最新刊期
卷 11 ， 期 6 ， 2025
- 摘要：Public data, as a key component of data elements, encompasses diverse operational models. Local governments, as the main entities authorizing public data, have introduced policies to explore models for the authorization and operation of public data. Analysis of policy texts on the approval and operation of public data reveals that current public data authorization and operation have become one of the main models for the development and utilization of public data, yet different regions have different focal points. Research suggests that a “hybrid authorization model” should be encouraged, as it increases institutional incentives for data source departments, facilitates the entry of public data into trading, and promotes the effective use of public data to enhance the effectiveness of authorization and operational policies.关键词：public data;authorized operation;model comparison;policy analysis319|187|0更新时间：2025-11-15
- 摘要：The concept of “Data Elements ×” serves as a perspective for demand-side data usage, which is interdependent and mutually reinforcing with the development and utilization of public data. This paper analyzes the development and utilization of public data from the perspective of “Data Elements ×”, focusing on the “in-market exchange of public data products and services”. Specifically, we examine the impacts of in-market circulation and exchange of public data products and services on usage thresholds, usage costs, and usage levels of data from three aspects: element attributes, supply-demand matching, and effect realization. Furthermore, by combining existing practices and research in the field of data circulation and transactions, we propose related suggestions for the development of trustworthy data spaces, the improvement of evaluation mechanisms, and the establishment of a circulation and transaction service ecosystem. These suggestions could be beneficial to unleash the vitality of data usage and promote a more prosperous data industry ecosystem.关键词：Data Element ×;in-market circulation and exchange;value release291|184|0更新时间：2025-11-15
- 摘要：Recently, the open-source routes of Alibaba and DeepSeek have sparked a global open-source wave. The European Union clearly signaled its support for the development of open source at the Paris AI Action Summit. Google and Microsoft have successively open-sourced cutting-edge multimodal models, and OpenAI has also signaled that it will open-source large language models. Overall, leading global large model enterprises are accelerating their open-source initiatives, which are conducive to reducing innovation barriers in the AI field. However, the risk of open-source has become an important unresolved issue in the short term. Open source helps curb the widening technological generation gap in the United States, and the European Union seeks technological autonomy through this approach. For China, the overseas expansion of large open-source models presents both opportunities and challenges. In this regard, it is suggested that China consolidate the foundation of the open-source ecosystem, build a secure and reliable industrial system, seize the window of opportunity for the international diffusion of open-source technologies, and further broaden the overseas output path of artificial intelligence technologies and rules.关键词：open-source ecosystem;overseas expansion of industry163|144|0更新时间：2025-11-15
- 摘要：As a new form of the digital economy, the computility economy plays a crucial role in promoting high-quality economic development. To facilitate its healthy and stable growth, an in-depth investigation was conducted into its theoretical connotation with a unified evaluation framework established to measure development levels. The formation mechanism of the computility economy was analyzed, its conceptual connotation and characteristic elements were defined, and a comprehensive evaluation system was constructed, encompassing three dimensions: infrastructure, digital applications, and high-quality development. The research findings reveal that computility economy has emerged as the core driving force for digital economic development. At the same time, significant regional disparities persisted, categorizing development patterns into three distinct echelons: comprehensively leading regions, breakthrough-oriented regions, and potential-catch-up regions. The paper proposes a three-pronged strategy focusing on infrastructure enhancement, application ecosystem cultivation, and high-quality development promotion, which could jointly facilitate the maturity and expansion of China’s computility economy industry.关键词：digital economy;high-quality development;comprehensive index213|352|0更新时间：2025-11-15
FORUM
-
Building high-quality datasets for enterprise-level artificial intelligence: methods and pathways AI导读
摘要：Currently, the artificial intelligence (AI) datasets in our country face challenges, including the lack of quality evaluation methods and an unclear capability-building system. This article reviews the composition and classification of AI datasets and combines structured data quality assessment to propose a set of evaluation methods for AI dataset quality. Based on industry practices, it distills a high-quality AI data engineering system, summarizes the pathways for enterprises to develop capability building, and provides policy recommendations for constructing high-quality datasets in our country.关键词：data quality assessment;AI data engineering318|213|0更新时间：2025-11-15 - 摘要：Artificial intelligence large language model training data usually has the characteristics of large data scale, high data quality and rich data types. At present, although domestic data resources are rich, high-quality chinese large language model training data is still scarce, and there is still a certain gap between the quantity and quality of chinese large language model training data and the world's leading countries. Based on the situation of public data sets and typical general large language model data sets at home and abroad, this paper deeply studied and compared the relevant situation of large language model data sets at home and abroad, analyzed the challenges and problems faced by the development of large language model data sets in China, and puted forward countermeasures and suggestions to enrich the supply of large language model data sets in artificial intelligence.关键词：large language model;data set192|111|0更新时间：2025-11-15
APPLICATION
- 摘要：With the rapid development of large language model (LLM) technology, the demand for processing long-text inputs has been increasing. However, long-text inference faces challenges such as high memory consumption and latency. To improve the efficiency of LLMs in long-text inference, a comprehensive review and analysis of existing optimization techniques were conducted. The study first revealed three key factors that affect efficiency: the first is the huge model size, the second is the attention mechanism operation with quadratic computational complexity, and the third is the autoregressive decoding strategy. These factors together restrict the overall performance of the model. Subsequently, a taxonomy was proposed, categorizing optimization techniques into model optimization, computation optimization, and system optimization, with detailed introductions to key technologies such as quantization, sparse attention, and operator fusion. The research results demonstrate that these optimization techniques can effectively enhance the performance of long-text inference. Finally, future research directions were outlined, emphasizing the importance offurther optimizing LLMs for long-text inference to meet the growing demands of context length.关键词：long-context inference;model optimization;computation optimization;system optimization241|323|0更新时间：2025-11-15
- 摘要：Text data quality is an important factor affecting the performance of language models. and its evaluation methodology is considered decisive for model training effectiveness. To address the issues of high computational costs and incomplete evaluation metrics in existing text data quality assessment methods, a deep learning-oriented text data quality evaluation (TDQE) method was proposed. Specifically, (1) the Dropout module of a text summarization model was utilized to generate multiple stochastic sub-networks, producing embedded representations of data samples to capture semantic consistency, thereby evaluating sample robustness; (2) a text similarity matching model was employed to compute the alignment between data samples and their summaries, assessing sample accuracy; (3) weighted robustness and accuracy metrics were designed to quantify overall text data quality. Comparative experiments were conducted onpublic datasets between TDQE and state-of-the-art methods, and the results demonstrated that TDQE outperformed existing mainstream algorithms.关键词：text data;data evaluation;quality evaluation204|144|0更新时间：2025-11-15
- 摘要：Most of the existing solutions directly process image information, although topic analysis, knowledge graph, and other methods are introduced, there are problems in the processing of image information, such as a single perspective, weakening the story generation process, and a lack of structured design. To solve the above problems, this paper proposes a visual story generation model based on image classification planning learning, introduces image classification and planning learning methods, divides images into seven types: people, animals, food, natural landscapes, architecture, indoor scenes, and others, and sets corresponding questions for each type, and uses the Visual Question pre-trained language model to generate answers and complete planning design, to guide visual story generation. The model is divided into four stages: the first stage extracts visual information from pictures; The second stage is to use the pre-trained language model to classify and guide the generation of planning information; the thirdstage updates the vocabulary information of the dataset; the fourth stage integrates the visual and planning information generated in the previous stages to complete the visual storytelling task. Compared with the existing COVS, the results of the proposed model are improved by 2.07%, 4.29%, 0.44%, 1.78%, 0.91% and 1.07% on BLEU-1, BLEU-2, CIDEr, Distinct-3, Distinct-4, and TTR.关键词：image classification;planning learning;visual question95|129|0更新时间：2025-11-15
- 摘要：In the securities industry, the processing and analysis of user data are critical technologies that significantly impact business decision-making and risk control. However, the vast scale and complexity of user data securities companies led to significant Shuffle operations and data skew issues in big data computations. Existing optimization methods either relied on hardware upgrades or were limited by domain-specific constraints, failing to address the problem effectively. To resolve this, a multi-group merging algorithm (MGMA) based on user relationships was proposed, which improved computational efficiency and reduces resource consumption through effective grouping and optimization strategies. Experimental results showed that, compared to the no optimized（NO） control group, MGMA algorithm achieved a 20% data skew rate, 72% memory usage, and 61% computation time. All three indicators surpass those of the other four comparison optimization methods.关键词：data skew;preprocessing;data of securities companies107|190|0更新时间：2025-11-15
- 摘要：Graph representation learning has emerged as a crucial research area in recent years, aiming to generate vector representations that accurately capture the structure and features of graphs. These vectors play a vital role in downstream tasks such as node classification, link prediction, and anomaly detection. This paper presents a comprehensive survey of graph representation learning methods, categorizing them into traditional graph embedding methods and Graph Neural Network (GNN) based approaches. We discuss various techniques within these categories, including matrix factorization, random walks, graph convolutional networks, and graph Transformers. Furthermore, we delve into the specific applications of GNN in heterogeneous graph embedding, encompassing both static and dynamic aspects. This paper also explores the challenges and future directions of graph representation learning, including scalability and dynamics.关键词：graph neural networks;graph embedding;heterogeneous graph;deep learning368|206|0更新时间：2025-11-15
- 摘要：To address the challenges of detecting small safety helmets in coal-mine tunnels—where images are dim, blurry and poorly lit—a refined YOLOv8 model is proposed. First, an SPD-Conv layer is inserted into the backbone to strengthen feature extraction for small objects. Next, the SPPF module is replaced with SPPFCSPC to enrich multi-scale features. Finally, the CoTAttention mechanism is introduced in the neck to focus on salient cues. Trained and tested on theunderground image set CUMT-HelmeT, the improved Model3 surpasses the original YOLOv8n by 9.08% F1, 17.7% Precision, 3.8% Recall, 3.5% mAP50 and 1.9% mAP50-95, markedly reducing missed and false detections under low-illumination and hazy conditions. Compared with YOLOv10-X, the enhanced YOLOv8-X further improves by 6.23% F1, 4.7% Precision, 7% Recall, 3.9% mAP50 and 1.8% mAP50-95, confirming its superiority in complex underground environments.关键词：SPD-Conv;SPPFCSPC;attention mechanism308|166|0更新时间：2025-11-15
STUDY
- Postal code：100079
- Tel：（010）53879208、53859533 Email：bdr@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：273110 Visits Today：16
0