Implementing a cognitive Sense-Think-Act-Learn (STAL) architecture for automated cultural heritage visualization, this paper illustrates a multi-agent system. The system consists of three specialized agents: an Exhibition Curator Agent that applies Large Language Models (LLMs) to classify and generate exhibitions, a Conversational Guide Agent that provides interactive visitor engagement, and an Image Acquisition Agent that performs automated visual content enrichment. Evaluated on a dataset including 6,398 historical events across multiple civilizations and epochs, the system effectively automates the curation of the data by adopting a hybrid approach via deterministic classification algorithms as well as LLM-based analysis. The architecture allows adaptive learning based on administrative feedback, improving classification accuracy over time. This work provides a concrete framework to leverage generative $A I$ in cultural heritage digitization while also addressing issues related to scale, multilingual content, and domain-specific curation needs.

This paper explores the application of FPGA (Field Programmable Gate Array) technology based on the Basys 3 board in video game development. Given that the hardware of most classic games is no longer functional, the programmable nature of FPGA allows for precise replication, allowing modern devices to run these games and provide an authentic gaming experience. The paper presents the design and emulation of a video game on the Basys 3 FPGA development system. It also examines the use of open-source tools for the development of FPGA applications, comparing them with commercial alternatives. The Basys 3 system was prepared for game emulation by implementing a RISC-V processor, a VGA controller, and memory in Verilog. The game was initially developed in $\mathbf{C}$ for Windows and later ported to the FPGA environment, with particular attention paid to memory management to ensure proper image display via the VGA controller. The results demonstrate that FPGA systems are viable platforms for video game emulation and complex application development, with open-source tools proving efficient and effective.

This paper presents a novel UART-based debugging interface for resource-constrained RISC-V soft-core implementations on FPGAs. Unlike traditional JTAG-based approaches that require dedicated hardware and tools, our design leverages the ubiquitous UART peripheral to provide comprehensive debug capabilities through a dual-mode architecture. The interface operates in standard ASCII mode for command-line interaction and switches to a binary protocol for advanced operations including bulk instruction memory programming (IMPR), singleinstruction hot-patching (IMWR), and direct memory/CSR bus read/write operations (BUSR/BUSW). A key innovation is the bus mastering mechanism that enables real-time memory inspection and modification without permanent CPU halting, facilitating live debugging and in-field firmware updates. The FSM-based protocol incorporates checksum verification and timeout recovery for robust operation. Implemented on an 80 MHz RISC-V based SoC for WireGuard VPN acceleration, the interface consumes less than 2% additional FPGA resources while providing functionality comparable to more complex debug modules. Experimental results demonstrate successful hot-patching of running programs, sub-second firmware updates, and effective production diagnostics without requiring specialized JTAG hardware or tools.

The global transition to renewable energy faces challenges, particularly in integrating variable sources such as wind and solar. Battery Energy Storage Systems (BESS) provide a key solution for grid stabilization and peak load management. Peak shaving stores energy during low-demand periods and releases it during high-demand periods, reducing costs and stabilizing the grid. This research aims to model and analyze optimal BESS operation for peak shaving in industrial environments, highlighting both technical performance and contributions to sustainable energy systems. MATLAB/Simulink simulations evaluate effects on grid dependency, energy efficiency, and economic benefits, showing how BESS with photovoltaic generation can enhance efficiency, reduce grid reliance, and support environmentally friendly energy management.

In the context of business systems, efficient data analysis through various Online Analytical Processing (OLAP) models represents a key challenge for performance optimization and timely decision-making. This study examines tabular and multidimensional OLAP models within the SQL Server (MSSQL) and Visual Studio environments to determine which model facilitates more effective data processing in a specific business context. An experimental analysis was conducted using the Stats dataset, where the same business question was addressed through both models, comparing their response times and query execution efficiency. Particular attention was paid to execution speed, data processing methodologies, and resource optimization strategies. Results indicated that the tabular model, which relies on in-memory technology and the Data Analysis Expressions (DAX) language, reduces data processing time by approximately 38 % and offers simpler modeling capabilities, making it suitable for analyses where rapid result retrieval is critical. In contrast, the multidimensional model, utilizing Multidimensional Expressions (MDX), provides advanced analytical features and greater scalability, rendering it more appropriate for complex analyses involving large datasets and predefined aggregations. Based on this comparison, it was concluded that the choice between tabular and multidimensional OLAP models depends on specific analytical requirements. If speed and flexibility are prioritized, the tabular model enables faster execution, whereas the multidimensional model offers enhanced control over analytical processes.

Online Analytical Processing (OLAP) technology facilitates efficient multidimensional data analysis, providing users with valuable insights for decision-making processes. Previous studies have explored the implementation of OLAP technology across various domains; however, a limited number of investigations have compared the Multidimensional Analysis Project and Pentaho on the same database or within a single study. This research contributes to existing literature by evaluating the performance and flexibility of these two tools using Microsoft SQL Server as a benchmark dataset, which represents the database of a specific blog application or an application based on user interactions with diverse posts. The manuscript details the modeling and implementation processes for OLAP cubes in both systems, emphasizing fundamental aspects of OLAP technology, key functionalities, performance metrics, and adaptability characteristics. Furthermore, a comparative analysis between Microsoft's solution and Pentaho was conducted, highlighting their respective advantages and limitations within the context of data analytics.

The development of reliable Decision Support Systems (DSS) for talent identification requires a rigorous analytical framework capable of processing high-dimensional educational data. This paper presents the mathematical formulation of the machine learning pipeline utilized for classifying student potential, focusing on the algebraic structure of data representation and the optimization of predictive algorithms. We formally define the mapping of unstructured textual attributes into sparse vector spaces using One-Hot Encoding and analyze the dimensionality reduction effects. The study details the training dynamics of classification models, specifically examining the cost function minimization in Decision Trees via the Gini Impurity index and the stochastic aggregation mechanisms within Random Forest ensembles. Furthermore, to address the challenge of class imbalance, we provide a formal definition of performance metrics, including the harmonic mean of precision and recall and the arithmetic mean of indicator functions for Global Top-K Accuracy. By establishing these mathematical foundations, the paper demonstrates how formal optimization directly correlates with the discriminative power and stability of AI-driven educational assessments.

This study investigates the application of machine learning clustering techniques, specifically Dynamic Time Warping (DTW), to define typical load profiles (TLPs) for industrial facilities. Utilizing $\mathbf{1 5}$-minute smart meter data from a plastics manufacturing plant, the research analyzes total factory consumption alongside individual chiller and compressor loads. Cluster quality is assessed using the Silhouette score, Dunn index, and mean intra-cluster distance. Results indicate that while DTW effectively captures temporal shapes, industrial profiles are highly enterprise-specific and noise-intensive, resulting in fair-to-weak cluster quality. The findings suggest that primary electricity datasets and basic temporal metadata are insufficient for high-quality profiling compared to existing household models. The study concludes that integrating production-related metadata, such as work orders, is essential for improving industrial consumption forecasting and capacity planning.

As web applications have grown to become more dynamic, frontend rendering strategies have also become more central in the architectural discussions. The three main strategies are client-side, server-side, and hybrid rendering, yet empirical comparisons of them under controlled conditions are limited. Each of them has their own set of trade-offs, and this study systematically evaluates them using three functionally and visually identical frontends which are connected to a single backend. The metrics which were collected described the rendering strategies' performance, user experience, search engine optimisation and crawlability, and resource utilisation. Results show that server-side and hybrid rendering significantly improve initial load performance and search engine optimisation compared to client-side rendering, reducing First Contentful Paint by approximately 65% on average, while client-side rendering reduces server resource usage but suffers from delayed rendering and poor crawlability. Hybrid rendering achieves the best balance between performance and search engine optimisation, but with higher resource consumption. These findings highlight that rendering strategy selection should be driven by application requirements, considering crawlability, infrastructure, and performance objectives.

Photovoltaic (PV) power plants have become an important source of clean electricity. To keep them running efficiently, it is essential to detect faults and unusual behavior. However, this is not easy because data from PV power plants are noisy and strongly affected by environment, seasonal changes, and weather conditions. Reliable anomaly detection could help with preventing power losses, reducing maintenance costs, and extending the lifetime of the system. This paper presents a two step approach for detecting anomalies in PV plant data. First, K-means clustering is used to group days with similar conditions based on daily energy metrics. Then, the Local Outlier Factor (LOF) algorithm is applied within each cluster to find whether any anomalies are present. For visualization and interpretability, results are presented in a graphic form. Initial application demonstrates the framework's ability to segment seasonal data and identify extreme deviations from normal operating clusters. Although larger datasets are still needed for full validation, the method provides a foundation for season-aware monitoring.

Three-dimensional geometric morphometric methods have emerged as a pivotal tool in veterinary anatomy, taxonomy, clinical research, and studies of morphological diversity. This article summarizes the key stages, applications, clinical potential, and recommendations for data standardization in 3D morphometrics. Datasets are typically acquired using radiological modalities, including computed tomography (CT), magnetic resonance imaging (MRI), and 3D surface scanning, each offering specific advantages and constraints contingent on the research context. Standardized landmark sets are essential in 3D morphometric studies to ensure reproducibility and comparability of results across independent investigations. Consistent use of reference landmarks enables repeatable analyses, but the number of landmarks directly influences the required sample size and statistical power. Consequently, a minimal yet balanced landmark configuration is critical. This article proposes a standardized, minimal landmark set for the skulls of horses, cattle, and sheep to enhance inter-study reproducibility and comparability. Landmark selection prioritizes anatomically distinct points to avoid excessive landmarking, which may complicate analyses or compromise interpretability. Applications of 3D morphometric methods include orthopedic surgical planning, biomechanical modeling, and assessment of congenital anomalies, providing enhanced precision in diagnostics and research. In conclusion, 3D geometric morphometric methods represent a robust analytical framework in veterinary anatomy, morphology, and clinical research. Their significance is poised to grow through integration with automated landmarking, artificial intelligence-driven analyses, and international data-sharing networks, thereby advancing scientific inquiry in novel dimensions.

Pigmentation is a key functional trait influencing camouflage, predator-prey interactions and energetic efficiency in marine organisms, yet its physiological and ecological consequences remain poorly understood in deep-sea sharks. Here, we describe a deep-sea shark (Heptranchias perlo) exhibiting a mosaic pigmentation disorder characterized by the coexistence of hypermelanotic, hypopigmented and amelanotic regions, indicating disruption of normal melanophore distribution and regulation. Histological examination revealed no structural or inflammatory abnormalities, supporting a non-pathological origin of the pigmentation anomaly. In contrast, condition indices indicated pronounced energetic depletion, with reduced condition factor and hepatosomatic index, while lipid extraction and Fourier-transform infrared and ultraviolet-visible spectroscopy revealed substantial depletion and altered composition of hepatic lipid reserves consistent with chronic negative energy balance relative to phenotypically normal conspecifics. We propose that disruption of countershading in hexanchiform sharks may reduce camouflage efficiency and increase energetic costs, contributing to the observed physiological compromise in sharks. Despite being based on a single individual, this integrative analysis links pigmentation anomalies to functional and energetic consequences, and underscores the need to move beyond descriptive accounts toward mechanistic assessments of coloration in marine predators, particularly in deep-sea elasmobranchs that are inherently rarely encountered.

In the era of Cloud Computing and High Performance Computing as a Service, multiple customers run their software and hardware on the same physical devices. This opens new opportunities for attackers, like Denial of Service (DoS) attacks on shared cloud Field Programmable Gate Arrays (FP-GAs). DoS attacks work by employing a malicious hardware design as a tenant on a shared FP-GA device, draining the FPGA's power budget. This paper aims at identifying potentially malicious circuits before they are deployed to the FPGA, using formal verification. Our methodology, PowerCourt, provides an accurate worst-case power estimation based on switching activity. It computes metadata about the design that enables service providers to classify designs as malicious. The method checks formal properties that verify whether a design can exceed a custom threshold and whether this threshold can be exceeded repeatedly. We demonstrate the effectiveness of our approach using various hardware designs. For example, for a trojan that enables excessive switching, we show that we can detect it offline and disable it during runtime.

The thermal stability and structural evolution of a GaN high-electron mobility transistor (HEMT) heterostructure grown on a Si (111) substrate were investigated using in situ high-temperature x-ray diffraction, reciprocal space mapping (RSM), Raman spectroscopy, and rocking-curve (RC) analysis at varying temperatures. The heterostructure, consisting of a p-GaN cap, an AlGaN barrier, and a GaN channel supported by two AlGaN/AlGaN superlattice buffer layers, maintained clear and periodic satellite peaks up to a temperature of 1000 °C, confirming structural integrity. Symmetric and asymmetric RSM results reveal that both the Si and GaN diffraction peaks shift with increasing temperature, consistent with thermal expansion, and show no significant broadening or relaxation throughout the heating process. The c-lattice constant follows the theoretical expansion predicted by the multi-frequency Einstein model, whereas the a-lattice expansion is slower due to in-plane strain constraints imposed by the underlying Si substrate and buffer layers. Irreversible strain relaxation and thermal mismatch-induced stress redistribution induce a residual compressive strain of roughly 0.3% in the GaN channel after cooling, which has been further confirmed in Raman spectra through a blue shift (∼1 cm−1) of the GaN E2 (high) phonon mode, corresponding to an in-plane strain of −0.15% ± 0.16%. RC analysis revealed an increase in both screw and edge dislocation densities of 28% and 12%, respectively. These results collectively demonstrate that the GaN HEMT heterostructure exhibits robust crystalline stability up to 1000 °C without cracking due to strain relaxation, with only minor strain redistribution and limited dislocation activity, providing experimental evidence for GaN devices' applications under high-temperature conditions.

Background/Objectives: Spontaneous intracerebral hemorrhage (sICH) is a particularly severe subtype of stroke, characterized by high rates of mortality and long-term disability, for which robust prognostic markers are still lacking. The aim of this study was to assess the relationship of the ICH score, the National Institutes of Health Stroke Scale (NIHSS) score, and serum high-sensitivity cardiac troponin I (hs-cTnI) levels with 30-day mortality in patients with sICH. Methods: We conducted a prospective observational cohort study enrolling 100 consecutive patients diagnosed with sICH based on neuroimaging findings. Demographic data, clinical parameters, neuroimaging findings, and serum hs-cTnI levels were collected on admission. Subsequently, the ICH score, its individual components, and the NIHSS score were assessed. Results: Patients who died were older and had significantly higher ICH and NIHSS scores, lower Glasgow Coma Scale (GCS) scores, larger hematoma volumes, more frequent intraventricular hemorrhage (IVH), and elevated hs-cTnI levels compared to survivors. Serum hs-cTnI concentrations were significantly correlated with ICH and NIHSS scores, lower GCS scores, larger hematoma volumes, and the presence of IVH. On univariate logistic regression, higher ICH score, NIHSS score, and hs-cTnI level were associated with mortality, whereas multivariate analysis identified the GCS score, hematoma volume, and IVH score as significant independent factors related to fatal outcome. Conclusions: Individual components of the ICH score may provide useful information on outcomes in patients with sICH. Higher serum hs-cTnI levels were associated with 30-day mortality but were not independent predictors. These markers may assist in patient monitoring and support established clinical procedures in therapeutic decision-making. Nevertheless, larger multicenter studies are needed to further clarify their clinical implications in sICH management.