Solving the p-Laplace equation numerically is challenging due to inherent degeneracies and singularities. We present a new hybrid strategy combining Continuous Data Assimilation, Picard iteration with Anderson Acceleration, and Newton’s Method. By incorporating coarse-mesh observations via a nudging term, the scheme provides essential spatial regularization. Our results show that this integration stabilizes nonlinear iterations and significantly improves convergence speed. Notably, the CDA-enhanced solver captures localized singular behaviors and suppresses oscillations in highly nonlinear regimes where standard methods typically fail.
Understanding nonlinear feature interactions is crucial in science and engineering, yet standard multilayer perceptrons (MLPs) often capture such interactions only implicitly, leading to entangled representations that can impair robustness and interpretability. We investigate product-unit residual networks (PURe) that integrate multiplicative product units with residual connections to explicitly model cross-feature couplings while stabilizing optimization. We conduct a systematic evaluation on an interaction-driven synthetic benchmark and two real-world datasets, assessing predictive accuracy, robustness to Gaussian feature noise, and performance under limited training data, and we compare real- and complex-valued variants under a matched parameter budget. Beyond accuracy, SHapley Additive exPlanations (SHAP)-based interaction analyses show that PURe learns more concentrated and structurally coherent interaction patterns than MLP baselines. Overall, PURe achieves competitive or improved performance, better robustness and sample efficiency in low-data regimes, and enhanced interaction-level interpretability.
In CT angiography, the accurate segmentation of abdominal aortic aneurysms (AAAs) is difficult due to large anatomical variability, low-contrast vessel boundaries, and the close proximity of organs whose intensities resemble vascular structures, often leading to false positives. To address these challenges, we propose an anatomy-aware segmentation framework that integrates organ exclusion masks derived from TotalSegmentator into the training process. These masks encode explicit anatomical priors by identifying non-vascular organsand penalizing aneurysm predictions within these regions, thereby guiding the U-Net to focus on the aorta and its pathological dilation while suppressing anatomically implausible predictions. Despite being trained on a relatively small dataset, the anatomy-aware model achieves high accuracy, substantially reduces false positives, and improves boundary consistency compared to a standard U-Net baseline. The results demonstrate that incorporating anatomical knowledge through exclusion masks provides an efficient mechanism to enhance robustness and generalization, enabling reliable AAA segmentation even with limited training data.
Artificial Intelligence (AI) has emerged as a powerful and effective tool with several applications in health science. An inherent drawback of drug therapies is the potential for side effects, which are adverse reactions that negatively impact human health. In recent years, AI has been applied in pharmacology and pharmacovigilance, e.g., for studying and analysing drug side effects. Likewise, network science has become widely used as an effective and efficient tool for modelling interactions between biological objects. In this paper, we presented a framework for predicting candidate drug side effects by using Machine Learning (ML) techniques applied to biological multilayer networks. Experimentation supports the application of the ML-based models implemented in the proposed framework for predicting novel (candidate) drug side effects from biological multilayer networks.
Linear elasticity and Navier-Stokes equations are fundamental tools in material science, enabling the modeling of solid deformations and fluid flows under various conditions. These equations are widely used to simulate stresses, strains, and fluid interactions in processes like 3D printing, welding, casting, and extrusion. Physics-Informed Neural Networks (PINNs), introduced in 2019, have gained significant attention for solving complex physical problems, including fluid mechanics, wave propagation, and inverse problems. Despite their growing popularity, PINNs face challenges in training efficiency and accuracy. This paper investigates the applicability of modern PINN methodologies to material science problems involving Navier-Stokes and linear elasticity equations. For linear elasticity, a randomized selection of collocation points is employed to enhance training. For Navier-Stokes equations, hard constraints on initial and boundary conditions are implemented to avoid multi-objective optimization. These approaches aim to address training difficulties and improve PINN performance in simulating material science phenomena.
We present the preliminary results of a Physics-Informed Neural Network (PINN) for a Wigner-Fokker-Planck (WFP) equation modeling open quantum systems such as electron transport in semiconductors. The WFP equation is a mathematical model that considers diffusion and friction introduced by the environment into an open quantum (sub-)system, describing the problem via a continuous quantum variable formulation. Recent developments in scientific machine learning have demonstrated that PINNs are useful in providing data-driven solutions to partial differential equations (PDE) and for data-driven discovery [20] in the estimation of model parameters, particularly when constrained to small or noisy data. PINNs minimize a residual additional to the typical Neural Network approach related to the satisfaction of a PDE that represents the “Physics” of the problem, along with the traditional loss function that estimates the fit to the data. The former residual implicitly trains the model to respect conservation principles following from the PDE that models the Physics phenomena. Since optimization is not solely dependent on minimizing the fit to the training data, such as in traditional machine learning models, it can perform exceedingly well for inverse problems to estimate model parameters, such as the diffusion and friction parameters in the particular case of our Wigner-Fokker-Planck model, when constrained to small data. This work used the PINN methodology to solve a data-driven discovery problem for the Wigner-Fokker-Planck equation. In particular, we solved an inverse problem with synthetic data obtained from a Monte Carlo forward solver of the Wigner-Fokker-Planck equation to estimate the elements of our diffusion matrix as parameters of the model representing noise introduced by the environment into the open quantum system.
This study examines the effectiveness of visual navigation cues in guiding player movement within 3D virtual environments. Using experimental levels with lighting contrast and object geometry cues at three intensity levels, the research identifies the minimum cue strength needed to influence player navigation effectively. Differences in cue responsiveness between casual and advanced players were also analyzed. The findings reveal threshold values for subtle yet effective navigation methods, providing practical insights for level designers. These results can enhance the development of user-friendly and engaging game environments, catering to players of varying skill levels.
Replication involves creating and managing multiple duplicates of objects on various storage devices, workers, or computers. Replication is widely used in RAID (Redundant Array of Independent Disks) systems, database systems, storage clusters, cloud storage networks, and compute clusters, among others. The primary goal of duplicating or replicating data is to enhance its availability, thereby improving fault tolerance and performance. The replication factor (degree of duplication, multiplicity) determines the number of copies or replicas of objects, indicating the level of redundancy and thus giving a guarantee. In many cases, such as multi-disk storage, distributed file systems, and database systems, policies fix the replication factor to a global value such as 3, for example, to balance storage expenses and performance. This research presents a global search using an evolutionary optimization to find a set of partial replications, combining them to fit the objects into the system and further reduce costs.
Despite the critical role played by power grids in smart cities, traditional systems often detect issues too late leading to costly outages and safety risks. This paper presents a novel approach by leveraging graph-based digital twins created with the help of Microsoft Azure IoT TwinMaker to enable an always-on view of real-time power grids. For one to derive a virtual model of this grid that reflects what happens in its counterpart in the physical plane, digital twins have this application. Simulation of the grid’s behavior helps reveal existing problems to operators without affecting the infrastructure. The system incorporates advanced machine learning techniques, including a genetic algorithm, which can predict failures in advance by taking realtime and previously recorded data from the digital twin. This approach is proactive, thereby preventing disruptions and operational risks. It ensures scalability and efficient processing of large datasets when running machine learning models in the cloud on Microsoft Azure, thus adapting the system to the complex needs of modern power grids in smart cities. Using machine learning in the digital twin rather than in the physical grid minimizes real-world testing and avoids unanticipated downtime, thereby saving costs. This approach not only enhances the accuracy of fault detection but also improves the lasting ability and steadiness of the grid. The solution aims to support the rising energy demands of smart cities by offering a scalable, cost-effective, and reliable way of managing power grids.
In psychological research, network models are widely used to study symptoms of mental health disorders. However, these models often fail to account for uncertainty, leading to potentially misleading inferences. To address this issue, this study examines the robustness of psychological networks by analyzing a dataset of risk factors for suicidal behavior with multiple network algorithms. We compare two causal discovery algorithms—Hill Climbing (HC) and TABU search—and the Gaussian Graphical Model (GGM), a widely used statistical network model in psychology. Uncertainty is assessed along two dimensions: (1) the impact of noise, by introducing varying levels of white noise into the dataset, and (2) the effect of sample size reduction, by systematically decreasing the number of observations. Our results indicate that both HC and TABU search are highly sensitive to noise and sample size, with HC slightly outperforming TABU in terms of precision and recall. GGM performance declines gradually with increasing noise and sample size reduction, leading to sparser networks. For all algorithms, recall declined at a faster rate than precision. Finally, we examine the robustness of edges leading to suicidal ideation, finding that the edge from Depression to suicidal ideation remains relatively stable across conditions. This is a promising result, since many suicide interventions are based on treating depressive mood. Our results emphasize the importance of considering uncertainty in network-based psychological research, particularly when applying causal discovery algorithms.
Medical image analysis often relies on models optimized for specific tasks, such as classification or detection. However, this single-task approach limits the utilization of shared features across tasks and becomes particularly inefficient when data availability is limited. In this paper, we investigate the potential of multi-task learning (MTL) with Visual Transformers to optimize both classification and reconstruction tasks, especially when training data is scarce. Using datasets like BRATS (binary classification of brain tumor presence on MRI slices) and ultrasound images of muscles (for identifying pathologies such as Myopathy, Myelopathy, and Polyneuropathy), we evaluate MTL against standalone and pre-training paradigms. Results indicate that MTL significantly enhances model performance, particularly in classification tasks, by leveraging shared representations and improving attention mechanisms. Our findings demonstrate that MTL mitigates the challenges of limited data availability by effectively transferring knowledge between tasks, making it a valuable strategy for medical imaging applications.
This study introduces a computational model designed to simulate the human immune response to SARS-CoV-2, validated against data from multiple clinical studies. The model captures the temporal dynamics of mature CD4 ^+ T cells, mature CD8 ^+ T cells, viral load, and antibody levels across three COVID-19 severity profiles: mild, severe, and critical. In all simulated scenarios, the model-generated trajectories remained primarily within the confidence intervals of empirical data, demonstrating its capacity to qualitatively reproduce key trends in immune responses across varying disease severities.
Designing modern systems requires technical knowledge and sustainability competencies from engineers, as almost every system or product designed has a sustainability impact. This gained additional significance after adopting the Sustainability Development Goals in 2015 and their subsequent integration into the activities of enterprises and other entities and the products and services they offer. For this reason, the study’s main goal was to determine the role of sustainability competences in activities performed by systems engineers. Sustainability competences and systems engineering competencies were analyzed. A significant overlap in many areas was found. Subsequently, the perception of these competencies by representatives of enterprises was examined. The results indicate that system engineers’ perception of sustainability competences is not uniform, particularly depending on the size of the enterprise and position in the company hierarchy. Engineering team leaders and management pay more attention to soft skills. The study’s results may help understand the relationships between competencies, support the development of HR processes, and entities offering training programs for employees at different career stages.
This study explores a formalized dynamical systems model of the General Escape Theory of Suicide using Sobol and PAWN global sensitivity analyses. The findings highlight the importance of self-feedback loops, the effect of stressors on aversive internal states, and the interaction effects between aversive internal states and the urge to escape on suicidal ideation and non-suicidal escape behaviors. Time-dependent sensitivity analysis also reveals the long-term stability of parameter importance over time. These results hold potential for informing clinical interventions by identifying the most important influences for individual suicidal ideation.
This paper introduces a novel, unconventional method to enhance the visual clarity of monochrome images derived from non-visual data sources. The approach involves a two-step process: image pseudo-colorization followed by decolorization. Surprisingly, this counterintuitive technique can significantly improve discernibility of image features, irrespective of their size, shape, or original visual prominence. The paper delves into the algorithmic details of this method and presents experimental results on a representative dataset of IR, X-ray, MRI, and ultrasound images. When disregarding factors related to natural image appearance (which are irrelevant in non-visual domains), this method outperforms conventional image enhancement techniques, including sophisticated ones, in terms of standard image quality criteria, i.e., sharpness, contrast, and overall detail perceptibility. This superiority is substantiated by both subjective evaluations and objective metrics. The success of this technique hinges on the careful selection of color maps and the application of a specific, recently proposed decolorization scheme. The technique is well-suited for various visual data analysis tasks in non-visual domains, primarily in AI-based solutions.
The issue of adjusting neural network structure is one of the core problems in artificial intelligence. The issue of adjusting neural network structure is one of the core problems in artificial intelligence. A highly desirable scenario is a dynamic architecture that evolves structurally during the training process. In this paper, we propose a new and powerful tool that facilitates dynamic changes in network structure. We introduce a novel form of residual connections based on matrix extensions, enabling adaptable weight matrices and enhancing structural flexibility. The approach enhance the potential for structural modifications. We conducted a series of comprehensive experiments confirming that the new residual connections scheme behaves very well. The new type of connection improves performance by enabling better error flow during the error backpropagation phase, resulting in more efficient training. Our method demonstrates superior performance and enhanced trackability during the training process. The paper is supplemented by Python source code to ensure reproducibility. This method marks a significant starting point, showing immense potential for more advanced dynamic neural network models and transfer learning with dynamic models.
In this article, the authors present three didactic examples of using Chat GPT in mathematical programming optimization tasks: linear programming, nonlinear programming and convex programming. These examples are analyzed in terms of the correctness of the methods used and the solutions obtained. The article also describes a didactic experiment with the participation of Informatics and Econometrics students of Warsaw University of Life Sciences, consisting in solving optimization tasks on their own using Chat GPT. The final conclusions of the article also present a comparison of the approach based on the CAS methodology and the approach using Chat GPT.
This article examines recent advancements in the use of Artificial Intelligence (AI) for automation and robotics, highlighting efficiency gains across sectors. It reviews key trends, including robotic process automation in different areas and compares AI implementation frameworks such as MLOps and DevOps. The study presents educational approaches to teaching automation at Warsaw University of Technology, detailing project structures, tools and practical outcomes. Quantitative results on time savings are discussed in relation to existing research. The paper concludes with insights on incorporating AI into education to better prepare students for the digital future.
This research explores the impact of structural limiting perception on the performance of Particle Swarm Optimization by restricting the range of information sharing among particles. By introducing localized communication models through Ring and Tree topologies, the study demonstrates significant improvements over the standard global-best PSO, particularly on a range of Traveling Salesman Problem instances from the TSPLIB. The results show that constraining particle perception enhances both solution quality and convergence behavior, with the Tree topology emerging as the most effective structure. The topological modifications maintain swarm diversity, prevent premature convergence, and facilitate continuous exploration while exploiting promising search regions. These findings suggest that structural constraints on information sharing can enhance PSO’s robustness and effectiveness without adding computational complexity, offering a flexible approach applicable to various PSO variants and problem domains beyond TSP.
The purpose of the article is to develop a new dimensionality reduction algorithm for data that are described by many features of different nature. A method of feature selection is based on a new concept of metrical importance of the features. The concept of feature importance is based on metrical properties of data and is inspired by the principle component analysis. Numerical experiments confirm the effectiveness of the method and certain accordance of it with other concepts of feature importance.
