Neurocomputing最新文献

{"title":"Monocular thermal SLAM with neural radiance fields for 3D scene reconstruction","authors":"Yuzhen Wu ,&nbsp;Lingxue Wang ,&nbsp;Lian Zhang ,&nbsp;Mingkun Chen ,&nbsp;Wenqu Zhao ,&nbsp;Dezhi Zheng ,&nbsp;Yi Cai","doi":"10.1016/j.neucom.2024.129041","DOIUrl":"10.1016/j.neucom.2024.129041","url":null,"abstract":"<div><div>Visual simultaneous localization and mapping (SLAM) faces significant challenges in environments with variable lighting and smoke. Excelling in such visually degraded settings, thermal imaging captures scene radiance effectively. To address the limitations of traditional thermal SLAM in 3D scene reconstruction, we propose ThermalSLAM-NeRF, a novel integration of thermal SLAM with neural radiance fields (NeRF). This method significantly enhances the quality of high dynamic range thermal images by improving their signal-to-noise ratio, contrast, and detail. It also employs online photometric calibration to ensure grayscale consistency between consecutive frames. We utilize a sparse direct method for pose estimation, selecting keyframes based on photometric error and tracking quality. The NeRF map is reconstructed using a multi-view keyframe sequence. Our evaluations on datasets containing over 15,000 thermal images show that ThermalSLAM-NeRF achieves an average improvement of 59.30% in trajectory accuracy over existing state-of-the-art SLAM methods. This approach uniquely tracks all sequences and constructs comprehensive NeRF maps, enabling robust and precise pose estimation without the need for extensive pre-training.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129041"},"PeriodicalIF":5.5,"publicationDate":"2024-11-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759559","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 user behavior-aware multi-task learning model for enhanced short video recommendation","authors":"Yuewei Wu ,&nbsp;Ruiling Fu ,&nbsp;Tongtong Xing ,&nbsp;Zhenyu Yu ,&nbsp;Fulian Yin","doi":"10.1016/j.neucom.2024.129076","DOIUrl":"10.1016/j.neucom.2024.129076","url":null,"abstract":"<div><div>In the rapidly evolving landscape of digital media consumption, accurately predicting user preferences and behaviors is critical for the effectiveness of recommendation systems, especially for short video content. Traditional recommendation methods often ignore the association between multiple user behavior types and struggle with dynamically adapting to user behavior changes, leading to suboptimal personalization and user engagement. To address these issues, this paper introduces a user behavior-aware multi-task learning model for enhanced short video recommendation (UBA-SVR) by leveraging insights into dynamic user interactions. In our approach, we construct a user behavior-aware transformer to comprehensively capture users’ dynamic interests and generate the fusion feature representation. Subsequently, we introduce a hierarchical knowledge extraction framework to process features in multi-stage and adopt a task-aware attention mechanism within the tower network structure to facilitate effective information sharing and differentiation among tasks. Furthermore, we employ a dynamic joint loss optimization strategy to adaptively adjust the loss weights for different tasks to promote collaborative enhancement. Extensive experiments on two real-world datasets demonstrate that the proposed method achieves significant improvements in multiple prediction tasks simultaneously.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129076"},"PeriodicalIF":5.5,"publicationDate":"2024-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759700","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":"Learning a more compact representation for low-rank tensor completion","authors":"Xi-Zhuo Li,&nbsp;Tai-Xiang Jiang,&nbsp;Liqiao Yang,&nbsp;Guisong Liu","doi":"10.1016/j.neucom.2024.129036","DOIUrl":"10.1016/j.neucom.2024.129036","url":null,"abstract":"<div><div>Transform-based tensor nuclear norm (TNN) methods have gained considerable attention for their effectiveness in addressing tensor recovery challenges. The integration of deep neural networks as nonlinear transforms has been shown to significantly enhance their performance. Minimizing transform-based TNN is equivalent to minimizing the <span><math><msub><mrow><mi>ℓ</mi></mrow><mrow><mn>1</mn></mrow></msub></math></span> norm of singular values in the transformed domain, which can be interpreted as finding a sparse representation with respect to the bases supported by singular vectors. This work aims to advance deep transform-based TNN methods by identifying a more compact representation through learnable bases, ultimately improving recovery accuracy. We specifically employ convolutional kernels as these learnable bases, demonstrating their capability to generate more compact representation, i.e., sparser coefficients of real-world tensor data compared to singular vectors. Our proposed model consists of two key components: a transform component, implemented through fully connected neural networks (FCNs), and a convolutional component that replaces traditional singular matrices. Then, this model is optimized using the ADAM algorithm directly on the incomplete tensor in a zero-shot manner, meaning all learnable parameters within the FCNs and convolution kernels are inferred solely from the observed data. Experimental results indicate that our method, with this straightforward yet effective modification, outperforms state-of-the-art approaches on video and multispectral image recovery tasks.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129036"},"PeriodicalIF":5.5,"publicationDate":"2024-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759557","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":"Global Span Semantic Dependency Awareness and Filtering Network for nested named entity recognition","authors":"Yunlei Sun,&nbsp;Xiaoyang Wang,&nbsp;Haosheng Wu,&nbsp;Miao Hu","doi":"10.1016/j.neucom.2024.129035","DOIUrl":"10.1016/j.neucom.2024.129035","url":null,"abstract":"<div><div>Span-based methods for nested named entity recognition (NER) are effective in handling the complexities of nested entities with hierarchical structures. However, these methods often overlook valid semantic dependencies among global spans, resulting in a partial loss of semantic information. To address this issue, we propose the Global Span Semantic Dependency Awareness and Filtering Network (GSSDAF). Our model begins with BERT for initial sentence encoding. Following this, a span semantic representation matrix is generated using a multi-head biaffine attention mechanism. We introduce the Global Span Dependency Awareness (GSDA) module to capture valid semantic dependencies among all spans, and the Local Span Dependency Enhancement (LSDE) module to selectively enhance key local dependencies. The enhanced span semantic representation matrix is then decoded to classify the spans. We evaluated our model on seven public datasets. Experimental results demonstrate that our model effectively handles nested NER, achieving higher F1 scores compared to baselines. Ablation experiments confirm the effectiveness of each module. Further analysis indicates that our model can learn valid semantic dependencies between global spans, significantly improving the accuracy of nested entity recognition. Our code is available at <span><span>https://github.com/Shaun-Wong/GSSDAF</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129035"},"PeriodicalIF":5.5,"publicationDate":"2024-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759561","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 novel multi-morphological representation approach for multi-source EEG signals","authors":"Yunyuan Gao ,&nbsp;Yici Liu ,&nbsp;Ming Meng ,&nbsp;Feng Fang ,&nbsp;Michael Houston ,&nbsp;Yingchun Zhang","doi":"10.1016/j.neucom.2024.129010","DOIUrl":"10.1016/j.neucom.2024.129010","url":null,"abstract":"<div><div>Advances in artificial intelligence have significantly enhanced intelligent assistance and rehabilitation medicine by leveraging electroencephalogram (EEG) signal recognition. Nevertheless, eliminating cross-subject variability remains a significant challenge in expending the application of EEG signal recognition to the broader society. The transfer learning strategy has been utilized to address this issue; however, multi-source domains are often treated as a single entity in transfer learning, leading to underutilization of the information from multiple sources. Furthermore, many EEG signal transfer approaches overlook the low-dimensional structural information and multivariate statistical features inherent in EEG signals, leading to inadequate interpretability and suboptimal performance. Thus, in this study, a novel multi-morphological representation approach (MMRA) was proposed for multi-source EEG signal recognition to address these issues. MMRA utilized multi-manifold mapping to extract the common invariant representation shared between the multi-source domains and target domain. It took into account the low-dimensional structure and multivariate statistical features of EEG signals to enhance the acquisition of high-quality common invariant representations. Subsequently, the multi-source domains were decomposed to extract one-to-one features. The Maximum Mean Discrepancy (MMD) loss was further applied to guide the model in obtaining high-quality private invariant representations. The performance of the proposed MMRA method was evaluated using three publicly available motor imagery datasets and a driving fatigue dataset. Experimental results demonstrated that our proposed MMRA method outperformed other state-of-the-art methods in scenarios involving multiple subjects. In conclusion, the MMRA method developed in this study can serve as a novel tool offering enhanced performance to analyze EEG signals across various subjects.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129010"},"PeriodicalIF":5.5,"publicationDate":"2024-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759708","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":"An HVS-derived network for assessing the quality of camouflaged targets with feature fusion","authors":"Qiyang Sun,&nbsp;Xia Wang,&nbsp;Changda Yan,&nbsp;Xin Zhang,&nbsp;Shiwei Xu","doi":"10.1016/j.neucom.2024.129016","DOIUrl":"10.1016/j.neucom.2024.129016","url":null,"abstract":"<div><div>High-value assets on the battlefield typically require adequate camouflage to evade detection and annihilation by enemy scouts. Consequently, artificial camouflage technology is extensively acknowledged and utilized as a crucial defensive tactic in the military sphere. The quality of camouflage performance was assessed by military observers through the human visual system (HVS). This method involved locating the camouflaged objects and rating the camouflaged degree against the background. Current camouflage assessment methods typically involved the manual extraction and aggregation of objective features throughout an image. These approaches fall short in constructing a correlation mapping between objective features and subjective perceptions of camouflaged objects, culminating in imprecise assessments and discrepancies. To address these issues, this paper presents the first three-stage full-reference learning framework for locating camouflaged objects, extracting camouflage features, and assessing camouflage quality. Given the lack of datasets specifically designed for evaluating camouflage quality, we have contributed a datasets focused on human-camouflaged targets. The experimental results show that the three-stage framework is remarkably accurate in assessing the camouflage quality, leading to an explainable network. The camouflaged people quality assessment(CPQA) dataset is available at <span><span>http://github.com/samsunq/CPQA_Datasets.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129016"},"PeriodicalIF":5.5,"publicationDate":"2024-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759560","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":"Soft prompt-tuning for unsupervised domain adaptation via self-supervision","authors":"Yi Zhu ,&nbsp;Shuqin Wang ,&nbsp;Yun Li ,&nbsp;Yunhao Yuan ,&nbsp;Jipeng Qiang","doi":"10.1016/j.neucom.2024.129008","DOIUrl":"10.1016/j.neucom.2024.129008","url":null,"abstract":"<div><div>Unsupervised domain adaptation methods aim to facilitate learning tasks in unlabeled target domains using labeled information from related source domains. Recently, prompt-tuning has emerged as a powerful instrument to incorporate templates that reformulate input examples into equivalent cloze-style phrases. However, there are still two great challenges for domain adaptation: (1) Existing prompt-tuning methods only rely on the general knowledge distributed in upstream pre-trained language models to alleviate the domain discrepancy. How to incorporate specific features in the source and target domains into prompt-tuning model is still divergent and under-explored; (2) In the prompt-tuning, either the crafted template methods are time-consuming and labor-intensive, or automatic prompt generation methods cannot achieve satisfied performance. To address these issues, in this paper, we propose an innovative Soft Prompt-tuning method for Unsupervised Domain Adaptation via Self-Supervision, which combines two novel ideas: Firstly, instead of only stimulating knowledge distributed in the pre-trained model, we further employ hierarchically clustered optimization strategies in a self-supervised manner to retrieve knowledge for the verbalizer construction in prompt-tuning. Secondly, we construct prompts with the special designed verbalizer that facilitate the transfer of learning representations across domains, which can consider both the automatic template generation and cross-domain classification performance. Extensive experimental results demonstrate that our method even outperforms SOTA baselines that utilize external open knowledge with much less computational time.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129008"},"PeriodicalIF":5.5,"publicationDate":"2024-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759552","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":"RTA: A reinforcement learning-based temporal knowledge graph question answering model","authors":"Yu Zhu ,&nbsp;Tinghuai Ma ,&nbsp;Shengjie Sun ,&nbsp;Huan Rong ,&nbsp;Yexin Bian ,&nbsp;Kai Huang","doi":"10.1016/j.neucom.2024.128994","DOIUrl":"10.1016/j.neucom.2024.128994","url":null,"abstract":"<div><div>Temporal Knowledge Graph Question Answering (TKGQA) is crucial research, focusing on finding an entity or a timestamp to answer temporal questions in the corresponding temporal knowledge graph. Currently, the main challenge in the temporal KGQA task is answering complex temporal questions, often necessitating complex multi-hop temporal reasoning in the TKG. In this paper, we propose a method for the TKGQA task called Reinforcement learning Temporal knowledge graph question <strong>A</strong>nswering (<strong>RTA</strong>). First, in the question understanding stage, our model extracts context information to select topic entities of the given question, which can effectively deal with scenarios involving multiple entities in complex temporal questions. Furthermore, reasoning complexity escalates significantly with complex temporal questions, as varying timestamps alter the relations between entities. Therefore, we introduce reinforcement learning into the reasoning process. In the policy network, a dynamic path-matching module is specifically included to aggregate the features of relational paths to effectively capture the dynamic changes of the relations between entities on the reasoning paths. At the same time, the weights are calculated to obtain the degree of attention of each candidate action. Then the score of each candidate action is obtained through a weighted summation mechanism which helps the agent learn the optimal path reasoning policy for effective exploration. Finally, we evaluate our method on the CRONQUESTIONS dataset and validate its superiority over all baseline methods. Specifically, our approach proves effective in handling complex temporal questions.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 128994"},"PeriodicalIF":5.5,"publicationDate":"2024-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759556","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":"Distributed continuous-time algorithm for nonsmooth aggregative optimization over weight-unbalanced digraphs","authors":"Zheng Zhang ,&nbsp;Guang-Hong Yang","doi":"10.1016/j.neucom.2024.129022","DOIUrl":"10.1016/j.neucom.2024.129022","url":null,"abstract":"<div><div>This paper studies the problem of distributed continuous-time aggregative optimization with set constraints under a weight-unbalanced digraph, where the nonsmooth objective function of each agent relies both on its own decision and on the aggregation of all agents’ decisions. To eliminate the impact of unbalanced digraphs, a consensus-based estimator that tracks the aggregation information is designed through a gradient rescaling technique. Considering that cost functions are nondifferentiable in many scenarios, such as electric power management that takes price caps into account, a novel distributed continuous-time optimization algorithm via generalized gradient is presented in a two-time scale. Moreover, the convergence of the algorithm is established through nonsmooth analysis and singular perturbation theory. Compared to the existing results, which depend on undirected graphs, the proposed strategy is applicable to general digraphs, which may be weight-unbalanced. Further, the assumption on the differentiability of objective functions is relaxed. Finally, two numerical examples are provided to verify the findings.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129022"},"PeriodicalIF":5.5,"publicationDate":"2024-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142757617","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":"An adaptation of hybrid binary optimization algorithms for medical image feature selection in neural network for classification of breast cancer","authors":"Olaide N. Oyelade ,&nbsp;Enesi Femi Aminu ,&nbsp;Hui Wang ,&nbsp;Karen Rafferty","doi":"10.1016/j.neucom.2024.129018","DOIUrl":"10.1016/j.neucom.2024.129018","url":null,"abstract":"<div><div>The performance of neural network is largely dependent on their capability to extract very discriminant features supporting the characterization of abnormalities in the medical image. Several benchmark architectures have been proposed and the use of transfer learning has further made these architectures return good performances. Study has shown that the use of optimization algorithms for selection of relevant features has improved classifiers. However continuous optimization algorithms have mostly been used though it allows variables to take value within a range of values. The advantage of binary optimization algorithms is that it allows variables to be assigned only two states, and this have been sparsely applied to medical image feature optimization. This study therefore proposes hybrid binary optimization algorithms to efficiently identify optimal features subset in medical image feature sets. The binary dwarf mongoose optimizer (BDMO) and the particle swarm optimizer (PSO) were hybridized with the binary Ebola optimization search algorithm (BEOSA) on new nested transfer functions. Medical images passed through convolutional neural networks (CNN) returns extracted features into a continuous space which are piped through these new hybrid binary optimizers. Features in continuous space a mapped into binary space for optimization, and then mapped back into the continuous space for classification. Experimentation was conducted on medical image samples using the Curated Breast Imaging Subset of Digital Database for Screening Mammography (DDSM+CBIS). Results obtained from the evaluation of the hybrid binary optimization methods showed that they yielded outstanding classification accuracy, fitness, and cost function values of 0.965, 0.021 and 0.943. To investigate the statistical significance of the hybrid binary methods, the analysis of variance (ANOVA) test was conducted based on the two-factor analysis on the classification accuracy, fitness, and cost metrics. Furthermore, results returned from application of the binary hybrid methods medical image analysis showed classification accuracy of 0.8286, precision of 0.97, recall of 0.83, and F1-score of 0.99, AUC of 0.8291. Findings from the study showed that contrary to the popular approach of using continuous metaheuristic algorithms for feature selection problem, the binary metaheuristic algorithms are well suitable for handling the challenge. Complete source code can be accessed from: <span><span>https://github.com/NathanielOy/hybridBinaryAlgorithm4FeatureSelection</span><svg><path></path></svg></span></div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"617 ","pages":"Article 129018"},"PeriodicalIF":5.5,"publicationDate":"2024-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759555","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}