Neurocomputing最新文献

{"title":"Dstsa-gcn: Advancing skeleton-based gesture recognition with semantic-aware spatio-temporal topology modeling","authors":"Hu Cui ,&nbsp;Renjing Huang ,&nbsp;Ruoyu Zhang ,&nbsp;Tessai Hayama","doi":"10.1016/j.neucom.2025.130066","DOIUrl":"10.1016/j.neucom.2025.130066","url":null,"abstract":"<div><div>Graph convolutional networks (GCNs) have emerged as a powerful tool for skeleton-based action and gesture recognition, thanks to their ability to model spatial and temporal dependencies in skeleton data. However, existing GCN-based methods face critical limitations: (1) they lack effective <em>spatio-temporal topology modeling</em> that captures dynamic variations in skeletal motion, and (2) they struggle to model <em>multiscale structural relationships</em> beyond local joint connectivity. To address these issues, we propose a novel framework called Dynamic Spatial-Temporal Semantic Awareness Graph Convolutional Network (DSTSA-GCN). DSTSA-GCN introduces three key modules: Group Channel-wise Graph Convolution (GC-GC), Group Temporal-wise Graph Convolution (GT-GC), and Multi-Scale Temporal Convolution (MS-TCN). GC-GC and GT-GC operate in parallel to independently model channel-specific and frame-specific correlations, enabling robust topology learning that accounts for temporal variations. Additionally, both modules employ a grouping strategy to adaptively capture multiscale structural relationships. Complementing this, MS-TCN enhances temporal modeling through group-wise temporal convolutions with diverse receptive fields. Extensive experiments demonstrate that DSTSA-GCN significantly improves the topology modeling capabilities of GCNs, achieving state-of-the-art performance on benchmark datasets for gesture and action recognition, including SHREC’17 Track, DHG-14/28, NTU-RGB+D, NTU-RGB+D-120 and NW-ULCA. The code will be publicly available <span><span>https://hucui2022.github.io/dstsa_gcn/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130066"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715512","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":"CS4TE: A Novel Coded Self-Attention and Semantic Synergy Network for Triple Extraction","authors":"Huiyong Lv ,&nbsp;Yurong Qian ,&nbsp;Jiaying Chen ,&nbsp;Shuxiang Hou ,&nbsp;Hongyong Leng ,&nbsp;Mengnan Ma","doi":"10.1016/j.neucom.2025.130034","DOIUrl":"10.1016/j.neucom.2025.130034","url":null,"abstract":"<div><div>The joint entity relation extraction approach holds great potential for extracting triples from unstructured text. However, in current research, two prevalent shortcomings significantly impact the efficacy of triple extraction task. Firstly, since entities constitute only a small proportion of sentences and token embedding contain a substantial amount of irrelevant information, these factors present significant challenges to the performance of classification models. Secondly, the typical process of predicting triples begins with identifying entities and then predicting triples solely based on the obtained entity representation, this process often overlooks the contextual semantic information associated with the entities. In this work, we propose CS4TE: A Novel Coded Self-Attention and Semantic Synergy Network for Triple Extraction. Specifically, we propose a novel Coded Self-Attention Mechanism designed to refine text representation by effectively masking irrelevant information and enhancing entity representation. Additionally, we propose a Semantic Synergy Network, which innovatively integrates semantic information with token pairs to predict triples, addressing the limitations of previous research that often overlooked semantic information. Finally, our model outperforms state-of-the-art baseline models on two public datasets in the joint entity-relation extraction task, and extensive experiments have been conducted to demonstrate the effectiveness of our method from multiple perspectives.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"636 ","pages":"Article 130034"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143706023","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":"DQT-CALF: Content adaptive neural network based In-Loop filter in VVC using dual query transformer","authors":"Yunfeng Liu,&nbsp;Cheolkon Jung","doi":"10.1016/j.neucom.2025.130064","DOIUrl":"10.1016/j.neucom.2025.130064","url":null,"abstract":"<div><div>As auxiliary inputs for the neural network-based in-loop filter (NNLF) in the versatile video coding (VVC), the prediction frame, partition map and quantization parameter (QP) map are effectively used for enhancing the reconstruction frame, i.e., main input. The prediction frame and partition map play a crucial role in reducing compression artifacts, while the QP map distinguishes between inputs with different QPs. However, direct concatenation of the auxiliary inputs with the reconstruction frame may not fully leverage their advantages potentially causing conflicting effects. In this paper, we propose a content adaptive NNLF in VVC using dual query transformer (DQT), named DQT-CALF. We adopt DQT based on a dual query mechanism to effectively fuse low-frequency global features and high-frequency local features, thereby learning rich feature representations. The dual query mechanism enhances comprehensiveness and representation ability of the model, reduces the risk of information loss, and is suitable for fusion of data multiple types. DQT-CALF mainly includes three parts of feature extraction (head), feature enhancement (body), and reconstruction (tail). In feature extraction, we use a parallel network structure for the main and auxiliary inputs and assign feature maps to different weights according to the richness of the input information. For the auxiliary inputs, we do not directly concatenate them with the reconstruction frame, but generate a spatial attention map based on them to enhance the reconstruction frame. In feature enhancement, we design a multi-type feature fusion module that divides the input tensor into the low-frequency and high-frequency features from the frequency viewpoint and the local and global features from the spatial viewpoint. The two groups of features are processed separately and are mutually transformed by the high-frequency local feature generation and low-frequency global feature generation, respectively. In reconstruction, we reconstruct the frame using a 3x3 convolution layer and a pixel-shuffle layer with a long skip connection. Experimental results demonstrate that DQT-CALF achieves average BD rate gains of {8.79% (Y), 22.09% (U), 22.99% (V)} and {9.68% (Y), 22.27% (U), 22.52% (V)} over the VTM-11.0_NNVC-3.0 anchor under all intra (AI) and random access (RA) configurations, respectively.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130064"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739375","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":"SAC: Collaborative learning of structure and content features for Android malware detection framework","authors":"Jin Yang ,&nbsp;Huijia Liang ,&nbsp;Hang Ren ,&nbsp;Dongqing Jia ,&nbsp;Xin Wang","doi":"10.1016/j.neucom.2025.130053","DOIUrl":"10.1016/j.neucom.2025.130053","url":null,"abstract":"<div><div>With the rapid development of Internet of Things (IoT) technology, Android devices have increasingly become primary targets for malware attacks. Although significant research has been conducted in the field of malware detection, existing methods still face challenges when dealing with complex samples. In particular, a more comprehensive analysis is required in the domain of feature extraction.</div><div>To enhance the accuracy of malware detection, we propose the SAC framework. This method utilizes Dalvik Executable (DEX) files as the data source and achieves deep integration of multi-view features by collaboratively modeling image and graph data types. Specifically, to accurately capture the local features of malware and improve the identification of critical behavioral patterns, we designed a task-oriented convolutional neural network (CNN) named IFNeXt, which integrates visualization analysis with an inverted bottleneck structure. Furthermore, we introduced a dual-channel graph convolutional network (GCN) that models the hierarchical structure of bytecode as a directed graph, capturing the co-occurrence relationships and semantic similarities between method calls. This approach enables a deeper exploration of the global structural features of malware.</div><div>The SAC framework fully leverages the complementary advantages of image and graph data structures, providing a more comprehensive characterization of malware features from both content and structural perspectives. Experimental results demonstrate that our method achieves a detection accuracy of 99.43% on multiple real-world public datasets, significantly outperforming existing state-of-the-art detection techniques. This indicates the potential and innovation of our approach in enhancing the security of the Android platform.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130053"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715507","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":"Efficient image and video processing via symmetric inertial proximal ADMM with RPCA model","authors":"Zhi-Bin Zhu ,&nbsp;Ying Liu ,&nbsp;Jia-Qi Huang ,&nbsp;Yue-Hong Ding","doi":"10.1016/j.neucom.2025.130054","DOIUrl":"10.1016/j.neucom.2025.130054","url":null,"abstract":"<div><div>This study proposes a novel symmetric inertial approximation alternating direction method of multiplier (SIPADMM) to address the significant challenges in solving separable two-block nonconvex nonsmooth optimization problems with linear constraints. The primary contribution of this work lies in the development of a unique algorithmic framework that successfully integrates the strengths of proximal and inertial methods while introducing a symmetric treatment of the Lagrange multiplier. A distinctive feature of our approach is the innovative utilization of numerical information from the previous three iteration steps in the proximal method, which significantly enhances the algorithm’s performance in handling nonconvex nonsmooth optimization problems. Furthermore, for the robust principal component analysis model, leveraging the properties of the hyperbolic tangent function, we have introduced a fresh nonconvex approximation function for the rank function, leading to the proposal of a new variant of the nonconvex model. Under carefully established assumptions, we obtain the convergence of the proposed algorithm through the construction of a specially designed auxiliary function <span><math><mi>H</mi></math></span>. Our theoretical analysis demonstrates that if the auxiliary function <span><math><mi>H</mi></math></span> satisfies the Kurdyka–Łojasiewicz inequality, every bounded sequence generated by the algorithm converges to a critical point of the minimization problem. In order to enhance the credibility of our approach and model in real-world scenarios, the proposed algorithm is utilized across various practical problems, especially in dealing with shadow and highlight removal in face images, and front–background separation tasks in videos. Experimental results consistently demonstrate the effectiveness of our algorithm in handling nonconvex approximation models based on rank functions.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130054"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739373","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":"Single-morphing attack detection using few-shot learning and triplet-loss","authors":"Juan E. Tapia ,&nbsp;Daniel Schulz ,&nbsp;Christoph Busch","doi":"10.1016/j.neucom.2025.130033","DOIUrl":"10.1016/j.neucom.2025.130033","url":null,"abstract":"<div><div>Face morphing attack detection is challenging and presents a concrete and severe threat to face verification systems. A reliable detection mechanism for such attacks, tested with a robust cross-dataset protocol and unknown morphing tools, is still a research challenge. This paper proposes a framework based on the Few-Shot-Learning approach that shares image information based on the Siamese network using triplet-semi-hard-loss to tackle the morphing attack detection and boost the learning classification process. This network compares a bona fide or potentially morphed image with triplets of morphing face images. Our results show that this new network clusters the morphed images and assigns them to the right classes to obtain a lower equal error rate in a cross-dataset scenario. Few-shot learning helps to boost the learning process by sharing only small image numbers from an unknown dataset. Experimental results using cross-datasets trained with FRGCv2 and tested with FERET datasets reduced the BPCER₁₀ from 43% to 4.91% using ResNet50. For the AMSL open-access dataset is reduced for MobileNetV2 from BPCER₁₀ of 31.50% to 2.02%. For the SDD open-access synthetic dataset, the BPCER₁₀ is reduced for MobileNetV2 from 21.37% to 1.96%.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"636 ","pages":"Article 130033"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143706024","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Adversarial erasure network based on multi-instance learning for weakly supervised video anomaly detection","authors":"Xin Song ,&nbsp;Penghui Liu ,&nbsp;Suyuan Li ,&nbsp;Siyang Xu ,&nbsp;Ke Wang","doi":"10.1016/j.neucom.2025.130030","DOIUrl":"10.1016/j.neucom.2025.130030","url":null,"abstract":"<div><div>Weakly supervised video anomaly detection (WSVAD) aims to precisely locate temporal windows of abnormal events in untrimmed videos using only video-level labels. By accurately locating anomalies, WSVAD has great application potential in the security domain and contributes to the progress of smart city development. However, the lack of frame-level annotations during training makes it highly challenging to infer the status of each frame. Multiple-Instance Learning (MIL) is the dominant method in WSVAD. Due to the limitation of video-level annotations, most MIL-based methods detect obvious abnormal segments to represent the overall anomaly level of the video while overlooking weak abnormal segments. To focus on the discrimination of weak anomalies, we propose a novel WSVAD framework named Adversarial Erasure Network (AE-Net). AE-Net consists of two key components: (1) a dual-branch architecture that highlights weak anomalies by erasing the most obvious abnormal features and combining the erased features with the original ones. (2) a novel triplet loss function that improves weak anomaly representation by separating abnormal and normal features in the erased feature space. Through the above design, AE-Net can reduce false negatives in real-world anomaly detection. Extensive experiments on three WSVAD benchmarks demonstrate that our method outperforms most existing state-of-the-art methods. Specifically, AE-Net achieves an AUC of 88.40% on the UCF-Crime dataset and 98.27% on the ShanghaiTech dataset, which demonstrates that AE-Net can effectively distinguish between normal and abnormal events. Moreover, AE-Net achieves an AP of 85.13% on the XD-Violence dataset, which highlights that AE-Net can accurately detect abnormal events.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"636 ","pages":"Article 130030"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143686897","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":"AMSFormer: A transformer with adaptive multi-scale partitioning and multi-level spectral filtering for time-series forecasting","authors":"Honghao Liu ,&nbsp;Yining Diao ,&nbsp;Ke Sun ,&nbsp;Zhaolin Wan ,&nbsp;Zhiyang Li","doi":"10.1016/j.neucom.2025.130067","DOIUrl":"10.1016/j.neucom.2025.130067","url":null,"abstract":"<div><div>Time series forecasting is essential in numerous real-world scenarios, yet it remains a challenging task due to complex temporal dependencies and the coexistence of multi-scale, periodic, and non-periodic patterns. In this paper, we present an innovative approach called the Adaptive Multi-Scale Transformer (AMSFormer) to tackle these challenges. AMSFormer integrates both frequency-domain and time-domain information to model local and global features collaboratively. By leveraging Fourier transforms, AMSFormer adaptively partitions time series data into patches that align with the data’s intrinsic patterns, enabling dynamic and efficient processing. A convolutional attention mechanism is employed to extract fine-grained local features at multiple scales while maintaining low computational overhead. For global feature extraction, AMSFormer utilizes a hierarchical frequency-domain filter to isolate key periodic components and suppress noise, enhancing the stability and accuracy of global pattern modeling. Extensive experiments on several real-world benchmark datasets demonstrate that AMSFormer consistently outperforms state-of-the-art models, highlighting its robust generalization ability and wide applicability across various forecasting tasks.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130067"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739384","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":"Unifying the syntax and semantics for math word problem solving","authors":"Xingyu Tao ,&nbsp;Yi Zhang ,&nbsp;Zhiwen Xie ,&nbsp;Zhuo Zhao ,&nbsp;Guangyou Zhou ,&nbsp;Yongchun Lu","doi":"10.1016/j.neucom.2025.130042","DOIUrl":"10.1016/j.neucom.2025.130042","url":null,"abstract":"<div><div>Math word problem solving is a complex task for natural language processing systems, requiring both comprehension of problem descriptions and deduction of accurate solutions. Existing studies have shown that graph-based approaches can achieve competitive results by applying multilayer graph neural networks to syntactic structure graphs. However, challenges such as incorrect parsing of syntactic dependency trees and insensitivity to numerical information may lead to misinterpretations in the representation. In this paper, we introduce a novel synthetic graph, the <strong>N</strong>umber-<strong>C</strong>entered <strong>S</strong>ynthetic <strong>S</strong>emantic <strong>G</strong>raph (NC-SSG), to address these challenges by reorganizing the dependency tree layout around numerical elements. We propose a double-channel graph transformer to enhance the connections between numbers and their contextual elements, thereby improving the understanding of problem descriptions. Additionally, we present a question-driven tree decoder to generate more accurate solutions, aiming to overcome shallow heuristics. Our approach mitigates the impact of parsing errors in syntactic dependency trees, yielding more precise representations and solutions. Experimental evaluations on two benchmark datasets demonstrate that our solver outperforms previous methods and achieves competitive performance compared to large language models.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"636 ","pages":"Article 130042"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143706078","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":"Class-view graph knowledge distillation: A new idea for learning MLPs on graphs","authors":"Yingjie Tian ,&nbsp;Shaokai Xu ,&nbsp;Muyang Li","doi":"10.1016/j.neucom.2025.130035","DOIUrl":"10.1016/j.neucom.2025.130035","url":null,"abstract":"<div><div>Graph Neural Networks (GNNs), while effective for processing non-Euclidean structured data, suffer from computationally intensive neighbor fetching, which hinders their deployment in low-latency applications. Cross-architecture graph knowledge distillation (KD), which trains high-performance Multi-layer Perceptrons (MLPs) to emulate teacher GNNs, offers a promising solution. However, existing GNN-MLP distillation methods rely on the sample-view KD paradigm, where student models directly mimic the teacher’s parameter space. Given the fundamentally different architectures and parameter spaces of GNNs and MLPs, this direct mimicry approach limits effective knowledge transfer. Inspired by the inherent properties of MLPs, we propose a novel class-view KD paradigm for GNN-MLP distillation. Unlike sample-view KD, our method guides student MLPs to generate more discriminative parameter configurations within their own parameter space while preserving the similarity of prediction distributions with the teacher, rather than directly imitating the teacher’s parameter configurations. Extensive experiments on public benchmark datasets demonstrate that class-view KD outperforms sample-view KD across various evaluation metrics and can be seamlessly integrated into existing GNN-MLP distillation methods to improve performance without additional computational cost. The code is available at <span><span>https://github.com/xsk160/Class-View-Graph-KD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130035"},"PeriodicalIF":5.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143725089","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}