Neurocomputing最新文献

{"title":"A context-aware temporal knowledge graph completion method based on logical paths","authors":"Yuchao Zhang ,&nbsp;Xiangjie Kong ,&nbsp;Kailun Ye ,&nbsp;Shangfei Zheng ,&nbsp;Qihong Pan ,&nbsp;Guojiang Shen ,&nbsp;Jianxin Li","doi":"10.1016/j.neucom.2026.133049","DOIUrl":"10.1016/j.neucom.2026.133049","url":null,"abstract":"<div><div>Temporal Knowledge Graph Completion (TKGC) plays a critical role in dynamic knowledge expansion by predicting entity relationships at specific time points based on existing facts. This enhances the performance of downstream applications. Existing TKGC methods mainly generate dynamic entity representations by capturing the structural characteristics of facts and the sequential properties of data. Some approaches also attempt to extract temporal logical rules from historical query data. However, these methods do not fully account for the contextual relevance of queries, which limits the accuracy of entity predictions. To address this gap, we propose a <strong>C</strong>ontext-<strong>A</strong>ware TKGC method based on <strong>L</strong>ogical <strong>P</strong>aths (LPCA). This method integrates embedding representations and logical rules to create an interpretable TKGC completion framework, improving both prediction transparency and accuracy. Our approach mines query-specific logical paths, incorporating a time-aware sampling mechanism that prioritizes temporally recent facts to enhance temporal relevance and adaptability. Additionally, we design a self-attention mechanism to capture enriched contextual features of these temporal logical paths, modeling dependencies among the head entity, tail entity, query relation, and associated path elements to strengthen semantic and structural relevance. Experimental results show that the proposed method improves MRR scores by 3.59%, 0.99%, and 2.67% on the ICEWS14, ICEWS18, and ICEWS05-15 datasets, respectively.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133049"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386786","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":"Explainable AI: Context-aware layer-wise integrated gradients for explaining transformer models","authors":"Melkamu Abay Mersha,&nbsp;Jugal Kalita","doi":"10.1016/j.neucom.2026.133050","DOIUrl":"10.1016/j.neucom.2026.133050","url":null,"abstract":"<div><div>Transformer models achieve state-of-the-art performance across domains and tasks, yet their deeply layered representations make their predictions difficult to interpret. Existing explainability methods rely on final-layer attributions, capture either local token-level attributions or global attention patterns without unification, and lack context-awareness of inter-token dependencies and structural components. They also fail to capture how relevance evolves across layers and how structural components shape decision-making. To address these limitations, we proposed the <strong>Context-Aware Layer-wise Integrated Gradients (CA-LIG) Framework</strong>, a unified hierarchical attribution framework that computes layer-wise Integrated Gradients within each Transformer block and fuses these token-level attributions with class-specific attention gradients. This integration yields signed, context-sensitive attribution maps that capture supportive and opposing evidence while tracing the hierarchical flow of relevance through the Transformer layers. We evaluate the CA-LIG Framework across diverse tasks, domains, and Transformer model families, including sentiment analysis and long and multi-class document classification with BERT, hate speech detection in a low-resource language setting with XLM-R and AfroLM, and image classification with Masked Autoencoder vision Transformer model. Across all tasks and architectures, CA-LIG provides more faithful attributions, shows stronger sensitivity to contextual dependencies, and produces clearer, more semantically coherent visualizations than established explainability methods. These results indicate that CA-LIG provides a more comprehensive, context-aware, and reliable explanation of Transformer decision-making, advancing both the practical interpretability and conceptual understanding of deep neural models.</div><div>The implementation code will be made publicly available at <span><span>https://github.com/melkamumersha/Context-Aware-XAI</span><svg><path></path></svg></span> upon acceptance of the paper.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133050"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386793","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":"Robust learning with time series noisy labels via self-supervised learning and soft labels refurbishment","authors":"Jiarong Liu ,&nbsp;Kaixiang Yang ,&nbsp;Jian He ,&nbsp;Chengrong Yang ,&nbsp;Shuang-hua Yang ,&nbsp;Yujue Zhou","doi":"10.1016/j.neucom.2026.133059","DOIUrl":"10.1016/j.neucom.2026.133059","url":null,"abstract":"<div><div>Label noise significantly degrades the generalization performance of Deep Neural Networks (DNNs). While Learning with Noisy Labels (LNL) is well-established in computer vision, its application to time-series data presents unique challenges: (1) Feature extraction in supervised learning is corrupted by noisy labels, as the critical assumption that clean samples yield lower loss than noisy ones is frequently violated. This leads to learned representations with poor class separability and ambiguous boundaries, which degrades downstream task performance. (2) Conventional soft-labeling methods generate soft-labels from a single training epoch, ignoring historical and cross-model information. This often leads to unstable supervision and inconsistent soft-labels. To address these challenges, this paper proposes the Robust Representation Learning Network (RoRLNet) for noisy time-series classification. RoRLNet employs a two-stage robust learning paradigm that decouples feature extraction from classifier training. In the first stage, it learns noise-robust spatio-temporal representations by integrating MixDecomposition, a data augmentation strategy based on trend-seasonality decomposition, with MSSFE, a multi-scale self-supervised feature extractor. In the second stage, it trains the classifier using EnBootstrap, a soft-label correction module that stabilizes supervision by ensembling predictions from multiple models and historical epochs. Extensive experiments on multiple benchmarks under diverse noise conditions demonstrate that RoRLNet consistently outperforms state-of-the-art methods, by 7.76%. The source code is available at: <span><span>https://github.com/JingGu-hub/RoRLNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133059"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386325","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":"Reinforcement learning control for once-through boiler-turbine units based on mechanistic directed multi-subgraph integration","authors":"Zilong Liu,&nbsp;Chiqiang Liu,&nbsp;Dazi Li","doi":"10.1016/j.neucom.2026.133010","DOIUrl":"10.1016/j.neucom.2026.133010","url":null,"abstract":"<div><div>Once-through boiler-turbine (OTBT) unit, as the core component of large-scale coal-fired power units, exhibits strong interdependencies among its state variables, which pose substantial challenges to conventional control strategies such as PID. Existing reinforcement learning (RL) approaches in complex process control often suffer from insufficient utilization of mechanistic knowledge and the presence of steady-state error. To address these issues, this paper proposes a new RL control framework named multi-subgraph integration reinforcement learning (MSIRL) for the OTBT unit. Based on the mechanistic model, a directed graph is constructed to characterize the dependencies of state variables, through which a graph attention network (GAT) extracts both local and global coupling features that are subsequently integrated into the Actor-Critic framework for end-to-end training. An integral compensation module is integrated at the model output to actively mitigate steady-state error induced by model uncertainties or external disturbances. Under setpoint tracking and disturbance rejection scenarios, the proposed method demonstrates significant advantages in both dynamic response speed and steady-state accuracy compared with conventional PID and standard RL approaches. Finally, ablation experiments further validate the critical roles of mechanistic directed graph embedding and integral compensation in enhancing the control performance of the system.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133010"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386329","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":"3MU-Net: A multi-layer, multi-view and multi-modal segmentation model for PET/CT images of lung tumors","authors":"Yuxia Niu ,&nbsp;Tao Zhou ,&nbsp;Yang Liu ,&nbsp;Qitao Liu","doi":"10.1016/j.neucom.2026.133027","DOIUrl":"10.1016/j.neucom.2026.133027","url":null,"abstract":"<div><div>Multimodal medical images provide reliable and accurate information on the lesion categories, which is essential for diagnosing lung cancer. In medical images, the adequate extraction of lesion cross-modal features and lesion contextual semantic information is a key issue. 3MU-Net is proposed by this paper, which is a Multi-layer, Multi-view and Multi-modal lung tumors segmentation model. The main contributions of 3MU-Net are as follows: Firstly, CMformer uses parallel CNN branch and Transformer branch to learn cross-modal medical image features, which builds cross-modal pixel dependencies between global features and local features. Secondly, Cross-view Context-aware Processor is designed, there are four Deep-Shallow Feature Enhancement Modules and a Cross-view Attention Module in Cross-view Context-aware Processor. Deep-Shallow Feature Enhancement Module uses a bidirectional learning method to guide the information interaction between shallow-level features and deep-level features, which reduces the semantic information gap between adjacent layers. In addition, the Deep-Shallow Feature Enhancement Module aggregates coarse-grained semantic features and fine-grained detailed features. Cross-view Attention Module is used to obtain multi-scale lesion information through multi-view feature extraction, which extends the diversity of spatial features in the lesion area. Finally, the effectiveness of 3MU-Net is validated on a clinical multimodal lung medical image dataset. The results for Miou, Dice, Voe, Rvd, Acc, and Recall are 90.97%, 95.23%, 93.87%, 94.13%, 97.97%, and 94.45%, respectively. It is great significance for lung tumors segmentation. Code is available at: <span><span>https://github.com/xiaoniu1030/</span><svg><path></path></svg></span> 3MU-Net.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133027"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146173296","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":"AVSCNet: A dual-branch network for synchronization detection and content consistency learning in audio-video forgery detection","authors":"Guangwei Zhu ,&nbsp;Hai-Sheng Li ,&nbsp;Haiying Xia ,&nbsp;Shuxiang Song","doi":"10.1016/j.neucom.2026.133067","DOIUrl":"10.1016/j.neucom.2026.133067","url":null,"abstract":"<div><div>Deepfake videos significantly threaten digital media credibility and public trust. While existing multimodal detection methods have advanced, they struggle to generalize across diverse real-world scenarios. Most current approaches focus exclusively on either synchronization detection or content consistency checking, limiting their effectiveness. To tackle these challenges, this study introduces a new dual-branch architecture that simultaneously learns synchronization features and content consistency representations. The model includes a synchronization branch to capture temporal misalignments and a content branch to detect semantic anomalies, with decoupling loss to enhance task specificity. In the content branch, a conditional generation task is introduced to reconstruct the fused feature sequence based on the content token, enhancing the resilience of feature representations through self-supervised learning. The proposed method also includes a hierarchical cross-modal interaction mechanism with cross-attention and fine-grained embeddings. Cross-attention combines features from different modalities to improve feature representations. Fine-grained embeddings provide the model with detailed information. Experimental results show that our approach attains an AUC of 98.30% on the FakeAVCeleb dataset, approaching the current SOTA. When evaluated across datasets, it outperformed the SOTA approaches by 0.08%, 13.46%, and 10.12% on the DeepfakeTIMIT, LAV-DF, and MAVOS-DD datasets, respectively, with AUC scores of 99.11%, 86.97%, and 67.23%. Our code is available at <span><span>https://github.com/zhudedede5-droid/AVSCNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133067"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386723","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 dendritic neuron model enhanced by the synaptic-attention mechanism and fusion-dendritic layer","authors":"Runcong Ma ,&nbsp;Yonghua Pang ,&nbsp;Canlong Zhang ,&nbsp;Xudong Luo","doi":"10.1016/j.neucom.2026.133030","DOIUrl":"10.1016/j.neucom.2026.133030","url":null,"abstract":"<div><div>Among various neural network architectures, the Dendritic Neuron Model (DNM) provides a biologically plausible framework with unique computational properties due to its nonlinearity, interpretability and efficiency. The nonlinearity is attributed to the multiplicative aggregation of dendritic synaptic features, however, when facing high-dimensional data, the multiplicative aggregation of multiple normalized features can lead to the exponential decay of output and catastrophic vanishing of gradients during backpropagation. To alleviate this issue, we proposed the Fusion-Dendritic Layer that uses mean and multiplicative aggregation instead of the original multiplicative aggregation, and a theoretical and empirical analysis confirm the effectiveness of this improvement. Furthermore, we integrated a novel Synaptic-Attention module after the synaptic layer, thus enabling the model to focus on task-relevant information, which accelerates model convergence. Experiments on 31 public datasets showed that the improved DNM achieved remarkable efficiency with very few parameters and effectively addressed the stability limitations of standard DNMs. Compared with various recent DNM variants, our DNM also obtained higher accuracy and faster convergence speed when processing high-dimensional data. The code is available at <span><span>https://github.com/PPOMZ/SAM-FDL-DNM</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"676 ","pages":"Article 133030"},"PeriodicalIF":6.5,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147386788","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":"Low-rank surrogate modeling and stochastic zero-order optimization for training of neural networks with black-box layers","authors":"Andrei Chertkov ,&nbsp;Artem Basharin ,&nbsp;Mikhail Saygin ,&nbsp;Evgeny Frolov ,&nbsp;Stanislav Straupe ,&nbsp;Ivan Oseledets","doi":"10.1016/j.neucom.2026.133808","DOIUrl":"10.1016/j.neucom.2026.133808","url":null,"abstract":"<div><div>The growing demand for energy-efficient, high-performance AI systems has led to increased attention on alternative computing platforms (e.g., photonic, neuromorphic) due to their potential to accelerate learning and inference. However, integrating such physical components into deep learning pipelines remains challenging, as physical devices often offer limited expressiveness, and their non-differentiable nature renders on-device backpropagation difficult or infeasible. This motivates the development of hybrid architectures that combine digital neural networks with reconfigurable physical layers, which effectively behave as black boxes. In this work, we present a framework for the end-to-end training of such hybrid networks. This framework integrates stochastic zeroth-order optimization for updating the physical layer’s internal parameters with a dynamic low-rank surrogate model that enables gradient propagation through the physical layer. A key component of our approach is the implicit projector-splitting integrator algorithm, which updates the lightweight surrogate model after each forward pass with minimal hardware queries, thereby avoiding costly full matrix reconstruction. We demonstrate our method across diverse deep learning tasks, including: computer vision (deep convolutional network on the CIFAR-10 dataset), audio classification (ECAPA-TDNN model on the UrbanSound8K dataset), and language modeling (a decoder-only transformer model on the FineWeb corpus). Notably, across all modalities, the proposed approach achieves near-digital baseline accuracy and consistently enables effective end-to-end training of hybrid models incorporating various non-differentiable physical components (spatial light modulators, microring resonators, and Mach-Zehnder interferometers). This work bridges hardware-aware deep learning and gradient-free optimization, thereby offering a practical pathway for integrating non-differentiable physical components into scalable, end-to-end trainable AI systems.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"691 ","pages":"Article 133808"},"PeriodicalIF":6.5,"publicationDate":"2026-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147827379","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":"SPEED: Structured kernel block pruning with filter groups for efficient and elastic SW-HW co-design in FPGA-based CNN accelerators","authors":"Kwanghyun Koo ,&nbsp;Sunwoong Kim ,&nbsp;Hyun Kim","doi":"10.1016/j.neucom.2026.132958","DOIUrl":"10.1016/j.neucom.2026.132958","url":null,"abstract":"<div><div>On-device AI has received increasing attention due to its ability to provide personalized performance, reduce server load, and address privacy concerns. In this context, efforts have been made to deploy deep learning models on power-efficient hardware platforms, such as field-programmable gate arrays (FPGAs). Specifically, various pruning techniques have been devised to improve performance and energy consumption. However, prior pruning methods fail to achieve balanced hardware utilization, which limits actual performance gains. This paper proposes <em>SPEED</em>, a hardware-aware structured pruning framework integrated into FPGA-based convolutional neural network (CNN) accelerators. <em>SPEED</em> introduces a novel processing unit (PU)-aware kernel block pruning technique for balanced computation across a PU array. Additionally, it proposes an adaptive kernel merging technique to minimize information loss during pruning. Experiments on ResNet18, ResNet50, and YOLACT using ImageNet and Pascal VOC2012 datasets show that <em>SPEED</em> achieves comparable accuracy to software-based pruning methods while achieving higher throughput and lower latency, validated on two types of processing elements. Specifically, for ResNet18, <em>SPEED</em> removes 57.9% of parameters and 44.6% of FLOPs with only a 0.91% drop in Top-1 accuracy, and for ResNet50, it removes 73.2% of parameters and 66.0% of FLOPs with a 1.20% drop in Top-1 accuracy. FPGA benchmarking results show that <em>SPEED</em> efficiently converts reductions in floating-point operations into actual speedups, with little increase in hardware resource usage. When deployed on an FPGA board, <em>SPEED</em> improves FPS by 42.2% and enhances power efficiency by 42.7% compared to the baseline. Case studies in CNN classification and instance segmentation models demonstrate the effectiveness of <em>SPEED</em> as a practical pruning solution for FPGA-based CNN accelerators.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"675 ","pages":"Article 132958"},"PeriodicalIF":6.5,"publicationDate":"2026-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146172996","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":"Tuning metaheuristic parameters with the use of Large Language Models","authors":"Alicja Martinek ,&nbsp;Ewelina Bartuzi-Trokielewicz ,&nbsp;Szymon Łukasik ,&nbsp;Amir H. Gandomi","doi":"10.1016/j.neucom.2026.132976","DOIUrl":"10.1016/j.neucom.2026.132976","url":null,"abstract":"<div><div>Since their explosion in popularity, the impact of Large Language Models (LLMs) has been evident in almost every aspect of life. This study examines whether LLMs can be utilized for tuning metaheuristic algorithms through the selection of their parameters. To verify this hypothesis, ten instances each of three well-known combinatorial optimization problems, Graph Coloring, Job-Shop Scheduling, and Traveling Salesman, were solved using heuristic optimizers guided by LLMs, including genetic algorithm, ant colony optimization, particle swarm optimization, and simulated annealing. Parameter values were generated by prompting several state-of-the-art LLMs with problem complexity descriptors and the set of tunable parameters. A two-stage procedure was employed: an initial run based on general problem characteristics, followed by a feedback run that used performance metrics such as average fitness, variance, and convergence behavior. Default settings from the Python-based Mealpy library served as the baseline for comparison.</div><div>Results, aggregated over 900 optimizer runs, show that LLMs are capable of proposing parameter configurations that outperform defaults in terms of final objective value and convergence speed. This effect is particularly pronounced in simulated annealing and Traveling Salesman problem settings. The findings suggest that LLMs possess a high degree of generalization and contextual understanding in the domain of optimization and can serve as practical assistants in heuristic algorithm design and tuning.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"675 ","pages":"Article 132976"},"PeriodicalIF":6.5,"publicationDate":"2026-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146172924","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}