Knowledge-Based Systems最新文献_第4页

A meta-heuristic stochastic algorithm for the numerical treatment of cancer model through the chemotherapy and stem cells 通过化疗和干细胞对癌症模型进行数值治疗的元启发式随机算法

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-04 DOI: 10.1016/j.knosys.2026.115493

Zulqurnain Sabir , M.A. Abdelkawy , Dumitru Baleanu , Ozlem Defterli

Objective

The aim of current research is to present the numerical performances of the cancer treatment model based on chemotherapy and stem cells using one of the heuristic computing neural network procedures. The cancer treatment model through chemotherapy and stem cells is categorized into stem cells, affected cells, tumor cells, and chemotherapy-based concentration drug.

Method

A process of artificial neural network is applied using the hybrid optimization of global and local search schemes, which are taken as genetic algorithm (GA) and an active set (AS). An error-based fitness function is designed by using the differential model and then optimized by the hybridization of both global and local search schemes. GA is applied to exploit the global result and give a primary guess to the AS that further improves the results locally. AS is rooted in the GA, where GA produces new populaces and AS optimizes the fitness function for every individual. The hybridization of these two schemes is used iteratively for purifying the results. Ten numbers of neurons and log-sigmoid activation functions has been used to solve the cancer treatment model based on chemotherapy and stem cells.

Results

For the correctness of the stochastic solver, the obtained numerical results have been compared with any traditional scheme. Moreover, the reliability and capability of the scheme are performed through the absolute error around 10^-05 to 10^-07 along with different statistical approaches for solving the mathematical model.

Novelty

The proposed artificial neural network structure along with the hybrid optimization of global and local search schemes has never been implemented before to solve the cancer treatment model based on chemotherapy and stem cells.

目的利用一种启发式计算神经网络程序，对基于化疗和干细胞的肿瘤治疗模型进行数值计算。通过化疗和干细胞治疗癌症的模式分为干细胞、受累细胞、肿瘤细胞和以化疗为基础的浓缩药物。方法采用遗传算法（GA）和活动集（as）两种全局和局部搜索混合优化的人工神经网络处理方法。利用差分模型设计了基于误差的适应度函数，并结合全局和局部搜索方案进行了优化。利用遗传算法对全局结果进行挖掘，并对自治系统进行初步猜测，进一步对局部结果进行改进。AS以遗传算法为基础，遗传算法产生新的种群，并对每个个体的适应度函数进行优化。用这两种方案的杂交迭代来净化结果。10个神经元和对数s型激活函数被用来解决基于化疗和干细胞的癌症治疗模型。结果为了验证随机解算器的正确性，将所得到的数值结果与任何传统格式进行了比较。采用不同的统计方法求解数学模型，通过对10-05 ~ 10-07之间的绝对误差，验证了方案的可靠性和性能。本文提出的人工神经网络结构以及全局和局部混合优化搜索方案在解决基于化疗和干细胞的癌症治疗模型中是前所未有的。

{"title":"A meta-heuristic stochastic algorithm for the numerical treatment of cancer model through the chemotherapy and stem cells","authors":"Zulqurnain Sabir , M.A. Abdelkawy , Dumitru Baleanu , Ozlem Defterli","doi":"10.1016/j.knosys.2026.115493","DOIUrl":"10.1016/j.knosys.2026.115493","url":null,"abstract":"<div><h3>Objective</h3><div>The aim of current research is to present the numerical performances of the cancer treatment model based on chemotherapy and stem cells using one of the heuristic computing neural network procedures. The cancer treatment model through chemotherapy and stem cells is categorized into stem cells, affected cells, tumor cells, and chemotherapy-based concentration drug.</div></div><div><h3>Method</h3><div>A process of artificial neural network is applied using the hybrid optimization of global and local search schemes, which are taken as genetic algorithm (GA) and an active set (AS). An error-based fitness function is designed by using the differential model and then optimized by the hybridization of both global and local search schemes. GA is applied to exploit the global result and give a primary guess to the AS that further improves the results locally. AS is rooted in the GA, where GA produces new populaces and AS optimizes the fitness function for every individual. The hybridization of these two schemes is used iteratively for purifying the results. Ten numbers of neurons and log-sigmoid activation functions has been used to solve the cancer treatment model based on chemotherapy and stem cells.</div></div><div><h3>Results</h3><div>For the correctness of the stochastic solver, the obtained numerical results have been compared with any traditional scheme. Moreover, the reliability and capability of the scheme are performed through the absolute error around 10<sup>-05</sup> to 10<sup>-07</sup> along with different statistical approaches for solving the mathematical model.</div></div><div><h3>Novelty</h3><div>The proposed artificial neural network structure along with the hybrid optimization of global and local search schemes has never been implemented before to solve the cancer treatment model based on chemotherapy and stem cells.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115493"},"PeriodicalIF":7.6,"publicationDate":"2026-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174780","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}

引用次数: 0

Graph-Prototype distillation with prototype-Guided contrastive training for multimodal emotion recognition in conversations 对话中多模态情感识别的图-原型升华与原型引导对比训练

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-04 DOI: 10.1016/j.knosys.2026.115484

Bengong Yu , Jun Wang , Chenyue Li , Zhonghao Xi , Xianxian Zhao , Yue Li

Multimodal Emotion Recognition in Conversations aims to determine utterance-level emotions robustly when heterogeneous textual, acoustic, and visual signals intertwine and the dialogue context evolves across turns. Although graph-based dialogue methods have made progress, decision calibration, geometric alignment, and class-level organization are often modeled in isolation when modality conflicts coexist with cross-turn context shifts. This promotes information diffusion and structural redundancy, thereby hampering the separability of weakly distinguishable emotions and overall robustness. To address these issues, we introduce Graph-Prototype Distillation with Prototype-Guided Contrastive Training (GPGC), which jointly constrains representation alignment, distributional consistency, and prototype alignment on a unified intra-modal graph-aggregated representation, thereby tightening intra-class dispersion from both probabilistic and geometric perspectives and stabilizing class-prototype directions. Prototype-guided momentum contrast is further employed to leverage a cross-batch stable dictionary and guided positives to consistently enlarge margins against hard negatives while reducing the interference of noisy samples during optimization. Systematic evaluations on two widely used MERC benchmarks and an in-the-wild multimodal sentiment benchmark demonstrate consistent improvements in both overall performance and stability.

对话中的多模态情感识别旨在识别异质文本、声音和视觉信号交织在一起、对话语境跨回合演变时的话语级情感。尽管基于图的对话方法已经取得了进展，但当模态冲突与交叉转弯上下文转换共存时，决策校准、几何对齐和类级组织往往是孤立的建模。这促进了信息扩散和结构冗余，从而阻碍了弱可区分情绪的可分离性和整体鲁棒性。为了解决这些问题，我们引入了带有原型引导对比训练（GPGC）的图-原型蒸馏，它在统一的模态内图聚合表示上联合约束表示对齐、分布一致性和原型对齐，从而从概率和几何角度加强类内分散，并稳定类-原型方向。原型引导动量对比进一步用于利用跨批稳定字典和引导阳性，以一致地扩大硬阴性的边际，同时减少优化过程中噪声样本的干扰。对两个广泛使用的MERC基准和一个野外多模态情绪基准的系统评估表明，两者在整体性能和稳定性方面都有一致的改进。

{"title":"Graph-Prototype distillation with prototype-Guided contrastive training for multimodal emotion recognition in conversations","authors":"Bengong Yu , Jun Wang , Chenyue Li , Zhonghao Xi , Xianxian Zhao , Yue Li","doi":"10.1016/j.knosys.2026.115484","DOIUrl":"10.1016/j.knosys.2026.115484","url":null,"abstract":"<div><div>Multimodal Emotion Recognition in Conversations aims to determine utterance-level emotions robustly when heterogeneous textual, acoustic, and visual signals intertwine and the dialogue context evolves across turns. Although graph-based dialogue methods have made progress, decision calibration, geometric alignment, and class-level organization are often modeled in isolation when modality conflicts coexist with cross-turn context shifts. This promotes information diffusion and structural redundancy, thereby hampering the separability of weakly distinguishable emotions and overall robustness. To address these issues, we introduce Graph-Prototype Distillation with Prototype-Guided Contrastive Training (GPGC), which jointly constrains representation alignment, distributional consistency, and prototype alignment on a unified intra-modal graph-aggregated representation, thereby tightening intra-class dispersion from both probabilistic and geometric perspectives and stabilizing class-prototype directions. Prototype-guided momentum contrast is further employed to leverage a cross-batch stable dictionary and guided positives to consistently enlarge margins against hard negatives while reducing the interference of noisy samples during optimization. Systematic evaluations on two widely used MERC benchmarks and an in-the-wild multimodal sentiment benchmark demonstrate consistent improvements in both overall performance and stability.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115484"},"PeriodicalIF":7.6,"publicationDate":"2026-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175311","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}

引用次数: 0

GADet: Geometry-Aware oriented object detection for remote sensing 面向遥感的几何感知目标检测

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-04 DOI: 10.1016/j.knosys.2026.115475

Haodong Li , Yan Gong , Xinyu Zhang , Ziying Song , Lei Yang , Haicheng Qu

Oriented object detection in remote sensing images is a key technology for accurately perceiving the geometric properties of objects on the Earth’s surface, playing a significant role in smart cities, national defense and security, and disaster emergency response. However, existing anchor-free methods have obvious limitations in geometric feature adaptation and orientation-aware modeling, and their large number of parameters makes real-time deployment difficult. To address these issues, we propose the geometry-aware detector GADet, a single-stage anchor-free detector comprising three key components: a geometrically structured adaptive convolution (GSA-Conv) module for enhanced feature extraction, a rotation-sensitive attention (RSA) module for robust orientation awareness, and a channel-isomorphic adaptive (CIA) pruning method for model compression. Comprehensive experiments demonstrate that GADet achieves mAP scores of 76.90%, 70.20%, and 97.47% on the DOTA-v1.0, DIOR-R, and UCAS-AOD datasets, respectively, while running at 56.5 FPS, achieving the optimal balance between accuracy and efficiency compared to recent state-of-the-art methods.

遥感图像定向目标检测是准确感知地球表面物体几何特性的关键技术，在智慧城市、国防安全、灾害应急响应等方面具有重要作用。然而，现有的无锚方法在几何特征自适应和方向感知建模方面存在明显的局限性，且其参数较多，给实时部署带来困难。为了解决这些问题，我们提出了几何感知检测器GADet，这是一种单级无锚检测器，由三个关键组件组成：用于增强特征提取的几何结构自适应卷积（GSA-Conv）模块，用于鲁棒方向感知的旋转敏感注意（RSA）模块，以及用于模型压缩的通道同构自适应（CIA）修剪方法。综合实验表明，GADet在DOTA-v1.0、DIOR-R和UCAS-AOD数据集上的mAP得分分别为76.90%、70.20%和97.47%，运行速度为56.5 FPS，与目前最先进的方法相比，达到了精度和效率的最佳平衡。

{"title":"GADet: Geometry-Aware oriented object detection for remote sensing","authors":"Haodong Li , Yan Gong , Xinyu Zhang , Ziying Song , Lei Yang , Haicheng Qu","doi":"10.1016/j.knosys.2026.115475","DOIUrl":"10.1016/j.knosys.2026.115475","url":null,"abstract":"<div><div>Oriented object detection in remote sensing images is a key technology for accurately perceiving the geometric properties of objects on the Earth’s surface, playing a significant role in smart cities, national defense and security, and disaster emergency response. However, existing anchor-free methods have obvious limitations in geometric feature adaptation and orientation-aware modeling, and their large number of parameters makes real-time deployment difficult. To address these issues, we propose the geometry-aware detector GADet, a single-stage anchor-free detector comprising three key components: a geometrically structured adaptive convolution (GSA-Conv) module for enhanced feature extraction, a rotation-sensitive attention (RSA) module for robust orientation awareness, and a channel-isomorphic adaptive (CIA) pruning method for model compression. Comprehensive experiments demonstrate that GADet achieves mAP scores of 76.90%, 70.20%, and 97.47% on the DOTA-v1.0, DIOR-R, and UCAS-AOD datasets, respectively, while running at 56.5 FPS, achieving the optimal balance between accuracy and efficiency compared to recent state-of-the-art methods.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115475"},"PeriodicalIF":7.6,"publicationDate":"2026-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174788","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}

引用次数: 0

Emo-STCapsnet: A spatio-temporal modeling approach with enhanced CapsNet for speech emotion recognition Emo-STCapsnet：一种基于增强CapsNet的时空建模方法，用于语音情感识别

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-03 DOI: 10.1016/j.knosys.2026.115447

Yonghong Fan , Heming Huang , Huiyun Zhang , Ziqi Zhou

Speech emotion recognition (SER) aims to enable computers to accurately identify emotional states embedded in speech signals, a critical area in human-computer interaction. Effective spatio-temporal feature extraction, which captures consistent emotional patterns while minimizing inter-emotion variability, is critical for SER. However, existing approaches often fall short in learning comprehensive spatio-temporal features. To address this, Emo-STCapsNet, a spatio-temporal modeling approach with enhanced capsule network, is proposed. It integrates four components: a temporal dynamic activation block to capture multi-scale temporal variations, a two-stream attentive fusion for past and future context integration to establish global emotional representations, a convolutional block for high-level feature abstraction from the bidirectional temporal representations, and an attention-enhanced CapsNet that leverages vectorized entity representations and dynamic routing mechanisms to more effectively capture hierarchical spatial relationships among emotional features compared to conventional methods like CNNs. Experimental results on the benchmark SER datasets IEMOCAP, EMODB, and CASIA demonstrate the superior performance of Emo-STCapsNet, achieving accuracies of 71.86%, 93.46%, and 87.92%, respectively. Comparative results highlight the superiority of Emo-STCapsNet approach over other methods. Extensive ablation studies further validate the effectiveness of the architecture of Emo-STCapsNet and underscore the necessity of comprehensive spatio-temporal feature learning in SER.

语音情绪识别（SER）旨在使计算机能够准确识别嵌入语音信号中的情绪状态，这是人机交互的一个关键领域。有效的时空特征提取，既能捕获一致的情绪模式，又能最大限度地减少情绪间的可变性，对SER至关重要。然而，现有的方法在学习综合时空特征方面往往存在不足。为了解决这个问题，Emo-STCapsNet提出了一种具有增强胶囊网络的时空建模方法。它由四个部分组成：一个用于捕获多尺度时间变化的时间动态激活块，一个用于过去和未来情境整合的两流关注融合以建立全局情感表征，一个用于从双向时间表征中提取高级特征的卷积块，以及一个注意力增强的CapsNet，它利用矢量化实体表示和动态路由机制，与cnn等传统方法相比，更有效地捕捉情感特征之间的层次空间关系。在基准SER数据集IEMOCAP、EMODB和CASIA上的实验结果表明，Emo-STCapsNet具有优异的性能，准确率分别达到71.86%、93.46%和87.92%。对比结果表明Emo-STCapsNet方法优于其他方法。大量的消融研究进一步验证了Emo-STCapsNet架构的有效性，并强调了在SER中进行全面的时空特征学习的必要性。

{"title":"Emo-STCapsnet: A spatio-temporal modeling approach with enhanced CapsNet for speech emotion recognition","authors":"Yonghong Fan , Heming Huang , Huiyun Zhang , Ziqi Zhou","doi":"10.1016/j.knosys.2026.115447","DOIUrl":"10.1016/j.knosys.2026.115447","url":null,"abstract":"<div><div>Speech emotion recognition (SER) aims to enable computers to accurately identify emotional states embedded in speech signals, a critical area in human-computer interaction. Effective spatio-temporal feature extraction, which captures consistent emotional patterns while minimizing inter-emotion variability, is critical for SER. However, existing approaches often fall short in learning comprehensive spatio-temporal features. To address this, Emo-STCapsNet, a spatio-temporal modeling approach with enhanced capsule network, is proposed. It integrates four components: a temporal dynamic activation block to capture multi-scale temporal variations, a two-stream attentive fusion for past and future context integration to establish global emotional representations, a convolutional block for high-level feature abstraction from the bidirectional temporal representations, and an attention-enhanced CapsNet that leverages vectorized entity representations and dynamic routing mechanisms to more effectively capture hierarchical spatial relationships among emotional features compared to conventional methods like CNNs. Experimental results on the benchmark SER datasets IEMOCAP, EMODB, and CASIA demonstrate the superior performance of Emo-STCapsNet, achieving accuracies of 71.86%, 93.46%, and 87.92%, respectively. Comparative results highlight the superiority of Emo-STCapsNet approach over other methods. Extensive ablation studies further validate the effectiveness of the architecture of Emo-STCapsNet and underscore the necessity of comprehensive spatio-temporal feature learning in SER.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115447"},"PeriodicalIF":7.6,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175315","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}

引用次数: 0

FeNeC: Enhancing continual learning via feature clustering with neighbor- or logit-based classification FeNeC：通过基于邻居或基于逻辑的分类的特征聚类来增强持续学习

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-03 DOI: 10.1016/j.knosys.2026.115479

Kamil Książek , Hubert Jastrzębski , Krzysztof Pniaczek , Bartosz Trojan , Michał Karp , Jacek Tabor

The ability of deep learning models to learn continuously is essential for adapting to new data categories and evolving data distributions. In recent years, approaches leveraging frozen feature extractors after an initial learning phase have been extensively studied. Many of these methods estimate per-class covariance matrices and prototypes based on backbone-derived feature representations. Within this paradigm, we introduce FeNeC (Feature Neighborhood Classifier) and FeNeC-Log, its variant based on the log-likelihood function. Our approach significantly extends the concept of per-class prototypes by constructing multiple, fine-grained sub-prototypes for each class, thereby enhancing the representation of class distributions. Utilizing the Mahalanobis distance, our models classify samples either through a nearest neighbor assignment to these sub-prototypes or trainable logit values assigned to consecutive classes. Our proposition can be seen as a generalization that reduces to existing single-prototype approaches in a special case, while extending them with the ability for more flexible adaptation to data. We demonstrate that our FeNeC variants establish state-of-the-art results across several benchmarks, proving particularly effective on CIFAR-100 and the complex ImageNet-Subset, where our method outperforms the strong FeCAM baseline by over 1% in average incremental accuracy and 1.5% in last task accuracy.

深度学习模型持续学习的能力对于适应新的数据类别和不断变化的数据分布至关重要。近年来，在初始学习阶段之后利用冻结特征提取器的方法得到了广泛的研究。这些方法中有许多是基于主干衍生的特征表示来估计每个类的协方差矩阵和原型的。在这个范例中，我们引入了FeNeC （Feature Neighborhood Classifier）和FeNeC- log，它是基于对数似然函数的变体。我们的方法通过为每个类构建多个细粒度的子原型，显著扩展了每个类原型的概念，从而增强了类分布的表示。利用马氏距离，我们的模型通过对这些子原型的最近邻分配或分配给连续类的可训练logit值对样本进行分类。我们的命题可以被看作是一种泛化，在特殊情况下减少到现有的单原型方法，同时扩展它们以更灵活地适应数据的能力。我们证明了我们的FeNeC变体在几个基准测试中建立了最先进的结果，证明在CIFAR-100和复杂的imagenet -子集上特别有效，其中我们的方法在平均增量精度上优于强FeCAM基线超过1%，在最后任务精度上优于1.5%。

{"title":"FeNeC: Enhancing continual learning via feature clustering with neighbor- or logit-based classification","authors":"Kamil Książek , Hubert Jastrzębski , Krzysztof Pniaczek , Bartosz Trojan , Michał Karp , Jacek Tabor","doi":"10.1016/j.knosys.2026.115479","DOIUrl":"10.1016/j.knosys.2026.115479","url":null,"abstract":"<div><div>The ability of deep learning models to learn continuously is essential for adapting to new data categories and evolving data distributions. In recent years, approaches leveraging frozen feature extractors after an initial learning phase have been extensively studied. Many of these methods estimate per-class covariance matrices and prototypes based on backbone-derived feature representations. Within this paradigm, we introduce FeNeC (Feature Neighborhood Classifier) and FeNeC-Log, its variant based on the log-likelihood function. Our approach significantly extends the concept of per-class prototypes by constructing multiple, fine-grained sub-prototypes for each class, thereby enhancing the representation of class distributions. Utilizing the Mahalanobis distance, our models classify samples either through a nearest neighbor assignment to these sub-prototypes or trainable logit values assigned to consecutive classes. Our proposition can be seen as a generalization that reduces to existing single-prototype approaches in a special case, while extending them with the ability for more flexible adaptation to data. We demonstrate that our FeNeC variants establish state-of-the-art results across several benchmarks, proving particularly effective on CIFAR-100 and the complex ImageNet-Subset, where our method outperforms the strong FeCAM baseline by over 1% in average incremental accuracy and 1.5% in last task accuracy.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115479"},"PeriodicalIF":7.6,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174785","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}

引用次数: 0

Boundary-aware and multi-angle modeling-based object tracking in polarimetric images 基于边界感知和多角度建模的偏振图像目标跟踪

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-03 DOI: 10.1016/j.knosys.2026.115442

Qiaohui Wang , Fan Shi , Mianzhao Wang , Xinbo Geng , Meng Zhao

Object tracking is a fundamental task in computer vision with applications ranging from surveillance to autonomous driving. Although RGB-based tracking methods have seen significant advancements by leveraging color and texture features, they often struggle under challenging conditions such as low light, occlusions, and fast motion. Polarimetric imaging, which encodes surface properties, material characteristics, and geometric structures, offers unique advantages as a complementary modality. However, its potential remains underexplored due to the lack of large-scale datasets and specialized algorithms designed for polarization-specific features. To address this gap, we introduce POL, the first large-scale benchmark dataset for polarimetric vision that enables comprehensive evaluations under diverse conditions. Building on this dataset, we propose PMTT, a cross-modal transformer framework that integrates polarimetric and RGB data. The Detailed Feature Prompter (DFP) module extracts boundary and multi-angle features from polarimetric images, while the Spatial-Channel Attention (SCA) mechanism enhances feature recognition in complex environments. Extensive experiments confirm that PMTT superior performance and robustness, highlighting the transformative potential of polarimetric imaging for dynamic object tracking.

目标跟踪是计算机视觉的一项基本任务，其应用范围从监视到自动驾驶。尽管基于rgb的跟踪方法通过利用颜色和纹理特征取得了重大进展，但它们经常在低光、遮挡和快速运动等具有挑战性的条件下挣扎。偏振成像，编码表面特性，材料特性和几何结构，提供了独特的优势，作为一种互补的模式。然而，由于缺乏大规模的数据集和专门针对极化特征设计的算法，其潜力仍未得到充分开发。为了解决这一差距，我们引入了POL，这是偏振视觉的第一个大规模基准数据集，可以在不同条件下进行综合评估。在此数据集的基础上，我们提出了PMTT，这是一个集成极化和RGB数据的跨模态变压器框架。细节特征提示器（DFP）模块从偏振图像中提取边界和多角度特征，而空间通道注意（SCA）机制增强了复杂环境下的特征识别。大量的实验证实了PMTT优越的性能和鲁棒性，突出了偏振成像在动态目标跟踪方面的变革潜力。

{"title":"Boundary-aware and multi-angle modeling-based object tracking in polarimetric images","authors":"Qiaohui Wang , Fan Shi , Mianzhao Wang , Xinbo Geng , Meng Zhao","doi":"10.1016/j.knosys.2026.115442","DOIUrl":"10.1016/j.knosys.2026.115442","url":null,"abstract":"<div><div>Object tracking is a fundamental task in computer vision with applications ranging from surveillance to autonomous driving. Although RGB-based tracking methods have seen significant advancements by leveraging color and texture features, they often struggle under challenging conditions such as low light, occlusions, and fast motion. Polarimetric imaging, which encodes surface properties, material characteristics, and geometric structures, offers unique advantages as a complementary modality. However, its potential remains underexplored due to the lack of large-scale datasets and specialized algorithms designed for polarization-specific features. To address this gap, we introduce POL, the first large-scale benchmark dataset for polarimetric vision that enables comprehensive evaluations under diverse conditions. Building on this dataset, we propose PMTT, a cross-modal transformer framework that integrates polarimetric and RGB data. The Detailed Feature Prompter (DFP) module extracts boundary and multi-angle features from polarimetric images, while the Spatial-Channel Attention (SCA) mechanism enhances feature recognition in complex environments. Extensive experiments confirm that PMTT superior performance and robustness, highlighting the transformative potential of polarimetric imaging for dynamic object tracking.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115442"},"PeriodicalIF":7.6,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175321","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}

引用次数: 0

DGFFA: Joint multimodal entity-relation extraction via dual-channel graph fusion and fine-grained alignment DGFFA：通过双通道图融合和细粒度对齐的联合多模态实体关系提取

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-02 DOI: 10.1016/j.knosys.2026.115470

Wenjie Liu , Xingwen Li , Zhijie Ren

Joint multimodal entity and relation extraction (JMERE) is a key task in multimodal knowledge graph completion (MKGC), aimed at integrating textual and visual information for better knowledge representation and semantic reasoning. However, existing paradigms often struggle with suboptimal cross-modal alignment and typically neglect the intrinsic correlations between entities and relations within word-pair structures. To tackle these challenges, we propose a JMERE framework via Dual-Channel Graph Fusion and Fine-Grained Alignment, namely DGFFA. Specifically, a fine-grained cross-modal alignment module is designed, which leverages token-patch similarity priors from a pre-trained vision-language model to guide optimal-transport matching, which suppresses noisy visual regions and yields more precise multimodal correspondences. To fully leverage the connections between entities and relationships, a dual-channel graph architecture was designed to jointly optimize the representations of nodes and edges in a unified prediction space, thereby effectively modeling bidirectional dependencies. Extensive experiments demonstrate that our model consistently outperforms state-of-the-art methods such as EEGA and TESGA, achieving average improvements of 2.4%, 3.2%, and 1.6% in Precision, Recall, and F1 on JMERE tasks. Our approach not only offers a new paradigm for multimodal entity-relation extraction, but also contributes novel insights into multimodal knowledge graph construction and unified multimodal reasoning.

联合多模态实体与关系提取（JMERE）是多模态知识图完成（MKGC）中的关键任务，旨在整合文本信息和视觉信息，以获得更好的知识表示和语义推理。然而，现有的范式经常与次优的跨模态对齐作斗争，并且通常忽略了词对结构中实体和关系之间的内在相关性。为了解决这些挑战，我们提出了一个通过双通道图融合和细粒度对齐的JMERE框架，即DGFFA。具体来说，设计了一个细粒度的跨模态对齐模块，该模块利用来自预训练的视觉语言模型的标记补丁相似性先验来指导最佳传输匹配，从而抑制噪声视觉区域并产生更精确的多模态对应。为了充分利用实体和关系之间的联系，设计了双通道图架构，在统一的预测空间中共同优化节点和边的表示，从而有效地建模双向依赖关系。大量的实验表明，我们的模型始终优于最先进的方法，如EEGA和TESGA，在JMERE任务的Precision， Recall和F1方面实现了2.4%，3.2%和1.6%的平均改进。该方法不仅为多模态实体关系提取提供了新的范式，而且对多模态知识图谱的构建和统一的多模态推理也有新的见解。

{"title":"DGFFA: Joint multimodal entity-relation extraction via dual-channel graph fusion and fine-grained alignment","authors":"Wenjie Liu , Xingwen Li , Zhijie Ren","doi":"10.1016/j.knosys.2026.115470","DOIUrl":"10.1016/j.knosys.2026.115470","url":null,"abstract":"<div><div>Joint multimodal entity and relation extraction (JMERE) is a key task in multimodal knowledge graph completion (MKGC), aimed at integrating textual and visual information for better knowledge representation and semantic reasoning. However, existing paradigms often struggle with suboptimal cross-modal alignment and typically neglect the intrinsic correlations between entities and relations within word-pair structures. To tackle these challenges, we propose a JMERE framework via Dual-Channel Graph Fusion and Fine-Grained Alignment, namely DGFFA. Specifically, a fine-grained cross-modal alignment module is designed, which leverages token-patch similarity priors from a pre-trained vision-language model to guide optimal-transport matching, which suppresses noisy visual regions and yields more precise multimodal correspondences. To fully leverage the connections between entities and relationships, a dual-channel graph architecture was designed to jointly optimize the representations of nodes and edges in a unified prediction space, thereby effectively modeling bidirectional dependencies. Extensive experiments demonstrate that our model consistently outperforms state-of-the-art methods such as EEGA and TESGA, achieving average improvements of 2.4%, 3.2%, and 1.6% in Precision, Recall, and F1 on JMERE tasks. Our approach not only offers a new paradigm for multimodal entity-relation extraction, but also contributes novel insights into multimodal knowledge graph construction and unified multimodal reasoning.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115470"},"PeriodicalIF":7.6,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174779","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}

引用次数: 0

FedPDM: Representation enhanced federated learning with privacy preserving diffusion models FedPDM：带有隐私保护扩散模型的表示增强联邦学习

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-02 DOI: 10.1016/j.knosys.2026.115452

Wei Guo , Fuzhen Zhuang , Yiqi Tong , Xiao Zhang , Zhaojun Hu , Jiejie Zhao , Jin Dong

Most existing semi-parameter-sharing federated learning (FL) frameworks utilize generative models to achieve partial parameter sharing with the server, which effectively enhances the data privacy of each client. However, these generative models often suffer from model utility degradation due to poor representation robustness. Meanwhile, representation inconsistency between local and global models exacerbates the client drift problem under non-IID scenarios. Furthermore, existing semi-parameter-sharing FL frameworks overlook representation leakage risks associated with generator sharing, while failing to balance privacy and utility. To alleviate these challenges, we propose FedPDM, a semi-parameter-sharing FL framework built upon a privacy-preserving diffusion model (PDM). Specifically, our proposed PDM enables model alignment with features from the privacy extractor without requiring direct exposure of this extractor, effectively mitigating utility degradation caused by poor representation robustness. Moreover, a feature-level penalty term is introduced into the optimization objective of PDM to avoid representation leakage. We further design a two-stage aggregation strategy that addresses representation inconsistency through initialization correction with a Gaussian constraint for knowledge distillation. Finally, we provide the first theoretical convergence analysis for semi-parameter-sharing FL, demonstrating that our framework converges at a rate of

O (1 / T)

. Extensive experiments on four datasets show that FedPDM achieves average accuracy improvements of 1.78% to 5.56% compared with various state-of-the-art baselines.

现有的半参数共享联邦学习（FL）框架大多利用生成模型实现与服务器的部分参数共享，有效增强了各客户端的数据隐私性。然而，由于表示鲁棒性差，这些生成模型往往存在模型效用下降的问题。同时，局部模型和全局模型之间的表示不一致加剧了非iid场景下的客户端漂移问题。此外，现有的半参数共享FL框架忽略了与生成器共享相关的表示泄漏风险，同时未能平衡隐私和实用性。为了缓解这些挑战，我们提出了一种基于隐私保护扩散模型（PDM）的半参数共享FL框架FedPDM。具体来说，我们提出的PDM使模型与隐私提取器的特征保持一致，而不需要直接暴露该提取器，有效地减轻了由于表示鲁棒性差而导致的效用下降。此外，在PDM的优化目标中引入了特征级惩罚项，以避免表示泄漏。我们进一步设计了一种两阶段聚合策略，通过初始化校正和知识蒸馏的高斯约束来解决表示不一致问题。最后，我们提供了半参数共享FL的第一个理论收敛分析，证明了我们的框架以0 （1/T）的速度收敛。在4个数据集上的大量实验表明，与各种最先进的基线相比，FedPDM的平均准确率提高了1.78% ~ 5.56%。

{"title":"FedPDM: Representation enhanced federated learning with privacy preserving diffusion models","authors":"Wei Guo , Fuzhen Zhuang , Yiqi Tong , Xiao Zhang , Zhaojun Hu , Jiejie Zhao , Jin Dong","doi":"10.1016/j.knosys.2026.115452","DOIUrl":"10.1016/j.knosys.2026.115452","url":null,"abstract":"<div><div>Most existing semi-parameter-sharing federated learning (FL) frameworks utilize generative models to achieve partial parameter sharing with the server, which effectively enhances the data privacy of each client. However, these generative models often suffer from model utility degradation due to <em>poor representation robustness</em>. Meanwhile, <em>representation inconsistency</em> between local and global models exacerbates the client drift problem under non-IID scenarios. Furthermore, existing semi-parameter-sharing FL frameworks overlook <em>representation leakage</em> risks associated with generator sharing, while failing to balance privacy and utility. To alleviate these challenges, we propose FedPDM, a semi-parameter-sharing FL framework built upon a privacy-preserving diffusion model (PDM). Specifically, our proposed PDM enables model alignment with features from the privacy extractor without requiring direct exposure of this extractor, effectively mitigating utility degradation caused by poor representation robustness. Moreover, a feature-level penalty term is introduced into the optimization objective of PDM to avoid representation leakage. We further design a two-stage aggregation strategy that addresses representation inconsistency through initialization correction with a Gaussian constraint for knowledge distillation. Finally, we provide the first theoretical convergence analysis for semi-parameter-sharing FL, demonstrating that our framework converges at a rate of <span><math><mrow><mi>O</mi><mo>(</mo><mn>1</mn><mo>/</mo><mi>T</mi><mo>)</mo></mrow></math></span>. Extensive experiments on four datasets show that FedPDM achieves average accuracy improvements of 1.78% to 5.56% compared with various state-of-the-art baselines.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115452"},"PeriodicalIF":7.6,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174781","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}

引用次数: 0

PLeFF-Net: parallel Le-Net forward fractional network for sarcoma cancer detection using histopathological image PLeFF-Net：利用组织病理图像检测肉瘤的平行Le-Net正向分数网络

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-02 DOI: 10.1016/j.knosys.2026.115476

Balashanmuga Vadivu Palanivel , Gopalsamy Venkadakrishnan Sriramakrishnan , Vadamodula Prasad , Subbiah Vairamuthu , Deena Gnanasekaran , Ilavarasan Sargunan

A sarcoma is considered a rare kind of tumor that generally occurs in various connective tissues that surround and support different organs and bones in the human body. Sarcoma cancer mainly affects the connective tissues, like nerves, blood vessels, muscles, fat, joints, and bones. The primary symptoms of sarcoma cancer are based on the location and size of tumors. Various imaging techniques have been recently applied to identify sarcoma cancer. However, these approaches do not adequately detect the cell response of individuals. Thus, a novel Deep Learning (DL) model, Parallel Le-Net Forward Fractional Network (PLeFF-Net), is proposed for the detection of sarcoma cancer from histopathological images. The histopathological images are primarily preprocessed utilizing homomorphic filtering, and then a Fully Convolutional Neural Network (FCNN) is exploited to segment the cells. Later, image-level features, shape-based features, color-based features, and network-level features are extracted from the segmented areas. For sarcoma cancer detection, the PLeFF-Net model receives the extracted features as its input. Using the k-fold cross-validation for k-value 8 on dataset 2, the proposed PLeFF-Net revealed superior performance with maximum accuracy of 93.789%, True Positive Rate (TNR) of 94.667%, True Negative Rate (TNR) of 92.100%, Precision of 93.357%, and F1-score of 94.007%.

肉瘤被认为是一种罕见的肿瘤，通常发生在围绕和支持人体不同器官和骨骼的各种结缔组织中。肉瘤主要影响结缔组织，如神经、血管、肌肉、脂肪、关节和骨骼。肉瘤的主要症状是基于肿瘤的位置和大小。近年来，各种影像技术已被应用于肉瘤的鉴别。然而，这些方法不能充分检测个体的细胞反应。因此，提出了一种新的深度学习（DL）模型，并行Le-Net前向分数网络（PLeFF-Net），用于从组织病理学图像中检测肉瘤癌症。利用同态滤波对组织病理图像进行预处理，然后利用全卷积神经网络（FCNN）对细胞进行分割。然后，从分割的区域中提取图像级特征、形状级特征、颜色级特征和网络级特征。对于肉瘤癌症检测，PLeFF-Net模型接收提取的特征作为其输入。在数据集2上对k值8进行k-fold交叉验证，所提出的PLeFF-Net的最大准确率为93.789%，真阳性率（TNR）为94.667%，真阴性率（TNR）为92.100%，精度为93.357%，f1得分为94.007%。

{"title":"PLeFF-Net: parallel Le-Net forward fractional network for sarcoma cancer detection using histopathological image","authors":"Balashanmuga Vadivu Palanivel , Gopalsamy Venkadakrishnan Sriramakrishnan , Vadamodula Prasad , Subbiah Vairamuthu , Deena Gnanasekaran , Ilavarasan Sargunan","doi":"10.1016/j.knosys.2026.115476","DOIUrl":"10.1016/j.knosys.2026.115476","url":null,"abstract":"<div><div>A sarcoma is considered a rare kind of tumor that generally occurs in various connective tissues that surround and support different organs and bones in the human body. Sarcoma cancer mainly affects the connective tissues, like nerves, blood vessels, muscles, fat, joints, and bones. The primary symptoms of sarcoma cancer are based on the location and size of tumors. Various imaging techniques have been recently applied to identify sarcoma cancer. However, these approaches do not adequately detect the cell response of individuals. Thus, a novel Deep Learning (DL) model, Parallel Le-Net Forward Fractional Network (PLeFF-Net), is proposed for the detection of sarcoma cancer from histopathological images. The histopathological images are primarily preprocessed utilizing homomorphic filtering, and then a Fully Convolutional Neural Network (FCNN) is exploited to segment the cells. Later, image-level features, shape-based features, color-based features, and network-level features are extracted from the segmented areas. For sarcoma cancer detection, the PLeFF-Net model receives the extracted features as its input. Using the k-fold cross-validation for k-value 8 on dataset 2, the proposed PLeFF-Net revealed superior performance with maximum accuracy of 93.789%, True Positive Rate (TNR) of 94.667%, True Negative Rate (TNR) of 92.100%, Precision of 93.357%, and F1-score of 94.007%.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115476"},"PeriodicalIF":7.6,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175180","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}

引用次数: 0

A software-defined wrapper discriminant federated learning-reinforcement attention adversarial regression approach for privacy and task management in cloud edge computing 一种用于云边缘计算隐私和任务管理的软件定义包装器判别联合学习强化注意对抗回归方法

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-02 DOI: 10.1016/j.knosys.2026.115430

K. Thangaraj , Koppula Srinivas Rao , Kayam Saikumar , Shruti Garg

In recent years, the widespread adoption of the Internet of Things has played an important role in advancing artificial intelligence by continuously generating large volumes of data used for model training and decision-making processes. The conventional cloud edge computing paradigm faces challenges in handling the massive data generated by Internet of Things. These challenges include high latency, excessive bandwidth usage, limited scalability, and privacy risks. To rectify these limitations, this research proposes a novel Software-defined Wrapper Discriminant Federated learning-Reinforcement Attention Adversarial Regression (SWDF-RAAR) model. Unlike other existing studies, the SWDF-RAAR model jointly addresses privacy analysis and intelligent resource task management within a unified structure. For privacy analysis, the model integrates software-defined networking, linear discriminant analysis with wrapper-style bi-directional removal technique, federated learning, and extreme learning machines. Resource task management is performed using a combination of dynamic perceptrons, scaled dot-product attention, a multilayer perceptron with graph convolution, deep Q-learning aided by generative adversarial networks, and a support vector machine. The main contribution of this model is to develop a scalable, lightweight, and privacy-preserving model for detecting intrusions while enhancing the efficiency of resource task management in cloud edge computing environments. Experimental analysis on various security based datasets and cloud-edge parameters demonstrated that the proposed model attained 98.73% detection accuracy, 93% scalability, 95% quality of service, 96% network efficiency, and 5.3 ms latency.

近年来，物联网的广泛采用，通过不断产生用于模型训练和决策过程的大量数据，在推进人工智能方面发挥了重要作用。传统的云边缘计算模式在处理物联网产生的海量数据时面临挑战。这些挑战包括高延迟、过多的带宽使用、有限的可伸缩性和隐私风险。为了纠正这些局限性，本研究提出了一种新的软件定义包装器判别联邦学习-强化注意对抗回归（SWDF-RAAR）模型。与其他现有研究不同，SWDF-RAAR模型在统一的结构中联合解决了隐私分析和智能资源任务管理。对于隐私分析，该模型集成了软件定义网络、线性判别分析（带有包装式双向删除技术）、联邦学习和极限学习机。资源任务管理使用动态感知器、缩放点积注意、具有图卷积的多层感知器、生成对抗网络辅助的深度q学习和支持向量机的组合来执行。该模型的主要贡献是开发了一种可扩展、轻量级和隐私保护的模型，用于检测入侵，同时提高了云边缘计算环境中资源任务管理的效率。在各种基于安全的数据集和云边缘参数上的实验分析表明，该模型达到了98.73%的检测准确率、93%的可扩展性、95%的服务质量、96%的网络效率和5.3 ms的延迟。

{"title":"A software-defined wrapper discriminant federated learning-reinforcement attention adversarial regression approach for privacy and task management in cloud edge computing","authors":"K. Thangaraj , Koppula Srinivas Rao , Kayam Saikumar , Shruti Garg","doi":"10.1016/j.knosys.2026.115430","DOIUrl":"10.1016/j.knosys.2026.115430","url":null,"abstract":"<div><div>In recent years, the widespread adoption of the Internet of Things has played an important role in advancing artificial intelligence by continuously generating large volumes of data used for model training and decision-making processes. The conventional cloud edge computing paradigm faces challenges in handling the massive data generated by Internet of Things. These challenges include high latency, excessive bandwidth usage, limited scalability, and privacy risks. To rectify these limitations, this research proposes a novel Software-defined Wrapper Discriminant Federated learning-Reinforcement Attention Adversarial Regression (SWDF-RAAR) model. Unlike other existing studies, the SWDF-RAAR model jointly addresses privacy analysis and intelligent resource task management within a unified structure. For privacy analysis, the model integrates software-defined networking, linear discriminant analysis with wrapper-style bi-directional removal technique, federated learning, and extreme learning machines. Resource task management is performed using a combination of dynamic perceptrons, scaled dot-product attention, a multilayer perceptron with graph convolution, deep Q-learning aided by generative adversarial networks, and a support vector machine. The main contribution of this model is to develop a scalable, lightweight, and privacy-preserving model for detecting intrusions while enhancing the efficiency of resource task management in cloud edge computing environments. Experimental analysis on various security based datasets and cloud-edge parameters demonstrated that the proposed model attained 98.73% detection accuracy, 93% scalability, 95% quality of service, 96% network efficiency, and 5.3 ms latency.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115430"},"PeriodicalIF":7.6,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174786","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}

引用次数: 0