62 results about "Tensor representation" patented technology

The invention provides a spatial intelligent three-dimensional modeling method based on a two-dimensional structure design sketch image, and the method comprises the steps: carrying out the preprocessing of an input two-dimensional structure design sketch image, and obtaining a preprocessed design sketch image, extracting two-dimensional structure design features from the preprocessed design sketch image, and encoding the two-dimensional structure design features to generate structured tensor representation; deducing space components, function partitions and constraint logic based on structured tensor representation to construct a sketch semantic graph, nodes of the sketch semantic graph representing physical structure units, and edges of the sketch semantic graph representing connection relations and stress constraints among the physical structure units; determining a node embedding vector corresponding to each node in the sketch semantic graph so as to perform three-dimensional space coding on the node embedding vector to obtain a node potential vector, and generating a geometric reasoning sequence during three-dimensional structure geometric reasoning according to a mechanical logic definition on the basis of edges in the sketch semantic graph, and performing three-dimensional structure geometric reasoning based on the node potential vectors and the geometric reasoning sequence to generate a three-dimensional structure model.

The invention relates to the technical field of financial data analysis and artificial intelligence, in particular to a multi-source heterogeneous financial data fusion and intelligent analysis system which comprises a data standardization processing module, a time sequence event fusion module, a knowledge graph construction module, a relation reasoning module and a self-adaptive anomaly detection module. The data standardization processing module is used for converting heterogeneous financial data from different sources into unified tensor representation; the time sequence event fusion module adopts a double-clue cooperation mechanism to establish a mapping relation between continuous time sequence data and discrete events; the knowledge graph construction module extracts financial entities and relationships thereof, and constructs a multi-level knowledge graph; the relation reasoning module performs deep reasoning based on a graph attention mechanism; and the adaptive anomaly detection module dynamically adjusts the detection threshold according to the market environment. According to the system, implicit association in heterogeneous financial data can be deeply mined, market anomalies are recognized in advance, and comprehensive support is provided for financial decision making.

The invention provides a method for predicting miRNA-lncRNA-disease ternary correlation through deep tensor decomposition and hypergraph convolution, and relates to the technical field of miRNA-lncRNA-disease ternary correlation prediction. Comprising six steps of integration of multi-source heterogeneous data, generation of three-dimensional tensor representation, hypergraph convolution modeling high-order interaction, graph attention network feature refining, depth graph convolution network enhancement and correlation prediction. Node features in a graph attention self-adaptive refining similarity network are integrated, global structure learning is enhanced by adopting a depth graph convolutional network, and the combination can generate stable and information-rich embedding for final ternary correlation prediction, so that potential complex correlation among various biological entities such as diseases, genes and drugs can be accurately extracted, and the prediction accuracy is improved. The potential relation and mechanism between the biological entities are further disclosed, and comprehensive ternary correlation prediction is achieved.

The present application relates to a data processing method and system, in particular to a vocational education multi-source data federal governance method and system, belonging to the field of vocational education informatization and big data technology, which first acquires and preprocesses vocational education multi-source data through network crawler and OCR technology; then constructs a knowledge graph in the form of triplets and maps it to a multi-dimensional tensor space; then evaluates the compatibility of multi-source data through topological feature analysis, and performs feature fusion that maintains topological invariants; finally, based on a quantum probability field model, knowledge reasoning and services are provided, the present application uses tensor representation to improve the expression ability of complex relationships, ensures the quality of heterogeneous data fusion through topological persistence, and introduces quantum probability theory to handle uncertainty in knowledge services, effectively solving the problems of multi-source heterogeneous vocational education data, complex relationships and insufficient service accuracy, and providing a comprehensive data federal governance solution for the vocational education field.

The invention provides a monocular depth estimation method and product based on convolution compensation dual-channel self-attention, and relates to the field of computer vision. Comprising the following steps: converting an event flow of a target scene into three-dimensional tensor representation; obtaining event image fusion multi-scale spatial features based on the image of the target scene and the three-dimensional tensor representation; modeling spatial context correlation in a spatial dimension by utilizing event image fusion multi-scale spatial features through a context modeling self-attention branch to obtain a context modeling self-attention result; through a modal fusion self-attention branch, using the event image to fuse the modal correlation of the multi-scale spatial features in the channel dimension modeling image and the event, and obtaining a modal fusion self-attention result; and pixel-level depth prediction is carried out by using a context modeling self-attention result and a modal fusion self-attention result to obtain a depth map, so that complementary characteristics between an event and an image are fully mined, fine-grained depth fusion expression is realized, and depth estimation precision and generalization ability are effectively improved.

The invention discloses a multi-source data driven intelligent charging stop resource matching method and system and a medium, and relates to the technical field of charging stop resource matching, and the method comprises the steps: respectively executing the multi-source data collection of a user side, a resource side and an environment side, and building a multi-source data set; mapping the multi-source data set to a multi-dimensional tensor space; weighting each dimension in the multi-dimensional tensor space through a space-time dynamic coding mechanism to form a dynamic tensor representation reflecting a real-time state and a future prediction state; constructing a user intention-resource state coupling tensor; calculating a dynamic matching potential index; and establishing an intelligent matching result. According to the method and the device, the technical problems of insufficient matching accuracy and low efficiency caused by insufficient fusion of multi-source data in charging stop resource matching and lack of dynamic adaptation and space-time scheduling consideration in the prior art are solved, and the technical effect of improving the matching accuracy and efficiency of the charging stop resource and the user demand is achieved.

The invention relates to a method and a device for constructing multi-modal and multi-dimensional solid waste management data. The method comprises the following steps: acquiring a solid waste scene image and a remote sensing monitoring image of a solid waste management scene; respectively labeling the two types of images according to a preset labeling strategy, and converting the two types of images into a structured data set through data cleaning, data enhancement and unified tensor representation; based on the data set, driving a general large model through a solid waste field expert Prompt template, automatically generating multiple groups of questions and candidate answers, and forming a solid waste visual question and answer corpus after verification; and finally, evaluating a corpus by using a plurality of evaluation difficulties, and iteratively optimizing the large model and the labeling strategy according to a result. By creating a systematized, multi-modal and multi-task data construction scheme and evaluation benchmark in the field of solid waste disposal management for the first time, the problems of shortage of targeted and universal training data, lack of model evaluation mechanisms and the like in the field of solid waste management are solved, the intelligent level and efficiency in the field of solid waste management are greatly improved, and iterative innovative design is achieved.

The invention relates to the technical field of natural language processing text classification, in particular to a multi-label text classification method based on semantic representation enhancement and dynamic weighted depolarization contrast learning and application of the multi-label text classification method. Preprocessing data in the training data set to obtain input tensor representation; enhancing text semantic representation by fusing multi-layer hidden semantic representation of a depth model; secondly, an improved label graph convolutional network is constructed, regularization, layer normalization, residual connection and label perception attention pooling are introduced into a graph neural network, fine-grained representation of the relation between labels is achieved, and label-text interaction is enhanced; and finally, introducing depolarization weighted contrast loss to construct dynamic weighted depolarization contrast learning, endowing a negative sample with a higher weight, and reducing false negative sample interference, thereby overcoming label semantic overlapping, and aiming at solving the problem of how to enhance the characterization capability of the model in a multi-label semantic overlapping and label incomplete scene.

The invention discloses a generation method and device for a mixed precision quantization operator, and relates to the field of deep learning, and the method comprises the steps: converting a quantization program inputted by a user into an initial Qtile calculation graph based on Qtile tensor representation; performing equivalent transformation on the initial Qtile calculation graph to generate a plurality of candidate Qtile calculation graphs which are equivalent to the initial Qtile calculation graph; according to the reuse frequency of operator input data in each candidate Qtile calculation graph, determining an optimal data layout position and an optimal data loading strategy of each candidate Qtile calculation graph; on the basis of the optimal data layout, the optimal data loading strategy and a code generation technology based on a template, codes of all the candidate Qtile computational diagrams are generated, and the codes of the candidate Qtile computational diagrams are achieved through codes of mixed precision quantization operators. According to the invention, an optimized high-performance mixing precision quantization operator can be output.

The invention discloses a multi-modal denoising method for a hyperspectral image, which is characterized by comprising the following steps of: S1, respectively performing low-rank tensor representation on the hyperspectral image and a registered multispectral image by utilizing Tucker decomposition, and extracting core tensors of the hyperspectral image and the registered multispectral image; s2, establishing a correlation model between the hyperspectral core tensor and the multispectral core tensor through model-driven linear mapping or data-driven multi-layer perceptron network; s3, iteratively solving a core tensor, a factor matrix and correlation model parameters by adopting an alternating direction multiplier method; and S4, reconstructing a denoised hyperspectral image by using the optimized hyperspectral core tensor and factor matrix. Compared with the prior art, the method has the advantages that the spectral details of the hyperspectral image and the high-signal-to-noise-ratio spatial information of the multispectral image are fully mined and utilized, the restoration precision in the mixed noise scene is effectively improved through the double-Tucker decomposition framework and the core tensor association strategy, and the high-quality hyperspectral image is obtained.

The invention provides a flood storage and detention area multi-scheduling scheme comparison algorithm based on an AI large model, and relates to the technical field of water conservancy, and the algorithm comprises the steps: receiving and analyzing input data and a scheduling instruction, then automatically generating a plurality of candidate scheduling schemes, carrying out the efficient simulation deduction of the candidate schemes, and carrying out the multi-dimensional comprehensive evaluation and comparative analysis. And finally, generating an interpretable comparison result and carrying out interaction. According to the method, human semantic instructions are accurately converted into structured preference tags and JSON task description through semantic analysis, task formalization and context enhancement by using a large language model, and then the semantic tags are dynamically converted into specific weight coefficients of a reinforcement learning proxy award function, so that the learning efficiency is improved. Therefore, a high-level intention is directly mapped into a computable optimization target, a mixed generation mechanism combining a rule template and a reinforcement learning agent is driven, a diversified and physically feasible scheduling scheme represented by a three-dimensional tensor is output, and the problem of dependence on fixed experience is solved.

Apparatuses, systems, methods, and computer program products are disclosed for hardware geometric regularization. A method includes receiving training data comprising labeled data points. A method includes selecting a class of self-adjoint differential operator equations. A method includes selecting a set of orthogonal polynomials as a spectral basis. A method includes iteratively optimizing parameters of the differential operator equations based on the training data using a gradient-based optimizer to minimize an objective function. A method includes solving the optimized differential operator equation using a spectral method with the selected orthogonal polynomials to generate a reproducing kernel represented by a kernel tensor. A method includes outputting a machine learning model comprising the reproducing kernel combined with support vectors derived from the training data, the model configured to infer labels for unseen data points by estimating similarity measures using the reproducing kernel.

The invention relates to a dynamic three-dimensional head reconstruction method based on compact tensor representation and an electronic device, a compact tensor representation method is used to encode texture attributes of a 3D Gaussian model, static three planes are used to store appearance information of neutral expressions, and lightweight one-dimensional feature lines are used to represent dynamic texture details, so that the dynamic three-dimensional head reconstruction method based on compact tensor representation is realized. Therefore, the storage cost is obviously reduced, and real-time rendering is realized. The invention further provides an adaptive truncation opacity punishment and category balance sampling strategy to improve the generalization ability of the model among different expressions. Compared with the prior art, the method has the advantages of high dynamic texture capture precision, low storage requirement and high rendering efficiency, can realize 300FPS rendering speed and single person 10M storage while ensuring high-quality dynamic detail reconstruction, and is suitable for wider application scenes.

In some implementations, a device may obtain a training corpus, from a set of pre-processed log data, associated with an alphanumeric format. The device may encode the training corpus to obtain encoded data using a set of tokens. The device may calculate a sequence length based on a statistical parameter associated with the training corpus. The device may generate a set of input sequences and a set of target sequences based on the encoded data, where each input sequence and each target sequence has a length equal to the sequence length. The device may generate a training data set based on combining the set of input sequences and the set of target sequences. The device may train a deep neural network (DNN) using the training data set and based on one or more hyperparameters to obtain a set of embedding tensors associated with an embedding layer of the DNN.

The invention provides a visual SLAM (Simultaneous Localization and Mapping) algorithm based on multi-scale neural tensor representation, which belongs to the technical field of indoor scene real-time reconstruction, and can efficiently capture the global structure and fine details of a scene by utilizing a tensor decomposition technology, and reduce the memory usage amount and the calculation overhead at the same time. In addition, the invention further introduces an efficient pixel-based color query method, and each pixel only needs neural network forward regression once in the method, so that the real-time performance of the neural radiation field-based SLAM (Simultaneous Localization and Mapping) is further improved. Experiments on a synthetic data set and a real data set show that the method is superior to the most advanced method in the aspects of reconstruction quality and camera tracking precision, and the efficiency in the aspects of running time and memory use is very high.

The invention discloses an equipment data processing method and system based on semantic annotation, equipment and a medium, and relates to the technical field of deep learning, and the method comprises the steps: collecting multi-source heterogeneous equipment data, constructing a unified data set through analysis and annotation, fusing multi-source features, constructing equipment unified representation and an initial knowledge graph, and receiving an event stream in real time to generate new knowledge. The graph is dynamically updated through conflict resolution, and real-time updating and dynamic evolution of the knowledge graph are achieved. According to the method, accurate construction and intelligent evolution of the equipment knowledge graph are realized by constructing a flow of multi-source data semantic fusion, multi-dimensional feature tensor representation and dynamic conflict resolution, and the real-time performance, the accuracy and the high credibility of a knowledge system are ensured.

This arithmetic device is configured so as to comprise a contraction unit (1) that expresses a tensor of rank L (where L is an integer equal to or greater than 2) as the product of a plurality of minor tensors, each of which has a smaller number of elements than the number of elements in the tensor of rank L, and performs a contraction process on the plurality of minor tensors to acquire factor information present among the plurality of minor tensors.

The application provides a monocular depth estimation method and product based on convolution compensation double-channel self-attention, and relates to the field of computer vision. The method comprises the following steps: converting an event stream of a target scene into a three-dimensional tensor representation; obtaining event image fusion multi-scale spatial features based on an image of the target scene and the three-dimensional tensor representation; modeling spatial context correlation of the event image fusion multi-scale spatial features in a spatial dimension by a context modeling self-attention branch to obtain a context modeling self-attention result; modeling modality correlation of the event image fusion multi-scale spatial features in a channel dimension by a modality fusion self-attention branch to obtain a modality fusion self-attention result; and performing pixel-level depth prediction by using the context modeling self-attention result and the modality fusion self-attention result to obtain a depth map, so as to fully mine the complementary characteristics between events and images, realize fine-grained depth fusion expression, and effectively improve the depth estimation precision and generalization ability.

The present application relates to the technical field of motion data recovery, and provides a human motion data recovery method and device based on tensor representation, which can better maintain the spatial structure and time structure of the human motion data to be recovered by determining the third-order tensor of the human motion data to be recovered. Considering the time sequence of the human motion data to be recovered, a frame difference regularization term is introduced into the objective function, so that the non-smoothness in the recovery process of the human motion data to be recovered can be avoided. The objective function combines the tensor core norm term of the recovery variable and the frame difference regularization term of the recovery variable, simultaneously considers tensor tube rank minimization and frame difference smoothing, can significantly reduce the recovery error, and obtain a more superior recovery result.

This application relates to the field of voice interaction technology, and particularly to a voice-driven intelligent agent interaction control method and device for water conservation. The proposed scheme involves constructing an intent dialogue tree structure to manage semantic segments, thereby achieving contextual continuity and task succession in multi-turn voice interactions. Furthermore, through semantic tensor representation and tensor field simulation, the optimal attachment path for semantic segments is determined, and the current semantics are completed based on the parent node's response content. This allows for the generation of query vectors by matching plugins, guiding the intelligent agent to generate accurate responses. The device integrates modules for voice acquisition, semantic parsing, structure management, plugin scheduling, and voice broadcasting, making it suitable for voice interaction control in intelligent water-saving terminals.

The invention discloses a hyperspectral anomaly detection method based on prior coupling driving non-convex tensor representation, and belongs to the technical field of hyperspectral image processing. Aiming at the problem that background representation is not ideal due to the fact that an existing tensor representation method depends on a plurality of independent regularization items for background prior modeling and a loose convex substitute is adopted for approximation, a basic tensor representation model is constructed, and meanwhile, a prior-coupled non-convex background regularization item and an abnormal regularization item are constructed; constructing an augmented Lagrangian equation, converting the model into an unconstrained optimization problem, performing model iteration according to the augmented Lagrangian equation, and taking an abnormal tensor when the iteration is completed as an optimal abnormal tensor; and obtaining an anomaly detection result according to the optimal anomaly tensor. According to the method, the global low-rank priori and the local smoothness priori of the background are coupled in a non-convex regularization item, and the background priori is better described, so that the accuracy of anomaly detection is improved, and the false alarm rate is reduced.

This application relates to the field of EEG signal processing technology, and provides a method and system for artifact removal from EEG signals, including: acquiring multi-channel EEG signals to be de-artifacted, and constructing a three-dimensional tensor representation of the multi-channel EEG signals; constructing a channel map to characterize the spatial topological relationship of EEG electrodes based on the spatial distribution relationship between the electrodes corresponding to the multi-channel EEG signals; extracting three-dimensional features from the three-dimensional tensor representation using a three-dimensional convolutional neural network; performing cross-channel feature propagation and fusion based on the channel map during the extraction process; obtaining modulation weights for modulating the original complex spectrum based on the three-dimensional features and performing modulation to obtain the temporal representation corresponding to the modulation result; obtaining fused features based on the multi-channel EEG signals, the channel map, and the temporal representation; processing the fused features to obtain the de-artifacted EEG signal. This application can improve the artifact removal effect of EEG signals.

This invention discloses a hyperspectral fusion imaging method and apparatus based on low-rank representation of deep nonlinear transform tensors. The method includes the following steps: simultaneously acquiring a prior multispectral image while performing hyperspectral compression measurements in the observation scene, using both data points as inputs; establishing a novel low-rank regularization of deep nonlinear transform tensors for hyperspectral images based on tensor representation and deep learning theory to fully characterize the global high-dimensional low-rank correlation of hyperspectral images in the deep nonlinear transform domain; using this regularization term as the objective function, and modeling the compression imaging and spectral degradation processes as two data fidelity terms respectively, constructing a fusion imaging model and a loss function; minimizing the loss function through a learning algorithm to optimize the model, and fusing and reconstructing a complete hyperspectral image. The novel method and apparatus proposed in this invention can achieve high-precision fusion imaging of the observation scene without manual intervention or explicit intermediate steps.

To execute a plurality of times of multiplication of a Galois field at a higher speed.SOLUTION: The arithmetic circuit executes a first operation corresponding to p multiplications executed in series, each of the p multiplications being represented by p (p is an integer of 2 or more) third order tensors that input two elements of a Galois field and output one element that is a multiplication result of the two elements, by an AND operation that calculates an AND value of a plurality of elements used in the p multiplications, and an XOR operation based on the AND value and a contraction tensor obtained by contracting a 3p tensor obtained by a direct product of the p third order tensors.SELECTED DRAWING: Figure 2

A tensor representation of a machine learning inferences to be performed is built by forming complementary tensor subsets that respectively correspond to complementary subsets of one or more leaf nodes of one or more decision trees based on statistics of the one or more leaf nodes of the one or more decision trees and data capturing attributes of one or more split nodes of the one or more decision trees and the one or more leaf nodes of the decision trees. The complementary tensor subsets are ranked such that a first tensor subset and a second tensor subset of the complementary tensor subsets correspond to a first leaf node subset and a second leaf node subset of the complementary subsets of the one or more leaf nodes.

The application relates to a hyperspectral image anomaly detection method based on high-order tensor representation, introduces a promotion operation, performs high-order tensor expression on data, utilizes a tensor low Tensor-Train rank approximation framework, embeds suitable regularization constraints into hyperspectral prior knowledge while retaining the overall structure of the hyperspectral image, adopts a spatial spectral total variation norm regularization term expression for the spatial dimension segmentation continuous prior of the background tensor, adopts a logarithmic sum norm regularization term expression for the spectral dimension low rank prior of the background, and adopts an L 2,1 norm regularization term expression for the group sparse prior of the anomaly target tensor. Finally, in the tensor framework, the target equation is subjected to a convex optimization through an alternating direction multiplier method, the anomaly target is effectively extracted, the principle is clear, the experimental verification effect is superior, and the robustness is strong. The application provides theory, a model and support for tensor expression model processing and analysis of hyperspectral remote sensing images.

The invention belongs to the technical field of image processing, and particularly relates to a hyperspectral image dimension reduction and classification method and system based on discriminant spatial-spectral tensor locality preserving projection and a medium, and the method comprises the steps: inputting hyperspectral image data, and constructing three-dimensional tensor representation; performing blocking processing on the tensor, and extracting local tensor blocks; calculating a weighted covariance matrix based on a nuclear spectrum angle weighting strategy; adopting affine invariant Riemannian measurement to construct a weight matrix of the adjacency graph; solving the multi-modal projection matrix through an alternative optimization algorithm to realize dimension reduction; and finally, inputting the dimensionality-reduced features into a classification model and outputting a classification result. According to the method, by fusing the spatial-spectral tensor representation and the Riemannian geometry, the problems that a traditional method destroys spatial coherence and is sensitive to noise are solved, and the dimension reduction effect and the classification precision are remarkably improved.

The application discloses a kind of based on physical event anchoring and space-time decoupling's sea transport freight right fraud real-time blocking system, comprising: physical event four-dimensional tensor construction module: for building the physical event including space coordinates, time series, equipment signature and event type four-dimensional data tensor representation;Quantum fingerprint anchoring module: for generating anti-copy unique identification to physical event data by 12-bit quantum circuit;Helve space-time topology engine: for calculating the continuity score of the physical event of generating anti-copy unique identification;Three-level risk response execution module: for automatically triggering on-chain blocking operation according to physical event continuity score, three-level risk response is judged;Self-evolution risk control system: for dynamically optimizing rule weight and freight right entropy model by fraud mode clustering, generates the defense mechanism of triggering on-chain blocking operation.The system marks that international trade trust mechanism has entered the trend and possibility of new era of quantum level credibility.

The invention discloses an event data processing method and device based on decoupling enhancement fusion. The method comprises the following steps: acquiring event data; performing fusion compression processing on the event data, and randomly sampling a fixed number of event points to obtain a fusion event point cloud; respectively inputting the fusion event point clouds into the corresponding independent branches to obtain a spatial feature sequence, a time feature sequence and a combined spatial-temporal feature sequence; respectively carrying out feature aggregation and enhancement on the spatial feature sequence, the time feature sequence and the combined spatial-temporal feature sequence by adopting a space filling curve; carrying out fusion modeling on the enhanced spatial feature sequence, the time feature sequence and the combined spatial-temporal feature sequence by adopting an attention mechanism to obtain a unified global feature sequence; performing feature tensor on the global feature sequence to obtain a universal grid tensor representation; therefore, the problems of space-time isomerism of event data, limitation of feature extraction, manual scale design and poor cross-task migration performance are effectively solved.

The invention relates to the technical field of power grid risk assessment, and discloses a power grid risk assessment method based on integrated load flow calculation and topology analysis, and the method comprises the steps: collecting and preprocessing power grid operation data, and organizing the preprocessed data into a four-dimensional high-order tensor structure; constructing a power grid topological relation tensor network; constructing an optimization problem through an L1 norm form, and solving the optimization problem by using an augmented Lagrangian multiplier method; the complex coupling relation among time, space and parameters is revealed by analyzing the relation among different dimension factor matrixes; performing risk assessment on different granularity levels under a multi-scale tensor analysis framework, and coordinating analysis results of each level to form comprehensive assessment; designing a tensor completion estimation algorithm, constructing an optimization model by using low-rank hypothesis, and solving a complete data set through a tensor low-rank decomposition method; according to the method, the high-order tensor representation and tensor network decomposition technology is adopted, so that the dimension and complexity of data are effectively reduced, and the calculation complexity is reduced.