31 results about "Vector process" patented technology

A software-accessible special purpose register is architected into a processing unit in order to implement persistent vector multiplexer control of a vector-based execution unit. A persistent swizzle instruction is defined in an instruction set for the vector-based execution unit and is used to cause state information to be stored in the special purpose register such that the operand vectors processed by subsequent vector instructions executed by the vector-based execution unit will be selectively shuffled using the persisted state information. As a result, when multiple vector instructions require a common custom word ordering for one or more operand vectors, a single persistent swizzle instruction may be used to select the desired custom word ordering for all of the vector instructions.

Accommodating a processor to process a number of different data formats includes loading a data word in a first format from a first storage device; reordering, before it reaches the arithmetic unit, the first format of the data word to a second format compatible with the native order of the arithmetic unit; and vector processing the data word in the arithmetic unit.

The present invention belongs to the technical field of radar target detection and discloses a range-extended target adaptive detection method under inverse Gaussian texture complex Gaussian clutter for the target detection under a wideband radar system. The method comprises a step of establishing the binary hypothesis model of radar target detection, a step of using a spherically invariant vector process model to construct an inverse Gaussian texture complex Gaussian clutter signal, a step of using a generalized likelihood ratio to obtain a range-extended target adaptive detector under the inverse Gaussian texture complex Gaussian clutter, a step of determining the detection threshold of the adaptive detector, a step of obtaining radar echo data and determining the detection statistic amount of the radar echo data according to the adaptive detector, and a step of judging whether the radar echo data has a range-extended target according to the detection statistic amount and the detection threshold, and if the detection statistic amount is larger than the detection threshold, determining that the radar echo data has the range-extended target, otherwise, determining that the radar echo data has no range-extended target.

The invention provides a named entity recognition method, a named entity recognition device, a storage medium and electronic equipment. The method comprises the steps of obtaining a to-be-recognized target text, and determining a target segmented word of the target text; determining a character image of a target character contained in the target segmented word, and determining a glyph vector of the character image; generating a character vector according to the glyph vector and the semantic vector of the target character, and generating a word vector according to all character vectors in the target segmented word; and performing named entity identification processing on the target text based on the word vector of the target segmented word, and determining a named entity in the target text.According to the named entity recognition method, the named entity recognition device, the storage medium and the electronic equipment, glyph features are introduced in the word vector generation process and named entity recognition can be performed more accurately in combination with the glyph features and semantic features. The recognition accuracy can be improved, and the defect of inaccuraterecognition caused by insufficient modeling information of rare words can be overcome.

The invention relates to an emotion classification method and system. The emotion classification method comprises the following steps: according to a data flow interface provided by a social network platform, obtaining data with emoticons, and marking the data according to the emoticons; according to a CBOW-emotion dictionary training model, constructing an emotion word vector dictionary by taking the marked data as input; according to provided training data, querying the emotion word vector dictionary, converting the training data into first mathematic vectors, sending the first mathematic vectors into a classifier, and training a classification model; and according to the classification model, performing emotion classification on social network data. According to the invention, texts are represented by emotion word vectors, the word-to-vector process is free from manual intervention, the cost for manually constructing and selecting features by use of a conventional method is decreased, and the classification effect is better than that of the conventional classification method.

The invention discloses an LDPC linear programming decoding method based on an acceleration alternating direction multiplier method. The LDPC linear programming decoding method mainly solves the problems that an existing message class transmission algorithm has the error floor and can be easily influenced by a short link. According to the technical scheme, the LDPC linear programming decoding method comprises the steps that first, decoding parameters are initialized; second, iteration updating is carried out on an auxiliary vector, a solution vector and a Lagrangian multiplier vector in the decoding parameters in sequence; third, acceleration processing is carried out to correct the solution vector and the Lagrangian multiplier vector; fourth, according to the corrected solution vector, infinite norms of vectors corresponding to all check nodes are calculated, and the maximum infinite norm is worked out; fifth, whether the decoding process ends or not is judged according to the obtained maximum value and the number of times of iteration; sixth, the solution vector processed through the last time of iteration is output to serve as a decoded code word. The LDPC linear programming decoding method is high in convergence rate and free of the error floor, efficiency of a decoding module in a communication system can be remarkably improved, and the LDPC linear programming decoding method is applicable to the technical field of communication.

The invention discloses a method for comprehensively evaluating key parameter errors of complicated mechanical systems. The method includes rejecting useless parameters by means of analyzing evaluation influence factors, and then constructing systems for evaluating the key parameter hierarchical structure errors of the complicated mechanical systems on the basis of the idea of AHP (analytic hierarchy processes); converting sorting computation on evaluation indexes into pairwise comparison on the evaluation indexes by the aid of characteristic vector processes on the basis of improved Delphi processes, measuring the importance of the various evaluation indexes by the aid of scales, computing pairwise ratios and constructing judgment matrixes; allocating weights of indexes for comprehensively evaluating key parameters and then comprehensively evaluating and analyzing the key parameters by the aid of power exponent comprehensive evaluation processes; acquiring influence of parameters of bottom layers on parameters of each upper layer according to power exponent evaluation models and ultimately acquiring influence of the key parameters on the precision of the complicated mechanical systems by parity of reasoning. Vectors corresponding to characteristic values of the judgment matrixes are relative weight number sorting relation vectors of the evaluation indexes. The method has the advantage that the influence degrees of input parameters on the precision of the complicated mechanical systems can be effectively analyzed by the aid of the method.

The invention discloses an intelligent method and a system for monitoring a hydrant's safety state. The method is performed through the following steps: 101) acquiring the measuring information of all sensors and conducting time synchronization processing to all acquired measuring information, wherein the measuring information includes water pressure inside a hydrant well, the inclination angle of a well cover, or environmental brightness inside the well; 102) adopting principal component analysis to degrade the dimensions of the measuring information after time synchronization and extracting the characteristic vectors of the dimension degraded space in the same process, according to the values of the characteristic vectors, picking a plurality of the characteristic vectors as the characteristic vectors for current hydrant's safety state judgment; 103) utilizing historical data to learn the relationship between the characteristic vectors and the safety state so as to establish a machine learning model; and 104) based on the characteristic vectors processed in step 102) and by combining with the SVM recognition function built in step 103), identifying the current safety sate of the hydrant.

A data processing apparatus includes a register bank having a plurality of registers for storing vectors being processed; a pipelined processor for processing the stream of vector instructions; the pipelined processor comprising circuitry configured to detect data dependencies for the vectors processed by the stream of vector instructions and stored in the plurality of registers and to determine constraints on timing of execution for the vector instructions such that no register data hazards arise. Register data hazards arise where two accesses to a same register, at least one of said accesses being a write, occur in an order different to an order of said instruction stream such that an access occurring later in said instruction stream starts before an access occurring earlier in said instruction stream has completed. The pipelined processor includes data element hazard determination circuitry.

The invention belongs to the technical field of relation extraction, and discloses a bidirectional GRU relation extraction data processing method and system, a terminal and a medium, and the method comprises the steps: carrying out the preprocessing of a reference data set SemEval-2010Task 8; performing word vectorization on the corpus through an EMLo pre-training model; performing preliminary denoising processing on the word vectors through a multi-head attention mechanism; coding the word vector by using a Bi-GRU network to obtain a hidden layer vector containing context information in the sentence; the hidden layer vector serves as input and is transmitted to a keyword attention layer, and the attention weight is calculated by combining output of the hidden layer with the entity pair relative position feature and the entity hiding similarity feature; and inputting the hidden layer vector processed by the attention mechanism into a classification layer to obtain a final relation extraction result. Experimental results show that the model provided by the invention achieves the most advanced performance without any other NLP tools.

The invention discloses a two-dimensional DOA estimation method based on an L-shaped array in the technical field of array signal angle estimation. The two-dimensional DOA estimation method comprisesthe following steps: 1, establishing a time domain model of array receiving; 2, decomposing an actual steering vector, and processing the decomposed vector by utilizing a Hadamard product; placing thelast Q elements of the vector processed by the Hadamard product at the tail of the actual steering vector to construct a virtual steering vector; 3, constructing estimation of a received signal autocorrelation matrix according to the virtual steering vector, and carrying out eigenvalue decomposition to obtain estimation of a noise subspace; and 4, constructing a spatial spectrum function, and estimating the DOA of the incident signal according to spectral peak search. The underdetermined problem that the number of signal sources to be estimated is greater than the number of physical array elements under the condition of limited array elements is effectively solved. The complexity and hardware cost of equipment in practical application are reduced, and the signal DOA estimation precision is effectively improved.

The invention relates to a method for concealing time-domain errors on the basis of inter-frame modes and movement repair. The method is mainly technically characterized by comprising steps of classifying processes for concealing the errors according to inter-frame mode information of upper, lower, left and right correct receive blocks adjacent to a current damaged macro block and respectively repairing damaged movement vectors; acquiring reconstruction values of the damaged block in reference frames by the aid of a movement compensation process and repaired movement vectors. The method has the advantages that the inter-frame mode information of the correct receive blocks adjacent to the damaged macro block is sufficiently utilized, the movement vectors of the damaged block are repaired by the aid of the different processes for concealing the errors, the movement compensation process on the basis of sub-blocks is used for complicated areas, a boundary matching process is used for relatively flat areas, a zero movement vector process is used for static areas, accordingly, the movement vectors of the damaged block can be accurately recovered, an error concealment effect is improved, the algorithm complexity is proper, and the instantaneity of the method in mobile application can be guaranteed.

The invention belongs to the technical field of natural language understanding, and particularly relates to a multi-intention recognition method and system based on bert + bolstm + crf and xgboost models. According to the technical scheme, the bert is used for processing the preprocessed data set to obtain the dynamic word vector, and the dynamic word vector is different from a word vector obtained by using a word2vec or glove model in the past. Word vectors output by the Bert model have dynamic characteristics, and the problem that one word is polysemy can be solved. The word vector is converted into a sentence vector through bi < l > stm + crf, a bi < l > stm + crf model can process context text information with a long distance at the same time, and an optimal sentence vector prediction sequence is obtained through the relation of neighbor labels. An Xgboost model is used in the aspect of recognition of the subjective map, and the model is high in recognition precision and more flexible, so that the model is used in the subjective map. After all the ideograms are obtained, a TF-I DF model is utilized to select a standard intention, and the standard intention is taken as an intention judgment basis. And inputting the sentence vector processed by the bert + bi l stm + crf model into a new bert model, and finally outputting a sub-intention.

The invention discloses an image classification method based on a bi-directional neural network structure, comprising the following steps: 1. The Directional layer replaces the full-connection layer in a traditional convolution network, the image classification model based on bi-directional neural network is built, 2. network forward propagation is performed, and by adding transformation matricesL and R, the rectangular structure of the last convolution layer is preserved, steps 2, 3 are repeated, the bi-directional neural network is fine-tuned until the classification network converges. 5, the class number of the image is obtained through forward propagation on the trained model. A bidirectional neural network structure is used on the premise of keeping the depth and width of the networkconstant, the dimension of the feature is transformed by designing the transformation matrix, which ensures the matrix form of convolution, effectively preserves the structure information of the original feature space of the image, and avoids the loss of spatial structure information in the process of matrix drawing into vectors in the whole connection layer.

The invention discloses a program stream control method for a vector processor and a system. The method includes steps of outputting a vector data processing state when a vector data processing unit executes a vector data operation instruction outputted by a fetch unit; acquiring a the vector data processing state by the aid of a program stream control unit; generating a program stream changing mark according to the vector data processing state; starting to sequentially obtain instructions from program stream changing addresses by the aid of the fetch unit according to the program steam changing mark; and outputting acquired instructions to the vector data processing unit. The system comprises the fetch unit, the vector data processing unit and the program stream control unit, the vector data processing unit consists of a state generating unit used for generating the vector data processing state, and an output end of the state generating unit is connected with the program stream control unit. The program stream control method for the vector process and the system have the advantages that a program stream can be changed during processing, vector data processing time is saved, data processing volume is reduced, and data processing efficiency is high.

The invention discloses a method for measuring interruption duration of an embedded system with high precision, and relates to the technical field of computers. Comprising the following steps: S1, specifying an interrupt exception type corresponding to an exception vector number; s2, when the interrupt exception occurs, the system jumps to a general exception entry function of the system; s3, the general exception entry function judges exception types through exception type numerical values; s4, calling an interrupt vector processing macro function in the interrupt entry function; s5, at the inlet of the interrupt vector processing macro function, firstly executing the interrupt duration measurement starting macro function to obtain an input interrupt timestamp, then calling and executing the interrupt processing function to obtain an output interrupt timestamp, and finally executing the interrupt duration measurement ending timing and data statistics function macro function at the outlet of the interrupt vector processing macro function; and S6, after the interrupt processing is completed, returning to a system state. According to the non-invasive performance analysis method, hardware resources can be fully utilized and the debugging efficiency can be improved on the premise of not modifying a program function flow.

The invention discloses a synonym construction method and a related device, which are used for constructing synonyms, do not need to depend on enough user behaviors and are wide in application range. The method comprises the steps of obtaining a first corpus subjected to word segmentation processing, wherein the first corpus is a set containing synonym statements; processing the first corpus according to a preset word vector calculation model to obtain a word vector representing each word in the first corpus; processing the word vectors according to a preset supervised learning model, wherein the supervised learning model is used for improving word vector similarity among words with the same semantics; and determining synonyms according to the word vectors processed by the supervised learning model.

The present invention relates to high efficiency processor designs, in the field of digital signal processing and intelligent computing, a method for slide window vector processing is disclosed, includes a set of vector instructions, a vector instruction designates a first source vector to perform corresponding operation with a second source vector according to an operation type, a sliding windowoperation skip step number is specified by a value in a third scalar register, and result elements obtained by the operation are sequentially written into a result vector register. The invention alsodiscloses a device for sliding window vector processing. A method and apparatus for processing a sliding window vector, The traditional vector and windowed vector operations are uniformly represented,and the vector processing is realized by a unified vector instruction and control method, which improves the performance of the vector processing device, and can flexibly represent various sliding window vector operations, and meets the needs of many new tasks with window operations.

The invention discloses an FFT (Fast Fourier Transform) code bit inverse sequence algorithm vectorization implementation method and device based on shuffling operation. The method comprises the following steps of: loading data which needs to be subjected to code bit inverse sequence processing into respective q registers in p vector processing units VPE of a vector processing part VPU to form vector data; configuring a shuffling mode memory, and generating a plurality of shuffling rules to indicate the position of the data source of the current position; traversing and selecting each current shuffling mode; based on shuffling rules corresponding to the current shuffling modes, shuffling vector data in respective q registers in p vector processing units (VPEs) loaded to a vector processing unit (VPU) until all the shuffling rules are traversed; and finally outputting vector data obtained by shuffling operation. The method is simple to implement, high in implementation efficiency, low in power consumption and good in effect, and can give full play to the calculation performance of the vector processing component to implement simple and efficient FFT'code bit reverse sequence 'calculation.