Complex & Intelligent Systems最新文献

{"title":"Enhancing geometric modeling in convolutional neural networks: limit deformable convolution","authors":"Wei Wang, Yuanze Meng, Han Li, Guiyong Chang, Shun Li, Chenghong Zhang","doi":"10.1007/s40747-025-01799-8","DOIUrl":"https://doi.org/10.1007/s40747-025-01799-8","url":null,"abstract":"<p>Convolutional neural networks (CNNs) are constrained in their capacity to model geometric transformations due to their fixed geometric structure. To overcome this problem, researchers introduce deformable convolution, which allows the convolution kernel to be deformable on the feature map. However, deformable convolution may introduce irrelevant contextual information during the learning process and thus affect the model performance. DCNv2 introduces a modulation mechanism to control the diffusion of the sampling points to control the degree of contribution of offsets through weights, but we find that such problems still exist in practical use. Therefore, we propose a new limit deformable convolution to address this problem, which enhances the model ability by adding adaptive limiting units to constrain the offsets and adjusts the weight constraints on the offsets to enhance the image-focusing ability. In the subsequent work, we perform lightweight work on the limit deformable convolution and design three kinds of LDBottleneck to adapt to different scenarios. The limit deformable network, equipped with the optimal LDBottleneck, demonstrated an improvement in mAP75 of 1.4% compared to DCNv1 and 1.1% compared to DCNv2 on the VOC2012+2007 dataset. Furthermore, on the CoCo2017 dataset, different backbones equipped with our limit deformable module achieved satisfactory results. The source code for this work is publicly available at https://github.com/1977245719/LDCN.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"9 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143538782","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 comprehensive transplanting of black-box adversarial attacks from multi-class to multi-label models","authors":"Zhijian Chen, Qi Zhou, Yujiang Liu, Wenjian Luo","doi":"10.1007/s40747-025-01805-z","DOIUrl":"https://doi.org/10.1007/s40747-025-01805-z","url":null,"abstract":"<p>Adversarial examples generated by perturbing raw data with carefully designed, imperceptible noise have emerged as a primary security threat to artificial intelligence systems. In particular, black-box adversarial attack algorithms, which only rely on model input and output to generate adversarial examples, are easy to implement in real scenarios. However, previous research on black-box attacks has primarily focused on multi-class classification models, with relatively few studies on black-box attack algorithms for multi-label classification models. Multi-label classification models exhibit significant differences from multi-class classification models in terms of structure and output. The former can assign multiple labels to a single sample, with these labels often exhibiting correlations, while the latter classifies a sample as the class with the highest confidence. Therefore, existing multi-class attack algorithms cannot directly attack multi-label classification models. In this paper, we study the transplantation methods of multi-class black-box attack algorithms to multi-label classification models and propose the multi-label versions for eight classic black-box attack algorithms, which include three score-based attacks and five decision-based (label-only) attacks, for the first time. Experimental results indicate that the transplanted black-box attack algorithms demonstrate effective attack performance across various attack types, except for extreme attacks. Especially, most transplanted attack algorithms achieve more than 60% success rate on the ML-GCN model and more than 30% on the ML-LIW model under the hiding all attack type. However, the performance of these transplanted attack algorithms shows variation among different attack types due to the correlations between labels.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"17 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143538781","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 local search with chain search path strategy for real-world many-objective vehicle routing problem","authors":"Ying Zhou, Lingjing Kong, Hui Wang, Yiqiao Cai, Shaopeng Liu","doi":"10.1007/s40747-025-01825-9","DOIUrl":"https://doi.org/10.1007/s40747-025-01825-9","url":null,"abstract":"<p>This article focuses on a new application-oriented variant of vehicle routing problem. This problem comes from the daily distribution scenarios of a real-world logistics company. It is a large-scale (with customer sizes up to 2000), many-objective (with six objective functions) NP-hard problem with six constraints. Then, a local search with chain search path strategy (LS-CSP) is proposed to effectively solve the problem. It is a decomposition-based algorithm. First, the considered problem is decomposed into multiple single-objective subproblems. Then, local search is applied to solve these subproblems one by one. The advantage of the LS-CSP lies in a chain search path strategy, which is designed for determining the order of solving the subproblems. This strategy can help the algorithm find a high-quality solution set quickly. Finally, to assess the performance of the proposed LS-CSP, three instance sets containing 132 instances are provided, and four state-of-the-art decomposition-based approaches are adopted as the competitors. Experimental results show the effectiveness of the proposed algorithm for the considered problem.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"16 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143532583","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":"CycleGuardian: a framework for automatic respiratory sound classification based on improved deep clustering and contrastive learning","authors":"Yun Chu, Qiuhao Wang, Enze Zhou, Ling Fu, Qian Liu, Gang Zheng","doi":"10.1007/s40747-025-01800-4","DOIUrl":"https://doi.org/10.1007/s40747-025-01800-4","url":null,"abstract":"<p>Auscultation plays a pivotal role in early respiratory and pulmonary disease diagnosis. Despite the emergence of deep learning-based methods for automatic respiratory sound classification post-Covid-19, limited datasets impede performance enhancement. Distinguishing between normal and abnormal respiratory sounds poses challenges due to the coexistence of normal respiratory components and noise components in both types. Moreover, different abnormal respiratory sounds exhibit similar anomalous features, hindering their differentiation. Besides, existing state-of-the-art models suffer from excessive parameter size, impeding deployment on resource-constrained mobile platforms. To address these issues, we design a lightweight network CycleGuardian and propose a framework based on an improved deep clustering and contrastive learning. We first generate a hybrid spectrogram for feature diversity and grouping spectrograms to facilitate intermittent abnormal sound capture. Then, CycleGuardian integrates a deep clustering module with a similarity-constrained clustering component to improve the ability to capture abnormal features and a contrastive learning module with group mixing for enhanced abnormal feature discernment. Multi-objective optimization enhances overall performance during training. In experiments, we use the ICBHI2017 dataset, following the official split method and without any pre-trained weights, our method achieves Sp: 82.06<span>(%)</span>, Se: 44.47<span>(%)</span>, and Score: 63.26<span>(%)</span> with a network model size of 38 M. Compared to the current model, our method leads by nearly 7<span>(%)</span>, achieving the current best performances. Additionally, we deploy the network on Android devices, showcasing a comprehensive intelligent respiratory sound auscultation system.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"52 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143532574","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":"Small sample smart contract vulnerability detection method based on multi-layer feature fusion","authors":"Jinlin Fan, Yaqiong He, Huaiguang Wu","doi":"10.1007/s40747-025-01782-3","DOIUrl":"https://doi.org/10.1007/s40747-025-01782-3","url":null,"abstract":"<p>The identification of vulnerabilities in smart contracts is necessary for ensuring their security. As a pre-trained language model, BERT has been employed in the detection of smart contract vulnerabilities, exhibiting high accuracy in tasks. However, it has certain limitations. Existing methods solely depend on features extracted from the final layer, thereby disregarding the potential contribution of features from other layers. To address these issues, this paper proposes a novel method, which is named multi-layer feature fusion (MULF). Experiments investigate the impact of utilizing features from other layers on performance improvement. To the best of our knowledge, this is the first instance of multi-layer feature sequence fusion in the field of smart contract vulnerability detection. Furthermore, there is a special type of patched contract code that contains vulnerability features which need to be studied. Therefore, to overcome the challenges posed by limited smart contract vulnerability datasets and high false positive rates, we introduce a data augmentation technique that incorporates function feature screening with those special smart contracts into the training set. To date, this method has not been reported in the literature. The experimental results demonstrate that the MULF model significantly enhances the performance of smart contract vulnerability identification compared to other models. The MULF model achieved accuracies of 98.95% for reentrancy vulnerabilities, 96.27% for timestamp dependency vulnerabilities, and 87.40% for overflow vulnerabilities, which are significantly higher than those achieved by existing methods.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"5 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143532575","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":"Decentralized non-convex online optimization with adaptive momentum estimation and quantized communication","authors":"Yunshan Lv, Hailing Xiong, Fuqing Zhang, Shengying Dong, Xiangguang Dai","doi":"10.1007/s40747-025-01818-8","DOIUrl":"https://doi.org/10.1007/s40747-025-01818-8","url":null,"abstract":"<p>In this work, we consider the decentralized non-convex online optimization problem over an undirected network. To solve the problem over a communication-efficient manner, we propose a novel quantized decentralized adaptive momentum gradient descent algorithm based on the adaptive momentum estimation methods, where quantified information is exchanged between agents. The proposed algorithm not only can effectively reduce the data transmission volume but also contribute to improved convergence. Theoretical analysis proves that the proposed algorithm can achieve sublinear dynamic regret under appropriate step-size and quantization level, which matches the convergence of the decentralized online algorithm with exact-communication. Extensive simulations are given to demonstrate the efficacy of the algorithm.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"66 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143532577","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":"SDGANets: a semantically enhanced dual graph-aware network for affine and registration of remote sensing images","authors":"Xie Zhuli, Wan Gang, Liu Jia, Bu Dongdong","doi":"10.1007/s40747-025-01792-1","DOIUrl":"https://doi.org/10.1007/s40747-025-01792-1","url":null,"abstract":"<p>Remote sensing image pairs of different time phases have complex and changeable semantic contents, and traditional convolutional registration methods are challenging in modeling subtle local changes and global large-scale deformation differences in detail. This results in poor registration performance and poor feature representation. To address these problems, a semantically enhanced dual-graph perception framework is proposed. This framework aims to gradually achieve semantic alignment and precise registration of remote sensing image pairs of different time phases via coarse to fine stages. On the one hand, a newly designed large-selection kernel convolution attention module is used to learn affine transformation parameters. Attention to global semantics perceives the large pixel displacement deviation caused by large-scale deformation, and the association relationship is established between remote sensing image pairs of different time phases. At the same time, dual-graph perception modules are embedded in multiple subspace structures, and the subtle local changes of remote sensing image pairs are modeled through the dynamic aggregation ability of graph perception nodes to achieve coarse registration of remote sensing images. On the other hand, a U-shaped module guided by global attention with deformable convolution is used to refine the local spatial structural features and global contextual semantic information of the rough registration, establish dependencies between channels, and correct the pixel displacement deviation of remote sensing image pairs of different phases through position encoding. It is worth noting that the newly designed weighted loss function supervises the learning of each module and the entire network structure from the perspective of inverse consistency, promoting the network’s optimal performance. Finally, the experimental results on the AerialData and GFRS datasets show that the proposed framework has good registration performance, with mean absolute error (MAE) of 3.64 and 3.81, respectively.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"86 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143518617","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":"Multimodal multilevel attention for semi-supervised skeleton-based gesture recognition","authors":"Jinting Liu, Minggang Gan, Yuxuan He, Jia Guo, Kang Hu","doi":"10.1007/s40747-025-01807-x","DOIUrl":"https://doi.org/10.1007/s40747-025-01807-x","url":null,"abstract":"<p>Although skeleton-based gesture recognition using supervised learning has achieved promising results, the reliance on extensive annotated data poses significant costs. This paper addresses the challenge of semi-supervised skeleton-based gesture recognition, to effectively learn feature representations from labeled and unlabeled data. To resolve this problem, we propose a novel multimodal multilevel attention network designed for semi-supervised learning. This model utilizes the self-attention mechanism to polymerize multimodal and multilevel complementary semantic information of the hand skeleton, designing a multimodal multilevel contrastive loss to measure feature similarity. Specifically, our method explores the relationships between joint, bone, and motion to learn more discriminative feature representations. Considering the hierarchy of the hand skeleton, the skeleton data is divided into multilevel to capture complementary semantic information. Furthermore, the multimodal contrastive loss measures similarity among these multilevel representations. The proposed method demonstrates improved performance in semi-supervised skeleton-based gesture recognition tasks, as evidenced by experiments on the SHREC-17 and DHG 14/28 datasets.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"25 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143518621","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":"$$text {H}^2text {CAN}$$ : heterogeneous hypergraph attention network with counterfactual learning for multimodal sentiment analysis","authors":"Changqin Huang, Zhenheng Lin, Qionghao Huang, Xiaodi Huang, Fan Jiang, Jili Chen","doi":"10.1007/s40747-025-01806-y","DOIUrl":"https://doi.org/10.1007/s40747-025-01806-y","url":null,"abstract":"<p>Multimodal sentiment analysis (MSA) has garnered significant attention for its immense potential in human-computer interaction. While cross-modality attention mechanisms are widely used in MSA to capture inter-modality interactions, existing methods are limited to pairwise interactions between two modalities. Additionally, these methods can not utilize the causal relationship to guide attention learning, making them susceptible to bias information. To address these limitations, we introduce a novel method called Heterogeneous Hypergraph Attention Network with Counterfactual Learning <span>((text {H}^2text {CAN}).)</span> The method constructs a heterogeneous hypergraph based on sentiment expression characteristics and employs Heterogeneous Hypergraph Attention Networks (HHGAT) to capture interactions beyond pairwise constraints. Furthermore, it mitigates the effects of bias through a Counterfactual Intervention Task (CIT). Our model comprises two main branches: hypergraph fusion and counterfactual fusion. The former uses HHGAT to capture inter-modality interactions, while the latter constructs a counterfactual world using Gaussian distribution and additional weighting for the biased modality. The CIT leverages causal inference to maximize the prediction discrepancy between the two branches, guiding attention learning in the hypergraph fusion branch. We utilize unimodal labels to help the model adaptively identify the biased modality, thereby enhancing the handling of bias information. Experiments on three mainstream datasets demonstrate that <span>(text {H}^2text {CAN})</span> sets a new benchmark.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"33 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143518813","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 new representation in genetic programming with hybrid feature ranking criterion for high-dimensional feature selection","authors":"Jiayi Li, Fan Zhang, Jianbin Ma","doi":"10.1007/s40747-025-01784-1","DOIUrl":"https://doi.org/10.1007/s40747-025-01784-1","url":null,"abstract":"<p>Feature selection is a common method for improving classification performance. Selecting features for high-dimensional data is challenging due to the large search space. Traditional feature ranking methods that search for top-ranked features cannot remove redundant and irrelevant features and may also ignore interrelated features. Evolutionary computation (EC) techniques are widely used in feature selection due to their global search capability. However, EC can easily fall into local optima when dealing with feature selection for high-dimensional applications. The top-ranked features are more likely to construct effective feature subsets and help EC reduce the search space. This paper proposes a feature selection method based on Genetic Programming (GP) with hybrid feature ranking criterion called GPHC, which combines multiple feature ranking methods into the GP structure using a novel GP representation to search for effective feature subsets. Experiments on eight high-dimensional datasets show that GPHC achieves significantly better classification performance compared to five feature ranking methods. Further comparisons between GPHC and other evolutionary algorithms demonstrate that GPHC has advantages in terms of classification performance, the number of features, and convergence speed.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"96 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143518618","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}