The deep learning model is overfitted and the accuracy of the test set is reduced when the deep learning model is trained in the network intrusion detection parameters, due to the traditional loss function convergence...The deep learning model is overfitted and the accuracy of the test set is reduced when the deep learning model is trained in the network intrusion detection parameters, due to the traditional loss function convergence problem. Firstly, we utilize a network model architecture combining Gelu activation function and deep neural network;Secondly, the cross-entropy loss function is improved to a weighted cross entropy loss function, and at last it is applied to intrusion detection to improve the accuracy of intrusion detection. In order to compare the effect of the experiment, the KDDcup99 data set, which is commonly used in intrusion detection, is selected as the experimental data and use accuracy, precision, recall and F1-score as evaluation parameters. The experimental results show that the model using the weighted cross-entropy loss function combined with the Gelu activation function under the deep neural network architecture improves the evaluation parameters by about 2% compared with the ordinary cross-entropy loss function model. Experiments prove that the weighted cross-entropy loss function can enhance the model’s ability to discriminate samples.展开更多
In wireless sensor network,virtual backbone is a cost effective broadcasting method.Connected dominating set formation is proposed to construct a virtual backbone.However,it is NP-Hard to find a minimum connected domi...In wireless sensor network,virtual backbone is a cost effective broadcasting method.Connected dominating set formation is proposed to construct a virtual backbone.However,it is NP-Hard to find a minimum connected dominating set in an arbitrary graph.In this paper,based on cross-entropy method,we present a novel backbone formulation algorithm(BFA-CE)in wireless sensor network.In BFA-CE,a maximal independent set is got at first and nodes in the independent set are required to get their action sets.Based on those action sets,a backbone is generated with the cross-entropy method.Simulation results show that our algorithm can effectively reduce the size of backbone network within a reasonable message overhead,and it has lower average node degree.This approach can be potentially used in designing efficient broadcasting strategy or working as a backup routing of wireless sensor network.展开更多
Due to the limited resources and budgets in many real-life projects, it is unaffordable to use full factorial experimental designs and thus fractional factorial(FF) designs are used instead. The aliasing of factorial ...Due to the limited resources and budgets in many real-life projects, it is unaffordable to use full factorial experimental designs and thus fractional factorial(FF) designs are used instead. The aliasing of factorial effects is the price we pay for using FF designs and thus some significant effects cannot be estimated. Therefore, some additional observations(runs) are needed to break the linages among the factorial effects. Folding over the initial FF designs is one of the significant approaches for selecting the additional runs. This paper gives an in-depth look at fold-over techniques via the following four significant contributions. The first contribution is on discussing the adjusted switching levels foldover technique to overcome the limitation of the classical one. The second contribution is on presenting a comparison study among the widely used fold-over techniques to help experimenters to recommend a suitable fold-over technique for their experiments by answering the following two fundamental questions:Do these techniques dramatically lessen the confounding of the initial designs, and do the resulting combined designs(combining initial design with its fold-over) via these techniques have considerable difference from the optimality point of view considering the markedly different searching domains in each technique? The optimality criteria are the aberration, confounding, Hamming distance and uniformity. Many of these criteria are given in sequences(patterns) form, which are inconvenient and costly to represent and compare, especially when the designs have many factors. The third innovation is on developing a new criterion(dictionary cross-entropy loss) to simplify the existing criteria fromsequence to scalar. The new criterion leads to a more straightforward and easy comparison study. The final contribution is on establishing a general framework for the connections between initial designs and combined designs based on any fold-over technique.展开更多
This paper considers the problem of estimating the finite population total in two-phase sampling when some information on auxiliary variable is available. The authors employ an informationtheoretic approach which make...This paper considers the problem of estimating the finite population total in two-phase sampling when some information on auxiliary variable is available. The authors employ an informationtheoretic approach which makes use of effective distance between the estimated probabilities and the empirical frequencies. It is shown that the proposed cross-entropy minimization estimator is more efficient than the usual estimator and has some desirable large sample properties. With some necessary modifications, the method can be applied to two-phase sampling for stratification and non-response. A simulation study is presented to assess the finite sample performance of the proposed estimator.展开更多
Accurately diagnosing skin lesion disease is a challenging task.Although present methods often use the multi-branch structure to get more clues,the rigescent methods of cropping zone and fusing branch results fail to ...Accurately diagnosing skin lesion disease is a challenging task.Although present methods often use the multi-branch structure to get more clues,the rigescent methods of cropping zone and fusing branch results fail to handle the instability of the disease zone and the difference in branch results,which leads to improper cropping and degrades Deep Convolutional Neural Networks(DCNN)’s performance.To address these problems,we propose a Multi-scale DCNN with Dynamic weight and Part cross-entropy loss model(namely MDP-DCNN)to bootstrap skin lesion diagnosis.Inspired by the object detection method,the multi-scale structure adjusts the cropping position based on the Gradient-weighted Class Activation Mapping(Grad-CAM)center.It enables the model to adapt to the disease zone variety in position and size.The dynamic weight structure alleviates the negative influence of branch differences by comparing the grey-cropped zone and its CAM.Moreover,we also propose the part cross-entropy loss to deal with the over-fitting problem.This optimizes the non-targeted label to decrease the influence on other labels’stability when the prediction is wrong.We conduct our model on the ISIC-2017 and ISIC-2018 datasets.Experiments demonstrate that MDP-DCNN achieves excellent results in skin lesion classification without external data.Multi-scale DCNN with dynamic weight and part loss function verifies its advantages in enhancing diagnosis accuracy.展开更多
Exploiting random access for the underlying connectivity provisioning has great potential to incorporate massive machine-type communication(MTC)devices in an Internet of Things(Io T)network.However,massive access atte...Exploiting random access for the underlying connectivity provisioning has great potential to incorporate massive machine-type communication(MTC)devices in an Internet of Things(Io T)network.However,massive access attempts from versatile MTC devices may bring congestion to the IIo T network,thereby hindering service increasing of IIo T applications.In this paper,an intelligence enabled physical(PHY-)layer user signature code acquisition(USCA)algorithm is proposed to overcome the random access congestion problem with reduced signaling and control overhead.In the proposed scheme,the detector aims at approximating the optimal observation on both active user detection and user data reception by iteratively learning and predicting the convergence of the user signature codes that are in active.The crossentropy based low-complexity iterative updating rule is present to guarantee that the proposed USCA algorithm is computational feasible.A closed-form bit error rate(BER)performance analysis is carried out to show the efficiency of the proposed intelligence USCA algorithm.Simulation results confirm that the proposed USCA algorithm provides an inherent tradeoff between performance and complexity and allows the detector achieves an approximate optimal performance with a reasonable computational complexity.展开更多
In this paper, we use the discontinuous exact penalty functions to solve the constrained minimization problems with an integral approach. We examine a general form of the constrained deviation integral and its analyti...In this paper, we use the discontinuous exact penalty functions to solve the constrained minimization problems with an integral approach. We examine a general form of the constrained deviation integral and its analytical properties. The optimality conditions of the penalized minimization problems are proven. To implement the al- gorithm, the cross-entropy method and the importance sampling are used based on the Monte-Carlo technique. Numerical tests show the effectiveness of the proposed algorithm.展开更多
Objective A classification and diagnosis model for psoriasis based on deep residual network is proposed in this paper.Which using deep learning technology to classify and diagnose psoriasis can help reduce the burden ...Objective A classification and diagnosis model for psoriasis based on deep residual network is proposed in this paper.Which using deep learning technology to classify and diagnose psoriasis can help reduce the burden of doctors,simplify the diagnosis and treatment process,and improve the quality of diagnosis.Methods Firstly,data enhancement,image resizings,and TFRecord coding are used to preprocess the input of the model,and then a 34-layer deep residual network(ResNet-34)is constructed to extract the characteristics of psoriasis.Finally,we used the Adam algorithm as the optimizer to train ResNet-34,used cross-entropy as the loss function of ResNet-34 in this study to measure the accuracy of the model,and obtained an optimized ResNet-34 model for psoriasis diagnosis.Results The experimental results based on k-fold cross validation show that the proposed model is superior to other diagnostic methods in terms of recall rate,F1-score and ROC curve.Conclusion The ResNet-34 model can achieve accurate diagnosis of psoriasis,and provide technical support for data analysis and intelligent diagnosis and treatment of psoriasis.展开更多
We propose a stochastic level value approximation method for a quadratic integer convex minimizing problem in this paper. This method applies an importance sampling technique, and make use of the cross-entropy method ...We propose a stochastic level value approximation method for a quadratic integer convex minimizing problem in this paper. This method applies an importance sampling technique, and make use of the cross-entropy method to update the sample density functions. We also prove the asymptotic convergence of this algorithm, and report some numerical results to illuminate its effectiveness.展开更多
This study combined fault identification with a deep learning algorithm and applied a convolutional neural network(CNN)design based on an improved balanced crossentropy(BCE)loss function to address the low accuracy in...This study combined fault identification with a deep learning algorithm and applied a convolutional neural network(CNN)design based on an improved balanced crossentropy(BCE)loss function to address the low accuracy in the intelligent identification of seismic faults and the slow training speed of convolutional neural networks caused by unbalanced training sample sets.The network structure and optimal hyperparameters were determined by extracting feature maps layer by layer and by analyzing the results of seismic feature extraction.The BCE loss function was used to add the parameter which is the ratio of nonfaults to the total sample sets,thereby changing the loss function to find the reference of the minimum weight parameter and adjusting the ratio of fault to nonfault data.The method overcame the unbalanced number of sample sets and improved the iteration speed.After a brief training,the accuracy could reach more than 95%,and gradient descent was evident.The proposed method was applied to fault identification in an oilfield area.The trained model can predict faults clearly,and the prediction results are basically consistent with an actual case,verifying the effectiveness and adaptability of the method.展开更多
A new incremental clustering method is presented, which partitions dynamic data sets by mapping data points in high dimension space into low dimension space based on （fuzzy） cross-entropy（CE）. This algorithm is di...A new incremental clustering method is presented, which partitions dynamic data sets by mapping data points in high dimension space into low dimension space based on （fuzzy） cross-entropy（CE）. This algorithm is divided into two parts： initial clustering process and incremental clustering process. The former calculates fuzzy cross-entropy or cross-entropy of one point relafive to others and a hierachical method based on cross-entropy is used for clustering static data sets. Moreover, it has the lower time complexity. The latter assigns new points to the suitable cluster by calculating membership of data point to existed centers based on the cross-entropy measure. Experimental compafisons show the proposed methood has lower time complexity than common methods in the large-scale data situations cr dynamic work environments.展开更多
A novel correction algorithm is proposed for multi-class classification problems with corrupted training data.The algorithm is non-intrusive,in the sense that it post-processes a trained classification model by adding...A novel correction algorithm is proposed for multi-class classification problems with corrupted training data.The algorithm is non-intrusive,in the sense that it post-processes a trained classification model by adding a correction procedure to the model prediction.The correction procedure can be coupled with any approximators,such as logistic regression,neural networks of various architectures,etc.When the training dataset is sufficiently large,we theoretically prove(in the limiting case)and numerically show that the corrected models deliver correct classification results as if there is no corruption in the training data.For datasets of finite size,the corrected models produce significantly better recovery results,compared to the models without the correction algorithm.All of the theoretical findings in the paper are verified by our numerical examples.展开更多
As one of the most serious geological disasters in deep underground engineering,rockburst has caused a large number of casualties.However,because of the complex relationship between the inducing factors and rockburst ...As one of the most serious geological disasters in deep underground engineering,rockburst has caused a large number of casualties.However,because of the complex relationship between the inducing factors and rockburst intensity,the problem of rockburst intensity prediction has not been well solved until now.In this study,we collect 292 sets of rockburst data including eight parameters,such as the maximum tangential stress of the surrounding rock σ_(θ),the uniaxial compressive strength of the rockσc,the uniaxial tensile strength of the rock σ_(t),and the strain energy storage index W_(et),etc.from more than 20 underground projects as training sets and establish two new rockburst prediction models based on the kernel extreme learning machine(KELM)combined with the genetic algorithm(KELM-GA)and cross-entropy method(KELM-CEM).To further verify the effect of the two models,ten sets of rockburst data from Shuangjiangkou Hydropower Station are selected for analysis and the results show that new models are more accurate compared with five traditional empirical criteria,especially the model based on KELM-CEM which has the accuracy rate of 90%.Meanwhile,the results of 10 consecutive runs of the model based on KELM-CEM are almost the same,meaning that the model has good stability and reliability for engineering applications.展开更多
In this paper we demonstrate the need for risk-adjusted approaches to planning expansion of livestock production. In particular, we illustrate that under exposure to risk, a portfolio of producers is needed where more...In this paper we demonstrate the need for risk-adjusted approaches to planning expansion of livestock production. In particular, we illustrate that under exposure to risk, a portfolio of producers is needed where more efficient producers co-exist and cooperate with less efficient ones given that the latter are associated with lower, uncorre, lated or even negatively correlated contingencies. This raises important issues of cooperation and risk sharing among diverse producers. For large-scale practical allocation problems when information on the contingencies may be disperse, not analytically tractable, or be available on aggregate levels, we propose a downscaling procedure based on behavioral principles utilizing spatial risk preference structure, It allows for estimation of production allocation at required resolutions accounting for location specific risks and suitability constraints. The approach provides a tool for harmonization of data from various spatial levels. We applied the method in a case study of livestock production allocation in China to 2030.展开更多
It is very important to efficiently represent the target scattering characteristics in applications of polarimetric radar remote sensing. Three probability mass functions are introduced in this paper for target repres...It is very important to efficiently represent the target scattering characteristics in applications of polarimetric radar remote sensing. Three probability mass functions are introduced in this paper for target representation： using similarity parameters to describe target average scattering mechanism, using the eigenvalues of a target coherency matrix to describe target scattering randomness, and using radar received power to describe target scattering intensity. The concept of cross-entropy is employed to measure the difference between two scatterers based on the probability mass functions. Three parts of difference between scatterers are measured separately as the difference of average scattering mechanism, the difference of scattering randomness and the difference of scattering intensity, so that the usage of polarimetric data can be highly efficient and flexible. The supervised/unsupervised image classification schemes and their simplified versions are established based on the minimum cross-entropy principle. They are demonstrated to have better classification performance than the maximum likelihood classifier based on the Wishart distribution assumption, both in supervised and in unsupervised classification.展开更多
In this paper,we propose a stochastic level-value estimation method to solve a kind of box-constrained global optimization problem.For this purpose,we first derive a generalized variance function associated with the c...In this paper,we propose a stochastic level-value estimation method to solve a kind of box-constrained global optimization problem.For this purpose,we first derive a generalized variance function associated with the considered problem and prove that the largest root of the function is the global minimal value.Then,Newton’s method is applied to find the root.The convergence of the proposed method is established under some suitable conditions.Based on the main idea of the cross-entropy method to update the sampling density function,an important sampling technique is proposed in the implementation.Preliminary numerical experiments indicate the validity of the proposed method.展开更多
文摘The deep learning model is overfitted and the accuracy of the test set is reduced when the deep learning model is trained in the network intrusion detection parameters, due to the traditional loss function convergence problem. Firstly, we utilize a network model architecture combining Gelu activation function and deep neural network;Secondly, the cross-entropy loss function is improved to a weighted cross entropy loss function, and at last it is applied to intrusion detection to improve the accuracy of intrusion detection. In order to compare the effect of the experiment, the KDDcup99 data set, which is commonly used in intrusion detection, is selected as the experimental data and use accuracy, precision, recall and F1-score as evaluation parameters. The experimental results show that the model using the weighted cross-entropy loss function combined with the Gelu activation function under the deep neural network architecture improves the evaluation parameters by about 2% compared with the ordinary cross-entropy loss function model. Experiments prove that the weighted cross-entropy loss function can enhance the model’s ability to discriminate samples.
基金supported partially by the science and technology project of CQ CSTC(No.cstc2012jjA40037)
文摘In wireless sensor network,virtual backbone is a cost effective broadcasting method.Connected dominating set formation is proposed to construct a virtual backbone.However,it is NP-Hard to find a minimum connected dominating set in an arbitrary graph.In this paper,based on cross-entropy method,we present a novel backbone formulation algorithm(BFA-CE)in wireless sensor network.In BFA-CE,a maximal independent set is got at first and nodes in the independent set are required to get their action sets.Based on those action sets,a backbone is generated with the cross-entropy method.Simulation results show that our algorithm can effectively reduce the size of backbone network within a reasonable message overhead,and it has lower average node degree.This approach can be potentially used in designing efficient broadcasting strategy or working as a backup routing of wireless sensor network.
基金supported by the Beijing Normal University-Hong Kong Baptist University United International College under Grant Nos.R201810,R201912 and R202010the Zhuhai Premier Discipline Grant。
文摘Due to the limited resources and budgets in many real-life projects, it is unaffordable to use full factorial experimental designs and thus fractional factorial(FF) designs are used instead. The aliasing of factorial effects is the price we pay for using FF designs and thus some significant effects cannot be estimated. Therefore, some additional observations(runs) are needed to break the linages among the factorial effects. Folding over the initial FF designs is one of the significant approaches for selecting the additional runs. This paper gives an in-depth look at fold-over techniques via the following four significant contributions. The first contribution is on discussing the adjusted switching levels foldover technique to overcome the limitation of the classical one. The second contribution is on presenting a comparison study among the widely used fold-over techniques to help experimenters to recommend a suitable fold-over technique for their experiments by answering the following two fundamental questions:Do these techniques dramatically lessen the confounding of the initial designs, and do the resulting combined designs(combining initial design with its fold-over) via these techniques have considerable difference from the optimality point of view considering the markedly different searching domains in each technique? The optimality criteria are the aberration, confounding, Hamming distance and uniformity. Many of these criteria are given in sequences(patterns) form, which are inconvenient and costly to represent and compare, especially when the designs have many factors. The third innovation is on developing a new criterion(dictionary cross-entropy loss) to simplify the existing criteria fromsequence to scalar. The new criterion leads to a more straightforward and easy comparison study. The final contribution is on establishing a general framework for the connections between initial designs and combined designs based on any fold-over technique.
基金supported by the National Natural Science Foundation of China under Grant No.61070236
文摘This paper considers the problem of estimating the finite population total in two-phase sampling when some information on auxiliary variable is available. The authors employ an informationtheoretic approach which makes use of effective distance between the estimated probabilities and the empirical frequencies. It is shown that the proposed cross-entropy minimization estimator is more efficient than the usual estimator and has some desirable large sample properties. With some necessary modifications, the method can be applied to two-phase sampling for stratification and non-response. A simulation study is presented to assess the finite sample performance of the proposed estimator.
基金supported by the scholarship from China Scholarship Council(CSC)(No.CSC N°201806280502).
文摘Accurately diagnosing skin lesion disease is a challenging task.Although present methods often use the multi-branch structure to get more clues,the rigescent methods of cropping zone and fusing branch results fail to handle the instability of the disease zone and the difference in branch results,which leads to improper cropping and degrades Deep Convolutional Neural Networks(DCNN)’s performance.To address these problems,we propose a Multi-scale DCNN with Dynamic weight and Part cross-entropy loss model(namely MDP-DCNN)to bootstrap skin lesion diagnosis.Inspired by the object detection method,the multi-scale structure adjusts the cropping position based on the Gradient-weighted Class Activation Mapping(Grad-CAM)center.It enables the model to adapt to the disease zone variety in position and size.The dynamic weight structure alleviates the negative influence of branch differences by comparing the grey-cropped zone and its CAM.Moreover,we also propose the part cross-entropy loss to deal with the over-fitting problem.This optimizes the non-targeted label to decrease the influence on other labels’stability when the prediction is wrong.We conduct our model on the ISIC-2017 and ISIC-2018 datasets.Experiments demonstrate that MDP-DCNN achieves excellent results in skin lesion classification without external data.Multi-scale DCNN with dynamic weight and part loss function verifies its advantages in enhancing diagnosis accuracy.
基金supported in part by Natural Science Foundation of Heilongjiang Province of China under Grant YQ2021F003in part by the National Natural Science Foundation of China under Grant 61901140+1 种基金in part by China Postdoctoral Science Foundation Funded Project under Grant 2019M650067in part by Science and Technology on Communication Networks Laboratory under Grant SCX21641X003。
文摘Exploiting random access for the underlying connectivity provisioning has great potential to incorporate massive machine-type communication(MTC)devices in an Internet of Things(Io T)network.However,massive access attempts from versatile MTC devices may bring congestion to the IIo T network,thereby hindering service increasing of IIo T applications.In this paper,an intelligence enabled physical(PHY-)layer user signature code acquisition(USCA)algorithm is proposed to overcome the random access congestion problem with reduced signaling and control overhead.In the proposed scheme,the detector aims at approximating the optimal observation on both active user detection and user data reception by iteratively learning and predicting the convergence of the user signature codes that are in active.The crossentropy based low-complexity iterative updating rule is present to guarantee that the proposed USCA algorithm is computational feasible.A closed-form bit error rate(BER)performance analysis is carried out to show the efficiency of the proposed intelligence USCA algorithm.Simulation results confirm that the proposed USCA algorithm provides an inherent tradeoff between performance and complexity and allows the detector achieves an approximate optimal performance with a reasonable computational complexity.
基金supported by the National Natural Science Foundation of China (No. 10771133)the KeyDisciplines of Shanghai Municipality (Operations Research and Cybernetics) (No. S30104)
文摘In this paper, we use the discontinuous exact penalty functions to solve the constrained minimization problems with an integral approach. We examine a general form of the constrained deviation integral and its analytical properties. The optimality conditions of the penalized minimization problems are proven. To implement the al- gorithm, the cross-entropy method and the importance sampling are used based on the Monte-Carlo technique. Numerical tests show the effectiveness of the proposed algorithm.
基金We thank for the funding support from the Key Research and Development Plan of China(No.2017YFC1703306)Youth Project of Natural Science Foundation of Hunan Province(No.2019JJ50453)+2 种基金Project of Hunan Health Commission(No.202112072217)Open Fund Project of Hunan University of Traditional Chinese Medicine(No.2018JK02)General Project of Education Department of Hunan Province(No.19C1318).
文摘Objective A classification and diagnosis model for psoriasis based on deep residual network is proposed in this paper.Which using deep learning technology to classify and diagnose psoriasis can help reduce the burden of doctors,simplify the diagnosis and treatment process,and improve the quality of diagnosis.Methods Firstly,data enhancement,image resizings,and TFRecord coding are used to preprocess the input of the model,and then a 34-layer deep residual network(ResNet-34)is constructed to extract the characteristics of psoriasis.Finally,we used the Adam algorithm as the optimizer to train ResNet-34,used cross-entropy as the loss function of ResNet-34 in this study to measure the accuracy of the model,and obtained an optimized ResNet-34 model for psoriasis diagnosis.Results The experimental results based on k-fold cross validation show that the proposed model is superior to other diagnostic methods in terms of recall rate,F1-score and ROC curve.Conclusion The ResNet-34 model can achieve accurate diagnosis of psoriasis,and provide technical support for data analysis and intelligent diagnosis and treatment of psoriasis.
基金Project supported by the National Natural Science Foundation of China (No.10671117)Shanghai Leading Academic Discipline Project (No.J050101)the Youth Science Foundation of Hunan Education Department of China (No.06B037)
文摘We propose a stochastic level value approximation method for a quadratic integer convex minimizing problem in this paper. This method applies an importance sampling technique, and make use of the cross-entropy method to update the sample density functions. We also prove the asymptotic convergence of this algorithm, and report some numerical results to illuminate its effectiveness.
基金supported by the Natural Science Foundation of Shandong Province(ZR202103050722).
文摘This study combined fault identification with a deep learning algorithm and applied a convolutional neural network(CNN)design based on an improved balanced crossentropy(BCE)loss function to address the low accuracy in the intelligent identification of seismic faults and the slow training speed of convolutional neural networks caused by unbalanced training sample sets.The network structure and optimal hyperparameters were determined by extracting feature maps layer by layer and by analyzing the results of seismic feature extraction.The BCE loss function was used to add the parameter which is the ratio of nonfaults to the total sample sets,thereby changing the loss function to find the reference of the minimum weight parameter and adjusting the ratio of fault to nonfault data.The method overcame the unbalanced number of sample sets and improved the iteration speed.After a brief training,the accuracy could reach more than 95%,and gradient descent was evident.The proposed method was applied to fault identification in an oilfield area.The trained model can predict faults clearly,and the prediction results are basically consistent with an actual case,verifying the effectiveness and adaptability of the method.
文摘A new incremental clustering method is presented, which partitions dynamic data sets by mapping data points in high dimension space into low dimension space based on （fuzzy） cross-entropy（CE）. This algorithm is divided into two parts： initial clustering process and incremental clustering process. The former calculates fuzzy cross-entropy or cross-entropy of one point relafive to others and a hierachical method based on cross-entropy is used for clustering static data sets. Moreover, it has the lower time complexity. The latter assigns new points to the suitable cluster by calculating membership of data point to existed centers based on the cross-entropy measure. Experimental compafisons show the proposed methood has lower time complexity than common methods in the large-scale data situations cr dynamic work environments.
文摘A novel correction algorithm is proposed for multi-class classification problems with corrupted training data.The algorithm is non-intrusive,in the sense that it post-processes a trained classification model by adding a correction procedure to the model prediction.The correction procedure can be coupled with any approximators,such as logistic regression,neural networks of various architectures,etc.When the training dataset is sufficiently large,we theoretically prove(in the limiting case)and numerically show that the corrected models deliver correct classification results as if there is no corruption in the training data.For datasets of finite size,the corrected models produce significantly better recovery results,compared to the models without the correction algorithm.All of the theoretical findings in the paper are verified by our numerical examples.
基金funded by National Natural Science Foundation of China(Grants Nos.41825018 and 42141009)the Second Tibetan Plateau Scientific Expedition and Research Program(Grants No.2019QZKK0904)。
文摘As one of the most serious geological disasters in deep underground engineering,rockburst has caused a large number of casualties.However,because of the complex relationship between the inducing factors and rockburst intensity,the problem of rockburst intensity prediction has not been well solved until now.In this study,we collect 292 sets of rockburst data including eight parameters,such as the maximum tangential stress of the surrounding rock σ_(θ),the uniaxial compressive strength of the rockσc,the uniaxial tensile strength of the rock σ_(t),and the strain energy storage index W_(et),etc.from more than 20 underground projects as training sets and establish two new rockburst prediction models based on the kernel extreme learning machine(KELM)combined with the genetic algorithm(KELM-GA)and cross-entropy method(KELM-CEM).To further verify the effect of the two models,ten sets of rockburst data from Shuangjiangkou Hydropower Station are selected for analysis and the results show that new models are more accurate compared with five traditional empirical criteria,especially the model based on KELM-CEM which has the accuracy rate of 90%.Meanwhile,the results of 10 consecutive runs of the model based on KELM-CEM are almost the same,meaning that the model has good stability and reliability for engineering applications.
文摘In this paper we demonstrate the need for risk-adjusted approaches to planning expansion of livestock production. In particular, we illustrate that under exposure to risk, a portfolio of producers is needed where more efficient producers co-exist and cooperate with less efficient ones given that the latter are associated with lower, uncorre, lated or even negatively correlated contingencies. This raises important issues of cooperation and risk sharing among diverse producers. For large-scale practical allocation problems when information on the contingencies may be disperse, not analytically tractable, or be available on aggregate levels, we propose a downscaling procedure based on behavioral principles utilizing spatial risk preference structure, It allows for estimation of production allocation at required resolutions accounting for location specific risks and suitability constraints. The approach provides a tool for harmonization of data from various spatial levels. We applied the method in a case study of livestock production allocation in China to 2030.
基金This work was supported in part by the National Natural Science Foundation of China(Grant No．40271077)the National Important Fundamental Research Plan of China(Grant No．2001CB309401)+3 种基金the Science Foundation of National Defence of Chinathe Research Fund for the Doctoral Program of Higher Education of Chinathe Aerospace Technology Foundation of Chinaand the Fundam ental Research Foundation of Tsinghua University．
文摘It is very important to efficiently represent the target scattering characteristics in applications of polarimetric radar remote sensing. Three probability mass functions are introduced in this paper for target representation： using similarity parameters to describe target average scattering mechanism, using the eigenvalues of a target coherency matrix to describe target scattering randomness, and using radar received power to describe target scattering intensity. The concept of cross-entropy is employed to measure the difference between two scatterers based on the probability mass functions. Three parts of difference between scatterers are measured separately as the difference of average scattering mechanism, the difference of scattering randomness and the difference of scattering intensity, so that the usage of polarimetric data can be highly efficient and flexible. The supervised/unsupervised image classification schemes and their simplified versions are established based on the minimum cross-entropy principle. They are demonstrated to have better classification performance than the maximum likelihood classifier based on the Wishart distribution assumption, both in supervised and in unsupervised classification.
文摘In this paper,we propose a stochastic level-value estimation method to solve a kind of box-constrained global optimization problem.For this purpose,we first derive a generalized variance function associated with the considered problem and prove that the largest root of the function is the global minimal value.Then,Newton’s method is applied to find the root.The convergence of the proposed method is established under some suitable conditions.Based on the main idea of the cross-entropy method to update the sampling density function,an important sampling technique is proposed in the implementation.Preliminary numerical experiments indicate the validity of the proposed method.
