Neurocomputing最新文献

{"title":"Graph neural architecture for dynamic hybrid flowshop: Addressing stochastic events and uncertain processing sequences","authors":"Weixiang Xu ,&nbsp;Xiaochuan Luo (Co-ordinator) ,&nbsp;Yejian Zhao ,&nbsp;Yulin Zhang","doi":"10.1016/j.neucom.2025.130636","DOIUrl":"10.1016/j.neucom.2025.130636","url":null,"abstract":"<div><div>Dynamic hybrid flow shop scheduling (DHFSP) presents substantial difficulties in attaining optimal production efficiency. These challenges arise particularly when addressing indeterminate job sequences and stochastic events. Current dynamic scheduling methodologies employing deep reinforcement learning (DRL) predominantly target static-scale problems while exhibiting limited scalability to evolving complexities in industrial production systems. To enable responsive shop-floor decisions, this work introduces a hierarchical interaction attention-driven multi-graph network (HIAMG) framework engineered for DHFSP with uncertain job sequences and stochastic events. The proposed architecture implements graph-structured scenario encoding to address neural network dimension constraints. It is coupled with a novel graph embedding paradigm that hierarchically models job-machine interdependencies. Through explicit incorporation of stochastic events in scheduling simulations, our approach achieves enhanced operational fidelity reflecting real manufacturing settings. Integrating attention-based mechanisms further empowers the model to dynamically prioritize critical scheduling parameters and self-tune to environmental fluctuations. Comprehensive simulations across diverse operational regimes demonstrate that HIAMG surpasses conventional scheduling benchmarks in both performance metrics and configuration adaptability.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130636"},"PeriodicalIF":5.5,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144364847","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":"LVSA: Lightweight and verifiable secure aggregation for federated learning","authors":"Gongli Li ,&nbsp;Zhe Zhang ,&nbsp;Ruiying Du","doi":"10.1016/j.neucom.2025.130712","DOIUrl":"10.1016/j.neucom.2025.130712","url":null,"abstract":"<div><div>Federated learning (FL) is a decentralized machine learning paradigm that facilitates collaborative training of global models through the exchange of local gradients while maintaining the confidentiality of raw data. However, recent studies have identified gradient leakage attacks and server-forged aggregation results as significant threats to user data privacy. This issue is especially pronounced in large-scale mobile devices (e.g., tablets, smartphones, and smartwatches), which store highly sensitive user data, making the protection of such data critical. In addition, it is essential to consider the limitations of mobile devices, such as potential power outages, disconnections, and their limited computational and communication resources. To address these challenges, LVSA, a lightweight and verifiable secure aggregation scheme is proposed. LVSA employs a non-interactive masking scheme to protect gradient privacy and allows any user to drop out at any stage. Moreover, a lightweight verification method based on the inner product is introduced, which eliminates complex computations and is more suitable for devices with limited computational resources. Security analysis shows that LVSA not only protects users’ original gradients from being leaked, but also verifies the correctness of the aggregation results. Experimental analysis shows that when the gradient dimension reaches <span><math><msup><mrow><mn>10</mn></mrow><mrow><mn>6</mn></mrow></msup></math></span>, the computation time in LVSA is two orders of magnitude faster than the most advanced existing schemes. In addition, the communication overhead for users is reduced by more than eight times compared to other schemes offering the same functionality.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130712"},"PeriodicalIF":5.5,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144364896","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":"Heterogeneous multi-agent reinforcement learning for zero-shot scalable collaboration","authors":"Xudong Guo,&nbsp;Daming Shi,&nbsp;Junjie Yu,&nbsp;Wenhui Fan","doi":"10.1016/j.neucom.2025.130716","DOIUrl":"10.1016/j.neucom.2025.130716","url":null,"abstract":"<div><div>The emergence of multi-agent reinforcement learning (MARL) is significantly transforming various fields like autonomous vehicle networks. However, real-world multi-agent systems typically contain multiple roles, and the scale of these systems dynamically fluctuates. Consequently, in order to achieve zero-shot scalable collaboration, it is essential that strategies for different roles can be updated flexibly according to the scales, which is still a challenge for current MARL frameworks. To address this, we propose a novel MARL framework named <em>S</em>calable and <em>H</em>eterogeneous <em>P</em>roximal <em>P</em>olicy <em>O</em>ptimization <em>(SHPPO)</em>, integrating heterogeneity into parameter-shared PPO-based MARL networks. We first leverage a latent network to learn strategy patterns for each agent adaptively. Second, we introduce a heterogeneous layer to be inserted into decision-making networks, whose parameters are specifically generated by the learned latent variables. Our approach is scalable as all the parameters are shared except for the heterogeneous layer, and gains both inter-individual and temporal heterogeneity, allowing SHPPO to adapt effectively to varying scales. SHPPO exhibits superior performance in classic MARL environments like Starcraft Multi-Agent Challenge (SMAC) and Google Research Football (GRF), showcasing enhanced zero-shot scalability, and offering insights into the learned latent variables’ impact on team performance by visualization.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130716"},"PeriodicalIF":5.5,"publicationDate":"2025-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144365887","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":"Non-separation method-based finite-time synchronization of discrete-time fractional-order complex-valued neural networks with time-varying delays","authors":"Weiheng He ,&nbsp;Feifei Du ,&nbsp;Shilong Zhang","doi":"10.1016/j.neucom.2025.130732","DOIUrl":"10.1016/j.neucom.2025.130732","url":null,"abstract":"<div><div>In this paper, the finite-time synchronization (FTS) of discrete-time fractional-order complex-valued neural networks (FOCVNNs) with time-varying delays is studied using a non-separation method. In the literature, fractional-order Gronwall-type inequalities are typically used to analyze the FTS of fractional-order delayed systems, with the norm being estimated by an increasing function. However, for stable delay systems where the solution norm tends to zero, such estimations are not optimal. To improve the estimate for stable systems, a nabla Caputo difference inequality is first rigorously derived. Next, a delayed complex-valued adaptive nonlinear controller is designed, and a verifiable FTS criterion based on the established inequality is proposed, ensuring that the error norm is estimated by a decreasing function. Finally, numerical simulations are performed to validate the effectiveness of the proposed theoretical results.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130732"},"PeriodicalIF":5.5,"publicationDate":"2025-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144364846","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":"MLGINet: An efficient network for short-term load forecasting based on multi-scale learning and local–global interactive learning","authors":"Xinzhi Ding ,&nbsp;Dayi Li ,&nbsp;Geshuai Luo ,&nbsp;Wei Huang ,&nbsp;Yiping Chen ,&nbsp;Qingyong Zhang ,&nbsp;Quan Zhou","doi":"10.1016/j.neucom.2025.130704","DOIUrl":"10.1016/j.neucom.2025.130704","url":null,"abstract":"<div><div>For the power grid to operate in a safe, stable, and dependable manner, accurate short-term load forecasting (STLF) is crucial. As smart grid develops, numerous deep learning models have been proposed for STLF. Some studies devote to decompose the input signal to obtain different signal components by transforming the signal into other domains to mitigate noise and singular value interference. However, existing approaches struggle to perceive subtle signal fluctuations and to integrate short-range and long-range dependencies to ensure the robustness and accuracy of the model. To address these issues, we introduce multi-scale learning to mine multi-granularity semantic knowledge and local–global interactive learning to fuse instructive feature information, thereby improving discriminative representation and optimizing the model’s decision boundary. Specifically, a novel multi-scale sampling (MSS) block including sliding window sampling reorganization and interval sampling reorganization is embedded in backbone to extract discrete signal components at different scales by adjusting sampling field and sampling interval. Accordingly, the load sequence’s dynamic change and periodic trend can aid in the learning of model. Afterwards, the multi-scale interactive attention including intra-signal attention and inter-signal attention is designed to facilitate multi-granularity semantic knowledge interaction within and between multi-scale signal features and filter out valuable local and global information. The designed strategy enhances the model’s ability to resist forgetting and improves its sensitivity to multi-scale perturbations. Experimental results on public datasets collected from Australia, Morocco and China demonstrate that our method outperforms other nine state-of-the-art (SOTA) models by at least on average 35.50% in MAE, MSE, RMSE and MAPE. The model provides a potential idea for load forecasting methods.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130704"},"PeriodicalIF":5.5,"publicationDate":"2025-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144365115","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":"Network dynamics mechanisms underlying the instigation and propagation of cortical spreading depression","authors":"Chang He ,&nbsp;Denggui Fan ,&nbsp;Weiping Wang ,&nbsp;Qingyun Wang ,&nbsp;Gerold Baier","doi":"10.1016/j.neucom.2025.130733","DOIUrl":"10.1016/j.neucom.2025.130733","url":null,"abstract":"<div><div>Cortical spreading depression (CSD) waves are widely recognized as the pathophysiological mechanism underlying migraine aura. Modeling the macroscopic phenomenological characteristics of CSD wave propagation is challenging due to the inability to capture biophysical features, while microscopic studies based on excitatory–inhibitory (E/I) neuron pairs struggle to link effectively with wave propagation behaviors. In order to couple the electrical activity of micro neurons with the macroscopic propagation behavior of the cortex, we adopt a network perspective and constructed a dual-layer ring network model. Within this unified framework, we identify four factors influencing CSD instigation and propagation: (i) the type and number of pathological neurons, (ii) the extracellular potassium concentration, (iii) the ratio of excitatory to inhibitory connections within the cortical network, and (iv) the architecture of network connectivity incorporating both short and long-range connections. Model results indicate counterintuitively that the number of initially pathological neurons does not significantly correlate with CSD propagation duration. The extracellular potassium concentration required for CSD instigation within the network is lower than that for single neurons, suggesting that coexisting cluster discharges alongside CSD may contribute to the comorbidity of epilepsy and migraine. An excessive imbalance in the E/I ratio can induce global re-entrant and retracting phenomena of CSD, whereas a higher proportion of long-range connections within the network can effectively reduce the probability of such occurrences. These findings suggest that designing intervention strategies that comprehensively consider these influential factors can effectively decrease the instigation probability of CSD or enhance the stability of brain networks during CSD propagation.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"649 ","pages":"Article 130733"},"PeriodicalIF":5.5,"publicationDate":"2025-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144365506","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":"SC-BSN: Shifted Convolutions Based Blind-Spot Network for self-supervised image denoising","authors":"Guo Yang,&nbsp;Chengyun Song,&nbsp;Minglong Xue,&nbsp;Jian Yu","doi":"10.1016/j.neucom.2025.130714","DOIUrl":"10.1016/j.neucom.2025.130714","url":null,"abstract":"<div><div>Self-supervised image denoising methods have garnered significant attention recently due to their ability to train solely on noisy images without requiring paired clean-noisy data. However, real-world noise is often spatially correlated, leading to poor performance in self-supervised algorithms that assume pixel-wise independent noise. To address this limitation, we design multi-branch directional shifted operations to create blind spots in different regions, which effectively disrupt noise correlation. Further, the Shifted Convolutions Blind-Spot Network (SC-BSN) is proposed for self-supervised denoising. This network leverages three distinct blind-spot branches with varying shifted distances to effectively balance noise correlation suppression and the preservation of local spatial structures. Finally, we develop the Complementary Random-Replacing Refinement (CR3) to complement denoising results instead of relying on the iterative averaging of R3. The new post-processing technique efficiently retains the details of denoised images. Experimental results demonstrate that SC-BSN outperforms existing self-supervised denoising methods across multiple datasets, achieving superior performance in both visual quality and quantitative metrics.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130714"},"PeriodicalIF":5.5,"publicationDate":"2025-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144322433","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":"Fixed/preassigned-time bipartite synchronization for delayed fractional-order multilayer multi-link signed networks via adaptive aperiodically semi-intermittent control","authors":"Qiu Peng,&nbsp;Xiaotang Zhang,&nbsp;Manchun Tan","doi":"10.1016/j.neucom.2025.130724","DOIUrl":"10.1016/j.neucom.2025.130724","url":null,"abstract":"<div><div>In this paper, the bipartite synchronization (BS) problem of fractional-order multilayer multi-link signed networks (FOMMKSNs) with internal time-varying delay, intra-layer and inter-layer coupling time-varying delays is studied in fixed-time or preassigned-time. Firstly, a mathematical model is established by integrating the fractional-order dynamics of nodes, the edges of multilayer and multi-link networks, and the positive and negative weights, making it more diverse and practical. Secondly, an expanded fixed-time aperiodically intermittent strategy theory is proposed, building on the existing theory, along with an improved estimation of the settling time (ST). Then, utilizing the signed graph theory and Lyapunov method, several sufficient conditions for fixed-time BS (FXTBS) of FOMMKSNs are derived through an adaptive aperiodically semi-intermittent control scheme. Furthermore, based on the FXTBS results, a new controller is designed to study the BS of FOMMKSNs within a specified time, where the ST is independent of any initial values and parameters of the network and the controller. Finally, the effectiveness of the obtained results is verified by two simulation examples.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130724"},"PeriodicalIF":5.5,"publicationDate":"2025-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144331272","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":"Individual vs. group dynamics: Impacts on equilibrium states in self-organizing systems","authors":"Xueyuan Li,&nbsp;Danilo Vasconcellos Vargas","doi":"10.1016/j.neucom.2025.130575","DOIUrl":"10.1016/j.neucom.2025.130575","url":null,"abstract":"<div><div>Recent research highlights the critical need to learn complex structures from data and adapt to evolving data patterns. In this work, we introduce the Decentralized SyncMap model, which shifts the focus from group-based to individual-based interactions in order to reveal more meaningful relationships while operating in lower-dimensional spaces. To improve the model’s robustness and adaptability by incorporating past information, we further propose a temporal memory mechanism. The temporal memory of attractors enhances the model’s ability to capture rare but important features, while the temporal memory of repellers helps the model identify local chunk patterns. By leveraging these individual dynamics, the model naturally preserves internal structures that traditional group-level methods tend to overlook. Our experiments on synthetic chunking tasks show that the Decentralized SyncMap model achieves an average NMI of 0.876, outperforming Word2Vec and both the Original and Symmetrical SyncMap models by 12.6%, 10.6%, and 5.5%, respectively. On real-world datasets, it attains an average NMI of 0.840, performing better than competing approaches by 400.1%, 52.7%, and 4.6%. These results suggest that a model emphasizing individual dynamics provides a more effective means of uncovering latent patterns in complex, evolving data.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130575"},"PeriodicalIF":5.5,"publicationDate":"2025-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144322533","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":"Multimodal prototypical networks with Co-metric fusion for few-shot hyperspectral image classification","authors":"Yuhang Li ,&nbsp;Jinrong He ,&nbsp;Hanchi Liu ,&nbsp;Yurong Zhang ,&nbsp;Zhaokui Li","doi":"10.1016/j.neucom.2025.130782","DOIUrl":"10.1016/j.neucom.2025.130782","url":null,"abstract":"<div><div>In the field of Hyperspectral image (HSI) classification, prototype-based network methods have achieved significant research progress. These methods utilize pixel-level information from images to construct central prototypes for each class, providing effective solutions for few-shot learning. However, traditional prototype networks have some inherent flaws; they primarily rely on a single image modality and fail to fully leverage the potential complementarity between different modalities, using only a single modality to generate class prototypes, which limits the model's performance in representing class prototypes and enhancing discriminative capabilities. And subtle inter-class differences are also a challenging task in cross-domain scenarios. To overcome these challenges, this study proposes an innovative Multimodal Prototypical Networks with Co-metric Fusion (MPCF). By integrating prototype information from both image and text modalities, MPCF significantly enhances the performance of few-shot learning. The method not only captures the spectral and spatial features of images to construct image prototypes but also extracts textual features from category descriptions to generate text prototypes. Furthermore, by integrating contrastive learning strategies with the Co-metric fusion mechanism, the method effectively harnesses the information from different modalities. This integration allows for the capture of category information across multiple dimensions, significantly boosting the model's discriminative power among various classes and enhancing its capacity to address few-shot learning scenarios. Experiments conducted on several public benchmark HSI datasets (Indian Pines-84.06 %, Houston-80.41 %, Salinas-92.63 %) demonstrate that MPCF exhibits excellent performance under few-shot and cross-domain conditions, achieving higher classification accuracy and stronger robustness compared to state-of-the-art methods. The related code will be made publicly available at the following URL: <span><span>https://github.com/AIYAU/MPCF</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"648 ","pages":"Article 130782"},"PeriodicalIF":5.5,"publicationDate":"2025-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144364898","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}