similarity scores between two images, without explaining their main differences and similarities. Our metric, on the other hand, provides visual explanations to support the calculated score, making the similarity assessment transparent and justified.

The quality of coastal waters, particularly aquaculture zones, is crucial to sustainable development and human health. Filter-feeding organisms farming, such as mussels (e.g., Mytilus galloprovincialis), is highly susceptible to bacterial contamination (e.g., E. Coli) from both natural and anthropogenic sources, posing substantial risks to consumers. Traditional contamination monitoring methods, based on periodic sampling and microbiological analysis, are often too slow, costly, and limited in spatial and temporal coverage to meet the needs of large-scale aquaculture management. Traditional monitoring methods can be combined with numerical modeling and artificial intelligence (AI). In this framework, we present Artificial Intelligence-based Water QUAlity Model Plus Plus (AIQUAM++) to predict levels of potential bacterial contamination in mussels. AIQUAM++ integrates an ensemble of machine learning (ML) models to address the Time Series Classification (TSC) problem posed by the dynamic nature of bacterial contamination in mussels, achieving robust predictive accuracy and capturing both short-term contamination spikes and long-term pollutant trends. The architecture of AIQUAM++ is designed for scalability and efficiency, utilizing a hierarchical parallelization approach that combines MPI (Message Passing Interface) for process-level parallelism, Open Multiprocessing (OpenMP) for threading, and CUDA for GPU acceleration. This configuration allows AIQUAM++ to process and analyze extensive datasets, making it suitable for application in multiple mussel farming regions to minimize the health risks associated with consuming contaminated mussels. We tested AIQUAM++ in the Gulf of Naples (Campania, Italy) and demonstrated the model's effectiveness. The model achieved an accuracy of 90% in contamination level classification, allowing timely and accurate predictions. AIQUAM++ represents a significant advance in the aquaculture economic sector, inserting itself as a decision-support tool for local authorities and aquaculture operators, enabling timely response to contamination events, and acting as a powerful, flexible tool to manage environmental quality in marine ecosystems.

The increasing complexity and scale of data-intensive scientific workflows necessitate advancements in workflow engines (WFEs) to handle real-time data streams and reduce input/output (I/O) bottlenecks. This paper introduces an innovative approach to enhancing the DAGonStar scientific workflow engine by integrating CAPIO, a middleware capable of injecting I/O streaming capabilities into traditional scientific workflows and optimized for high-speed data access and low latency. By combining DAGonStar's robust task orchestration and dependency management with CAPIO, we aim to significantly improve scientific workflows' performance and scalability. We present the design and implementation of this integration, detailing the architectural modifications required to enable seamless interaction between DAGonStar and CAPIO. The paper includes comprehensive benchmarks and performance evaluations demonstrating the impact of CAPIO on workflow execution times and data handling efficiency. Our findings indicate that the enhanced DAGonStar, equipped with CAPIO, offers a powerful solution for managing and processing large-scale, real-time data streams, thereby advancing the capabilities of scientific computing infrastructure.

The emergence of exascale computing systems presents both opportunities and challenges in scientific computing, particularly for complex mathematical models requiring high-performance implementations. This paper addresses these challenges in the context of biomedical applications, specifically focusing on tumor angiogenesis modeling. We present a parallel implementation for solving a system of partial differential equations that describe the dynamics of tumor-induced blood vessel formation. Our approach leverages the Julia programming language and its CUDA capabilities, combining a high-level paradigm with efficient GPU acceleration. The implementation incorporates advanced optimization strategies for memory management and kernel organization, demonstrating significant performance improvements for large-scale simulations while maintaining numerical accuracy. Experimental results confirm the performance gains and reliability of the proposed parallel implementation.

In the identification of new planetary candidates in transit surveys, the employment of deep learning models proved to be essential to efficiently analyze a continuously growing volume of photometric observations. To further improve the robustness of these models, it is necessary to exploit the complementarity of data collected from different transit surveys such as NASA’s Kepler, Transiting Exoplanet Survey Satellite (TESS), and, in the near future, the ESA Planetary Transits and Oscillation of stars mission. In this work, we present a deep learning model, named DART-Vetter , that is able to distinguish planetary candidates from false positives signals detected by any potential transiting survey. DART-Vetter is a convolutional neural network that processes only the light curves folded on the period of the relative signal, featuring a simpler and more compact architecture with respect to other triaging and/or vetting models available in the literature. We trained and tested DART-Vetter on several data sets of publicly available and homogeneously labelled TESS and Kepler light curves in order to prove the effectiveness of our model. Despite its simplicity, DART-Vetter achieves highly competitive triaging performance, with a recall rate of 91% on an ensemble of TESS and Kepler data, when compared to Exominer and Astronet-Triage . Its compact, open source, and easy to replicate architecture makes DART-Vetter a particularly useful tool for automatizing triaging procedures or assisting human vetters, showing a discrete generalization on threshold-crossing events with multiple event statistic > 20 and orbital period < 50 days.

Coastal landforms, particularly sea cliffs and associated wave-cut platforms, preserve key evidence of past sea-level fluctuations, tectonic activity, and paleoclimate variability. In this study, we implement a supervised machine learning approach, trained on an original, expert-labeled geomorphological dataset, to detect and classify inherited and active coastal features - such as paleo-sea cliffs and polycyclic sea cliffs - along the south-Tyrrhenian. Using DTM and morphometric indicators, our model, based on a RandomForestClassifier trained on expert-based cartography and independently validated, accurately identifies the spatial signatures of Quaternary coastal evolution. These results are cross validated against independent geomorphological mapping and sea-level reconstruction datasets. The integration of geomorphological classification with sea level markers enables us to reconstruct coastal morphogenesis in relation to the last interglacial cycle. Our findings highlight the potential of machine learning to automate the identification of coastal paleo-landscapes, providing insight into the imprint of climatic forcing on their morphology. This approach offers a scalable framework for investigating past climate-landscape interactions and for supporting future coastal hazard assessments under changing climate conditions.