204 results about "Vector element" patented technology

The invention discloses a natural resource big data off-line comprehensive analysis, annotation and proof system and method, and belongs to the technical field of natural resource informatization and geographic information processing. The system comprises the following components: a spatial data fusion module, which is used for realizing unified management and efficient processing of heterogeneous spatial data; the spatial index optimization module is used for realizing quick retrieval and high-performance query optimization of spatial data; the intelligent annotation and arrangement module is used for realizing low-code rapid configuration and business process automatic arrangement of different natural resource investigation scenes; the off-line analysis module is used for realizing buffer area analysis, space superposition and terrain analysis in an off-line environment; the real-time visualization module is used for realizing multi-level map display, track recording, vector element editing and multimedia evidence collection; and the data quality management module is used for realizing automatic quality inspection of geometry, topology and attributes and statistical analysis, version control and export management of annotation results.

Systems and methods are disclosed for memory protection for memory protection for gather-scatter operations. For example, an integrated circuit may include a processor core; a memory protection circuit configured to check for memory protection violations with a protection granule; and an index range circuit configured to: memoize a maximum value and a minimum value of a tuple of indices stored in a vector register of the processor core as the tuple of indices is written to the vector register; determine a range of addresses for a gather-scatter memory instruction that takes the vector register as a set of indices based on a base address of a vector in memory, the memoized minimum value, and the memoized maximum value; and check, using the memory protection circuit during a single clock cycle, whether accessing elements of the vector within the range of addresses will cause a memory protection violation.

The invention discloses an automatic verification method and system for a topological relation of ocean charting data, and relates to the technical field of charting verification, and the method comprises the steps: obtaining a chart vector element, splitting a broken line into a line segment unit according to a node, splitting a surface domain into a ring domain unit according to an outer ring and an inner hole, and extracting a control point; constructing fractal spectrums for the line segment units based on a multi-resolution box dimension method, comparing the fractal spectrums with normal spectrums, and marking complexity abnormal sections; constructing a polymorphic behavior matrix according to predefined geometric expectation and semantic rules, analyzing conflicts, and identifying topology anomalies; a tide level field is obtained, a coupling constraint field is constructed in combination with equal-depth closed loop hierarchical nesting, and tide depth coupling abnormity of a channel boundary and a closed loop is verified; a lighthouse, a navigation mark and a tide level station control point are screened, and connectivity abnormity is verified according to a semantic rule; and summarizing to generate a verification report. Fractal analysis, a semantic matrix, isodepth coupling and a communication rule are combined, break point errors are accurately distinguished, topological anomaly is detected in a fusion mode, and the time-space consistency of a sea chart and the connectivity of control points are guaranteed.

Disclosed embodiments relate to systems and methods for performing instructions to convert to 16-bit floating-point format. In one example, a processor includes fetch circuitry to fetch an instruction having fields to specify an opcode and locations of a first source vector comprising N single-precision elements, and a destination vector comprising at least N 16-bit floating-point elements, the opcode to indicate execution circuitry is to convert each of the elements of the specified source vector to 16-bit floating-point, the conversion to include truncation and rounding, as necessary, and to store each converted element into a corresponding location of the specified destination vector, decode circuitry to decode the fetched instruction, and execution circuitry to respond to the decoded instruction as specified by the opcode.

Methods, apparatus and systems for wireless sensing based on deep learning are described. For example, a described method comprises: transmitting a wireless signal through a wireless multipath channel of a venue, wherein the wireless multipath channel is impacted by a motion of an object in the venue; receiving the wireless signal through the wireless multipath channel of the venue, wherein the received wireless signal differs from the transmitted wireless signal due to the wireless multipath channel and the motion of the object; obtaining a time series of channel information (TSCI) of the wireless multipath channel based on the received wireless signal; computing a plurality of autocorrelation functions based on the TSCI, each autocorrelation function (ACF) computed based on CI of the TSCI in a respective sliding time window; constructing at least one ACF vector, wherein each respective ACF vector is a vector associated with a respective ACF comprising multiple vector elements each associated with a respective time lag, each vector element being a value of the respective ACF evaluated at the respective time lag; rearranging the at least one ACF vector into rearranged ACF data, wherein each ACF vector is a one-dimensional (1D) ACF-block; and performing a wireless sensing task based on a task engine to do a processing using the rearranged ACF data as an input.

Systems and methods for performing signed matrix operations using a linear photonic processor are provided. The linear photonic processor is formed as an array of first amplitude modulators and second amplitude modulators, the first amplitude modulators configured to encode elements of a vector into first optical signals and the second amplitude modulators configured to encode a product between the vector elements and matrix elements into second optical signals. An apparatus may be used to implement a signed value of an output of the linear processor. The linear photonic processor may be configured to perform matrix-vector and / or matrix-matrix operations.

The invention discloses a construction safety monitoring system for house construction, which belongs to the technical field of safety monitoring and comprises a construction monitoring data integration module, a construction monitoring image denoising module, a construction safety monitoring model building module and a real-time construction safety monitoring module. The method specifically comprises the following steps: respectively extracting three types of noise features and obtaining a weight coefficient, cascading a double-branch enhanced feature map and connecting the feature map with an initial feature map residual error, fusing a security mask, constructing a three-scale encoder and an asymmetric decoder, mapping a deep fusion feature into a noise feature and removing the noise feature, and providing a clear and complete image data support; dividing personnel feature map windows based on double scales, performing weighted fusion on same-scale features and aligning resolutions, extracting local, mesoscale and global features in parallel, weighting query vectors element by element by using importance weights of all positions, screening effective spatial features through convolution and gating weights, and obtaining a query result; and the real-time performance and the accuracy of house construction safety monitoring are improved.

The invention relates to a position precision evaluation method, device and equipment, a vehicle, a storage medium and a product. The method comprises the following steps: acquiring track data and visual data acquired in a driving process of the vehicle; generating vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data comprises at least one vector element; and based on the trajectory data, the visual data and the vector data, evaluating the position precision of the vector elements to obtain a position precision evaluation result. The internal evaluation of the position precision is realized according to the characteristics of the trajectory data, the visual data and the vector data, the evaluation of the position precision can be realized without the help of a high-precision map, and the accuracy and efficiency of the evaluation of the position precision are improved.

The invention provides a sparse operator cooperative computing architecture for scientific computation, and relates to the technical field of sparse computation.The architecture generates a first task through a block task generator according to non-zero element distribution of an original sparse matrix; coding and recombining the first task through a dot product scheduler to generate a second task, and distributing the second task to a segmented dot product calculation unit; and carrying out parallel calculation on the dot product of the plurality of vector elements in the second task through the segmented dot product calculation unit, outputting a plurality of intermediate results, and directly routing the plurality of intermediate results to the target storage position. According to the sparse operator cooperative computing architecture for scientific computing, through flexible data transmission and computing, the data transmission overhead is reduced, the computing efficiency is improved, and then the universality and performance of the sparse tensor core are improved.

The invention relates to the technical field of power generation prediction, in particular to a generating capacity prediction method of a zero-carbon zero-energy building photovoltaic system. The method comprises the following steps: obtaining the availability of each vector element of each particle according to the correlation between MA output values and moments in an initial prediction model at different moments and the distribution of each vector element of each particle; the speed weight of each particle is obtained by combining the difference between each kind of power generation influence factor data and the power generation influence factor prediction data at all moments and the distribution characteristics of different kinds of power generation influence factor data at the corresponding moments; according to the speed weights of different particles, a speed updating formula and a position updating formula are corrected, and a global optimal position is obtained; and according to the global optimal position corresponding vector and the initial prediction model at the real-time moment, obtaining an optimization prediction model, and obtaining power generation capacity optimization prediction data at the next moment. According to the method, the prediction model is optimized by obtaining the proper weight of the historical moment data during prediction, and the prediction accuracy is improved.

The invention discloses a three-dimensional visualization method, system and device based on heterogeneous spatial data fusion and dynamic sectioning, and a medium, belongs to the technical field of urban information model visualization, and aims to solve the technical problem of how to improve data fusion efficiency, sectioning response delay and mobile terminal frame rate in a million patch scene. Light-weight and high-interaction visual display of scenes such as smart cities and underground pipe galleries is realized, and the defects of multi-source spatial data splitting, high sectioning interaction delay and poor cross-terminal compatibility are overcome. According to the technical scheme, heterogeneous data dynamic normalization is carried out, wherein local coordinates of a building information model, WGS84 coordinates of a geographic information system and computer aided design data are unified to an Euclidean coordinate system through a space dynamic reference conversion algorithm, and normalized data are obtained; constructing a semantic data relation graph; generating a dynamic sectioning plane in real time; vector elements are highlighted in a linkage manner; and carrying out cross-terminal adaptive rendering.

According to the multi-source vector grid big data city remote sensing image fusion registration method, firstly, on a high-resolution remote sensing image, an object feature analysis method is adopted to extract ground features, and proper parameters are set for specific images and interested target ground features to optimize the extraction effect; and 2, extraction of a ground object target: not only is inherent characteristics of the object considered, but also comprehensive analysis is carried out by combining context and topology of the object and global characteristics of an image, and for different ground objects, a method of combining DSM data is adopted, and corresponding appropriate characteristic combination extraction is selected. And 3, searching homonymous elements between the extracted result and the existing vector data, matching the vector elements by adopting an approximation degree measurement model integrating position, shape and area characteristics, and performing image registration by utilizing the homonymous elements obtained by matching, thereby realizing accurate and efficient fusion registration of the multi-source vector grating big data urban remote sensing image.

The invention discloses a non-key unit fire risk assessment method fusing human factor analysis, and the method comprises the steps: building a fire human factor analysis time sequence analysis model, dividing assessment subspaces, and collecting human factor dynamic data; extracting fire risk parameters such as electrical equipment use frequency and the like, and fusing the fire risk parameters with corresponding human factor data features to form vectors; a vector element relationship is captured by means of a harmful multi-head attention mechanism architecture, and an attention weighted feature vector is generated. And deducing a human factor change trend by combining a time sequence analysis model according to the vector, and determining risk influence factors of each subspace. A risk propagation network is established, weights are set by considering physical positions of subspaces and personnel flow, and risk diffusion is simulated. And finally, through a comprehensive evaluation model, in combination with a risk diffusion simulation result, a human factor model deduction result and an attention weighted feature vector, determining an overall fire risk level of a non-key unit, thereby providing a scientific basis for fire prevention and control.

The invention provides a digital watermark processing method and device for vector data and electronic equipment, and the method comprises the steps: extracting the vertex of each vector element in the vector data, and obtaining a vertex set for any vector element; extracting a feature point set from the vertex set based on the shape feature of any vector element; all the feature points in the feature point set are mapped into an original coordinate system, an outermost convex hull of the feature point set is constructed based on the position relation of all the feature points in the original coordinate system, and the outermost convex hull is a minimum convex polygon surrounding all the feature points; establishing a relative Cartesian coordinate system based on the gravity center position of the outermost convex hull; coordinates of all vertexes in the vertex set in the original coordinate system are converted into relative coordinates in a relative Cartesian coordinate system, and a converted coordinate set is obtained; and embedding the digital watermark information into the converted coordinate set to obtain a target vector element containing the digital watermark information.

The invention discloses a spatial data labeling method for delineating a prospecting target area, which comprises the following steps: collecting mining area data, preprocessing to obtain spatial data, and dividing the spatial data by using a Thiessen polygon based on the spatial data to obtain a typical ore deposit research to-be-labeled area model area, the method comprises the following steps: obtaining an initial value according to geological structure features and element features of a model area of a typical ore deposit research to-be-labeled area, obtaining a first index and a second index based on the initial value, obtaining a first label and a second label by using the first index through an evidence weight method, and correcting the first index to obtain a correction label; and marking the model area of the typical ore deposit research to-be-marked area to obtain a marked area, expressing the marked area through polygonal vector elements, training the vector elements and the marked area by using a deep learning algorithm to obtain a spatial data marking model, and outputting a visual marking result. The method not only can improve the efficiency of spatial data annotation, but also has good interpretability.

Methods, apparatus and systems that relate to hardware accelerators of artificial neural network (ANN) performance that significantly reduce the energy and area costs associated with performing vector dot-product operations in the ANN training and inference tasks. Specifically, the methods, apparatus and systems reduce the cost of bit-level flexibility stemming from aggregation logic by amortizing related costs across vector elements and reducing complexity of the cooperating narrower bitwidth units.

The invention discloses a PDF document-oriented chart area accurate positioning and extraction method, and belongs to the technical field of information processing.The PDF document-oriented chart area accurate positioning and extraction method comprises the following specific steps of 1, analyzing a PDF vector instruction, analyzing a binary stream of a PDF document, extracting a bottom layer drawing instruction set, and extracting a bottom layer drawing instruction set; a set of structured vector elements is generated, each element containing a type, coordinates, and operator attributes. According to the method, through a vector-vision multi-mode fusion mechanism, the chart positioning precision is greatly improved, the error of an extraction region is remarkably reduced through sub-pixel-level boundary optimization, the PDF single page is converted into a vector element set and a visual feature map, and the vector element set and the visual feature map are processed together, so that the single page processing time is shortened, a cross-page chart splicing mechanism supports continuous document processing, and the processing efficiency is improved. And complex formats such as image-text surrounding and semitransparent covering can be effectively processed, the output structured data can be directly used for a data analysis assembly line, and the manual intervention cost is greatly reduced.

A computing system including a hardware accelerator configured to receive vector processing instructions from a processor. The vector processing instructions include an initial read address, an input increment size, and a vector processing operation. During vector processing iterations performed at vector processor tiles included in a vector processor tile array, the hardware accelerator reads vector elements into respective vector processor tiles in an input stream. The vector elements are read into the vector processor tiles from locations in memory that start at the initial read address and advance by the input increment size at successive vector processing iterations. At each of the vector processor tiles, the hardware accelerator computes a vector processing result at least in part by performing the vector processing operation on the vector element read into the vector processor tile. The hardware accelerator outputs the vector processing results from the vector processor tiles in an output stream.

A processor and a vector sort instruction for the processor to execute are provided, in which the vector sort instructions includes instructions for comparing a first element of a set of vector elements of a vector to a remainder of the set of vector elements; determining, based on the comparing, a control vector that specifies a respective sorted position for each element of the set of vector elements; and reordering the set of vector elements based on the control vector.

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.

An on-chip memory device within a hardware attention unit of a neural network accelerator, comprising: a) at least one dedicated on-chip memory block (1) for storing key vectors and at least one dedicated on-chip memory block (2) for storing value vectors, the memory blocks being located within the attention unit; b) a validity bitmap (3) with one bit per token position indicating whether valid key / value vectors are stored for the respective position; c) a store logic that, when calculating new token positions, stores the calculated key / value vectors element-wise in the on-chip memory and, after completion of all dimensions, sets the corresponding bit in the validity bitmap (3); d) a load logic that, in subsequent calculations, makes stored key / value vectors available from the on-chip memory with a latency of one clock cycle;unde) a hit check logic (10) which combinatorially (without clock cycle latency) indicates, based on the validity bitmap (3), whether valid K / V vectors are available for a queried token position.;

A vector processor includes a mask register configured to hold mask values, an instruction decoder configured to set dependency information included in instruction execution information when a decoded instruction is a subsequent instruction having data dependency with one or more previous instructions, and to set all-set information included in instruction execution information when a decoded instruction sets all of the mask values, a vector processing unit configured to execute vector arithmetic operations based on the instruction execution information, and to store in the data register a result of an arithmetic operation of a vector element corresponding to each mask value that is in a set state, and a dependency reset unit configured to reset the dependency information corresponding to a destination operand of the subsequent instruction and the mask register, when the all-set information is set for the mask register and the mask register is designated by the subsequent instruction.

The invention relates to the field of data lookup, and discloses a locality perception lookup table lookup method, which comprises the following steps of: S1, reordering between rows: clustering and reordering the rows of an input vector and a weight matrix according to the same quantized value to enable elements with the same quantized value to be continuously distributed; for a common matrix GeMV scene, cache locality is improved through rearrangement and subsequent blocking operation, and repeated loading of a lookup table LUT is reduced. And the data locality of the GeMV scene is remarkably improved through inter-line reordering and a virtual mapping strategy. On the basis of reordering of input vector element values, elements with the same value are aggregated, so that the LUT line loading times are reduced to 256 times at most from being in direct proportion to the input dimension d, and frequent replacement of the LUT lines in the WRAM is avoided; meanwhile, a weight matrix is divided into BR * BC sub-matrixes, the characteristic that vertical adjacent rows share an accumulator is utilized, multi-row sub-matrix calculation is loaded at a time, and virtual rearrangement maintains the row matching relation between an input vector and the weight matrix through index mapping.

A computing system including a hardware accelerator configured to receive vector processing instructions from a processor. The vector processing instructions include an initial read address, an input increment size, and a vector processing operation. During vector processing iterations performed at vector processor tiles included in a vector processor tile array, the hardware accelerator reads vector elements into respective vector processor tiles in an input stream. The vector elements are read into the vector processor tiles from locations in memory that start at the initial read address and advance by the input increment size at successive vector processing iterations. At each of the vector processor tiles, the hardware accelerator computes a vector processing result at least in part by performing the vector processing operation on the vector element read into the vector processor tile. The hardware accelerator outputs the vector processing results from the vector processor tiles in an output stream.

The invention provides an LDPC code alternating direction multiplier method decoding method and related equipment, and the method comprises the steps: determining a to-be-projected vector based on an LDPC code, and initializing an indication vector according to the to-be-projected vector; the parity of the number of the target elements is determined, and if the parity represents that the number is an even number, vector elements in the indication vector are overturned to obtain a target indication vector; if the parity representation number is an odd number, taking the indication vector as a target indication vector; and if it is judged that the to-be-projected vector corresponding to the target indication vector has the cut set, calculating a projection coefficient on the projection hyperplane, expanding the projection coefficient to a preset multiple, and correcting a projection result based on the expanded projection coefficient so as to determine a decoding result. In the embodiment provided by the invention, the expanded projection coefficient enables more projection results to fall on the position in the parity check multi-cell body and close to the projection hyperplane, and better decoding performance can be ensured under the condition of low iteration times.

Devices and methods of operating such devices are disclosed. Vector processing circuitry performs data processing in a plurality of parallel processing channels, wherein the data processing is performed in a subset of the plurality of parallel processing channels, the subset being determined by set bit values ​​of a vector predicate. Predicate monitoring circuitry is responsive to the vector predicate and generates a predicate summary value based on the bit values ​​of the vector predicate. A first value of the predicate summary value indicates that a sparse condition is true for the vector predicate, the sparse condition being true when the bit value of the vector predicate includes a set bit corresponding to a vector element at a higher index and an immediately following unset bit corresponding to a vector element at a lower index. A second value of the predicate summary value indicates that the sparse condition is not true for the vector predicate. Thus, improved predicate-controlled vector processing is supported.

The application discloses a kind of space data labeling methods for delineating prospecting target area, comprising: collecting mine data and pre-processing to obtain spatial data, based on spatial data using thiesen polygon division spatial data to obtain typical deposit research to be labeled area model area, according to the geological structure characteristics and element characteristics of typical deposit research to be labeled area model area to obtain initial value, based on initial value to obtain first index, second index, using first index to obtain first label, second label by evidence weight method, correct to obtain correction label for first index, label typical deposit research to be labeled area model area to obtain labeled area and be expressed by polygon vector element, using deep learning algorithm training vector element and labeled area to obtain spatial data labeling model, output visual labeling result.The method can not only the efficiency of space data labeling, but also has good interpretability.

Disclosed embodiments provide techniques for improved performance in processing vector instructions. A processor core is accessed. The processor core is coupled to a memory hierarchy, and the processor core includes one or more vector execution units (VUs), and one or more load store units (LSUs). The processor core includes a vector register file (VRF). The VRF includes multiple vector registers, and each vector register includes multiple vector elements. Vector elements that have a source or destination in contiguous memory are identified. Load store units (LSUs) take advantage of the contiguous memory condition by executing a vector load or vector store operation as a single memory access, requiring a reduced number of clock cycles. The single memory access satisfies each memory operation for each vector element within the vector register file.

A computing device, including a hardware accelerator configured to train a machine learning model by computing a first product matrix including a plurality of first dot products. Computing the first product matrix may include receiving a first matrix including a plurality of first vectors and a second matrix including a plurality of second vectors. Each first vector may include a first shared exponent and a plurality of first vector elements. Each second vector may include a second shared exponent and a plurality of second vector elements. For each first vector, computing the first product matrix may further include computing the first dot product of the first vector and a second vector. The first dot product may include a first dot product exponent, a first dot product sign, and a first dot product mantissa. Training the first machine learning model may further include storing the first product matrix in memory.