Aiming at the problems of bispectral analysis when applied to machinery fault diagnosis, a machinery fault feature extraction method based on sparseness-controlled non-negative tensor factorization （SNTF） is propose...Aiming at the problems of bispectral analysis when applied to machinery fault diagnosis, a machinery fault feature extraction method based on sparseness-controlled non-negative tensor factorization （SNTF） is proposed. First, a non-negative tensor factorization（NTF） algorithm is improved by imposing sparseness constraints on it. Secondly, the bispectral images of mechanical signals are obtained and stacked to form a third-order tensor. Thirdly, the improved algorithm is used to extract features, which are represented by a series of basis images from this tensor. Finally, coefficients indicating these basis images＇ weights in constituting original bispectral images are calculated for fault classification. Experiments on fault diagnosis of gearboxes show that the extracted features can not only reveal some nonlinear characteristics of the system, but also have intuitive meanings with regard to fault characteristic frequencies. These features provide great convenience for the interpretation of the relationships between machinery faults and corresponding bispectra.展开更多
Aiming at the slow convergence and low accuracy problems of the traditional non-negative tensor factorization, a local hierarchical non-negative tensor factorization method is proposed by applying the local objective ...Aiming at the slow convergence and low accuracy problems of the traditional non-negative tensor factorization, a local hierarchical non-negative tensor factorization method is proposed by applying the local objective function theory to non- negative tensor factorization and combining the three semi-non- negative matrix factorization（NMF） model. The effectiveness of the method is verified by the facial feature extraction experiment. Through the decomposition of a series of an air compressor＇s vibration signals composed in the form of a bispectrum by this new method, the basis images representing the fault features and corresponding weight matrices are obtained. Then the relationships between characteristics and faults are analyzed and the fault types are classified by importing the weight matrices into the BP neural network. Experimental results show that the accuracy of fault diagnosis is improved by this new method compared with other feature extraction methods.展开更多
The rapid development of location-based social networks(LBSNs) provides people with an opportunity of better understanding their mobility behavior which enables them to decide their next location.For example,it can he...The rapid development of location-based social networks(LBSNs) provides people with an opportunity of better understanding their mobility behavior which enables them to decide their next location.For example,it can help travelers to choose where to go next,or recommend salesmen the most potential places to deliver advertisements or sell products.In this paper,a method for recommending points of interest(POIs)is proposed based on a collaborative tensor factorization(CTF)technique.Firstly,a generalized objective function is constructed for collaboratively factorizing a tensor with several feature matrices.Secondly,a 3-mode tensor is used to model all users' check-in behaviors,and three feature matrices are extracted to characterize the time distribution,category distribution and POI correlation,respectively.Thirdly,each user's preference to a POI at a specific time can be estimated by using CTF.In order to further improve the recommendation accuracy,PCTF(Partitionbased CTF) is proposed to fill the missing entries of a tensor after clustering its every mode.Experiments on a real checkin database show that the proposed method can provide more accurate location recommendation.展开更多
Dear Editor,This letter presents a latent-factorization-of-tensors(LFT)-incorporated battery cycle life prediction framework.Data-driven prognosis and health management(PHM)for battery pack(BP)can boost the safety and...Dear Editor,This letter presents a latent-factorization-of-tensors(LFT)-incorporated battery cycle life prediction framework.Data-driven prognosis and health management(PHM)for battery pack(BP)can boost the safety and sustainability of a battery management system(BMS),which relies heavily on the quality of the measured BP data like the voltage(V),current(I),and temperature(T).展开更多
The nonnegative tensor （matrix） factorization finds more and more applications in various disciplines including machine learning, data mining, and blind source separation, etc. In computation, the optimization probl...The nonnegative tensor （matrix） factorization finds more and more applications in various disciplines including machine learning, data mining, and blind source separation, etc. In computation, the optimization problem involved is solved by alternatively minimizing one factor while the others are fixed. To solve the subproblem efficiently, we first exploit a variable regularization term which makes the subproblem far from ill-condition. Second, an augmented Lagrangian alternating direction method is employed to solve this convex and well-conditioned regularized subproblem, and two accelerating skills are also implemented. Some preliminary numerical experiments are performed to show the improvements of the new method.展开更多
How to extract robust feature is an important research topic in machine learning community. In this paper, we investigate robust feature extraction for speech signal based on tensor structure and develop a new method ...How to extract robust feature is an important research topic in machine learning community. In this paper, we investigate robust feature extraction for speech signal based on tensor structure and develop a new method called constrained Nonnegative Tensor Factorization （cNTF）. A novel feature extraction framework based on the cortical representation in primary auditory cortex （A1） is proposed for robust speaker recognition. Motivated by the neural firing rates model in A1, the speech signal first is represented as a general higher order tensor, cNTF is used to learn the basis functions from multiple interrelated feature subspaces and find a robust sparse representation for speech signal. Computer simulations are given to evaluate the performance of our method and comparisons with existing speaker recognition methods are also provided. The experimental results demonstrate that the proposed method achieves higher recognition accuracy in noisy environment.展开更多
Non-convex methods play a critical role in low-rank tensor completion for their approximation to tensor rank is tighter than that of convex methods.But they usually cost much more time for calculating singular values ...Non-convex methods play a critical role in low-rank tensor completion for their approximation to tensor rank is tighter than that of convex methods.But they usually cost much more time for calculating singular values of large tensors.In this paper,we propose a double transformed tubal nuclear norm(DTTNN)to replace the rank norm penalty in low rank tensor completion(LRTC)tasks.DTTNN turns the original non-convex penalty of a large tensor into two convex penalties of much smaller tensors,and it is shown to be an equivalent transformation.Therefore,DTTNN could take advantage of non-convex envelopes while saving time.Experimental results on color image and video inpainting tasks verify the effectiveness of DTTNN compared with state-of-the-art methods.展开更多
How can we efficiently store and mine dynamically generated dense tensors for modeling the behavior of multidimensional dynamic data?Much of the multidimensional dynamic data in the real world is generated in the form...How can we efficiently store and mine dynamically generated dense tensors for modeling the behavior of multidimensional dynamic data?Much of the multidimensional dynamic data in the real world is generated in the form of time-growing tensors.For example,air quality tensor data consists of multiple sensory values gathered from wide locations for a long time.Such data,accumulated over time,is redundant and consumes a lot ofmemory in its raw form.We need a way to efficiently store dynamically generated tensor data that increase over time and to model their behavior on demand between arbitrary time blocks.To this end,we propose a Block IncrementalDense Tucker Decomposition(BID-Tucker)method for efficient storage and on-demand modeling ofmultidimensional spatiotemporal data.Assuming that tensors come in unit blocks where only the time domain changes,our proposed BID-Tucker first slices the blocks into matrices and decomposes them via singular value decomposition(SVD).The SVDs of the time×space sliced matrices are stored instead of the raw tensor blocks to save space.When modeling from data is required at particular time blocks,the SVDs of corresponding time blocks are retrieved and incremented to be used for Tucker decomposition.The factor matrices and core tensor of the decomposed results can then be used for further data analysis.We compared our proposed BID-Tucker with D-Tucker,which our method extends,and vanilla Tucker decomposition.We show that our BID-Tucker is faster than both D-Tucker and vanilla Tucker decomposition and uses less memory for storage with a comparable reconstruction error.We applied our proposed BID-Tucker to model the spatial and temporal trends of air quality data collected in South Korea from 2018 to 2022.We were able to model the spatial and temporal air quality trends.We were also able to verify unusual events,such as chronic ozone alerts and large fire events.展开更多
The discrimination of neutrons from gamma rays in a mixed radiation field is crucial in neutron detection tasks.Several approaches have been proposed to enhance the performance and accuracy of neutron-gamma discrimina...The discrimination of neutrons from gamma rays in a mixed radiation field is crucial in neutron detection tasks.Several approaches have been proposed to enhance the performance and accuracy of neutron-gamma discrimination.However,their performances are often associated with certain factors,such as experimental requirements and resulting mixed signals.The main purpose of this study is to achieve fast and accurate neutron-gamma discrimination without a priori information on the signal to be analyzed,as well as the experimental setup.Here,a novel method is proposed based on two concepts.The first method exploits the power of nonnegative tensor factorization(NTF)as a blind source separation method to extract the original components from the mixture signals recorded at the output of the stilbene scintillator detector.The second one is based on the principles of support vector machine(SVM)to identify and discriminate these components.In addition to these two main methods,we adopted the Mexican-hat function as a continuous wavelet transform to characterize the components extracted using the NTF model.The resulting scalograms are processed as colored images,which are segmented into two distinct classes using the Otsu thresholding method to extract the features of interest of the neutrons and gamma-ray components from the background noise.We subsequently used principal component analysis to select the most significant of these features wich are used in the training and testing datasets for SVM.Bias-variance analysis is used to optimize the SVM model by finding the optimal level of model complexity with the highest possible generalization performance.In this framework,the obtained results have verified a suitable bias–variance trade-off value.We achieved an operational SVM prediction model for neutron-gamma classification with a high true-positive rate.The accuracy and performance of the SVM based on the NTF was evaluated and validated by comparing it to the charge comparison method via figure of merit.The results indicate that the proposed approach has a superior discrimination quality(figure of merit of 2.20).展开更多
Non-negative Tucker decomposition(NTD) has been developed as a crucial method for non-negative tensor data representation.However, NTD is essentially an unsupervised method and cannot take advantage of label informati...Non-negative Tucker decomposition(NTD) has been developed as a crucial method for non-negative tensor data representation.However, NTD is essentially an unsupervised method and cannot take advantage of label information. In this paper, we claim that the low-dimensional representation extracted by NTD can be treated as the predicted soft-clustering coefficient matrix and can therefore be learned jointly with label propagation in a unified framework. The proposed method can extract the physicallymeaningful and parts-based representation of tensor data in their natural form while fully exploring the potential ability of the given labels with a nearest neighbors graph. In addition, an efficient accelerated proximal gradient(APG) algorithm is developed to solve the optimization problem. Finally, the experimental results on five benchmark image data sets for semi-supervised clustering and classification tasks demonstrate the superiority of this method over state-of-the-art methods.展开更多
Social tagging is one of the most important characteristics of Web 2.0 services, and social tagging systems （STS） are becoming more and more popular for users to annotate, organize and share items on the Web. Moreov...Social tagging is one of the most important characteristics of Web 2.0 services, and social tagging systems （STS） are becoming more and more popular for users to annotate, organize and share items on the Web. Moreover, online social network has been incorporated into social tagging systems. As more and more users tend to interact with real friends on the Web, personalized user recommendation service provided in social tagging systems is very appealing. In this paper, we propose a personalized user recommendation method, and our method handles not only the users＇ interest networks, but also the social network information. We empirically show that our method outperforms a state-of-the-art method on real dataset from Last.fro dataset and Douban.展开更多
基金The National Natural Science Foundation of China (No.50875048)the Natural Science Foundation of Jiangsu Province (No.BK2007115)the National High Technology Research and Development Program of China (863 Program)(No.2007AA04Z421)
文摘Aiming at the problems of bispectral analysis when applied to machinery fault diagnosis, a machinery fault feature extraction method based on sparseness-controlled non-negative tensor factorization （SNTF） is proposed. First, a non-negative tensor factorization（NTF） algorithm is improved by imposing sparseness constraints on it. Secondly, the bispectral images of mechanical signals are obtained and stacked to form a third-order tensor. Thirdly, the improved algorithm is used to extract features, which are represented by a series of basis images from this tensor. Finally, coefficients indicating these basis images＇ weights in constituting original bispectral images are calculated for fault classification. Experiments on fault diagnosis of gearboxes show that the extracted features can not only reveal some nonlinear characteristics of the system, but also have intuitive meanings with regard to fault characteristic frequencies. These features provide great convenience for the interpretation of the relationships between machinery faults and corresponding bispectra.
基金The National Natural Science Foundation of China(No.50875078)the Natural Science Foundation of Jiangsu Province(No.BK2007115)the National High Technology Research and Development Program of China(863 Program)(No.2007AA04Z421)
文摘Aiming at the slow convergence and low accuracy problems of the traditional non-negative tensor factorization, a local hierarchical non-negative tensor factorization method is proposed by applying the local objective function theory to non- negative tensor factorization and combining the three semi-non- negative matrix factorization（NMF） model. The effectiveness of the method is verified by the facial feature extraction experiment. Through the decomposition of a series of an air compressor＇s vibration signals composed in the form of a bispectrum by this new method, the basis images representing the fault features and corresponding weight matrices are obtained. Then the relationships between characteristics and faults are analyzed and the fault types are classified by importing the weight matrices into the BP neural network. Experimental results show that the accuracy of fault diagnosis is improved by this new method compared with other feature extraction methods.
基金supported in part by the National Nature Science Foundation of China(91218301,61572360)the Basic Research Projects of People's Public Security University of China(2016JKF01316)Shanghai Shuguang Program(15SG18)
文摘The rapid development of location-based social networks(LBSNs) provides people with an opportunity of better understanding their mobility behavior which enables them to decide their next location.For example,it can help travelers to choose where to go next,or recommend salesmen the most potential places to deliver advertisements or sell products.In this paper,a method for recommending points of interest(POIs)is proposed based on a collaborative tensor factorization(CTF)technique.Firstly,a generalized objective function is constructed for collaboratively factorizing a tensor with several feature matrices.Secondly,a 3-mode tensor is used to model all users' check-in behaviors,and three feature matrices are extracted to characterize the time distribution,category distribution and POI correlation,respectively.Thirdly,each user's preference to a POI at a specific time can be estimated by using CTF.In order to further improve the recommendation accuracy,PCTF(Partitionbased CTF) is proposed to fill the missing entries of a tensor after clustering its every mode.Experiments on a real checkin database show that the proposed method can provide more accurate location recommendation.
文摘Dear Editor,This letter presents a latent-factorization-of-tensors(LFT)-incorporated battery cycle life prediction framework.Data-driven prognosis and health management(PHM)for battery pack(BP)can boost the safety and sustainability of a battery management system(BMS),which relies heavily on the quality of the measured BP data like the voltage(V),current(I),and temperature(T).
文摘The nonnegative tensor （matrix） factorization finds more and more applications in various disciplines including machine learning, data mining, and blind source separation, etc. In computation, the optimization problem involved is solved by alternatively minimizing one factor while the others are fixed. To solve the subproblem efficiently, we first exploit a variable regularization term which makes the subproblem far from ill-condition. Second, an augmented Lagrangian alternating direction method is employed to solve this convex and well-conditioned regularized subproblem, and two accelerating skills are also implemented. Some preliminary numerical experiments are performed to show the improvements of the new method.
基金supported by the National Natural Science Foundation of China under Grant No.60775007the National Basic Research 973 Program of China under Grant No.2005CB724301the Science and Technology Commission of Shanghai Municipality under Grant No.08511501701
文摘How to extract robust feature is an important research topic in machine learning community. In this paper, we investigate robust feature extraction for speech signal based on tensor structure and develop a new method called constrained Nonnegative Tensor Factorization （cNTF）. A novel feature extraction framework based on the cortical representation in primary auditory cortex （A1） is proposed for robust speaker recognition. Motivated by the neural firing rates model in A1, the speech signal first is represented as a general higher order tensor, cNTF is used to learn the basis functions from multiple interrelated feature subspaces and find a robust sparse representation for speech signal. Computer simulations are given to evaluate the performance of our method and comparisons with existing speaker recognition methods are also provided. The experimental results demonstrate that the proposed method achieves higher recognition accuracy in noisy environment.
基金financially supported by the National Nautral Science Foundation of China(No.61703206)
文摘Non-convex methods play a critical role in low-rank tensor completion for their approximation to tensor rank is tighter than that of convex methods.But they usually cost much more time for calculating singular values of large tensors.In this paper,we propose a double transformed tubal nuclear norm(DTTNN)to replace the rank norm penalty in low rank tensor completion(LRTC)tasks.DTTNN turns the original non-convex penalty of a large tensor into two convex penalties of much smaller tensors,and it is shown to be an equivalent transformation.Therefore,DTTNN could take advantage of non-convex envelopes while saving time.Experimental results on color image and video inpainting tasks verify the effectiveness of DTTNN compared with state-of-the-art methods.
基金supported by the Institute of Information&Communications Technology Planning&Evaluation (IITP)grant funded by the Korean government (MSIT) (No.2022-0-00369)by the NationalResearch Foundation of Korea Grant funded by the Korean government (2018R1A5A1060031,2022R1F1A1065664).
文摘How can we efficiently store and mine dynamically generated dense tensors for modeling the behavior of multidimensional dynamic data?Much of the multidimensional dynamic data in the real world is generated in the form of time-growing tensors.For example,air quality tensor data consists of multiple sensory values gathered from wide locations for a long time.Such data,accumulated over time,is redundant and consumes a lot ofmemory in its raw form.We need a way to efficiently store dynamically generated tensor data that increase over time and to model their behavior on demand between arbitrary time blocks.To this end,we propose a Block IncrementalDense Tucker Decomposition(BID-Tucker)method for efficient storage and on-demand modeling ofmultidimensional spatiotemporal data.Assuming that tensors come in unit blocks where only the time domain changes,our proposed BID-Tucker first slices the blocks into matrices and decomposes them via singular value decomposition(SVD).The SVDs of the time×space sliced matrices are stored instead of the raw tensor blocks to save space.When modeling from data is required at particular time blocks,the SVDs of corresponding time blocks are retrieved and incremented to be used for Tucker decomposition.The factor matrices and core tensor of the decomposed results can then be used for further data analysis.We compared our proposed BID-Tucker with D-Tucker,which our method extends,and vanilla Tucker decomposition.We show that our BID-Tucker is faster than both D-Tucker and vanilla Tucker decomposition and uses less memory for storage with a comparable reconstruction error.We applied our proposed BID-Tucker to model the spatial and temporal trends of air quality data collected in South Korea from 2018 to 2022.We were able to model the spatial and temporal air quality trends.We were also able to verify unusual events,such as chronic ozone alerts and large fire events.
基金L’Ore´al-UNESCO for the Women in Science Maghreb Program Grant Agreement No.4500410340.
文摘The discrimination of neutrons from gamma rays in a mixed radiation field is crucial in neutron detection tasks.Several approaches have been proposed to enhance the performance and accuracy of neutron-gamma discrimination.However,their performances are often associated with certain factors,such as experimental requirements and resulting mixed signals.The main purpose of this study is to achieve fast and accurate neutron-gamma discrimination without a priori information on the signal to be analyzed,as well as the experimental setup.Here,a novel method is proposed based on two concepts.The first method exploits the power of nonnegative tensor factorization(NTF)as a blind source separation method to extract the original components from the mixture signals recorded at the output of the stilbene scintillator detector.The second one is based on the principles of support vector machine(SVM)to identify and discriminate these components.In addition to these two main methods,we adopted the Mexican-hat function as a continuous wavelet transform to characterize the components extracted using the NTF model.The resulting scalograms are processed as colored images,which are segmented into two distinct classes using the Otsu thresholding method to extract the features of interest of the neutrons and gamma-ray components from the background noise.We subsequently used principal component analysis to select the most significant of these features wich are used in the training and testing datasets for SVM.Bias-variance analysis is used to optimize the SVM model by finding the optimal level of model complexity with the highest possible generalization performance.In this framework,the obtained results have verified a suitable bias–variance trade-off value.We achieved an operational SVM prediction model for neutron-gamma classification with a high true-positive rate.The accuracy and performance of the SVM based on the NTF was evaluated and validated by comparing it to the charge comparison method via figure of merit.The results indicate that the proposed approach has a superior discrimination quality(figure of merit of 2.20).
基金This work was supported by the National Natural Science Foundation of China(Grant Nos.62073087,U191140003,6197309 and 61973090)the Key-Area Research and Development Program of Guangdong Province(Grant Nos.2019B010154002 and 2019B010118001)。
文摘Non-negative Tucker decomposition(NTD) has been developed as a crucial method for non-negative tensor data representation.However, NTD is essentially an unsupervised method and cannot take advantage of label information. In this paper, we claim that the low-dimensional representation extracted by NTD can be treated as the predicted soft-clustering coefficient matrix and can therefore be learned jointly with label propagation in a unified framework. The proposed method can extract the physicallymeaningful and parts-based representation of tensor data in their natural form while fully exploring the potential ability of the given labels with a nearest neighbors graph. In addition, an efficient accelerated proximal gradient(APG) algorithm is developed to solve the optimization problem. Finally, the experimental results on five benchmark image data sets for semi-supervised clustering and classification tasks demonstrate the superiority of this method over state-of-the-art methods.
基金supported by the National Basic Research Program of China (2009CB320505)the Hi-Tech Research and Development Program of China (20011AA 01A102)+2 种基金the Nuclear High-Based Project of China (2012ZX01039004-008)the National Nature Science Foundation of China (61002011)the Electronic Information Industry Development Fund Program 'The Development and Industrialization of Key Supporting Software in Cloud Computing (Cloud Storage Service)'
文摘Social tagging is one of the most important characteristics of Web 2.0 services, and social tagging systems （STS） are becoming more and more popular for users to annotate, organize and share items on the Web. Moreover, online social network has been incorporated into social tagging systems. As more and more users tend to interact with real friends on the Web, personalized user recommendation service provided in social tagging systems is very appealing. In this paper, we propose a personalized user recommendation method, and our method handles not only the users＇ interest networks, but also the social network information. We empirically show that our method outperforms a state-of-the-art method on real dataset from Last.fro dataset and Douban.
