333 results about "Tensor decomposition" patented technology

The invention discloses a tensor decomposition-based acceleration and compression method for a deep convolutional neural network. The method at least comprises the steps of 1: obtaining an original deep convolutional neural network; 2: performing tensor decomposition on a weight tensor of each layer in the original deep convolutional neural network to obtain a plurality of low-rank sub-tensors; and 3: replacing the weight tensor of each layer in the original deep convolutional neural network with the low-rank sub-tensors to obtain a new deep convolutional neural network. Through an embodiment of the method, the acceleration and compression of a large deep convolutional neural network are realized.

The invention discloses a recommended system and a method with facing social network for context awareness based on the tensor decomposition, and relates to the field of the data mining and the information retrieval. Firstly, the method makes use of a social network massive data set to collect users and projects and contexts, to pay attention to the list information, to establish an original the user-the project-the context mark matrix, to calculate the users similarity, and to establish a user-user similarity matrix; Secondly, aim at the extreme sparsity of the original mark matrix, a sparse mark matrix is predicated and filled by using the tensor decomposition; Thirdly, aim at a problem that the user similarity matrix is sparse, a sparse user similarity matrix is predicated and filled by using the matrix decomposition; Finally, according to some similar interest tendencies of some similar users in the social network, a social normalization item is taken to optimizing the mark matrix. The method deals with the problem that a traditional predicated mark matrix does not consider that the context information and the relationship between users have an effect on marking. Also, the method deals with an obstruction which is caused by the sparsity of the mark matrix brings to the recommended system, thus the accuracy of the recommended system is improved. The method can be widely applied to the fields of the social network, the electronic commerce and the like.

The invention belongs to the technical field of image processing, relates to the POLSAR image processing technology, and discloses a polarization image sorting method based on tensor decomposition and dimension reduction. According to the method, polarization data and a polarization characteristic quantity matrix are utilized for setting three-dimensional polarization characteristic tensor, low-dimension characteristic tensor is obtained according to the dimension reduction method based on tensor decomposition, training samples are selected from the low-dimension characteristic tensor for classification of an SVM, and on the basis of not destroying the space relation between the structure of the three-dimensional polarization characteristic tensor and adjacent pixels, redundancy between the characteristic quantities is removed, dimension disasters are avoided, the classification effect is well improved, and the efficiency and robustness of the algorithm are improved. The polarization image sorting method based on tensor decomposition and dimension reduction can be applied to classification of various complex terrains.

The invention discloses a knowledge graph representation learning method based on path tensor decomposition, comprising the following steps of: step 1, extracting an entity set, a relation set and a triple set in a knowledge graph, and embedding the entity set and the relation set which satisfy a triple into a low-dimensional continuous vector space; step 2, acquiring a path between entities through a PRA (Progressive Refinement Approach) algorithm; step 3, carrying out tensor decomposition on the path where all entities exist possibly, and calculating a decomposition loss function value; step 4, repeating step 3 until a convergent preset value or the maximum number of iteration is reached; step 5, if the maximum number of iteration or the convergent preset value is reached, calculating a next triple-related path, and repeating steps 2 to 4 until all triples of a training set are executed; and step 6, outputting the corresponding entity set and relation set in a training model. The representation learning method can improve the inference accuracy of knowledge discovery and enhance the prediction precision.

The invention discloses a multi-channel audio signal compressing method based on tensor decomposition, and belongs to the technical field of audio signal processing, in particular to the technical field of spatial audio coding and decoding. The method comprises the following steps: overlapping and framing an audio signal of each channel and carrying out time frequency transform on each frame of signal to obtain a frequency domain coefficient; combining all channels and the frequency domain coefficients of all frame sequences to establish a three-order tensor signal; carrying out tensor decomposition on the three-order tensor signal so as to obtain a low-rank nuclear tensor for coding transmission; reconstructing a tensor signal by using the low-rank nuclear tensor combined and recovered at a decoding end and a low-rank projection matrix trained in advance; and carrying out inverse transformation and overlap-add on the reconstructed tensor signal in each channel to recover a multi-channel audio signal. The multi-channel audio signal compressing method has the advantages as follows: as the multi-channel audio signal is analyzed, coded and decoded through the combination of time frequency transform and tensor decomposition and redundant information is removed by using correlations between channels and within the channels, the compression efficiency of the multi-channel audio signal can be increased to a greater degree.

The invention discloses a voice signal characteristics extracting method based on tensor decomposition and belongs to the technical field of voice signal processing. The voice signal characteristics extracting method based on the tensor decomposition comprises the following steps: having multi-layer wavelet decomposition to voice signals after framing, respectively extracting MR frequency cepstrum coefficients, the corresponding first order difference coefficient and second order difference coefficient from a plurality of component information after the wavelet decomposition to form a characteristic parameter vector, establishing a third order voice tensor and having tensor decomposition to the third order voice tensor, and calculating component information order and characteristic projection of characteristic parameters. Marticulated results are characteristics carried by each frame of voice signals. Compare with the traditional characteristic parameters, the voice signal characteristics extracting method based on the tensor decomposition has the advantages of enhancing representational ability to the voice signals, acquiring characteristics which carries more comprehensive voice signals, and improving the effects of voice signal processing systems such as voice identifying signal processing system, speaker identifying signal processing system.

The invention relates to an asymptotic variational method-based method for simulating and optimizing a composite material laminated plate, which belongs to the field of analysis of material mechanics. The method specifically comprises the following steps of: constructing a three-dimensional plate energy equation represented by a one-dimensional generalized strain and warping function on the basisof a rotation tensor decomposition concept; strictly splitting the original three-dimensional problem analysis into nonlinear two-dimensional plate analysis and cross-section analysis along a thickness direction on the basis of an asymptotic variational method; asymptotically correcting an approximate energy functional of a reduced-order model to a second order by using the inherent small parameter of the plate and converting the approximate energy functional into the form of Reissner model for practical application through an equilibrium equation; accurately reconstructing a three-dimensional stress/strain/displacement field by using an obtained global response asymptotic correction warping function; and optimizing the composite material laminated plate by using an optimization strategy of bending and torsion rigidity coefficients obtained by maximization cross-section analysis. The method has high practicability and high generality, and the resolving speed and efficiency of this type of problems can be remarkably increased.

The invention relates to a multi-data set and collaborative tensor decomposition-based scenic spot recommendation method. According to the method, user behavior information of different data sources is utilized to provide fine-grit travel recommendation services for the users. The method comprises the following steps of: firstly obtaining information of a plurality of data sets from a social network site; constructing a user-scenic spot-time tensor on the basis of user travel history information, combining a collaborative tensor decomposition model to decompose and complement the tensor so as to obtain travel preferences of the users; and finally recommending proper scenic spots according to the complemented user-scenic spot-time tensor and travel cities and time scene information input by users. According to the method, time perception-based scenic spot recommendation services can be provided for the users.

The present invention discloses a deep neural network compression method based on block item tensor decomposition. The method comprises the steps of: obtaining a framework of a deep neural network, respectively converting a weight matrix W and an input vector x to a high-order tensor (u)W(/u) and a high-order tensor (u)X(/u), performing block item tensor decomposition processing of the high-ordertensor (u)W(/u), replacing a full-connection layer of the deep neural network with a block item tensor layer, and employing a back-propagating algorithm to perform training of the deep neural networkreplaced in the step D. The deep neural network compression method based on the block item tensor decomposition employs a block item tensor decomposition method to construct the block item tensor layer to replace the full- connection layer of the original deep neural network, the features of symmetry and index expression capability of the block item tensor layer are employed to greatly compress the parameter amount of the full-connection layer and maintain the classification precision of the original network.

The invention relates to a multilayer convolution neural network optimization system and a multilayer convolution neural network optimization method. The multilayer convolution neural network optimization system comprises an image positioning module, a sampling module based on CP decomposition, a micro-sampling module, a parameter tuning module based on a BP algorithm, and a convolution neural network feature output module, wherein the image positioning module sets a parameter matrix theta by means of a regression function according to dimensionality of a convolutional layer; the sampling module based on CP decomposition performs tensor decomposition on a result after convolution operation to obtain two rank-one tensors p and q; the micro-sampling module adopts a bilinear interpolation algorithm for carrying out linear interpolation on pixel points in different directions of an image, and obtains network output V; the parameter tuning module based on a BP algorithm updates a parameter theta; and the convolution neural network feature output module is used for introducing an updated parameter theta<hat> into a network, carrying out iterative operation and outputting convolution neural network features. The multilayer convolution neural network optimization system and the multilayer convolution neural network optimization method are conductive to extracting space-invariant features, and improving operational efficiency.

The invention relates to a convolutional neural network generation method and device, and belongs to the technical field of video coding and decoding. The method comprises the following steps: training an initial convolutional neural network to obtain a trained convolutional neural network; determining a to-be-processed convolutional neural network based on the trained convolutional neural network; and performing tensor decomposition operation on at least one target convolutional layer in the to-be-processed convolutional neural network to obtain a target convolutional neural network. According to the invention, the problem that the operation efficiency is low when the current convolutional neural network performs convolution operation can be solved.

The invention discloses an earthquake data compression method based on tensor adaptive rank truncation. According to the method, based on given compression conditions and a different-dimension singular value setting truncation rank acquired through instant higher-order singular value decomposition, tensor decomposition is carried out further according to singular value distribution determination truncation rank magnitude; on the condition that a compression rate is guaranteed, a compression peak value signal to noise ratio is improved, and the compression effect is improved; through designing the tensor decomposition flow based on adaptive rank truncation, manual parameter regulation and control in the compression process can be reduced, and data compression work is simplified. On the basis of the singular value distribution setting truncation rank magnitude, data anisotropy characteristics are utilized, a singular value standard deviation is introduced to calculate weight of a truncation rank corresponding to different dimensions, the compression peak value signal to noise ratio is improved, and the data compression effect is further improved.

The invention, which belongs to the field of the vehicle-mounted self-organizing network technology, relates to a car networking crowdsourcing method facing urban spatial information collection. A car driving habit is excavated by using a trajectory prediction algorithm; according to the car driving habit, a specific car is selected to perform a task; and collection of relevant task information is completed under the circumstance that normal car driving is not affected. A location prediction algorithm based on tensor decomposition is put forward; according to historical trajectory data, a ternary group is constructed; a three-order tensor is constructed based on the ternary group; the tensor is decomposed based on a BPRC specification by means of PITF; iterative parameter optimization is carried out to complement tensor elements; and ranking is carried out based on tensor element values to complete prediction. According to a prediction result, a bipartite graph of a car and a road task is constructed; and maximum matching of the car with the road task is solved by using a Kuhn-Munkres algorithm, so that the matching success probability is maximized. Compared with the prior art, the car networking crowdsourcing method has the following beneficial effects: with combination of the car driving characteristics, information collection redundancy can be reduced and the information collection efficiency can be enhanced.

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 invention discloses a seismic signal restoration method based on dictionary learning regularization sparse representation. The method comprises the steps that S1 a tensor product method is appliedto a tensor dictionary learning process to construct an objective function; S2 an alternating iterative algorithm is used to solve a tensor sparse coefficient; S3 a tensor dictionary is trained by the Lagrange dual method; and S4 the tensor dictionary and the tensor sparse coefficient are iteratively updated to reconstruct a missing seismic signal. According to the seismic signal restoration method based on dictionary learning regularization sparse representation, firstly a t-product operator is introduced into tensor decomposition, and accordingly a new objective function is constructed; the ADM algorithm and the Lagrange dual algorithm are used to solve the sparse coefficient and the tensor dictionary; sparse coding of the seismic signal is realized; sparse representation of tensor data and recovery of the missing seismic signal are finally realized; and the effect of seismic data reconstruction is improved.

The invention discloses a compressed representation learning method based on tensor decomposition. The method comprises the following steps that firstly, a representation learned through micro neuralnetwork preprocessing is converted into a tensor to be decomposed; tensor decomposition is carried out on the basis of an optimization algorithm, a subspace of the tensor decomposition is solved, low-rank reconstruction is carried out, and finally, low-rank representation extracted through processing tensor decomposition of another micro neural network is fused into representation learned by a backbone network to play a role of regularization; and a truncated single-step gradient optimization method is combined to improve an optimization algorithm with a multi-step time axis iterative model. According to the method, regularization and supplementation are successfully provided for large-scale pre-training and representation learning in a calculation-friendly and parameter-saving mode, the effectiveness of the method is verified by a large number of tasks and applications of computer vision, and a remarkable effect is achieved in image recognition, semantic segmentation and target detection; and the attention mechanism commonly used by computer vision is destroyed by lighter calculation and parameter quantity.

The invention discloses a social relation-perceiving IPTV user behavioral modeling and program recommendation method, having the main works comprising: (1) analyzing the behavior of an IPTV user baseson a historical viewing log, and constructing viewing preference models of the user and a group to which the user belongs by combining the user group clustering result; (2) avoiding harmful effects brought by data sparsity to recommendation by using dense program category similarity matrixes, improving tensor decomposition models, and optimizing the recommendation result; and (3) realizing user oriented personalized IPTV program recommendation by comprehensively considering the viewing preferences of the IPTV user and the group to which the user belongs. Compared with the existing recommendation algorithm, the method provided by the invention obviously improves the recommendation accuracy, the recall rate, and the like.

The invention discloses a GPU cluster-based multidimensional big data factorization method, aims to solve the problem that a conventional grid parallel factor analysis model cannot process large-scale and high-dimension multidimensional data analysis, and provides an effective pattern processing unit-based multidimensional big data multi-mode decomposition method, namely a hierarchical parallel factor analysis framework. The framework is based on the conventional grid parallel factor analysis model, comprises a process of integrating tensor subsets under a coarse-grained model and a process of calculating all of the tensor subsets and fusing factor subsets under a fine-grained model, and is operated on a cluster formed by a plurality of nodes; each node comprises a plurality of pattern processing units. Tensor decomposition on pattern processing unit equipment can fully utilize a powerful parallel computing capability and a paralleling resource generated in tensor decomposition; experimental results show that through the adoption of the method, executive time for acquiring tensor factors can be greatly shortened, the large-scale data processing capability is improved, and the problem that the computing resource is insufficient is well solved.

The invention discloses a knowledge base completion method based on variational interference and a tensor neural network. Tensor decomposition thought and a Bayesian framework are introduced into the knowledge base completion method, and the prior knowledge of implicit variables is considered. The interaction between the implicit variables is discussed and nonlinear expression is performed through the neural network, and consideration of nondeterminacy is added. The accuracy of the knowledge base method is improved obviously, and compared with the existing technology, the knowledge base completion method has been greatly improved.

The invention claims to protect a recommendation system and method for improving user role undifferentiated treatment and data sparseness, and belongs to the fields of data mining and information retrieval. The method comprises the following steps: firstly, in order to solve the problem of user role undifferentiated treatment, a time-slicing method is used to divide scoring time zones based on user scoring density, and at the same time, an information entropy theory is introduced to measure user roles in the time zones and dynamically divide the user roles, achieving hierarchical processing of scoring data; and secondly, for the problem of data sparseness, a tensor decomposition scoring prediction model based on ''user-item-role'' is constructed, missing values of the data are processed with a CP decomposition method to obtain a scoring prediction result and generate a target user recommending list. The invention effectively improves the problem generated by data sparseness, and improves the recommendation efficiency.

A method of performing an inference task on a knowledge graph comprising semantic triples of entities, wherein entity types are subject, object and predicate, and wherein each semantic triple comprises one of each entity type, using a quantum computing device, wherein a first entity of a first type and a second entity of a second type are given and the inference task is to infer a third entity of the third type. By performing specific steps and choosing values according to specific prescriptions, an efficient and resource-saving method is developed that utilizes the power of quantum computing systems for inference tasks on large knowledge graphs. An advantageous value for a cutoff threshold for a cutoff based on singular values of a singular value tensor decomposition is prescribed, and a sequence of steps is developed in which only the squares of the singular values are of consequence and their signs are not.