Pub Date : 2026-05-01Epub Date: 2026-02-11DOI: 10.1016/j.neucom.2026.133039
Yan Kang , Jiajun Tang , Baochen Fan , Hu Yuan
As complex networks grow, community detection aids social network clustering but also risks exposing sensitive ties. Community deception alters network structures to hide target communities and protect privacy. Existing deception approaches primarily rely on either node-level or edge-level interventions, yet they often neglect the heterogeneous influence of individual nodes and edges, resulting in suboptimal concealment performance. To address these limitations, we propose CDNE, a novel community deception model from both node and edge perspectives. The model integrates node-based community deceptions into edge-based community deceptions for the first time, thus expanding the feasible manipulation space and enabling more flexible and effective deceptions. Moreover, we theoretically study the effects of inter-community and intra-community edge adding and deleting operations as a deception optimization function. Experiments on twenty-five community structure partitions generated by five real-world network datasets and five community detection algorithms show that CDNE consistently outperforms existing state-of-the-art deception methods.
{"title":"CDNE: Community deception from node and edge perspectives","authors":"Yan Kang , Jiajun Tang , Baochen Fan , Hu Yuan","doi":"10.1016/j.neucom.2026.133039","DOIUrl":"10.1016/j.neucom.2026.133039","url":null,"abstract":"<div><div>As complex networks grow, community detection aids social network clustering but also risks exposing sensitive ties. Community deception alters network structures to hide target communities and protect privacy. Existing deception approaches primarily rely on either node-level or edge-level interventions, yet they often neglect the heterogeneous influence of individual nodes and edges, resulting in suboptimal concealment performance. To address these limitations, we propose CDNE, a novel community deception model from both node and edge perspectives. The model integrates node-based community deceptions into edge-based community deceptions for the first time, thus expanding the feasible manipulation space and enabling more flexible and effective deceptions. Moreover, we theoretically study the effects of inter-community and intra-community edge adding and deleting operations as a deception optimization function. Experiments on twenty-five community structure partitions generated by five real-world network datasets and five community detection algorithms show that CDNE consistently outperforms existing state-of-the-art deception methods.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133039"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146173295","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-17DOI: 10.1016/j.neucom.2026.133078
Han Xue , Yuanyuan Zhang , Chenyang Shi , Seakweng Vong
This paper addresses the exponential stability of delayed neural networks by extending existing integral inequalities. Specifically, we consider a refined Lyapunov-Krasovskii function that incorporates decay rate terms and employs the new integral inequalities containing additional information on the delay-dependent terms to derive less conservative stability criteria for delayed neural networks. Numerical simulations validate the effectiveness of the proposed approach, demonstrating its potential for practical implementation in the control design of delayed neural networks and highlighting its advantages over traditional methods.
{"title":"New integral inequalities for exponential stability of neural networks with time-varying delay","authors":"Han Xue , Yuanyuan Zhang , Chenyang Shi , Seakweng Vong","doi":"10.1016/j.neucom.2026.133078","DOIUrl":"10.1016/j.neucom.2026.133078","url":null,"abstract":"<div><div>This paper addresses the exponential stability of delayed neural networks by extending existing integral inequalities. Specifically, we consider a refined Lyapunov-Krasovskii function that incorporates decay rate terms and employs the new integral inequalities containing additional information on the delay-dependent terms to derive less conservative stability criteria for delayed neural networks. Numerical simulations validate the effectiveness of the proposed approach, demonstrating its potential for practical implementation in the control design of delayed neural networks and highlighting its advantages over traditional methods.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133078"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386326","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Physics-informed neural networks (PINNs) and related methods struggle to resolve sharp gradients in singularly perturbed boundary value problems without resorting to some form of domain decomposition, which often introduces complex interface penalties. While the Extreme Theory of Functional Connections (X-TFC), thanks to the TFC component, avoids multi-objective optimization by employing exact boundary condition enforcement, it remains computationally inefficient for boundary layers and incompatible with decomposition. We propose Gated X-TFC, a novel framework for both forward and inverse problems, that overcomes these limitations through a soft, learned domain decomposition. Our method replaces hard interfaces with a differentiable logistic gate that dynamically adapts radial basis function (RBF) kernel widths across the domain, eliminating the need for interface penalties. This approach yields not only superior accuracy but also significant improvements in computational efficiency: on a benchmark one dimensional (1D) convection–diffusion, Gated X-TFC achieves an order-of-magnitude lower error than standard X-TFC while using 80% fewer collocation points and reducing training time by 66%. In addition, we introduce an operator-conditioned meta-learning layer that learns a probabilistic mapping from PDE parameters to optimal gate configurations, enabling fast, uncertainty-aware warm-starting for new problem instances. We further demonstrate extensibility to multiple subdomains and higher dimensions by solving a twin boundary–layer equation and a 2D Poisson problem with a sharp Gaussian source. Overall, Gated X-TFC delivers a simple alternative to PINNs that is both accurate and computationally efficient for challenging boundary-layer regimes. Future work will focus on nonlinear problems. For reproducibility, all the codes are available at https://github.com/vikas-dwivedi-2022/gated_xtfc
{"title":"Gated X-TFC: Soft domain decomposition for forward and inverse problems in sharp-gradient PDEs","authors":"Vikas Dwivedi , Enrico Schiassi , Bruno Sixou , Monica Sigovan","doi":"10.1016/j.neucom.2026.133090","DOIUrl":"10.1016/j.neucom.2026.133090","url":null,"abstract":"<div><div>Physics-informed neural networks (PINNs) and related methods struggle to resolve sharp gradients in singularly perturbed boundary value problems without resorting to some form of domain decomposition, which often introduces complex interface penalties. While the Extreme Theory of Functional Connections (X-TFC), thanks to the TFC component, avoids multi-objective optimization by employing exact boundary condition enforcement, it remains computationally inefficient for boundary layers and incompatible with decomposition. We propose Gated X-TFC, a novel framework for both forward and inverse problems, that overcomes these limitations through a soft, learned domain decomposition. Our method replaces hard interfaces with a differentiable logistic gate that dynamically adapts radial basis function (RBF) kernel widths across the domain, eliminating the need for interface penalties. This approach yields not only superior accuracy but also significant improvements in computational efficiency: on a benchmark one dimensional (1D) convection–diffusion, Gated X-TFC achieves an order-of-magnitude lower error than standard X-TFC while using 80% fewer collocation points and reducing training time by 66%. In addition, we introduce an operator-conditioned meta-learning layer that learns a probabilistic mapping from PDE parameters to optimal gate configurations, enabling fast, uncertainty-aware warm-starting for new problem instances. We further demonstrate extensibility to multiple subdomains and higher dimensions by solving a twin boundary–layer equation and a 2D Poisson problem with a sharp Gaussian source. Overall, Gated X-TFC delivers a simple alternative to PINNs that is both accurate and computationally efficient for challenging boundary-layer regimes. Future work will focus on nonlinear problems. For reproducibility, all the codes are available at <span><span>https://github.com/vikas-dwivedi-2022/gated_xtfc</span><svg><path></path></svg></span></div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133090"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386371","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-09DOI: 10.1016/j.neucom.2026.132999
Zhenghui Luo, Min Meng, Jigang Wu
Cross-modal retrieval aims to establish semantic associations between heterogeneous modalities, among which image-text retrieval is a key application scenario that seeks to achieve efficient semantic alignment between images and texts. Existing approaches often rely on fixed patch selection strategies for fine-grained alignment. However, such static strategies struggle to adapt to complex scene variations. Moreover, fine-grained alignment methods tend to fall into local optima by overemphasizing local feature details while neglecting global semantic context. Such limitations significantly hinder both retrieval accuracy and generalization performance. To address these challenges, we propose a Dynamic Patch Selection and Dual-Granularity Alignment (DPSDGA) framework that jointly enhances global semantic consistency and local feature interactions for robust cross-modal alignment. Specifically, we introduce a dynamic sparse module that adaptively adjusts the number of retained visual patches based on scene complexity, effectively filtering redundant information while preserving critical semantic features. Furthermore, we design a dual-granularity alignment mechanism, which combines global contrastive learning with local fine-grained alignment to enhance semantic consistency across modalities. Extensive experiments on two benchmark datasets, Flickr30k and MS-COCO, demonstrate that our method significantly outperforms existing approaches in image-text retrieval.
{"title":"Dynamic patch selection and dual-granularity alignment for cross-modal retrieval","authors":"Zhenghui Luo, Min Meng, Jigang Wu","doi":"10.1016/j.neucom.2026.132999","DOIUrl":"10.1016/j.neucom.2026.132999","url":null,"abstract":"<div><div>Cross-modal retrieval aims to establish semantic associations between heterogeneous modalities, among which image-text retrieval is a key application scenario that seeks to achieve efficient semantic alignment between images and texts. Existing approaches often rely on fixed patch selection strategies for fine-grained alignment. However, such static strategies struggle to adapt to complex scene variations. Moreover, fine-grained alignment methods tend to fall into local optima by overemphasizing local feature details while neglecting global semantic context. Such limitations significantly hinder both retrieval accuracy and generalization performance. To address these challenges, we propose a Dynamic Patch Selection and Dual-Granularity Alignment (DPSDGA) framework that jointly enhances global semantic consistency and local feature interactions for robust cross-modal alignment. Specifically, we introduce a dynamic sparse module that adaptively adjusts the number of retained visual patches based on scene complexity, effectively filtering redundant information while preserving critical semantic features. Furthermore, we design a dual-granularity alignment mechanism, which combines global contrastive learning with local fine-grained alignment to enhance semantic consistency across modalities. Extensive experiments on two benchmark datasets, Flickr30k and MS-COCO, demonstrate that our method significantly outperforms existing approaches in image-text retrieval.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 132999"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146172689","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-11DOI: 10.1016/j.neucom.2026.133007
Zhihao Li, Shanshan Zhang, Jian Yang
3D occupancy prediction estimates the occupancy status in 3D space and is crucial for safe autonomous driving. Existing query-based methods improve efficiency but suffer from challenges such as poor quality of sampled features, semantic inconsistency of prediction results, and so on. To address these challenges, we propose ASHSR, an efficient occupancy prediction network driven by three innovative modules in the decoder. First, considering the limitations of the existing sampling approaches, we propose a view-aware Anti-occlusion Sampling Module (ASM), which significantly improves the quality of sampled features by adaptively avoiding meaningless sampling along the optical ray. Second, to resolve the ambiguity of the one-to-many prediction paradigm in single-query prediction, we design a Hard Sample Reweighting (HSR) module, which effectively improves the prediction purity and semantic consistency by assigning higher weights to the identified hard samples. Finally, to mitigate the cross-class data imbalance in driving scenes, we develop a hierarchical Coarse-to-Fine Supervision (CFS) mechanism to customize reasonable supervision signals across decoder layers, which significantly enhances the perception of rare categories. Extensive experiments on the Occ3D-nuScenes dataset demonstrate that our ASHSR significantly outperforms the existing state-of-the-art methods in terms of the RayIoU metric while maintaining superior inference efficiency (FPS).
{"title":"ASHSR: Enhancing query-based occupancy prediction via anti-occlusion sampling and hard sample reweighting","authors":"Zhihao Li, Shanshan Zhang, Jian Yang","doi":"10.1016/j.neucom.2026.133007","DOIUrl":"10.1016/j.neucom.2026.133007","url":null,"abstract":"<div><div>3D occupancy prediction estimates the occupancy status in 3D space and is crucial for safe autonomous driving. Existing query-based methods improve efficiency but suffer from challenges such as poor quality of sampled features, semantic inconsistency of prediction results, and so on. To address these challenges, we propose ASHSR, an efficient occupancy prediction network driven by three innovative modules in the decoder. First, considering the limitations of the existing sampling approaches, we propose a view-aware Anti-occlusion Sampling Module (ASM), which significantly improves the quality of sampled features by adaptively avoiding meaningless sampling along the optical ray. Second, to resolve the ambiguity of the one-to-many prediction paradigm in single-query prediction, we design a Hard Sample Reweighting (HSR) module, which effectively improves the prediction purity and semantic consistency by assigning higher weights to the identified hard samples. Finally, to mitigate the cross-class data imbalance in driving scenes, we develop a hierarchical Coarse-to-Fine Supervision (CFS) mechanism to customize reasonable supervision signals across decoder layers, which significantly enhances the perception of rare categories. Extensive experiments on the Occ3D-nuScenes dataset demonstrate that our ASHSR significantly outperforms the existing state-of-the-art methods in terms of the RayIoU metric while maintaining superior inference efficiency (FPS).</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133007"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146172690","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-04DOI: 10.1016/j.neucom.2026.132953
Junho Lee , Jisu Yoon , Jisong Kim , Jun Won Choi
Semantic segmentation models trained on source domains often fail to generalize to unseen domains due to domain shifts caused by varying environmental conditions. While existing approaches rely solely on text prompts for domain randomization, their generated styles often deviate from real-world distributions. To address this limitation, we propose a novel two-stage framework for Domain Generalization in Semantic Segmentation (DGSS). First, we introduce Image-Prompt-driven Instance Normalization (I-PIN), which leverages both style images and text prompts to optimize style parameters, achieving more accurate style representations compared to text-only approaches. Second, we present Dual-Path Style-Invariant Feature Learning (DSFL) which employs inter-style and intra-style consistency losses, ensuring consistent predictions across different styles while promoting feature alignment within semantic classes. Extensive experiments demonstrate that our approach consistently outperforms existing state-of-the-art methods across multiple challenging domains, effectively addressing the domain shift problem in semantic segmentation.
{"title":"Image-text driven style randomization for domain generalized semantic segmentation","authors":"Junho Lee , Jisu Yoon , Jisong Kim , Jun Won Choi","doi":"10.1016/j.neucom.2026.132953","DOIUrl":"10.1016/j.neucom.2026.132953","url":null,"abstract":"<div><div>Semantic segmentation models trained on source domains often fail to generalize to unseen domains due to domain shifts caused by varying environmental conditions. While existing approaches rely solely on text prompts for domain randomization, their generated styles often deviate from real-world distributions. To address this limitation, we propose a novel two-stage framework for <em>Domain Generalization in Semantic Segmentation</em> (DGSS). First, we introduce <em>Image-Prompt-driven Instance Normalization</em> (I-PIN), which leverages both style images and text prompts to optimize style parameters, achieving more accurate style representations compared to text-only approaches. Second, we present <em>Dual-Path Style-Invariant Feature Learning</em> (DSFL) which employs inter-style and intra-style consistency losses, ensuring consistent predictions across different styles while promoting feature alignment within semantic classes. Extensive experiments demonstrate that our approach consistently outperforms existing state-of-the-art methods across multiple challenging domains, effectively addressing the domain shift problem in semantic segmentation.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 132953"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146172681","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-14DOI: 10.1016/j.neucom.2026.133043
Tao Wen , Yixue Shen , Xia Fang , Zhongbei Tian , Clive Roberts
Fault diagnosis technology based on Transfer Learning (TL) has been widely investigated, as it enables knowledge transfer from related domains to the target domain, thereby reducing the need for large amounts of labeled data in the target domain. In the railway industry, Railway Point Machines (RPMs) are essential track-connection devices. However, due to the scarcity of fault data for RPMs, the accuracy of RPM fault diagnosis remains low. To address this challenge, this paper proposes a Two-step Transfer Learning (TSTL) method for RPM fault diagnosis. First, a small set of target-domain samples transformed by continuous wavelet transform (CWT) is used to construct a transition dataset, which is fed into a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) to generate synthetic samples. Then, a Convolutional Neural Network (CNN) with a top-down attention mechanism is employed to perform the first phase of TL from the source domain to the generated domain. Next, the parameters of selected frozen layers in the transferred CNN are retained to conduct the second phase of learning from the generated domain to the target domain. Experimental results demonstrate that the proposed method achieves 98.7% accuracy for fault diagnosis across different positions on the same RPM, and 95.6% accuracy across different RPMs at the same position.
{"title":"A two-step transfer learning approach for railway point machine fault diagnosis under small sample conditions","authors":"Tao Wen , Yixue Shen , Xia Fang , Zhongbei Tian , Clive Roberts","doi":"10.1016/j.neucom.2026.133043","DOIUrl":"10.1016/j.neucom.2026.133043","url":null,"abstract":"<div><div>Fault diagnosis technology based on Transfer Learning (TL) has been widely investigated, as it enables knowledge transfer from related domains to the target domain, thereby reducing the need for large amounts of labeled data in the target domain. In the railway industry, Railway Point Machines (RPMs) are essential track-connection devices. However, due to the scarcity of fault data for RPMs, the accuracy of RPM fault diagnosis remains low. To address this challenge, this paper proposes a Two-step Transfer Learning (TSTL) method for RPM fault diagnosis. First, a small set of target-domain samples transformed by continuous wavelet transform (CWT) is used to construct a transition dataset, which is fed into a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) to generate synthetic samples. Then, a Convolutional Neural Network (CNN) with a top-down attention mechanism is employed to perform the first phase of TL from the source domain to the generated domain. Next, the parameters of selected frozen layers in the transferred CNN are retained to conduct the second phase of learning from the generated domain to the target domain. Experimental results demonstrate that the proposed method achieves 98.7% accuracy for fault diagnosis across different positions on the same RPM, and 95.6% accuracy across different RPMs at the same position.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133043"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386778","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-07DOI: 10.1016/j.neucom.2026.132831
S. Vidhya , P.M. Siva Raja , R.P. Sumithra , Moses Garuba , Xiao-Zhi Gao
Brain tumor segmentation and classification from MRI images are critical tasks in neuro-oncology, but they remain challenging due to tumor heterogeneity, ambiguous boundaries, and variability across multi-modal imaging sequences. Existing deep learning methods often struggle with subregion delineation and generalization, leading to incomplete or inaccurate results. In this paper, we propose Squeeze-and-Excitation Mamba with DeiSwin+ + (SEMD-Net), a unified deep learning framework designed to improve both tumor segmentation and imaging-based glioma grade classification (HGG vs. LGG). The model integrates a multi-branch learning architecture capable of capturing both global and localized tumor features. It combines spatial-channel attention mechanisms and transformer-based representation learning to address the challenges of boundary precision, tissue variability, and intra-class heterogeneity. We evaluated our method on the BraTS 2020 and 2021 datasets using five-fold cross-validation. SEMD-Net achieved strong performance on key segmentation metrics, including a mean Dice Similarity Coefficient (DSC) of 0.88 ± 0.02, IoU of 0.86 ± 0.03, and a Hausdorff Distance (HD) of 3.5 ± 0.8 mm. For glioma subtype classification, the model reached 97.5 % accuracy, 98.9 % precision, 98.8 % recall, and an F1-score of 98.85 %, outperforming benchmark methods such as U-Net, DeepSemantic, and AdaBoost. These results suggest that SEMD-Net effectively balances segmentation accuracy and classification robustness, offering a promising solution for integrated brain tumor analysis. While further validation on external datasets is ongoing, the proposed framework shows strong potential for clinical application in automated MRI-based diagnosis.
{"title":"SEMD-Net: A transformer based approach for brain tumor segmentation and classification","authors":"S. Vidhya , P.M. Siva Raja , R.P. Sumithra , Moses Garuba , Xiao-Zhi Gao","doi":"10.1016/j.neucom.2026.132831","DOIUrl":"10.1016/j.neucom.2026.132831","url":null,"abstract":"<div><div>Brain tumor segmentation and classification from MRI images are critical tasks in neuro-oncology, but they remain challenging due to tumor heterogeneity, ambiguous boundaries, and variability across multi-modal imaging sequences. Existing deep learning methods often struggle with subregion delineation and generalization, leading to incomplete or inaccurate results. In this paper, we propose Squeeze-and-Excitation Mamba with DeiSwin+ + (SEMD-Net), a unified deep learning framework designed to improve both tumor segmentation and imaging-based glioma grade classification (HGG vs. LGG). The model integrates a multi-branch learning architecture capable of capturing both global and localized tumor features. It combines spatial-channel attention mechanisms and transformer-based representation learning to address the challenges of boundary precision, tissue variability, and intra-class heterogeneity. We evaluated our method on the BraTS 2020 and 2021 datasets using five-fold cross-validation. SEMD-Net achieved strong performance on key segmentation metrics, including a mean Dice Similarity Coefficient (DSC) of 0.88 ± 0.02, IoU of 0.86 ± 0.03, and a Hausdorff Distance (HD) of 3.5 ± 0.8 mm. For glioma subtype classification, the model reached 97.5 % accuracy, 98.9 % precision, 98.8 % recall, and an F1-score of 98.85 %, outperforming benchmark methods such as U-Net, DeepSemantic, and AdaBoost. These results suggest that SEMD-Net effectively balances segmentation accuracy and classification robustness, offering a promising solution for integrated brain tumor analysis. While further validation on external datasets is ongoing, the proposed framework shows strong potential for clinical application in automated MRI-based diagnosis.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 132831"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386789","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-09DOI: 10.1016/j.neucom.2026.133009
Xudong Ye , Qi Zhang , Yapeng Wang , Xu Yang , Zuobin Ying , Jingzhang Sun , Qi Zhong , Xia Du
Adversarial patches pose a significant and real threat to deep neural networks, capable of inducing misclassification in realistic physical scenarios. Developing reliable and robust defense methods against these attacks is a critical application, and current research remains unsatisfactory. In this paper, we propose a novel framework that exploits the fact that unnatural perturbations introduced by adversarial patches can produce prediction biases significantly different from those of clean images during denoising. In the localization stage, our method focuses on the critical denoising steps through an adaptive temporal sampling strategy and introduces an energy metric that fuses kinetic and potential energy to quantify the degree of anomaly in the denoised trajectory. Furthermore, by combining this with the adaptive similarity weighting mechanism and the striding trajectory consistency analysis, our method effectively suppresses the interference of background noise, so as to achieve accurate locking of the patch area. In the restoration phase, the same diffusion model is applied to the patch region to restore the original visual content and integrity. This two-stage architecture shares a unified diffusion model, enabling the localization and inpainting processes to enhance the overall defense performance through information complementarity. Extensive experiments on the INRIA, COCO2017, and APRICOT datasets show that our approach achieves state-of-the-art detection performance under both digital and physical attack types without compromising the recognition accuracy of clean images.
{"title":"Diffbias: Harnessing diffusion models’ prediction bias for adversarial patch defense","authors":"Xudong Ye , Qi Zhang , Yapeng Wang , Xu Yang , Zuobin Ying , Jingzhang Sun , Qi Zhong , Xia Du","doi":"10.1016/j.neucom.2026.133009","DOIUrl":"10.1016/j.neucom.2026.133009","url":null,"abstract":"<div><div>Adversarial patches pose a significant and real threat to deep neural networks, capable of inducing misclassification in realistic physical scenarios. Developing reliable and robust defense methods against these attacks is a critical application, and current research remains unsatisfactory. In this paper, we propose a novel framework that exploits the fact that unnatural perturbations introduced by adversarial patches can produce prediction biases significantly different from those of clean images during denoising. In the localization stage, our method focuses on the critical denoising steps through an adaptive temporal sampling strategy and introduces an energy metric that fuses kinetic and potential energy to quantify the degree of anomaly in the denoised trajectory. Furthermore, by combining this with the adaptive similarity weighting mechanism and the striding trajectory consistency analysis, our method effectively suppresses the interference of background noise, so as to achieve accurate locking of the patch area. In the restoration phase, the same diffusion model is applied to the patch region to restore the original visual content and integrity. This two-stage architecture shares a unified diffusion model, enabling the localization and inpainting processes to enhance the overall defense performance through information complementarity. Extensive experiments on the INRIA, COCO2017, and APRICOT datasets show that our approach achieves state-of-the-art detection performance under both digital and physical attack types without compromising the recognition accuracy of clean images.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133009"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146172693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2026-05-01Epub Date: 2026-02-10DOI: 10.1016/j.neucom.2026.132926
Liya Ma , Siqi Sun , Chen Li , Lu Liu , Lihong Wang
Most real-world graphs are dynamic and pose privacy risks when published and shared. Several researchers have used differential privacy to protect graph privacy. However, it is still challenging to maintain a continuous privacy–utility tradeoff when differential privacy methods are applied to dynamic graphs. To address this, we propose a novel Dynamic Graph Generative Adversarial Network with Edge-level Differential Privacy (DyGAN-EDP) model, which applies structural self-attention and an LSTM neural network to learn complex nonlinear patterns associated with graph structure evolution, which is beneficial for maintaining utility. To protect privacy, noise is introduced into the gradient of the generator during the training process to synthesize dynamic graphs, which can guarantee edge-level differential privacy. Moreover, a noise adjustment mechanism based on mutual information estimation is introduced, which provides stronger privacy protection when the graph structure changes significantly over time; at other times, it lowers the noise scale to ensure data utility. Experimental validations on three real-world dynamic graphs demonstrate the model’s ability to balance privacy and utility.
{"title":"A novel dynamic graph generative adversarial network with edge-level differential privacy","authors":"Liya Ma , Siqi Sun , Chen Li , Lu Liu , Lihong Wang","doi":"10.1016/j.neucom.2026.132926","DOIUrl":"10.1016/j.neucom.2026.132926","url":null,"abstract":"<div><div>Most real-world graphs are dynamic and pose privacy risks when published and shared. Several researchers have used differential privacy to protect graph privacy. However, it is still challenging to maintain a continuous privacy–utility tradeoff when differential privacy methods are applied to dynamic graphs. To address this, we propose a novel Dynamic Graph Generative Adversarial Network with Edge-level Differential Privacy (<strong>DyGAN-EDP</strong>) model, which applies structural self-attention and an LSTM neural network to learn complex nonlinear patterns associated with graph structure evolution, which is beneficial for maintaining utility. To protect privacy, noise is introduced into the gradient of the generator during the training process to synthesize dynamic graphs, which can guarantee edge-level differential privacy. Moreover, a noise adjustment mechanism based on mutual information estimation is introduced, which provides stronger privacy protection when the graph structure changes significantly over time; at other times, it lowers the noise scale to ensure data utility. Experimental validations on three real-world dynamic graphs demonstrate the model’s ability to balance privacy and utility.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 132926"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146172680","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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