To improve data distribution efficiency a load-balancing data distribution LBDD method is proposed in publish/subscribe mode.In the LBDD method subscribers are involved in distribution tasks and data transfers while r...To improve data distribution efficiency a load-balancing data distribution LBDD method is proposed in publish/subscribe mode.In the LBDD method subscribers are involved in distribution tasks and data transfers while receiving data themselves.A dissemination tree is constructed among the subscribers based on MD5 where the publisher acts as the root. The proposed method provides bucket construction target selection and path updates furthermore the property of one-way dissemination is proven.That the average out-going degree of a node is 2 is guaranteed with the proposed LBDD.The experiments on data distribution delay data distribution rate and load distribution are conducted. Experimental results show that the LBDD method aids in shaping the task load between the publisher and subscribers and outperforms the point-to-point approach.展开更多
Comparing the city-size distribution at the urban agglomeration（UA） scale is important for understanding the processes of urban development. However, comparative studies of city-size distribution among China＇s thre...Comparing the city-size distribution at the urban agglomeration（UA） scale is important for understanding the processes of urban development. However, comparative studies of city-size distribution among China＇s three largest UAs, the Beijing-Tianjin-Hebei agglomeration（BTHA）, the Yangtze River Delta agglomeration（YRDA）, and the Pearl River Delta agglomeration（PRDA）, remain inadequate due to the limitation of data availability. Therefore, using urban data derived from time-series nighttime light data, the common characteristics and distinctive features of city-size distribution among the three UAs from 1992 to 2015 were compared by the Pareto regression and the rank clock method. We identified two common features. First, the city-size distribution became more even. The Pareto exponents increased by 0.17, 0.12, and 0.01 in the YRDA, BTHA, and PRDA, respectively. Second, the average ranks of small cities ascended, being 0.55, 0.08 and 0.04 in the three UAs, respectively. However, the average ranks of large and medium cities in the three UAs experienced different trajectories, which are closely related to the similarities and differences in the driving forces for the development of UAs. Place-based measures are encouraged to promote a coordinated development among cities of differing sizes in the three UAs.展开更多
We analyze co-seismic displacement field of the 26 December 2004, giant Sumatra–Andaman earthquake derived from Global Position System observations,geological vertical measurement of coral head, and pivot line observ...We analyze co-seismic displacement field of the 26 December 2004, giant Sumatra–Andaman earthquake derived from Global Position System observations,geological vertical measurement of coral head, and pivot line observed through remote sensing. Using the co-seismic displacement field and AK135 spherical layered Earth model, we invert co-seismic slip distribution along the seismic fault. We also search the best fault geometry model to fit the observed data. Assuming that the dip angle linearly increases in downward direction, the postfit residual variation of the inversed geometry model with dip angles linearly changing along fault strike are plotted. The geometry model with local minimum misfits is the one with dip angle linearly increasing along strike from 4.3oin top southernmost patch to 4.5oin top northernmost path and dip angle linearly increased. By using the fault shape and geodetic co-seismic data, we estimate the slip distribution on the curved fault. Our result shows that the earthquake ruptured ＊200-km width down to a depth of about 60 km.0.5–12.5 m of thrust slip is resolved with the largest slip centered around the central section of the rupture zone78N–108N in latitude. The estimated seismic moment is8.2 9 1022 N m, which is larger than estimation from the centroid moment magnitude（4.0 9 1022 N m）, and smaller than estimation from normal-mode oscillation data modeling（1.0 9 1023 N m）.展开更多
1 Introduction Geochemical mapping at national and continental scales continues to present challenges worldwide due to variations in geologic and geotectonic units.Use of the proper sampling media can provide rich inf...1 Introduction Geochemical mapping at national and continental scales continues to present challenges worldwide due to variations in geologic and geotectonic units.Use of the proper sampling media can provide rich information on展开更多
In non-independent and identically distributed(non-IID)data environments,model performance often degrades significantly.To address this issue,two improvement methods are proposed:FedReg and FedReg^(*).FedReg is a meth...In non-independent and identically distributed(non-IID)data environments,model performance often degrades significantly.To address this issue,two improvement methods are proposed:FedReg and FedReg^(*).FedReg is a method based on hybrid regularization aimed at enhancing federated learning in non-IID scenarios.It introduces hybrid regularization to replace traditional L2 regularization,combining the advantages of L1 and L2 regularization to enable feature selection while preventing overfitting.This method better adapts to the diverse data distributions of different clients,improving the overall model performance.FedReg^(*)combines hybrid regularization with weighted model aggregation.In addition to the benefits of hybrid regularization,FedReg^(*)applies a weighted averaging method in the model aggregation process,calculating weights based on the cosine similarity between each client gradient and the global gradient to more reasonably distribute client contributions.By considering variations in data quality and quantity among clients,FedReg^(*)highlights the importance of key clients and enhances the model’s generalization performance.These improvement methods enhance model accuracy and communication efficiency.展开更多
Due to the limited scenes that synthetic aperture radar(SAR)satellites can detect,the full-track utilization rate is not high.Because of the computing and storage limitation of one satellite,it is difficult to process...Due to the limited scenes that synthetic aperture radar(SAR)satellites can detect,the full-track utilization rate is not high.Because of the computing and storage limitation of one satellite,it is difficult to process large amounts of data of spaceborne synthetic aperture radars.It is proposed to use a new method of networked satellite data processing for improving the efficiency of data processing.A multi-satellite distributed SAR real-time processing method based on Chirp Scaling(CS)imaging algorithm is studied in this paper,and a distributed data processing system is built with field programmable gate array(FPGA)chips as the kernel.Different from the traditional CS algorithm processing,the system divides data processing into three stages.The computing tasks are reasonably allocated to different data processing units(i.e.,satellites)in each stage.The method effectively saves computing and storage resources of satellites,improves the utilization rate of a single satellite,and shortens the data processing time.Gaofen-3(GF-3)satellite SAR raw data is processed by the system,with the performance of the method verified.展开更多
Privacy is a critical requirement in distributed data mining. Cryptography-based secure multiparty computation is a main approach for privacy preserving. However, it shows poor performance in large scale distributed s...Privacy is a critical requirement in distributed data mining. Cryptography-based secure multiparty computation is a main approach for privacy preserving. However, it shows poor performance in large scale distributed systems. Meanwhile, data perturbation techniques are comparatively efficient but are mainly used in centralized privacy-preserving data mining (PPDM). In this paper, we propose a light-weight anonymous data perturbation method for efficient privacy preserving in distributed data mining. We first define the privacy constraints for data perturbation based PPDM in a semi-honest distributed environment. Two protocols are proposed to address these constraints and protect data statistics and the randomization process against collusion attacks: the adaptive privacy-preserving summary protocol and the anonymous exchange protocol. Finally, a distributed data perturbation framework based on these protocols is proposed to realize distributed PPDM. Experiment results show that our approach achieves a high security level and is very efficient in a large scale distributed environment.展开更多
In wastewater treatment process(WWTP), the accurate and real-time monitoring values of key variables are crucial for the operational strategies. However, most of the existing methods have difficulty in obtaining the r...In wastewater treatment process(WWTP), the accurate and real-time monitoring values of key variables are crucial for the operational strategies. However, most of the existing methods have difficulty in obtaining the real-time values of some key variables in the process. In order to handle this issue, a data-driven intelligent monitoring system, using the soft sensor technique and data distribution service, is developed to monitor the concentrations of effluent total phosphorous(TP) and ammonia nitrogen(NH_4-N). In this intelligent monitoring system, a fuzzy neural network(FNN) is applied for designing the soft sensor model, and a principal component analysis(PCA) method is used to select the input variables of the soft sensor model. Moreover, data transfer software is exploited to insert the soft sensor technique to the supervisory control and data acquisition(SCADA) system. Finally, this proposed intelligent monitoring system is tested in several real plants to demonstrate the reliability and effectiveness of the monitoring performance.展开更多
It is difficult to parallelize a subsistent sequential algorithm. Through analyzing the sequential algorithm of a Global Atmospheric Data Objective Analysis System, this article puts forward a distributed parallel alg...It is difficult to parallelize a subsistent sequential algorithm. Through analyzing the sequential algorithm of a Global Atmospheric Data Objective Analysis System, this article puts forward a distributed parallel algorithm that statically distributes data on a massively parallel processing (MPP) computer. The algorithm realizes distributed parailelization by extracting the analysis boxes and model grid point Iatitude rows with leaped steps, and by distributing the data to different processors. The parallel algorithm achieves good load balancing, high parallel efficiency, and low parallel cost. Performance experiments on a MPP computer arc also presented.展开更多
Multimodal sentiment analysis utilizes multimodal data such as text,facial expressions and voice to detect people’s attitudes.With the advent of distributed data collection and annotation,we can easily obtain and sha...Multimodal sentiment analysis utilizes multimodal data such as text,facial expressions and voice to detect people’s attitudes.With the advent of distributed data collection and annotation,we can easily obtain and share such multimodal data.However,due to professional discrepancies among annotators and lax quality control,noisy labels might be introduced.Recent research suggests that deep neural networks(DNNs)will overfit noisy labels,leading to the poor performance of the DNNs.To address this challenging problem,we present a Multimodal Robust Meta Learning framework(MRML)for multimodal sentiment analysis to resist noisy labels and correlate distinct modalities simultaneously.Specifically,we propose a two-layer fusion net to deeply fuse different modalities and improve the quality of the multimodal data features for label correction and network training.Besides,a multiple meta-learner(label corrector)strategy is proposed to enhance the label correction approach and prevent models from overfitting to noisy labels.We conducted experiments on three popular multimodal datasets to verify the superiority of ourmethod by comparing it with four baselines.展开更多
Distributed Data Mining is expected to discover preciously unknown, implicit and valuable information from massive data set inherently distributed over a network. In recent years several approaches to distributed data...Distributed Data Mining is expected to discover preciously unknown, implicit and valuable information from massive data set inherently distributed over a network. In recent years several approaches to distributed data mining have been developed, but only a few of them make use of intelligent agents. This paper provides the reason for applying Multi-Agent Technology in Distributed Data Mining and presents a Distributed Data Mining System based on Multi-Agent Technology that deals with heterogeneity in such environment. Based on the advantages of both the CS model and agent-based model, the system is being able to address the specific concern of increasing scalability and enhancing performance.展开更多
It is crucial,while using healthcare data,to assess the advantages of data privacy against the possible drawbacks.Data from several sources must be combined for use in many data mining applications.The medical practit...It is crucial,while using healthcare data,to assess the advantages of data privacy against the possible drawbacks.Data from several sources must be combined for use in many data mining applications.The medical practitioner may use the results of association rule mining performed on this aggregated data to better personalize patient care and implement preventive measures.Historically,numerous heuristics(e.g.,greedy search)and metaheuristics-based techniques(e.g.,evolutionary algorithm)have been created for the positive association rule in privacy preserving data mining(PPDM).When it comes to connecting seemingly unrelated diseases and drugs,negative association rules may be more informative than their positive counterparts.It is well-known that during negative association rules mining,a large number of uninteresting rules are formed,making this a difficult problem to tackle.In this research,we offer an adaptive method for negative association rule mining in vertically partitioned healthcare datasets that respects users’privacy.The applied approach dynamically determines the transactions to be interrupted for information hiding,as opposed to predefining them.This study introduces a novel method for addressing the problem of negative association rules in healthcare data mining,one that is based on the Tabu-genetic optimization paradigm.Tabu search is advantageous since it removes a huge number of unnecessary rules and item sets.Experiments using benchmark healthcare datasets prove that the discussed scheme outperforms state-of-the-art solutions in terms of decreasing side effects and data distortions,as measured by the indicator of hiding failure.展开更多
HT-7 is the first superconducting tokamak device for fusion research in China. Many experiments have been done in the machine since 1994, and lots of satisfactory results have been achieved in the fusion research fiel...HT-7 is the first superconducting tokamak device for fusion research in China. Many experiments have been done in the machine since 1994, and lots of satisfactory results have been achieved in the fusion research field on HT-7 tokamak [1]. With the development of fusion research, remote control of experiment becomes more and more important to improve experimental efficiency and expand research results. This paper will describe a RCS (Remote Control System), the combined model of Browser/Server and Client/Server, based on Internet of HT-7 distributed data acquisition system (HT7DAS). By means of RCS, authorized users all over the world can control and configure HT7DAS remotely. The RCS is designed to improve the flexibility, opening, reliability and efficiency of HT7DAS. In the paper, the whole process of design along with implementation of the system and some key items are discussed in detail. The System has been successfully operated during HT-7 experiment in 2002 campaign period.展开更多
Integrating heterogeneous data sources is a precondition to share data for enterprises. Highly-efficient data updating can both save system expenses, and offer real-time data. It is one of the hot issues to modify dat...Integrating heterogeneous data sources is a precondition to share data for enterprises. Highly-efficient data updating can both save system expenses, and offer real-time data. It is one of the hot issues to modify data rapidly in the pre-processing area of the data warehouse. An extract transform loading design is proposed based on a new data algorithm called Diff-Match,which is developed by utilizing mode matching and data-filtering technology. It can accelerate data renewal, filter the heterogeneous data, and seek out different sets of data. Its efficiency has been proved by its successful application in an enterprise of electric apparatus groups.展开更多
Product data management (PDM) has been accepted as an important tool for the manufacturing industries. In recent years, more and mor e researches have been conducted in the development of PDM. Their research area s in...Product data management (PDM) has been accepted as an important tool for the manufacturing industries. In recent years, more and mor e researches have been conducted in the development of PDM. Their research area s include system design, integration of object-oriented technology, data distri bution, collaborative and distributed manufacturing working environment, secur ity, and web-based integration. However, there are limitations on their rese arches. In particular, they cannot cater for PDM in distributed manufacturing e nvironment. This is especially true in South China, where many Hong Kong (HK) ma nufacturers have moved their production plants to different locations in Pearl R iver Delta for cost reduction. However, they retain their main offices in HK. Development of PDM system is inherently complex. Product related data cover prod uct name, product part number (product identification), drawings, material speci fications, dimension requirement, quality specification, test result, log size, production schedules, product data version and date of release, special tooling (e.g. jig and fixture), mould design, project engineering in charge, cost spread sheets, while process data includes engineering release, engineering change info rmation management, and other workflow related to the process information. Accor ding to Cornelissen et al., the contemporary PDM system should contains manageme nt functions in structure, retrieval, release, change, and workflow. In system design, development and implementation, a formal specification is nece ssary. However, there is no formal representation model for PDM system. Theref ore a graphical representation model is constructed to express the various scena rios of interactions between users and the PDM system. Statechart is then used to model the operations of PDM system, Fig.1. Statechart model bridges the curr ent gap between requirements, scenarios, and the initial design specifications o f PDM system. After properly analyzing the PDM system, a new distributed PDM (DPDM) system is proposed. Both graphical representation and statechart models are constructed f or the new DPDM system, Fig.2. New product data of DPDM and new system function s are then investigated to support product information flow in the new distribut ed environment. It is found that statecharts allow formal representations to capture the informa tion and control flows of both PDM and DPDM. In particular, statechart offers a dditional expressive power, when compared to conventional state transition diagr am, in terms of hierarchy, concurrency, history, and timing for DPDM behavioral modeling.展开更多
Recently, researches on distributed data mining by making use of grid are in trend. This paper introduces a data mining algorithm by means of distributed decision-tree,which has taken the advantage of conveniences and...Recently, researches on distributed data mining by making use of grid are in trend. This paper introduces a data mining algorithm by means of distributed decision-tree,which has taken the advantage of conveniences and services supplied by the computing platform-grid,and can perform a data mining of distributed classification on grid.展开更多
Federated Learning(FL)is currently a widely used collaborative learning framework,and the distinguished feature of FL is that the clients involved in training do not need to share raw data,but only transfer the model ...Federated Learning(FL)is currently a widely used collaborative learning framework,and the distinguished feature of FL is that the clients involved in training do not need to share raw data,but only transfer the model parameters to share knowledge,and finally get a global model with improved performance.However,recent studies have found that sharing model parameters may still lead to privacy leakage.From the shared model parameters,local training data can be reconstructed and thus lead to a threat to individual privacy and security.We observed that most of the current attacks are aimed at client-specific data reconstruction,while limited attention is paid to the information leakage of the global model.In our work,we propose a novel FL attack based on shared model parameters that can deduce the data distribution of the global model.Different from other FL attacks that aim to infer individual clients’raw data,the data distribution inference attack proposed in this work shows that the attackers can have the capability to deduce the data distribution information behind the global model.We argue that such information is valuable since the training data behind a welltrained global model indicates the common knowledge of a specific task,such as social networks and e-commerce applications.To implement such an attack,our key idea is to adopt a deep reinforcement learning approach to guide the attack process,where the RL agent adjusts the pseudo-data distribution automatically until it is similar to the ground truth data distribution.By a carefully designed Markov decision proces(MDP)process,our implementation ensures our attack can have stable performance and experimental results verify the effectiveness of our proposed inference attack.展开更多
基金The National Key Basic Research Program of China(973 Program)
文摘To improve data distribution efficiency a load-balancing data distribution LBDD method is proposed in publish/subscribe mode.In the LBDD method subscribers are involved in distribution tasks and data transfers while receiving data themselves.A dissemination tree is constructed among the subscribers based on MD5 where the publisher acts as the root. The proposed method provides bucket construction target selection and path updates furthermore the property of one-way dissemination is proven.That the average out-going degree of a node is 2 is guaranteed with the proposed LBDD.The experiments on data distribution delay data distribution rate and load distribution are conducted. Experimental results show that the LBDD method aids in shaping the task load between the publisher and subscribers and outperforms the point-to-point approach.
基金National Natural Science Foundation of China,No.41621061,No.41501092 Talents Training Program from the Beijing Municipal Commission of Education No.201500002012G058
文摘Comparing the city-size distribution at the urban agglomeration（UA） scale is important for understanding the processes of urban development. However, comparative studies of city-size distribution among China＇s three largest UAs, the Beijing-Tianjin-Hebei agglomeration（BTHA）, the Yangtze River Delta agglomeration（YRDA）, and the Pearl River Delta agglomeration（PRDA）, remain inadequate due to the limitation of data availability. Therefore, using urban data derived from time-series nighttime light data, the common characteristics and distinctive features of city-size distribution among the three UAs from 1992 to 2015 were compared by the Pareto regression and the rank clock method. We identified two common features. First, the city-size distribution became more even. The Pareto exponents increased by 0.17, 0.12, and 0.01 in the YRDA, BTHA, and PRDA, respectively. Second, the average ranks of small cities ascended, being 0.55, 0.08 and 0.04 in the three UAs, respectively. However, the average ranks of large and medium cities in the three UAs experienced different trajectories, which are closely related to the similarities and differences in the driving forces for the development of UAs. Place-based measures are encouraged to promote a coordinated development among cities of differing sizes in the three UAs.
基金supported by the Special Fund of Fundamental Scientific Research Business Expense for Higher School of Central Government(Projects for creation teams ZY20110101)NSFC 41090294talent selection and training plan project of Hebei university
文摘We analyze co-seismic displacement field of the 26 December 2004, giant Sumatra–Andaman earthquake derived from Global Position System observations,geological vertical measurement of coral head, and pivot line observed through remote sensing. Using the co-seismic displacement field and AK135 spherical layered Earth model, we invert co-seismic slip distribution along the seismic fault. We also search the best fault geometry model to fit the observed data. Assuming that the dip angle linearly increases in downward direction, the postfit residual variation of the inversed geometry model with dip angles linearly changing along fault strike are plotted. The geometry model with local minimum misfits is the one with dip angle linearly increasing along strike from 4.3oin top southernmost patch to 4.5oin top northernmost path and dip angle linearly increased. By using the fault shape and geodetic co-seismic data, we estimate the slip distribution on the curved fault. Our result shows that the earthquake ruptured ＊200-km width down to a depth of about 60 km.0.5–12.5 m of thrust slip is resolved with the largest slip centered around the central section of the rupture zone78N–108N in latitude. The estimated seismic moment is8.2 9 1022 N m, which is larger than estimation from the centroid moment magnitude（4.0 9 1022 N m）, and smaller than estimation from normal-mode oscillation data modeling（1.0 9 1023 N m）.
基金supported by the Special Scientific Research Fund of Public Welfare Profession of Ministry of Land and Resources of the People’s Republic of China (No. 201011057)
文摘1 Introduction Geochemical mapping at national and continental scales continues to present challenges worldwide due to variations in geologic and geotectonic units.Use of the proper sampling media can provide rich information on
文摘In non-independent and identically distributed(non-IID)data environments,model performance often degrades significantly.To address this issue,two improvement methods are proposed:FedReg and FedReg^(*).FedReg is a method based on hybrid regularization aimed at enhancing federated learning in non-IID scenarios.It introduces hybrid regularization to replace traditional L2 regularization,combining the advantages of L1 and L2 regularization to enable feature selection while preventing overfitting.This method better adapts to the diverse data distributions of different clients,improving the overall model performance.FedReg^(*)combines hybrid regularization with weighted model aggregation.In addition to the benefits of hybrid regularization,FedReg^(*)applies a weighted averaging method in the model aggregation process,calculating weights based on the cosine similarity between each client gradient and the global gradient to more reasonably distribute client contributions.By considering variations in data quality and quantity among clients,FedReg^(*)highlights the importance of key clients and enhances the model’s generalization performance.These improvement methods enhance model accuracy and communication efficiency.
基金Project(2017YFC1405600)supported by the National Key R&D Program of ChinaProject(18JK05032)supported by the Scientific Research Project of Education Department of Shaanxi Province,China。
文摘Due to the limited scenes that synthetic aperture radar(SAR)satellites can detect,the full-track utilization rate is not high.Because of the computing and storage limitation of one satellite,it is difficult to process large amounts of data of spaceborne synthetic aperture radars.It is proposed to use a new method of networked satellite data processing for improving the efficiency of data processing.A multi-satellite distributed SAR real-time processing method based on Chirp Scaling(CS)imaging algorithm is studied in this paper,and a distributed data processing system is built with field programmable gate array(FPGA)chips as the kernel.Different from the traditional CS algorithm processing,the system divides data processing into three stages.The computing tasks are reasonably allocated to different data processing units(i.e.,satellites)in each stage.The method effectively saves computing and storage resources of satellites,improves the utilization rate of a single satellite,and shortens the data processing time.Gaofen-3(GF-3)satellite SAR raw data is processed by the system,with the performance of the method verified.
基金Project supported by the National Natural Science Foundation of China (Nos. 60772098 and 60672068)the New Century Excel-lent Talents in University of China (No. NCET-06-0393)
文摘Privacy is a critical requirement in distributed data mining. Cryptography-based secure multiparty computation is a main approach for privacy preserving. However, it shows poor performance in large scale distributed systems. Meanwhile, data perturbation techniques are comparatively efficient but are mainly used in centralized privacy-preserving data mining (PPDM). In this paper, we propose a light-weight anonymous data perturbation method for efficient privacy preserving in distributed data mining. We first define the privacy constraints for data perturbation based PPDM in a semi-honest distributed environment. Two protocols are proposed to address these constraints and protect data statistics and the randomization process against collusion attacks: the adaptive privacy-preserving summary protocol and the anonymous exchange protocol. Finally, a distributed data perturbation framework based on these protocols is proposed to realize distributed PPDM. Experiment results show that our approach achieves a high security level and is very efficient in a large scale distributed environment.
基金Supported by the National Natural Science Foundation of China(61622301,61533002)Beijing Natural Science Foundation(4172005)Major National Science and Technology Project(2017ZX07104)
文摘In wastewater treatment process(WWTP), the accurate and real-time monitoring values of key variables are crucial for the operational strategies. However, most of the existing methods have difficulty in obtaining the real-time values of some key variables in the process. In order to handle this issue, a data-driven intelligent monitoring system, using the soft sensor technique and data distribution service, is developed to monitor the concentrations of effluent total phosphorous(TP) and ammonia nitrogen(NH_4-N). In this intelligent monitoring system, a fuzzy neural network(FNN) is applied for designing the soft sensor model, and a principal component analysis(PCA) method is used to select the input variables of the soft sensor model. Moreover, data transfer software is exploited to insert the soft sensor technique to the supervisory control and data acquisition(SCADA) system. Finally, this proposed intelligent monitoring system is tested in several real plants to demonstrate the reliability and effectiveness of the monitoring performance.
文摘It is difficult to parallelize a subsistent sequential algorithm. Through analyzing the sequential algorithm of a Global Atmospheric Data Objective Analysis System, this article puts forward a distributed parallel algorithm that statically distributes data on a massively parallel processing (MPP) computer. The algorithm realizes distributed parailelization by extracting the analysis boxes and model grid point Iatitude rows with leaped steps, and by distributing the data to different processors. The parallel algorithm achieves good load balancing, high parallel efficiency, and low parallel cost. Performance experiments on a MPP computer arc also presented.
基金supported by STI 2030-Major Projects 2021ZD0200400National Natural Science Foundation of China(62276233 and 62072405)Key Research Project of Zhejiang Province(2023C01048).
文摘Multimodal sentiment analysis utilizes multimodal data such as text,facial expressions and voice to detect people’s attitudes.With the advent of distributed data collection and annotation,we can easily obtain and share such multimodal data.However,due to professional discrepancies among annotators and lax quality control,noisy labels might be introduced.Recent research suggests that deep neural networks(DNNs)will overfit noisy labels,leading to the poor performance of the DNNs.To address this challenging problem,we present a Multimodal Robust Meta Learning framework(MRML)for multimodal sentiment analysis to resist noisy labels and correlate distinct modalities simultaneously.Specifically,we propose a two-layer fusion net to deeply fuse different modalities and improve the quality of the multimodal data features for label correction and network training.Besides,a multiple meta-learner(label corrector)strategy is proposed to enhance the label correction approach and prevent models from overfitting to noisy labels.We conducted experiments on three popular multimodal datasets to verify the superiority of ourmethod by comparing it with four baselines.
文摘Distributed Data Mining is expected to discover preciously unknown, implicit and valuable information from massive data set inherently distributed over a network. In recent years several approaches to distributed data mining have been developed, but only a few of them make use of intelligent agents. This paper provides the reason for applying Multi-Agent Technology in Distributed Data Mining and presents a Distributed Data Mining System based on Multi-Agent Technology that deals with heterogeneity in such environment. Based on the advantages of both the CS model and agent-based model, the system is being able to address the specific concern of increasing scalability and enhancing performance.
文摘It is crucial,while using healthcare data,to assess the advantages of data privacy against the possible drawbacks.Data from several sources must be combined for use in many data mining applications.The medical practitioner may use the results of association rule mining performed on this aggregated data to better personalize patient care and implement preventive measures.Historically,numerous heuristics(e.g.,greedy search)and metaheuristics-based techniques(e.g.,evolutionary algorithm)have been created for the positive association rule in privacy preserving data mining(PPDM).When it comes to connecting seemingly unrelated diseases and drugs,negative association rules may be more informative than their positive counterparts.It is well-known that during negative association rules mining,a large number of uninteresting rules are formed,making this a difficult problem to tackle.In this research,we offer an adaptive method for negative association rule mining in vertically partitioned healthcare datasets that respects users’privacy.The applied approach dynamically determines the transactions to be interrupted for information hiding,as opposed to predefining them.This study introduces a novel method for addressing the problem of negative association rules in healthcare data mining,one that is based on the Tabu-genetic optimization paradigm.Tabu search is advantageous since it removes a huge number of unnecessary rules and item sets.Experiments using benchmark healthcare datasets prove that the discussed scheme outperforms state-of-the-art solutions in terms of decreasing side effects and data distortions,as measured by the indicator of hiding failure.
基金The project supported by the Meg-science Engineering Project of the Chinese Academy of Sciences
文摘HT-7 is the first superconducting tokamak device for fusion research in China. Many experiments have been done in the machine since 1994, and lots of satisfactory results have been achieved in the fusion research field on HT-7 tokamak [1]. With the development of fusion research, remote control of experiment becomes more and more important to improve experimental efficiency and expand research results. This paper will describe a RCS (Remote Control System), the combined model of Browser/Server and Client/Server, based on Internet of HT-7 distributed data acquisition system (HT7DAS). By means of RCS, authorized users all over the world can control and configure HT7DAS remotely. The RCS is designed to improve the flexibility, opening, reliability and efficiency of HT7DAS. In the paper, the whole process of design along with implementation of the system and some key items are discussed in detail. The System has been successfully operated during HT-7 experiment in 2002 campaign period.
基金Supported by National Natural Science Foundation of China (No. 50475117)Tianjin Natural Science Foundation (No.06YFJMJC03700).
文摘Integrating heterogeneous data sources is a precondition to share data for enterprises. Highly-efficient data updating can both save system expenses, and offer real-time data. It is one of the hot issues to modify data rapidly in the pre-processing area of the data warehouse. An extract transform loading design is proposed based on a new data algorithm called Diff-Match,which is developed by utilizing mode matching and data-filtering technology. It can accelerate data renewal, filter the heterogeneous data, and seek out different sets of data. Its efficiency has been proved by its successful application in an enterprise of electric apparatus groups.
文摘Product data management (PDM) has been accepted as an important tool for the manufacturing industries. In recent years, more and mor e researches have been conducted in the development of PDM. Their research area s include system design, integration of object-oriented technology, data distri bution, collaborative and distributed manufacturing working environment, secur ity, and web-based integration. However, there are limitations on their rese arches. In particular, they cannot cater for PDM in distributed manufacturing e nvironment. This is especially true in South China, where many Hong Kong (HK) ma nufacturers have moved their production plants to different locations in Pearl R iver Delta for cost reduction. However, they retain their main offices in HK. Development of PDM system is inherently complex. Product related data cover prod uct name, product part number (product identification), drawings, material speci fications, dimension requirement, quality specification, test result, log size, production schedules, product data version and date of release, special tooling (e.g. jig and fixture), mould design, project engineering in charge, cost spread sheets, while process data includes engineering release, engineering change info rmation management, and other workflow related to the process information. Accor ding to Cornelissen et al., the contemporary PDM system should contains manageme nt functions in structure, retrieval, release, change, and workflow. In system design, development and implementation, a formal specification is nece ssary. However, there is no formal representation model for PDM system. Theref ore a graphical representation model is constructed to express the various scena rios of interactions between users and the PDM system. Statechart is then used to model the operations of PDM system, Fig.1. Statechart model bridges the curr ent gap between requirements, scenarios, and the initial design specifications o f PDM system. After properly analyzing the PDM system, a new distributed PDM (DPDM) system is proposed. Both graphical representation and statechart models are constructed f or the new DPDM system, Fig.2. New product data of DPDM and new system function s are then investigated to support product information flow in the new distribut ed environment. It is found that statecharts allow formal representations to capture the informa tion and control flows of both PDM and DPDM. In particular, statechart offers a dditional expressive power, when compared to conventional state transition diagr am, in terms of hierarchy, concurrency, history, and timing for DPDM behavioral modeling.
文摘Recently, researches on distributed data mining by making use of grid are in trend. This paper introduces a data mining algorithm by means of distributed decision-tree,which has taken the advantage of conveniences and services supplied by the computing platform-grid,and can perform a data mining of distributed classification on grid.
文摘Federated Learning(FL)is currently a widely used collaborative learning framework,and the distinguished feature of FL is that the clients involved in training do not need to share raw data,but only transfer the model parameters to share knowledge,and finally get a global model with improved performance.However,recent studies have found that sharing model parameters may still lead to privacy leakage.From the shared model parameters,local training data can be reconstructed and thus lead to a threat to individual privacy and security.We observed that most of the current attacks are aimed at client-specific data reconstruction,while limited attention is paid to the information leakage of the global model.In our work,we propose a novel FL attack based on shared model parameters that can deduce the data distribution of the global model.Different from other FL attacks that aim to infer individual clients’raw data,the data distribution inference attack proposed in this work shows that the attackers can have the capability to deduce the data distribution information behind the global model.We argue that such information is valuable since the training data behind a welltrained global model indicates the common knowledge of a specific task,such as social networks and e-commerce applications.To implement such an attack,our key idea is to adopt a deep reinforcement learning approach to guide the attack process,where the RL agent adjusts the pseudo-data distribution automatically until it is similar to the ground truth data distribution.By a carefully designed Markov decision proces(MDP)process,our implementation ensures our attack can have stable performance and experimental results verify the effectiveness of our proposed inference attack.
