Neural Networks最新文献

{"title":"A spiking network model of the cerebellum for predicting movements with diverse complex spikes","authors":"Tomohiro Mitsuhashi ,&nbsp;Yusuke Kuniyoshi ,&nbsp;Koji Ikezoe ,&nbsp;Kazuo Kitamura ,&nbsp;Tadashi Yamazaki","doi":"10.1016/j.neunet.2025.107962","DOIUrl":"10.1016/j.neunet.2025.107962","url":null,"abstract":"<div><div>Smooth and coordinated motor control is believed to be achieved through prediction by forward models in the cerebellum, which generate predicted movements from motor commands. These models are acquired via supervised learning, where instruction signals, originating from the inferior olive and represented as complex spikes (CSs) in Purkinje cells, guide learning. Previous studies show that CSs represent a wide variety of motor- and nonmotor-related activities, but how this diversity contributes to forward model acquisition remains unclear. We hypothesized that predicted movements are learned through the combination of various types of CSs. To test this, we developed a spiking network model of the cerebellum as a supervised learning machine, using instruction signals based on <span><math><msup><mtext>Ca</mtext><mtext>2+</mtext></msup></math></span> imaging data from a self-initiated lever-pull task in mice. While individual signals did not fully represent lever movements, the combination of Purkinje cell activities, trained by different instruction signals, allowed neurons in the cerebellar nucleus to represent lever trajectory. Additionally, the same set of instruction signals trained the model to generate different movement trajectories. We further confirmed that a mouse musculoskeletal model successfully reproduced lever-pulling movements. These findings suggest that forward models in the cerebellum are achieved through a combination of diverse CSs with different spatiotemporal profiles, providing an over-complete basis for movement prediction.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107962"},"PeriodicalIF":6.3,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144885539","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":"Dynamic graph transformation with multi-task learning for enhanced spatio-temporal traffic prediction","authors":"Nana Bu ,&nbsp;Zongtao Duan ,&nbsp;Wen Dang ,&nbsp;Jianxun Zhao","doi":"10.1016/j.neunet.2025.107963","DOIUrl":"10.1016/j.neunet.2025.107963","url":null,"abstract":"<div><div>Traffic prediction plays an essential role in intelligent transportation systems by supporting urban traffic management and public safety. A major challenge lies in addressing both the limitations of static assumptions and the inherent complexity they introduce when modeling dynamic and heterogeneous traffic systems. Traditional methods often simplify complex spatio-temporal data into a single-dimensional framework, potentially overlooking intricate node interactions and detailed network characteristics. This fundamental challenge manifests primarily in single-task approaches. When extended to multi-task learning scenarios, the complexity and limitations of this modeling challenge becomes more pronounced. To address these issues, this paper introduce a novel framework, Dynamic Graph Transformation with Multi-Task Learning (DGT-MTL) for spatio-temporal traffic prediction. DGT-MTL features a dynamic adjacency matrix generation module that balances static stability with dynamic flexibility. Additionally, it employs a multi-scale graph learning module to effectively capture fine-grained, latent features. An adaptive multi-task learning module is incorporated to uncover hidden correlations and dynamic relationships between road segments. Experiments conducted across six standard benchmarks demonstrate DGT-MTL’s superior performance compared to contemporary approaches, achieving over 15 % improvements in both ROC-AUC and F1 score metrics. Further experiments demonstrate its effectiveness and robustness in handling complex traffic prediction.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107963"},"PeriodicalIF":6.3,"publicationDate":"2025-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144860813","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":"Structure-preserving contrastive graph clustering with dual-channel label alignment","authors":"Guang-Yu Zhang ,&nbsp;Yan-Di Huang ,&nbsp;Dong Huang ,&nbsp;Chang-Dong Wang ,&nbsp;Yang Liu ,&nbsp;Enbo Huang","doi":"10.1016/j.neunet.2025.107954","DOIUrl":"10.1016/j.neunet.2025.107954","url":null,"abstract":"<div><div>The past few years have witnessed the rapid development of contrastive graph clustering (CGC). Although a series of achievements have been made, there still remain two challenging problems in the literature. First, previous works typically generate different views via some pre-defined graph augmentation strategies, but inappropriate augmentations may alter the latent semantics of the original data. Second, they often overlook the discriminative unsupervised information when constructing positive and negative sample pairs, resulting in compromised clustering performance. Third, some of them are restricted to only static neighborhood connections for contrastive learning, which neglect the dynamical structural relationship via robust neighboring graph learning. To cope with these issues, this paper proposes a Structure-preserving Contrastive Graph Clustering approach with Dual-channel Label Alignment (SCGC-DLA). In terms of the high-and-low frequency issues, the low-pass and hybrid graph filters are designed for generating two views of reliable augmentations, which can supply rich and complementary information to each other. Further, we construct a structure-preserving matrix, which is derived from the edge betweenness centrality (EBC) perspective design and allows us to efficiently capture the topological relationships among different embedding representations. Under the guidance of the non-dominated sorting theory, the clustering distribution information of dual-channel is used to construct high-confidence pseudo labels. Especially, the generated high-confidence pseudo labels are aligned with latent semantic labels. Finally, the overall network is guided by a self-supervised learning scheme and therefore the final clustering could be obtained. Substantial results on five benchmarks prove the robustness and effectiveness of our approach compared to several state-of-the-arts.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107954"},"PeriodicalIF":6.3,"publicationDate":"2025-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144889120","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-view parallel convolutional network for organ segmentation in mediastinal region on CT images","authors":"Yining Xie ,&nbsp;Wei Zhou ,&nbsp;Jiayi Ma ,&nbsp;Fengjiao Wang ,&nbsp;Jing Zhao","doi":"10.1016/j.neunet.2025.107972","DOIUrl":"10.1016/j.neunet.2025.107972","url":null,"abstract":"<div><div>In lung CT images, mediastinal organ segmentation is crucial for localizing different mediastinal regions. However, existing medical image segmentation methods exhibit significant limitations in modeling the diverse topological structures of organs, sensitivity to intra-class morphological variations, and inter-class feature differentiation. To address these limitations, we propose a novel multi-view parallel convolutional network (MVPCNet), built on an efficient U-shaped encoder-decoder framework. The shallow and deep information encoders are respectively composed of alternating multi-view parallel convolution module (MVPM) and the dual-path backbone structure (DPBS) at different scales. MVPM is designed as a parallel convolutional structure to enhance the model’s ability to capture complex structural features, enabling complementary extraction of morphological and detailed features. DPBS comprises the efficient dual-channel bottleneck structures (EDC-BS) and the region fusion small-kernel deformable attention mechanism (RF-SKDA). EDC-BS employs a branched convolutional architecture, effectively reducing computational complexity while ensuring accurate recognition of the same organ across varying morphologies. RF-SKDA captures the spatial structural information of different organs by combining regional and global average pooling, and further extracts organ-specific morphological features through the deformable convolutions. The decoder utilizes lightweight parameterization through depthwise separable convolutions and integrates multi-scale features during the decoding process. Experimental results demonstrate that MVPCNet achieves an average Dice Coefficient of 90.59 % and an mIoU of 82.80 % on mediastinal organ dataset. With a parameter size of only 8.21 MB, it outperforms advanced medical segmentation algorithms and classical lightweight semantic segmentation models.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107972"},"PeriodicalIF":6.3,"publicationDate":"2025-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879971","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":"Enhancing signed graph neural networks through curriculum-based training","authors":"Zeyu Zhang ,&nbsp;Lu Li ,&nbsp;Xingyu Ji ,&nbsp;Kaiqi Zhao ,&nbsp;Xiaofeng Zhu ,&nbsp;Philip S. Yu ,&nbsp;Jiawei Li ,&nbsp;Maojun Wang","doi":"10.1016/j.neunet.2025.107975","DOIUrl":"10.1016/j.neunet.2025.107975","url":null,"abstract":"<div><div>Signed graphs are powerful models for representing complex relations with both positive and negative connections. Recently, Signed Graph Neural Networks (SGNNs) have emerged as potent tools for analyzing such graphs. To our knowledge, no prior research has been conducted on devising a training plan specifically for SGNNs. The prevailing training approach feeds samples (edges) to models in a random order, resulting in equal contributionsfrom each sample during the training process, but fails to account for varying learning difficulties based on the graph’s structure. We contend that SGNNs can benefit from a curriculum that progresses from easy to difficult, similar to human learning. The main challenge is evaluating the difficulty of edges in a signed graph. Weaddress this by theoretically analyzing the difficulty of SGNNs in learning adequate representations for edges in unbalanced cycles and propose a lightweight difficulty measurer. This forms the basis for our innovative <u>C</u>urriculum representation learning framework for <u>S</u>igned <u>G</u>raphs, referred to as <strong>CSG</strong>. The process involves using the measurer to assign difficulty scores to training samples, adjusting their order using a scheduler and training the SGNN model accordingly. We empirically our approach on six real-world signed graph datasets. Our method demonstrates remarkable results, enhancing the accuracy of popular SGNN models by up to 23.7 % and showing a reduction of 8.4 % in standard deviation, enhancing model stability. Our implementation is available in PyTorch (<span><span>https://github.com/Alex-Zeyu/CSG</span><svg><path></path></svg></span>).</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107975"},"PeriodicalIF":6.3,"publicationDate":"2025-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144864548","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":"Focusing on pedestrians like human for clothes changing person re-identification","authors":"Wenjie Pan ,&nbsp;Jianqing Zhu ,&nbsp;Xiaolin Cui ,&nbsp;Huanqiang Zeng ,&nbsp;Yibing Zhan","doi":"10.1016/j.neunet.2025.107960","DOIUrl":"10.1016/j.neunet.2025.107960","url":null,"abstract":"<div><div>Current approaches focus mainly on the design of networks to learn key identity features from local body components for clothes-changing person re-identification (CC-ReID). In this paper, we propose a humanoid focus-inspired image augmentation (HFIA) method, which is intuitive image processing rather than a sophisticated network architecture designed to enhance local nuances of pedestrian images. Based on pedestrian silhouettes, we roughly divide a pedestrian image into five body components, that is, head-shoulder, upper left torso, upper right torso, lower left torso, and lower right torso. The HFIA has two key designs to deal with these components: the central emphasis strategy (CES) and the component continuity processing (CCP). For each component, leveraging the natural tendency of human visual attention towards central regions, the CES constructs an enlargement grid, where the closer the center, the greater the enlargement. To maintain the continuity of assembly, the CCP performs an overall alignment of component centers, that is, all components share the same normalized vertical coordinate and the left and right torsos have mirrored horizontal coordinates. Furthermore, the CCP implements a smoothing post-processing to uniformly erase the discontinuity between the head-shoulder, upper left torso, and upper right torso. Experiments show the state-of-the-art performance of HFIA.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107960"},"PeriodicalIF":6.3,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144864549","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":"C3aptioner: Improving change captioning by leveraging momentum cross-view and cross-modality contrastive learning","authors":"Lin Deng ,&nbsp;Borui Kang ,&nbsp;Yuzhong Zhong ,&nbsp;Maoning Wang ,&nbsp;Jianwei Zhang","doi":"10.1016/j.neunet.2025.107957","DOIUrl":"10.1016/j.neunet.2025.107957","url":null,"abstract":"<div><div>The primary goal of change captioning is to identify subtle visual differences between two similar images and express them in natural language. Existing research has been significantly influenced by the task of vision change detection and has mainly concentrated on the identification and description of visual changes. However, we contend that an effective change captioner should go beyond mere detection and description of what has changed. Two additional aspects are crucial: 1) retaining significant and unique semantic elements that persist across both images, and 2) forging a robust link between visual cues and their concomitant descriptive linguistic elements. This paper addresses these challenges by presenting the C<span><math><msup><mrow></mrow><mn>3</mn></msup></math></span>aptioner, which seamlessly incorporates dual momentum contrastive learning objectives into change captioning. Our model architecture consists of intra-image and inter-image Transformer encoders for visual feature extraction, complemented by unimodal language and multimodal decoders. Specifically, we introduce a cross-view contrastive learning objective to capture essential invariant features by aligning cross-view representations with a momentum-updated queue of negative samples, addressing the challenge of viewpoint variations. Additionally, our cross-modality contrastive learning objective aligns and interacts visual and textual modalities using a separate momentum-maintained queue, resolving the modality gap that hampers existing methods. This dual contrastive approach enables C<span><math><msup><mrow></mrow><mn>3</mn></msup></math></span>aptioner to model both changed and unchanged elements while establishing strong vision-language correspondence, resulting in more contextually rich and human-like descriptions. Extensive experiments across five distinct datasets confirm that our approach achieves state-of-the-art performance, with particularly significant improvements in challenging scenarios involving extreme viewpoint changes. Source code is available at <span><span>https://github.com/DenglinGo/C-3aptioner</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107957"},"PeriodicalIF":6.3,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144889122","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":"CausalCOMRL: Context-based offline meta-reinforcement learning with causal representation","authors":"Zhengzhe Zhang ,&nbsp;Wenjia Meng ,&nbsp;Haoliang Sun ,&nbsp;Gang Pan","doi":"10.1016/j.neunet.2025.107955","DOIUrl":"10.1016/j.neunet.2025.107955","url":null,"abstract":"<div><div>Context-based offline meta-reinforcement learning (OMRL) methods have achieved appealing success by leveragingpre-collected offline datasets to develop task representations that guide policy learning. However, current context-based OMRL methods often introduce spurious correlations, where task components are incorrectly correlated due to confounders. These correlations can degrade policy performance when the confounders in the test taskdiffer from those in the training task. To address this problem, we propose CausalCOMRL, a context-based OMRL method that integrates causal representation learning. This approach uncovers causal relationships among the task components and incorporates the causal relationships into task representations, enhancing the generalizability of RL agents. We further improve the distinction of task representations from different tasks by using mutual information optimization and contrastive learning. Utilizing these causal task representations, we employSAC to optimize policies on meta-RL benchmarks. Experimental results show that CausalCOMRL achieves better performance than other methods on most benchmarks.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107955"},"PeriodicalIF":6.3,"publicationDate":"2025-08-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144864547","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":"Context selectivity with dynamic availability enables lifelong continual learning.","authors":"Martin L L R Barry, Wulfram Gerstner, Guillaume Bellec","doi":"10.1016/j.neunet.2025.107728","DOIUrl":"https://doi.org/10.1016/j.neunet.2025.107728","url":null,"abstract":"<p><p>\"You never forget how to ride a bike\", - but how is that possible? The brain is able to learn complex skills, stop the practice for years, learn other skills in between, and still retrieve the original knowledge when necessary. The mechanisms of this capability, referred to as lifelong learning (or continual learning, CL), are unknown. We suggest a bio-plausible meta-plasticity rule building on classical work in CL which we summarize in two principles: (i) neurons are context selective, and (ii) a local availability variable partially freezes the plasticity if the neuron was relevant for previous tasks. In a new neuro-centric formalization of these principles, we suggest that neuron selectivity and neuron-wide consolidation is a simple and viable meta-plasticity hypothesis to enable CL in the brain. In simulation, this simple model balances forgetting and consolidation leading to better transfer learning than contemporary CL algorithms on image recognition and natural language processing CL benchmarks.</p>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"107728"},"PeriodicalIF":6.3,"publicationDate":"2025-08-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144823063","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":"Domain-aware self-prompting for cross-domain sequential recommendations with natural language explanations","authors":"Tesfaye Fenta Boka ,&nbsp;Zhendong Niu ,&nbsp;Tekie Tsegay Tewolde ,&nbsp;Ramadhani Duma","doi":"10.1016/j.neunet.2025.107969","DOIUrl":"10.1016/j.neunet.2025.107969","url":null,"abstract":"<div><div>Cross-domain sequential recommendation faces persistent challenges in addressing domain shift, data sparsity, and the trade-off between performance, efficiency, and explainability. Existing methods often struggle with inefficient cross-domain adaptation or fail to generate coherent explanations that bridge user preferences across domains. To overcome these limitations, we propose <strong>Domain-Aware Self-Prompting (DASP)</strong>, a novel framework that integrates cross-domain recommendation with natural language explanation generation. DASP introduces three key innovations: (1) a domain-invariant self-prompt generator that captures shared user preferences via contrastive alignment across domains; (2) lightweight domain adapters with meta-learned initialization for parameter-efficient adaptation to target domains; and (3) a cross-domain explanation generator that grounds recommendations in semantically aligned multi-domain prompts using large language models. Extensive experiments on Amazon Movie-Book and Food-Kitchen datasets demonstrate DASP’s superiority, achieving <strong>10.7 %</strong> and <strong>10.5 %</strong> improvements in HR@10 and NDCG@10 over state-of-the-art baselines on the Movie-Book dataset, while reducing training time by <strong>54 %</strong> compared to full large language models fine-tuning approaches. Qualitative and quantitative analyses validate DASP’s ability to generate interpretable explanations that link cross-domain preferences, offering a scalable and trustworthy solution for cross-domain sequential recommendation. Our work bridges critical gaps in efficiency, adaptability, and explainability for real-world multi-domain recommendation systems.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107969"},"PeriodicalIF":6.3,"publicationDate":"2025-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144852320","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}