Pub Date : 2025-10-08DOI: 10.1016/j.jocs.2025.102713
Ramraj Thirupathyraj
Network Science, delving into complex networks with intricate topologies and structural interactions, plays a pivotal role in understanding various natural systems. Computational studies highlight the importance of influential nodes in capturing network characteristics and functionalities. Previous research underscores the inadequacy of relying on a single node characteristic to identify influence, emphasizing the need for integrating multiple characteristics. In this study, we propose an indicator by incorporating the network’s topological features into the Krylov subspace to effectively capture influence propagation among nodes and their neighbors. This new indicator, in an asymmetric form, considers distinct node influence effects and inherent dynamics asymmetry. Furthermore, when integrated with other locality-based measures, it enhances the cohesion of a unified model. This model is employed to identify influential nodes within complex networks. Empirical evaluations of Susceptible–Infected–Recovered (SIR) propagation dynamics across ten authentic networks demonstrate that our proposed unified model operates within polynomial time and surpasses numerous traditional methods in terms of accuracy. Utilizing this approach to identify influential nodes offers potential applications across a range of domains, such as social networks, malware analysis, and neuro-perception networks.
{"title":"Detecting cardinal nodes in unweighted complex networks by examining their trajectories within Krylov subspace and various topological features","authors":"Ramraj Thirupathyraj","doi":"10.1016/j.jocs.2025.102713","DOIUrl":"10.1016/j.jocs.2025.102713","url":null,"abstract":"<div><div>Network Science, delving into complex networks with intricate topologies and structural interactions, plays a pivotal role in understanding various natural systems. Computational studies highlight the importance of influential nodes in capturing network characteristics and functionalities. Previous research underscores the inadequacy of relying on a single node characteristic to identify influence, emphasizing the need for integrating multiple characteristics. In this study, we propose an indicator by incorporating the network’s topological features into the Krylov subspace to effectively capture influence propagation among nodes and their neighbors. This new indicator, in an asymmetric form, considers distinct node influence effects and inherent dynamics asymmetry. Furthermore, when integrated with other locality-based measures, it enhances the cohesion of a unified model. This model is employed to identify influential nodes within complex networks. Empirical evaluations of Susceptible–Infected–Recovered (SIR) propagation dynamics across ten authentic networks demonstrate that our proposed unified model operates within polynomial time and surpasses numerous traditional methods in terms of accuracy. Utilizing this approach to identify influential nodes offers potential applications across a range of domains, such as social networks, malware analysis, and neuro-perception networks.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102713"},"PeriodicalIF":3.7,"publicationDate":"2025-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265400","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-08DOI: 10.1016/j.jocs.2025.102726
Burhaneddin İzgi , Murat Özkaya , Nazım Kemal Üre , Matjaž Perc
Existing methodologies for solving Markov reward games mostly rely on state–action frameworks and iterative algorithms to address these challenges. However, these approaches often impose significant computational burdens, particularly when applied to large-scale games, due to their inherent complexity and the need for extensive iterative calculations. In this paper, we propose a new neural network architecture for solving Markov reward games in the form of a decision tree with relatively large state and action sets, such as 2-actions-3-stages, 3-actions-3-stages, and 4-actions-3-stages, by trimming the decision tree. In this context, we generate datasets of Markov reward games with sizes ranging from to using the holistic matrix norm-based solution method and obtain the necessary components, such as the payoff matrices and the corresponding solutions of the games, for training the neural network. We then propose a vectorization process to prepare the outcomes of the matrix norm-based solution method and adapt them for training the proposed neural network. The neural network is trained using both the vectorized payoff and transition matrices as input, and the prediction system generates the optimal strategy set as output. In the model, we approach the problem as a classification task by labeling the optimal and non-optimal branches of the decision tree with ones and zeros, respectively, to identify the most rewarding paths of each game. As a result, we propose a novel neural network architecture for solving Markov reward games in real time, enhancing its practicality for real-world applications. The results reveal that the system efficiently predicts the optimal paths for each decision tree, with f1-scores slightly greater than 0.99, 0.99, and 0.97 for Markov reward games with 2-actions-3-stages, 3-actions-3-stages, and 4-actions-3-stages, respectively.
{"title":"Machine learning tree trimming for faster Markov reward game solutions","authors":"Burhaneddin İzgi , Murat Özkaya , Nazım Kemal Üre , Matjaž Perc","doi":"10.1016/j.jocs.2025.102726","DOIUrl":"10.1016/j.jocs.2025.102726","url":null,"abstract":"<div><div>Existing methodologies for solving Markov reward games mostly rely on state–action frameworks and iterative algorithms to address these challenges. However, these approaches often impose significant computational burdens, particularly when applied to large-scale games, due to their inherent complexity and the need for extensive iterative calculations. In this paper, we propose a new neural network architecture for solving Markov reward games in the form of a decision tree with relatively large state and action sets, such as 2-actions-3-stages, 3-actions-3-stages, and 4-actions-3-stages, by trimming the decision tree. In this context, we generate datasets of Markov reward games with sizes ranging from <span><math><mrow><mn>1</mn><msup><mrow><mn>0</mn></mrow><mrow><mn>3</mn></mrow></msup></mrow></math></span> to <span><math><mrow><mn>1</mn><msup><mrow><mn>0</mn></mrow><mrow><mn>5</mn></mrow></msup></mrow></math></span> using the holistic matrix norm-based solution method and obtain the necessary components, such as the payoff matrices and the corresponding solutions of the games, for training the neural network. We then propose a vectorization process to prepare the outcomes of the matrix norm-based solution method and adapt them for training the proposed neural network. The neural network is trained using both the vectorized payoff and transition matrices as input, and the prediction system generates the optimal strategy set as output. In the model, we approach the problem as a classification task by labeling the optimal and non-optimal branches of the decision tree with ones and zeros, respectively, to identify the most rewarding paths of each game. As a result, we propose a novel neural network architecture for solving Markov reward games in real time, enhancing its practicality for real-world applications. The results reveal that the system efficiently predicts the optimal paths for each decision tree, with f1-scores slightly greater than 0.99, 0.99, and 0.97 for Markov reward games with 2-actions-3-stages, 3-actions-3-stages, and 4-actions-3-stages, respectively.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102726"},"PeriodicalIF":3.7,"publicationDate":"2025-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265399","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-07DOI: 10.1016/j.jocs.2025.102728
Yurun Ge , Lucas Böttcher , Tom Chou , Maria R. D’Orsogna
Allocating limited resources to a set of alternatives with uncertain long-term benefits is a common challenge in innovation management, research funding, and participatory budgeting. Related problems arise in emerging applications such as ranking outputs of large language models and coordinating decisions in agentic systems. All settings include multiple agents tasked with estimating the true value of a potentially large number of alternatives. These estimates, or quantities derived from them, are then aggregated to select a final portfolio that maximizes overall benefit, ideally using efficient methods. Standard sorting algorithms are ill-suited as they do not account for uncertainties associated with each agent’s estimate. Furthermore, the cognitive load on agents can be demanding, especially if the number of alternatives to evaluate is large. Building on the Quicksort algorithm and the Bradley–Terry model, we develop four new, efficient aggregation protocols based on agent-assigned win probabilities of pairwise comparisons that are then globally aggregated. The pairwise comparisons we introduce not only reduce cognitive load on agents, but lead to aggregation protocols that outperform existing ones, which we confirm via numerical simulations. Our methods can be combined with sampling strategies to further reduce the number of pairwise comparisons.
{"title":"Efficient portfolio selection through preference aggregation with Quicksort and the Bradley–Terry model","authors":"Yurun Ge , Lucas Böttcher , Tom Chou , Maria R. D’Orsogna","doi":"10.1016/j.jocs.2025.102728","DOIUrl":"10.1016/j.jocs.2025.102728","url":null,"abstract":"<div><div>Allocating limited resources to a set of alternatives with uncertain long-term benefits is a common challenge in innovation management, research funding, and participatory budgeting. Related problems arise in emerging applications such as ranking outputs of large language models and coordinating decisions in agentic systems. All settings include multiple agents tasked with estimating the true value of a potentially large number of alternatives. These estimates, or quantities derived from them, are then aggregated to select a final portfolio that maximizes overall benefit, ideally using efficient methods. Standard sorting algorithms are ill-suited as they do not account for uncertainties associated with each agent’s estimate. Furthermore, the cognitive load on agents can be demanding, especially if the number of alternatives to evaluate is large. Building on the Quicksort algorithm and the Bradley–Terry model, we develop four new, efficient aggregation protocols based on agent-assigned win probabilities of pairwise comparisons that are then globally aggregated. The pairwise comparisons we introduce not only reduce cognitive load on agents, but lead to aggregation protocols that outperform existing ones, which we confirm via numerical simulations. Our methods can be combined with sampling strategies to further reduce the number of pairwise comparisons.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102728"},"PeriodicalIF":3.7,"publicationDate":"2025-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265398","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-04DOI: 10.1016/j.jocs.2025.102721
Achraf Zinihi , Moulay Rchid Sidi Ammi , Ahmed Bachir
This paper presents a computationally efficient hybrid approach for multi-city epidemic modeling, utilizing a topology-based SIR model for individual cities coupled via empirical transportation networks to account for migration between them. Within each city, the epidemiological dynamics are described using an SAIRD model. This study introduces two key innovations: the self-consistent determination of coupling parameters to maintain the populations of individual cities, and the incorporation of distance-dependent temporal delays in migration. Our model is applied to China’s 3 populated cities. The results demonstrate the model’s effectiveness in capturing the complex dynamics of epidemic spread across multiple urban centers.
{"title":"Multi-city modeling of epidemics using a topology-based SIR model: Neural network-enhanced SAIRD model","authors":"Achraf Zinihi , Moulay Rchid Sidi Ammi , Ahmed Bachir","doi":"10.1016/j.jocs.2025.102721","DOIUrl":"10.1016/j.jocs.2025.102721","url":null,"abstract":"<div><div>This paper presents a computationally efficient hybrid approach for multi-city epidemic modeling, utilizing a topology-based SIR model for individual cities coupled via empirical transportation networks to account for migration between them. Within each city, the epidemiological dynamics are described using an SAIRD model. This study introduces two key innovations: the self-consistent determination of coupling parameters to maintain the populations of individual cities, and the incorporation of distance-dependent temporal delays in migration. Our model is applied to China’s 3 populated cities. The results demonstrate the model’s effectiveness in capturing the complex dynamics of epidemic spread across multiple urban centers.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102721"},"PeriodicalIF":3.7,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265396","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-04DOI: 10.1016/j.jocs.2025.102722
Abhijnan Dikshit, Leifur Leifsson
Composite Bayesian optimization (CBO) methods are attractive methods for black-box optimization problems. Though CBO methods offer significant benefits, extending CBO to high-dimensional input and output spaces has been less explored. The limited scalability and accuracy of multi-output Gaussian process (GP) models makes them less attractive for engineering design problems. Standard neural network-based models provide an alternative, but require the implementation of expensive and complex uncertainty quantification methods to enable CBO. As such, this paper develops Bayesian optimization using non-intrusive reduced-order models (ROMBO), a framework for high-dimensional CBO using deep learning reduced-order models. The framework utilizes autoencoders to create a nonlinear embedding of the output space that is modeled using a multi-task GP model. A Monte Carlo expected improvement acquisition function is used to balance exploration of the design space and exploitation of the composite objective function. The proposed framework is characterized using three synthetic problems and an inverse design problem for a transonic airfoil. It is compared with a standard BO implementation and a CBO implementation that generates an embedding of the outputs using proper orthogonal decomposition (POD). The results demonstrate that the ROMBO framework can achieve up to one to four orders of magnitude lower objective function values as compared to the other two methods. Additionally, ROMBO is more sample efficient than the other two methods, achieving far lower objective function values in fewer sampling iterations. This work demonstrates that ROMBO is a promising framework for enabling the use of CBO for complex high-dimensional design problems.
{"title":"A scalable composite Bayesian optimization framework for engineering design using deep learning reduced-order models","authors":"Abhijnan Dikshit, Leifur Leifsson","doi":"10.1016/j.jocs.2025.102722","DOIUrl":"10.1016/j.jocs.2025.102722","url":null,"abstract":"<div><div>Composite Bayesian optimization (CBO) methods are attractive methods for black-box optimization problems. Though CBO methods offer significant benefits, extending CBO to high-dimensional input and output spaces has been less explored. The limited scalability and accuracy of multi-output Gaussian process (GP) models makes them less attractive for engineering design problems. Standard neural network-based models provide an alternative, but require the implementation of expensive and complex uncertainty quantification methods to enable CBO. As such, this paper develops Bayesian optimization using non-intrusive reduced-order models (ROMBO), a framework for high-dimensional CBO using deep learning reduced-order models. The framework utilizes autoencoders to create a nonlinear embedding of the output space that is modeled using a multi-task GP model. A Monte Carlo expected improvement acquisition function is used to balance exploration of the design space and exploitation of the composite objective function. The proposed framework is characterized using three synthetic problems and an inverse design problem for a transonic airfoil. It is compared with a standard BO implementation and a CBO implementation that generates an embedding of the outputs using proper orthogonal decomposition (POD). The results demonstrate that the ROMBO framework can achieve up to one to four orders of magnitude lower objective function values as compared to the other two methods. Additionally, ROMBO is more sample efficient than the other two methods, achieving far lower objective function values in fewer sampling iterations. This work demonstrates that ROMBO is a promising framework for enabling the use of CBO for complex high-dimensional design problems.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102722"},"PeriodicalIF":3.7,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265402","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-03DOI: 10.1016/j.jocs.2025.102727
Ikram Ul Haq , Saira Shukat , Ikram Ullah , Waqar Ul Hassan , Hong-Na Zhang , Xiao-Bin Li , Feng-Chen Li
The magnetohydrodynamic (MHD) micropolar nanofluid with stratification is evaluated in this work by integrated numerical computing using the Levenberg Marquardt backpropagation (LMBB) optimization technique, an artificial neural network (ANN) approach. After that, model is condensed to a set of problems with boundary values, which are resolved utilizing the proposed method LMBB algorithm and a numerical technique BVP4c. The LMBB approach is an iterative approach for figuring out the least of a function that is not linear, is distinct as the addition of squares. The outcomes are also cross-checked against those of earlier studies and the MATLAB’s BVP4c solver for validation. The mapping of velocity, concentration and temperature profiles from the input to results is another use of neural networking. These results show the accuracy level of the predictions and improvements made by ANN. To generalize a dataset, the BVP4c techniques’ performance is utilized to lower error of mean square. Data based on the ratio of training (80 %), validation (10 %) and testing (10 %) is used by the ANN-based LMBB backpropagation optimization technique. Histograms and function fitness are utilized to verify the algorithm’s dependability. For fluid dynamics, numerical methods and ANN perform incredibly well together, and this could result in new developments across a wide range of fields. The results of this study may aid in the optimization of fluid systems, leading to higher productivity and efficiency in a range of engineering applications.
{"title":"Intelligent computing for magnetohydrodynamic micropolar nanofluid with stratification using Levenberg–Marquardt backpropagation algorithm","authors":"Ikram Ul Haq , Saira Shukat , Ikram Ullah , Waqar Ul Hassan , Hong-Na Zhang , Xiao-Bin Li , Feng-Chen Li","doi":"10.1016/j.jocs.2025.102727","DOIUrl":"10.1016/j.jocs.2025.102727","url":null,"abstract":"<div><div>The magnetohydrodynamic (MHD) micropolar nanofluid with stratification is evaluated in this work by integrated numerical computing using the Levenberg Marquardt backpropagation (LMBB) optimization technique, an artificial neural network (ANN) approach. After that, model is condensed to a set of problems with boundary values, which are resolved utilizing the proposed method LMBB algorithm and a numerical technique BVP4c. The LMBB approach is an iterative approach for figuring out the least of a function that is not linear, is distinct as the addition of squares. The outcomes are also cross-checked against those of earlier studies and the MATLAB’s BVP4c solver for validation. The mapping of velocity, concentration and temperature profiles from the input to results is another use of neural networking. These results show the accuracy level of the predictions and improvements made by ANN. To generalize a dataset, the BVP4c techniques’ performance is utilized to lower error of mean square. Data based on the ratio of training (80 %), validation (10 %) and testing (10 %) is used by the ANN-based LMBB backpropagation optimization technique. Histograms and function fitness are utilized to verify the algorithm’s dependability. For fluid dynamics, numerical methods and ANN perform incredibly well together, and this could result in new developments across a wide range of fields. The results of this study may aid in the optimization of fluid systems, leading to higher productivity and efficiency in a range of engineering applications.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102727"},"PeriodicalIF":3.7,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265397","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.jocs.2025.102724
Kuo-Ching Ying , Pourya Pourhejazy , Shih-Cheng Lin
Scheduling problems predominantly assume that the same operators work fixed shifts during the day and night. Scheduling with a single-shift approach can result in infeasible or suboptimal production planning solutions when a multiple-shift system is implemented. This study introduces a new scheduling extension that incorporates shift work constraints. A new mathematical model based on the Permutation Flowshop Scheduling Problem is proposed, and the Iterated Greedy algorithm is adapted to solve it. The objective is to minimize the maximum completion time (makespan) and thereby improve the system performance while considering shift work constraints. Experiments reveal that the overall response time in 10-hour and 12-hour shifts is better than that of 8-hour shifts, despite the shorter overall active hours on the shop floor. Additional experiments confirm that the proposed Adjusted Iterated Greedy algorithm outperforms the Variable Neighbourhood Search algorithm in solving medium- and large-scale problems.
{"title":"An exploration of the shift work consideration in production scheduling","authors":"Kuo-Ching Ying , Pourya Pourhejazy , Shih-Cheng Lin","doi":"10.1016/j.jocs.2025.102724","DOIUrl":"10.1016/j.jocs.2025.102724","url":null,"abstract":"<div><div>Scheduling problems predominantly assume that the same operators work fixed shifts during the day and night. Scheduling with a single-shift approach can result in infeasible or suboptimal production planning solutions when a multiple-shift system is implemented. This study introduces a new scheduling extension that incorporates shift work constraints. A new mathematical model based on the Permutation Flowshop Scheduling Problem is proposed, and the Iterated Greedy algorithm is adapted to solve it. The objective is to minimize the maximum completion time (makespan) and thereby improve the system performance while considering shift work constraints. Experiments reveal that the overall response time in 10-hour and 12-hour shifts is better than that of 8-hour shifts, despite the shorter overall active hours on the shop floor. Additional experiments confirm that the proposed Adjusted Iterated Greedy algorithm outperforms the Variable Neighbourhood Search algorithm in solving medium- and large-scale problems.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102724"},"PeriodicalIF":3.7,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145219534","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-20DOI: 10.1016/j.jocs.2025.102717
Ting Wang , Jaroslaw Knap
Scientific machine learning has become an increasingly important tool in materials science and engineering. It is particularly well suited to tackle material problems involving many variables or to allow rapid construction of surrogates of material models, to name just a few. Mathematically, many problems in materials science and engineering can be cast as variational problems. However, handling of uncertainty, ever present in materials, in the context of variational formulations remains challenging for scientific machine learning. In this article, we propose a deep-learning-based numerical method for solving variational problems under uncertainty. Our approach seamlessly combines deep-learning approximation with Monte Carlo sampling. The resulting numerical method is powerful yet remarkably simple. We assess its performance and accuracy on a number of variational problems.
{"title":"Stochastic deep-Ritz for parametric uncertainty quantification","authors":"Ting Wang , Jaroslaw Knap","doi":"10.1016/j.jocs.2025.102717","DOIUrl":"10.1016/j.jocs.2025.102717","url":null,"abstract":"<div><div>Scientific machine learning has become an increasingly important tool in materials science and engineering. It is particularly well suited to tackle material problems involving many variables or to allow rapid construction of surrogates of material models, to name just a few. Mathematically, many problems in materials science and engineering can be cast as variational problems. However, handling of uncertainty, ever present in materials, in the context of variational formulations remains challenging for scientific machine learning. In this article, we propose a deep-learning-based numerical method for solving variational problems under uncertainty. Our approach seamlessly combines deep-learning approximation with Monte Carlo sampling. The resulting numerical method is powerful yet remarkably simple. We assess its performance and accuracy on a number of variational problems.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102717"},"PeriodicalIF":3.7,"publicationDate":"2025-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145118610","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-18DOI: 10.1016/j.jocs.2025.102720
Xiaohan Jing , Lin Qiu , Hong Zhao , Zeqian Zhang , Yaoming Zhang , Yan Gu
In this study, an accurate and stable space-time radial basis function (STRBF) collocation method is developed to solve two- and three-dimensional dynamic coupled thermo-mechanical problems. The proposed method enhances numerical precision by strategically positioning source points beyond the computational domain through space-time scaling factors. To address the challenge of selecting the optimal shape parameter, a new coupled STRBF is formulated by combining the Multiquadric function with the conical spline. Furthermore, a multiscale computational strategy is implemented to mitigate numerical instability in the resulting linear system. The effectiveness of the developed approach is demonstrated through four numerical examples involving complex geometries and different initial and boundary conditions. Numerical results show that, compared to the traditional RBF collocation method, the developed scheme not only enhances computational accuracy but also significantly reduces the dependence on the choice of shape parameter, making it a promising method for dealing with transient coupled thermo-mechanical problems.
{"title":"An accurate and stable space-time radial basis function collocation method for transient coupled thermo-mechanical analysis","authors":"Xiaohan Jing , Lin Qiu , Hong Zhao , Zeqian Zhang , Yaoming Zhang , Yan Gu","doi":"10.1016/j.jocs.2025.102720","DOIUrl":"10.1016/j.jocs.2025.102720","url":null,"abstract":"<div><div>In this study, an accurate and stable space-time radial basis function (STRBF) collocation method is developed to solve two- and three-dimensional dynamic coupled thermo-mechanical problems. The proposed method enhances numerical precision by strategically positioning source points beyond the computational domain through space-time scaling factors. To address the challenge of selecting the optimal shape parameter, a new coupled STRBF is formulated by combining the Multiquadric function with the conical spline. Furthermore, a multiscale computational strategy is implemented to mitigate numerical instability in the resulting linear system. The effectiveness of the developed approach is demonstrated through four numerical examples involving complex geometries and different initial and boundary conditions. Numerical results show that, compared to the traditional RBF collocation method, the developed scheme not only enhances computational accuracy but also significantly reduces the dependence on the choice of shape parameter, making it a promising method for dealing with transient coupled thermo-mechanical problems.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102720"},"PeriodicalIF":3.7,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145158211","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-18DOI: 10.1016/j.jocs.2025.102719
Paraskevas Dimitriou, Vasileios Karyotis
Link prediction is a fundamental task in network analysis, aiming at predicting missing or future connections between nodes in a network. With the growing availability of complex network data in fields like social networks, biological systems, the Internet, and scientific collaboration networks, accurate link prediction methods are becoming increasingly critical. Neighborhood or graph based link prediction algorithms are applied identically to different types of networks so that any differences in their structures are not exploited efficiently. Machine or deep learning based link prediction algorithms apply to each kind of network differently depending on the type of network, due to the unique characteristics of each domain, but frequently, most of them give poor results. In this paper, we propose a novel approach for link prediction, leveraging the power of machine learning and evolutionary algorithms. Our method utilizes local network information by encoding the network topology into link embeddings through a heuristic machine learning architecture. We introduce a novel tool to extract features from network structure effectively and combine them in an effective way through an evolutionary algorithm improving the discriminative power of link embeddings. We evaluate our method on eleven benchmark datasets and demonstrate its superior performance compared to a series (eleven in total) of effective and state-of-the-art algorithms. Our approach advances the state-of-the-art in link prediction yielding better results than other methods in all the networks we have applied it to.
{"title":"Heuristic Custom Similarity Index (HCSI): A novel machine learning approach for link prediction","authors":"Paraskevas Dimitriou, Vasileios Karyotis","doi":"10.1016/j.jocs.2025.102719","DOIUrl":"10.1016/j.jocs.2025.102719","url":null,"abstract":"<div><div>Link prediction is a fundamental task in network analysis, aiming at predicting missing or future connections between nodes in a network. With the growing availability of complex network data in fields like social networks, biological systems, the Internet, and scientific collaboration networks, accurate link prediction methods are becoming increasingly critical. Neighborhood or graph based link prediction algorithms are applied identically to different types of networks so that any differences in their structures are not exploited efficiently. Machine or deep learning based link prediction algorithms apply to each kind of network differently depending on the type of network, due to the unique characteristics of each domain, but frequently, most of them give poor results. In this paper, we propose a novel approach for link prediction, leveraging the power of machine learning and evolutionary algorithms. Our method utilizes local network information by encoding the network topology into link embeddings through a heuristic machine learning architecture. We introduce a novel tool to extract features from network structure effectively and combine them in an effective way through an evolutionary algorithm improving the discriminative power of link embeddings. We evaluate our method on eleven benchmark datasets and demonstrate its superior performance compared to a series (eleven in total) of effective and state-of-the-art algorithms. Our approach advances the state-of-the-art in link prediction yielding better results than other methods in all the networks we have applied it to.</div></div>","PeriodicalId":48907,"journal":{"name":"Journal of Computational Science","volume":"92 ","pages":"Article 102719"},"PeriodicalIF":3.7,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145158210","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号