35 results about "Tensor factorization" patented technology

A method and a system for fast tensor-vector multiplication provide factoring an original tensor into a kernel and a commutator, multiplying the kernel obtained by the factoring of the original tensor, by the vector and thereby obtaining a matrix, and summating elements and sums of elements of the matrix as defined by the commutator obtained by the factoring of the original tensor, and thereby obtaining a resulting tensor which corresponds to a product of the original tensor and the vector.

A method and a system for digital filtering comprising fast tensor-vector multiplication provide factoring an original tensor into a kernel and a commutator, multiplying the kernel obtained by the factoring of the original tensor, by the vector and thereby obtaining a matrix, and summating elements and sums of elements of the matrix as defined by the commutator obtained by the factoring of the original tensor, and thereby obtaining a resulting tensor which corresponds to a product of the original tensor and the vector.

The invention provides a user attribute and social network detection system based on space-time locus. A user behavior mode is analyzed to predict true identity attributes of a user, and data processing, social network modeling, characteristic extraction and classification prediction subsystems are included; space-time locus data is analyzed; according to the space-time locus data, an original method based on the point mutual information is proposed to establish a social network model of the user; an original non-negative tensor factorization (NTF) algorithm is proposed to automatically extract hidden characteristics of the user; the hidden characteristics of the user are utilized to predict user attributes through utilizing multiple types of classifiers. The system is advantaged in that the system is applied to user attribute authenticity detection, can be further applied to user social network detection and can be further applied to precise information push and friendly recommendation according to predicted attributes and the social network.

The present invention relates to an urban traffic flow prediction method based on tensor filling. The method comprises the following steps: the step S1: collecting traffic flow state data adjacent toa prediction point; the step S2: performing initial filling according to the collected traffic data; the step S3: design a traffic data tensor; and the step S4: performing dynamic filling prediction based on tensor decomposition dynamic filling according to the designed traffic flow data tensor. The urban traffic flow prediction method based on tensor filling overcomes defects that prediction precision is not good in the condition that a traffic flow data source has deficiencies in the prior art; and moreover, a mode of tensor window division dynamic filling is employed while a traffic flow tensor model is constructed, so that prediction precision is improved in the condition that the traffic flow data source has deficiencies.

The present teaching relates to recommending content items to a user based on tensor factorization. In one example, a request is received for recommending content items to the user. Tensor data related to a plurality of users and a plurality of content items are obtained based on the request. The tensor data is decomposed into a plurality of sub-tensors based on a prior probability distribution. At least one bound is determined for a tensor factorization model that is generated based on the prior probability distribution. One or more items interesting to the user are predicted based on the at least one bound and the plurality of sub-tensors. At least one of the one or more items is recommended to the user as a response to the request.

The invention discloses a multi-view face synthesis method based on tensor resolution and Delaunay triangulation, and mainly solves the problem of difficulty in face image synthesis under continuous changing views in the prior art. The method comprises the following steps of: extracting outline information of a face image by a characteristic point labeled method; separating out a view coefficientmatrix of human face data in a training set by a tensor resolution method; fitting sample bands of the view coefficient matrix; building characteristic points of a new view by adopting a tensor resolution formula; synthesizing the face image of the new view by adopting the Delaunay triangulation and linear affine transformation according to the known face image. The method has the advantages of authentic face synthesis result, wide synthesis view ranges and low operation complexity and can be applied to the field of computer vision or face synthesis under different views in a multimedia technology.

An exemplary system, method and computer-accessible medium for determining a plurality of tissue parameters of a tissue(s), can include, for example, receiving information related to a plurality of rotational invariants contained within a diffusion magnetic resonance (dMR) image(s) of the tissue(s), and generating the tissue parameters using a set of rotational invariants related to the plurality of tissue parameters using such information. The tissue parameters can be generated by factorizing a response of an individual fiber segment of the tissue(s) based on the set of rotational invariants. The response of the individual fiber segments can be factorized from an orientational distribution function (“ODF”). The individual fiber segments can be factorized using a scalar tensor factorization(s) of the rotational invariants. The set of rotational invariants can be of a rotation group SO(3).

We describe techniques for restoring an audio signal. In embodiments these employ masked positive semi-definite tensor factorization to process the signal in the time-frequency domain. Broadly speaking the methods estimate latent variables which factorize a tensor representation of the (unknown) variance / covariance of an input audio signal, using a mask so that the audio signal is separated into desired and undesired audio source components. In embodiments a masked positive semi-definite tensor factorization of ψftk=MftkUfkVtk is performed, where M defines the mask and U, V the latent variables. A restored audio signal is then constructed by modifying the input signal to better match the variance / covariance of the desired components.

The invention belongs to the technical field of digital image processing and mode identification and specifically relates to a digital-vein feature extraction method based on a nonnegative-matrix factorization theory. The method includes: extracting a region of interest (ROI) from a digital-vein panel pool; performing image matrix column vectorization on the ROI and obtaining a digital-vein data set; training the digital-vein data set and obtaining a feature matrix and a factor matrix; and extracting a test-sample RIO and performing column vectorization on the test-sample ROI which is then changed into a test-sample vector and projecting on the feature matrix so that an obtained projecting factor is a to-be-identified feature. The digital-vein feature extraction method based on the nonnegative-matrix factorization theory applies the nonnegative-matrix factorization theory which is more aligned with cognition and has more physical significance to a digital-vein image feature extraction task and further studies and promotes the sparseness of the digital-vein feature based images.

The invention discloses an ultralow rank tensor data filling method, used to solve a technical problem that an existing tensor data processing method is low in precision. The technical scheme comprises: decomposing a tensor to a low-rank structure and a non-low-rank structure, using a Gaussian Mixture Model (MOG) to perform priori description on the non-low-rank structure, using a Gibbs to sample to obtain sample average of the non-low-rank structure E and the low-rank structure X which is decomposed based on CP, and using an average approximate band to obtain implicit tensor. On one hand, the method fully excavates sparsity of CANDECOMP / PARAFAC(CP) decomposed weight, and a low-rank model based on sparsity is established, on the other hand, the Gaussian Mixture Model (MOG) is used to simulate the complex non-low-rank structure. The two points ensure that even though under the condition of lower than 10% observation rate, the method can adaptively fill tensor. Through tests, relative reconstruction error is reduced, and reconstruction precision is improved.

The present development details a method and apparatus for performing channel emulation of doubly selective scenarios, where the simulation and emulation duration is arbitrarily long for a stationary or non-stationary channel, with non-separable dispersion which is achieved by combining the techniques of channel orthogonalization, decomposition of the correlation tensor in the Doppler domain into frequency-dependent correlation matrices, followed by a matrix factorization of each of the mentioned matrices and, finally, the use of the windowing method to generate arbitrarily long achievements which thereby allows the concatenation of channel realizations coming from the same or different NSSF, thus achieving reproduction of stationary or non-stationary channels, respectively.

The invention provides a multiattribute grading system oriented grading prediction method, comprising the following steps: S1: inputting data: collecting grading of a user for a plurality of attributes of a commodity, inputting grading data of the attributes; S2: constructing a model: using a three-dimensional tensor with U, C and A three dimensions (as shown in the specification) for expressing multiattribute grading information; S3: setting parameters and obtaining an initialized matrix (as shown in the specification); S4: resolving the tensor, calculating to obtain an intermediary matrix (as shown in the specification); and S5: filling the tensor and completing grading prediction. By using the method provided by the invention, the grading information of a plurality of attributes can be effectively integrated for predicting the grading of the user for each attribute of the commodity.

An approach to processing of acoustic signals acquired at a user's device include one or both of acquisition of parallel signals from a set of closely spaced microphones, and use of a multi-tier computing approach in which some processing is performed at the user's device and further processing is performed at one or more server computers in communication with the user's device. The acquired signals are processed using time versus frequency estimates of both energy content as well as direction of arrival. In some examples, a non-negative matrix or tensor factorization approach is used to identify multiple sources each associated with a corresponding direction of arrival of a signal from that source. In some examples, data characterizing direction of arrival information is passed from the user's device to a server computer where direction-based processing is performed.

The invention relates to the medical imaging technology, in particular to a diffusion tensor imaging method for processing nerve fiber reconstruction in a nerve complex distribution area, device and magnetic resonance equipment. A multivariate high-order tensor model is adopted to describe the diffusion motion in a voxel, and in combination of a high-order tensor decomposition theory, the directions of all fibers in the voxel are solved. The method disclosed in the invention is particularly suitable for reconstructing nerve fibers in the nerve complex distribution area and can effectively process nerve fiber bundle intersection, bifurcation and other conditions. The method, the device and the equipment have important application value in neuroscience, medical imaging and the like.

The present invention provides a spatial-temporal trajectory and social network user attribute prediction system. According to the system, the behavior pattern of a user is analyzed, so that the real identity attributes of the user can be predicted. The system includes a data processing subsystem, a feature extraction subsystem and a classification prediction subsystem; spatial-temporal trajectory data and social network data are analyzed; an original non-negative tensor factorization (NTF) algorithm is put forward to automatically extract the hidden features of the user; and a plurality of classifiers are adopted to predict the attribute of the user through using the hidden features of the user. The spatial-temporal trajectory and social network user attribute prediction system of the invention can be used for user attribute authenticity detection and can be also used for performing precise popularization according to predicted attributes.

The invention relates to a road network travel time prediction method based on a tensor neural network. The method comprises the following steps: (1) constructing a road network travel time tensor based on multi-dimensional features; (2) decomposing a road network travel time tensor to obtain a feature matrix of the tensor in each dimension; (3) predicting and obtaining a time sequence characteristic matrix through a long and short time recording neural network; and (4) reconstructing a road network travel time tensor, and obtaining predicted road network travel time stored in a tensor form. According to the tensor neural network provided by the invention, large-scale road network time sequence prediction can be achieved only by predicting the factor matrix of the tensor through a big data compression technology, and the tensor neural network has strong robustness for noise and data loss.

The invention discloses a multilinear ICA (independent component analysis)-based spectrum tensor dimension reduction classification method. In the method, factors affecting spectrum characteristics of a terrain are used as within-class factors, a within-class factor, a class and a pixel spectrum serve as a mode respectively to build a three-order tensor, and dimension reduction based on low rank tensor decomposition is carried out on the three-order tensor; the three-order tensor D is subjected to multilinear ICA decomposition, and a class space matrix Cclass and a within-class factor space matrix Cwithin-class are obtained; and a supervised classifier is adopted to classify classless test hyperspectral images d. After the model is built, the hyperspectral images can be classified, no adjustment is needed, and according to other tensor modeling methods, the best classification effects are achieved only by repeated setting and parameter adjustment. All pixel spectrums of the same class are mapped to the same coefficient vector, influences from various factors are reduced to the minimum, the classification precision is improved, and the result is stable. When unknown pixel spectrums are classified, which factor affects the spectrums can be inferred.

The invention provides a time sequence regularization tensor decomposition-based QoS (Quality of Service) prediction method in mobile edge computing. The method comprises the following steps of: acquiring a QoS record of a user access service; constructing a three-dimensional tensor model representing a user, service and time relationship based on the QoS record; using a CP decomposition method to decompose the three-dimensional tensor model, and introducing constraint regular terms related to a user dimension, a service dimension and a time dimension in the decomposition process to perform decomposition constraint on the weights of the user, the service and the time of the three-dimensional tensor model; and introducing a time sequence regular term to carry out decomposition constraint on a time sequence relation of a time dimension of the three-dimensional tensor model, respectively obtaining a user factor matrix, a service factor matrix and a time factor matrix, and predicting a QoS value of a specific service called by a specific user at a specific moment. According to the method, tensor decomposition and time sequence prediction are combined to improve the accuracy of QoS prediction in the mobile edge computing environment.

Provided is a computer-implemented method and platform for context aware sorting of items available for configuration of a system during a selection session, the method including the steps of providing a numerical input vector, V, representing items selected in a current selection session as context; calculating a compressed vector, Vcomp, from the numerical input vector, V, using an artificial neural network, ANN, adapted to capture non-linear dependencies between items; multiplying the compressed vector, Vcomp, with a weight matrix, EI, derived from a factor matrix, E, obtained as a result of a tensor factorization of a stored relationship tensor, Tr, representing relations, r, between selections of items performed in historical selection sessions, available items and their attributes to compute an output score vector, S; and sorting automatically the available items for selection in the current selection session according to relevance scores of the computed output score vector, S.

The invention discloses a label recommendation method and device and a readable medium, and relates to the field of label recommendation. The method comprises the steps of obtaining a target account and a target resource; determining a recommendation value of a label through a label recommendation model decomposed by the tensor, the label recommendation model being composed into a core tensor anda factor matrix by the tensor, a target sub tensor in n sub tensors of the core tensor being correspondingly multiplied by the factor matrix, and the other sub tensors being 0; and determining n labels with the highest recommendation value in the label data as labels recommended to the target account. In the process of recommending the labels through the label recommendation model, the element inother sub tensors except the target sub tensor in the core tensor is 0, namely in the equality relationship, the part participating in the equality relationship in the core tensor only comprises thepart corresponding to the target sub tensor, and the rest part does not participate in the equality relationship, so that the problem of overhigh time complexity caused by the complete three-dimensional core tensor is avoided.

The invention discloses a top-k elephant flow prediction method and system based on discretization tensor filling. The method comprises: acquiring tensors containing missing flow data from known flowdata; decomposing the tensor into three discrete binary factor matrixes to form a real value factor matrix; expressing real value tensor data by using binary factor vectors in three dimension directions of a tensor source node, time and a target node respectively, wherein the elements of the three factor matrixes are used as the binary factor vectors; representing the missing flow data at each moment by using the inner product of the binary factor vectors in the three-dimensional direction, and calculating a Hamming distance through a high-efficiency data prediction method based on bit operation to replace the inner product; calculating a Hamming distance by using a top-k prediction acceleration method based on binary code segmentation, and determining whether corresponding real value tensor data is a top-k elephant flow or not according to the Hamming distance; and retrieving all the real value tensor data, and returning the first k pieces of maximum real value tensor data to obtain atop-k elephant flow. The problem of calculation complexity in the prior art is solved, and time and space complexity are reduced.

An exemplary system, method and computer-accessible medium for determining a plurality of tissue parameters of a tissue(s), can include, for example, receiving information related to a plurality of rotational invariants contained within a diffusion magnetic resonance (dMR) image(s) of the tissue(s), and generating the tissue parameters using a set of rotational invariants related to the plurality of tissue parameters using such information. The tissue parameters can be generated by factorizing a response of an individual fiber segment of the tissue(s) based on the set of rotational invariants. The response of the individual fiber segments can be factorized from an orientational distribution function (“ODF”). The individual fiber segments can be factorized using a scalar tensor factorization(s) of the rotational invariants. The set of rotational invariants can be of a rotation group SO(3).

The invention discloses a medical privacy data protection method based on federated learning tensor factorization, and the method comprises the following specific steps: 1, each medical institution needs to maintain a locally decomposed tensor factor matrix and a global tensor non-patient factor matrix, initialization is carried out when the federation process is started; 2, each medical institution is enabled to perform local tensor factorization training, and gradient descent by using a loss function; 3, calculating of a corresponding factor matrix updating gradient according to the locally decomposed factor matrix and the global non-patient factor matrix is carried out; according to the medical privacy data protection method based on federated learning tensor factorization, the user data privacy can be further protected while the communication efficiency is improved, meanwhile, the calculation amount of homomorphic encryption is reduced, and the problem that the accuracy of an aggregated global factor matrix is low due to local training of non-independent identically-distributed clients can be solved.

The invention provides a high-speed rail rail damage classification method, the main idea of ​​which is: firstly, the time domain and frequency domain local features of the lossy signal are extracted by wavelet analysis method, and a three-dimensional tensor signal is established for the same measurement point combined with different carriages, and the data Extending to multi-dimensional space to obtain non-negative tensors, using the alternating least squares algorithm as the iterative criterion for non-negative tensor decomposition, and then introducing singular value decomposition to improve the initialization of non-negative tensors, using the improved non-negative tensor decomposition method The hidden features are extracted, and finally the extreme learning machine algorithm is introduced to realize the real-time classification of rail damage. The method of the invention can accurately classify rail damage signals, improves the speed and accuracy of damage classification and has better robustness. The rail damage classification method proposed by the invention is superior to existing methods, can obtain better recognition effect, and can be widely used in the field of rail damage classification.

The invention discloses a multi-view face synthesis method based on tensor resolution and Delaunay triangulation, and mainly solves the problem of difficulty in face image synthesis under continuous changing views in the prior art. The method comprises the following steps of: extracting outline information of a face image by a characteristic point labeled method; separating out a view coefficientmatrix of human face data in a training set by a tensor resolution method; fitting sample bands of the view coefficient matrix; building characteristic points of a new view by adopting a tensor resolution formula; synthesizing the face image of the new view by adopting the Delaunay triangulation and linear affine transformation according to the known face image. The method has the advantages of authentic face synthesis result, wide synthesis view ranges and low operation complexity and can be applied to the field of computer vision or face synthesis under different views in a multimedia technology.

The present teaching relates to recommending content items to a user based on tensor factorization. In one example, a request is received for recommending content items to the user. Tensor data related to a plurality of users and a plurality of content items are obtained based on the request. The tensor data is decomposed into a plurality of sub-tensors based on a prior probability distribution. At least one bound is determined for a tensor factorization model that is generated based on the prior probability distribution. One or more items interesting to the user are predicted based on the at least one bound and the plurality of sub-tensors. At least one of the one or more items is recommended to the user as a response to the request.

The invention discloses an objective quality evaluation method for high dynamic range images without reference based on tensor space, which starts from the image data itself, uses a third-order tensor to represent a color high dynamic range image, and then uses Tucker in tensor decomposition The decomposition algorithm performs tensor decomposition on the distorted high dynamic range image to obtain three feature images, namely the first feature image, the second feature image and the third feature image, and the colors of the high dynamic range image are integrated in the three feature images Information, then extract the manifold structure features of the first feature image and the perceptual detail contrast features of the second and third feature images, and then use the support vector regression method in machine learning to calculate the objective quality evaluation value of the distorted high dynamic range image, so that The objective evaluation of the quality of the color high dynamic range image without reference is realized, and the evaluation effect is obviously improved, thereby effectively improving the correlation between the objective evaluation result and the subjective perception.

The invention discloses an image annotation reinforcing method based on user information modeling, which comprises the steps of: firstly, representing an original relationship of a user, an image and an annotation into a three-order tensor; then representing the observed original annotation into a positive and negative sample pair, decomposing the tensor until an ordering relation of the positiveand negative sample pair is minimized; and finally, carrying out subspace alternation by using a core tensor and a factor matrix obtained through decomposition to obtain image and annotation representations on a common subspace. The condition that large-scale pictures share website data proves that the image annotation reinforcing method obtains a better annotation reinforcing effect.

The invention discloses a pre-stack seismic reflection mode analysis method based on a tensor discriminant dictionary, comprising the following steps: S1, acquiring pre-stack seismic signals; S2, adding neighborhood gather information of target gathers, and target gather data Composing multi-dimensional pre-stack seismic data together; S3, making training data; S4, constructing the objective function, and formulating a classification strategy that is conducive to the classification of pre-stack seismic data on the basis of the discriminant tensor dictionary learning algorithm; S5, iteratively updating the model; S6. Put the pre-stack seismic data into the model trained in step S6, and obtain the reflection mode result of the target layer in the whole work area. The invention is based on Tucker tensor decomposition, can process multi-dimensional data well, better utilize multi-dimensional information of pre-stack seismic data, can have higher classification accuracy for pre-stack seismic reflection mode analysis, and is robust to noise better.