最新刊期
卷 6 ， 期 2 ， 2024
- 摘要：Computational knowledge vision is emphasized as a novel perspective or field in this paper. Computational knowledge vision injects a priori knowledge into a visual perception model so that the model gains comprehension, which enables the model to learn and reason effectively to a certain extent like an ordinary adult. Computational knowledge vision not only provides a "bridge" between human priori knowledge and perceptual visual machines, but also systematically adapts to different levels of visual tasks on this "bridge". In addition, this paper proposes the visual hierarchy and its connection to knowledge, stating that knowledge is a justified true belief. Finally, our recent work is summarized and a new direction is proposed, which suggests that knowledge is also a thought framework in vision.关键词：vision;knowledge;computational knowledge vision351|342|0更新时间：2024-06-15
Review Intelligence
- 摘要：Large language models and vision-language models have demonstrated significant potential in various downstream applications, making it become a research hotspot. However, the issues such as hallucinations and knowledge transfer impact the performance of these models. Firstly, this paper explores the fundamental principles of prompt engineering and alignment techniques, and proposes the concept of "prescriptive", which is based on optimizing prompts and expert feedback verification and can be adjusted in real-time. This aims to further enhance the performance of large language models in cross-domain applications. Secondly, the core technologies of prompt engineering, such as the principles of multi-step reasoning for handling complex tasks, are analyzed in depth. Additionally, the development status of prompt engineering is discussed based on practical applications in various fields. Finally, this paper summarizes the challenges faced by prompt engineering and looks into its future development directions. The development of prompt engineering in theory and application, provide comprehensive solutions for improving the performance of large models in practical applications.关键词：prompt engineering;alignment technology;generative artificial intelligence;large language models;vision-language models562|449|3更新时间：2024-06-15
Surveys and Prospectives
- 摘要：Museums are responsible for the preservation, research and dissemination of vast cultural heritage. With the rapid development of artificial intelligence technology, society's demand for the intelligence level of museums is increasing. General artificial intelligence represented by large language models and artificial intelligence agents has achieved milestone progress, providing new technological support for the construction of museums in the new era. The artificial intelligence agent-driven human-machine hybrid museum management architecture was proposed based on parallel intelligence and parallel museum systems, thereby further leveraging the framework of parallel interaction and parallel execution of parallel museums. The system architecture and key technologies of intelligent agent-driven parallel museums were elaborated. Parallel execution and interaction optimization between artificial museums and real museums were realized by building a production factory and operation platform for intelligent agents based on artificial systems, using computational experiments to model museums and train intelligent agents, and constructing a human-machine hybrid intelligent agent team composed of biological workers, digital workers, and robotic workers. Finally, typical cases of intelligent agent-driven parallel museums were introduced.关键词：parallel museum;intelligent museum;parallel system;AI agent326|311|1更新时间：2024-06-15
- 摘要：As foundational models in specialized fields like sound art have rapidly advanced, the convergence of artificial intelligence with music creation and performance has become increasingly pronounced. A novel framework for music creation and performance called the parallel music system was proposed. Grounded in parallel systems theory and the artificial system, computational experiments, parallel execution (ACP) approach, this system stimulates the real-world music creation and performance environments to build highly authentic virtual scenarios. It facilitates real-time, seamless interactions between the virtual and real music systems and outlines the technological strategies for music creation and performance in the era of large models, incorporating a hybrid team of humans, digital beings, and robots. Additionally, the theoretical framework of parallel music and its critical technologies are discussed, the innovative contributions and potential in the modern music field are evaluated, the applications in scenarios like music generation and music therapy are examined. The objective is to harness the emerging AI technologies of the large model era to enhance traditional paths in music creation, drive technological advancements in music performance, unlock the creative potential of musicians, and offer fresh perspectives and inspirations for music creation and performance in the new era.关键词：parallel systems;large models;parallel arts;parallel music;parallel human;ACP approach342|300|0更新时间：2024-06-15
- 摘要：Tourism, as an activity that satisfies people's desire for diverse life experiences and knowledge, has had profound impacts on the economy, culture, and other fields. However, with the rapid development of technologies, such as the Internet of Things and multimodal large language models, traditional tourism is unable to meet the demand for intelligent and personalized travel experiences. To address this, this paper proposes an interactive personalized tourism service system based on the concept of parallel intelligence, utilizing the ACP approach and multimodal large language models. The system constructs a comprehensive tourism model and leverages retrieval-augmented generation and multi-agent collaboration systems to create a new paradigm for personalized tourism services. Additionally, this paper explores the application ecosystem of parallel tourism, expanding the tourism ecosystem from four aspects: clothing, food, accommodation and transportation. This paper analyzes the integration of other industries with personalized tourism services. Parallel tourism is poised to bring new possibilities for the development of the tourism service industry.关键词：parallel tourism;parallel systems;ACP approach;multimodal large language models;retrieval-augmented generation;metaverse343|306|1更新时间：2024-06-15
- 摘要：In recent years, multimodal content generation models have rapidly evolved, significantly improving the efficiency of creative workers. However, existing methods still require expertise and cannot realize the automation of the whole creation process, making it difficult to widely serve the public. This paper proposes the concept and architecture of intelligent art factory, based on advanced foundation models and AI agent systems. Creative production oriented agents and digital teams, in conjunction with the basic workflow, functions, and methods of the intelligent art factory are established. The creative process automation and creative knowledge automation are achieved through collaboration among biological humans, digital humans, and robots in cyber-physical-social spaces. The intelligent art factory seeks to further liberate the creativity of the public.关键词：intelligent art factory;parallel art;foundation model;AI agent;human-machine co-creation304|299|1更新时间：2024-06-15
Special Topic: New Liberal Arts
- 摘要：Facing the future requirements of distributed warfare at sea, the strategic planning of missile penetration is firstly analyzed based on the background of intelligent missile salvo penetration against surface ships in distributed warfare scenario. Secondly, a strategic planning method of intelligent missile penetration based on reward-shaping reinforcement learning is designed by using multi-class reward function. Then, the operation scenario of the missile penetration ship is constructed on the Mozi joint operation simulation system. The comparison experiment shows that the success rate of the intelligent missile penetration attack controlled by the model learned by the reward molding method is 79%, which verifies the effectiveness of the reward-based reinforcement learning method. Finally, after action review, it is found that there are emerging four kinds of penetration strategies of intelligent missiles in the reward shaping experiment, including concentrated and roundabout attack, scattered penetration multi-direction attack, group delay attack and cruise detection guide attack.关键词：penetrating anti-ship;intelligent missile;salvo penetration;task planning;wargaming299|328|0更新时间：2024-06-15
-
Intelligent testing method for railway CTC interface data based on fuzzy natural language processing
摘要：Fuzzy natural language processing applies fuzzy theoretical knowledge to the task of natural language processing (NLP). With the continuous development of large language model and artificial intelligence, research on text data continues to deepen. As a large and complex system, the interface data between various subsystems and server software are stored and transmitted in log text format. Due to its large number of texts and miscellaneous text types, a fuzzy NLP method was proposed to solve the problem of manual testing the interface data of centralized traffic control (CTC) system. The fuzzy C-means (FCM) clustering algorithm divided the log text into different label categories, which was used as the label input for named entity recognition in NLP tasks, and BERT was introduced on the traditional BiLSTM-CRF model for text encoding, which understood the relationship between texts more accurately and improved the accuracy of text recognition. An intelligent verification tool for log-text interface testing of railway CTC system was presented based on an improved training model, which enhanced the current manual testing process of CTC system, assisted testing staff in verifying the interface testing, and increased the level of intelligence and automation in testing work.关键词：natural language processing;fuzzy text clustering;railway centralized traffic control system;named entity recognition;intelligent testing331|329|2更新时间：2024-06-15 - 摘要：The acceleration of digital transformation has made data a core element driving technological innovation and economic growth. However, the widespread phenomenon of data island severely hinders data circulation and collaboration across different fields and organizations, making it difficult for data supply to meet the diverse needs of users, thus creating a dilemma in matching data supply and demand. Serving as the core link in the federated ecology and an important window for external empowerment, federated services provide an innovative solution for data circulation and utilization under the principle of distributed data co-governance, thereby enabling the intelligent matching of supply and demand. Firstly, this paper delves into the pivotal role of federated services within the federated ecology, and examines the relationship between federated services and federated ecology's key components. Secondly, it presents a five-layer basic framework of federated services and discusses its core advantages. On this basis, the paper proposes the key supporting technologies for federated services, including blockchain and decentralized autonomous organizations and operations, large language models and scenario engineering, distributed computing and edge computing, as well as encryption technology and privacy computing. Finally, taking the supply-demand matching problem of data and scenarios as a case study, the paper illustrates how federated services achieve intelligent matching of demand and supply. The federated services proposed in this paper holds significant implications for constructing a smart service paradigm based on distributed data co-governance.关键词：federated services;blockchain;decentralized autonomous organization and operation;large language models;decentralized physical infrastructure networks;federated data289|303|0更新时间：2024-06-15
- 摘要：In response to the challenges encountered in quantitative trading of digital currencies, characterized by the presence of a multitude of intricate factors and a high-dimensional factor state space, an enhanced optimistic actor-critic(OAC) model, referred to as OAC_LSTM_ATT, had been proposed. This model incorporated long short-term memory (LSTM) and a multi-head attention mechanism to optimize the network architecture of OAC, thereby augmenting its capacity for modeling time-series data and generalization. Through this integration, the intelligent agent operating in the quantitative trading environment was capable of making more adaptable and precise trading decisions, consequently elevating the quality and efficacy of trading strategies. Experimental findings revealed that, in the Bitcoin market, the cumulative return achieved was 16.36%, with a maximum drawdown of 9.08%, a Sharpe ratio of 0.014, and a volatility of 13.09%. Corresponding metrics in the Ethereum market amounted to 16.30%, 8.56%, 0.014, and 13.42%. When compared to models such as PPO, LSTM_PPO, A2C, OAC_LSTM_ATT demonstrates superior performance in terms of both effectiveness and stability, thereby offering valuable insights for the development of quantitative trading strategies.关键词：quantitative trading;deep reinforcement learning;attention mechanism;long short-term memory;digital currency315|319|0更新时间：2024-06-15
- 摘要：Parkinson's disease (PD) patients were often accompanied by cognitive impairment, which seriously affected the quality of life, so the over-prediction of cognitive impairment in Parkinson's disease was crucial for clinical diagnosis and intervention. However, Parkinson's disease was affected by the coupling of multivariate factors, such as age, gender, and disease duration, which made the overprediction of cognitive impairment a serious challenge. Aiming at the multivariate coupling characteristics of cognitive impairment in Parkinson's disease, in this study, multivariate logistic regression was used to construct a novel column-linear graphical model aiming at over-predicting the risk of cognitive impairment (CI) in Parkinson's disease patients. First, the least absolute shrinkage and selection operator (LASSO) algorithm was applied to analyze the risk factors that may affect the cognitive ability of patients, and the clinical variables with high correlation were screened out. Second, multivariate logistic regression was used to analyze the correlation between variables and construct a visualized novel column-line diagram model to achieve the risk over prediction of cognitive impairment in Parkinson's disease. Finally, the results of model performance evaluation show that the novel cognitive impairment prediction model proposed in this paper has good accuracy, consistency and clinical practicability, which can significantly improve the diagnostic efficiency of clinicians; in addition, the model also realizes the visual comparison and analysis of the number and distribution of patients with different values of the same predictor, which can assist clinicians in formulating personalized healthcare management and consulting programs according to the individual risk of each patient, and help to start the intervention and treatment of the patients at an early stage, and it has a certain clinical diagnostic value.关键词：Parkinson's disease;cognitive impairment;multivariate logistic regression;nomogram272|299|0更新时间：2024-06-15
- 摘要：The existing seam carving tamper detection algorithms have the problems of low detection accuracy and weak robustness for the case of low scaling factor. A seam carving tamper detection algorithm integrated with hybrid attention mechanism is proposed. Firstly, BayarConv2D constrained convolution is used to preprocess the image, fully learn the noise characteristics of the image, and fuse the features with RGB image through matrix multiplication. Then, ResNet is used as the backbone network for feature learning, and the residual propagation and residual feedback mechanisms are introduced to highlight the operation traces of seam carving. Finally, the hybrid attention mechanism is used to simultaneously extract the features between adjacent locations and channels to better capture the global features, and then input them into the full connection layer to achieve classification. The experimental results show that on the BOSSBase1.01 dataset, when the scaling factor is 1% and 9%, the detection accuracy of the proposed method reaches 89.48% and 97.94% respectively, which is better than existing mainstream methods. At the same time, it has lower computational complexity and better robustness, and can resist JPEG compression attacks.关键词：mixed attention mechanism;image forensics;Seam Carving detection;low scaling factor280|316|0更新时间：2024-06-15
- 摘要：The concentration of nitrogen dioxide (NO2) in the air significantly affects environmental protection and public health. Current methods for NO2 concentration prediction lack sufficient characterization of spatiotemporal correlations. Therefore, this study proposes a novel approach using interval type-2 intuitionistic fuzzy neural networks (IT2IFNNs) for spatiotemporal prediction of NO2 concentrations. Firstly, the framework of IT2IFNNs is elucidated, incorporating variable coefficient weighting for its membership and non-membership outputs, and employing a random vector functional-link neural network (RVFLNN) as the rule consequent. Secondly, a hierarchical clustering algorithm is employed to determine the fuzzy rule base and optimize the output weight values of the network consequents using least squares estimation. Finally, numerical validation is conducted using real NO2 concentration data collected in Beijing from January to March 2018. Experimental results demonstrate that compared to existing methods, the proposed approach achieves superior prediction accuracy and efficiency in both short-term and long-term spatiotemporal prediction tasks.关键词：nitrogen dioxide concentration spatiotemporal prediction;interval type-2 intuitionistic fuzzy neural networks;structural identification;parameter optimization;least squares method271|295|1更新时间：2024-06-15
- 摘要：An improved UAVAI-YOLO model was proposed to address the problem of poor target detection in UAV aerial images. Firstly, in order to obtain richer semantic information for the model, the original convolution of the C2f module of the original backbone part was replaced with the improved DCN convolution. Secondly, in order to increase the P2 feature layer without increasing the number of model parameters, the Conv_C module was proposed to downscale the output channel of the backbone network, and at the same time, because of avoiding the loss of semantic information due to channel downsizing, the original convolution of the C2f module in the neck part was replaced by the improved ODConv dynamic convolution. Then, the BIFPN module was introduced to make full use of the contextual semantic information. Finally, Wise-IoU was used to replace the original loss function to improve the accuracy of the model target detection frame. Experimental results on the publicly available VisDrone2019 dataset and UAVDT dataset showed that the UAVAI-YOLO model improves 4.4% and 1.1% compared to the original YOLOv8n model mAP0.5, respectively, high detectability accuracy compared to other mainstream object detection models.关键词：UAV aerial images;small object detection;YOLOv8;DCN;attention mechanisms435|354|0更新时间：2024-06-15
- 摘要：Aiming at the problem of poor generalization ability in hand-crafted architectures that overly rely on expert experience, a neural network architecture search method with an adaptive smoothness strategy was proposed. Firstly, an improved candidate operation selection strategy and a continuous relaxation method were used to convert discrete search space into continuous space, and a weight-sharing mechanism was employed to enhance search efficiency. Secondly, a regularization operation with an adaptive smoothness strategy was added to the loss function, whose smoothness degree was controlled by a temperature parameter. Finally, the loss function was calculated using an exponential normalization method to avoid loss value overflow. Experimental results on 3D point cloud datasets and protein-protein interaction datasets showed that the proposed method achieved higher classification accuracy and more stable performance under the same training samples and iterations.关键词：regularization;neural architecture search;search space;point cloud;classification252|299|0更新时间：2024-06-15
Papers and Reports
- 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