Neural Networks最新文献

{"title":"Nonlinear multi-view clustering for non-negative matrix factorization","authors":"Jinrong Cui ,&nbsp;Bang Liufu ,&nbsp;Yulu Fu ,&nbsp;Meihua Wang ,&nbsp;Zhihui Lai","doi":"10.1016/j.neunet.2025.107744","DOIUrl":"10.1016/j.neunet.2025.107744","url":null,"abstract":"<div><div>Deep multi-view clustering methods have made significant progress in recent years, benefiting from their wide parameter space and the ability to consider more details in the learning procedure. However, the existing methods suffer from the following problems: (1) Deep models based on data-driven learning often involve opaque update processes and struggle to maintain stability across varying data scales. (2) Non-negative Matrix Factorization (NMF)-based clustering methods generally exhibit limited robustness due to their restricted nonlinear fitting capabilities and narrow parameter spaces. To address these issues, we propose a nonlinear non-negative matrix factorization multi-view clustering framework. Our framework integrates traditional NMF optimization principles into a deep model to enhance interpretability and stability. In addition, it employs partially parameterized NMF iterations to improve nonlinear fitting ability, thereby expanding the parameter space and enhancing model robustness. We also introduce a cross-view contrastive loss to guide the model in learning inter-view diversity and cluster-friendly structural features. Experiments on multiple datasets show that our method outperforms state-of-the-art clustering methods.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107744"},"PeriodicalIF":6.0,"publicationDate":"2025-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144545840","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CTFS: A consolidated transformer framework for instance and semantic segmentation tasks","authors":"Kun Dai ,&nbsp;Fuyuan Qiu ,&nbsp;Hongbo Gao ,&nbsp;Tao Xie ,&nbsp;Chuqing Cao ,&nbsp;Ruifeng Li ,&nbsp;Lijun Zhao ,&nbsp;Ke Wang","doi":"10.1016/j.neunet.2025.107745","DOIUrl":"10.1016/j.neunet.2025.107745","url":null,"abstract":"<div><div>Instance segmentation and semantic segmentation are fundamental tasks that support many computer vision applications. Recently, researchers have investigated the feasibility of constructing a unified transformer framework and leveraging multi-task learning techniques to optimize instance and semantic segmentation tasks simultaneously. However, these methods learn the proportion and distribution of task-shared parameters concurrently during the training process, which inevitably presents a challenge to sufficiently optimize the network. In addition, conventional gradient rectification algorithms attempt to address gradient conflicts from an overall perspective, but they fall short of adequately resolving conflicts among individual elements within gradient vectors. In this study, we develop a consolidated Transformer framework CTFS to address these issues. To address the first issue, we introduce an affinity-guided sharing strategy (AGSS) that learns the proportion and distribution of task-shared parameters in two separate stages. This approach leverages the proportion of task-shared parameters as prior knowledge to guide the subsequent learning process, reducing the difficulty of network optimization. To address the second issue, we propose a fine-grained gradient rectification strategy (FGRS) that effectively mitigates gradient conflicts for each element in gradient vectors during backpropagation. Built upon the standard Swin Transformer without complicating its network architecture, CTFS achieves impressive performance on both the COCO dataset for the instance segmentation task and the ADE20K dataset for the semantic segmentation task.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107745"},"PeriodicalIF":6.0,"publicationDate":"2025-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144489826","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Rethinking mixture of rain removal via depth-guided adversarial learning","authors":"Yongzhen Wang ,&nbsp;Xuefeng Yan ,&nbsp;Yanbiao Niu ,&nbsp;Lina Gong ,&nbsp;Yanwen Guo ,&nbsp;Mingqiang Wei","doi":"10.1016/j.neunet.2025.107739","DOIUrl":"10.1016/j.neunet.2025.107739","url":null,"abstract":"<div><div>Rainy weather significantly deteriorates the visibility of scene objects, particularly when images are captured through outdoor camera lenses or windshields. Through careful observation of numerous rainy photos, we have discerned that the images are typically affected by various rainwater artifacts such as raindrops, rain streaks, and rainy haze, which impair the image quality from near to far distances, resulting in a complex and intertwined process of image degradation. However, current deraining techniques are limited in their ability to address only one or two types of rainwater, which poses a challenge in removing the mixture of rain (MOR). In this study, we naturally associate scene depth with the MOR effect and propose an effective image deraining paradigm for the Mixture of Rain Removal, termed DEMore-Net. Going beyond the existing deraining wisdom, DEMore-Net is a joint learning paradigm that integrates depth estimation and MOR removal tasks to achieve superior rain removal. The depth information can offer additional meaningful guidance information based on distance, thus better helping DEMore-Net remove different types of rainwater. Moreover, this study explores normalization approaches in image deraining tasks and introduces a new Hybrid Normalization Block (HNB) to enhance the deraining performance of DEMore-Net. Extensive experiments conducted on synthetic datasets and real-world MOR photos fully validate the superiority of DEMore-Net. Code is available at <span><span>https://github.com/yz-wang/DEMore-Net</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107739"},"PeriodicalIF":6.0,"publicationDate":"2025-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144502054","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A lightweight spiking neural network for EEG-based motor imagery classification","authors":"Herui Zhang,&nbsp;Haoran Wang,&nbsp;Jiayu An,&nbsp;Shitao Zheng,&nbsp;Dongrui Wu","doi":"10.1016/j.neunet.2025.107741","DOIUrl":"10.1016/j.neunet.2025.107741","url":null,"abstract":"<div><div>Spiking neural networks (SNNs) aim to simulate the human brain neural network, using sparse spike event streams for effective and energy-efficient spatio-temporal signal processing. This paper proposes a lightweight SNN model for electroencephalogram (EEG) based motor imagery (MI) classification, a classical brain–computer interface paradigm. The model has three desirable characteristics: (1) it has a brain-inspired architecture; (2) it is energy efficient; and, (3) it is dataset agnostic. Within-subject and cross-subject experiments on three public datasets demonstrated the superiority of our SNN model over four classical convolutional neural network based models in EEG based MI classification.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107741"},"PeriodicalIF":6.0,"publicationDate":"2025-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144489825","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Few-shot event-based action recognition","authors":"Zanxi Ruan ,&nbsp;Nan Pu ,&nbsp;Jiangming Chen ,&nbsp;Songqun Gao ,&nbsp;Yanming Guo ,&nbsp;Qiuyu Kong ,&nbsp;Yuxiang Xie ,&nbsp;Yingmei Wei","doi":"10.1016/j.neunet.2025.107750","DOIUrl":"10.1016/j.neunet.2025.107750","url":null,"abstract":"<div><div>Despite the evident superiority of event cameras in practical vision applications (e.g., action recognition), owing to their distinctive sensing mechanism, existing event-based action recognition methods rely heavily on large-scale training data. However, the expensive cost of camera deployment and the requirement of data privacy protection make it challenging to collect substantial data in real-world scenarios. To address this limitation, we explore a novel yet practical task, Few-Shot Event-Based Action Recognition (FSEAR), which aims at leveraging a minimal number of intractable event action data for model training and accurately classifying unlabeled data into a specific category. Accordingly, we design a new framework for FSEAR, including a Noise-Aware Event Encoder (NAE) and a Distilled Prototypical Distance Fusion (DPDF). The former efficiently filters noise within the spatiotemporal domain while retaining vital information related to action timing. The latter conducts multi-scale measurements across geometric, directional, and distributional dimensions. These two modules benefit mutually and thus effectively exploit the potential characteristics of event data. Extensive experiments on four distinct event action recognition datasets have demonstrated the significant advantages of our model over other few-shot learning methods. Our code and models will be publicly released.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107750"},"PeriodicalIF":6.0,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144480379","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Learning multi-regularized mutation-aware correlation filter for object tracking via an adaptive hybrid model","authors":"Sathiyamoorthi Arthanari,&nbsp;Jae Hoon Jeong,&nbsp;Young Hoon Joo","doi":"10.1016/j.neunet.2025.107746","DOIUrl":"10.1016/j.neunet.2025.107746","url":null,"abstract":"<div><div>Discriminative Correlation Filters (DCF) have emerged as a popular and effective approach in object tracking. With promising performance and efficiency, DCF-based trackers achieved impressive attention and reliable tracking results in several challenging scenarios. Although DCF-based trackers improve tracking performance, they still suffer from unexpected factors such as appearance mutations, filter degradation, and target distortion, which leads to decreased tracker performance. To address these challenges, a novel Multi-Regularized Mutation-Aware Correlation Filter (MRMACF) approach is presented. To do this, we propose a mutation-aware strategy with an adaptive hybrid model that employs the mutation threat mechanism technique to effectively handle the appearance mutations and filter degradation issues when the filter deviates from the target location. The mutation threat mechanism identifies sudden and significant changes in the target object’s appearance, which is achieved by an adaptive hybrid model approach that compares the current appearance with recent historical models. Following that, we introduce an improved sparse spatial feature selection approach that incorporates row and column-based feature selection methods into the sparse spatial technique, which aims to identify crucial features within the target region and successfully address the problem of target distortion. Moreover, the surrounding-aware approach is presented that extracts the surrounding samples of the target region to utilize the context information, which prevents the filter deviation from the target and improves the discriminative ability. Specifically, the adaptive hybrid model approach is proposed to mitigate both tracking drift and the mutation threat of target by incorporating target position information from previous frames. Furthermore, we showcase the efficiency of the proposed MRMACF approach against existing modern trackers using the OTB-2013, OTB-2015, TempleColor-128, UAV-123, UAVDT, VOT-2018, LaSOT, and GOT-10K benchmark datasets. Specifically, our proposed method achieved the highest performance on the OTB-2015 dataset, with a DP score of 93.2% and an AUC score of 69.8%, respectively.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107746"},"PeriodicalIF":6.0,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144471026","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Path-aware multi-scale learning for heterogeneous graph neural network","authors":"Jin Fan ,&nbsp;Jiajun Yang ,&nbsp;Zhangyu Gu ,&nbsp;Huifeng Wu ,&nbsp;Danfeng Sun ,&nbsp;Feiwei Qin ,&nbsp;Jia Wu","doi":"10.1016/j.neunet.2025.107743","DOIUrl":"10.1016/j.neunet.2025.107743","url":null,"abstract":"<div><div>Heterogeneous Graph Neural Networks (HGNNs) are a powerful tool for modeling data with diverse node and edge types, found in applications like social networks, recommendation systems, and knowledge graphs, including tasks such as node classification, link prediction, and graph classification. Based on information aggregation methods, HGNNs can be broadly categorized into meta-path-free and meta-path-based HGNNs. Recently, meta-path-based HGNNs have made significant advancements in both performance and interpretability. However, these methods often overlook the redundancy among meta-paths and fail to fully leverage the inherent information within the paths, such as path length and path type. Furthermore, their insufficient utilization of global information hinders comprehensive representation learning. To address these issues, we propose a path-aware multi-scale heterogeneous graph neural network named PM-HGNN. To better capture global information, PM-HGNN employs a global similarity-based mean aggregator to pre-compute neighbor aggregation information. Additionally, PM-HGNN exploits the inherent relevance and differences between meta-paths, enabling redundancy reduction and the dynamic assignment of weights. Experiments conducted on four real-world heterogeneous graph datasets revealed that PM-HGNN consistently exceeds the performance of current state-of-the-art methods in tasks related to node classification.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107743"},"PeriodicalIF":6.0,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144489824","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Attacking cooperative multi-agent reinforcement learning by adversarial minority influence","authors":"Simin Li ,&nbsp;Jun Guo ,&nbsp;Jingqiao Xiu ,&nbsp;Yuwei Zheng ,&nbsp;Pu Feng ,&nbsp;Xin Yu ,&nbsp;Jiakai Wang ,&nbsp;Aishan Liu ,&nbsp;Yaodong Yang ,&nbsp;Bo An ,&nbsp;Wenjun Wu ,&nbsp;Xianglong Liu","doi":"10.1016/j.neunet.2025.107747","DOIUrl":"10.1016/j.neunet.2025.107747","url":null,"abstract":"<div><div>This study probes the vulnerabilities of cooperative multi-agent reinforcement learning (c-MARL) under adversarial attacks, a critical determinant of c-MARL’s worst-case performance prior to real-world implementation. Current observation-based attacks, constrained by white-box assumptions, overlook c-MARL’s complex <em>multi-agent</em> interactions and <em>cooperative</em> objectives, resulting in impractical and limited attack capabilities. To address these shortcomes, we propose <em>Adversarial Minority Influence</em> (AMI), a practical and strong for c-MARL. AMI is a practical black-box attack and can be launched without knowing victim parameters. AMI is also strong by considering the complex <em>multi-agent</em> interaction and the <em>cooperative</em> goal of agents, enabling a single adversarial agent to <em>unilaterally</em> misleads majority victims to form <em>targeted</em> worst-case cooperation. This mirrors minority influence phenomena in social psychology. To achieve maximum deviation in victim policies under complex agent-wise interactions, our <em>unilateral</em> attack aims to characterize and maximize the impact of the adversary on the victims. This is achieved by adapting a unilateral agent-wise relation metric derived from mutual information, thereby mitigating the adverse effects of victim influence on the adversary. To lead the victims into a jointly detrimental scenario, our <em>targeted</em> attack deceives victims into a long-term, cooperatively harmful situation by guiding each victim towards a specific target, determined through a trial-and-error process executed by a reinforcement learning agent. Through AMI, we achieve the first successful attack against real-world robot swarms and effectively fool agents in simulated environments into collectively worst-case scenarios, including Starcraft II and Multi-agent Mujoco. The source code and demonstrations can be found at: <span><span>https://github.com/DIG-Beihang/AMI</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107747"},"PeriodicalIF":6.0,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144534936","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Event-triggered ADP-based tracking controller for partially unknown nonlinear uncertain systems with input and state constraints","authors":"Raju Dahal,&nbsp;Indrani Kar","doi":"10.1016/j.neunet.2025.107752","DOIUrl":"10.1016/j.neunet.2025.107752","url":null,"abstract":"<div><div>This paper addresses the robust tracking control problem for nonlinear systems with unmatched uncertainties and partially unknown dynamics while also taking into account the input and state constraints. An event-triggered ADP framework is utilized to tackle this issue. Initially, an identifier neural network (NN) is designed to estimate the unknown system dynamics. Next, an augmented system is constructed using the reference trajectory and tracking error. The uncertainty is then divided into matched and unmatched components, converting the tracking control problem into an optimal regulation problem for an auxiliary system. A novel event-triggered safe HJB equation is developed by integrating a control barrier function (CBF) and a nonquadratic term within the cost function to enforce the safety constraints. A critic NN is utilized to solve this safe HJB equation. The controller is updated based on a triggering rule formulated using the Lyapunov approach. Lyapunov stability theory is applied to demonstrate that the closed-loop system is stable and that the identifier network and the critic network parameters remain uniformly ultimately bounded (UUB) under constraints and disturbances. The effectiveness of the proposed theoretical approach is validated using a simulation example.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107752"},"PeriodicalIF":6.0,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144517747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unified semantic space learning for cross-modal retrieval","authors":"Jie Zhu ,&nbsp;Jianan Liu ,&nbsp;Shufang Wu ,&nbsp;Feng Zhang","doi":"10.1016/j.neunet.2025.107756","DOIUrl":"10.1016/j.neunet.2025.107756","url":null,"abstract":"<div><div>With the increasing amount of multimodal data on the Internet, cross-modal retrieval has gradually become a hot research topic and has achieved significant progress, especially since graph convolutional networks were introduced. Most methods based on graph convolutional networks tend to focus on incorporating the correlations among samples and the correlations among labels into the common representations, but neglect the correlations among the semantic contents. Moreover, the semantic similarity between instances and semantic contents is also underutilized. To address these issues, we propose a Unified Semantic Space Learning (USSL) method, which not only explores the correlations of the semantic contents but also maps images, texts, labels, and multi-labels into a unified semantic space, facilitating the calculation of similarities between samples and between samples and semantic contents. To fully explore the correlations of the semantic contents, we construct a label-multi-label graph and learn the correlations of the semantic contents in a data-driven manner using our proposed Group Semantic Sharing Graph Convolutional Network. Furthermore, we propose an isomorphic InfoNCE loss to bridge the heterogeneity gap between the samples and semantic contents, along with an intra-modality InfoNCE loss and an inter-modality InfoNCE loss to maintain the semantic and structural consistencies of the learned modality-invariant common representations. Through comparative experiments on three representative cross-modal datasets, we have demonstrated the superiority of our proposed method.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"191 ","pages":"Article 107756"},"PeriodicalIF":6.0,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144471035","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}