Abstract
In order to facilitate the adoption of deep learning in areas where decisions are of critical importance, understanding the model's internal workings is paramount. Nevertheless, since most models are considered black boxes, this task is usually not trivial, especially when the user does not have access to the network's intermediate outputs. In this paper, we propose IBISA, a model-agnostic attribution method that reaches state-of-the-art performance by optimizing sampling masks using the Information Bottleneck Principle. Our method improves on the previously known RISE and IBA techniques by placing the bottleneck right after the image input without complex formulations to estimate the mutual information. The method also requires only twenty forward passes and ten backward passes through the network, which is significantly faster than RISE, which needs at least 4000 forward passes. We evaluated IBISA using a VGG-16 and a ResNET-50 model, showing that our method produces explanations comparable or superior to IBA, RISE, and Grad-CAM but much more efficiently. © 2025 The Authors

Abstract
Artificial intelligence is gaining attraction in more ways than ever before. The popularity of language models and AI-based businesses has soared since ChatGPT was made available to the public via the OpenAI web platform. It gains popularity in a very short period because of its real-world problem-solving capability. Considering the widespread use of ChatGPT and the people relying on it, this study determined how reliable ChatGPT can be used for learning in the medical domain. The capability of ChatGPT was evaluated using the questions of Harvard University gross anatomy and the United States Medical Licensing Examination (USMLE). The outcome of the ChatGPT was analyzed using a 2-way ANOVA and post-hoc analysis. Both tests showed systematic covariation between format and prompt. Furthermore, the physician adjudicators independently rated the outcome’s accuracy, concordance, and insight into the answers given by ChatGPT. As a result of the analysis, ChatGPT-generated answers were more context-oriented and represented a better model for deductive reasoning than regular Google search results. Furthermore, ChatGPT obtained 58.8% on logical questions and 60% on ethical questions. This means that the ChatGPT is approaching the passing range for logical questions and has crossed the threshold for ethical questions. These results indicate that ChatGPT and other language-learning models can be invaluable tools for e-learners. © 2025 Elsevier B.V., All rights reserved.

Abstract
Epileptic seizure classification based on seizure semiology requires automated, quantitative approaches to support the diagnosis of epilepsy, which affects 1 % of the world's population. Current approaches address the problem on a seizure level, neglecting the detailed evaluation of the classification of the underlying action features, also known as Movements of Interest (MOIs), which are critical for epileptologists in determining their classifications. Moreover, it hinders objective comparison of these approaches and attribution of performance differences due to datasets, intra-dataset MOI distribution, or architecture variations. Objective evaluation of action recognition techniques is crucial, with MOIs serving as foundational elements of semiology for clinical in-bed applications to facilitate epileptic seizure classification. However, until now, there were no MOI datasets available nor benchmarks comparing different action recognition approaches for this clinical problem. Therefore, as a pilot, we introduced a novel, simulated seizure semiology dataset carried out by 8 experienced epileptologists in an EMU bed, consisting of 7 MOI classes. We compare several computer vision methods for MOI classification, two image-based (I3D and Uniformerv2), and two skeleton-based (ST-GCN++ and PoseC3D) action recognition approaches. This study emphasizes the advantages of a 2-stage skeleton-based action recognition approach in a transfer learning setting (4 classes) and the multi-scale challenge of MOI classification (7 classes), advocating for the integration of skeleton-based methods with hand gesture recognition technologies in the future. The study's controlled MOI simulation dataset provides us with the opportunity to advance the development of automated epileptic seizure classification systems, paving the way for enhancing their performance and having the potential to contribute to improved patient care.

Abstract
This study explores the application of the Lean Inception methodology in developing “EcoRider: Green Adventure,” an educational game aimed at enhancing motorcycle safety and promoting environmental awareness. Funded by the A-MoVeR project under the European Recovery and Resilience Facility, the game educates players on advanced safety technologies such as radars, cameras, LiDAR, and artificial intelligence (AI) algorithms. Players navigate complex urban scenarios, learning to manage potential hazards and promoting ecofriendly urban mobility. Using a qualitative case study approach, the research evaluates the effectiveness of integrating these technologies into the game’s design and gameplay. The game features multiple levels with increasing difficulty, requiring players to strategically place sensors and use AI models to overcome challenges. The application of the Lean Inception methodology has been essential in aligning the development team’s efforts, ensuring a cohesive approach to delivering a minimum viable product that satisfies both educational and technological objectives. Future work will be on refining the game, expanding its scope and exploring additional applications in the wider context of sustainable and safe mobility. © 2025 Elsevier B.V., All rights reserved.

Abstract
Mineral identification is a challenging task in geological sciences, which often implies multiple analyses of the physical and chemical properties of the samples for an accurate result. This task is particularly critical for the mining industry, where proper and fast mineral identification may translate into major efficiency and performance gains, such as in the case of the lithium mining industry. In this study, a mineral identification algorithm optimized for analyzing lithium-bearing samples using Laser-induced breakdown spectroscopy (LIBS) imaging, is put to the test with a set of representative samples. The algorithm incorporates advanced spectral processing techniques-baseline removal, Gaussian filtering, and data normalization-alongside unsupervised clustering to generate interpretable classification maps and auxiliary charts. These enhancements facilitate rapid and precise labelling of mineral compositions, significantly improving the interpretability and interactivity of the user interface. Extensive testing on diverse mineral samples with varying complexities confirmed the algorithm's robustness and broad applicability. Challenges related to sample granulometry and LIBS resolution were identified, suggesting future directions for optimizing system resolution to enhance classification accuracy in complex mineral matrices. The integration of this advanced algorithm with LIBS technology holds the potential to accelerate the mineral evaluation, paving the way for more efficient and sustainable mineral exploration.

Abstract
The integration of precise landing capabilities into unmanned aerial vehicles (UAVs) is crucial for enabling autonomous operations, particularly in challenging environments such as the offshore scenarios. This work proposes a heterogeneous perception system that incorporates a multimodal fiducial marker, designed to improve the accuracy and robustness of autonomous landing of UAVs in both daytime and nighttime operations. This work presents ViTAL-TAPE, a visual transformer-based model, that enhance the detection reliability of the landing target and overcomes the changes in the illumination conditions and viewpoint positions, where traditional methods fail. VITAL-TAPE is an end-to-end model that combines multimodal perceptual information, including photometric and radiometric data, to detect landing targets defined by a fiducial marker with 6 degrees-of-freedom. Extensive experiments have proved the ability of VITAL-TAPE to detect fiducial markers with an error of 0.01 m. Moreover, experiments using the RAVEN UAV, designed to endure the challenging weather conditions of offshore scenarios, demonstrated that the autonomous landing technology proposed in this work achieved an accuracy up to 0.1 m. This research also presents the first successful autonomous operation of a UAV in a commercial offshore wind farm with floating foundations installed in the Atlantic Ocean. These experiments showcased the system's accuracy, resilience and robustness, resulting in a precise landing technology that extends mission capabilities of UAVs, enabling autonomous and Beyond Visual Line of Sight offshore operations.