Neural Networks最新文献

{"title":"Attentional dual-stream interactive perception network for efficient infrared small aerial target detection","authors":"Lihao Zhou,&nbsp;Huawei Wang","doi":"10.1016/j.neunet.2026.108563","DOIUrl":"10.1016/j.neunet.2026.108563","url":null,"abstract":"<div><div>Drones and other flying objects can be regarded as small targets from a long-distance perspective. Considering the occlusion and interference caused by the external environment, the infrared detection methods are adopted to help identify and manage small aerial targets. However, remote infrared imaging often leads to small target feature detail loss. And the general methods have low detection efficiency, difficult to deeply extract target features. To better address the above problems, we propose an attentional dual-stream interactive perception network (ADIPNet) in this paper. Based on dual-stream U-Net, ADIPNet mainly combines the multi-patch series-parallel attention module (MSPA), edge anchoring module with regret (EAR), context scene perception module (CSP) and dual-stream interaction fusion module (DSIF). MSPA manually constructs the weight of patch regions at multiple scales and then performs the nested self-attention so as to fully mine global target information. EAR unites two types of global features using local mapping and matrix product, which helps accurately capture small target edge. CSP exchanges context information multiple times and conducts mutual complementation of semantic scenarios to enhances the perception of small target features. Finally, DSIF conducts cross attention for high-level encoded features on double U-Nets, further improving the network’s understanding of complex scenario information. The proposed ADIPNet alleviates the insufficient feature extraction of infrared small targets. Compared with other state-of-the-art methods, mIoU respectively reaches 80.52% and 72.54% on two large infrared datasets. It achieves more accurate detection of small aerial targets with low operating cost, possessing potential application prospect in various infrared surveillance systems.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108563"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145981540","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":"CocoAdapter: Efficient end-to-end temporal action detection via self-constrained multi-cognitive adapters","authors":"Lizao Zhang ,&nbsp;Qiuhong Tian ,&nbsp;Junxiao Ning ,&nbsp;Yihan Yuan ,&nbsp;Ziyu Yang ,&nbsp;Yang Yu","doi":"10.1016/j.neunet.2025.108531","DOIUrl":"10.1016/j.neunet.2025.108531","url":null,"abstract":"<div><div>End-to-end training in temporal action detection (TAD) has shown great potential for performance improvement by jointly optimizing the video encoder and action classification head. However, memory bottlenecks have limited the performance of end-to-end TAD. To alleviate the memory overhead during training, this paper explores the application of adapters in TAD and proposes a specialized TAD-oriented self-<strong>co</strong>nstraint multi-<strong>co</strong>gnitive <strong>adapter</strong> (<strong>CocoAdapter</strong>). Based on CocoAdapter, we construct a novel baseline, CocoTad. Our proposed CocoAdapter utilizes self-constraint projection layers to adjust multiple cognitive convolutional groups based on network depth, enabling a fine-tuning process tailored to the TAD task. As a result, the network only needs to update the parameters in CocoAdapter to achieve end-to-end training, significantly reducing memory consumption during training. We evaluate our model on four representative datasets. Experimental results demonstrate that our proposed CocoTad surpasses previous state-of-the-art methods in terms of mAP.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108531"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146004471","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":"Graph-enhanced dual low-rank correlation embedding for spatio-temporal EEG fusion in depression recognition","authors":"Lu Zhang ,&nbsp;Jisheng Dang ,&nbsp;Shu Zhang ,&nbsp;Wencheng Gan ,&nbsp;Juan Wang ,&nbsp;Bin Hu ,&nbsp;Gang Feng ,&nbsp;Hong Peng","doi":"10.1016/j.neunet.2026.108609","DOIUrl":"10.1016/j.neunet.2026.108609","url":null,"abstract":"<div><div>Electroencephalography (EEG) signals contain rich spatiotemporal information reflecting brain activity, making them valuable for analyzing cognitive, emotional, and neurological disorders. However, effectively integrating these two types of information to capture both discriminative and complementary features remains a significant challenge. To address this, we propose a Graph-Enhanced Dual Low-Rank Correlation Embedding (GEDLCE) method, which integrates spatiotemporal EEG features to improve depression recognition. GEDLCE enforces low-rank constraints at both feature and sample levels, enabling extraction of shared latent factors across multiple feature sets. To preserve the intrinsic geometric structure of the data, GEDLCE employs two graph Laplacian terms to model local relationships in the sample space. Furthermore, GEDLCE introduces a graph embedding term that utilizes label information to enhance its discriminative capability. In addition, GEDLCE incorporates an enhanced correlation analysis to exploit inter-view correlations while reducing intra-view redundancy. Finally, GEDLCE jointly optimizes low-rank representations, correlation constraints, and graph embedding within a unified framework. Experiments on EEG datasets show that GEDLCE effectively captures critical information, achieves superior performance in depression recognition, and shows promise for early diagnosis and disease monitoring.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108609"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146114755","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":"Multi-modal feature alignment networks for multi-label image classification","authors":"Wenlan Kuang ,&nbsp;Zhixin Li","doi":"10.1016/j.neunet.2026.108629","DOIUrl":"10.1016/j.neunet.2026.108629","url":null,"abstract":"<div><div>Multi-label image classification is a classification task that assigns labels to multiple objects in an input image. Recent research ideas mainly focus on solving the semantic consistency of visual features and label features. However, since images contain complex scene content, the features captured by visual feature extraction networks based on grid or sequence representation may introduce redundant information or lack continuity when identifying irregular objects. In order to fully mine the visual information of complex objects in images and enhance the inter-modal interaction of images and labels, we introduce a flexible graph structure to explore the internal information of objects and design a multi-modal feature alignment (MMFA) network for multi-label image classification. To enhance the context awareness and semantic association of different patch regions, we propose a semantic-augmented interaction module that combines two kinds of visual semantic information with label embeddings for interactive learning. Finally, we refine the dependence between local intrinsic information and overall semantics by redefining semantic queries through semantically enhanced visual spatial features and graph aggregation features. Experiments on three large-scale public datasets: Microsoft COCO, Pascal VOC 2007 and NUS-WIDE demonstrate the effectiveness of our proposed MMFA and achieve state-of-the-art performance.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108629"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039362","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":"Two-hidden-layer ReLU neural networks and finite elements","authors":"Pengzhan Jin","doi":"10.1016/j.neunet.2026.108559","DOIUrl":"10.1016/j.neunet.2026.108559","url":null,"abstract":"<div><div>We point out that (continuous or discontinuous) piecewise linear functions on a convex polytope mesh can be represented by two-hidden-layer ReLU neural networks in a weak sense. In addition, the numbers of neurons of the two hidden layers required to weakly represent are accurately given based on the numbers of polytopes and hyperplanes involved in this mesh. The results naturally hold for constant and linear finite element functions. Such weak representation establishes a bridge between two-hidden-layer ReLU neural networks and finite element functions, and leads to a perspective for analyzing approximation capability of ReLU neural networks in <em>L^p</em> norm via finite element functions. Moreover, we discuss the strict representation for tensor finite element functions via the recent tensor neural networks.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108559"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145981627","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":"Relation-aware pre-trained network with hierarchical aggregation mechanism for cold-start drug recommendation","authors":"Xiaobo Li ,&nbsp;Xiaodi Hou ,&nbsp;Shilong Wang ,&nbsp;Hongfei Lin ,&nbsp;Yijia Zhang","doi":"10.1016/j.neunet.2026.108618","DOIUrl":"10.1016/j.neunet.2026.108618","url":null,"abstract":"<div><div>Drug recommendation systems have garnered considerable interest in the healthcare, striving to offer precise and customized drug prescriptions that align with patients’ specific health needs. However, existing methods primarily focus on modeling temporal dependencies between visits for patients with multiple encounters, often neglecting the challenge of data sparsity in single-visit patients. To address above limitation, we propose a novel Relation-aware Pre-trained Network with hierarchical aggregation mechanism for drug recommendation (RPNet), which employs a pre-training and fine-tuning framework to enhance drug recommendation in cold-start scenario. Specifically, we introduce: 1) A code matching discrimination task during pre-training, designed to model the complex relationships between diagnosis and procedure entities. This task employs a mask-replace contrastive learning strategy, which pulls similar samples closer while pushing dissimilar ones apart, thereby capturing robust feature representations; 2) A hierarchical aggregation mechanism that enhances drug information integration by first selecting relevant visits based on rarity discrimination and then retrieving similar patients’ drug insights via similarity matching during fine-tuning. Extensive experiments on two real-world datasets demonstrate the superiority of the proposed RPNet, notably improving the F1 metric by 1.32% and 1.19%. The code of our model is available at <span><span>https://github.com/Lxb0102/RPNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108618"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146079254","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":"Self-supervised exceptional prototypical network for few-shot grading of gastric intestinal metaplasia","authors":"Xuanchi Chen ,&nbsp;Yonghui Xu ,&nbsp;Zhen Li ,&nbsp;Mingzhe Zhang ,&nbsp;Han Yu ,&nbsp;Lizhen Cui ,&nbsp;Xiangwei Zheng","doi":"10.1016/j.neunet.2026.108584","DOIUrl":"10.1016/j.neunet.2026.108584","url":null,"abstract":"<div><div>Automatic grading of Gastric Intestinal Metaplasia (GIM) is valuable in assisting the diagnosis of early gastric cancer. Recently, prototypical networks are served as a effective method for medical image processing in few-shot scenarios. However, existing prototypical networks suffer from the following two limitations when applied to GIM grading: 1) Variable camera angles of gastric endoscopes result in diverse sampling granularities of GIM lesions, leading to a multitude of multiscale features. Fully supervised encoders struggle to learn robust multiscale features due to limited labeled endoscopic images and privacy concerns. 2) Class prototypes based on sample means ignore the latent class information of exceptional cases, resulting in one-sided inferences of category prototypes and decision boundaries. To address these challenges, we propose a Self-supervised Exceptional Prototypical Network (Swin-EPN) for few-shot grading of GIM. Specifically, three tailored pretext tasks are designed to jointly pretrain a swin transformer, which is integrated as the model’s embedding layer to learning robust multiscale features. We propose an exceptional prototype mining module that identifies exceptional prototypes by defining a prototype score for each sample and updating potential exceptional prototypes in an exceptional prototype bank. These exceptional prototypes are served as supplementary information to class prototypes, and are leveraged to guide the delineation of class decision boundaries. We validated Swin-EPN on a private GIM dataset from a local grade-A tertiary hospital in both 1-shot and 5-shot scenarios, achieving accuracy improvements of 6.12% and 5.61% respectively compared to state-of-the-art (SOTA) models.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108584"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145981729","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":"Uncovering various neuronal responses in a fractional-order generalized HR system","authors":"Krishnendu Bera ,&nbsp;Chinmay Chakraborty ,&nbsp;Eva Kaslik ,&nbsp;Urszula Foryś ,&nbsp;Sanjeev K. Sharma ,&nbsp;Argha Mondal","doi":"10.1016/j.neunet.2026.108548","DOIUrl":"10.1016/j.neunet.2026.108548","url":null,"abstract":"<div><div>This study investigates neuronal electrical activities in a fractional-order generalized Hindmarsh-Rose (HR) system and explores an extended model incorporating an induced electric field. Stability and bifurcation analyses examine the impact of external electrical stimulation on neuronal dynamics. The results show how electric field parameters, including amplitude and frequency, modulate neuronal excitability and stability. The H-R model is a mathematical representation that captures diverse neuronal activities, and the introduction of fractional-order derivatives allows us to explore non-local dynamics in greater depth. We analyze the effects of fractional-order derivatives on the system’s behavior, including the generation of action potential dynamics. We discuss some biophysical aspects of the different firing patterns that we encounter. In addition, the study employs both analytical and numerical methods to investigate the stability of bursting and spiking patterns, using linear stability analysis to examine the transitions between stable and unstable states. Simulations reveal significant memory effects even with a slight decrease in fractional order. This underscores the versatility of fractional-order models in bridging mathematical theory with biologically plausible phenomena. The findings of this study demonstrate the potential of fractional-order systems in capturing the intricacies of neuronal responses, highlighting the need for further exploration of these phenomena in excitable biophysical systems.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108548"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145999305","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":"Efficient multi-agent communication via entity-aware causal network","authors":"Yifan Bo ,&nbsp;Bowen Huang ,&nbsp;Jinghan Feng ,&nbsp;Shuo Zhang ,&nbsp;Biao Leng","doi":"10.1016/j.neunet.2026.108538","DOIUrl":"10.1016/j.neunet.2026.108538","url":null,"abstract":"<div><div>Communication is considered as a crucial approach for solving complicated multi-agent reinforcement learning (MARL) cooperative tasks. However, existing approaches rely on predefined agent orders and identifiers to learn targeted communication. The predefined approaches ignore the prior knowledge that the selection of communication targets is solely related to agents’ states rather than their orders or identifiers, which leads to poor scalability and inefficient sampling. To address these limitations, we introduce the <strong>Entity-Aware Causal (EAC)</strong> framework, which tackles MARL communication from an entity-centric perspective. The core idea is to enhance communication efficiency through entity-aware communication target selection and causal inference belief mechanism, we make three main contributions. Firstly, we design an entity-aware hypernetwork that identifies communication targets based on individual state information and employs a masked-attention mechanism to enable scalable and sparse communication topology. Secondly, we propose a causal inference beliefs mechanism to strengthen the belief of the communication between entities and reduce redundant message exchanges. Finally, our algorithm outperforms baseline multi-agent cooperative reinforcement learning algorithms across SMAC, SMAC_v2, GRF, and MPE benchmarks. We further demonstrate the robustness of the algorithm across various network topologies and sparsity levels.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108538"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146020007","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 novel cross-domain fault diagnosis method for multi-condition industrial processes based on meta-domain adaptation with progressive meta-learning","authors":"Xin Qin ,&nbsp;Xuan Guo ,&nbsp;Jie Dong ,&nbsp;Kaixiang Peng","doi":"10.1016/j.neunet.2026.108561","DOIUrl":"10.1016/j.neunet.2026.108561","url":null,"abstract":"<div><div>Complex industrial processes are characterized by high dynamics, diverse operating conditions, and strong inter-system coupling, often leading to reduced production efficiency and product quality fluctuations. Employing advanced fault diagnosis technologies has become an effective approach to support high-quality and efficient execution of industrial processes. However, the increasing prevalence of customized manufacturing has introduced substantial variability in working conditions, under which traditional fault diagnosis methods struggle to perform effectively. Each working condition can be abstracted as a domain. Therefore, employing domain adaptation techniques to achieve multi-condition fault diagnosis is one of the key approaches to addressing the above challenge. Based on the above observation, a novel neural network-based cross-domain fault diagnosis method for multi-condition industrial processes via meta-domain adaptation with progressive meta-learning is proposed. First, an adversarial dual-scale neural network is designed to address the challenge of feature alignment across multiple source domains, comprising a one-dimensional convolutional neural network feature extractor and a multi-layer perceptrons domain discriminator. A progressive adversarial strength adjustment strategy is proposed to better extract domain-invariant yet discriminative shared features, thereby enhancing domain generalization. Second, to tackle practical issues such as imbalanced condition distributions, limited sample availability, and intra-source heterogeneity, a meta-learning mechanism is employed to reduce internal distributional discrepancies within source domains. Additionally, multi-kernel maximum mean discrepancy is employed to explicitly align source and target features, facilitating robust generalization under substantial domain shifts. Finally, the constructed cross-domain feature extractor and fault classifier are used to achieve fault diagnosis in industrial processes. The proposed method is evaluated on the benchmark Tennessee Eastman process and a real hot strip mill process, demonstrating its effectiveness and superiority.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"198 ","pages":"Article 108561"},"PeriodicalIF":6.3,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145967132","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}