Mehdhar S.A.M. Al-Gaashani , Abduljabbar S. Ba Mahel , Ammar Muthanna
{"title":"Truncated MobileNetV2 Sparse Vision Graph Attention Model for Explainable Monkeypox Disease Classification","authors":"Mehdhar S.A.M. Al-Gaashani , Abduljabbar S. Ba Mahel , Ammar Muthanna","doi":"10.1016/j.knosys.2025.114503","DOIUrl":"10.1016/j.knosys.2025.114503","url":null,"abstract":"<div><div>The current outbreak of monkeypox (mpox) presents challenges for timely and accurate diagnosis due to the disease’s diverse and unusual skin lesion patterns. Traditional deep learning models, such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), struggle with these irregular features because they rely on rigid, grid-based methods. To address this, we introduce the Truncated MobileNetV2 Sparse Vision Graph Attention (TMSVGA) model. TMSVGA combines components of MobileNetV2, which focuses on identifying smaller details, with a Sparse Vision Graph Attention block enhanced by a Squeeze-and-Excitation (SE) mechanism to improve channel-wise attention. This approach enhances the understanding of complex and long-distance relationships, emphasizing diagnostically significant regions and improving classification precision. We optimized TMSVGA using the Optuna framework for automated hyperparameter tuning. Additionally, Gradient-weighted Class Activation Mapping (Grad-CAM) and Local Interpretable Model-Agnostic Explanations (LIME) provided interpretable visualizations, highlighting influential regions in decision-making. The TMSVGA model was validated on the Monkeypox Skin Images Dataset (MSID), achieving 96.79 % accuracy, 96.90 % precision, 95.34 % recall, 96.08 % F1-score, and 95.37% Matthews Correlation Coefficient (MCC). These results demonstrate that TMSVGA outperforms existing models, particularly in handling irregular lesion patterns. By achieving high diagnostic accuracy and precision, our study showcases the potential of Vision Graph Neural Networks (ViGNNs) in advancing medical image analysis for diseases with non-uniform spatial patterns. Furthermore, the lightweight architecture of TMSVGA ensures suitability for mobile and resource-constrained diagnostic applications.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114503"},"PeriodicalIF":7.6,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145222222","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}
Jiaqi Nie , Rankun Chen , Jingxiang Huang , Ben Yang , Xuetao Zhang
{"title":"Robust multi-view discrete clustering with unified graph learning","authors":"Jiaqi Nie , Rankun Chen , Jingxiang Huang , Ben Yang , Xuetao Zhang","doi":"10.1016/j.knosys.2025.114510","DOIUrl":"10.1016/j.knosys.2025.114510","url":null,"abstract":"<div><div>Graph-based multi-view clustering (GMVC) has garnered significant attention due to its ability to overcome sample space shape constraints. However, existing GMVC methods encounter two major challenges: (1) Their effectiveness diminishes because they solely rely on sample-constructed graphs and the two-stage mismatch caused by additional discretization; (2) Their robustness deteriorates substantially when applied to real-world datasets that contain complex noise. To address these limitations, we propose a robust multi-view discrete clustering model with unified graph learning (RCUGL). This model integrates richer graph structural information and accommodates complex noise clustering tasks. Specifically, we incorporated low-rank approximation graphs reconstructed from spectral embeddings and graphs constructed by samples into a unified graph to provide enriched structural insights. Subsequently, within the framework of the correntropy, discrete spectral analysis was performed directly on the unified graph to derive cluster assignments. Given the non-convex and discrete nature of the proposed RCUGL model, we developed a half-quadratic-based coordinate descent optimisation algorithm to ensure rapid and reliable convergence. Extensive experiments demonstrate that RCUGL substantially improves clustering effectiveness, comparable to state-of-the-art methods.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114510"},"PeriodicalIF":7.6,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159230","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}
Yuping Lai , Zidong Wang , Ziqing Lin , Yuhan Cao , Zihao Li , Qing Ye
{"title":"An efficient network intrusion detection model based on beta mixture models","authors":"Yuping Lai , Zidong Wang , Ziqing Lin , Yuhan Cao , Zihao Li , Qing Ye","doi":"10.1016/j.knosys.2025.114506","DOIUrl":"10.1016/j.knosys.2025.114506","url":null,"abstract":"<div><div>With the rapid development of computer networks and network applications, ensuring network security has become a critical concern and has garnered significant attention from both academia and industry. Network intrusion detection (NID) plays a pivotal role in safeguarding cybersecurity and maintaining system stability. Most existing NID approaches rely on traditional machine learning (ML) or deep learning (DL) techniques to identify threats and potential attacks based on network traffic data. However, these methods often suffer from high computational complexity and large model sizes, which significantly impede their deployment in resource-constrained environments such as the Internet of Things (IoT), edge computing infrastructures, and wireless sensor networks. In this study, we propose an efficient NID framework based on the Beta Mixture Model (BMM) classifier. The proposed method integrates the BMM with the recently introduced Extended Stochastic Variational Inference (ESVI) framework to effectively characterize both normal and intrusive behavior patterns. The ESVI framework enables simultaneous parameter estimation and model complexity control in a principled and computationally efficient manner. Experimental evaluations show that, compared to NID methods utilizing established finite mixture models, traditional ML, or state-of-the-art DL techniques, our approach substantially reduces computational overhead while achieving comparable detection performance.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114506"},"PeriodicalIF":7.6,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159233","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}
Qiyin Lin , Feiyu Gu , Mingjun Qiu , Chen Wang , Jian Zhuang , Jun Hong
{"title":"TransPhyX: A data-driven method for dynamic physical field prediction in stochastic load time-series","authors":"Qiyin Lin , Feiyu Gu , Mingjun Qiu , Chen Wang , Jian Zhuang , Jun Hong","doi":"10.1016/j.knosys.2025.114492","DOIUrl":"10.1016/j.knosys.2025.114492","url":null,"abstract":"<div><div>Dynamic prediction of in-service physical fields (e.g. stress, strain, and temperature fields) constitutes a cornerstone technology for digital governance of mechanical equipment. The stochasticity and time-varying characteristics of external excitation loads (e.g. thermal, vibrational, and impact loads) introduce significant complexity in physical field prediction. Online monitoring of physical fields at the assembly interfaces of mechanical systems is critical for ensuring structural safety, extending service life, and optimizing design. This study proposes TransPhyX (Transformer-Based Physical Field Prediction with XGBoost Precoder), a hybrid data-driven framework designed to overcome these challenges. The novelty of TransPhyX lies in: (1) a recursive stochastic load generation and parametric dataset construction method tailored for dynamic prediction tasks; (2) a modular hybrid architecture that decouples transient load encoding (via XGBoost) and dynamic sequence modeling (via Transformer), improving spatiotemporal continuity and generalization; and (3) an Outlier Removal Ensemble (ORE) algorithm that fuses multi-scale predictions to eliminate anomalies and enhance robustness. Validated on flip-chip thermal management and flange-bolt stress prediction, TransPhyX achieves 99.79 % prediction fidelity with a 97.79 % reduction in computational costs compared to FEM, outperforming AutoGAN and TransUNet baselines in both accuracy and stability. These contributions establish TransPhyX as a rapid, high-fidelity solution for real-time structural health monitoring and digital twin implementation in stochastic loading environments.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114492"},"PeriodicalIF":7.6,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145222442","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}
Muxin Liao , Jiayang Wang , Hong Deng , Yingqiong Peng , Hua Yin , Yinglong Wang , Guoguang Hua
{"title":"Domain-generalized token linking in vision foundation models for semantic segmentation","authors":"Muxin Liao , Jiayang Wang , Hong Deng , Yingqiong Peng , Hua Yin , Yinglong Wang , Guoguang Hua","doi":"10.1016/j.knosys.2025.114497","DOIUrl":"10.1016/j.knosys.2025.114497","url":null,"abstract":"<div><div>[S U M M A R Y] Vision Foundation Models (VFMs) achieve remarkable performance compared with traditional methods based on convolutional neural networks and vision transformer networks in Domain-Generalized Semantic Segmentation (DGSS). These VFM-based DGSS methods focus on adopting efficient parameter fine-tuning strategies that use a set of learnable tokens to fine-tune VFMs to the downstream DGSS task, yet struggle to mine domain-invariant information from VFMs since the backbone of VFMs is frozen during the fine-tuning stage. To address this issue, a Domain-Generalized Token Linking (DGTL) approach is proposed to mine domain-invariant information from VFMs for improving the performance in unseen target domains, which contains a Text-guided Dual Token Linking (TDTL) module and a Text-guided Distribution Normalization (TDN) strategy. For the TDTL module, first, a set of learnable tokens is linked to the text embeddings for building the relations between the learnable tokens and text embeddings, which is beneficial for learning domain-invariant tokens since the text embeddings generated from the CLIP model are domain-invariant. Second, the feature-level and mask-level linking strategies are proposed to link the learned domain-invariant tokens to the features and masks to guide the mining of domain-invariant information from the VFM. For the TDN strategy, the pairwise similarity between the predictive masks associated with the learnable tokens and the text embeddings is utilized to explicitly align the semantic distribution of visual features in the learnable tokens with the text embeddings. Extensive experiments demonstrate that the DGTL approach achieves superior performance to recent methods across multiple DGSS benchmarks. The code is released on GitHub:<span><span>https://github.com/seabearlmx/DGTL</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114497"},"PeriodicalIF":7.6,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159954","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}
{"title":"P-EVFL: Efficient verifiable federated learning with privacy","authors":"Juan Ma , Xiangshen Ma , Yuling Chen","doi":"10.1016/j.knosys.2025.114480","DOIUrl":"10.1016/j.knosys.2025.114480","url":null,"abstract":"<div><div>Federated learning has recently become popular and widely used in various areas. However, it still faces challenges like the leakage of the client’s local model updates and the server forging aggregation results. To address these issues, we propose <em>an efficient verifiable federated learning scheme with privacy</em> (P-EVFL), which seeks to ensure privacy and verifiability with a lower overhead. Specifically, we first design a lightweight masking technique to protect the honest clients’ local model updates. Next, we introduce homomorphic hash functions to develop a verifiable method to ensure the integrity of the aggregation results. Besides, to reduce the overhead of the verification process, a verification algorithm based on a Merkle tree is proposed. We also conduct comprehensive experiments and compare our scheme with other state-of-the-art schemes. The experimental results show that in a scenario with 100 clients, our scheme reduces the computational overhead by up to 8.15 % and the communication overhead by up to 67.38 %.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114480"},"PeriodicalIF":7.6,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159237","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}
{"title":"A multi-granularity facial aesthetic evaluation model based on image-text modality","authors":"Huanyu Chen , Yong Wang , Weisheng Li , Bin Xiao","doi":"10.1016/j.knosys.2025.114502","DOIUrl":"10.1016/j.knosys.2025.114502","url":null,"abstract":"<div><div>Facial Beauty Prediction (FBP) is an emerging research direction at the intersection of artificial intelligence and aesthetics, which has attracted increasing attention in recent years. However, most existing methods rely solely on unimodal data and fail to comprehensively capture the multi-dimensional information of facial aesthetics. To address this challenge, we propose a multigranularity facial aesthetic evaluation model based on image-text modality (ITM-MGFA). By incorporating multi-granularity cognitive theory into the FBP task, the model effectively integrates both coarse-grained and fine-grained aesthetic features extracted from the CLIP encoder through a multigranularity representation module, a task-oriented dynamic alignment module, and a hierarchical interaction optimization module. This facilitates deep cross-modal interaction and fusion, significantly enhancing the model’s capability to model complex aesthetic attributes. Experimental results demonstrate that ITM-MGFA, leveraging the fusion of cross-modal information, achieves higher accuracy in facial aesthetic assessment task compared to traditional unimodal methods, offering a new direction for FBP research. Furthermore, the model can be applied in various scenarios, such as: simulation postoperative assessment of personalized cosmetic surgery in the medical aesthetics; selection of optimal facial aesthetic enhancement solutions on social media; and recommendation of matching solutions in cosmetic recommendation.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114502"},"PeriodicalIF":7.6,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159956","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}
Yunfei Zhang , Xiaoyang Huo , Tianyi Chen , Si Wu , Hau-San Wong
{"title":"Class-conditional image synthesis with intra-class relation preservation","authors":"Yunfei Zhang , Xiaoyang Huo , Tianyi Chen , Si Wu , Hau-San Wong","doi":"10.1016/j.knosys.2025.114487","DOIUrl":"10.1016/j.knosys.2025.114487","url":null,"abstract":"<div><div>Modeling class-conditional data distributions remains challenging, since the intra-class variation may be very large. Different from generic class-conditional Generative Adversarial Networks (GANs), we take inspiration from the observation that there may exist multiple modes with diverse visual appearances in a single class, and propose an Intra-class Prototype-based Relation Preservation (IPRP) approach to improve class-conditional image synthesis. Toward this end, a generator is designed to learn class-specific data distribution, conditioned on intra-class prototype-based relation. To associate label embeddings with the cluster prototypes, we incorporate an auxiliary prototypical network to perform adversarial interpolation, and the synthesized data are required to encapsulate their relation to the corresponding prototypes in the form of interpolation coefficients. The prototypical network can be further leveraged to improve the class-conditional real-fake identification performance by injecting semantics-aware features into a discriminator. This design allows the generator to better capture intra-class modes We conduct extensive experiments to demonstrate that IPRP outperforms the competing class-conditional GANs in terms of data diversity and semantic accuracy.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114487"},"PeriodicalIF":7.6,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159958","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}
{"title":"A supervised learning approach to dynamic weighted fusion in multi-source ordered decision systems","authors":"Xiaoyan Zhang, Jiajia Lin","doi":"10.1016/j.knosys.2025.114485","DOIUrl":"10.1016/j.knosys.2025.114485","url":null,"abstract":"<div><div>With the rapid advancement of new-generation artificial intelligence technologies, machines can process and analyze large-scale data more accurately and efficiently and for more complex tasks. Enhancing the usability and value of the information derived from various information systems across multiple dimensions is essential. However, traditional data dominance relationships cannot reflect people’s different levels of attention to antithetic features, leading to higher complexity and lower classification accuracy. Therefore, it is necessary to consider the weight relationships between attributes in the data, which refers to the degree of correlation between each attribute and the decision in multi-source information systems. Based on these weights and dominance relationships, we consider an entropy-based weighted information fusion method for processing supervised data in multi-source ordered decision systems. We intend four incremental fusion mechanisms to adjust information sources and attribute changes to save running time. Furthermore, experiments are conducted on nine real datasets to demonstrate our method’s effectiveness. The results show that the inevitable accuracy comparisons by the proposed method are superior to most fusion methods. In addition, the dynamic mechanisms, compared to static mechanisms, can significantly reduce running time.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114485"},"PeriodicalIF":7.6,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145120252","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}
Yi Yin , Zongxian Long , Shengli Jin , Yawei Li , Fang Wang , Xin Xu
{"title":"Multiphysics coupling AI prediction method for thermomechanical behavior of steel ladle linings","authors":"Yi Yin , Zongxian Long , Shengli Jin , Yawei Li , Fang Wang , Xin Xu","doi":"10.1016/j.knosys.2025.114495","DOIUrl":"10.1016/j.knosys.2025.114495","url":null,"abstract":"<div><div>Predicting the thermomechanical responses of refractory linings in steel ladles is critical to optimizing production efficiency and ensuring safety in the iron and steel smelting industry. However, traditional numerical simulation methods suffer challenges of high computational costs and insufficient generalizability, while data-driven models are limited by a lack of physical rationality and poor interpretability. Aiming at overcoming these challenges, an artificial intelligence (AI) model, named the steel ladle Kolmogorov–Arnold network (SLKAN), is designed to predict the thermomechanical behavior of ladle linings. Based on the Kolmogorov–Arnold theorem and material constitutive equations, SLKAN precisely predicts the thermomechanical behavior of ladle linings. The model offers substantial advantages in predicting the maximum tensile stress in the steel shell and the maximum compressive stress at the working lining hot face: the coefficient of determination (R<sup>2</sup>) value for compressive stress prediction reaches 0.9942, with a mean absolute error (MAE) of 9.4136 and a root mean squared error (RMSE) of 0.0192; the R<sup>2</sup> value for tensile stress prediction is 0.9578, with an MAE of 41.4855 and an RMSE of 0.0385. Further analysis indicates that the function expressions of SLKAN hold clear physical significance. This study provides an interpretable, efficient AI solution for multiphysics coupling modeling in complex industrial scenarios and offers theoretical guidance for the application of AI in predicting the lifespan of steel-smelting equipment.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114495"},"PeriodicalIF":7.6,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145120322","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}