The transport sector includes data exchange across organizational boundaries. However, once uploaded, the data are under the control of existing platforms, which offer only limited usage and access controls for this data. For this purpose, the Transit data trustee platform was developed to enforce fine-grained usage control at the attribute level, supported by a dedicated access rights management service. The centrally stored data are only shared for the purposes of the data owner. Inside the platform, modules such as attributelevel usage control, central logging, geo-fencing, and a modular and multi-layered data trustee architecture ensure the proper use of the data.

Proprietary design in commercial windrow-detection systems restricts transparency and limits progress in open autonomous forage-harvesting research. We present a multi-modal dataset combining stereo vision and LiDAR from tractor-mounted sensors during real baling operations. The dataset includes synchronized sensor data with GNSS trajectories, partly released as ROS2 Humble bags on Zenodo, with additional data available on request. Using this dataset, we implement a real-time (>20 Hz) centroid-based windrow-following method on an NVIDIA Jetson AGX Orin. Across the critical 4-10 m guidance range, stereo and LiDAR depth measurements show strong agreement (0.965 +/- 0.021), indicating that low-cost stereo sensors can approach LiDAR performance. Our open-source ROS 2 pipeline provides a reproducible benchmark for GPS-free windrow detection and supports development of practical autonomous forage-harvesting systems. Dataset: https://zenodo.org/records/17486318

Digitizing processes such as energy modeling, risk management, or building permit procedures in the public sector require high-quality, up-to-date data of the urban building stock. Though this information is critical in many smart-city applications, it is often missing for many (older) buildings, and its collection is tedious. In this study, we investigate how to automate the classification of buildings into type and age on a large scale. In comparison to previous work, we use typologies specifically suited for urban planning. We assess and compare different machine-learning approaches using three distinct data sources: street-view images, LiDAR point clouds, and Level of Detail 2 data. Our comprehensive study demonstrates that image-based classification using YOLOv8 achieves the highest performance, reaching up to 97

Abstract Cities are increasingly threatened by the effects of rising temperatures. Due to the heat island effect, cities experience greater heat stress, leading to increased energy demand and reduced quality of life. Vertical greenery systems (VGSs), such as green façades and living walls, offer a spatially efficient strategy to mitigate these impacts. The environmental and social benefits of vertical greenery are well established. However, large-scale implementation is lacking methods that integrate relevant data to compute the variety of factors determining a surface’s suitability. This paper aims to fill this gap by introducing a computational method for evaluating the potential of individual building walls for vertical greening. Using the city of Leipzig as a case study, the approach integrates Level of Detail 2 (LoD2) building data with street view imagery to compute key factors such as orientation or Window-to-Wall Ratio (WWR). These factors are then integrated into a composite index prioritizing walls with high potential for reducing urban heat stress. The study thus provides a practical tool for urban planners and policymakers to support targeted climate adaptation strategies.

One of the major hurdles in Multi-Criteria Decision Analysis (MCDA) is the re-identification of pre-existing decision models. Due to factors like limited access to domain experts, some models become impractical, leading to the need for methods that aid in their re-identification. Addressing the challenge of re-identifying decision models brings up issues related to updating the last version of models to preserve their effectiveness, particularly in nonlinear decision scenarios. Common MCDA methods, which are based on linearity assumptions, encounter difficulties when dealing with nonlinearity, making it necessary to investigate practical methods for re-identifying nonlinear models. In this paper, we present innovative methods for re-identifying MCDA models utilizing optimization and machine learning techniques. Firstly, we introduce Support Vector Regression - Characteristic Objects Method (SVR-COMET), which combines the SVR and COMET methods. Secondly, we developed several extensions of the Stochastic Identification ofWeights (SITW) algorithm. These methods were evaluated against a benchmark comprising four selected MCDA techniques. Comparisons were performed using Spearman’s weighted correlation coefficient ( rw ). The findings from the study indicate that the proposed methods for re-identifying MCDA models are capable of mapping both linear and non-linear decision-making models.

Prostate cancer (PCa) is the second most common cancer in men worldwide, presenting significant challenges in both diagnosis and treatment. To determine the most effective therapy for each patient, accurate staging and grading of the cancer are essential, but often difficult due to the complexities involved in both staging and grading. Staging prostate cancer involves assessing its extent within the prostate and its spread to other parts of the body. This requires precise imaging and interpretation of scans, such as magnetic resonance imaging (MRI), and can be challenging due to the prostate lesions' small size, often resulting in missed lesions. Grading evaluates how much cancer cells differ from normal cells, typically using the Gleason scoring system, where pathologists examine tissue samples under a microscope. This process is challenging because it heavily depends on the pathologist's expertise, leading to variability and inconsistent results. To address these challenges, we employ recent advances in deep learning, specifically self-supervised learning (SSL), and transformer-architectures to develop an open-source AI framework aimed at enhancing the staging and grading process. Our framework is also equipped with a module interface that allows for integrating individual image analysis use cases. To demonstrate the potential of our framework, we simulate actual clinical trials to test the AI system under realistic conditions. Our AI system has demonstrated the capability to accurately identify cancerous lesions in the majority of MRI and histopathology cases. Moreover, it can grade a substantial number of cases, including identifying various subtypes of prostate cancer, and assists in routine tasks. The results of our studies indicate that AI may indeed enhance the accuracy of staging and grading in prostate cancer diagnosis and has the potential to make diagnostic practice more efficient and reproducible.

Effective deployment and updating of machine learning models are crucial for maintaining operational efficiency in industrial environments. Model performance often degrades over time due to Concept Drift, necessitating regular updates of initially trained machine learning models. This study addresses gaps in the literature, particularly the lack of comparisons of model adaptation techniques across different machine learning regression model types and the empirical evaluation of these update techniques in real-world scenarios. In the course of this work, experiments were conducted on deploying machine learning models in two virtual metrology cases from semiconductor fabrication, investigating the performance of model-type independent update techniques. The study examines how update frequency and the amount of new data used for the model updates influence performance of the update techniques. Results confirm that update techniques using a sliding window including historical labeled data are optimal across all frequencies and batch sizes. The optimal update technique out of the family of sliding window update techniques varies across the different use cases and model types. The findings highlight the need for case-specific testing to determine the best update technique as well as the best update frequency and amount of data used for updates.

The growth of e-commerce has intensified challenges in the last-mile delivery, which represents a critical and costly segment of the supply chain. The increasing demand for integrative and sustainable supply solutions highlights the need for approaches that balance the environmental, social, and economic dimensions of last-mile delivery. A promising solution involves consolidating deliveries with zero-emission vehicles to central, barrier-free pick-up stations in urban outskirts, reducing environmental impact while addressing accessibility issues. Simulation-based approaches are emphasized as a tool to address the last-mile delivery challenges of urban outskirts. This paper presents a systematic literature review on simulation-based approaches to sustainable and inclusive last-mile delivery in urban outskirts. It addresses the underexplored logistical challenges in these areas and examines how customer needs impact on the three pillars of sustainability: social inclusion, ecological responsibility, and economic viability. By synthesizing current peer-reviewed research from 2020 to 2025, this study identifies critical research gaps, particularly with regard to designing inclusive infrastructures and developing adaptive simulation tools for rural areas and urban outskirts. This work contributes to a comprehensive research agenda, offering insights into inclusive, efficient, and environmentally conscious last-mile delivery solutions for urban outskirts.

With the increase in data sharing and a growing number of data consumers, access control can become a challenging task to manage. This paper introduces the development of the Finer-Grained Attribute-based Policy (FiGAP) machine, an innovative framework designed to manage access within a Data Trust. FiGAP leverages fine-grained access control mechanisms with inheritance to facilitate enhanced data sharing across different organizational boundaries. Unlike other access control models, this paper develops and thoroughly explains a distinct framework that allows for the establishment of an adaptable Data Trust, potentially as a service. Additionally, this approach also provides an exceptionally detailed level of permission control.

This paper introduces a novel decision-making framework designed to enhance personalized support in Decision Support Systems (DSS). Existing methods often rely on single-dimensional modeling, which can introduce bias and uncertainty in criteria weighting. To address this, the study introduces a new approach for integrating subjective preferences with the multiple Expected Solution Points (ESPs) approach within the Characteristic Objects Method (COMET). This dual-dimensional approach tailors decision-making to align more closely with individual preferences and desired outcomes. Applied to the housing location selection problem, the proposed approach captures diverse user needs, delivering highly personalized and relevant recommendations. A comparative analysis with selected subjective weighting methods validates the stability and reliability of the multiple ESP-COMET model. By advancing personalization in decision support systems, this research offers a robust framework for more accurate and user-centered decision-making across complex scenarios.

Aprosumer is a person who plays the role of a consumer and a producer simultaneously. He shares his knowledge and experience from using a product, gives feedback, and helps to improve products and services. Such an attitude is very desired by the producers as they benefit from it. The most common way to get feedback is to conduct a survey. Sometimes, it is hard to convince clients to fill out the questionnaire. Machine learning and deep learning techniques give opportunities to predict survey responses based on a sample of real survey results. The aim of the research was to determine the possibility of classifying prosumer survey respondents depending on their method of assessing each decision-making factor separately using selected machine learning and deep learning methods. Decision trees, LightGBM, and neural network models were used. The input data were the respondents' characteristics and the output data was the importance of four factors in the prosumption decision-making process. The accuracy of built models was the highest for prosumption destimulants using DNN, for information contribution using Decision trees, for communication channels using DNN and for stimulants using LightGBM. The presented research shows that there is a possibility of predicting the importance of decision-making factors based on respondents' characteristics. It may be especially important for smaller companies that do not have much money or time to take large market surveys.

The growing number of cars and limited street space present significant challenges for cities, applying not only to moving but extending to stationary traffic. The quest for parking spaces exacerbates traffic congestion, noise, and air pollution, particularly in residential areas. To develop effective parking solutions for these challenges, a trustful data foundation on available parking space capacities, its usage and parking type is crucial. Gathering this data is currently time-consuming, requiring manual labeling and street inspections. Moreover, it must be repeated to keep the data current. Research on parking space management has heavily focused on monitoring designated parking lots with fixed cameras to identify free or occupied parking spaces. However, due to privacy concerns fixed cameras are not applicable for the larger part of the street space in European cities. This paper introduces a novel computer vision-based method for automatically collecting parking space capacities and parking type information. Our approach combines both street view and aerial imagery, which are recorded by a moving camera source. We tackle challenges in geo-referencing images, identifying parking types, classifying moving and stationary cars and dealing with partial occlusions in images. By not permanently recording the same environment, our approach lowers the surveillance risk, making parking capacity estimation scalable. We conduct a thorough evaluation of our methods and release a novel validation data set to allow for further research. In the future, more moving camera sources will be available when attached to city cleaning vehicles or to delivery drones.

Efficient satellite communications play an enormously important role in all of our daily lives. This includes the transmission of data for communication purposes, the operation of IoT applications or the provision of data for ground stations. More and more, AI-based methods are finding their way into these areas. This paper gives an overview of current research in the field of intelligent optimization of satellite communication. For this purpose, a text-mining based literature review was conducted and the identified papers were thematically clustered and analyzed. The identified clusters cover the main topics of routing, resource allocation and, load balancing. Through such a clustering of the literature in overarching topics, a structured analysis of the research papers was enabled, allowing the identification of latest technologies and approaches as well as research needs for intelligent optimization of satellite communication.

The application of developments in the real world is the final aim of all scientific works. In the case of Data Science and Machine Learning, this means there are additional tasks to care about, compared to the rather academic part of "just" building a model based on the available data. In the well accepted Cross Industry Standard for Data Mining (CRISP-DM), one of these tasks is the maintenance of the deployed application. This task can be of extreme importance, since in real-world applications the model performance often decreases over time, usually due to Concept Drift. This directly leads to the need to adapt/update the used Machine Learning model. In this work, available model-agnostic model update methods are evaluated on a real-world industry application, here Virtual Metrology in semiconductor fabrication. The results show that for the real-world use case sliding window techniques performed best. The models used in the experiments were an XGBoost and Neural Network. For the Neural Network, Model-Agnostic Meta-Learning and Learning to learn by Gradient Descent by Gradient Descent were applied as update techniques (among others) and did not show any improvement compared to the baseline of not updating the Neural Network. The implementation of the update techniques was validated on an artificial use case for which they worked well.

The paper aims to create an understanding of the importance of remote sensing data and its efficient analysis. The authors show the benefits that the combination of unmanned aerial vehicles and deep learning algorithms can create. The focus of the paper is on a practical example: using a semi-supervised neural network and remote sensing data to distinguish between healthy and unhealthy trees. This scenario occurs in similar contexts in the real world and can provide added value for the observation of fruit trees and forest stands. Particularly in view of climate change and its effects, it is extremely important to ensure good health monitoring of ecosystems in order to initiate appropriate measures at an early stage. An Open-Source-framework serves as the basis. It is based on a semi-supervised approach and promises to create a good neural network with comparatively little annotated data. This work shows how Deepforest can be extended from simple tree detection to multi-class health monitoring.