Neurocomputing最新文献

{"title":"Joint entropy search for multi-objective Bayesian optimization with constraints and multiple fidelities","authors":"Daniel Fernández-Sánchez,&nbsp;Daniel Hernández-Lobato","doi":"10.1016/j.neucom.2025.131674","DOIUrl":"10.1016/j.neucom.2025.131674","url":null,"abstract":"<div><div>Bayesian optimization (BO) methods can be used to solve efficiently problems with several objectives and constraints. Each objective and constraint is considered a black-box function that is expensive to evaluate, lacking a closed-form expression. BO methods use a model of each black-box to guide the search for the problem’s solution. Specifically, they make intelligent decisions about where each black-box function should be evaluated next with the goal of finding the solution using a few evaluations only. Sometimes, however, the black-boxes may be evaluated at different fidelity levels. A lower fidelity is simply a cheap proxy for the corresponding black-box. These lower fidelities correlate with the actual black-boxes to optimize and can, therefore, be used to reduce the overall cost of solving the optimization problem. Here, we propose Multi-fidelity Joint Entropy Search for Multi-objective Bayesian Optimization with Constraints (MF-JESMOC), a BO method for solving the aforementioned problems. MF-JESMOC chooses the next point, and fidelity level at which to evaluate the black-boxes, as the combination that is expected to reduce the most the joint entropy of the Pareto set and the Pareto front, normalized by the fidelity’s evaluation cost. We use Deep Gaussian processes to model each black-box and the dependencies between fidelities. These are powerful probabilistic models that can learn the dependency structure among fidelity levels of each black-box. Several experiments show that MF-JESMOC outperforms other state-of-the-art methods for multi-objective BO with constraints and different fidelity levels in both synthetic and real-world problems.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"657 ","pages":"Article 131674"},"PeriodicalIF":6.5,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145222624","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}

{"title":"Interactive attention and contrastive learning for few-shot relation extraction","authors":"Yan Li ,&nbsp;Yao Wang ,&nbsp;Zhaojie Wang ,&nbsp;Wei Wang ,&nbsp;Bailing Wang ,&nbsp;Guodong Xin","doi":"10.1016/j.neucom.2025.131551","DOIUrl":"10.1016/j.neucom.2025.131551","url":null,"abstract":"<div><div>Relation extraction is a critical task in natural language processing, often challenged by the problem of insufficient samples in real world scenarios. Therefore, studying few-shot relation extraction is of great significance. Currently, prototype networks and meta-learning-based parameter optimization are the mainstream methods to study this kind of problem. However, these methods still face sample confusion during classification, and the trained models are prone to overfitting. To solve these problems, this paper proposes a few-shot relation extraction method based on interactive attention. During the model training stage, we introduce two contrastive learning approaches to better capture sample features and reduce sample confusion. Contrastive learning strengthens the connections between instances and their corresponding relationship descriptions, thus improving relation extraction. In the testing phase, the model employs an attention mechanism to calculate the attention scores between the query set and the support set and employs a new classification layer to mitigate overfitting. We conducted experiments on two real-world few-shot relation extraction datasets, and the results demonstrate that our method achieved superior performance on both in-domain and cross-domain datasets, proving the effectiveness of the proposed approach. The code is available at <span><span>https://github.com/xyzew/IACL.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131551"},"PeriodicalIF":6.5,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145270628","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}

{"title":"The development and future of digital rights management: A review","authors":"Xue Feng ,&nbsp;Yijie Pan ,&nbsp;Nai-an Xiao","doi":"10.1016/j.neucom.2025.131672","DOIUrl":"10.1016/j.neucom.2025.131672","url":null,"abstract":"<div><div>Digital rights management (DRM) serves as a critical technological mechanism for copyright protection, ensuring the legitimate use of digital content, and facilitating innovative business models in content distribution and access. This paper begins by introducing the fundamental concepts and typical architecture of DRM systems. It then provides a detailed analysis of the four distinct evolutionary phases of DRM, with a focus on key technologies including usage control, rights expression, content sharing, and decentralization. The paper further examines DRM standards, legal implications, and the tension between enforcement and fair use. Finally, it outlines future challenges and suggests promising directions for future research.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131672"},"PeriodicalIF":6.5,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145236196","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}

{"title":"KING: An efficient optimization approach","authors":"Dong Zhao ,&nbsp;Zhen Wang ,&nbsp;Yupeng Li ,&nbsp;Ali Asghar Heidari ,&nbsp;Zongda Wu ,&nbsp;Yi Chen ,&nbsp;Huiling Chen","doi":"10.1016/j.neucom.2025.131645","DOIUrl":"10.1016/j.neucom.2025.131645","url":null,"abstract":"<div><div>Real-world engineering optimization problems are often highly challenging due to narrow feasible regions, numerous local optima, and intricate constraints. Metaheuristic algorithms (MAs) have shown promise in addressing these issues owing to their global search capability, flexibility, and adaptability. However, a critical challenge with MAs is effectively balancing the global search (exploration) and local search (exploitation) phases, which significantly influences the efficiency and precision of convergence. Many MAs require problem-specific adjustments to control convergence behavior, thereby increasing computational cost and implementation effort. Moreover, existing improvements are often tailored to specific problems, lacking comprehensive validation in terms of generality, robustness, and scalability. To overcome these limitations, this paper proposes a novel high-performance optimization algorithm with enhanced adaptability, named the Three Kingdoms Optimization Algorithm (KING), inspired by historical dynamics of the Three Kingdoms period in China. We establish an analogy between key components of MAs—such as population initialization, exploration, and exploitation—and four historical phases: the ascent of the might, joint confrontation, three-legged tripod, and whole country united. KING incorporates a new reinforcement convergence mechanism to systematically guide the search process while maintaining an effective balance between exploration and exploitation, enabling rapid and efficient convergence. Additionally, a dynamic, tolerance-based constraint-handling technique is introduced to strengthen its capability in solving complex constrained problems. The performance of KING is extensively evaluated on the IEEE CEC 2017 and IEEE CEC 2022 benchmark test suites, comparing it with classical algorithms, high-performance variants, and state-of-the-art methods across problems of varying scales. Experimental results demonstrate that KING outperforms the compared algorithms in convergence speed, solution accuracy, and stability. Its superiority is further validated through applications to four real-world engineering problems. The proposed algorithm proves to be an effective and reliable tool for engineering optimization. Its source code will be made publicly available at <span><span>https://aliasgharheidari.com/KING.html</span><svg><path></path></svg></span> and other websites.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"657 ","pages":"Article 131645"},"PeriodicalIF":6.5,"publicationDate":"2025-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145223328","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}

{"title":"Both reliable and unreliable predictions matter: Domain adaptation for bearing fault diagnosis without source data","authors":"Wenyi Wu ,&nbsp;Hao Zhang ,&nbsp;Zhisen Wei ,&nbsp;Xiao-Yuan Jing ,&nbsp;Qinghua Zhang ,&nbsp;Songsong Wu","doi":"10.1016/j.neucom.2025.131661","DOIUrl":"10.1016/j.neucom.2025.131661","url":null,"abstract":"<div><div>Rolling bearing fault diagnosis is crucial for maintaining the reliability and safety of industrial systems. Recently, it has attracted increasing attention to transferring a diagnosis model from the source domain to the target domain without source data in real-world diagnosis scenarios due to confidentiality and efficiency concerns. However, existing approaches are sub-optimal as they simply exploit confidently pseudo-labeled target samples, and simultaneously overlook the intrinsic structural characteristics of the feature space. Besides, the reliability of fault pseudo-labels is always estimated with entropy, whose accuracy could be improved through more sophisticated strategies. To address these issues, we propose to explore the correlation between features and pseudo-labels in the target domain to maintain the balance between feature discriminability and feature diversity. In addition, we develop a voting-based strategy associated with data augmentation for more accurate reliability estimation of fault pseudo-labels. The proposed method is able to utilize both the reliable samples and unreliable samples for diagnosis model transfer via self-supervised training and distribution structure discovering respectively. Extensive experiments on two bearing fault benchmarks demonstrate the effectiveness and superiority of our proposed method. The source code is publicly available at: <span><span>https://github.com/BdLab405/SDALR</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"657 ","pages":"Article 131661"},"PeriodicalIF":6.5,"publicationDate":"2025-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145222623","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}

{"title":"Conditional plane-based multi-scene representation for novel view synthesis","authors":"Uchitha Rajapaksha ,&nbsp;Hamid Laga ,&nbsp;Dean Diepeveen ,&nbsp;Mohammed Bennamoun ,&nbsp;Ferdous Sohel","doi":"10.1016/j.neucom.2025.131657","DOIUrl":"10.1016/j.neucom.2025.131657","url":null,"abstract":"<div><div>Existing explicit and implicit-explicit hybrid neural representations for novel view synthesis are scene-specific. In other words, they represent only a single scene and require retraining for every novel scene. Implicit scene-agnostic methods rely on large multilayer perception (MLP) networks conditioned on learned features. They are computationally expensive during training and rendering times. In contrast, we propose a novel plane-based representation that learns to represent multiple static and dynamic scenes during training and renders per-scene novel views during inference. The method consists of a deformation network, explicit feature planes, and a conditional decoder. Explicit feature planes are used to represent a time-stamped view space volume and a shared canonical volume across multiple scenes. The deformation network learns the deformations across shared canonical object space and time-stamped view space. The conditional decoder estimates the color and density of each scene constrained by a scene-specific latent code. We evaluated and compared the performance of the proposed representation on static (NeRF) and dynamic (Plenoptic videos) datasets. The results show that explicit planes combined with tiny MLPs can efficiently train multiple scenes simultaneously. The project page: <span><span>https://anonpubcv.github.io/cplanes/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"657 ","pages":"Article 131657"},"PeriodicalIF":6.5,"publicationDate":"2025-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145222621","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}

{"title":"A lightweight convolution and vision transformer integrated model with multi-scale self-attention mechanism","authors":"Yi Zhang ,&nbsp;Lingxiao Wei ,&nbsp;Bowei Zhang ,&nbsp;Ziwei Liu ,&nbsp;Kai Yi ,&nbsp;Shu Hu","doi":"10.1016/j.neucom.2025.131670","DOIUrl":"10.1016/j.neucom.2025.131670","url":null,"abstract":"<div><div>Vision Transformer (ViT) has prevailed in computer vision tasks due to its strong long-range dependency modelling ability. However, its large model size and weak local feature modeling ability hinder its application in real scenarios. To balance computational efficiency and performance in downstream vision tasks, we propose an efficient ViT model with sparse attention (dubbed SAEViT) and convolution blocks. Specifically, a Sparsely Aggregated Attention (SAA) module has been proposed to perform adaptive sparse sampling and recover the feature map via deconvolution operation, which significantly reduces the computational complexity of attention operations. In addition, a Channel-Interactive Feed-Forward Network (CIFFN) layer is developed to enhance inter-channel information exchange through feature decomposition and redistribution, which mitigates the redundancy in traditional feed-forward networks (FFN). Finally, a hierarchical pyramid structure with embedded depth-wise separable convolutional blocks (DWSConv) is devised to further strengthen convolutional features. Extensive experiments on mainstream datasets show that SAEViT achieves Top-1 accuracies of 76.3 % and 79.6 % on the ImageNet-1 K classification task with only 0.8 GFLOPs and 1.3 GFLOPs, respectively, demonstrating a lightweight solution for fundamental vision tasks.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131670"},"PeriodicalIF":6.5,"publicationDate":"2025-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271200","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}

{"title":"Fused adaptive tensor log-determinant and local smoothness regularizer for multi-view clustering","authors":"Fei Wang,&nbsp;Gui-Fu Lu","doi":"10.1016/j.neucom.2025.131564","DOIUrl":"10.1016/j.neucom.2025.131564","url":null,"abstract":"<div><div>The prevailing techniques for multi-view subspace clustering (MVC) methods often depend on the assumption of low-rankness, which asserts that data can be effectively represented in a low-dimensional subspace. While these approaches capture the structure of the data globally and remove noise and redundancy, they all neglect local smoothness prior, which has been extensively used to reduce noise in the image field. Besides, existing techniques often depend on the tensor nuclear norm (TNN)to approximate the intrinsically non-convex tensor rank function. However, the TNN approach equates all singular values, which gives rise to excessive penalization of the principal rank components and ultimately leads to sub-optimal tensor representations. In response to these challenges, we introduce an innovative method called fused adaptive tensor Log-determinant and local smoothness regularizer (FATLLSR) for multi-view clustering. Specifically, we initially derive the self-expressive matrix for each view and subsequently integrate these matrices into a tensor. Then in order to simultaneously explore low-rankness and local smoothness prior, FATLLSR is designed and is used to constrain the obtained tensor. By using FATLLSR, we can not only relax tensor multi-rank constraint better than TNN but also utilize the local smoothness information hidden in multi-view data, making our method more robust to noise and redundancy. These techniques are integrated to constitute a unified model that is effectively handled using the augmented Lagrange multiplier (ALM). As demonstrated by its performance on different datasets, FATLLSR achieves outstanding clustering performance compared to the most advanced methods. The code is publicly available at <span><span>https://github.com/wangfii/FATLLSR</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131564"},"PeriodicalIF":6.5,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271202","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}

{"title":"A survey of neural signal decoding based on domain adaptation","authors":"Suchen Li ,&nbsp;Zhuo Tang ,&nbsp;Mengmeng Li ,&nbsp;Lifang Yang ,&nbsp;Zhigang Shang","doi":"10.1016/j.neucom.2025.131653","DOIUrl":"10.1016/j.neucom.2025.131653","url":null,"abstract":"<div><div>An important objective in brain-computer interfaces (BCIs) is to develop robust and reliable neural signal decoders. However, the decoders will encounter challenges under cross-subject or cross-session conditions due to the randomness, non-stationarity, and individual variability of brain electrical activity. Reducing distributional differences is an exceptionally intuitive way to eliminate inter-subject/session differences and enhance decoder generalizability. In this context, domain adaptation (DA) emerges as a valuable technique, enabling the rapid transfer of knowledge acquired from large datasets with labeled data to new subjects or sessions. This paper provides a comprehensive survey of DA research in neural decoding from 2014 to the present. We categorize neural decoding methods related to DA by considering instance-based, feature-based, and model-based, which is motivated by three fundamental challenges in DA: How can one effectively select suitable source domains or samples for transfer? How can inter-domain distributional differences be minimized through feature space transformation? And how can decoder parameters be optimally shared? Additionally, several decoding methods that combine deep learning with DA are highlighted, given the significant advantages of deep learning over traditional feature extraction techniques. Furthermore, our paper explores the application of DA in complex scenarios, such as multiple source domains and low-resource settings. In summary, we have reviewed domain-adaptive decoding algorithms and their application considerations, while identifying various challenges that need to be addressed in future research.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"657 ","pages":"Article 131653"},"PeriodicalIF":6.5,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145223204","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}

{"title":"A 3D UNet-based fusion network for brain tumor segmentation with missing modalities","authors":"Yutian Xiao ,&nbsp;Xiaomao Fan ,&nbsp;Yuanyuan Liao ,&nbsp;Chongguang Yang ,&nbsp;Yang Zhao","doi":"10.1016/j.neucom.2025.131642","DOIUrl":"10.1016/j.neucom.2025.131642","url":null,"abstract":"<div><div>Multimodal magnetic resonance imaging provides complementary information for brain tumor segmentation, significantly enhancing the accuracy of diagnosis and prognosis. However, the common issue of missing modalities in clinical practice severely undermines the performance of existing methods, as they predominantly rely on complete multimodal data and struggle to effectively handle dynamic inter-modality correlations and tumor region specificity. To address this challenge, we propose a novel fusion network based on 3D U-Net, termed MPDF-UNET. Its core innovation lies in the introduction of the Modality Priors and Dynamic Features fusion (MPDF) module, which adaptively learns the unique representations of different MRI modalities under conditions of partial modality loss while effectively integrating complementary information across modalities. Additionally, we develop a modality combination sampling strategy that dynamically adjusts the distribution of modality combinations in the training data. This strategy encourages the network to fully exploit prior knowledge from each modality, thereby enhancing model robustness under conditions of missing modalities. To mitigate the impact of missing modality-associated dynamic feature information, we further propose a feature loss function. By imposing constraints on dynamic features, this loss function facilitates the learning of modality priors, alleviating the degradation of the network’s representational capacity caused by missing modalities. Experiments conducted on BRATS2018 and BRATS2020 benchmark datasets demonstrate the superiority of MPDF-UNET. Notably, the model achieves significant improvements in the fine-grained segmentation of enhancing tumors, surpassing current SOTA. Specifically, on BRATS2018 dataset, our method improves the Dice score of enhancing tumor segmentation by 7.78 % on average compared to the best-performing baseline Region-aware Fusion Network (RFNet), demonstrating superior robustness under missing modalities. This work provides a reliable solution for incomplete or resource-limited multimodal data in clinical settings, demonstrating significant practical value.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"657 ","pages":"Article 131642"},"PeriodicalIF":6.5,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145269569","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}