The learning Bayesian network （BN） structure from data is an NP-hard problem and still one of the most exciting chal- lenges in the machine learning. In this work, a novel algorithm is presented which combines ideas...The learning Bayesian network （BN） structure from data is an NP-hard problem and still one of the most exciting chal- lenges in the machine learning. In this work, a novel algorithm is presented which combines ideas from local learning, constraint- based, and search-and-score techniques in a principled and ef- fective way. It first reconstructs the junction tree of a BN and then performs a K2-scoring greedy search to orientate the local edges in the cliques of junction tree. Theoretical and experimental results show the proposed algorithm is capable of handling networks with a large number of variables. Its comparison with the well-known K2 algorithm is also presented.展开更多
In order to reduce the calculation of the failure probability in the complex mechanical system reliability risk evaluation,and to implement importance analysis of system components effectively,the system fault tree wa...In order to reduce the calculation of the failure probability in the complex mechanical system reliability risk evaluation,and to implement importance analysis of system components effectively,the system fault tree was converted into five different Bayesian network models. The Bayesian network with the minimum conditional probability table specification and the highest computation efficiency was selected as the optimal network. The two heuristics were used to optimize the Bayesian network. The fault diagnosis and causal reasoning of the system were implemented by using the selected Bayesian network. The calculation methods of Fussel-Vesely( FV),risk reduction worth( RRW),Birnbaum measure( BM) and risk achievement worth( RAW) importances were presented. A certain engine was taken as an application example to illustrate the proposed method. The results show that not only the correlation of the relevant variables in the system can be accurately expressed and the calculation complexity can be reduced,but also the relatively weak link in the system can be located accurately.展开更多
Owing to the increase in unprecedented accidents with new root causes in almost all operational areas, the importance of risk management has dramatically risen. Risk assessment, one of the most significant aspects of ...Owing to the increase in unprecedented accidents with new root causes in almost all operational areas, the importance of risk management has dramatically risen. Risk assessment, one of the most significant aspects of risk management, has a substantial impact on the system-safety level of organizations, industries, and operations. If the causes of all kinds of failure and the interactions between them are considered, effective risk assessment can be highly accurate. A combination of traditional risk assessment approaches and modern scientific probability methods can help in realizing better quantitative risk assessment methods. Most researchers face the problem of minimal field data with respect to the probability and frequency of each failure. Because of this limitation in the availability of epistemic knowledge, it is important to conduct epistemic estimations by applying the Bayesian theory for identifying plausible outcomes. In this paper, we propose an algorithm and demonstrate its application in a case study for a light-weight lifting operation in the Persian Gulf of Iran. First, we identify potential accident scenarios and present them in an event tree format. Next, excluding human error, we use the event tree to roughly estimate the prior probability of other hazard-promoting factors using a minimal amount of field data. We then use the Success Likelihood Index Method(SLIM) to calculate the probability of human error. On the basis of the proposed event tree, we use the Bayesian network of the provided scenarios to compensate for the lack of data. Finally, we determine the resulting probability of each event based on its evidence in the epistemic estimation format by building on two Bayesian network types: the probability of hazard promotion factors and the Bayesian theory. The study results indicate that despite the lack of available information on the operation of floating objects, a satisfactory result can be achieved using epistemic data.展开更多
Recently the importance of intellectual property has been increased. There has been various ways of research on analy- sis of companies, forecast of technology and so on through patents and many investments of money a...Recently the importance of intellectual property has been increased. There has been various ways of research on analy- sis of companies, forecast of technology and so on through patents and many investments of money and time. Unlike traditional method of patent analysis such as company analysis, forecasting technologies, this research is to suggest the ways to forecast registration and rejection of patents which help minimize the efforts to register patents. To do so, in- formation such as inventors, applicants, application date, and IPC codes were extracted to be used as input variables for analyzing Bayesian network. Especially, among various forms of Bayesian network, we used Tree Augmented NBN (TAN) to forecast registration and rejection of patent. This is because, TAN was assumed to have dependence between variables. As a result of this Bayesian network, it was shown that there are nearly more than 80% of accuracy to fore- cast registration and rejection of patents. Therefore, we expect the minimization of time and cost of registration by forecasting registration and rejection of R&D patent through this research.展开更多
Many black box functions and datasets have regions of different variability. Models such as the Gaussian process may fall short in giving the best representation of these complex functions. One successful approach for...Many black box functions and datasets have regions of different variability. Models such as the Gaussian process may fall short in giving the best representation of these complex functions. One successful approach for modeling this type of nonstationarity is the Treed Gaussian process <span style="font-family:Verdana;">[1]</span><span></span><span><span></span></span><span style="font-family:Verdana;">, which extended the Gaussian process by dividing the input space into different regions using a binary tree algorithm. Each region became its own Gaussian process. This iterative inference process formed many different trees and thus, many different Gaussian processes. In the end these were combined to get a posterior predictive distribution at each point. The idea was that when the iterations were combined, smoothing would take place for the surface of the predicted points near tree boundaries. We introduce the Improved Treed Gaussian process, which divides the input space into a single main binary tree where the different tree regions have different variability. The parameters for the Gaussian process for each tree region are then determined. These parameters are then smoothed at the region boundaries. This smoothing leads to a set of parameters for each point in the input space that specify the covariance matrix used to predict the point. The advantage is that the prediction and actual errors are estimated better since the standard deviation and range parameters of each point are related to the variation of the region it is in. Further, smoothing between regions is better since each point prediction uses its parameters over the whole input space. Examples are given in this paper which show these advantages for lower-dimensional problems.</span>展开更多
Information diffusion on social media has become a key strategy in people’s daily interactions. This paper studies consumers’ participation in the product information diffusion, and analyzes the complexity of inform...Information diffusion on social media has become a key strategy in people’s daily interactions. This paper studies consumers’ participation in the product information diffusion, and analyzes the complexity of information diffusion which is affected by many factors. Prior investigations of information diffusion have primarily focused on the composition of diffusion networks with independent factors and the intricacy of the process has not been completely evaluated. The majority of prior investigations have focused on strategies and the moving forces in social media processes and the determination of influential seed nodes, with few evaluations conducted about the factors affecting consumers’ choices in information diffusion. In this study, a Bayesian network model of product information diffusion was created to examine the links between factors and consumer deportment. It revealed how those factors had an impact on each other and on consumer deportment choice. The innovation of the thesis is reflected in the exploration and analysis of the specific communication path of product information diffusion, which provides a better marketing idea and practical method for the development of mobile e-commerce. The research findings can help identify the quantitative relationships between the factors affecting the process of product information diffusion and user behavior.展开更多
基金supported by the National Natural Science Fundation of China (6097408261075055)the Fundamental Research Funds for the Central Universities (K50510700004)
文摘The learning Bayesian network （BN） structure from data is an NP-hard problem and still one of the most exciting chal- lenges in the machine learning. In this work, a novel algorithm is presented which combines ideas from local learning, constraint- based, and search-and-score techniques in a principled and ef- fective way. It first reconstructs the junction tree of a BN and then performs a K2-scoring greedy search to orientate the local edges in the cliques of junction tree. Theoretical and experimental results show the proposed algorithm is capable of handling networks with a large number of variables. Its comparison with the well-known K2 algorithm is also presented.
基金National Natural Science Foundations of China(Nos.61164009,61463021)the Science Foundation of Education Commission of Jiangxi Province,China(No.GJJ14420)+1 种基金the Young Scientists Object Program of Jiangxi Province,China(No.20144BCB23037)the Graduate Innovation Foundation of Jiangxi Province,China(No.YC2014-S364)
文摘In order to reduce the calculation of the failure probability in the complex mechanical system reliability risk evaluation,and to implement importance analysis of system components effectively,the system fault tree was converted into five different Bayesian network models. The Bayesian network with the minimum conditional probability table specification and the highest computation efficiency was selected as the optimal network. The two heuristics were used to optimize the Bayesian network. The fault diagnosis and causal reasoning of the system were implemented by using the selected Bayesian network. The calculation methods of Fussel-Vesely( FV),risk reduction worth( RRW),Birnbaum measure( BM) and risk achievement worth( RAW) importances were presented. A certain engine was taken as an application example to illustrate the proposed method. The results show that not only the correlation of the relevant variables in the system can be accurately expressed and the calculation complexity can be reduced,but also the relatively weak link in the system can be located accurately.
文摘Owing to the increase in unprecedented accidents with new root causes in almost all operational areas, the importance of risk management has dramatically risen. Risk assessment, one of the most significant aspects of risk management, has a substantial impact on the system-safety level of organizations, industries, and operations. If the causes of all kinds of failure and the interactions between them are considered, effective risk assessment can be highly accurate. A combination of traditional risk assessment approaches and modern scientific probability methods can help in realizing better quantitative risk assessment methods. Most researchers face the problem of minimal field data with respect to the probability and frequency of each failure. Because of this limitation in the availability of epistemic knowledge, it is important to conduct epistemic estimations by applying the Bayesian theory for identifying plausible outcomes. In this paper, we propose an algorithm and demonstrate its application in a case study for a light-weight lifting operation in the Persian Gulf of Iran. First, we identify potential accident scenarios and present them in an event tree format. Next, excluding human error, we use the event tree to roughly estimate the prior probability of other hazard-promoting factors using a minimal amount of field data. We then use the Success Likelihood Index Method(SLIM) to calculate the probability of human error. On the basis of the proposed event tree, we use the Bayesian network of the provided scenarios to compensate for the lack of data. Finally, we determine the resulting probability of each event based on its evidence in the epistemic estimation format by building on two Bayesian network types: the probability of hazard promotion factors and the Bayesian theory. The study results indicate that despite the lack of available information on the operation of floating objects, a satisfactory result can be achieved using epistemic data.
文摘Recently the importance of intellectual property has been increased. There has been various ways of research on analy- sis of companies, forecast of technology and so on through patents and many investments of money and time. Unlike traditional method of patent analysis such as company analysis, forecasting technologies, this research is to suggest the ways to forecast registration and rejection of patents which help minimize the efforts to register patents. To do so, in- formation such as inventors, applicants, application date, and IPC codes were extracted to be used as input variables for analyzing Bayesian network. Especially, among various forms of Bayesian network, we used Tree Augmented NBN (TAN) to forecast registration and rejection of patent. This is because, TAN was assumed to have dependence between variables. As a result of this Bayesian network, it was shown that there are nearly more than 80% of accuracy to fore- cast registration and rejection of patents. Therefore, we expect the minimization of time and cost of registration by forecasting registration and rejection of R&D patent through this research.
文摘Many black box functions and datasets have regions of different variability. Models such as the Gaussian process may fall short in giving the best representation of these complex functions. One successful approach for modeling this type of nonstationarity is the Treed Gaussian process <span style="font-family:Verdana;">[1]</span><span></span><span><span></span></span><span style="font-family:Verdana;">, which extended the Gaussian process by dividing the input space into different regions using a binary tree algorithm. Each region became its own Gaussian process. This iterative inference process formed many different trees and thus, many different Gaussian processes. In the end these were combined to get a posterior predictive distribution at each point. The idea was that when the iterations were combined, smoothing would take place for the surface of the predicted points near tree boundaries. We introduce the Improved Treed Gaussian process, which divides the input space into a single main binary tree where the different tree regions have different variability. The parameters for the Gaussian process for each tree region are then determined. These parameters are then smoothed at the region boundaries. This smoothing leads to a set of parameters for each point in the input space that specify the covariance matrix used to predict the point. The advantage is that the prediction and actual errors are estimated better since the standard deviation and range parameters of each point are related to the variation of the region it is in. Further, smoothing between regions is better since each point prediction uses its parameters over the whole input space. Examples are given in this paper which show these advantages for lower-dimensional problems.</span>
文摘Information diffusion on social media has become a key strategy in people’s daily interactions. This paper studies consumers’ participation in the product information diffusion, and analyzes the complexity of information diffusion which is affected by many factors. Prior investigations of information diffusion have primarily focused on the composition of diffusion networks with independent factors and the intricacy of the process has not been completely evaluated. The majority of prior investigations have focused on strategies and the moving forces in social media processes and the determination of influential seed nodes, with few evaluations conducted about the factors affecting consumers’ choices in information diffusion. In this study, a Bayesian network model of product information diffusion was created to examine the links between factors and consumer deportment. It revealed how those factors had an impact on each other and on consumer deportment choice. The innovation of the thesis is reflected in the exploration and analysis of the specific communication path of product information diffusion, which provides a better marketing idea and practical method for the development of mobile e-commerce. The research findings can help identify the quantitative relationships between the factors affecting the process of product information diffusion and user behavior.
