Neurocomputing最新文献

{"title":"CM-SQL: A cross-model consistency framework for text-to-SQL","authors":"Xiang Li ,&nbsp;Jinguo You ,&nbsp;Heng Li ,&nbsp;Jun Peng ,&nbsp;Xi Chen ,&nbsp;Ziheng Guo","doi":"10.1016/j.neucom.2025.131708","DOIUrl":"10.1016/j.neucom.2025.131708","url":null,"abstract":"<div><div>In recent years, large language models (LLMs) have been widely applied to the task of Text-to-SQL. Currently, most LLM-based Text-to-SQL methods primarily adopt the following approaches to improve the accuracy of generated SQL: (1) schema linking; and (2) leveraging the model’s self-consistency to check, modify, and select the generated SQL. However, due to issues such as hallucinations in LLMs, the database schema generated during the schema linking phase may contain errors or omissions. On the other hand, LLMs often exhibit overconfidence when evaluating the correctness of their outputs. To address these issues, we propose a cross-model consistency SQL generation framework (CM-SQL), which generates SQL outputs from different perspectives by feeding two database schemas into two LLMs. The framework combines the stability of fine-tuned models with the powerful reasoning capabilities of LLMs to evaluate the generated SQL. Additionally, we propose a local modification strategy to correct erroneous SQL. Finally, the outputs of the evaluation module and the LLM are used to select candidate SQLs, yielding the final SQL. We evaluated the proposed framework on the BIRD dev dataset using GPT-4o-mini and DeepSeek-V2.5, achieving an execution accuracy of 65.65 %. On the test set of the Spider dataset, the execution accuracy reached 87.6%, significantly outperforming most methods based on the same LLMs. Furthermore, our performance is comparable to many approaches that rely on more expensive models, such as GPT-4.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131708"},"PeriodicalIF":6.5,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271204","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":"UDA-DDA: Unsupervised domain adaptation with dynamic distribution alignment network for emotion recognition using EEG signals","authors":"Jiahao Tang ,&nbsp;Youjun Li ,&nbsp;Chun-Wang Su ,&nbsp;Xiangting Fan ,&nbsp;Yangxuan Zheng ,&nbsp;Haoyu Wang ,&nbsp;Hadia Naeem ,&nbsp;Peng Fang ,&nbsp;Jue Wang ,&nbsp;Nan Yao ,&nbsp;Xueping Li ,&nbsp;Zi-Gang Huang","doi":"10.1016/j.neucom.2025.131715","DOIUrl":"10.1016/j.neucom.2025.131715","url":null,"abstract":"<div><div>In this paper, we address the challenge of individual variability in affective brain-computer interfaces (aBCI), which employ electroencephalogram (EEG) signals to monitor and recognize human emotional states, thereby facilitating the advancement of emotion-aware technologies. The variability in EEG data across individuals poses a significant barrier to the development of effective and widely applicable aBCI models. To mitigate this issue, we propose a novel transfer learning framework called Unsupervised Domain Adaptation (UDA) with Dynamic Distribution Alignment (UDA-DDA). This approach aligns the marginal and conditional probability distributions of source and target domains by employing maximum mean discrepancy (MMD) and conditional maximum mean discrepancy (CMMD). Firstly, we introduce a dynamic distribution alignment mechanism to adjust differences throughout training and enhance adaptation. Additionally, a pseudo-label confidence filtering module is integrated into the unsupervised process to refine pseudo-label generation and optimize the estimation of conditional distributions. In order to demonstrate the effectiveness and robustness of the proposed UDA-DDA method, extensive experiments are conducted on EEG benchmark databases (SEED, SEED-IV and DEAP). Evaluations of the algorithm’s performance in comparison with other UDA with dynamic distribution alignment network methods indicate the proposed method achieves state-of-the-art results in emotion recognition across various scenarios, including cross-subject and cross-session conditions. This advancement significantly enhances the accuracy and generalization of emotion recognition, potentially fostering the development of personalized aBCI applications. The source code is accessible at <span><span>https://github.com/XuanSuTrum/UDA-DDA</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131715"},"PeriodicalIF":6.5,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271116","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":"FaceDisentGAN: Disentangled facial editing with targeted semantic alignment","authors":"Meng Xu ,&nbsp;Prince Hamandawana ,&nbsp;Xiaohan Ma ,&nbsp;Zekang Chen ,&nbsp;Rize Jin ,&nbsp;Tae-Sun Chung","doi":"10.1016/j.neucom.2025.131706","DOIUrl":"10.1016/j.neucom.2025.131706","url":null,"abstract":"<div><div>Facial attribute editing in generative adversarial networks (GANs) involves two essential objectives: (1) accurately modifying the desired facial attribute, and (2) avoiding the unintended modification of irrelevant facial attributes. To address these challenges, we propose FaceDisentGAN, a novel generative framework for disentangled facial attribute manipulation. Specifically, we introduce: (1) a disentanglement module that decomposes feature maps into orthogonal spatial components (vertical and horizontal) to isolate target-related and unrelated semantics; (2) a two-stage training strategy that first learns general facial representations and then refines them to balance generic feature learning with fine-grained detail preservation; and (3) two novel evaluation metrics—Overall Preservation Score (OPS) and Perfect Match Rate (PMR)—which measure, respectively, the average preservation of non-target attributes and the proportion of perfectly disentangled results. This combination provides both soft and strict assessments of disentanglement quality. Extensive experiments demonstrate that FaceDisentGAN achieves accurate target attribute editing while effectively minimizing feature entanglement, outperforming several existing methods in both visual fidelity and semantic control.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131706"},"PeriodicalIF":6.5,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271115","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":"Classroom activity recognition using hybrid 3D-CNNs and visualization of action features with Grad-CAM","authors":"Rajamanickam Yuvaraj ,&nbsp;Jack S Fogarty ,&nbsp;Ratnavel Rajalakshmi ,&nbsp;Ritika Sarkar","doi":"10.1016/j.neucom.2025.131694","DOIUrl":"10.1016/j.neucom.2025.131694","url":null,"abstract":"<div><div>In the era of advanced computer vision technology, it is possible to use automatic methods to detect and classify student and teacher activities in classroom environments, providing novel approaches to study or evaluate the quality of teaching or learning. However, to date, there has been little research developing and testing these methods to work towards an optimal activity recognition system. This paper proposes an automated framework using a 3D-convolutional neural network (CNN) to recognize classroom activities, including teacher and student behaviors, from classroom videos. The 3D-CNN captured spatiotemporal features from the video data. Then, an extreme learning machine (ELM) classifier was trained over the 3D-CNN features to recognize different activities in the classroom. Multi-layer perceptron (MLP) and support vector machine (SVM) classifiers were also examined in comparison to ELM. Gradient-weighted class activation mapping (Grad-CAM) was employed to provide visual explanations of what information the highest performing model learned from videos to classify classroom activities. To evaluate each model, classifications were carried out on the EduNet dataset, containing annotated classroom activities featuring students and teachers. Classroom videos from the internet were also utilized to further evaluate the performance of the proposed frameworks. The proposed 3D-CNN+ELM model achieved a maximum average recognition accuracy of 88.17 % on EduNet, as estimated by 5-fold cross-validation, which is 5.87 % higher than the standard baseline I3D-ResNet-50 model proposed by the EduNet authors. The model also achieved an accuracy of 80.00 % when applied to an independent dataset of videos sourced from the internet, indicating reasonable reliability and generalizability. The Grad-CAM outcomes indicate that the model focuses on valid features to determine its recognition; however, in some cases, the recognition can still be incorrect. With its high level of performance, the proposed automated framework may assist in providing information on a range of classroom actions, which may offer preliminary insights to support the evaluation of classroom teaching and learning in real-world educational environments.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"657 ","pages":"Article 131694"},"PeriodicalIF":6.5,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145269558","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":"Unsupervised temporal action segmentation with sample discrimination training and alignment-based boundary refinement","authors":"Feng Huang ,&nbsp;Xiao-Diao Chen ,&nbsp;Hongyu Chen ,&nbsp;Haichuan Song","doi":"10.1016/j.neucom.2025.131636","DOIUrl":"10.1016/j.neucom.2025.131636","url":null,"abstract":"<div><div>Unsupervised temporal action segmentation (UTAS) addresses the task of partitioning untrimmed videos into coherent action segments without manual annotations. While boundary-detection-based approaches have demonstrated superior performance, they exhibit two critical limitations. First, these methods often uniformly treat all frames during training, resulting in over-segmentation and suboptimal performance. Second, they primarily rely on intra-video features while neglecting potentially valuable inter-video correlations within the dataset. To address these challenges, we present a comprehensive UTAS framework with three key innovations: (1) A discriminative training mechanism that differentiates between boundary/non-boundary frames in the temporal domain and motion/background pixels in the spatial domain, employing weighted training strategies alongside multiple temporal-scale modeling. (2) A self-validation mechanism for cross-verifying predictions across different input sequences. (3) A boundary refinement approach based on video alignment, which constructs reference video sets according to feature distributions and establishes inter-video correspondences to improve boundary localization. Extensive evaluations on three benchmark datasets, <em>i.e.</em>, the Breakfast, the 50Salads, and the YouTube Instructions, demonstrate that our approach achieves state-of-the-art performance, with quantitative results showing significant improvements over existing methods.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131636"},"PeriodicalIF":6.5,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145236193","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A fake news detection framework integrating multi-domain and multimodal features","authors":"Longqin Guo ,&nbsp;Zeqian Chen ,&nbsp;Xiaoyang Liu","doi":"10.1016/j.neucom.2025.131711","DOIUrl":"10.1016/j.neucom.2025.131711","url":null,"abstract":"<div><div>With the widespread dissemination of video-based fake news on social media, identifying the authenticity of information in complex contexts has become increasingly challenging. News from different domains often differs significantly in vocabulary, expression styles, and modality distributions, leading to semantic ambiguity and increasing the difficulty of cross-news modeling. To address these challenges, this paper proposes a Multimodal Multi-Domain Fake News Detection framework (MMMD), which integrates textual, audio, and visual modalities. A domain gating mechanism is introduced to model domain-specific contextual structures, thereby enhancing the discriminative power of weak modalities (such as audio) and improving inter-modal coordination. Experiments conducted on multiple benchmark datasets show that MMMD outperforms mainstream multimodal methods in terms of accuracy, F1-score, and other metrics. Notably, on the FakeSV dataset, MMMD achieves a 6.87 % improvement in accuracy over the representative method SV-FEND. Furthermore, to address the high cost of domain annotation, a K-Means-based pseudo-label generation strategy is adopted. Comparative experiments across different numbers of clusters indicate that setting 10 yields performance close to that of human annotations, validating the method’s feasibility in low-supervision scenarios. Without relying on external user relationships, MMMD leverages domain-aware semantic structures and modality interaction mechanisms, providing an efficient and scalable solution for multimodal fake news detection in complex environments.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131711"},"PeriodicalIF":6.5,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271201","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":"Auto-weighted graph tensor and rank-constrained bipartite graph fusion for multi-view clustering","authors":"Jie Zhang ,&nbsp;Xiaoqian Zhang ,&nbsp;Jinghao Li ,&nbsp;Yongyi Yang ,&nbsp;Zhenwen Ren ,&nbsp;Rong Tang ,&nbsp;Dong Wang","doi":"10.1016/j.neucom.2025.131575","DOIUrl":"10.1016/j.neucom.2025.131575","url":null,"abstract":"<div><div>Tensor multi-view clustering generally outperforms non-tensor counterparts, as the tensor structure can effectively capture the higher-order correlations of data. Although the t-SVD-based tensor nuclear norm has shown remarkable performance, it treats the similar information across all views equally, overlooking the higher-order similarities between similar graphs. To address this issue, we propose a Pearson Correlation Coefficient-based <span><math><mtext>A</mtext></math></span>uto-weighted <span><math><mtext>G</mtext></math></span>raph <span><math><mtext>T</mtext></math></span>ensor and <span><math><mtext>R</mtext></math></span>ank-constrained <span><math><mtext>B</mtext></math></span>ipartite <span><math><mtext>G</mtext></math></span>raph <span><math><mtext>F</mtext></math></span>usion (AGTRBGF) approach for multi-view clustering. Specifically, the P-AGT learning method breaks free from the constraints of predefined weights, automatically assigning optimal weight values for each similarity graph by leveraging the higher-order similarities among the similar graphs of different views. Additionally, the Laplace rank is utilized to constrain the adaptive graph fusion, endowing learned consensus graph with strong diagonal structure and enhancing the model’s robustness. Experiments conducted on distinct datasets validate the effectiveness and superior clustering performance of AGTRBGF.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131575"},"PeriodicalIF":6.5,"publicationDate":"2025-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145236192","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":"IntSTR: An integrated spatio-temporal relation transformer for video object detection","authors":"Wentao Zheng ,&nbsp;Hong Zheng ,&nbsp;Yuquan Sun ,&nbsp;Ying Jing","doi":"10.1016/j.neucom.2025.131704","DOIUrl":"10.1016/j.neucom.2025.131704","url":null,"abstract":"<div><div>In recent years, Transformer-based video object detection (VOD) methods have achieved remarkable progress by replacing the hand-crafted components traditionally used in CNN-based detectors. However, many existing approaches rely on staged spatio-temporal modeling strategies, which increase model complexity and restrict early interaction between spatial and temporal information. To overcome these limitations, we propose IntSTR, a novel framework for unified spatio-temporal modeling. At its core, the spatio-temporal relation encoder (STRE) integrates spatio-temporal feature processing within a single encoder through cascaded attention modules. To strengthen temporal consistency, the temporal query relation (TQR) module explicitly captures geometric relations between object queries across adjacent frames with minimal computational overhead. In addition, the Temporal Feature Memory (TFM) maintains a dynamic memory bank that caches temporal contexts, enabling effective feature aggregation and efficient online processing. Extensive experiments on the ImageNet VID dataset validate the effectiveness of our approach. IntSTR achieves an excellent trade-off between accuracy and efficiency, reaching a competitive 87.2 % <span><math><msub><mtext>mAP</mtext><mrow><mn>50</mn></mrow></msub></math></span> with the ResNet-101 backbone while maintaining real-time performance at 33.4 FPS.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131704"},"PeriodicalIF":6.5,"publicationDate":"2025-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145236305","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":"Dynamic event-based asymptotic tracking and vibration control for constrained flexible manipulator systems with intermittent faults","authors":"Shan-Lin Liu ,&nbsp;Meina Zhai ,&nbsp;Rui Wang ,&nbsp;Yufeng Tian","doi":"10.1016/j.neucom.2025.131638","DOIUrl":"10.1016/j.neucom.2025.131638","url":null,"abstract":"<div><div>The angle constraint and vibration suppression issues of flexible manipulator (FM) systems subjected to intermittent faults are addressed in this article. Firstly, integral barrier Lyapunov functions (BLFs) that can directly constrain the angular position are introduced, eliminating the feasibility conditions of traditional BLFs. Secondly, a triggering mechanism with dynamic variables is provided to reduce the transmission of redundant information, thereby saving communication resources. To mitigate the impact of intermittent faults and handle system, the boundary estimation method and the neural networks (NNs) technology considering the influence of approximation error are adopted, which reduces the conservatism of the developed control algorithm. Through Lyapunov stability theory and Hamiltonian principle, a dynamic event-based fault-tolerant controller is designed, suppressing the offset of the FM while ensuring that the angular position asymptotically tracks the ideal position without exceeding the given constraint boundary. Eventually, the simulation results demonstrate the rationality of the developed control scheme.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131638"},"PeriodicalIF":6.5,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271119","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":"3D Gaussian splatting technologies and extensions: A review","authors":"Fengkai Luan ,&nbsp;Siliang Sun ,&nbsp;Hu Zhang,&nbsp;Yong Yin,&nbsp;Ke Wang,&nbsp;Jiaxing Yang","doi":"10.1016/j.neucom.2025.131629","DOIUrl":"10.1016/j.neucom.2025.131629","url":null,"abstract":"<div><div>In recent years, 3D Gaussian Splatting (3DGS) has achieved remarkable progress in the field of novel view synthesis. Unlike implicit neural radiance field (NeRF) methods that primarily focus on positional and viewpoint transformations, 3DGS leverages millions of Gaussian ellipsoids for scene reconstruction and employs parallel differentiable rasterization to substantially improve rendering efficiency. Given the rapid advancement and promising prospects of this technique, this survey presents a systematic overview of recent developments in 3DGS. We provide a detailed exposition of the fundamental theory underlying 3DGS, along with relevant benchmark datasets. Uniquely, this work organizes existing optimization strategies according to the stages of the Gaussian splatting pipeline. In addition, we review various downstream applications based on 3DGS and discuss prospective research directions. This survey aims to serve as a valuable reference for researchers across all stages of engagement and to foster further advancements in 3DGS.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"658 ","pages":"Article 131629"},"PeriodicalIF":6.5,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271497","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}