Abstract
City councils play a crucial role in local governance, directly influencing citizens' daily lives through decisions made during municipal meetings. These deliberations are formally documented in meeting minutes, which serve as official records of discussions, decisions, and voting outcomes. Despite their importance, municipal meeting records have received little attention in Information Retrieval (IR) and Natural Language Processing (NLP), largely due to the lack of annotated datasets, which ultimately limit the development of computational models. To address this gap, we introduce CitiLink-Minutes, a multilayer dataset of 120 European Portuguese municipal meeting minutes from six municipalities. Unlike prior annotated datasets of parliamentary or video records, CitiLink-Minutes provides multilayer annotations and structured linkage of official written minutes. The dataset contains over one million tokens, with all personal identifiers de-identified. Each minute was manually annotated by two trained annotators and curated by an experienced linguist across four complementary dimensions: (1) personal information, (2) metadata, (3) subjects of discussion, and (4) voting outcomes, totaling over 38,000 individual annotations. Released under FAIR principles and accompanied by baseline results on metadata extraction, topic classification, and vote labeling, CitiLink-Minutes demonstrates its potential for downstream NLP and IR tasks, while promoting transparent access to municipal decisions.

Abstract
Model-Driven Engineering (MDE) offers significant benefits but faces usability challenges, particularly in developing intuitive modeling interfaces. Most existing tools rely on textual or tree-based editors, which are often difficult for non-technical users to navigate. Creating custom graphical editors based on existing metamodels, a common way to describe modeling environments and languages, typically requires substantial manual effort and specialized knowledge of complex frameworks such as Sirius. These hurdles limit the broader adoption of MDE among developers and domain experts. Consequently, there is an increasing demand for automated, flexible, and user-friendly approaches to generating interfaces that can make MDE more accessible and practical across various domains. In this work, we present a method for automatically generating a visual modeling environment from existing metamodels. Based on this method, we devised Model2Block, a command-line application that automatically generates block-based visual languages from metamodels defined using the Ecore language. The tool analyzes an input metamodel and produces a corresponding set of blocks, which are then integrated into a web-based environment. This enables users to visually build models using blocks that represent valid model elements. We present an applicability study demonstrating that our method and tool were successfully applied to a publicly available dataset containing more than 50 metamodels. Moreover, we present a case study of a concrete modeling environment we generated from a metamodel, achieving, in this particular case, higher usability and lower workload than an existing tool. © 2026 Copyright held by the owner/author(s).

Abstract
This paper explores a Cycle-GAN architecture based on diffusion models for translating cardiac CT images with and without contrast, aiming to enhance the quality and accuracy of medical imaging. The combination of GANs and diffusion models has demonstrated promising results, particularly in generating high-quality, visually similar contrast-enhanced cardiac images. This effectiveness is evidenced by metrics such as a PSNR of 32.85, an SSIM of 0.766, and an FID of 42.348, highlighting the model's capability for accurate and detailed image generation. Although these results indicate substantial potential for improving diagnostic accuracy, challenges remain, particularly concerning the generation of image artefacts and brightness inconsistencies, which could affect the clinical validation of these images. These issues have important implications for the reliability of the images in real medical diagnoses. The results of this study suggest that future research should focus on optimizing these aspects, improving the handling of artefacts, and investigating alternative architectures further to enhance the quality and reliability of the generated images, ensuring their applicability in clinical settings

Abstract
This study investigates the impact of virtual agent realism in immersive Virtual Reality (iVR) on foreign-language vocabulary learning. Specifically, it compares the effectiveness of a realistic (human-like) pedagogical virtual agent versus an abstract (non-human-like) one in delivering instructional content. A between-subjects experiment was conducted with 17 participants, divided into two groups, were exposed to either the realistic or abstract agent in an iVR Search-and-Find vocabulary learning task. Learning outcomes were measured using pre- and post-tests (based on word matching translations for 10 German-Portuguese item pairs), while presence-related experiences were assessed via the Igroup Presence Questionnaire and Temple Presence Inventory. Both groups demonstrated significant vocabulary acquisition improvements. However, no significant differences were found between the realistic and abstract agent groups in either learning outcomes or presence scores. The findings suggest that the visual realism of virtual agents may not significantly influence language learning effectiveness or user presence in these iVR environments. These preliminary results imply that abstract agents could be as effective as realistic agents for this type of foreign-language instruction, potentially reducing development resources without compromising learning benefits.

Abstract
Local flexibility markets have emerged as a promising market-based approach to accommodate the increasing penetration of distributed energy resources in distribution grids in a cost-efficient manner. This paper presents a novel fast datadriven methodology for the segmentation of medium-voltage distribution networks into flexibility zones based on historical operation data. Each flexibility zone groups nodes that affect network constraints in the same way, so that activating active power flexibility at any node within a zone produces the same effect on the network. This allows the DSO to assess and procure flexibility needs at zonal level, while enabling aggregators to manage and optimize their flexibility portfolios by zone. A case study is conducted to validate the methodology and the performance of the proposed data-driven grid segmentation by comparing with the dynamic grid segmentation approach. The results demonstrate the advantages of the proposed grid segmentation in terms of computational efficiency in real-time flexibility procurement. © 2026 IEEE.

Abstract
In CPU-FPGA systems, C/C++ applications are commonly accelerated by offloading selected code regions to the FPGA using HighLevel Synthesis (HLS). Despite the growing capacity of modern FPGAs to support large designs, those code regions are typically limited to kernels of small regions. This is primarily due to the limitations of the HLS tools, such as the lack of support for dynamic memory allocation and multiple levels of pointer dereferencing. This paper presents an automated approach that uses source-tosource transformations, such as function inlining, memory allocation hoisting, and struct flattening, to expand code regions beyond standard kernels and function-based hotspots in complex multifunction applications. We present a method that, by leveraging information from the entire program, enhances communication within each region and mitigates shared-memory data communication bottlenecks, mapping the most promising buffers (partially or fully) to FPGA BRAMs. We evaluate our approach using four applications from the CortexSuite-Vision benchmark suite. As a baseline, we consider the code regions that can be offloaded without transformations, which are severely limited by HLS restrictions and capable of capturing at most 3% of the program's CPU execution time. Then, we examine how lifting each restriction, through code transformations, can extend code regions that capture over similar to 80% of the execution time. Finally, we examine the impact of these regions on FPGA resources and latency. Our approach yields speedups of up to 31x for the application hotspots, and speedups ranging from 2x to 5x for the entire CPU-FPGA applications when compared to their CPU execution time baselines.