Resistor networks, valued for their topological versatility and stable electrical properties, have emerged as a focal point across multiple disciplines. Yet, resistor networks with profound mathematical and physical significance remain largely unexplored. This study presents a detailed investigation of the double Beltrami horn surface resistor network and proposes an interpretable reasoning framework based on graph structures and grounded in physical laws. To improve the efficiency of large scale computation, the seventh type of discrete sine transform and Chebyshev polynomials of the first class are employed to derive the exact potential formula. In addition to generating potential distribution diagrams for various special scenarios, a fast algorithm is developed to significantly enhance the efficiency of potential computation. Furthermore, to expand the application potential of the resistor network, an efficient path planning algorithm based on the exact potential formula is proposed, and its applicability in dynamic environments is validated in preliminary experiments.
{"title":"Analysis of double Beltrami horn surface resistor networks and efficient path planning","authors":"Xiaoyu Jiang , Jianwei Dai , Yanpeng Zheng , Zhaolin Jiang","doi":"10.1016/j.knosys.2026.115489","DOIUrl":"10.1016/j.knosys.2026.115489","url":null,"abstract":"<div><div>Resistor networks, valued for their topological versatility and stable electrical properties, have emerged as a focal point across multiple disciplines. Yet, resistor networks with profound mathematical and physical significance remain largely unexplored. This study presents a detailed investigation of the double Beltrami horn surface resistor network and proposes an interpretable reasoning framework based on graph structures and grounded in physical laws. To improve the efficiency of large scale computation, the seventh type of discrete sine transform and Chebyshev polynomials of the first class are employed to derive the exact potential formula. In addition to generating potential distribution diagrams for various special scenarios, a fast algorithm is developed to significantly enhance the efficiency of potential computation. Furthermore, to expand the application potential of the resistor network, an efficient path planning algorithm based on the exact potential formula is proposed, and its applicability in dynamic environments is validated in preliminary experiments.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115489"},"PeriodicalIF":7.6,"publicationDate":"2026-02-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175229","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-07DOI: 10.1016/j.knosys.2026.115449
Jiaxuan Song , Yue Li , Duantengchuan Li , Xiaoguang Wang , Rui Zhang , Hui Zhang , Jinsong Chen
The existing graph neural network recommendation models aggregate neighborhood information by using a weighted sum strategy based on node popularity. However, this strategy struggles to accurately model the impact of item category features on user behavior. To alleviate this problem, we propose MDCRec, a novel graph convolutional recommendation framework integrating deep clustering. MDCRec utilizes the deep clustering module to mine item category information from the item review keyword documents and constructs multi-view subgraphs based on category information. Information aggregation based on node popularity is performed subsequently on each subgraph to obtain the node embeddings within each subgraph. Ultimately, based on the interaction distribution of users in each subgraph, the embeddings within multi-view subgraphs are aggregated into the final embeddings of nodes. MDCRec integrates item category information and user interests across categories into information aggregation, allowing recommendation models to capture more fine-grained relationships between items and user preferences. It can also work in tandem with other performance-enhancing techniques like contrastive learning to further boost model effectiveness. Experimental results on public real-world datasets indicate that most graph neural network recommendation models—including variants that use contrastive learning—integrated with the MDCRec information aggregation framework outperform the original popularity-based version. These models achieve varying degrees of performance gains, with average improvements of 1.75% in Recall@20 and 1.87% in NDCG@20. Our code is publicly available at https://github.com/dacilab/MDCRec.
{"title":"Integrating deep clustering and multi-view graph neural networks for recommender system","authors":"Jiaxuan Song , Yue Li , Duantengchuan Li , Xiaoguang Wang , Rui Zhang , Hui Zhang , Jinsong Chen","doi":"10.1016/j.knosys.2026.115449","DOIUrl":"10.1016/j.knosys.2026.115449","url":null,"abstract":"<div><div>The existing graph neural network recommendation models aggregate neighborhood information by using a weighted sum strategy based on node popularity. However, this strategy struggles to accurately model the impact of item category features on user behavior. To alleviate this problem, we propose MDCRec, a novel graph convolutional recommendation framework integrating deep clustering. MDCRec utilizes the deep clustering module to mine item category information from the item review keyword documents and constructs multi-view subgraphs based on category information. Information aggregation based on node popularity is performed subsequently on each subgraph to obtain the node embeddings within each subgraph. Ultimately, based on the interaction distribution of users in each subgraph, the embeddings within multi-view subgraphs are aggregated into the final embeddings of nodes. MDCRec integrates item category information and user interests across categories into information aggregation, allowing recommendation models to capture more fine-grained relationships between items and user preferences. It can also work in tandem with other performance-enhancing techniques like contrastive learning to further boost model effectiveness. Experimental results on public real-world datasets indicate that most graph neural network recommendation models—including variants that use contrastive learning—integrated with the MDCRec information aggregation framework outperform the original popularity-based version. These models achieve varying degrees of performance gains, with average improvements of 1.75% in Recall@20 and 1.87% in NDCG@20. Our code is publicly available at <span><span>https://github.com/dacilab/MDCRec</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115449"},"PeriodicalIF":7.6,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175150","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-07DOI: 10.1016/j.knosys.2026.115446
Hongpu Jiang , Jinxin Zuo , Yueming Lu , Haonan Li
Federated Self-Supervised Learning (FSSL) is a promising paradigm for extracting robust representations from decentralized unlabeled data. However, its effectiveness is often hindered by non-IID data distributions and label scarcity, which cause model divergence and limit generalization. In this paper, we propose Federated Self-Supervised and Global-Personalized Collaborative Learning (FedGP), a novel framework designed to bridge the gap between global knowledge integration and local client adaptation. The core of FedGP is the Collaborative Knowledge Anchoring (CKA) mechanism, which utilizes adaptive regularization to anchor shared global knowledge while enabling personalized refinement on local data. By dynamically balancing collaborative risks and local empirical losses via learnable coefficients, FedGP ensures stable convergence in heterogeneous environments. Extensive evaluations on multiple benchmarks, including a real-world private Flora dataset, demonstrate that FedGP consistently outperforms state-of-the-art FSSL methods. Our results confirm that FedGP achieves high-quality representation learning with significantly reduced communication overhead and annotation dependency, providing a scalable solution for privacy-preserving decentralized systems.
{"title":"Towards heterogeneity-aware federated self-supervised learning via knowledge anchoring","authors":"Hongpu Jiang , Jinxin Zuo , Yueming Lu , Haonan Li","doi":"10.1016/j.knosys.2026.115446","DOIUrl":"10.1016/j.knosys.2026.115446","url":null,"abstract":"<div><div>Federated Self-Supervised Learning (FSSL) is a promising paradigm for extracting robust representations from decentralized unlabeled data. However, its effectiveness is often hindered by non-IID data distributions and label scarcity, which cause model divergence and limit generalization. In this paper, we propose Federated Self-Supervised and Global-Personalized Collaborative Learning (FedGP), a novel framework designed to bridge the gap between global knowledge integration and local client adaptation. The core of FedGP is the Collaborative Knowledge Anchoring (CKA) mechanism, which utilizes adaptive regularization to anchor shared global knowledge while enabling personalized refinement on local data. By dynamically balancing collaborative risks and local empirical losses via learnable coefficients, FedGP ensures stable convergence in heterogeneous environments. Extensive evaluations on multiple benchmarks, including a real-world private Flora dataset, demonstrate that FedGP consistently outperforms state-of-the-art FSSL methods. Our results confirm that FedGP achieves high-quality representation learning with significantly reduced communication overhead and annotation dependency, providing a scalable solution for privacy-preserving decentralized systems.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115446"},"PeriodicalIF":7.6,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175319","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-07DOI: 10.1016/j.knosys.2026.115522
Changting Zhong , Hao Chen , Dabo Xin , Tong Xu , Zeng Meng , Xinwei Wang , Ali Riza Yildiz , Seyedali Mirjalili
Starfish optimization algorithm (SFOA) is a bio-inspired metaheuristic algorithm for global optimization, which has demonstrated accuracy and efficiency in popular benchmark functions. However, for complex practical problems such as engineering design and feature selection, SFOA still requires a better balance between exploration and exploitation to ensure robust performance in real-world applications. In this paper, we present an improved SFOA algorithm named ORLGSFOA, which integrates opposition-based learning, reinforcement learning, and the growth optimizer with the basic SFOA. The algorithm first incorporates the opposition-based learning strategy during initialization to improve the diversity and quality of the initial solutions. Then, the updating rule from the growth optimizer is hybridized with SFOA to balance exploration and exploitation. Moreover, ORLGSFOA integrates the reinforcement learning strategy to reward the winner from SFOA and growth optimizer by adding updating positions during optimization to enhance global convergence. Experiments demonstrate the superior performance of ORLGSFOA. In comprehensive benchmark tests on 65 functions from classical, CEC2017, and CEC2022 suites, ORLGSFOA outperformed 15 other metaheuristic algorithms by achieving more accurate solutions. Additionally, this effectiveness translates directly to real-world applications, as is evidenced by tests on seven engineering design problems. Besides, the effectiveness of ORLGSFOA in solving discrete combinatorial optimization problems is verified through 52 feature selection problems, and the algorithm is extended to the wind engineering scenarios. In conclusion, ORLGSFOA demonstrates powerful efficacy in addressing a wide range of challenges, including global optimization, engineering design, and feature selection problems. The source code of ORLGSFOA is publicly available at: https://ww2.mathworks.cn/matlabcentral/fileexchange/183223-orlgsfoa.
{"title":"Opposition and reinforcement learning growth-starfish optimization algorithm for engineering design and feature selection","authors":"Changting Zhong , Hao Chen , Dabo Xin , Tong Xu , Zeng Meng , Xinwei Wang , Ali Riza Yildiz , Seyedali Mirjalili","doi":"10.1016/j.knosys.2026.115522","DOIUrl":"10.1016/j.knosys.2026.115522","url":null,"abstract":"<div><div>Starfish optimization algorithm (SFOA) is a bio-inspired metaheuristic algorithm for global optimization, which has demonstrated accuracy and efficiency in popular benchmark functions. However, for complex practical problems such as engineering design and feature selection, SFOA still requires a better balance between exploration and exploitation to ensure robust performance in real-world applications. In this paper, we present an improved SFOA algorithm named ORLGSFOA, which integrates opposition-based learning, reinforcement learning, and the growth optimizer with the basic SFOA. The algorithm first incorporates the opposition-based learning strategy during initialization to improve the diversity and quality of the initial solutions. Then, the updating rule from the growth optimizer is hybridized with SFOA to balance exploration and exploitation. Moreover, ORLGSFOA integrates the reinforcement learning strategy to reward the winner from SFOA and growth optimizer by adding updating positions during optimization to enhance global convergence. Experiments demonstrate the superior performance of ORLGSFOA. In comprehensive benchmark tests on 65 functions from classical, CEC2017, and CEC2022 suites, ORLGSFOA outperformed 15 other metaheuristic algorithms by achieving more accurate solutions. Additionally, this effectiveness translates directly to real-world applications, as is evidenced by tests on seven engineering design problems. Besides, the effectiveness of ORLGSFOA in solving discrete combinatorial optimization problems is verified through 52 feature selection problems, and the algorithm is extended to the wind engineering scenarios. In conclusion, ORLGSFOA demonstrates powerful efficacy in addressing a wide range of challenges, including global optimization, engineering design, and feature selection problems. The source code of ORLGSFOA is publicly available at: <span><span>https://ww2.mathworks.cn/matlabcentral/fileexchange/183223-orlgsfoa</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115522"},"PeriodicalIF":7.6,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175143","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-07DOI: 10.1016/j.knosys.2026.115448
Xu-Hua Yang , Zhen-Lei Huang , Gang-Feng Ma , Jia-Ning Xu
Social recommendation incorporates social network information into personalized recommendation systems, thus effectively mitigating data sparsity and cold-start issues. However, real-world social networks often contain noise, which significantly hinders the capture of authentic social preference information. Existing social denoising methods fall into two categories: “hard” denoising based on network reconstruction (which may severely damage the original network topology) and “soft” denoising based on user representation (which often overlooks node dependencies during the denoising process). To address these limitations, we propose a Structure-Guided High Fidelity Denoising framework for Social Recommendation (SGDSR). First, we design a dual-diffusion module that incorporates structural information by introducing network topology constraints into the diffusion process. This effectively preserves key social signals during denoising. Then, we employ contrastive learning to align representations from dual-diffusion pathways, enhancing consistency. Finally, we propose a fusion-denoising mechanism that refines integrated network information to improve representation robustness. Extensive experiments on three real-world datasets demonstrate that SGDSR outperforms state-of-the-art baselines. The code is available at https://github.com/Only-SR/SGDSR.
{"title":"Dual-track diffusion: Structure-Guided high fidelity denoising for social recommendation","authors":"Xu-Hua Yang , Zhen-Lei Huang , Gang-Feng Ma , Jia-Ning Xu","doi":"10.1016/j.knosys.2026.115448","DOIUrl":"10.1016/j.knosys.2026.115448","url":null,"abstract":"<div><div>Social recommendation incorporates social network information into personalized recommendation systems, thus effectively mitigating data sparsity and cold-start issues. However, real-world social networks often contain noise, which significantly hinders the capture of authentic social preference information. Existing social denoising methods fall into two categories: “hard” denoising based on network reconstruction (which may severely damage the original network topology) and “soft” denoising based on user representation (which often overlooks node dependencies during the denoising process). To address these limitations, we propose a Structure-Guided High Fidelity Denoising framework for Social Recommendation (SGDSR). First, we design a dual-diffusion module that incorporates structural information by introducing network topology constraints into the diffusion process. This effectively preserves key social signals during denoising. Then, we employ contrastive learning to align representations from dual-diffusion pathways, enhancing consistency. Finally, we propose a fusion-denoising mechanism that refines integrated network information to improve representation robustness. Extensive experiments on three real-world datasets demonstrate that SGDSR outperforms state-of-the-art baselines. The code is available at <span><span>https://github.com/Only-SR/SGDSR</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115448"},"PeriodicalIF":7.6,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175186","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-06DOI: 10.1016/j.knosys.2026.115468
Xing Wu , Yimin Zhu , Shuo Duan , Xinyuan Zhang , Xing Wei , Bo Huang , Quan Qian
Computed tomography (CT) denoising is essential for clinical diagnosis and industrial inspection, but it is challenged by various noise and structural artifacts. Existing deep learning methods are limited by insufficient modeling of long-term dependencies, a disregard for intrinsic frequency-domain priors, and a significant domain gap caused by their reliance on unrealistic synthetic noise. To address these issues, a frequency-spatial complementary attention network (FSCANet) is proposed, which is based on the complementary fusion of frequency and spatial domains. The frequency domain branch explicitly decouples structural and phase information to model global context, while the spatial-domain branch improves local details. Simultaneously, a real-data-guided physics-informed noise model is introduced to bridge the domain gap by formalizing the physical noise generation process as a differentiable layer. FSCANet and the noise model are jointly optimized using a hybrid data-driven co-optimization strategy, resulting in a dynamic feedback loop that not only compels the noise model to generate physically interpretable noise but also drives FSCANet to achieve greater robustness. FSCANet achieves state-of-the-art performance on the DeepLesion dataset with a PSNR of 40.5861 dB and an SSIM of 0.9913, and demonstrates robust generalization on authentic clinical data from the Mayo dataset.
{"title":"Frequency-spatial complementary attention network for computed tomography","authors":"Xing Wu , Yimin Zhu , Shuo Duan , Xinyuan Zhang , Xing Wei , Bo Huang , Quan Qian","doi":"10.1016/j.knosys.2026.115468","DOIUrl":"10.1016/j.knosys.2026.115468","url":null,"abstract":"<div><div>Computed tomography (CT) denoising is essential for clinical diagnosis and industrial inspection, but it is challenged by various noise and structural artifacts. Existing deep learning methods are limited by insufficient modeling of long-term dependencies, a disregard for intrinsic frequency-domain priors, and a significant domain gap caused by their reliance on unrealistic synthetic noise. To address these issues, a frequency-spatial complementary attention network (FSCANet) is proposed, which is based on the complementary fusion of frequency and spatial domains. The frequency domain branch explicitly decouples structural and phase information to model global context, while the spatial-domain branch improves local details. Simultaneously, a real-data-guided physics-informed noise model is introduced to bridge the domain gap by formalizing the physical noise generation process as a differentiable layer. FSCANet and the noise model are jointly optimized using a hybrid data-driven co-optimization strategy, resulting in a dynamic feedback loop that not only compels the noise model to generate physically interpretable noise but also drives FSCANet to achieve greater robustness. FSCANet achieves state-of-the-art performance on the DeepLesion dataset with a PSNR of 40.5861 dB and an SSIM of 0.9913, and demonstrates robust generalization on authentic clinical data from the Mayo dataset.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115468"},"PeriodicalIF":7.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174787","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-06DOI: 10.1016/j.knosys.2026.115426
Yuanlin Zhao , Jiangang Ding , Yansong Wang , Yihui Shan , Lili Pei , Wei Li
Nearshore scenarios are frequently affected by fog and contain a variety of objects exhibiting distinct motion patterns. These inherent factors pose significant challenges for accurate object detection in nearshore scenarios. Common approach is to utilize Video Object Detection (VOD) to learn the spatial features and motion information of nearshore objects. However, this method becomes hindered in situations involving foggy conditions or when different objects share similar optical characteristics, thus impeding effective pipeline modeling. To address these challenges, we propose a nearshore Cross-modal Video Object Detector (CVODNet). By leveraging learnable feature interaction between Infrared (IR) and visible light videos, we reduce the obstacles in pipeline modeling caused by the transient loss of features from unimodal. Learning from correlated frames to obtain the optimal weights for moving objects. Finally, deformable convolution is employed to address the challenges of pixel-level misalignment in cross-modal data presented in video form. After end-to-end training, CVODNet achieves State-of-the-art (SOTA) performance in benchmark evaluations.
{"title":"Enabling nearshore cross-modal video object detector to learn more accurate spatial and temporal information","authors":"Yuanlin Zhao , Jiangang Ding , Yansong Wang , Yihui Shan , Lili Pei , Wei Li","doi":"10.1016/j.knosys.2026.115426","DOIUrl":"10.1016/j.knosys.2026.115426","url":null,"abstract":"<div><div>Nearshore scenarios are frequently affected by fog and contain a variety of objects exhibiting distinct motion patterns. These inherent factors pose significant challenges for accurate object detection in nearshore scenarios. Common approach is to utilize Video Object Detection (VOD) to learn the spatial features and motion information of nearshore objects. However, this method becomes hindered in situations involving foggy conditions or when different objects share similar optical characteristics, thus impeding effective pipeline modeling. To address these challenges, we propose a nearshore Cross-modal Video Object Detector (CVODNet). By leveraging learnable feature interaction between Infrared (IR) and visible light videos, we reduce the obstacles in pipeline modeling caused by the transient loss of features from unimodal. Learning from correlated frames to obtain the optimal weights for moving objects. Finally, deformable convolution is employed to address the challenges of pixel-level misalignment in cross-modal data presented in video form. After end-to-end training, CVODNet achieves State-of-the-art (SOTA) performance in benchmark evaluations.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115426"},"PeriodicalIF":7.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-06DOI: 10.1016/j.knosys.2026.115507
Chia-Cheng Hu
The rapid expansion of real-time Internet of Things (IoT) applications, particularly in highly dynamic environments such as Vehicular Fog Networks (VFNs), presents significant challenges for task offloading due to stringent latency constraints and fluctuating resource availability. To address these challenges, this paper introduces a hybrid knowledge-based framework that integrates Integer Linear Programming (ILP) with Case-Based Reasoning (CBR) to enable intelligent and adaptive task offloading in VFNs. The framework operates in two complementary phases: ILP is applied offline to derive optimal offloading strategies under diverse network conditions and construct a decision knowledge base, while CBR is executed online to retrieve and adapt relevant cases for real-time decision-making with minimal computational cost. By decoupling global optimization from online inference, the proposed system achieves high scalability and responsiveness.
Comprehensive simulations conducted in AGV-enabled VFNs demonstrate that the proposed framework achieves near-optimal performance, reducing average task latency by up to 20% and energy consumption by 15% compared with heuristic and learning-based baselines. Furthermore, the Decision Support System (DSS) sustains a retrieval latency below 150 ms even with a large-scale case database, ensuring real-time adaptability and scalability under varying network topologies and workloads. These results confirm the framework’s robustness and efficiency, offering a promising foundation for knowledge-driven task offloading in next-generation IoT and edge computing infrastructures.
{"title":"Knowledge-based optimization and reasoning for intelligent task offloading in dynamic vehicular fog networks","authors":"Chia-Cheng Hu","doi":"10.1016/j.knosys.2026.115507","DOIUrl":"10.1016/j.knosys.2026.115507","url":null,"abstract":"<div><div>The rapid expansion of real-time Internet of Things (IoT) applications, particularly in highly dynamic environments such as Vehicular Fog Networks (VFNs), presents significant challenges for task offloading due to stringent latency constraints and fluctuating resource availability. To address these challenges, this paper introduces a hybrid knowledge-based framework that integrates Integer Linear Programming (ILP) with Case-Based Reasoning (CBR) to enable intelligent and adaptive task offloading in VFNs. The framework operates in two complementary phases: ILP is applied offline to derive optimal offloading strategies under diverse network conditions and construct a decision knowledge base, while CBR is executed online to retrieve and adapt relevant cases for real-time decision-making with minimal computational cost. By decoupling global optimization from online inference, the proposed system achieves high scalability and responsiveness.</div><div>Comprehensive simulations conducted in AGV-enabled VFNs demonstrate that the proposed framework achieves near-optimal performance, reducing average task latency by up to 20% and energy consumption by 15% compared with heuristic and learning-based baselines. Furthermore, the Decision Support System (DSS) sustains a retrieval latency below 150 ms even with a large-scale case database, ensuring real-time adaptability and scalability under varying network topologies and workloads. These results confirm the framework’s robustness and efficiency, offering a promising foundation for knowledge-driven task offloading in next-generation IoT and edge computing infrastructures.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115507"},"PeriodicalIF":7.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175178","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-06DOI: 10.1016/j.knosys.2026.115463
Jia Peng , Xiaoyu Liu , Xiaodong Zhou , Lei Wang , Xianyou Zhu
Accurate identification of drug-target interactions (DTIs) is crucial for improving screening efficiency and reducing experimental costs in drug discovery. However, existing DTI prediction methods still face two major challenges: (1) Feature representation relies on single-modality data, making it difficult to comprehensively characterize the multi-level properties of drugs and targets; (2) Limited cross-modal fusion capabilities hinder the capture of complex associations between drugs and targets, resulting in constrained prediction performance. To address these issues, this study proposes a dual-branch collaborative multi-modal fusion DTI prediction framework (DBCA-DTI). This framework comprises two feature encoding branches: the first is a large language model-enhanced semantic feature branch, which utilizes pre-trained large language models to encode drug molecule and protein, accurately capturing their high-dimensional semantic information; the second is a physicochemical property feature branch, which combines RDKit-extracted drug structural descriptors with amino acid-based protein fundamental features to enhance the model’s feature expression depth and recognition capability in the physicochemical property dimension. Additionally, both branches employ a bidirectional adaptive gated cross-attention mechanism to enhance cross-modal interactions between drugs and targets. A multimodal feature fusion module integrates diverse outputs from both branches, boosting overall representational capacity and prediction robustness. Experimental results demonstrate that DBCA-DTI significantly outperforms existing mainstream methods across multiple public benchmark datasets. This study provides an efficient, flexible, and scalable solution for DTI prediction.The code is accessible at https://github.com/myseverus/DBCA-DTI.
{"title":"DBCA-DTI: A dual-branch multimodal framework based on bidirectional adaptive gated cross-attention mechanism for drug-target interaction prediction","authors":"Jia Peng , Xiaoyu Liu , Xiaodong Zhou , Lei Wang , Xianyou Zhu","doi":"10.1016/j.knosys.2026.115463","DOIUrl":"10.1016/j.knosys.2026.115463","url":null,"abstract":"<div><div>Accurate identification of drug-target interactions (DTIs) is crucial for improving screening efficiency and reducing experimental costs in drug discovery. However, existing DTI prediction methods still face two major challenges: (1) Feature representation relies on single-modality data, making it difficult to comprehensively characterize the multi-level properties of drugs and targets; (2) Limited cross-modal fusion capabilities hinder the capture of complex associations between drugs and targets, resulting in constrained prediction performance. To address these issues, this study proposes a dual-branch collaborative multi-modal fusion DTI prediction framework (DBCA-DTI). This framework comprises two feature encoding branches: the first is a large language model-enhanced semantic feature branch, which utilizes pre-trained large language models to encode drug molecule and protein, accurately capturing their high-dimensional semantic information; the second is a physicochemical property feature branch, which combines RDKit-extracted drug structural descriptors with amino acid-based protein fundamental features to enhance the model’s feature expression depth and recognition capability in the physicochemical property dimension. Additionally, both branches employ a bidirectional adaptive gated cross-attention mechanism to enhance cross-modal interactions between drugs and targets. A multimodal feature fusion module integrates diverse outputs from both branches, boosting overall representational capacity and prediction robustness. Experimental results demonstrate that DBCA-DTI significantly outperforms existing mainstream methods across multiple public benchmark datasets. This study provides an efficient, flexible, and scalable solution for DTI prediction.The code is accessible at <span><span>https://github.com/myseverus/DBCA-DTI</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115463"},"PeriodicalIF":7.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175320","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 : 2026-02-05DOI: 10.1016/j.knosys.2026.115488
Yang Yang, Xi Zhang, Jiaqi Zhang, Lanling Zeng
The diffusion models have demonstrated remarkable performance on the task of image restoration. Most of the existing image restoration methods leverage the diffusion model as a powerful prior. In this paper, we propose a novel method named PIRP that further integrates the gradient prior, which has been a popular prior in image restoration. The integration harnesses the strengths of both priors, thus being able to enhance the overall efficacy of image restoration. More importantly, the incorporation of the gradient prior improves the flexibility of the method by facilitating parameterized image restoration, i.e., it provides an effective way to tweak the parameters, which is essential in tailoring satisfactory results. Moreover, we propose a novel plug-and-play sampling method based on the proposed model, which is able to improve the image restoration quality without necessitating any retraining. To validate the effectiveness of the proposed method, we have conducted extensive experiments on multiple image restoration tasks, including single-image super-resolution, Gaussian deblurring, motion deblurring, and their noisy variants. Both qualitative and quantitative results on popular datasets demonstrate the advantages of the proposed method.
{"title":"Parameterized image restoration with diffusion and gradient priors","authors":"Yang Yang, Xi Zhang, Jiaqi Zhang, Lanling Zeng","doi":"10.1016/j.knosys.2026.115488","DOIUrl":"10.1016/j.knosys.2026.115488","url":null,"abstract":"<div><div>The diffusion models have demonstrated remarkable performance on the task of image restoration. Most of the existing image restoration methods leverage the diffusion model as a powerful prior. In this paper, we propose a novel method named PIRP that further integrates the gradient prior, which has been a popular prior in image restoration. The integration harnesses the strengths of both priors, thus being able to enhance the overall efficacy of image restoration. More importantly, the incorporation of the gradient prior improves the flexibility of the method by facilitating parameterized image restoration, i.e., it provides an effective way to tweak the parameters, which is essential in tailoring satisfactory results. Moreover, we propose a novel plug-and-play sampling method based on the proposed model, which is able to improve the image restoration quality without necessitating any retraining. To validate the effectiveness of the proposed method, we have conducted extensive experiments on multiple image restoration tasks, including single-image super-resolution, Gaussian deblurring, motion deblurring, and their noisy variants. Both qualitative and quantitative results on popular datasets demonstrate the advantages of the proposed method.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115488"},"PeriodicalIF":7.6,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175323","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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