Current Issue

15 March 2022, Volume 4 Issue 1

Previous Issue    Next Issue

Review Intelligence

Traffic situational awareness research and development enhanced by social media data: the state of the art and prospects

Yuanwen CHEN, Xiao WANG, Lingxi LI, Fei-Yue WANG

2022, 4(1):  1-13.  doi:10.11959/j.issn.2096-6652.202220

Asbtract ( 539 )   HTML ( 79)   PDF (744KB) ( 628 )   Knowledge map   

References | Related Articles | Metrics

Traffic situational awareness is an important research direction of intelligent transportation systems.Most of the existing research focused on how to use physical sensors to perceive the current traffic situation and predict the future traffic state.However, the performance of physical sensors is prone to instability or failure due to adverse weather, electromagnetic interference, energy limitation and other problems, resulting in sparse or missing collected data, which makes the perception of traffic situation lagging and inaccurate.Social media data provides a new and enhanced way of perceiving comprehensive traffic situation information in a timely manner.Facing with the current traffic situation where sudden abnormal traffic events occur frequently, social sensing and physical sensing data can complement with each other to further improve the efficiency of urban traffic management.The related work of traffic event detection and traffic state prediction enhancing based on social media data were analyzed, and how those research works provide decision support for the traffic management departments to plan and guide traffic reasonably and alleviate traffic congestion were explored.Finally, some future research directions of traffic situational awareness enhanced by social media data were proposed.

Surveys and Prospectives

A survey on canonical correlation analysis based multi-view learning

Chenfeng GUO, Dongrui WU

2022, 4(1):  14-26.  doi:10.11959/j.issn.2096-6652.202206

Asbtract ( 449 )   HTML ( 72)   PDF (5381KB) ( 372 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Multi-view learning (MVL) is a strategy for fusing data from different sources or subsets.Canonical correlation analysis (CCA) is very important in MVL, whose main idea is to maximize the correlation of different views.The traditional CCA can only calculate the linear correlation between two views.Moreover, it is unsupervised, and the label information is wasted in supervised learning tasks.Many nonlinear, supervised, or generalized extensions have been proposed to accommodate these limitations.Firstly, a comprehensive overview of representative CCA approaches was provided.Then their classical applications in pattern recognition, cross-modal retrieval and classification, and multi-view embedding were described.Finally, the challenges and future research directions of CCA-based MVL approaches were pointed out.

Special Topic: Crowd Intelligence

Crowd Intelligence

2022, 4(1):  27-28.  doi:10.11959/j.issn.2096-6652.202206-1

Asbtract ( 263 )   HTML ( 76)   PDF (500KB) ( 418 )   Knowledge map   

References | Related Articles | Metrics

A survey on federated learning in crowd intelligence

Qiang YANG, Yongxin TONG, Yansheng WANG, Lixin FAN, Wei WANG, Lei CHEN, Wei WANG, Yan KANG

2022, 4(1):  29-44.  doi:10.11959/j.issn.2096-6652.202218

Asbtract ( 1610 )   HTML ( 213)   PDF (2877KB) ( 2396 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Crowd intelligence is emerging as a new artificial intelligence paradigm owing to the rapid development of the Internet.However, the data isolation and data privacy preservation problems make it difficult to share data among the crowd and to build crowd intelligent applications.Federated learning is a novel solution that aims to collaboratively build models by breaking the data barriers in crowd.Firstly, the basic ideas of federated learning and a comparison with crowd intelligence were introduced.Secondly, federated learning algorithms were divided into three categories according to the crowd organization, and further optimization techniques on privacy, accuracy and efficiency were discussed.Thirdly, federated learning operators based on linear models, tree models and neural network models were presented respectively.Finally, mainstream federated learningopensource platforms and typical applications were introduced, followed by the conclusion.

Collective decision-making in open environment: concepts, challenges, and leading technologies

Xueqi CHENG, Bingbing XU, Qi CAO, Shenghua LIU, Juan CHEN, Lei LIN, Huawei SHEN

2022, 4(1):  45-54.  doi:10.11959/j.issn.2096-6652.202219

Asbtract ( 492 )   HTML ( 58)   PDF (1203KB) ( 615 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Research on collective decision-making is based on new decision-making theories and their methods are driven by group participation, human-computer interaction, and big data, to realize complex problem solving and intelligent decision-making in an open environment.However, the open environment is of high openness, complex interaction, and emerging behaviors, making collective decision-making face the challenges of difficult incentive mechanism design, uncontrollable decision-making individuals, diverse decision-making environments, and highly complex decosion-making information.Based on the era background of collective decision-making in an open environment, a new decision-making paradigm was proposed, its conceptual connotation, main challenges, and leading technologies for effective implementation of collective decision-making were sorted out.The successful cases of collective decision-making were analyzed, in order to support the application of new decision-making paradigms in the fields of economy, medical care, and people’s livelihood, and to promote progress and changes in the corresponding fields.

Collective knowledge graph: meta knowledge transfer and federated graph reasoning

Mingyang CHEN, Wen ZHANG, Xiangnan CHEN, Hongting ZHOU, Huajun CHEN

2022, 4(1):  55-64.  doi:10.11959/j.issn.2096-6652.202217

Asbtract ( 609 )   HTML ( 66)   PDF (833KB) ( 662 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Collective knowledge graphs refer to knowledge graphs that are managed and maintained in a decentralized or distributed manner through group collaboration.Compared with the existing centrally managed knowledge graph, the collective knowledge graph has the characteristics of knowledge right confirmation, privacy protection, crowd sourcing incentive, and credible traceability.Tring to explore the technical challenges faced by building and applying a collective knowledge graph platform.For meta knowledge transfer, the knowledge incompleteness of a single knowledge graph by knowledge transfer among multiple knowledge graphs from different sources under a decentralized and autonomous framework was considered.The main difficulty was to enhance the respective knowledge graph representation by sharing useful knowledge with each other as much as possible while fully protecting the autonomous ownership of knowledge.For federated graph reasoning, the knowledge graph reasoning in a distributed environment under the privacy-preserving by means of the federated learning mechanism was considered.Meta knowledge transfer focused on transferring entity-independent knowledge between knowledge graphs with overlapped relation set, while federated graph reasoning aimed at learning better entity embeddings for knowledge graphs with overlapped entity set.The model design and experimental validation for each of these two problems were conducted.

Emergence measurement of robot swarm intelligence based on swarm entropy

Pu FENG, Wenjun WU, Jie LUO, Xin YU, Yongkai TIAN

2022, 4(1):  65-74.  doi:10.11959/j.issn.2096-6652.202213

Asbtract ( 385 )   HTML ( 45)   PDF (5964KB) ( 279 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Swarm behavior can often produce value and complexity far beyond individual behavior.In order to more effectively derive swarm intelligence on the basis of individual intelligence, it is necessary to scientifically measure the level of swarm intelligence based on swarm entropy, and use swarm entropy as the guiding goal to promote the enhancement and evolution of swarm intelligence.Aiming at this important scientific problem, the unmanned car group as the research object was taken and a multi-agent soft Q learning method based on parameter sharing and group strategy entropy was proposed.Which by sharing the observation information of the agent, combined with the maximum entropy reinforcement learning method, to achieve continuous learning and updating of swarm strategies in exploratory tasks.At the same time, by defining swarm entropy as a measurement tool, characterizing the entropy change pattern in swarm learning, realizing the quantitative analysis of the gathering process of swarm intelligence.

A cooperative multi-agent reinforcement learning algorithm based on dynamic self-selection parameters sharing

Han WANG, Yang YU, Yuan JIANG

2022, 4(1):  75-83.  doi:10.11959/j.issn.2096-6652.202214

Asbtract ( 523 )   HTML ( 55)   PDF (3737KB) ( 881 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In multi-agent reinforcement learning, parameter sharing can effectively alleviate the inefficiency of learning caused by non-stationarity.However, maintaining the same policy forall agents during learning may have detrimental effects.To solve this problem, a new approach was introduced to give agents the ability to automatically identify agents that may benefit from parameter sharing and dynamically share parameters them during learning.Specifically, agents needed to encode empirical trajectories as implicit information that can represent their potential intentions, and selected peers to share parameters by comparing their intentions.Experiments show that the proposed method not only can improve the efficiency of parameter sharing, but also ensure the quality of policy learning in multi-agent system.

Crowd intelligent cooperative obstacle avoidance method inspired by brain attention mechanism

Yuming XIANG, Kun CHEN, Zhifeng ZHAO, Rongpeng Li, Honggang ZHANG

2022, 4(1):  84-96.  doi:10.11959/j.issn.2096-6652.202215

Asbtract ( 187 )   HTML ( 20)   PDF (11159KB) ( 441 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

A crowd intelligent (CI) system often acquires, calculates, and transmits a large amount of redundant information during the performing of exploration tasks, which inevitably results in inefficient use of the limited resources.Therefore, it emerges a strong incentive to design a task-driven mechanism for efficient utilization of computing and communication resources.A crowd intelligent cooperative obstacle avoidance method inspired by brain attention mechanism was proposed.Inspired by brain attention mechanism, the CI system introduced an intelligent selection module based on the deep Q network, by efficiently tuning the working state of sensors exploring the unknown environment and realized the acquisition and calculation of key necessary information with as little sensor overhead as possible.Meanwhile, based on the optimal reciprocal collision avoidance algorithm, a single agent fuses a small amount of limited information from neighbor agents to drive the intelligent selection module, so as to greatly reduce the redundancy of sensor acquisition and information calculation required for the obstacle avoidance task.The effectiveness of this proposed method was verified through extensive simulation analyses and practical realization empowered with Kehepera IV robots.The results show that the proposed method can significantly reduce the redundancy of sensor information in the CI system.More importantly, as the number of agents and the amount of information interaction increase, there also emerged a clear trend in the increase of performance gains.

Exploration of the continual learning ability that supports the application ecological evolution of the large-scale pretraining Peng Cheng series open source models

Yue YU, Xin LIU, Fangqing JIANG, Han ZHANG, Hui WANG, Wei ZENG

2022, 4(1):  97-108.  doi:10.11959/j.issn.2096-6652.202212

Asbtract ( 402 )   HTML ( 59)   PDF (3005KB) ( 530 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Large-scale pre-training models have achieved great success in the field of natural language processing by using large-scale corpora and pre-training tasks.With the gradual development of large models, the continual learning ability of large models has become a new research focus.The continual learning technology of the Peng Cheng series large models, the exploration of practice and the still facing challenges were mainly introduced, including the Peng Cheng series continual learning technology through task expansion, data increment and knowledge reasoning, Peng Cheng PANGU multi-task continual learning and the practical exploration of the continual learning ability of the Peng Cheng TONGYAN open source large model, the vocabulary update, semantic mapping and knowledge conflicts that the large model faces in the process of continual learning.

Papers and Reports

Graph-regularized Bayesian broad learning system

Junwei DUAN, Lincan XU, Yujuan QUAN, Long CHEN, C.L.Philip CHEN

2022, 4(1):  109-117.  doi:10.11959/j.issn.2096-6652.202203

Asbtract ( 304 )   HTML ( 27)   PDF (1571KB) ( 444 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

As a feed forward neural network, broad learning system (BLS) has attracted much attention because of its high accuracy, fast training speed, and the ability to effectively replace deep learning methods.However, it is sensitive to the number of feature nodes and the pseudo-inverse method is likely to result in the problem of over fitting for BLS model.To address the above issues, Bayesian inference and graph regularization was introduced in to the BLS model.By introducing the prior knowledge for Bayesian learning, the sparsity of the weights and the stability of the model could be effectively improved; while the graph information mining from the data could be fully considered to improve the generalization ability of the model by regularization.The UCI and NORB dataset were adopted for evaluating the performance of the proposed model.The experiment results demonstrated that the proposed graph-regularized Bayesian broad learning system model can further improve the accuracy of classification and has better stability.

Research on application of edge computing system based on KubeEdge

Hang ZHAO, Sheng LIU, Kun LUO, Shichao CHEN, Linghui KONG, Fan JIA

2022, 4(1):  118-128.  doi:10.11959/j.issn.2096-6652.202201

Asbtract ( 643 )   HTML ( 62)   PDF (2220KB) ( 692 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

With the increasing development of Internet of things, traditional data processing methods based on cloud computing have shown many problems, such as high bandwidth occupation and time delay.Edge computing can be a supplement for cloud computing with the characteristics of low latency and high reliability to process data.The KubeEdge edge computing system and its application were mainly discussed and analyzed.Firstly, the system architecture, functions, and key technologies of KubeEdge were introduced.Secondly, the KubeEdge was applied to the parts assembly scenario with dual-arm cooperative robot, and the functions and performances of a cloud-edge collaboration system based on KubeEdge for the robot assembly application were analyzed and tested.The experimental results show that the system can satisfy the functional and application requirements of the scenario, which also provides basic reference and guidance for practical applications of KubeEdge.

Knowledge graph construction for control systems in process industry

Tianhao MOU, Shaoyuan LI

2022, 4(1):  129-141.  doi:10.11959/j.issn.2096-6652.202216

Asbtract ( 645 )   HTML ( 95)   PDF (1908KB) ( 733 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Achieving intelligence in industrial control systems is a prevailing trend in recent years, with numerous new technologies and ideas prompted.Knowledge graph is a fundamental resource for artificial intelligence, and domain-specific knowledge graph construction attracts a lot of research attentions.However, knowledge graph construction for control systems is still in the early stage of exploitation.In this paper, structural characteristics and task requirements of process control systems were analyzed.Furthermore, a knowledge graph construction methodology architecture for process control systems was proposed.Firstly, a brief summary on existing related works was given.After that, the characteristics of process industry control systems were analyzed, and the corresponding knowledge graph construction principles and procedures were proposed.Cyber-physical assets management was taken as a case study for detailed explanation.Finally, a prospect for the future research directions of knowledge graph construction for process control systems was made.