最新刊期
卷 7 ， 期 4 ， 2025
- 摘要：As the U.S. heralds a new paradigm of AI-driven scientific discovery, global competition has escalated from isolated technical rivalries to a systemic struggle for "breathing sovereignty”, i.e., a nation's fundamental capacity for survival. The core risk is that without autonomous institutional standards and evaluation systems, mere technical pursuit is perpetually vulnerable to dependency and the severance of knowledge sources. This article argues for reconstructing the coupled "technology-institution-talent” ecosystem, calling for indigenous high-impact journals and evaluation mechanisms to reclaim the authority to define and value scientific innovation. Accordingly, it proposes strategic lifecycle education frameworks, such as CASE-K21 and iSTREAMs, to cultivate interdisciplinary talent capable of navigating knowledge automation and governing complex social systems. Furthermore, leveraging parallel intelligence and parallel society, it outlines a virtual-real interactive governance framework to offer solutions for cross-border data flows and global digital order disputes. Finally, the article emphasizes that technological autonomy is not isolationism; rather, mastering core ecological capabilities is the prerequisite for engaging in high-level openness and global cooperation with genuine confidence.关键词：artificial intelligence;smart education;iSTREAMs;K21;CASE;technological competition186|201|0更新时间：2025-12-15
Review Intelligence
- 摘要：The integration of artificial intelligence (AI) with deep excavation engineering represents a key path for the intelligent development of underground space. This paper provides a state-of-the-art review of the research status and development trends of AI applications in the full lifecycle of deep excavation engineering. First, based on the key stages of deep excavation, the applications of traditional AI algorithms, deep learning methods, and generative AI models are briefly summarized. Subsequently, it systematically summarizes and compares the fundamental principles, applicability, and limitations of AI in deep excavation engineering from three perspectives: intelligent geological investigation, intelligent design, and intelligent monitoring. For intelligent investigation, AI algorithms have enabled high-precision classification of rock and soil masses and the reconstruction of 3D geological models. For intelligent design, AI methods have enhanced the efficiency and creativity of design schemes through parameter inversion, optimization, and generation. Furthermore, AI is closely integrated with deep excavation monitoring, enabling single-point time-series prediction and multi-point spatiotemporal correlation prediction of deformation, as well as the establishment of a digital twin enhanced by multi-source data fusion. Finally, four directions for the integration of AI and deep excavation engineering are proposed: establishing comprehensive multi-source databases, developing universal intelligent algorithms for deep excavations, constructing models driven by data-theory-experience fused mechanism, and improving the user ecosystem for AI in deep excavations.关键词：deep excavation;artificial intelligence;intelligent geological survey;intelligent design;intelligent monitoring317|239|0更新时间：2025-12-15
Surveys and Prospectives
- 摘要：To address the challenge of SLAM systems struggling to extract sufficient stable feature points and being prone to incorrect matching in low-texture dynamic environments, which leads to poor accuracy and robustness in pose estimation, this paper proposes a monocular visual-inertial SLAM system specifically designed for low-texture dynamic scenarios. The proposed method achieves improved performance in feature extraction, matching, and dynamic feature point discrimination. First, a deep learning-based SuperPoint feature extraction and LightGlue feature matching modules are employed to replace the existing ORB-SLAM3 frontend, significantly enhancing the robustness of feature extraction and matching in weak texture areas. Second, by integrating YOLO-seg for dynamic region semantic segmentation and leveraging IMU pre-integration to estimate camera pose changes, a joint dynamic point removal mechanism is constructed to achieve finer-grained dynamic point filtering, thereby enhancing system accuracy and robustness in dynamic interference scenarios. Finally, the performance of the proposed method was validated through comparative experiments on public datasets and real-world scenarios, and the proposed system achieves a reduction of 88.4% or more in the root mean square error of absolute trajectory error, and 90% or more in the standard deviation in real-world low-texture dynamic scenarios, exhibiting superior positioning accuracy and robustness when compared with existing visual and visual-inertial SLAM approaches.关键词：dynamic SLAM;low-texture;deep learning;feature extraction;semantic segmentation;IMU pre-integration212|231|0更新时间：2025-12-15
- 摘要：Well-logging text generation is a key task in oil and gas exploration and development, and its quality directly affects the efficiency and accuracy of subsequent geological structure interpretation. Existing approaches mainly include template-based rule strategies, statistical summarization techniques, and small-scale data-driven models based on recurrent neural network (RNN)/Transformer architectures. However, these methods generally have problems such as insufficient utilization of domain knowledge, poor contextual and logical consistency in long-text generation, and the absence of multi-task collaborative learning mechanisms. To address the high specialization and complexity of Chinese well-logging texts, this paper proposed a multi-task model, K2-KGLogGen, which integrated the semantic understanding capability of knowledge-enhanced bidirectional encoder representations from transformer (K-BERT) with the text generation strength of knowledge graph-enhanced bidirectional and auto-regressive transformer (KG-BART). The model incorporated a well-logging domain knowledge graph to enhance semantic awareness, used a classification module to provide category-specific contextual guidance, and employed a self-attention fusion mechanism to achieve joint optimization of classification and generation. The experimental results show that in the classification task, the proposed model achieves significant improvements in F1-score compared to existing mainstream models. Specifically, it outperforms K-BERT (single-task) by approximately 2.2%, BERT by 3.2%, text convolutional neural network (TextCNN) by 4.7%, and support vector machine + term frequency-inverse document frequency (SVM+TF-IDF) by 9.3%. In the generation task, it attains ROUGE-1, ROUGE-2, and ROUGE-L scores of 0.63, 0.41 and 0.54, significantly outperforming Transformer, text-to-text transfer transformer (T5), unified language model (UniLM), pointer generator network (PGN), and BART. Ablation studies confirm that the self-attention mechanism and knowledge injection module are key contributors to performance gains. The results demonstrate the effectiveness of K2-KGLogGen in professional well-logging text generation and its potential applicability to other highly specialized technical text generation tasks.关键词：well logging text generation;multi-task learning;knowledge graph;K-BERT;KG-BART;attention fusion mechanism79|160|0更新时间：2025-12-15
- 摘要：To address three-dimensional collision risks associated with oversized cargo vehicles and enhance path smoothness and tracking accuracy, a closed-loop iterative path planning method is proposed for oversized cargo vehicles. First, a basic path is planned by incorporating kinematic constraints for oversized cargo vehicles into the hybrid A* algorithm. Second, the basic path is optimized by integrating the controller and vehicle kinematic model to improve path smoothness and reduce path tracking error. Concurrently, during path optimization, a tetrahedral mesh of the oversized cargo obtained via Delaunay triangulation is utilized for refined 3D collision risk identification. Based on the risk identification results, the 2D obstacle map is redefined to perform closed-loop iterative planning, determining the optimal path for the oversized cargo vehicle. Through simulation testing in three typical scenarios, the proposed method effectively identifies 3D collision risks and plans safe paths compared with four alternative approaches. The average curvature of the planned paths is 0.013 7 m-1, with articulation angles consistently below 24.59 °, and the root mean square error reduction exceeds 49.51 %, thereby planning optimal safe paths for large-item logistics vehicles.关键词：oversized cargo;path planning;collision detection;articulated vehicle92|169|0更新时间：2025-12-15
- 摘要：To address the issues of low audio realism and insufficient style diversity in sound effect generation, a generative adversarial network (GAN) model based on a multi-band attention mechanism was proposed. Firstly, a multi-band expansion mode was adopted to extract audio features at different sampling rates, and a relativistic average hinge GAN (RaHingeGAN) loss function was introduced to improve audio generation stability. Secondly, a Transformer attention mechanism was incorporated to enhance the expression of harmonic information and spectral structure, while an Alpha Dropout adaptive regularization layer was applied to mitigate overfitting. Finally, an audio style transfer module was designed to enhance style controllability. During the feature learning process, Mel frequency cepstral coefficient (MFCC), Gammatone frequency cepstrum coefficient (GFCC), and their higher-order differential coefficients were fused to capture the dynamic characteristics of the audio signal. Experiments demonstrated that the proposed GAN model based on the multi-band attention mechanism outperformed existing models in terms of sound effect realism, style diversity, and generation stability, effectively improving the quality of sound effect generation.关键词：sound effect generation;GAN;multi band;Transformer85|144|0更新时间：2025-12-15
- 摘要：To address the problem of severe noise in LiDAR point cloud data collected under adverse weather conditions and the low accuracy of traditional denoising methods, an adaptive joint filtering method based on local features and spatial distance relationships is proposed. First, an intensity threshold is set based on the characteristics of snowflake noise to distinguish potential noise points from valid points. Then, the spatial distribution of snowflake noise is modeled using a log-normal distribution, dividing the noise regions into three types. Corresponding denoising strategies are designed according to the characteristics of noise points in different regions. In high-density regions, an improved DSOR algorithm is applied, incorporating the intensity information of points into the threshold calculation, enabling the algorithm to consider both spatial structure and intensity features. In low-density regions, dynamic thresholds are set based on the local geometric features and neighborhood density of points to filter and remove isolated noise points. Finally, the filtered point clouds from high- and low-density regions are merged with the point cloud from noise-free regions to obtain the complete denoised point cloud. Experimental results show that the adaptive joint filtering method improves precision by 5.16% and recall by approximately 3.29% compared to DSOR, thereby effectively reducing the number of missed noise points. Compared with DDIOR, which also incorporates intensity information, the proposed method maintains a high precision while further enhancing the ability to identify noise points, achieving more stable denoising performance across different traffic scenarios.关键词：adverse weather;point cloud denoising;adaptive joint filtering;DSOR;intensity90|67|0更新时间：2025-12-15
Papers and Reports
- 摘要：Traffic incident detection is a core component of intelligent transportation systems (ITS), but existing methods are limited in processing unstructured social media text, associated geographic information, and collaborative multi-task learning. To address this, a traffic incident detection model based on integrated geographical knowledge enhancement and multi-task learning (GeoKE-MTL) was proposed to improve the accuracy and robustness of incident detection. The model consists of two main components: a knowledge enhancement module and a multi-task learning module. Experimental results show that on a self-built social media text dataset, GeoKE-MTL achieves F1 scores of 79.42% and 79.75% in incident location identification and traffic event identification tasks, respectively, outperforming mainstream baseline models in the incident location identification task. This study validates the effectiveness of integrating geographic knowledge enhancement with multi-task learning in improving detection performance, providing a new solution for real-time event perception in intelligent transportation systems.关键词：traffic incident detection;knowledge enhancement;multi-task learning;intelligent transportation system102|63|0更新时间：2025-12-15
- 摘要：In recent years, the cumulative number of registered developers on the open-source community GitHub has exceeded 100 million, leading to increasingly fierce competition for user attention among projects. Understanding the dynamics of project popularity in open-source communities not only helps researchers and developers grasp technological trends, but also provides valuable references for resource allocation, collaboration decisions, and research investment. However, existing studies often treat popularity as a continuous numerical regression or binary classification task, overlooking the stage-wise evolutionary patterns and gradual transitions exhibited in real project popularity dynamics. Open-source large language model (LLM) projects were taken as the research object. An integrated framework for popularity stage segmentation and prediction was proposed, enabling the automatic characterization and forecasting of project popularity stages. Specifically, feature vectors based on growth rate and cumulative scale of project attention within fixed windows were extracted, and a Gaussian mixture model (GMM) for soft clustering was employed to obtain four interpretable stages of popularity with probabilistic labels. Subsequently, a prediction model based on the temporal fusion transformer (TFT) was designed, which leveraged sequential data of five behavioral indicators, such as attention, number of forks, and number of commits, to forecast the stage probability distribution in future windows. Experimental results demonstrate that the proposed method outperforms existing baselines in terms of cross-entropy, mean squared error, mean absolute error, and accuracy, while maintaining high precision across all four patterns of popularity evolution, thereby validating its generalization capability. The proposed approach provides a practical and scalable framework for popularity stage prediction of open-source projects, offering strong utility and potential for applications.关键词：open-source community;large language model;machine learning79|52|0更新时间：2025-12-15
- 摘要：In recent years, numerous social agents empowered by large language model (LLM) have emerged on social media, which play a significant role in online social interactions. However, since not all messages shared on social media are verified as genuine, the involvement of these agents could amplify rumor propagation. Therefore, it becomes increasingly important to measure agents' susceptibility to rumors and reduce their acceptance of rumors. To tackle this issue, agents' susceptibility to rumors and how their opinion evolve on social media were systematically examined. The findings demonstrate that agents are highly susceptible to unknown rumors, tend to reinforce their beliefs over time through prolonged exposure. Furthermore, agents show a strong tendency to believe rumors during social interactions. To reduce agents' susceptibility to rumors, a “self-prompting” intervention strategy was proposed, which significantly reduced rumor acceptance rate among agents from 75.91% to 13.50% and effectively motivated agents with a neutral stance to take on anti-rumor positions. This research not only deepens our understanding of the mechanisms behind the rumor susceptibility of LLM-driven agents, but also provides an effective pathway to improve their anti-rumor capabilities, thereby offering support for the safe deployment of agents on social media and reduction of rumor propagation.关键词：large language model;social agent;rumor susceptibility;rumor intervention101|44|0更新时间：2025-12-15
- 摘要：The randomness of individual behavior, the complexity of social relationships, and the lag in prevention and control strategies make the transmission process of infectious diseases highly dynamic and uncertain. This poses new challenges to the analysis of the evolution of infectious diseases. Existing research has made significant progress in disease transmission, behavioral evolution, and policy intervention through multi-layer network modeling and individual behavior analysis. However, there is still insufficient systematic modeling of cross-layer closed-loop mechanisms, which makes it difficult to fully explain the coupling relationship between individual behavior, social network structure, and prevention and control strategies. This paper describes the dynamic interaction process of “policy, behavior, dissemination, and feedback” as a whole, and proposes a computational model for the evolution of infectious diseases based on social networks and individual behavior, which includes three interconnected levels: individual level, organizational level, and social level. Finally, the effectiveness of this model was verified by analyzing and calculating the number of surviving agents of various types under different strategies and the final stable state achieved in the experiment.关键词：infectious disease model;computational experiment;social network;individual behavior;prevention and control strategy64|28|0更新时间：2025-12-15
Topics\：Social Computing and Resilient Society
- Postal code：100079
- Tel：（010）53879206 Email：cjist@bjxintong.com.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：161982 Visits Today：145
0