Neurocomputing最新文献_第2页

MVHDiff: Leveraging 3D priors for consistent multi-view human image generation with diffusion models MVHDiff：利用3D先验与扩散模型一致的多视图人体图像生成

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-13 DOI: 10.1016/j.neucom.2026.133057

Yan Huang , Hongxin Fu , Zhonghang Li , Yongcan Luo , Tianyi Chen , Si Wu

Text-to-image models are increasingly applied to human image generation, leveraging multimodal information under multiple conditions to produce high-quality human images. Despite their ability to generate detailed images, these models often struggle to maintain perceptual consistency across multiple viewpoints. To address this limitation, we propose Multi-View Human Diffusion (MVHDiff), a novel framework that integrates 3D human model priors and text prompts to generate high-quality, multi-view-consistent human images. MVHDiff separately acquires textual descriptions of human appearance and pose, as well as spatial information regarding the subject’s orientation relative to the camera. Subsequently, a perceptual fusion module is employed to align these text features with the visual features extracted from the human image, thereby enabling the fused learning of prior information and image features. Further, MVHDiff finetunes both appearance descriptions and spatial viewpoint-related textual inputs, enabling precise text-based control over human attributes while ensuring semantic consistency across different spatial viewpoints. Experimental results demonstrate that MVHDiff significantly outperforms existing methods in generating text-guided human attributes with consistent multi-view representations, offering a robust solution for high-quality, text-driven human image generation.

文本-图像模型越来越多地应用于人体图像生成，利用多模态信息在多种条件下生成高质量的人体图像。尽管这些模型能够生成详细的图像，但它们往往难以在多个视点之间保持感知一致性。为了解决这一限制，我们提出了多视图人体扩散（MVHDiff），这是一个集成了3D人体模型先验和文本提示的新框架，可以生成高质量、多视图一致的人体图像。MVHDiff分别获取人体外观和姿势的文本描述，以及关于主体相对于相机方向的空间信息。随后，使用感知融合模块将这些文本特征与从人类图像中提取的视觉特征对齐，从而实现先验信息和图像特征的融合学习。此外，MVHDiff对外观描述和与空间视点相关的文本输入进行微调，在确保不同空间视点之间的语义一致性的同时，实现对人类属性的精确的基于文本的控制。实验结果表明，MVHDiff在生成具有一致多视图表示的文本引导人类属性方面显著优于现有方法，为高质量的文本驱动人类图像生成提供了鲁棒解决方案。

{"title":"MVHDiff: Leveraging 3D priors for consistent multi-view human image generation with diffusion models","authors":"Yan Huang , Hongxin Fu , Zhonghang Li , Yongcan Luo , Tianyi Chen , Si Wu","doi":"10.1016/j.neucom.2026.133057","DOIUrl":"10.1016/j.neucom.2026.133057","url":null,"abstract":"<div><div>Text-to-image models are increasingly applied to human image generation, leveraging multimodal information under multiple conditions to produce high-quality human images. Despite their ability to generate detailed images, these models often struggle to maintain perceptual consistency across multiple viewpoints. To address this limitation, we propose Multi-View Human Diffusion (MVHDiff), a novel framework that integrates 3D human model priors and text prompts to generate high-quality, multi-view-consistent human images. MVHDiff separately acquires textual descriptions of human appearance and pose, as well as spatial information regarding the subject’s orientation relative to the camera. Subsequently, a perceptual fusion module is employed to align these text features with the visual features extracted from the human image, thereby enabling the fused learning of prior information and image features. Further, MVHDiff finetunes both appearance descriptions and spatial viewpoint-related textual inputs, enabling precise text-based control over human attributes while ensuring semantic consistency across different spatial viewpoints. Experimental results demonstrate that MVHDiff significantly outperforms existing methods in generating text-guided human attributes with consistent multi-view representations, offering a robust solution for high-quality, text-driven human image generation.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133057"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386791","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}

引用次数: 0

Hierarchy-aware graph neural network and inverse-variance reinforcement learning for drug recommendation 层次感知图神经网络和反方差强化学习在药物推荐中的应用

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-11 DOI: 10.1016/j.neucom.2026.132989

Wei Hou , Xianxing Liu , Linxiao Li , Chunling Fu

Drug recommendation (DR) based on artificial intelligence plays a crucial role in healthcare research, offering precise and effective drug prescription suggestions for doctors. However, existing methods typically model DR as a sequential task, overlooking the complex correlations among medical entities present in electronic medical records (EMRs). To this end, we propose a novel DR model that integrates a hierarchy-aware graph neural network (GNN) with inverse-variance (IV) reinforcement learning (RL). Specifically, we represent patient and drug information using a knowledge graph, and employ a hyperbolic space-embedded GNN to encode the hierarchical structure among graph nodes. We propose an IV-RL mechanism to reduce excessive exploration of the model on inefficient or noisy data. By incorporating IV into the RL framework, the model can more efficiently sample from the training data, thereby enhancing learning performance. Extensive experiments, on the widely-used MIMIC-III, MIMIC-IV, and eICU, datasets demonstrate that our proposed method achieves superior performance and exhibits reliable DR capabilities. We believe that our proposed method provides a promising solution for accurate and effective DR, and opens up new opportunities for further research.

基于人工智能的药物推荐（DR）在医疗保健研究中发挥着至关重要的作用，为医生提供精确有效的药物处方建议。然而，现有方法通常将DR建模为顺序任务，忽略了电子病历（emr）中医疗实体之间的复杂相关性。为此，我们提出了一种新的DR模型，该模型将层次感知图神经网络（GNN）与逆方差（IV）强化学习（RL）相结合。具体来说，我们使用知识图来表示患者和药物信息，并使用双曲空间嵌入的GNN来编码图节点之间的层次结构。我们提出了一种IV-RL机制，以减少对低效率或噪声数据的过度探索。通过将IV纳入RL框架，模型可以更有效地从训练数据中采样，从而提高学习性能。在广泛使用的MIMIC-III、MIMIC-IV和eICU数据集上进行的大量实验表明，我们提出的方法具有优越的性能和可靠的DR能力。我们相信我们提出的方法为准确有效的DR提供了一个有希望的解决方案，并为进一步的研究开辟了新的机会。

{"title":"Hierarchy-aware graph neural network and inverse-variance reinforcement learning for drug recommendation","authors":"Wei Hou , Xianxing Liu , Linxiao Li , Chunling Fu","doi":"10.1016/j.neucom.2026.132989","DOIUrl":"10.1016/j.neucom.2026.132989","url":null,"abstract":"<div><div>Drug recommendation (DR) based on artificial intelligence plays a crucial role in healthcare research, offering precise and effective drug prescription suggestions for doctors. However, existing methods typically model DR as a sequential task, overlooking the complex correlations among medical entities present in electronic medical records (EMRs). To this end, we propose a novel DR model that integrates a hierarchy-aware graph neural network (GNN) with inverse-variance (IV) reinforcement learning (RL). Specifically, we represent patient and drug information using a knowledge graph, and employ a hyperbolic space-embedded GNN to encode the hierarchical structure among graph nodes. We propose an IV-RL mechanism to reduce excessive exploration of the model on inefficient or noisy data. By incorporating IV into the RL framework, the model can more efficiently sample from the training data, thereby enhancing learning performance. Extensive experiments, on the widely-used MIMIC-III, MIMIC-IV, and eICU, datasets demonstrate that our proposed method achieves superior performance and exhibits reliable DR capabilities. We believe that our proposed method provides a promising solution for accurate and effective DR, and opens up new opportunities for further research.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 132989"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386724","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}

引用次数: 0

One-step diffusion-based real-world image super-resolution with visual perception distillation 基于一步扩散的真实世界图像超分辨与视觉感知蒸馏

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-13 DOI: 10.1016/j.neucom.2026.133066

Xue Wu , Jingwei Xin , Jun Hao , Hui Gao , Jie Li , Nannan Wang , Xinbo Gao

Diffusion-based models have been widely used in various visual generation tasks, showing promising results in image super-resolution (SR), while typically being limited by dozens or even hundreds of sampling steps. Although existing methods aim to accelerate the inference speed of multi-step diffusion-based SR methods through knowledge distillation, their generated images exhibit insufficient semantic alignment with real images, resulting in suboptimal perceptual quality reconstruction, specifically reflected in the CLIPIQA score. These methods still face many challenges in perceptual quality and semantic fidelity. Based on the challenges, we propose VPD-SR, a novel visual perception diffusion distillation framework specifically designed for SR, aiming to construct an effective and efficient one-step SR model. Specifically, VPD-SR consists of two components: Explicit Semantic-aware Supervision (ESS) and High-Frequency Perception (HFP) loss. Firstly, the ESS leverages the powerful visual perceptual understanding capabilities of the CLIP model to extract explicit semantic supervision, thereby enhancing semantic consistency. Then, considering that high-frequency information contributes to the visual perception quality of images, in addition to the vanilla distillation loss, the HFP loss guides the student model to restore the missing high-frequency details in degraded images that are critical for enhancing perceptual quality. Lastly, we expand VPD-SR in an adversarial training manner to further enhance the authenticity of the generated content. Extensive experiments conducted on synthetic and real-world datasets demonstrate that the proposed VPD-SR achieves superior performance compared to both previous state-of-the-art methods and the teacher model with just one-step sampling.

基于扩散的模型已广泛应用于各种视觉生成任务，在图像超分辨率（SR）方面显示出良好的效果，但通常受到数十甚至数百个采样步骤的限制。尽管现有方法旨在通过知识蒸馏加快基于多步扩散的SR方法的推理速度，但其生成的图像与真实图像的语义一致性不足，导致感知质量重构不理想，具体体现在CLIPIQA评分上。这些方法在感知质量和语义保真度方面仍面临许多挑战。基于这些挑战，我们提出了一种新的视觉感知扩散蒸馏框架VPD-SR，旨在构建一个有效和高效的一步SR模型。具体来说，VPD-SR由两个部分组成：显式语义感知监督（ESS）和高频感知损失（HFP）。首先，ESS利用CLIP模型强大的视觉感知理解能力提取显式语义监督，从而增强语义一致性。然后，考虑到高频信息有助于图像的视觉感知质量，除了香草蒸馏损失外，HFP损失还指导学生模型恢复退化图像中缺失的高频细节，这些细节对增强感知质量至关重要。最后，我们以对抗训练的方式扩展VPD-SR，进一步增强生成内容的真实性。在合成数据集和真实数据集上进行的大量实验表明，与之前最先进的方法和仅一步采样的教师模型相比，所提出的VPD-SR具有优越的性能。

{"title":"One-step diffusion-based real-world image super-resolution with visual perception distillation","authors":"Xue Wu , Jingwei Xin , Jun Hao , Hui Gao , Jie Li , Nannan Wang , Xinbo Gao","doi":"10.1016/j.neucom.2026.133066","DOIUrl":"10.1016/j.neucom.2026.133066","url":null,"abstract":"<div><div>Diffusion-based models have been widely used in various visual generation tasks, showing promising results in image super-resolution (SR), while typically being limited by dozens or even hundreds of sampling steps. Although existing methods aim to accelerate the inference speed of multi-step diffusion-based SR methods through knowledge distillation, their generated images exhibit insufficient semantic alignment with real images, resulting in suboptimal perceptual quality reconstruction, specifically reflected in the CLIPIQA score. These methods still face many challenges in perceptual quality and semantic fidelity. Based on the challenges, we propose VPD-SR, a novel visual perception diffusion distillation framework specifically designed for SR, aiming to construct an effective and efficient one-step SR model. Specifically, VPD-SR consists of two components: Explicit Semantic-aware Supervision (ESS) and High-Frequency Perception (HFP) loss. Firstly, the ESS leverages the powerful visual perceptual understanding capabilities of the CLIP model to extract explicit semantic supervision, thereby enhancing semantic consistency. Then, considering that high-frequency information contributes to the visual perception quality of images, in addition to the vanilla distillation loss, the HFP loss guides the student model to restore the missing high-frequency details in degraded images that are critical for enhancing perceptual quality. Lastly, we expand VPD-SR in an adversarial training manner to further enhance the authenticity of the generated content. Extensive experiments conducted on synthetic and real-world datasets demonstrate that the proposed VPD-SR achieves superior performance compared to both previous state-of-the-art methods and the teacher model with just one-step sampling.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133066"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386787","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}

引用次数: 0

A systematic review of machine learning for digital stain processing in pathology 机器学习在病理学中数字染色处理的系统综述

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-16 DOI: 10.1016/j.neucom.2026.133064

Rabiah Al-Qudah , Abubakar Bala , Mrouj Almuhajri , Khiati Zakaria , Ching Y. Suen

Digital staining involves using methods such as Machine Learning (ML) to replace chemical staining in pathology. Staining adds contrast that makes cell details more visible under the microscope. However, chemical methods are slow, use toxic reagents, and require skilled personnel. In contrast, digital staining can generate images faster, reduce the need for reagents and specialized equipment, and minimize plastic and chemical waste, making the workflow more sustainable. This paper systematically reviews papers published on ML-based digital stain processing. We propose a new taxonomy that divides existing studies into five groups: stain normalization, stain augmentation, virtual staining, stain transformation, and hybrid approaches. In addition, we observed several trends from the reviewed papers. Finally, we outline open research directions.

数字染色涉及使用机器学习（ML）等方法来取代病理学中的化学染色。染色增加了对比度，使细胞细节在显微镜下更清晰可见。然而，化学方法是缓慢的，使用有毒的试剂，需要熟练的人员。相比之下，数字染色可以更快地生成图像，减少对试剂和专用设备的需求，并最大限度地减少塑料和化学废物，使工作流程更具可持续性。本文系统地综述了基于机器学习的数字染色处理的相关文献。我们提出了一种新的分类方法，将现有的研究分为五组：染色归一化、染色增强、虚拟染色、染色转化和混合方法。此外，我们还从审评的论文中观察到几个趋势。最后，概述了开放的研究方向。

引用次数: 0

Enhancing accuracy and interpretability in corporate credit rating classification with the transformer-LSTM model 利用变压器- lstm模型提高企业信用评级分类的准确性和可解释性

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-11 DOI: 10.1016/j.neucom.2026.133026

Mikio Kofune , Kensei Monden , Suguru Yamanaka

Corporate credit rating classification is essential for assessing a corporation’s debt repayment ability. Previous research has demonstrated that neural network models exhibit high classification accuracy, particularly when incorporating time-series features. However, a significant challenge remains regarding their interpretability, often limited by nonlinear and intricate calculation processes. To address this trade-off between interpretability and the utilization of time-series features, we introduce a novel approach: the Transformer-Long Short-Term Memory (T-LSTM). Specifically, the attention matrix embedded within the T-LSTM architecture provides interpretability by revealing the temporal importance of features. Comparative experiments show that our T-LSTM model surpasses standard machine learning baselines in the long-history setting. Empirical results demonstrate that the proposed model, trained on a 10-year history of longitudinal financial ratios, yields an absolute accuracy improvement of approximately 1 percentage point when compared with a strong sequential baseline such as a Long Short-Term Memory (LSTM) model trained on the same 10-year history, and up to about 17 percentage points when compared with representative single-year “snapshot” baselines that use only the most recent year’s ratios. Furthermore, the attention matrix successfully visualizes specific time points where information is most critical for rating classification. Consequently, the proposed model offers a highly accurate and interpretable solution for credit rating classification in the financial industry.

企业信用等级分类是评估企业偿债能力的必要手段。先前的研究表明，神经网络模型具有很高的分类精度，特别是当结合时间序列特征时。然而，它们的可解释性仍然是一个重大的挑战，通常受到非线性和复杂计算过程的限制。为了解决可解释性和时间序列特征利用之间的这种权衡，我们引入了一种新的方法：变压器-长短期记忆（T-LSTM）。具体来说，嵌入在T-LSTM架构中的注意力矩阵通过揭示特征的时间重要性来提供可解释性。对比实验表明，我们的T-LSTM模型在长期设置中超过了标准的机器学习基线。实证结果表明，与基于同样10年历史的长短期记忆（LSTM）模型等强序列基线相比，基于10年历史的纵向财务比率训练的拟议模型的绝对准确性提高了约1个百分点，与仅使用最近一年比率的代表性单年“快照”基线相比，精确度提高了约17个百分点。此外，注意矩阵成功地将信息对评级分类最关键的特定时间点可视化。因此，所提出的模型为金融行业的信用评级分类提供了一个高度准确和可解释的解决方案。

{"title":"Enhancing accuracy and interpretability in corporate credit rating classification with the transformer-LSTM model","authors":"Mikio Kofune , Kensei Monden , Suguru Yamanaka","doi":"10.1016/j.neucom.2026.133026","DOIUrl":"10.1016/j.neucom.2026.133026","url":null,"abstract":"<div><div>Corporate credit rating classification is essential for assessing a corporation’s debt repayment ability. Previous research has demonstrated that neural network models exhibit high classification accuracy, particularly when incorporating time-series features. However, a significant challenge remains regarding their interpretability, often limited by nonlinear and intricate calculation processes. To address this trade-off between interpretability and the utilization of time-series features, we introduce a novel approach: the Transformer-Long Short-Term Memory (T-LSTM). Specifically, the attention matrix embedded within the T-LSTM architecture provides interpretability by revealing the temporal importance of features. Comparative experiments show that our T-LSTM model surpasses standard machine learning baselines in the long-history setting. Empirical results demonstrate that the proposed model, trained on a 10-year history of longitudinal financial ratios, yields an absolute accuracy improvement of approximately 1 percentage point when compared with a strong sequential baseline such as a Long Short-Term Memory (LSTM) model trained on the same 10-year history, and up to about 17 percentage points when compared with representative single-year “snapshot” baselines that use only the most recent year’s ratios. Furthermore, the attention matrix successfully visualizes specific time points where information is most critical for rating classification. Consequently, the proposed model offers a highly accurate and interpretable solution for credit rating classification in the financial industry.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133026"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386380","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}

引用次数: 0

XID: A protocol for evaluating identity consistency under domain shifts and reidentification method XID：一种域移位下的恒等一致性评估协议和再识别方法

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-14 DOI: 10.1016/j.neucom.2026.133012

Seongyeop Yang , Minho Kim , Byeongkeun Kang , Yeejin Lee

Recent domain generalization methods for person reidentification aim to learn features that remain discriminative across domains to improve performance in unseen environments. Prior work has addressed domain shift through discrepancy reduction and alternative normalization strategies, while maintaining identity separability. However, these evaluations often rely on simplified settings with non-overlapping identities and limited visual diversity. To address this, we propose a new evaluation protocol that introduces identity transfer and significant appearance variation by constructing query and gallery sets from different domains. This setup enables a more realistic assessment of intra-class variation and inter-class discriminability. We further develop a learning framework specifically designed for this protocol, which enhances generalization by regulating achromatic information and projecting embeddings into a space that simulates unseen domains. The framework includes a self-regulating augmentation policy that adjusts transformation strength during training. Extensive experiments show consistent performance gains under both the proposed and standard protocols, establishing a more rigorous and practical benchmark.

最近用于人员再识别的领域泛化方法旨在学习跨领域仍然具有区别性的特征，以提高在未知环境中的性能。先前的工作通过减少差异和替代规范化策略来解决领域转移，同时保持身份可分离性。然而，这些评估往往依赖于简化的设置，没有重叠的身份和有限的视觉多样性。为了解决这个问题，我们提出了一个新的评估协议，该协议通过构造来自不同领域的查询和库集来引入身份转移和显著的外观变化。这种设置可以更现实地评估阶级内部的变化和阶级之间的区别。我们进一步开发了一个专门为该协议设计的学习框架，该框架通过调节消色差信息和将嵌入投影到模拟未见域的空间中来增强泛化。该框架包括一个自我调节的增强策略，在训练过程中调整转换强度。大量的实验表明，在提出的协议和标准协议下，性能都得到了一致的提高，从而建立了一个更严格和实用的基准。

{"title":"XID: A protocol for evaluating identity consistency under domain shifts and reidentification method","authors":"Seongyeop Yang , Minho Kim , Byeongkeun Kang , Yeejin Lee","doi":"10.1016/j.neucom.2026.133012","DOIUrl":"10.1016/j.neucom.2026.133012","url":null,"abstract":"<div><div>Recent domain generalization methods for person reidentification aim to learn features that remain discriminative across domains to improve performance in unseen environments. Prior work has addressed domain shift through discrepancy reduction and alternative normalization strategies, while maintaining identity separability. However, these evaluations often rely on simplified settings with non-overlapping identities and limited visual diversity. To address this, we propose a new evaluation protocol that introduces identity transfer and significant appearance variation by constructing query and gallery sets from different domains. This setup enables a more realistic assessment of intra-class variation and inter-class discriminability. We further develop a learning framework specifically designed for this protocol, which enhances generalization by regulating achromatic information and projecting embeddings into a space that simulates unseen domains. The framework includes a self-regulating augmentation policy that adjusts transformation strength during training. Extensive experiments show consistent performance gains under both the proposed and standard protocols, establishing a more rigorous and practical benchmark.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133012"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386318","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}

引用次数: 0

Data-driven adaptive secure collision-free formation tracking of networked marine surface vehicles under DoS attacks DoS攻击下网络水面舰艇数据驱动自适应安全无碰撞编队跟踪

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-14 DOI: 10.1016/j.neucom.2026.133082

Jun Hu , Jie Wu , Xisheng Zhan , Tao Han , Huaicheng Yan

This paper addresses the adaptive path planning and secure formation tracking (SFT) control problem for networked marine surface vehicles (NMSVs) subject to denial-of-service (DoS) attacks, model uncertainty, external disturbances, and actuator faults. A hierarchical adaptive formation planning and control (HAFPC) framework is proposed. In its path planning layer, reinforcement learning (RL) based greedy path map inference (GPMI) infers local maps in unknown environments, while a Euclidean-distance-field-based formation path planning (EDF-FPP) algorithm finds collision-free trajectories. In the network layer, a distributed resilient estimator is designed to accurately estimate the virtual leader information under DoS attacks in directed graphs. In the control layer, a neural network (NN)-based data-driven observer is first employed to address model uncertainty. Then, an adaptive offset function and a data-driven observer-based control (DDOBC) algorithm are adopted to achieve SFT, obstacle avoidance, and handle input saturation and actuator faults. Lyapunov stability theory establishes sufficient conditions for system convergence and stabilization. Numerical simulations validate the proposed framework’s effectiveness.

本文研究了受拒绝服务（DoS）攻击、模型不确定性、外部干扰和执行器故障影响的网络化海上水面车辆（nmsv）的自适应路径规划和安全编队跟踪（SFT）控制问题。提出了一种分层自适应编队规划与控制（HAFPC）框架。在路径规划层，基于强化学习（RL）的贪婪路径图推理（GPMI）在未知环境中推断局部地图，而基于欧氏距离场的编队路径规划（EDF-FPP）算法寻找无碰撞轨迹。在网络层，设计了一个分布式弹性估计器，以准确估计有向图中DoS攻击下的虚拟leader信息。在控制层，首先采用基于神经网络的数据驱动观测器来解决模型的不确定性。然后，采用自适应偏移函数和基于数据驱动观测器的控制（DDOBC）算法实现SFT、避障，并处理输入饱和和执行器故障。Lyapunov稳定性理论建立了系统收敛和稳定的充分条件。数值仿真验证了该框架的有效性。

{"title":"Data-driven adaptive secure collision-free formation tracking of networked marine surface vehicles under DoS attacks","authors":"Jun Hu , Jie Wu , Xisheng Zhan , Tao Han , Huaicheng Yan","doi":"10.1016/j.neucom.2026.133082","DOIUrl":"10.1016/j.neucom.2026.133082","url":null,"abstract":"<div><div>This paper addresses the adaptive path planning and secure formation tracking (SFT) control problem for networked marine surface vehicles (NMSVs) subject to denial-of-service (DoS) attacks, model uncertainty, external disturbances, and actuator faults. A hierarchical adaptive formation planning and control (HAFPC) framework is proposed. In its path planning layer, reinforcement learning (RL) based greedy path map inference (GPMI) infers local maps in unknown environments, while a Euclidean-distance-field-based formation path planning (EDF-FPP) algorithm finds collision-free trajectories. In the network layer, a distributed resilient estimator is designed to accurately estimate the virtual leader information under DoS attacks in directed graphs. In the control layer, a neural network (NN)-based data-driven observer is first employed to address model uncertainty. Then, an adaptive offset function and a data-driven observer-based control (DDOBC) algorithm are adopted to achieve SFT, obstacle avoidance, and handle input saturation and actuator faults. Lyapunov stability theory establishes sufficient conditions for system convergence and stabilization. Numerical simulations validate the proposed framework’s effectiveness.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133082"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386369","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}

引用次数: 0

Self-organizing interval type-2 fuzzy neural networks with rule interaction constraints 具有规则交互约束的自组织区间型-2模糊神经网络

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-11 DOI: 10.1016/j.neucom.2026.133048

Yuanwen Zhang, Ke Zhang

Interval Type-2 fuzzy neural networks (IT2FNNs) provide a robust framework for managing uncertainty in classification tasks; however, conventional architectures rely on linear aggregation and fail to capture pairwise rule interactions. This paper proposes a self-organizing IT2FNN designed to explicitly parameterize these rule dependencies. The framework incorporates a rule interaction layer utilizing factorization machines to identify synergistic and competitive effects between IF-THEN rules. To ensure semantic consistency, a knowledge-driven constraint mechanism derived from 2-additive Choquet integral theory guides the interaction parameters. Additionally, a differentiable gating strategy enables autonomous rule base refinement through sparse regularization. Evaluations on benchmark datasets demonstrate that the proposed architecture achieves superior classification accuracy while maintaining a parsimonious rule base, averaging approximately 10 active rules. This study indicates that explicit rule interaction modeling improves functional approximation, while structural constraints and pruning maintain the model’s compactness and linguistic interpretability.

区间2型模糊神经网络（IT2FNNs）为管理分类任务中的不确定性提供了一个鲁棒框架；然而，传统的体系结构依赖于线性聚合，无法捕获成对规则交互。本文提出了一种自组织IT2FNN，旨在显式参数化这些规则依赖关系。该框架结合了一个规则交互层，利用因子分解机器来识别IF-THEN规则之间的协同和竞争效应。为了保证语义的一致性，基于2加性Choquet积分理论的知识驱动约束机制指导交互参数。此外，可微门控策略通过稀疏正则化实现自治规则库的细化。对基准数据集的评估表明，所提出的体系结构在保持精简的规则库（平均约10条活动规则）的同时实现了卓越的分类精度。该研究表明，显式规则交互建模提高了功能逼近，而结构约束和剪枝保持了模型的紧凑性和语言可解释性。

{"title":"Self-organizing interval type-2 fuzzy neural networks with rule interaction constraints","authors":"Yuanwen Zhang, Ke Zhang","doi":"10.1016/j.neucom.2026.133048","DOIUrl":"10.1016/j.neucom.2026.133048","url":null,"abstract":"<div><div>Interval Type-2 fuzzy neural networks (IT2FNNs) provide a robust framework for managing uncertainty in classification tasks; however, conventional architectures rely on linear aggregation and fail to capture pairwise rule interactions. This paper proposes a self-organizing IT2FNN designed to explicitly parameterize these rule dependencies. The framework incorporates a rule interaction layer utilizing factorization machines to identify synergistic and competitive effects between IF-THEN rules. To ensure semantic consistency, a knowledge-driven constraint mechanism derived from 2-additive Choquet integral theory guides the interaction parameters. Additionally, a differentiable gating strategy enables autonomous rule base refinement through sparse regularization. Evaluations on benchmark datasets demonstrate that the proposed architecture achieves superior classification accuracy while maintaining a parsimonious rule base, averaging approximately 10 active rules. This study indicates that explicit rule interaction modeling improves functional approximation, while structural constraints and pruning maintain the model’s compactness and linguistic interpretability.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133048"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386420","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}

引用次数: 0

Dynamic enhanced binning framework 动态增强的分组框架

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-11 DOI: 10.1016/j.neucom.2026.133033

Wei Jiang , XiCheng Yang , YuXin Wang , LingYuLe Wang

This paper proposes Dynamic Enhanced Binning (DEB), a structured white-box framework designed to bridge the gap between the accuracy of Gradient Boosting Decision Trees (GBDTs) and the interpretability of linear models. DEB constructs a high-fidelity linear feature space by synergizing Soft Binning, Polynomial Expansion, and Explicit Interaction Terms, optimized via a lightweight Bayesian engine. Extensive experiments demonstrate that DEB achieves predictive accuracy competitive with XGBoost while delivering distinct operational advantages: a 3x speedup in real-time inference on industrial datasets, an 8.8x increase in training throughput on large-scale synthetic benchmarks, and a 32x reduction in structural complexity compared to ensemble baselines. By reconciling nonlinear approximation with theoretical transparency and

O (N)

scalability, DEB offers a robust solution for high-stakes, latency-sensitive industrial applications.

本文提出了动态增强分类（DEB），这是一个结构化的白盒框架，旨在弥合梯度增强决策树（gbdt）的准确性与线性模型的可解释性之间的差距。DEB通过软分区、多项式展开和显式交互项协同构建了高保真的线性特征空间，并通过轻量级贝叶斯引擎进行了优化。大量的实验表明，DEB实现了与XGBoost竞争的预测精度，同时提供了独特的操作优势：在工业数据集上的实时推理速度提高了3倍，在大规模合成基准上的训练吞吐量提高了8.8倍，与集成基线相比，结构复杂性降低了32倍。通过协调非线性近似与理论透明度和O(N)可扩展性，DEB为高风险，延迟敏感的工业应用提供了强大的解决方案。

引用次数: 0

Drift-aware variational autoencoder-based anomaly detection with two-level ensembling 基于漂移感知变分自编码器的两级集成异常检测

IF 6.5 2区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Neurocomputing

Pub Date : 2026-05-01 Epub Date: 2026-02-13 DOI: 10.1016/j.neucom.2026.133061

Jin Li , Kleanthis Malialis , Christos G. Panayiotou , Marios M. Polycarpou

In today’s digital world, the generation of vast amounts of streaming data in various domains has become ubiquitous. However, much of this data is unlabeled, making it challenging to identify events, particularly anomalies. This task becomes even more formidable in nonstationary environments where model performance can deteriorate over time due to concept drift. To address these challenges, this paper presents a novel method, VAE++ESDD, which employs incremental learning and two-level ensembling: an ensemble of Variational AutoEncoders (VAEs) for anomaly prediction, along with an ensemble of concept drift detectors. Each drift detector utilizes a statistical-based concept drift mechanism. To evaluate the effectiveness of VAE++ESDD, we conduct a comprehensive experimental study using real-world and synthetic datasets characterized by severely or extremely low anomalous rates and various drift characteristics. Our study reveals that the proposed method significantly outperforms both strong baselines and state-of-the-art methods.

在当今的数字世界中，在各个领域产生大量流数据已经变得无处不在。然而，这些数据中有很多是未标记的，这使得识别事件，特别是异常事件具有挑战性。在非平稳环境中，由于概念漂移，模型性能会随着时间的推移而恶化，因此这项任务变得更加艰巨。为了解决这些挑战，本文提出了一种新的方法，VAE++ESDD，它采用增量学习和两级集成：用于异常预测的变分自动编码器（VAEs）的集成，以及概念漂移检测器的集成。每个漂移检测器利用基于统计的概念漂移机制。为了评估VAE++ESDD的有效性，我们使用真实世界和合成数据集进行了全面的实验研究，这些数据集具有严重或极低的异常率和各种漂移特征。我们的研究表明，所提出的方法显着优于强基线和最先进的方法。

引用次数: 0