22results about How to "Good computing power" patented technology

The invention discloses a method for automatically searching an optimal modal fusion strategy. The method comprises the following steps: S1, constructing a multi-task learning loss function L (theta); s2, inner layer optimization: using a current fusion strategy parameter alpha, executing forward and backward propagation on training data, updating a model parameter theta, and minimizing a training loss function Ltr which comprises L (theta); s3, outer layer optimization: evaluating performance on verification data by using the updated model parameter theta, calculating a gradient about alpha, updating a fusion strategy parameter alpha, and minimizing a verification loss function Lval; and S4, the steps S2 and S3 are repeated until the preset number of iterations is reached or the model performance is not improved any more, and the obtained fusion strategy parameter is the optimal modal fusion strategy. According to the scheme, the fusion strategy can be dynamically adjusted according to different data set characteristics and modal missing conditions, so that the potential of the model in multi-modal learning is exerted to the greatest extent.

The invention discloses a reconfigurable computing array, a reconfiguration controller architecture, a chip and a reconfiguration computing method. The reconfigurable computing array comprises a main control module, a state machine module, a computing decoder module, a memory access resource controller module, a reconfiguration controller module, the reconfigurable computing array, an input buffer area module and an output buffer area module. The state machine module controls the whole reconstruction calculation process through state switching; the reconfiguration controller module configures the connection mode of PE units in the reconfigurable computing array according to the decoding result of the computing decoder module, and various computing functions are achieved so as to complete different types of data computing tasks. According to the method, the multiple computing units in the reconfigurable computing array are dynamically configured, so that a hardware architecture can be adjusted in real time according to different algorithm requirements, the reconfigurable computing array is accurately configured to complete different computing tasks, special hardware does not need to be designed for each task, and the computing efficiency is improved. Therefore, the flexibility of chip configuration and the resource utilization rate are improved.

The invention discloses a three-dimensional reconfigurable hardware acceleration core chip, belongs to the technical field of chips, and is characterized in that the three-dimensional reconfigurable hardware acceleration core chip comprises a reconfigurable operation array used for providing at least one unit-level calculation unit and at least one algorithm-level calculation unit; the storage array is used for storing operation data input by the AXI bus and output by the reconfigurable operation array; the controller set is used for controlling the at least one cell-level computing unit and the at least one algorithm-level computing unit so as to achieve cell-level computing operation and algorithm-level computing operation respectively and controlling operation data storage of the storage array. The three-dimensional reconfigurable hardware acceleration core chip based on static scheduling and a static data flow model is constructed by managing scheduling functions such as configuration decoding, reconfiguration control, calculation control, data distribution and storage control through an independent control system, and the chip realizes multi-dimensional reconfiguration of a space dimension, a time dimension and a resource dimension through storage and calculation decoupling.

Embodiments of the present specification provide a method, device, and database for selecting a computing node, wherein the method for selecting a computing node includes: obtaining multiple computing cost factors of a computing request and factor values ​​corresponding to each computing node, wherein the multiple computing cost factors include any two or more factors of node type, node load, and transmission cost; for each computing node, according to the factor weights corresponding to the multiple computing cost factors and the targeted computing nodes, weighted synthesis of the factor values ​​corresponding to the multiple computing cost factors is performed to predict the comprehensive computing cost corresponding to the computing request when it is executed on the targeted computing node; and selecting a computing node that executes the computing request according to the comprehensive computing costs corresponding to the computing request when it is executed on each computing node.

The invention relates to the technical field of artificial intelligence, and provides an operator fusion method and system, electronic equipment and a storage medium, and the method comprises the steps: receiving the operator number of to-be-fused operators and an equipment function identifier sent by a host, and the equipment function identifier is determined by the host based on the operator identifier of the to-be-fused operators; based on the device function identifier, loading a device function corresponding to the to-be-fused operator, the device function being developed in advance based on the to-be-fused operator, and being compiled and loaded into a memory of the device; and fusing the operators to be fused based on the number of the operators and the equipment function. According to the invention, automatic fusion during operator operation can be realized, the development amount of operator fusion is greatly reduced, and the efficiency of operator fusion is improved.

The present invention discloses a parking space detection method based on parking space marking points, comprising the following steps: step 1: marking the parking space entrance head marking points and the marking frame according to the parking space entrance; step 2: establishing a parking space detection model; step 3: pairing the marking points, and screening the parking space entrance according to the constraint conditions; step 4: based on the marked corner points, inferring the other two invisible corner points of the parking space to form a complete parking space, and determining the parking space type based on the parking space angle information; step 5: based on the inferred complete parking space, designing an occupancy classification m_DCNN model based on improved AlexNet; step 6: judging whether the parking space is vacant based on step 5; step 7: integrating parking space detection and occupancy classification information. The present invention combines parking space classification with marking point positioning to infer the precise positions of various parking spaces; designs the m_DCNN model, optimizes the network structure and computing performance, and further improves the accuracy of parking space detection.

The present application relates to the technical field of file access in a computer, and discloses a file access method and storage medium based on a GPGPU multi-channel register. By introducing the multi-channel access technology, each channel is equipped with a dedicated request queue cache and control logic; and then the multi-thread bundle parallel access is realized through an operand collection unit and an arbitration mechanism, so that the register file access efficiency will be significantly improved, the access conflict and delay will be greatly reduced, the pipeline stagnation waiting time will be shortened, and the overall computing performance will be significantly improved.

One embodiment of the present invention provides a semiconductor device having a novel structure. The semiconductor device includes: a first arithmetic device including a register; and a second arithmetic device including a storage circuit, a layer selection circuit, and an arithmetic circuit. The first arithmetic device and the second arithmetic device are provided in an element layer in which a plurality of second element layers are laminated on a first element layer. The register includes a flip-flop and a data retention circuit. The flip-flop and the operational circuit are arranged in the first element layer. The data retention circuit is disposed in each of the plurality of second element layers on the first element layer including the flip-flop. The memory circuit and the layer selection circuit are provided in each of a plurality of second element layers on a first element layer including the arithmetic circuit.

The application relates to a large-scale underwater explosion problem, in particular to a calculation method of underwater explosion load based on an LDGM and MPI. The calculation method of underwater explosion load based on the LDGM and MPI comprises the following steps: performing non-structural grid division on a water area geometric model, calculating data in the non-structural grid after spatial discretization and time discretization according to a LDGM second-order wave equation, performing regional decomposition by using METIS during the calculation process, dividing the non-structural network into multiple subdomains and distributing the subdomains to different processors, exchanging numerical flux on the subdomain boundaries between the processors, performing parallel processing, until time domain loop calculation is completed, and obtaining a pressure field of a flow field evolving with time. The LDGM is used to solve the flow field pressure load, can more accurately capture the strong gradient of a shock wave and local details of a cavitation area, avoids numerical dissipation and oscillation problems, and improves calculation efficiency through a parallel processing scheme.

The application discloses a long-term return type gold bull fund performance index ranking method and system, which is executed by a computer device and comprises the following steps: S1, acquiring public fund disclosure data and constructing a basic database; S2, generating a sample pool of participating fund companies by executing a preset sample screening algorithm; S3, precomputing the basic indexes of each trading day in the investigation period for the participating funds in the sample pool and constructing a participating fund basic index dataset; S4, calling a parallel index calculation process, reading the participating fund basic index dataset and performing feature extraction; S5, converting investor gain evaluation indexes and overall performance indexes into standard scores, performing weighted summation according to a preset scoring model to obtain comprehensive scores, and outputting a comprehensive ranking sequence of fund companies based on the comprehensive scores. The application combines the independent characteristics among various evaluation indexes, and greatly improves the overall computing performance in the environment of massive fund data through a parallel computing architecture.

The application discloses a rock damage and failure smooth molecular dynamics analysis method. The application combines particle failure with rock numerical damage simulation to construct a new rock damage numerical simulation mode. The particle discrete principle of the material point method is used to discretize the calculation material domain. The calculation domain is discretized into particles by using software such as ANSYS. The particles in the calculation domain are endowed with certain material properties. The Mohr-Coulomb criterion commonly used in rock mechanics is used as the failure criterion. When the material reaches the failure condition, the particles are subjected to failure treatment. The failed particles will be permanently failed. The method is simple and easy to implement. The high efficiency and stability of GIMP itself are utilized. The gradual damage and failure process of the rock is simulated by the failure process of the particles. The whole process of the rock damage and failure can be obtained. The research provides a certain reference for the understanding of the smooth particle dynamics method in rock damage numerical simulation.

The invention discloses a track line segmentation method for high-precision detection of the tail end of a track. The method comprises the following steps: constructing to-be-segmented track line image data; inputting the to-be-segmented track line image data into a track line segmentation network RTEW-Net for processing to obtain a train operation track line segmentation result; wherein the rail line segmentation network RTEW-Net is used for extracting rail line tail end preliminary features from the train operation rail line feature graph, performing feature detail maintenance on the preliminary features, and retaining the rail line tail end features in combination with the train operation rail line feature graph; the track line segmentation network RTEW-Net is obtained through training of a training set, and the training set is a track line image and a corresponding label. According to the invention, accurate segmentation of the track line is realized, and the running safety of the train is guaranteed.

The present invention relates to the field of Beidou short message technology, and proposes a data encryption method for Beidou short message transmission, comprising the following steps: S1 sets a hybrid homomorphic encryption framework, the hybrid homomorphic encryption framework includes a full homomorphic encryption algorithm and a partial homomorphic encryption algorithm; S2 receives a short message input by a transmitter, and pre-processes the short message; S3 performs a hybrid encryption operation on the pre-processed short message through the full homomorphic encryption algorithm and the partial homomorphic encryption algorithm in the hybrid homomorphic encryption framework, and adds differential privacy noise during the encryption operation to obtain an intermediate file; S4 generates a key according to the intermediate file and distributes it to the transmitter and the receiver. The data encryption method for Beidou short message transmission can work under limited bandwidth and high real-time requirements, can adapt to the security needs of different users, and maintain stable encryption performance in a complex and changeable use environment.

The application discloses a time delay reservoir computing method based on dispersion orthogonal polarization light feedback VCSEL, and the time delay RC system comprises an adjustable laser, a Mach-Zehnder modulator, a VCSEL, an optical isolator, an optical coupler, a polarization controller, a variable optical attenuator, a chirped fiber Bragg grating and a photodetector; the signal of an input layer is injected into the X polarization component of the VCSEL of a reservoir layer after being modulated by the Mach-Zehnder modulator; the emission light of the VCSEL is divided into two paths after passing through an optical circulator, one path is reflected back to the VCSEL by the chirped fiber Bragg grating after being rotated by the polarization controller, thereby constructing the reservoir layer of the time delay RC system; and the other path is transmitted to an output layer and converted into an electrical signal output by the photodetector. The application can further inhibit the time delay characteristics, thereby further inhibiting the generation of harmful resonance, and has stronger nonlinear characteristics and better computing performance.

The invention provides a remote sensing image classification method and system and a storage medium, and relates to the technical field of remote sensing image classification. Preprocessing images in the remote sensing image set; extracting multilayer features of the preprocessed remote sensing image through a pre-trained CNN model; performing dimension reduction processing on the extracted multi-layer features by applying PCA, extracting non-overlapping image blocks from the features after dimension reduction, performing spatial feature extraction on the extracted image blocks by applying 3D-CNN flow, and performing spectrum feature extraction on the extracted image blocks by applying CV-CNN flow; carrying out series fusion on the spatial features and the spectrum features; the fused features are enhanced through an SE attention mechanism; and inputting the enhanced features into an SVM (Support Vector Machine) for classification. The method integrates the advantages of multiple technologies, accurately extracts features, optimizes calculation, is high in adaptability, and effectively improves the classification precision and application value of the remote sensing image.

The invention discloses a method for anti-collision inspection of mining electric locomotives in the field of anti-collision inspection technology, comprising an explosion-proof camera, a Mini server, a storage module, a power supply module, a communication controller and a flameproof shell; step one: the explosion-proof camera collects real-time video data and transmits it to the Mini server, the user can set the duration of a single backup video and the total number of backup videos, the Mini server automatically backs up in the storage module according to the user's settings, when the total number of backed up videos exceeds the total number of videos set by the user, the Mini server can automatically clean up old videos, only one monocular camera is needed to realize the function, while ensuring accuracy, reducing costs; the system structure adopts a modular design, with strong reconfigurability, good expansibility and compatibility; the use of the Mini server can provide better computing performance and development support, and more functional frameworks can be built on the platform later.

The invention discloses an I-MU algorithm design method of a holder carrying a friction wheel launching mechanism. The I-MU algorithm design method comprises the following steps: S1, testing and selecting an attitude algorithm; s2, calibrating an IMU (Inertial Measurement Unit); s3, judging whether the data of the magnetometer are fused or not; s4, designing and selecting a low-pass filter; and S5, verifying the overall performance. The scheme designed by the invention is fused with the movement of the pan-tilt motor and the friction wheel, better attitude performance and control performance can be obtained, the code coupling degree is proper, and the calculation performance is optimized.

The invention provides a digital energy switch, a DESS system and a processing method of the DESS system. The switch comprises a switch CPU and a switch FPGA. One end of the switch CPU is respectively connected with the digital energy adapter and the digital energy concentrator through the switch FPGA, and the other end of the switch CPU is in two-way communication connection with an upper cloud end or a host. The switch FPGA comprises a data cache region, an information flow control module, an energy flow control module, an optical fiber communication module and an Ethernet communication module; according to the digital energy switch, the DESS system and the processing method of the DESS system, the limitation of a single CPU setting mode can be overcome, the task load borne by the CPU is reduced, the structure of the digital energy switch is optimized, the communication speed and the real-time performance of the switch are improved, and the anti-interference capability of the switch is enhanced.

This application relates to the fields of integrated circuit design and artificial intelligence hardware optimization technology, and discloses a radiation-resistant SRAM in-memory computing method and system with multi-source decision-making and partition self-healing. The method includes: steps for cross-layer benchmark archiving and pre-susceptibility topology initialization; steps for multi-source data acquisition and feature quantization processing in runtime; steps for semantic attribution and nonlinear weighted analysis of runtime fields; steps for spatiotemporal multidimensional state machine decision-making based on large model stage perception and thermodynamic constraints; steps for bit-plane decoupling mapping and asymmetric compression migration for large model features; and steps for multi-physics closed-loop self-healing and heterogeneous degradation reconstruction. The system corresponds to this method. This application solves the technical problems of difficulty in quantifying the performance degradation of in-memory computing chips under extreme radiation, difficulty in identifying fault mechanisms, and difficulty in controlling online self-healing.

The embodiment of the invention provides a cache use optimization method and device, a storage medium and a program product, and relates to the technical field of artificial intelligence, and the method comprises the steps: segmenting an input tensor into a plurality of sub-tensors according to the shape of the input tensor and the number of calculation units; and for each sub-tensor, a plurality of candidate strategy groups are selected based on the shape of the sub-tensor, and each candidate strategy group comprises a cache use strategy and a pipeline configuration strategy. And calculating corresponding predicted processing performance when convolution calculation is performed on the sub-tensor by adopting each candidate strategy group, and selecting an optimal target strategy group from the plurality of candidate strategy groups based on a plurality of obtained predicted processing performance. In this way, for input tensors of different shapes, the optimal cache use strategy and the optimal pipeline configuration strategy can be selected for convolution calculation, the efficiency of convolution calculation is improved, cache resources are fully utilized, and the method can be generalized to various scenes, so that the calculation performance of a convolution operator in various scenes is optimized.

The application discloses a micro-batch-based real-time statistical method and device, equipment and storage medium, and relates to the field of data statistics, which comprises the following steps: analyzing the received data to be processed, filtering the analyzed data, and obtaining processed data; dividing the processed data into several parts of processed data and inputting them into several preset concurrent statistical subunits to perform data statistics on the data input into the subunits to obtain temporary statistical results; determining several index identifiers corresponding to the temporary statistical results to group the temporary statistical results by the index identifiers, using a preset summary statistical subunit to perform summary statistics on the grouped temporary statistical results, and outputting target statistical results corresponding to the grouped temporary statistical results. In this way, the calculation capacity of statistical indicators can be optimized by different concurrent statistical subunits to support simultaneous statistical calculation under different dimensions and different time windows.

The present invention discloses a method and system for evaluating the stability of a converter grid-connected system considering periodic time variation, comprising: establishing an LTP state space model of a modular multi-level converter grid-connected system; selecting N+1 discrete points θ on a periodic interval [0, T]; j As the configuration point, at each discrete point, the derivative value l′ i (θ j ) is the element (j, i) to form a Chebyshev differential matrix; the periodic time-varying differential equations of the LTP state space model are converted into a set of algebraic equations using the Chebyshev differential matrix, and then a single-valued matrix is ​​obtained; all or part of the key eigenvalues ​​of the single-valued matrix are solved, and the stability of the modular multi-level converter grid-connected system is determined based on the key eigenvalues. The present invention can accurately evaluate the stability of the converter system considering the periodic time-varying characteristics, and has high efficiency compared with numerical methods such as time domain integration and is suitable for large-scale complex systems.