Context: The adoption of Microservice Architectures has grown significantly as a result of their ability to overcome key limitations of monolithic systems, such as limited scalability, maintenance difficulties, and technological lock-in. However, the design of microservice-based systems remains a complex and non-trivial task. This underscores the need for well-defined methodologies and techniques to support practitioners and software architects throughout the microservice design process. Objective: The objective of this study is to examine effective microservices design practices, microservice identification techniques, the tools employed to support design activities, and the factors that drive the decomposition of systems into new microservices. Methods: A Systematic Literature Review was conducted following the guidelines established by the PRISMA framework. From 609 studies that were retrieved from the search execution, 150 were selected and analyzed to answer the research questions. Results: This SLR contributes by examining design tools used for decomposition, analyzing microservice decomposition processes, identifying metrics for microservice decomposition, and framing microservice decomposition as a multi-objective optimization problem. Conclusions: This study highlights the complexity and variability inherent in microservices design, as well as the limited availability of tools to support this process. The findings identify microservices communication and service decomposition as key challenges, with the majority of existing research focusing primarily on the decomposition of monolithic systems.
In an era of frequent legislative changes and a growing volume of published laws, automating amendment-related processes is essential for effective analysis and interpretation of legal documents. Manually tracking and describing relationships between new and ammending acts in the Journal of Laws of the Republic of Poland is time-consuming and prone to error. These tasks can be supported through metadata extraction tools and process mining algorithms. This work presents a system for automatic extraction and visualization of amendment processes. Parsing tools were applied to obtain and process the texts of laws, transforming the extracted metadata into event logs and a Directly-Follows Graph (DFG) for process mining. A web-based application provides graphical representation of the data. The resulting tool automatically generates a graph illustrating relationships between acts and their evolution over time. A survey of legal professionals and non-expert users confirms improved efficiency and readability in identifying changes and visualizing legislative paths. The tool offers practical support for lawyers and analysts working with historical and current legislative processes.
Glaucoma is the second leading cause of irreversible blindness worldwide, affecting over 80 million people, with projections reaching 111 million by 2040. Early detection is critical, yet up to 50% of patients remain undiagnosed. This paper presents an integrated system for automated glaucoma screening combining a deep learning image analysis pipeline, an iOS mobile application, and a low-cost fundus camera prototype. The AI pipeline follows a two-stage architecture: a YOLOv9 model for region-of-interest (ROI) detection, followed by a UNet++ segmentation model for optic disc and cup delineation. The Cup-to-Disc Ratio (CDR) is computed from the resulting masks and used to classify glaucoma risk. Five YOLO model variants (v8, v9, v11, v12, v26) were evaluated on an augmented dataset of over 6100 fundus images; YOLOv9 achieved the best overall balance with precision of 98%. UNet++ reached approximately 90% accuracy on the segmentation task. The iOS application, built in Swift with an MVVM architecture and a FastAPI backend hosted on Microsoft Azure, streams real-time analysis progress to prevent frozen-screen effects. The fundus camera prototype, constructed from a Volk 20D lens and a PVC tube mounted on an iPhone 14 Pro, enables retinal image capture at a cost below 1300 PLN. The system targets clinical screening workflows while remaining accessible to individual users.
The article presents a package of services supporting energy efficiency within the CAISE platform environment, aimed at monitoring, managing, and optimizing energy consumption in cloud computing environments. The monitoring services include Computation Energy Consumption Monitoring (MEO) and Computation Intensity Monitoring (MIO), which enable real-time tracking of energy usage parameters and resource load. The management and administrative layer is formed by System Energy Configuration (KES) and Optimization Parameter Logging (RPO), which allow the definition of energy policies and the collection of data relevant for subsequent analysis and system tuning. A key component of the package is the Performance–Energy Optimization (OWE) service, which supports the selection of configurations and execution methods for computational tasks in order to achieve a balance between performance and energy efficiency. The paper discusses the functional assumptions, supported hardware–software configurations, and implementation aspects of these services. It also outlines the requirements and challenges related to their practical deployment, potential application perspectives and use cases in large-scale computing environments, as well as selected practical results for OWE.
Accurate detection of pulmonary nodules in chest CT is essential for early lung cancer diagnosis. This study presents a systematic evaluation of two-dimensional (2D) and three-dimensional (3D) deep learning detectors for pulmonary nodule detection using the LUNA16 benchmark dataset. While 3D deep learning approaches can exploit full volumetric information, 2D detectors remain attractive due to their lower computational complexity and simpler training requirements. Particular attention was devoted to the effect of annotation preprocessing in the 2D setting. Two 2D detectors (RetinaNet and YOLO) and two 3D detectors (3D RetinaNet and a hybrid CNN-Transformer architecture) were evaluated. For the 2D experiments, multiple dataset variants were prepared by filtering annotations according to minimum bounding-box area and by balancing positive and negative slices. Detection performance was evaluated using F1-score, precision, recall, mAP, mAR, and FROC-based metrics across multiple IoU thresholds. The results showed that preprocessing strategies substantially affected 2D detector performance. Filtering very small peripheral annotations and balancing the datasets improved localisation quality and detector stability for both RetinaNet and YOLO. Among the 2D configurations, the best results were obtained for the balanced filtered subsets. The 3D RetinaNet achieved the strongest overall performance, providing higher detection sensitivity, more stable localisation, and better generalisation between validation and independent test data. The CNN-Transformer also benefited from volumetric input and retained relatively high recall at lower IoU thresholds, although its performance remained lower than that of the 3D RetinaNet. The findings demonstrate the importance of volumetric context for robust pulmonary nodule detection and indicate that appropriate annotation preprocessing can substantially improve the effectiveness of slice-based 2D detection frameworks.
For Platform-as-a-Service (PaaS) environments characterized by elasticity, multi-tenancy, and layered dependencies, this paper proposes a structured method for selecting a vendor-neutral open-source observability stack. The proposed method is based on an analysis of commercial monitoring solutions and covers five functional categories: metrics monitoring, visualization, log monitoring, distributed tracing, and alert routing/incident response. Selected open-source tools are evaluated using technical and operational criteria, including integration, scalability, high availability, and maintainability, with a 1–5 Mean Opinion Score (MOS)-style rubric. Based on the results, the proposed toolchain consists of Prometheus, Grafana, Loki, Jaeger, and Alertmanager. It can serve as a practical monitoring framework for different PaaS environments.
Cloud computing resource management represents one of the most critical challenges in modern distributed systems, where efficient allocation of computational resources directly impacts system performance, energy consumption, and operational costs. This paper presents a novel hybrid approach combining deep reinforcement learning (DRL) with genetic algorithms (GA) for optimizing cloud resource scheduling in large-scale distributed environments. The proposed framework, termed Distributed Adaptive Learning Resource Scheduler (DALRS), integrates convolutional neural networks (CNN) for workload prediction with deep Q-networks (DQN) for dynamic resource allocation decisions. We evaluate our approach on a comprehensive simulation platform modeling realistic cloud infrastructure with heterogeneous resource configurations. Experimental results demonstrate that DALRS achieves 34.7% improvement in resource utilization, 28.3% reduction in task completion time, and 31.5% decrease in energy consumption compared to state-of-the-art baselines. Furthermore, the hybrid genetic algorithm component provides Pareto-optimal solutions balancing multiple objectives including cost, latency, and throughput. The paper also addresses scalability challenges through a distributed implementation using Apache Spark, enabling efficient processing of workloads exceeding 10,000 concurrent tasks. Our results validate the effectiveness of combining machine learning with evolutionary optimization for managing complex resource allocation problems in cloud infrastructure.
The aim of the study is to conduct a comparative analysis of HRV calculation methods and how they affect its measures (SDNN, SDNN Index, SDANN, RSA Index, Mean RR, RMSSD, pNN50) and their quality. Modern medical devices offer many possibilities, including heart rate and ECG measurement. An important issue in assessing heart function is HRV and its measures, which allow us to determine how the heart responds to different conditions, but the question of how to measure HRV and how its calculation works in different conditions remains open. The paper will show how HRV changes depending on frequency during rest, activity (walking), and sleep. HRV will be calculated using both an analytical method (using the Pan-Tompkins and U-Net algorithms) and a convolutional neural network.
This paper presents a static electrical model developed to analyze results from the Test Without Active Gases (TWAG) procedure, which characterizes fuel cell behavior in the absence of electrochemically active gases. The model topology is inspired by electric double-layer supercapacitor circuits and was derived from first principles using Lagrangian formalism. It was validated using five experimental TWAG discharge curves recorded at temperatures between 40°C and 120°C. Despite its simplicity and low computational cost, the model achieved satisfactory accuracy. The extracted parameters indicate potential for further refinement, such as introducing temperature-dependent components. The approach provides insight into the intrinsic electrochemical properties of high-temperature proton exchange membrane fuel cells in states without active gases and may serve as a foundation for broader diagnostic and modeling applications. Future developments may include extending the RC circuit, incorporating nonlinear elements, or applying the model to other fuel cell technologies. Testing on deliberately degraded cells could also help correlate model parameters with cell health.
For many entrepreneurs, the terms "supercomputers" or "High-Performance Computing" are primarily associated with academic institutions, theoretical research, or complex graphs understood only by a small group of experts. Some may also link them with services offered by Google or Amazon to businesses, but "available only to those with large budgets." This publication aims to demonstrate that supercomputers are increasingly becoming a part of the business landscape, and not necessarily just for the big companies. Small and medium-sized enterprises (SMEs), as well as startups, can successfully benefit from them. By combining a clear explanation of supercomputer capabilities with real-life case studies, the article aims to convince entrepreneurs that HPC is no longer a niche technology, but a real game-changer for SMEs across diverse sectors of the economy. We will discuss: • how the use of supercomputers can create a unique value proposition and expand the customer base, • how machine learning and generative AI can scale up business, • the path to financing access to supercomputers in Europe. In this paper, we will present two paths for applying for grants: one for entrepreneurs with no prior experience in HPC, and another for those looking to use supercomputers at a more advanced level. We will also provide essential information on the application process, key participation requirements, and the grant application schedule.
Coopetition is a strategy of cooperation between competing entities, with the primary goal of jointly creating value. The positive effects of coopetition may relate to costs, risks, economies of scale, research and development activities, as well as access to external knowledge and resources. Coopetition supports increased competitive advantage because it enables the creation of products and services that partners could not develop independently. The Polish National Competence Center (NCC) operates based on five independent supercomputing centers. This article presents the results of research confirming that the established collaboration is characterized by coopetition. The centers cooperate within the EuroCC project to form the Polish NCC, supporting one another in achieving the project's goals. The cooperation meets three primary conditions of coopetition: mutual dependence, mutual interests, and mutual benefits. It brings specific advantages such as increased innovation, shared competencies and resources, and enhanced competitiveness in the Polish market through joint marketing and training activities.
High-performance computing (HPC) has become a foundation for cutting-edge discovery and industrial innovation in Europe. The European High-Performance Computing Joint Undertaking (EuroHPC JU) funds and coordinates a continent-wide ecosystem of world-class systems, HPC software, skills, and services. Within this ecosystem, the National Competence Centre in HPC (NCC) in Poland consolidates the potential of six leading computing centers to serve researchers, enterprises - including SMEs - and public administration. This article introduces the strategic rationale of EuroHPC and the EuroCC program, explains how the Polish NCC is organized and funded, details its service portfolio, and documents representative scientific and industrial use cases. We conclude with forward-looking remarks on data-intensive research, and the role of competence centers in ensuring that compute power translates into measurable impact.
The EuroCC is a flagship project of the EuroHPC initiative intended to establish a network of National Competence Centers (NCCs) across participating member states. The Polish NCC has been dedicated to advancing skill sets and technical expertise in high-performance computing (HPC) among a broad user community. This work highlights NCC's main competence building efforts, their impact, and significance within the national HPC landscape.
High-performance computing (HPC) is a field of computer science focused on the use of highly efficient computer systems to perform complex calculations in a short time. In the article, we present the basic principles of HPC and explain how systems based on parallel computational architectures allow for the performance of multiple concurrent calculations. HPC technologies are crucial in various areas such as scientific research, industry, engineering, and data analysis. Applications include (but are not limited to) physics simulations, genome analyses, climate change forecasting or artificial intelligence research. The article also discusses the main building blocks of the HPC environment, such as supercomputers or software allowing parallel calculations and analyses of large data volumes. The main HPC challenges, including cost, scalability, and the need for specialist technical knowledge, are also considered. This article aims to present the reader with an overview of what HPC is, how it works, and what its current role is.
This article explores the opportunities available to Polish researchers thanks to access to the EuroHPC Joint Undertaking supercomputers. The EuroHPC JU initiative aims to create a unified high-performance computing (HPC) infrastructure across Europe, offering researchers access to powerful computational resources for scientific and industrial advancements. Polish researchers benefit from this initiative, gaining access to supercomputing resources to support all fields of science, industry, and the public sector. The article details the various types of EuroHPC JU access calls, including Extreme Scale Access calls for high-impact and high-gain innovative research, Regular Access calls for scientific innovation in respective domains, Development Access calls for code and algorithm development and optimization, Benchmark Access calls for code scalability or AI applications testing, and, leveraging the AI Factories network, the AI and Data-Intensive Applications Access call for ethical artificial intelligence, machine learning, and data-intensive applications. In particular, the article describes additional ways to get access to LUMI, one of the fastest supercomputers in Europe, provided via the PLGrid portal thanks to Poland’s participation in the LUMI consortium.
The paper presents the implementation of generative methods in classification tasks. A distinction is made between two types of tasks – supervised learning and unsupervised learning – along with example use cases. Within the scope of supervised methods, described are the Bayes classifier and the use of the multivariate Gaussian distribution. To solve the unsupervised learning task using a generative approach, the Gaussian Mixture Model (GMM) is presented. The paper also describes a generative neural network based on an autoencoder architecture, implemented as a Variational Autoencoder (VAE).
Deep reinforcement learning models such as Deep Q-Networks (DQNs) have achieved great performance in both simple and complex environments, but their decision-making process remains largely opaque. This work addresses the interpretability challenge by proposing a~method of extracting and comparing symbolic rules from a trained DQN and logic-based agents. The method is showcased in the popular Wumpus World domain. Rules extraction from the agents is done via training decision trees to mimic the agent's behavior. Comparison is done using Jaccard similarity and simple structural metrics. Results show that despite similar performance, DQNs and logic agents rely on partially overlapping but structurally largely distinct decision rules. This highlights the feasibility of translating subsymbolic policies into interpretable rules and reveals meaningful structural differences between learned and symbolic strategies.
Biological complexity emerges from the collective behaviour of proteins, whose folding and conformational dynamics are directly governed by the underlying potential energy surface. Here, we compare the potential energy landscapes of bovine pancreatic trypsin inhibitor (BPTI) obtained with the all-atom AMBER force field and the coarse-grained UNRES potential. For the native triply disulphide-bonded state of BPTI, both models sample predominantly folded, native-like conformations with relatively small structural deviations from the crystallographic structure, though UNRES explores structures with a broader range of RMSDs compared to experimental structure, radii of gyration, and solvent-accessible surface areas. Using comparable CPU time, UNRES generates a more diverse set of minima and transition states, yielding a more extensively sampled landscape with a globally similar topology. Together, these results highlight how all-atom and coarse-grained potentials differ in the representation of protein energy landscapes, while retaining consistent global features for a highly constrained, disulphide-rich protein such as BPTI.
Cyclodextrins (CDs) are cyclic oligosaccharides widely used as host molecules capable of forming inclusion complexes with a variety of guest compounds. In this study, for the first time, we investigated the potential interactions between α-, β -, and γ-CD and heparin (HP), a highly sulfated glycosaminoglycan, using a combined theoretical and experimen- tal approach. All-atom molecular dynamics (MD) simulations showed that HP does not enter the CD cavity, instead remaining adsorbed on the external surface at the secondary cavity of the CD. Free energy calculations using LIE and MM-GBSA confirmed that complex formation is enthalpically unfavorable in water, with desolvation penalties outweighing van der Waals and electrostatic contributions. The binding observed in the MD simulations could be, therefore, driven by the entropy. Potential of mean force (PMF) analysis further demonstrated that HP translocation through the γ-CD cavity requires overcoming a high energy barrier of ≈ 23 kcal/mol, indicating that the inclusion is not spontaneous. Complementary isothermal titration calorimetry (ITC) measurements in aqueous buffer also showed negligible enthalpy changes, suggesting that complex formation is not driven by detectable heat effects and is predomi- nantly entropically controlled. Overall, our findings highlight the limitations of natural cyclodextrins in encapsulating large, highly charged polysaccharides such as HP, emphasizing the need for their chemical modifications to enable effective host–guest complexation in such systems potentially designed to encapsulate this cargo
Graph Neural Networks (GNNs) have been increasingly adopted in modern, large-scale applications, such as social network analysis, recommendation systems, drug discovery, and more. However, the training cost of GNNs can be computationally prohibitive, especially when the graph is large and complex, necessitating the use of a mini-batching approach. In this paper, we propose a novel data structure called the Hierarchical Graph Adjacency Matrix (HGAM) to accelerate GNN training by avoiding redundant computations. With HGAM, we can accelerate the training speed of GNNs by up to four times. Additionally, we propose optimizations on top of HGAM to further enhance performance, achieving an overall speedup of up to 7.72 times for training 3-layer deep GNNs. We evaluated our techniques using three benchmark datasets—Reddit, ogbn-products, and ogbn-mag—and demonstrate that the proposed HGAM technique and related optimizations are advantageous for GNN training across modern hardware platforms.
