111 results about "Result vector" patented technology

The invention discloses a low sample hydro-generator fault diagnosis method based on transfer learning. The method comprises the following steps: collecting a current signal, a vibration signal, a temperature signal and a voiceprint signal; executing preprocessing; performing label labeling, and dividing the data into a source domain data set and a target domain data set; operating parameters of the hydro-generator are collected, processed and combined into working condition feature vectors; training a source domain base model on the source domain data set to generate a pre-training parameter set; establishing a physical constraint module, and binding the physical constraint module with the parameter updating process of the source domain base model; initializing a transfer learning model, and calling a physical constraint module to apply constraint to generate a physically constrained transfer learning model; performing increment fine adjustment; and outputting a diagnosis result vector through the transfer learning model after increment fine tuning. According to the method, transfer learning and physical constraints are combined, hydro-generator fault diagnosis is optimized through increment fine tuning, and the method has the advantages of high precision, small sample adaptability and physical consistency verification.

The present application discloses a data processing method and apparatus, a device, and a readable storage medium. The method comprises: combining M vision mapping vectors generated from media data and N text mapping vectors generated from text information into a mapping vector sequence, and in the mapping vector sequence, inserting a compression token vector between the M vision mapping vectors and the N text mapping vectors to obtain a vision compression sequence; performing attention processing on the vision compression sequence to obtain an attention result vector; determining a unit attention vector associated with the compression token vector in the attention result vector as a global compression vector, the vector length of the global compression vector being less than the sum of vector lengths of the M vision mapping vectors; and generating a question-answer result on the basis of the global compression vector and unit attention vectors associated with the N text mapping vectors. By using the present application, vision mapping vectors can be compressed, thereby reducing calculation costs while improving model performance.

The embodiment disclosed by the application belongs to the technical field of computing, and particularly relates to a method and device for performing stencil calculation and a computing device. The method comprises the following steps: acquiring a plurality of node data included in a calculation region, and grouping the plurality of node data into a data vector. According to at least one calculation condition included in the stencil calculation, the plurality of node data are subjected to judgment processing, and a judgment result corresponding to each calculation condition is obtained for each node data. For each calculation condition, the judgment result corresponding to the calculation condition is grouped into a judgment result vector, and the judgment result vector is obtained. Based on the data vector and the judgment result vector corresponding to each calculation condition, a calculation operation corresponding to each calculation condition in the stencil calculation is performed, and a calculation result of the stencil calculation is obtained. The application can convert the stencil calculation into vector calculation, and can improve the efficiency of the stencil calculation.

The invention discloses a multi-modal retrieval method and device, and relates to the technical field of multi-modal retrieval, and the method comprises the steps: obtaining a query vector corresponding to a user query text, and carrying out the query in a vector database according to the query vector, and generating a query result; vector codes corresponding to the multi-level content block structures corresponding to the multiple multi-modal table documents are stored in the vector database; the multi-level content block structure of any multi-modal table document comprises an atomic layer, a semantic layer and a relation layer; the atomic block comprises target element content and structure information of any cell in the multi-modal table document; the semantic block comprises aggregation content corresponding to any row, any column or any whole table in the multi-modal table document; the relation block comprises a text for describing the incidence relation between the multiple atomic blocks and the incidence relation between the multiple semantic blocks; the structure information comprises row and column indexes corresponding to each cell and a cross-modal association relationship among the multi-modal elements. According to the method, the multi-modal retrieval accuracy can be improved.

The embodiment of the invention provides a method and device for calculating a matrix multiplied by a vector, computing equipment and a storage medium, the matrix is a matrix of M * K, the vector is a vector of K * 1, and M and K are positive integers. The method comprises the following steps: converting an M * K matrix into a first tensor of M * (K / N) * N, N being a positive integer, Ngt; 1 and K are multiples of N; converting the vector of K * 1 into a second tensor of N * (K / N); using a tensor calculation kernel to carry out general matrix multiplication calculation of a first tensor of M * (K / N) * N and a second tensor of N * (K / N) in batches to obtain M result matrices; elements on diagonals of each of the M result matrices are added to result in a matrix of M * K multiplied by each element on a result vector of M * 1 of the vector of K * 1. According to the scheme, high-throughput and low-delay high-speed matrix vector multiplication (MMV) operation is realized by utilizing the tensor calculation kernel.

Aspects for forward propagation of a convolutional artificial neural network are described herein. The aspects may include a direct memory access unit configured to receive input data from a storage device and a master computation module configured to select one or more portions of the input data based on a predetermined convolution window. Further, the aspects may include one or more slave computation modules respectively configured to convolute a convolution kernel with one of the one or more portions of the input data to generate a slave output value. Further still, the aspects may include an interconnection unit configured to combine the one or more slave output values into one or more intermediate result vectors, wherein the master computation module is further configured to merge the one or more intermediate result vectors into a merged intermediate vector.

The application discloses an edge computing platform and system based on FPGA real-time target recognition detection, and relates to the field of microelectronic chips.The edge computing platform comprises an interconnected FPGA and DDR, and an ISP module, a pre-processing module, a VDMA module, an inference accelerator, a CPU and a character superposition module are arranged on the FPGA.The inference accelerator is used for deploying a preset convolutional neural network model, reading any frame of second image and weight parameter data about convolution kernels in the convolutional neural network model from the DDR, accelerating the execution of the algorithm of the convolutional neural network model by using a sphygmic array cluster module, and generating an inference result vector output to the DDR;the sphygmic array cluster module is integrated with a Winograd fast convolution algorithm and a multi-channel sphygmic array.Compared with the prior art, the application realizes the accelerated operation of the convolutional neural network model, and thus improves the inference efficiency.

The invention relates to the technical field of artificial intelligence and data processing, and discloses an intelligent information decision-making system based on reinforcement learning, and the system comprises an information perception module which is used for extracting concept nodes from an external information flow; the causal learning and evolution module is used for executing actions through a strategy network to update the dynamic causal graph and outputting a decision starting signal; the reward and strategy updating module is used for generating a mixed reward after the graph is updated and updating the strategy network; the candidate decision generation module is used for generating candidate actions by utilizing a strategy network and a causal graph after receiving the decision starting signal; the anti-fact deduction module is used for deducing the candidate actions by utilizing a causal graph to generate an expected result vector; and the decision refining module is used for integrating the candidate action and the expected result and outputting a final decision result. According to the method, through dynamic causal modeling, the problems of model stiffness and opaque decision are solved, and scientific and adaptive reliable decision is realized.

The invention relates to the field of cross application of electronic data processing and block chain technologies, in particular to a tax declaration data verification method based on a block chain. Comprising the following steps: outputting a global data abstract through an SHA-512 hash function; performing Hash processing to obtain field-level structural features, and generating structural perturbation terms in combination with the global data abstract; carrying out nonlinear transformation and weight weighted fusion, and packaging into a ciphertext; analyzing the tax declaration data ciphertext and the global abstract structure, and extracting a structural feature vector; matching the rule template library on the chain to generate a verification function set, outputting a Boolean type result vector, and generating a cross-chain key identifier; a comprehensive fusion credible score value is generated through nonlinear fusion, and whether the tax declaration data meet verification requirements in the aspects of structure verification and cross-chain consistency or not is judged. The technical problems that in the prior art, structural consistency multi-path verification is lacked, the self-adaptive rule generation capacity is weak, and a credible scoring mechanism does not have entropy perception discrimination capacity are solved.

The application provides an automatic disk IO method based on structural mechanics CAE simulation software, comprising the following steps: S1, constructing a memory-disk cooperative computing framework, and dynamically monitoring a data access state in a simulation solving process; S2, automatically exchanging non-active core data structures to a high-performance disk according to a data access frequency, a life cycle and a memory pressure; S3, loading back to the memory on demand when subsequent calculation needs to access the data exchanged to the high-performance disk; wherein the core data structures comprise a stiffness matrix, shape function data, an intermediate result vector and derivative physical quantities, and the exchange process is transparent to the solver. According to the application, the exchange process is highly transparent to the core computing logic of the solver, is automatically managed by the system, and does not need user intervention, so that efficient utilization of memory resources and stable operation of super-large-scale model simulation are realized.

This application provides an address comparison and checking method, scalar processor, and device for a compressed issue queue. The method acquires a first instruction and a second instruction from the compressed issue queue; and obtains an address comparison and checking result based on the address correlation between the first instruction and each second instruction. Furthermore, a comparison and checking result vector is formed, and this vector is updated in advance based on the instruction status in the compressed issue queue in the next cycle. This method, by obtaining the address comparison and checking result based on the address correlation between the first instruction and each second instruction, can flexibly support diverse memory access granularities with a small hardware area overhead while ensuring processor performance (timing). Additionally, by calculating the next state of the comparison and checking result in advance based on the enqueue / dequeue status of the issue queue determined in the current cycle, and updating it directly in the next cycle, it ensures a one-to-one correspondence between the address comparison and checking result and the new instruction position in the issue queue, while also optimizing timing.

The invention discloses a high-precision coordinate calculation method based on Matlab, and belongs to the field of coordinate calculation technology and data processing, and the method comprises the steps: firstly, obtaining the combination of all points and vectors through a specific combination generation function, and then initializing a result vector; aiming at each combination, calculating intersection point coordinates and parameters by virtue of an accurate geometric intersection point calculation function, storing the intersection point coordinates and parameters in a result vector, and distributing the intersection point coordinates and parameters to a working area according to a rule by virtue of a working area assignment function; then, a midpoint coordinate matrix is obtained by applying a midpoint calculation function based on the coordinates of the intersection points, the distance between the points is calculated, a clustering center is determined by adopting a clustering algorithm, points of an intersection point concentrated area are screened out according to distance characteristics in a cluster, and finally, high-precision coordinates are calculated through a weighted average calculation function in combination with priori knowledge and weights. According to the method, the coordinate calculation precision can be effectively improved, good noise immunity is achieved, priori knowledge can be fully fused, and the method is suitable for numerous fields with high coordinate precision requirements.

To provide a grouping device for graded answer sheets in a written examination, which contributes to suppressing variations in the grade results. [Solution] The grouping device for graded answers in a written exam includes a reading unit that reads graded answer data in a written exam, a morphological analysis unit that performs morphological analysis on the answer data, a vectorization unit that vectorizes the results of the morphological analysis into answer sentence vectors, a grouping implementation unit that groups answer data that matches the rule definition and groups answer data that does not conform to the rule definition based on the answer sentence vectors to generate groups, a group feature analysis unit that analyzes statistical information on the features of the generated groups, a scoring result analysis unit that analyzes statistical information on the scoring results of each group, and an output unit that outputs statistical information on the group features and statistical information on the scoring results.

Efficient scaling of in-network compute operations to large numbers of compute nodes is disclosed. Each compute node is connected to a same plurality of network compute nodes, such as compute-enabled network switches. Compute processes at the compute nodes generate local gradients or other vectors by, for instance, performing a forward pass on a neural network. Each vector comprises values for a same set of vector elements. Each network compute node is assigned to, based on the local vectors, reduce vector data for a different a subset of the vector elements. Each network compute node returns a result chunk for the elements it processed back to each of the compute nodes, whereby each compute node receives the full result vector. This configuration may, in some embodiments, reduce buffering, processing, and / or other resource requirements for the network compute node or network at large.

The invention relates to the technical field of power equipment state monitoring and intelligent operation and maintenance, and discloses a high-voltage cable operation state reliability evaluation method and system. The method at least comprises the following steps: converting a standardized multi-modal data set into a standardized evaluation data set by referring to a pre-constructed integrity evaluation system, and generating a training sample set and a verification sample set based on the standardized evaluation data set; constructing a composite neural network architecture comprising a feature extraction layer, a nonlinear transformation layer and a parameter optimization layer, and training the composite neural network architecture based on the training sample set to obtain an evaluation result vector comprising a plurality of reliability dimensions; and comparing the evaluation result vector with a verification sample set from a plurality of dimensions including numerical value consistency check, trend goodness of fit analysis and anomaly detection capability evaluation to obtain a structured evaluation report reflecting the reliability of the running state of the high-voltage cable. The intelligent operation and maintenance response efficiency of the high-voltage cable is remarkably improved.

The application discloses an AI scoring report intelligent generation and analysis method based on cloud cooperation, which comprises the following steps: step one, obtaining original scoring data of an object to be evaluated; step two, performing data preprocessing on the original scoring data; step three, obtaining a scoring feature tensor and a collaborative scoring feature sub-tensor set through decryption verification and feature construction; step four, generating a scoring prediction result vector through an improved Crossformer network based on the scoring feature tensor and the collaborative scoring feature sub-tensor set; step five, generating a structured AI scoring report text through template matching and field semantic mapping based on the scoring prediction result vector; step six, performing semantic consistency detection and logical conflict detection on the structured AI scoring report text; and step seven, updating the improved Crossformer network. The Crossformer network is used, and the credibility of AI scoring report intelligent generation is improved.

This application provides a text correction model training method and device, relating to the field of artificial intelligence technology. The text correction model training method includes: acquiring training data, which includes random erroneous text samples, near-phonetic erroneous text samples, and near-shape erroneous text samples; determining the input vector corresponding to each character in the training data, which includes a character representation vector, a positional representation vector, a pinyin representation vector, and a character shape representation vector; and training a pre-trained language model based on the input vector to obtain a text correction model. The text correction method includes: inputting the input vector corresponding to each character in the text to be corrected into the text correction model to obtain a text correction prediction result vector; and decoding the text correction prediction result vector to obtain the corrected target text. This application can not only reduce the cost of text correction but also improve the efficiency and accuracy of text correction.

This invention relates to an improved spatially smoothed source angle estimation method and apparatus based on a coprime linear array. The method includes: acquiring a signal from a source under test using an augmented coprime array to obtain a received signal; vectorizing the covariance matrix of the received signal, sorting the resulting vectors according to the element positions of the uniform linear array, and processing them according to a data processing strategy to obtain a virtual signal received by a virtual array; uniformly dividing the virtual array into overlapping virtual subarrays based on a pre-constructed spatial smoothing rule, and calculating the spatial smoothing covariance matrix of the received signals from the virtual subarrays; performing eigenvalue decomposition on the spatial smoothing covariance matrix using a predefined spatial spectrum estimation method to obtain a spatial spectrum function; and performing spectral peak search on the spatial spectrum function to determine the source angle of the source under test. The method provided by this invention improves the array degrees of freedom and the accuracy of source angle estimation by constructing a larger number of virtual array elements.

A method, system, device and medium for evaluating the contribution of participants in federated learning, the method evaluates the contribution of participant updates to the convergence of the global model from the perspective of model update similarity, uses the projection size and the included angle relationship of the local model update of the participant in the convergence direction of the global model to calculate the contribution value of a participant at a certain training round; the contribution is an evaluation of the uploaded model of the participant, including the data contribution and model training contribution of the participant, obtaining the contribution percentage normalization result of each participant in this federated learning, and forming a result vector evaluation result to provide a basis for constructing an incentive mechanism to ensure the fairness of federated learning; the system, device and medium based on the projection for evaluating the contribution of participants in federated learning, realize reasonable evaluation of the contribution of participants, and do not need test data set, have the advantages of simple and efficient evaluation, high precision.

The invention discloses a sparse matrix vector multiplication-oriented hardware accelerator, sparse matrix vector multiplication-oriented hardware accelerator equipment and an application method, and the hardware accelerator comprises an off-chip storage system, an on-chip network used for carrying out data exchange routing, and an on-chip processing system used for executing access and multiplication calculation and matrix in-row element merging, the off-chip storage system comprises HBM channels used for storing five types of data of a column index, a row index, a vector value, a matrix value and a result vector, each HBM channel comprises an HBM stack and a memory controller, and the HBM channels used for storing the vector values are connected with an on-chip network through second-level caches. And the other HBM channels are directly connected with the on-chip processing system. The method aims at improving on-chip data reuse of the hardware accelerator for sparse matrix vector multiplication, reducing off-chip memory access times and improving performance and energy efficiency performance of the hardware accelerator.

The embodiment of the invention provides an intelligent material placing method, device and equipment and a computer readable storage medium. The method comprises the following steps: acquiring a detection result vector, and generating a target placement matrix; generating a target mapping matrix based on the target placement matrix; generating a target collision-free track based on the target mapping matrix and the obtained target minimum beat; and according to the target collision-free track, target material placing operation is executed. In this way, the user experience is greatly improved.

The present invention discloses a method and device for array SNSPD readout based on compressed sensing, the method comprising: (1) constructing a deep learning network integrating compressed sensing sampling and reconstruction; (2) using natural scene pictures as sample inputs for deep learning network training; (3) forming a compressed sensing sampling matrix S according to the deep learning network. r 、S c ; (4) Compressed sensing sampling vector S ri ,S ci Load the bias current onto the array SNSPD; (5) Combine the outputs of all pixels in the array SNSPD and obtain the total number of pixels in the array SNSPD that respond based on the output signal amplitude. i ; (6) Repeat steps (4) and (5) until the sampling matrix S r 、S c The rows of are traversed to form the sampling result vector X={x i} T (7) The sampling result vector X is input into the preliminary reconstruction sub-network in the trained deep learning network, and the signal output by the encoding and decoding reconstruction sub-network is the reconstructed signal Y. The present invention is suitable for simultaneous response of multiple pixels, has a simple circuit and a high speed.

The embodiment of the invention provides a reasoning result obtaining method, electronic equipment and a computer readable storage medium, in the reasoning result obtaining method, after the electronic equipment obtains a query text, the query text is converted into a mark sequence, and the mark sequence is sent to the electronic equipment; then obtaining a first query matrix, a first key matrix and a first value matrix in a pre-filling stage according to the mark sequence, and executing dot product operation, normalization operation and weighted summation operation among the first query matrix, the first key matrix and the first value matrix in parallel; next, in a decoding stage, the matrix vector operation, the normalization operation and the weighted summation operation of the current result vector, the second key matrix and the second value matrix are subjected to parallel calculation, and finally, the next mark of the reasoning result is obtained according to the result of the weighted summation operation in the decoding stage, so that the next mark of the reasoning result can be obtained through parallel calculation in the pre-filling stage. The calculation delay is reduced, and the calculation bottleneck is relieved; in the decoding stage, through parallel computing, the bandwidth utilization rate is improved, and the memory access bottleneck is relieved.

According to the continuous sign language recognition method based on visual text prompt guidance, frame-level semantic conditional fusion is carried out before fusion, and on the premise that an external sensor and a skeleton pipeline are not introduced, the frame-level semantic conditional fusion is carried out; frame-level visual feature sequences extracted by a video encoder are constructed in parallel, and are averagely pooled to obtain a video-level visual prompt vector and a frame-level text prompt vector obtained by a text prompt extraction module. Unified linear projection and splicing are completed in the prompt guide fusion module, visual-text prompt vectors are generated through a multi-layer perceptron, broadcast copying is carried out along the time dimension, layer normalization is carried out to complete frame-by-frame fusion after the visual-text prompt vectors are added with feature residuals output by a main model of each frame, and then the visual-text prompt vectors are input into an encoder and a CTC to be subjected to end-to-end training; the discriminability and the time sequence consistency of feature expression are improved, so that the synchronous improvement of identification optimization and feature optimization is realized, and the robustness and the identification performance of the model are enhanced.

The invention provides a gene data lookup table coding method and device based on in-memory calculation. The method comprises the following steps: splitting three-dimensional lookup table data, and sequentially storing the split three-dimensional lookup table data into a plurality of SRAM (Static Random Access Memory) arrays; the method comprises the following steps: acquiring an input sequence of gene data with the length of L, and arranging the input sequence into L-2 triads (addr1, addr2, sym) according to the position of a symbol in the input sequence; determining a target array from the plurality of SRAM arrays according to addr1 in the triads, transmitting N triads to the target array of the triads at a time, and determining a to-be-searched data row from all data rows of the target array according to addr2 in the triads; three dimensions of the lookup table respectively correspond to three dimensions of an addr1 one-dimensional array, an addr2 one-dimensional array and a sym one-dimensional array; the target array completes comparison between all symbols of a data row to be searched and sym in the triple in a single period through a single-period byte search operation, and outputs a comparison result vector; and recording positions representing that the comparison results are the same in the comparison result vector to an output buffer area of the target array, and replacing symbols sym of the input sequence with the positions.

Embodiments of the present disclosure provide a method, apparatus, computing device and storage medium for calculating matrix-vector multiplication, where the matrix is an M*K matrix and the vector is a K*1 vector, M and K are positive integers. The method comprises: converting the M*K matrix into an M*(K / N)*N first tensor, where N is a positive integer, N>1, and K is a multiple of N; converting the K*1 vector into an N*(K / N) second tensor; using a tensor computing core to batch-process the general matrix multiplication of the M*(K / N)*N first tensor and the N*(K / N) second tensor to obtain M result matrices; and adding the elements on the diagonal of each of the M result matrices to obtain each element on the M*1 result vector of the M*K matrix multiplied by the K*1 vector. The above scheme uses a tensor computing core to implement high-throughput, low-latency high-speed matrix-vector multiplication (MMV) operation.

The application provides an optical computing device, method, equipment and storage medium. The device comprises a processor configured to obtain a target vector and a target matrix to be subjected to multiplication calculation; the processor is further configured to decompose the target matrix to obtain at least one sub-matrix and a weight vector; a modulator configured to configure a phase difference of an optical computing unit in a corresponding optical computing array based on an element attribute of a matrix element in the sub-matrix; a laser configured to generate an optical signal; the modulator is further configured to configure a phase of the optical computing unit in the corresponding optical computing array based on an element attribute of a vector element in the target vector; an optical computing array configured to receive the optical signal and output a corresponding optical computing result; and the processor is further configured to perform weighted calculation on the optical computing result based on the weight vector to obtain a calculation result vector of the target vector and the target matrix. The application can reduce the complexity of the configuration of the optical computing unit and avoid information loss.

The invention relates to the technical field of image processing, in particular to a geographic image semantic segmentation evaluation method and system based on road element edge accuracy, and the method comprises the steps: vectorizing a semantic segmentation map, and carrying out the resampling of the vectorized semantic segmentation map, and obtaining a segmentation result vector diagram and a truth value vector diagram; based on the segmentation result vector diagram, obtaining a pairing vector in the truth value vector diagram based on a vector distance, and constructing a pairing vector set; marking correct or wrong labels of the midpoints of the segmentation result vectors according to the relationship between the distance between the midpoints of the segmentation result vector diagrams and the corresponding contour in the true value vector diagrams and threshold values, and obtaining road edge accuracy evaluation index values according to the number of the correct labels; and evaluating a road semantic segmentation effect in a map acquisition process by using the road edge accuracy evaluation index value. According to the method, the workload of manual checking and editing of the road element segmentation result can be judged to a great extent, so that the geographic element collection efficiency is ensured.

The invention discloses a fault decision-making method for a fault diagnosis system, and the method comprises the following steps: S1, carrying out the feature extraction of multi-modal data: enabling the multi-modal data to enter a neural network operator (100) in a tensor form, and carrying out the hierarchical feature extraction, thereby obtaining a feature graph F; s2, calculation of a weight vector G: inputting the feature map F into a gating network (210) to generate the weight vector G, inputting the feature map F into an expert network (220) to generate a diagnosis result vector E, and obtaining an output result Y by an internal hybrid expert layer (200) according to the weight vector G and the diagnosis result vector E; and S3, generation of a structured fault result: the local expert system (300) performs reasoning on the output result Y in combination with a preset rule base to generate the structured fault result. According to the fault decision-making method for the fault diagnosis system provided by the invention, the calculation efficiency and the interpretability of the model can be improved while high precision is ensured.

A data processing device comprises a round mask generator, a controller configured to control values and a processor configured to iteratively process a vector of values, where each iteration comprises receiving a respective input vector, generating a processing result vector by applying a predefined processing algorithm to the input vector, and, in reaction to that the control value associated with the iteration indicates that the iteration is a dummy iteration, outputting the input vector re-masked with the round mask associated with the next iteration of the sequence of iterations and, in reaction to that the control value associated with the iteration indicates that the iteration is a real iteration, generating a masked processing result vector by masking the processing result vector with the round mask associated with the next iteration and outputting the masked processing result vector.