14results about How to "The acceleration effect is obvious" patented technology

The present application relates to an online batch data processing method and device for training deep neural networks, the method comprising the following steps: sampling a data batch from a given dataset; selecting a subset S from the data batch by using an orthogonalization-based objective function; training a target deep neural network based on the subset S. Compared with the prior art, the present application uses an orthogonalization method to remove the redundancy between subsets, provides an objective function based on orthogonalization de-redundancy correlation to measure the diversity and representativeness of the subsets at the same time, and provides an efficient optimization method based on the objective function and a corresponding data selection algorithm, which can efficiently select diverse and representative subsets.

This application discloses a data processing method, apparatus, neural network accelerator, and storage medium. The apparatus includes: a conversion processing module, configured to perform a first matrix transformation operation in the Winograd algorithm on the corresponding image data for each channel in a feature image, and perform a second matrix transformation operation in the Winograd algorithm on the corresponding convolution kernel, to obtain the corresponding converted image data and convolution kernel; at least one convolution processing unit, including a number of calculation groups matching the number of channels in the feature image, each calculation group supporting the execution of a dot product operation in the Winograd algorithm on the converted image data and convolution kernel corresponding to one channel in the feature image, to obtain the corresponding dot product data; the conversion processing module is further configured to accumulate the dot product data corresponding to different channels in the feature image, and perform a third matrix transformation operation in the Winograd algorithm on the obtained accumulated data, to obtain the convolution processing result corresponding to the feature image.

The invention provides an underground coal mine drill site object detection model construction method and device, and the method comprises the steps: obtaining a training data set based on a coal mine data set; an improved YOLOv5 model is determined, the improved YOLOv5 model replaces a Focus structure in a backbone network with a common convolution structure, a Swish activation function is replaced with a Leaky ReLU activation function, and a self-adaptive anchor frame calculation module at an input end is replaced with a data self-adaptive anchor frame calculation module; training the improved YOLOv5 model through the training data set to obtain an initial detection model; and performing iterative processing including pruning, quantification and training on the initial detection model to obtain a coal mine underground drill site object detection model supporting DPU hardware deployment. A coal mine underground drill site object detection model with high accuracy and high recognition speed is obtained, and a target detection method supporting DPU hardware deployment and light weight is realized.

The invention relates to the technical field of neural network acceleration, and relates to a CPU-FPGA-oriented graph neural network training acceleration method and system. The method comprises the steps that a CPU stores complete graph feature data and distributes different feature data to storage modules of different FPGAs through data channels according to a load balancing algorithm; partitioning the GNN network model by the CPU, and matching a partitioned model operator with a corresponding FPGA (Field Programmable Gate Array); the CPU completes issuing and sampling of a partition model according to a sampler algorithm, and sends the partition model to the FPGA; reading feature data after the FPGA is started, completing a calculation task by a load execution module, and writing a result back to a CPU memory; the CPU completes data updating and weight updating; and resending the updated weight data to the corresponding FPGA, and carrying out iterative training. According to the invention, operation of the graph neural network on a CPU-FPGA platform is successfully realized, and the overall acceleration effect is excellent.

This invention discloses an intelligent adaptive dose verification calculation method and system. The intelligent adaptive dose verification calculation method includes: obtaining preliminary dose distribution data for a patient based on a deep learning-based dose prediction model; dynamically assigning variance reduction parameters to particles in a Monte Carlo simulation based on the high and low dose distribution information in the preliminary dose distribution data, wherein the variance reduction parameters include simulation weights, with simulation weights assigned to particles predicted to be in high-dose regions being less than those assigned to particles predicted to be in low-dose regions; and using a Monte Carlo dose calculation engine, simulating the planned beam based on the dynamically assigned variance reduction parameters to obtain high-precision verification dose distribution data. This invention's intelligent adaptive dose verification calculation method ensures the physical authenticity of the final verification dose distribution and the credibility of the "gold standard."

The application relates to the technical field of computer networks, in particular to a scheduling method and device based on CNN matrix blocking, equipment and a storage medium, the method comprising the following steps: receiving at least one computing task of a CNN neural network submitted by a user, determining the priority of each computing task while responding to each computing task according to a preset job priority ordering mode; generating a task publishing request according to the idle resources of one or more FPGA board cards and the receiving condition of the published tasks, responding to the task publishing request according to a preset data dependency relationship and a preset job priority ordering; based on the published computing tasks to be received, performing matrix blocking through a preset FPGA resource scheduling algorithm, allocating corresponding FPGA resources, and deploying the FPGA resources to one or more FPGA board cards to perform parallel computing on the published computing tasks to be received. Therefore, the problems that the optimal acceleration effect cannot be achieved in the related art and FPGA resources are easily wasted are solved.

The present invention relates to an autonomous preemptive load balancing method for a slave core array based on a Shenwei multi-core processor, and belongs to the technical field of electronic information, including: 1) for large-scale tasks, a divide-and-conquer method is used for decomposition; 2) for the obtained small-scale tasks, a dynamic and static mixed task division is performed; in the static task area, a uniform allocation method is used, and in the dynamic task area, a dynamic allocation method is used; 3) each task execution information of the dynamic task area is mapped to an identification buffer; at the same time, a double buffering strategy is adopted, and two identification buffers are configured to achieve the reset of the buffer used in the next round while executing the current round of tasks; after the slave core successfully obtains the task execution information to be executed, the task read in the dynamic task area is calculated in the LDM private space of the slave core by using DMA. The present invention has a relatively obvious acceleration effect.

The invention belongs to the technical field of quantum computing, and particularly relates to a quantum circuit simulation method based on circuit cutting. By designing a cutting strategy, a complex quantum circuit is divided into a plurality of small sub-circuits, and each sub-circuit can be independently simulated; the method specifically comprises the steps of circuit preprocessing and dependency relationship analysis; sub-circuits are divided through circuit cutting, specifically, a heuristic cutting algorithm is adopted, and an optimal cutting point is determined according to the dependency relationship between the circuits; selecting a scheme capable of minimizing the overall simulation burden by evaluating the overhead of different cutting schemes to optimize the global calculation cost and resource allocation so as to ensure that the cut sub-circuits can be independently and efficiently simulated; and sub-circuit state vector multiplexing: storing the intermediate state vector of each sub-circuit, and searching similar state vectors for multiplexing when a new sub-circuit is calculated, so as to reduce the calculation amount and resource consumption. According to the method, the complexity of large-scale quantum circuit simulation can be reduced, and the calculation efficiency is improved.

The present invention is applicable to the technical field of optical imaging, and provides a photoacoustic remote sensing microscopic imaging method and an imaging system based on parallel processing acceleration. A graphics processing unit (GPU) is used to introduce a unified device architecture based on parallel processing into photoacoustic remote sensing microscopic imaging to improve data processing speed. The data transmission form of data stream is used to break through the limitation of the traditional PARS signal acquisition method subject to the reloading time of the acquisition card. A multi-atomic lock algorithm is used to avoid memory competition caused by multiple threads simultaneously accessing a shared memory location in concurrent threads, thereby greatly improving efficiency.

The present application relates to the technical field of quantum search simulation, and particularly relates to a quantum search simulation method and system for supercomputing, which constructs a quantum search simulation heterogeneous execution environment, the heterogeneous execution environment is composed of a general processor CPU host end and a coprocessor DCU device end, and the general processor CPU interacts with the coprocessor DCU through a communication bus; in the heterogeneous execution environment, the current computing task scale is determined according to the input number of quantum bits and the target item to be searched, the quantum register is created on the general processor CPU according to the search algorithm circuit diagram, the memory space for storing the probability amplitude data is created on the coprocessor DCU, and the equal-weight superposition state is initialized and prepared, the G iteration process of the quantum search algorithm is simulated, and the target item probability amplitude is obtained through G iteration. The present application transplants the quantum gate operation with intensive calculation to the DCU accelerator for execution, realizes the heterogeneous version running of the quantum search algorithm on the supercomputing platform, has strong scalability, and improves the computing performance of the quantum search simulation platform.

The invention relates to the technical field of medicine application, in particular to application of DMXAA in preparation of a medicine for accelerating orthodontic tooth movement. According to the application of DMXAA in preparation of the medicine for accelerating orthodontic tooth movement, orthodontic tooth movement can be remarkably promoted, the defects related to cortex osteotomy such as invasive surgery, postoperative bleeding and other postoperative risks can be avoided, the application is easy to operate, no negative influence is generated, and the practicability of the medicine is enhanced; the DMXAA can induce an inflammatory microenvironment in the orthodontic tooth movement process and is applied to acceleration of orthodontic tooth movement, statistical comparison of data information shows that the movement distance of orthodontic teeth can be increased by locally injecting DMXAA medicine under periosteum, and the DMXAA medicine has an obvious acceleration effect on orthodontic tooth movement.

The application discloses a reinforced concrete structure improvement time domain synthetic aperture imaging fast algorithm, and steps are as follows: a reinforced concrete two-dimensional finite element simulation model is established, and a half diffusion angle is calculated; simulation testing is carried out on an ultrasonic probe in the reinforced concrete two-dimensional finite element simulation model, and full matrix data are formed; the reinforced concrete two-dimensional finite element simulation model is pixelated, and a diffusion angle factor is calculated; whether synthetic aperture processing is carried out is compared according to the half diffusion angle in step one and the diffusion angle factor in step three; and imaging processing is carried out on the full matrix data in step two according to a time domain synthetic aperture full focusing algorithm. The application has the beneficial effects that the application determines the diffusion angle factor of each pixel point to each transmission ultrasonic probe, the area outside the diffusion angle of the ultrasonic probe is not subjected to sound path calculation and amplitude superposition, the algorithm efficiency is greatly improved, meanwhile, imaging of the pseudo defects outside the diffusion angle is avoided. The smaller the diffusion angle is, the more the data quantity is, and the more obvious the algorithm acceleration effect is.

The invention discloses an airflow accelerator for a compressed air dryer, and relates to the technical field of airflow adjusting.The airflow accelerator comprises an outer shell, a motor is assembled outside the outer shell, an air conveying pipe is arranged in the outer shell in a penetrating mode, a driving shaft of the motor is in transmission connection with an impeller located outside the air conveying pipe, and the airflow accelerator further comprises a reducing mechanism; the reducing mechanism comprises an accelerating pipe a slidably connected to the interior of the air conveying pipe, a telescopic assembly and a flow guide mechanism are arranged at the top of the accelerating pipe a, the flow guide mechanism comprises a top plate fixed between the inner wall of the outer shell and the outer wall of the air conveying pipe, and a plurality of accelerating pipes b communicating with the interior of the air conveying pipe are fixedly connected to the air conveying pipe; the length of the telescopic assembly can be adaptively adjusted along with the change of the flow velocity of the airflow, so that the device is suitable for the airflow at different flow velocities, the flow velocity of the airflow and energy loss in the acceleration process are balanced, unnecessary energy loss is avoided, and the acceleration effect of the airflow is guaranteed.

The invention relates to an irradiation plan optimization method based on novel sparse sampling, belongs to the technical field of radiation irradiation dose optimization, and solves the problems that invalid voxel data volume cannot be effectively simplified, the processing data volume is large and the result is not accurate enough in the prior art. Receiving an initial irradiation plan and generating a full dose deposition matrix corresponding to each radiation field; constructing a gradient image based on the full dose deposition matrix; performing dual-condition voxel screening on the constructed gradient image to obtain a final sampling voxel set, and constructing a sparse dose deposition matrix; and performing irradiation plan optimization and dose normalization on the sparse dose deposition matrix to obtain an optimized irradiation plan.