Top Cited
1
2019, 56(1): 69-89.
DOI: 10.7544/issn1000-1239.2019.20180760
Abstract：
With the burgeoning of the Internet of everything, the amount of data generated by edge devices increases dramatically, resulting in higher network bandwidth requirements. In the meanwhile, the emergence of novel applications calls for the lower latency of the network. It is an unprecedented challenge to guarantee the quality of service while dealing with a massive amount of data for cloud computing, which has pushed the horizon of edge computing. Edge computing calls for processing the data at the edge of the network and develops rapidly from 2014 as it has the potential to reduce latency and bandwidth charges, address the limitation of computing capability of cloud data center, increase availability as well as protect data privacy and security. This paper mainly discusses three questions about edge computing: where does it come from, what is the current status and where is it going? This paper first sorts out the development process of edge computing and divides it into three periods: technology preparation period, rapid growth period and steady development period. This paper then summarizes seven essential technologies that drive the rapid development of edge computing. After that, six typical applications that have been widely used in edge computing are illustrated. Finally, this paper proposes six open problems that need to be solved urgently in future development.
With the burgeoning of the Internet of everything, the amount of data generated by edge devices increases dramatically, resulting in higher network bandwidth requirements. In the meanwhile, the emergence of novel applications calls for the lower latency of the network. It is an unprecedented challenge to guarantee the quality of service while dealing with a massive amount of data for cloud computing, which has pushed the horizon of edge computing. Edge computing calls for processing the data at the edge of the network and develops rapidly from 2014 as it has the potential to reduce latency and bandwidth charges, address the limitation of computing capability of cloud data center, increase availability as well as protect data privacy and security. This paper mainly discusses three questions about edge computing: where does it come from, what is the current status and where is it going? This paper first sorts out the development process of edge computing and divides it into three periods: technology preparation period, rapid growth period and steady development period. This paper then summarizes seven essential technologies that drive the rapid development of edge computing. After that, six typical applications that have been widely used in edge computing are illustrated. Finally, this paper proposes six open problems that need to be solved urgently in future development.
2
2018, 55(11): 2452-2466.
DOI: 10.7544/issn1000-1239.2018.20170658
Abstract：
With the flourishing development of blockchain technology represented by bitcoin, the blockchain technology has moved from the era of programmable currency into the era of smart contract. The smart contract is an event-driven, state-based code contract and algorithm contract, which has been widely concerned and studied with the deep development of blockchain technology. The protocol and user interface are applied to complete all steps of the smart contract process. Smart contract enables users to implement personalized logic on the blockchain. The blockchain-based smart contract technology has the characteristics of de-centralization, autonomy, observability, verifiability and information sharing. It can also be effectively applied to build programmable finance and programmable society, which has been widely used in digital payment, financial asset disposal, multi-signature contract, cloud computing, Internet of things, sharing economy and other fields. The survey describes the basic concepts of smart contract technology, its whole life cycle, basic classification and structure, key technology, the art of the state, as well as its application scenarios and the main technology platforms. Its problems encountered at present are also discussed. Finally, based on the theoretical knowledge of the smart contract, we set up the Ethereum experimental environment and develop a system of crowdsale contract. The survey is aimed at providing helpful guidance and reference for future research of smart contract based on blockchain technology.
With the flourishing development of blockchain technology represented by bitcoin, the blockchain technology has moved from the era of programmable currency into the era of smart contract. The smart contract is an event-driven, state-based code contract and algorithm contract, which has been widely concerned and studied with the deep development of blockchain technology. The protocol and user interface are applied to complete all steps of the smart contract process. Smart contract enables users to implement personalized logic on the blockchain. The blockchain-based smart contract technology has the characteristics of de-centralization, autonomy, observability, verifiability and information sharing. It can also be effectively applied to build programmable finance and programmable society, which has been widely used in digital payment, financial asset disposal, multi-signature contract, cloud computing, Internet of things, sharing economy and other fields. The survey describes the basic concepts of smart contract technology, its whole life cycle, basic classification and structure, key technology, the art of the state, as well as its application scenarios and the main technology platforms. Its problems encountered at present are also discussed. Finally, based on the theoretical knowledge of the smart contract, we set up the Ethereum experimental environment and develop a system of crowdsale contract. The survey is aimed at providing helpful guidance and reference for future research of smart contract based on blockchain technology.
3
2019, 56(10): 2071-2096.
DOI: 10.7544/issn1000-1239.2019.20190540
Abstract：
While machine learning has achieved great success in various domains, the lack of interpretability has limited its widespread applications in real-world tasks, especially security-critical tasks. To overcome this crucial weakness, intensive research on improving the interpretability of machine learning models has emerged, and a plethora of interpretation methods have been proposed to help end users understand its inner working mechanism. However, the research on model interpretation is still in its infancy, and there are a large amount of scientific issues to be resolved. Furthermore, different researchers have different perspectives on solving the interpretation problem and give different definitions for interpretability, and the proposed interpretation methods also have different emphasis. Till now, the research community still lacks a comprehensive understanding of interpretability as well as a scientific guide for the research on model interpretation. In this survey, we review the explanatory problems in machine learning, and make a systematic summary and scientific classification of the existing research works. At the same time, we discuss the potential applications of interpretation related technologies, analyze the relationship between interpretability and the security of interpretable machine learning, and discuss the current research challenges and potential future research directions, aiming at providing necessary help for future researchers to facilitate the research and application of model interpretability.
While machine learning has achieved great success in various domains, the lack of interpretability has limited its widespread applications in real-world tasks, especially security-critical tasks. To overcome this crucial weakness, intensive research on improving the interpretability of machine learning models has emerged, and a plethora of interpretation methods have been proposed to help end users understand its inner working mechanism. However, the research on model interpretation is still in its infancy, and there are a large amount of scientific issues to be resolved. Furthermore, different researchers have different perspectives on solving the interpretation problem and give different definitions for interpretability, and the proposed interpretation methods also have different emphasis. Till now, the research community still lacks a comprehensive understanding of interpretability as well as a scientific guide for the research on model interpretation. In this survey, we review the explanatory problems in machine learning, and make a systematic summary and scientific classification of the existing research works. At the same time, we discuss the potential applications of interpretation related technologies, analyze the relationship between interpretability and the security of interpretable machine learning, and discuss the current research challenges and potential future research directions, aiming at providing necessary help for future researchers to facilitate the research and application of model interpretability.
4
2018, 55(2): 327-337.
DOI: 10.7544/issn1000-1239.2018.20170228
Abstract：
With the trend of Internet of Everything, the number of end devices (e.g. smartphones, smart glasses) has increased rapidly which makes the data produced by these devices have grown at rates far more than the growth rate of network bandwidth. At the same time, the emergence of novel applications demands lower latency of the network, such as augmented reality and manless driving. Edge computing integrates any computing, storage and network resources at the edge of the network into a unified platform that provides services for users. This new computing model gets around the bottleneck that is caused by network bandwidth and latency, has received widespread attention in both industrial and academic. In this survey, we first introduce the concepts of edge computing and provide the definition of it. To help the readers to better understand the characteristics of edge computing, we compare it with cloud computing. We then analyze the three representative instances of edge computing platform and make a systematic comparison of them. After that, we enumerate some typical applications based on the edge computing to describe the advantages of edge computing in mobile or Internet of Things applications. Finally, the paper lays out several grand challenges of edge computing.
With the trend of Internet of Everything, the number of end devices (e.g. smartphones, smart glasses) has increased rapidly which makes the data produced by these devices have grown at rates far more than the growth rate of network bandwidth. At the same time, the emergence of novel applications demands lower latency of the network, such as augmented reality and manless driving. Edge computing integrates any computing, storage and network resources at the edge of the network into a unified platform that provides services for users. This new computing model gets around the bottleneck that is caused by network bandwidth and latency, has received widespread attention in both industrial and academic. In this survey, we first introduce the concepts of edge computing and provide the definition of it. To help the readers to better understand the characteristics of edge computing, we compare it with cloud computing. We then analyze the three representative instances of edge computing platform and make a systematic comparison of them. After that, we enumerate some typical applications based on the edge computing to describe the advantages of edge computing in mobile or Internet of Things applications. Finally, the paper lays out several grand challenges of edge computing.
5
2020, 57(7): 1424-1448.
DOI: 10.7544/issn1000-1239.2020.20190358
Abstract：
There is a phenomenon that information extraction has long been concerned by a lot of research works in the field of natural language processing. Information extraction mainly includes three sub-tasks: entity extraction, relation extraction and event extraction, among which relation extraction is the core mission and a great significant part of information extraction. Furthermore, the main goal of entity relation extraction is to identify and determine the specific relation between entity pairs from plenty of natural language texts, which provides fundamental support for intelligent retrieval, semantic analysis, etc, and improves both search efficiency and the automatic construction of the knowledge base. Then, we briefly expound the development of entity relation extraction and introduce several tools and evaluation systems of relation extraction in both Chinese and English. In addition, four main methods of entity relation extraction are mentioned in this paper, including traditional relation extraction methods, and other three methods respectively based on traditional machine learning, deep learning and open domain. What is more important is that we summarize the mainstream research methods and corresponding representative results in different historical stages, and conduct contrastive analysis concerning different entity relation extraction methods. In the end, we forecast the contents and trend of future research.
There is a phenomenon that information extraction has long been concerned by a lot of research works in the field of natural language processing. Information extraction mainly includes three sub-tasks: entity extraction, relation extraction and event extraction, among which relation extraction is the core mission and a great significant part of information extraction. Furthermore, the main goal of entity relation extraction is to identify and determine the specific relation between entity pairs from plenty of natural language texts, which provides fundamental support for intelligent retrieval, semantic analysis, etc, and improves both search efficiency and the automatic construction of the knowledge base. Then, we briefly expound the development of entity relation extraction and introduce several tools and evaluation systems of relation extraction in both Chinese and English. In addition, four main methods of entity relation extraction are mentioned in this paper, including traditional relation extraction methods, and other three methods respectively based on traditional machine learning, deep learning and open domain. What is more important is that we summarize the mainstream research methods and corresponding representative results in different historical stages, and conduct contrastive analysis concerning different entity relation extraction methods. In the end, we forecast the contents and trend of future research.
6
2022, 59(1): 47-80.
DOI: 10.7544/issn1000-1239.20201055
Abstract：
In recent years, the application of deep learning related to graph structure data has attracted more and more attention. The emergence of graph neural network has made major breakthroughs in the above tasks, such as social networking, natural language processing, computer vision, even life sciences and other fields. The graph neural network can treat the actual problem as the connection between nodes in the graph and the message propagation problem, and the dependence between nodes can be modeled, so that the graph structure data can be handled well. In view of this, the graph neural network model and its application are systematically reviewed. Firstly, the graph convolutional neural network is explained from three aspects: spectral domain, spatial domain and pooling. Then, the graph neural network model based on the attention mechanism and autoencoder is described, and some graph neural network implemented by other methods are supplemented. Secondly, it summarizes the discussion and analysis on whether the graph neural network can be bigger and deeper. Furthermore, four frameworks of graph neural network are summarized. It also explains in detail the application of graph neural network in natural language processing and computer vision, etc. Finally, the future research of graph neural network is prospected and summarized. Compared with existing review articles on graph neural network, it elaborates the knowledge of spectral theory in detail, and comprehensively summarizes the development history of graph convolutional neural network based on the spectral domain. At the same time, a new classification standard, an improved model for the low efficiency of the spatial domain graph convolutional neural network, is given. And for the first time, it summarizes the discussion and analysis of graph neural network expression ability, theoretical guarantee, etc., and adds a new framework model. In the application part, the latest application of graph neural network is explained.
In recent years, the application of deep learning related to graph structure data has attracted more and more attention. The emergence of graph neural network has made major breakthroughs in the above tasks, such as social networking, natural language processing, computer vision, even life sciences and other fields. The graph neural network can treat the actual problem as the connection between nodes in the graph and the message propagation problem, and the dependence between nodes can be modeled, so that the graph structure data can be handled well. In view of this, the graph neural network model and its application are systematically reviewed. Firstly, the graph convolutional neural network is explained from three aspects: spectral domain, spatial domain and pooling. Then, the graph neural network model based on the attention mechanism and autoencoder is described, and some graph neural network implemented by other methods are supplemented. Secondly, it summarizes the discussion and analysis on whether the graph neural network can be bigger and deeper. Furthermore, four frameworks of graph neural network are summarized. It also explains in detail the application of graph neural network in natural language processing and computer vision, etc. Finally, the future research of graph neural network is prospected and summarized. Compared with existing review articles on graph neural network, it elaborates the knowledge of spectral theory in detail, and comprehensively summarizes the development history of graph convolutional neural network based on the spectral domain. At the same time, a new classification standard, an improved model for the low efficiency of the spatial domain graph convolutional neural network, is given. And for the first time, it summarizes the discussion and analysis of graph neural network expression ability, theoretical guarantee, etc., and adds a new framework model. In the application part, the latest application of graph neural network is explained.
7
2018, 55(12): 2587-2599.
DOI: 10.7544/issn1000-1239.2018.20180623
Abstract：
With the advent of the medical big data era, knowledge interconnection has received extensive attention. How to extract useful medical knowledge from massive data is the key for medical big data analysis. Knowledge graph technology provides a means to extract structured knowledge from massive texts and images.The combination of knowledge graph, big data technology and deep learning technology is becoming the core driving force for the development of artificial intelligence. The knowledge graph technology has a broad application prospect in the medical domain. The application of knowledge graph technology in the medical domain will play an important role in solving the contradiction between the supply of high-quality medical resources and the continuous increase of demand for medical services.At present, the research on medical knowledge graph is still in the exploratory stage. The existing knowledge graph technology generally has several problems such as low efficiency, multiple restrictions and poor expansion in the medical domain. This paper firstly analyzes the medical knowledge graph architecture and construction technology for the strong professionalism and complex structure of big data in the medical domain. Secondly, the key technologies and research progress of the three modules of knowledge extraction, knowledge expression, knowledge fusion and knowledge reasoning in medical knowledge map are summarized. In addition, the application status of medical knowledge maps in clinical decision support, medical intelligence semantic retrieval, medical question answering system and other medical services are introduced. Finally, the existing problems and challenges of current research are discussed and analyzed, and its development is prospected.
With the advent of the medical big data era, knowledge interconnection has received extensive attention. How to extract useful medical knowledge from massive data is the key for medical big data analysis. Knowledge graph technology provides a means to extract structured knowledge from massive texts and images.The combination of knowledge graph, big data technology and deep learning technology is becoming the core driving force for the development of artificial intelligence. The knowledge graph technology has a broad application prospect in the medical domain. The application of knowledge graph technology in the medical domain will play an important role in solving the contradiction between the supply of high-quality medical resources and the continuous increase of demand for medical services.At present, the research on medical knowledge graph is still in the exploratory stage. The existing knowledge graph technology generally has several problems such as low efficiency, multiple restrictions and poor expansion in the medical domain. This paper firstly analyzes the medical knowledge graph architecture and construction technology for the strong professionalism and complex structure of big data in the medical domain. Secondly, the key technologies and research progress of the three modules of knowledge extraction, knowledge expression, knowledge fusion and knowledge reasoning in medical knowledge map are summarized. In addition, the application status of medical knowledge maps in clinical decision support, medical intelligence semantic retrieval, medical question answering system and other medical services are introduced. Finally, the existing problems and challenges of current research are discussed and analyzed, and its development is prospected.
8
2020, 57(5): 1103-1122.
DOI: 10.7544/issn1000-1239.2020.20190460
Abstract：
With the rapid development of cloud computing and Internet of things, users’ demand for software systems tends to be diversified. Service oriented architecture (SOA) needs to strike a balance between stable service integration and flexible adaptation of requirements. Based on this situation, the microservice technology, which goes with independent process as well as independent deployment capability, emerges as the times require. It has a slew of advantages, such as distributed storage, high availability, scalability, and intelligent operation maintenance, which can make up for the shortcomings of the traditional SOA architecture. From the perspective of system integration, the paper firstly describes the application background of microservice, which include the core components of microservice, software technology development and architecture evolution to ensure the availability of microservice infrastructure. Secondly, in view of problems existing in practical applications, the paper analyzes the key technologies utilized in the specific application of the microservice architecture through the aspects of distributed communication, distributed data storage, distributed call chain, and testing complexity; then, a specific application case is given to confirm the technical feasibility of microservice. Finally, this paper intends to explore the challenges by microservice through the aspects of infrastructure, information exchange, data security, and network security. Meanwhile, the future development trend is analyzed so as to provide valuable theoretical and technical reference for the future innovation and development of microservice.
With the rapid development of cloud computing and Internet of things, users’ demand for software systems tends to be diversified. Service oriented architecture (SOA) needs to strike a balance between stable service integration and flexible adaptation of requirements. Based on this situation, the microservice technology, which goes with independent process as well as independent deployment capability, emerges as the times require. It has a slew of advantages, such as distributed storage, high availability, scalability, and intelligent operation maintenance, which can make up for the shortcomings of the traditional SOA architecture. From the perspective of system integration, the paper firstly describes the application background of microservice, which include the core components of microservice, software technology development and architecture evolution to ensure the availability of microservice infrastructure. Secondly, in view of problems existing in practical applications, the paper analyzes the key technologies utilized in the specific application of the microservice architecture through the aspects of distributed communication, distributed data storage, distributed call chain, and testing complexity; then, a specific application case is given to confirm the technical feasibility of microservice. Finally, this paper intends to explore the challenges by microservice through the aspects of infrastructure, information exchange, data security, and network security. Meanwhile, the future development trend is analyzed so as to provide valuable theoretical and technical reference for the future innovation and development of microservice.
9
2018, 55(1): 30-52.
DOI: 10.7544/issn1000-1239.2018.20170055
Abstract：
With the rapid development of social networks, electronic commerce, mobile Internet and other technologies, all kinds of Web data expand rapidly. There are a large number of emotional texts on the Internet, and they are very helpful to understand the netizen’s opinion and viewpoint if fully explored. The aim of emotion classification is to predict the emotion categories of emotive texts, which is the core of emotion analysis. In this paper, we first introduce the background knowledge of emotion analysis including different emotion classification systems and its application scenarios on public opinion management and control, business decisions, opinion search, information prediction, emotion management. Then we summarize the mainstream approaches of emotion classification, and make a detailed description and analysis on these approaches. Finally, we expound the problems of data sparsity, class imbalance learning, dependence for the strong domain knowledge and language imbalance existing in the emotion analysis work. The research progress of text emotion analysis is summarized and prospect combined with large data processing, the mixing of multiple media, deep learning development, mining on a specific topic and multilingual synergy.
With the rapid development of social networks, electronic commerce, mobile Internet and other technologies, all kinds of Web data expand rapidly. There are a large number of emotional texts on the Internet, and they are very helpful to understand the netizen’s opinion and viewpoint if fully explored. The aim of emotion classification is to predict the emotion categories of emotive texts, which is the core of emotion analysis. In this paper, we first introduce the background knowledge of emotion analysis including different emotion classification systems and its application scenarios on public opinion management and control, business decisions, opinion search, information prediction, emotion management. Then we summarize the mainstream approaches of emotion classification, and make a detailed description and analysis on these approaches. Finally, we expound the problems of data sparsity, class imbalance learning, dependence for the strong domain knowledge and language imbalance existing in the emotion analysis work. The research progress of text emotion analysis is summarized and prospect combined with large data processing, the mixing of multiple media, deep learning development, mining on a specific topic and multilingual synergy.
10
2019, 56(2): 319-327.
DOI: 10.7544/issn1000-1239.2019.20170749
Abstract：
Normally, small object is the object which only covers a small part of a whole image. Compared with regular object, small object has less information and the training data of small object is difficult to be marked. This leads to the poor performance when directly employing the previous object detection methods for small object detection. Moreover, the detection methods designed for small object are often too complex or not generic. In this paper, we propose a small object detection algorithm named multi-scale Faster-RCNN. According to the characteristics of convolutional neural network, the structure of Faster-RCNN is modified, such that the network can integrate both the low-level and high-level features for multi-scale object detection. Through such a manner, the accuracy of small object detection is improved. Simultaneously, with the goal of solving the problem that training data is difficult to be marked, we use training data crawled from search engine to train the model. Because the distribution of crawled data is different from the real test data’s, training images in which objects have high resolution are transformed by means of down sampling and up sampling. It makes the feature distribution of training images and test images more similar. The experiment results show that the mean average precision (mAP) of proposed approach can be up to 5% higher than the original Faster-RCNN’s in the task of small object detection.
Normally, small object is the object which only covers a small part of a whole image. Compared with regular object, small object has less information and the training data of small object is difficult to be marked. This leads to the poor performance when directly employing the previous object detection methods for small object detection. Moreover, the detection methods designed for small object are often too complex or not generic. In this paper, we propose a small object detection algorithm named multi-scale Faster-RCNN. According to the characteristics of convolutional neural network, the structure of Faster-RCNN is modified, such that the network can integrate both the low-level and high-level features for multi-scale object detection. Through such a manner, the accuracy of small object detection is improved. Simultaneously, with the goal of solving the problem that training data is difficult to be marked, we use training data crawled from search engine to train the model. Because the distribution of crawled data is different from the real test data’s, training images in which objects have high resolution are transformed by means of down sampling and up sampling. It makes the feature distribution of training images and test images more similar. The experiment results show that the mean average precision (mAP) of proposed approach can be up to 5% higher than the original Faster-RCNN’s in the task of small object detection.
11
2018, 55(9): 1871-1888.
DOI: 10.7544/issn1000-1239.2018.20180129
Abstract：
In recent years, deep neural networks (DNNs) have achieved remarkable success in many artificial intelligence (AI) applications, including computer vision, speech recognition and natural language processing. However, such DNNs have been accompanied by significant increase in computational costs and storage services, which prohibits the usages of DNNs on resource-limited environments such as mobile or embedded devices. To this end, the studies of DNN compression and acceleration have recently become more emerging. In this paper, we provide a review on the existing representative DNN compression and acceleration methods, including parameter pruning, parameter sharing, low-rank decomposition, compact filter designed, and knowledge distillation. Specifically, this paper provides an overview of DNNs, describes the details of different DNN compression and acceleration methods, and highlights the properties, advantages and drawbacks. Furthermore, we summarize the evaluation criteria and datasets widely used in DNN compression and acceleration, and also discuss the performance of the representative methods. In the end, we discuss how to choose different compression and acceleration methods to meet the needs of different tasks, and envision future directions on this topic.
In recent years, deep neural networks (DNNs) have achieved remarkable success in many artificial intelligence (AI) applications, including computer vision, speech recognition and natural language processing. However, such DNNs have been accompanied by significant increase in computational costs and storage services, which prohibits the usages of DNNs on resource-limited environments such as mobile or embedded devices. To this end, the studies of DNN compression and acceleration have recently become more emerging. In this paper, we provide a review on the existing representative DNN compression and acceleration methods, including parameter pruning, parameter sharing, low-rank decomposition, compact filter designed, and knowledge distillation. Specifically, this paper provides an overview of DNNs, describes the details of different DNN compression and acceleration methods, and highlights the properties, advantages and drawbacks. Furthermore, we summarize the evaluation criteria and datasets widely used in DNN compression and acceleration, and also discuss the performance of the representative methods. In the end, we discuss how to choose different compression and acceleration methods to meet the needs of different tasks, and envision future directions on this topic.
12
2018, 55(5): 945-957.
DOI: 10.7544/issn1000-1239.2018.20170049
Abstract：
Neural network-based architectures have been pervasively applied to sentiment analysis and achieved great success in recent years. However, most previous approaches usually classified with word feature only, which ignoring some characteristic features on the task of sentiment classification. One of the remaining challenges is to leverage the sentiment resources effectively because of the lack of length of Chinese micro-blog texts. To address this problem, we propose a novel sentiment classification method for Chinese micro-blog sentiment analysis based on multi-channels convolutional neural networks (MCCNN) to capture the characteristic information in micro-blog texts. With the help of the part of speech vector, the model could promote the full use of sentiment features through different part of speech tagging. Meanwhile, the position vector helps the model indicate the degree of importance of every word in the sentence, which impels the model to focus on the important words in the training process. Afterwards, a multi-channels architecture based on convolutional neural networks will be used to learn more feature information of micro-blog texts, and extract more hidden information through combining different vectors and original word embedding. Finally, the experiments on COAE2014 dataset and micro-blog dataset reveal better performance than the current main stream convolutional neural networks and traditional classifier.
Neural network-based architectures have been pervasively applied to sentiment analysis and achieved great success in recent years. However, most previous approaches usually classified with word feature only, which ignoring some characteristic features on the task of sentiment classification. One of the remaining challenges is to leverage the sentiment resources effectively because of the lack of length of Chinese micro-blog texts. To address this problem, we propose a novel sentiment classification method for Chinese micro-blog sentiment analysis based on multi-channels convolutional neural networks (MCCNN) to capture the characteristic information in micro-blog texts. With the help of the part of speech vector, the model could promote the full use of sentiment features through different part of speech tagging. Meanwhile, the position vector helps the model indicate the degree of importance of every word in the sentence, which impels the model to focus on the important words in the training process. Afterwards, a multi-channels architecture based on convolutional neural networks will be used to learn more feature information of micro-blog texts, and extract more hidden information through combining different vectors and original word embedding. Finally, the experiments on COAE2014 dataset and micro-blog dataset reveal better performance than the current main stream convolutional neural networks and traditional classifier.
13
2020, 57(2): 346-362.
DOI: 10.7544/issn1000-1239.2020.20190455
Abstract：
Large-scale data collection has vastly improved the performance of machine learning, and achieved a win-win situation for both economic and social benefits, while personal privacy preservation is facing new and greater risks and crises. In this paper, we summarize the privacy issues in machine learning and the existing work on privacy-preserving machine learning. We respectively discuss two settings of the model training process—centralized learning and federated learning. The former needs to collect all the user data before training. Although this setting is easy to deploy, it still exists enormous privacy and security hidden troubles. The latter achieves that massive devices can collaborate to train a global model while keeping their data in local. As it is currently in the early stage of the study, it also has many problems to be solved. The existing work on privacy-preserving techniques can be concluded into two main clues—the encryption method including homomorphic encryption and secure multi-party computing and the perturbation method represented by differential privacy, each having its advantages and disadvantages. In this paper, we first focus on the design of differentially-private machine learning algorithm, especially under centralized setting, and discuss the differences between traditional machine learning models and deep learning models. Then, we summarize the problems existing in the current federated learning study. Finally, we propose the main challenges in the future work and point out the connection among privacy protection, model interpretation and data transparency.
Large-scale data collection has vastly improved the performance of machine learning, and achieved a win-win situation for both economic and social benefits, while personal privacy preservation is facing new and greater risks and crises. In this paper, we summarize the privacy issues in machine learning and the existing work on privacy-preserving machine learning. We respectively discuss two settings of the model training process—centralized learning and federated learning. The former needs to collect all the user data before training. Although this setting is easy to deploy, it still exists enormous privacy and security hidden troubles. The latter achieves that massive devices can collaborate to train a global model while keeping their data in local. As it is currently in the early stage of the study, it also has many problems to be solved. The existing work on privacy-preserving techniques can be concluded into two main clues—the encryption method including homomorphic encryption and secure multi-party computing and the perturbation method represented by differential privacy, each having its advantages and disadvantages. In this paper, we first focus on the design of differentially-private machine learning algorithm, especially under centralized setting, and discuss the differences between traditional machine learning models and deep learning models. Then, we summarize the problems existing in the current federated learning study. Finally, we propose the main challenges in the future work and point out the connection among privacy protection, model interpretation and data transparency.
14
2018, 55(9): 1853-1870.
DOI: 10.7544/issn1000-1239.2018.20180127
Abstract：
Blockchain technology is a new emerging technology that has the potential to revolutionize many traditional industries. Since the creation of Bitcoin, which represents blockchain 1.0, blockchain technology has been attracting extensive attention and a great amount of user transaction data has been accumulated. Furthermore, the birth of Ethereum, which represents blockchain 2.0, further enriches data type in blockchain. While the popularity of blockchain technology bringing about a lot of technical innovation, it also leads to many new problems, such as user privacy disclosure and illegal financial activities. However, the public accessible of blockchain data provides unprecedented opportunity for researchers to understand and resolve these problems through blockchain data analysis. Thus, it is of great significance to summarize the existing research problems, the results obtained, the possible research trends, and the challenges faced in blockchain data analysis. To this end, a comprehensive review and summary of the progress of blockchain data analysis is presented. The review begins by introducing the architecture and key techniques of blockchain technology and providing the main data types in blockchain with the corresponding analysis methods. Then, the current research progress in blockchain data analysis is summarized in seven research problems, which includes entity recognition, privacy disclosure risk analysis, network portrait, network visualization, market effect analysis, transaction pattern recognition, illegal behavior detection and analysis. Finally, the directions, prospects and challenges for future research are explored based on the shortcomings of current research.
Blockchain technology is a new emerging technology that has the potential to revolutionize many traditional industries. Since the creation of Bitcoin, which represents blockchain 1.0, blockchain technology has been attracting extensive attention and a great amount of user transaction data has been accumulated. Furthermore, the birth of Ethereum, which represents blockchain 2.0, further enriches data type in blockchain. While the popularity of blockchain technology bringing about a lot of technical innovation, it also leads to many new problems, such as user privacy disclosure and illegal financial activities. However, the public accessible of blockchain data provides unprecedented opportunity for researchers to understand and resolve these problems through blockchain data analysis. Thus, it is of great significance to summarize the existing research problems, the results obtained, the possible research trends, and the challenges faced in blockchain data analysis. To this end, a comprehensive review and summary of the progress of blockchain data analysis is presented. The review begins by introducing the architecture and key techniques of blockchain technology and providing the main data types in blockchain with the corresponding analysis methods. Then, the current research progress in blockchain data analysis is summarized in seven research problems, which includes entity recognition, privacy disclosure risk analysis, network portrait, network visualization, market effect analysis, transaction pattern recognition, illegal behavior detection and analysis. Finally, the directions, prospects and challenges for future research are explored based on the shortcomings of current research.
15
2018, 55(3): 487-511.
DOI: 10.7544/issn1000-1239.2018.20170778
Abstract：
Edge computing is a new network architecture and open platform that integrates network, computing, storage, and application core capabilities on the edge of the network. Edge computing changes the way of traditional centralized cloud computing which moves computing and storage capabilities to the edge of the network. Edge computing can greatly reduce the congestion and burden of core network and transmission network, due to canceling the data backhaul. And it can lower the delay, bring high bandwidth. Also at the same time it can quickly respond to users’ requests and improve service quality. Now, edge computing has become an important enabling technology for the future 5G, and it has been written into 3GPP standard. More and more operators, equipment vendors and chip manufacturers join to construct the edge computing ecological. How to build a unified standardized edge computing platform for future ecological construction is very important. Therefore, this paper focuses on the progress of standardization of the current edge computing. From edge computing architecture was first proposed by ETSI, to edge computing has been listed as the key technology for future 5G development in 3GPP. The approval of projects of the edge computing in CCSA is also included. The introduction of each part has a great deal of analysis and explanation of the standard content. In the end, this paper introduces China Unicom’s edge computing research achievement in recent years, including the important edge computing experimental project, the pilot scheme for future edge computing of China Unicom and the exploration of the network deployment plan of edge computing. We are looking forward to discussing the edge computing commercial cooperation mode with all sectors so as to jointly build the network edge ecology and comprehensively accelerate the vigorous development of 5G services.
Edge computing is a new network architecture and open platform that integrates network, computing, storage, and application core capabilities on the edge of the network. Edge computing changes the way of traditional centralized cloud computing which moves computing and storage capabilities to the edge of the network. Edge computing can greatly reduce the congestion and burden of core network and transmission network, due to canceling the data backhaul. And it can lower the delay, bring high bandwidth. Also at the same time it can quickly respond to users’ requests and improve service quality. Now, edge computing has become an important enabling technology for the future 5G, and it has been written into 3GPP standard. More and more operators, equipment vendors and chip manufacturers join to construct the edge computing ecological. How to build a unified standardized edge computing platform for future ecological construction is very important. Therefore, this paper focuses on the progress of standardization of the current edge computing. From edge computing architecture was first proposed by ETSI, to edge computing has been listed as the key technology for future 5G development in 3GPP. The approval of projects of the edge computing in CCSA is also included. The introduction of each part has a great deal of analysis and explanation of the standard content. In the end, this paper introduces China Unicom’s edge computing research achievement in recent years, including the important edge computing experimental project, the pilot scheme for future edge computing of China Unicom and the exploration of the network deployment plan of edge computing. We are looking forward to discussing the edge computing commercial cooperation mode with all sectors so as to jointly build the network edge ecology and comprehensively accelerate the vigorous development of 5G services.
16
2020, 57(6): 1208-1217.
DOI: 10.7544/issn1000-1239.2020.20190485
Abstract：
The research on the interpretability of deep learning is closely related to various disciplines such as artificial intelligence, machine learning, logic and cognitive psychology. It has important theoretical research significance and practical application value in too many fields, such as information push, medical research, finance, and information security. In the past few years, there were a lot of well studied work in this field, but we are still facing various issues. In this paper, we clearly review the history of deep learning interpretability research and related work. Firstly, we introduce the history of interpretable deep learning from following three aspects: origin of interpretable deep learning, research exploration stage and model construction stage. Then, the research situation is presented from three aspects, namely visual analysis, robust perturbation analysis and sensitivity analysis. The research on the construction of interpretable deep learning model is introduced following four aspects: model agent, logical reasoning, network node association analysis and traditional machine learning model. Moreover, the limitations of current research are analyzed and discussed in this paper. At last, we list the typical applications of the interpretable deep learning and forecast the possible future research directions of this field along with reasonable and suitable suggestions.
The research on the interpretability of deep learning is closely related to various disciplines such as artificial intelligence, machine learning, logic and cognitive psychology. It has important theoretical research significance and practical application value in too many fields, such as information push, medical research, finance, and information security. In the past few years, there were a lot of well studied work in this field, but we are still facing various issues. In this paper, we clearly review the history of deep learning interpretability research and related work. Firstly, we introduce the history of interpretable deep learning from following three aspects: origin of interpretable deep learning, research exploration stage and model construction stage. Then, the research situation is presented from three aspects, namely visual analysis, robust perturbation analysis and sensitivity analysis. The research on the construction of interpretable deep learning model is introduced following four aspects: model agent, logical reasoning, network node association analysis and traditional machine learning model. Moreover, the limitations of current research are analyzed and discussed in this paper. At last, we list the typical applications of the interpretable deep learning and forecast the possible future research directions of this field along with reasonable and suitable suggestions.
17
2021, 58(1): 1-21.
DOI: 10.7544/issn1000-1239.2021.20190785
Abstract：
In recent years, the rapid development of Internet technology has greatly facilitated the daily life of human, and it is inevitable that massive information erupts in a blowout. How to quickly and effectively obtain the required information on the Internet is an urgent problem. The automatic text summarization technology can effectively alleviate this problem. As one of the most important fields in natural language processing and artificial intelligence, it can automatically produce a concise and coherent summary from a long text or text set through computer, in which the summary should accurately reflect the central themes of source text. In this paper, we expound the connotation of automatic summarization, review the development of automatic text summarization technique and introduce two main techniques in detail: extractive and abstractive summarization, including feature scoring, classification method, linear programming, submodular function, graph ranking, sequence labeling, heuristic algorithm, deep learning, etc. We also analyze the datasets and evaluation metrics that are commonly used in automatic summarization. Finally, the challenges ahead and the future trends of research and application have been predicted.
In recent years, the rapid development of Internet technology has greatly facilitated the daily life of human, and it is inevitable that massive information erupts in a blowout. How to quickly and effectively obtain the required information on the Internet is an urgent problem. The automatic text summarization technology can effectively alleviate this problem. As one of the most important fields in natural language processing and artificial intelligence, it can automatically produce a concise and coherent summary from a long text or text set through computer, in which the summary should accurately reflect the central themes of source text. In this paper, we expound the connotation of automatic summarization, review the development of automatic text summarization technique and introduce two main techniques in detail: extractive and abstractive summarization, including feature scoring, classification method, linear programming, submodular function, graph ranking, sequence labeling, heuristic algorithm, deep learning, etc. We also analyze the datasets and evaluation metrics that are commonly used in automatic summarization. Finally, the challenges ahead and the future trends of research and application have been predicted.
18
2022, 59(9): 1947-1965.
DOI: 10.7544/issn1000-1239.20210829
Abstract：
With the wave of the past decade, the development of artificial intelligence is in the critical period from perceptual intelligence to cognitive intelligence. Knowledge graph, as the core technique of knowledge engineering in the era of big data, is the combination of symbolism and connectionism, and is the cornerstone of realizing cognitive intelligence. It provides an effective solution for the knowledge organization and intelligent application in the Internet era. In recent years, some progress has been made in the key technologies and theories of knowledge graph, and typical applications of knowledge graph based on information system have gradually entered various industries, including intelligent question answering, recommendation system, personal assistant, etc. However, in the context of big data environment and new infrastructure of China, the increasing multi-modal data and new interaction ways have raised new demands and brought new challenges to the new generation of knowledge graph in terms of basic theory, architecture, and key technologies. We summarize the research and development status of key technologies of the new generation knowledge graph at home and abroad, including unstructured multi-modal data organization and understanding, large-scale dynamic knowledge graph representation learning and pre-training models, and neural-symbolic knowledge inference. We summarize, compare and analyze the latest research progress. Finally, the future technical challenges and research directions are prospected.
With the wave of the past decade, the development of artificial intelligence is in the critical period from perceptual intelligence to cognitive intelligence. Knowledge graph, as the core technique of knowledge engineering in the era of big data, is the combination of symbolism and connectionism, and is the cornerstone of realizing cognitive intelligence. It provides an effective solution for the knowledge organization and intelligent application in the Internet era. In recent years, some progress has been made in the key technologies and theories of knowledge graph, and typical applications of knowledge graph based on information system have gradually entered various industries, including intelligent question answering, recommendation system, personal assistant, etc. However, in the context of big data environment and new infrastructure of China, the increasing multi-modal data and new interaction ways have raised new demands and brought new challenges to the new generation of knowledge graph in terms of basic theory, architecture, and key technologies. We summarize the research and development status of key technologies of the new generation knowledge graph at home and abroad, including unstructured multi-modal data organization and understanding, large-scale dynamic knowledge graph representation learning and pre-training models, and neural-symbolic knowledge inference. We summarize, compare and analyze the latest research progress. Finally, the future technical challenges and research directions are prospected.
19
2018, 55(3): 478-486.
DOI: 10.7544/issn1000-1239.2018.20170801
Abstract：
Future 5G wireless networks are confronted with various challenges such as exponentially increasing mobile traffic and new services requiring high backhaul bandwidth and low latency. Integrating mobile edge computing (MEC) into 5G network architectures may be a promising solution. First of all, this paper introduces the functional framework of MEC systems. Then the standardization progress of MEC in 5G is presented. Supporting MEC, the functionalities of 5G core network are described in detail. Given MEC deployment strategies and the mobile network architectures of future 5G, a MEC coordinated 5G network architecture is proposed, which demonstrates that 5G will be a network featured by the coordination of communications and multi-level computing. The proposed network architecture can support various communication modes adaptively and enable the resource sharing efficiently with virtualization technologies. Some researches have been carried out on MEC coordinated 5G, such as basic theorems related to the 5G network capacity concerning both the communication and computing resources, and key technologies including the joint optimization of communication and computing resources, multicast based on computing and cache, and bandwidth-saving transmission. It can be seen that much more efforts need to be put on MEC coordinated 5G before the network can be fully understood.
Future 5G wireless networks are confronted with various challenges such as exponentially increasing mobile traffic and new services requiring high backhaul bandwidth and low latency. Integrating mobile edge computing (MEC) into 5G network architectures may be a promising solution. First of all, this paper introduces the functional framework of MEC systems. Then the standardization progress of MEC in 5G is presented. Supporting MEC, the functionalities of 5G core network are described in detail. Given MEC deployment strategies and the mobile network architectures of future 5G, a MEC coordinated 5G network architecture is proposed, which demonstrates that 5G will be a network featured by the coordination of communications and multi-level computing. The proposed network architecture can support various communication modes adaptively and enable the resource sharing efficiently with virtualization technologies. Some researches have been carried out on MEC coordinated 5G, such as basic theorems related to the 5G network capacity concerning both the communication and computing resources, and key technologies including the joint optimization of communication and computing resources, multicast based on computing and cache, and bandwidth-saving transmission. It can be seen that much more efforts need to be put on MEC coordinated 5G before the network can be fully understood.
20
2019, 56(3): 566-575.
DOI: 10.7544/issn1000-1239.2019.20180063
Abstract：
Intrusion detection system can efficiently detect attack behaviors, which will do great damage for network security. Currently many intrusion detection systems have low detection rates in these abnormal behaviors Probe (probing), U2R (user to root) and R2L (remote to local). Focusing on this weakness, a new hybrid multi-level intrusion detection method is proposed to identify network data as normal or abnormal behaviors. This method contains KNN (K nearest neighbors) outlier detection algorithm and multi-level random forests (RF) model, called KNN-RF. Firstly KNN outlier detection algorithm is applied to detect and delete outliers in each category and get a small high-quality training dataset. Then according to the similarity of network traffic, a new method of the division of data categories is put forward and this division method can avoid the mutual interference of anomaly behaviors in the detection process, especially for the detecting of the attack behaviors of small traffic. Based on this division, a multi-level random forests model is constructed to detect network abnormal behaviors and improve the efficiency of detecting known and unknown attacks. The popular KDD (knowledge discovery and data mining) Cup 1999 dataset is used to evaluate the performance of the proposed method. Compared with other algorithms, the proposed method is significantly superior to other algorithms in accuracy and detection rate, and can detect Probe, U2R and R2L effectively.
Intrusion detection system can efficiently detect attack behaviors, which will do great damage for network security. Currently many intrusion detection systems have low detection rates in these abnormal behaviors Probe (probing), U2R (user to root) and R2L (remote to local). Focusing on this weakness, a new hybrid multi-level intrusion detection method is proposed to identify network data as normal or abnormal behaviors. This method contains KNN (K nearest neighbors) outlier detection algorithm and multi-level random forests (RF) model, called KNN-RF. Firstly KNN outlier detection algorithm is applied to detect and delete outliers in each category and get a small high-quality training dataset. Then according to the similarity of network traffic, a new method of the division of data categories is put forward and this division method can avoid the mutual interference of anomaly behaviors in the detection process, especially for the detecting of the attack behaviors of small traffic. Based on this division, a multi-level random forests model is constructed to detect network abnormal behaviors and improve the efficiency of detecting known and unknown attacks. The popular KDD (knowledge discovery and data mining) Cup 1999 dataset is used to evaluate the performance of the proposed method. Compared with other algorithms, the proposed method is significantly superior to other algorithms in accuracy and detection rate, and can detect Probe, U2R and R2L effectively.
- First
- Prev
- 1
- 2
- 3
- 4
- 5
- Next
- Last
- Total 5 Pages
- To
- Go
