86 results about "Hybrid computation" patented technology

The Field Programmable Instrument Controller (FPIC) is a stand-alone low to high performance, clocked or unclocked multi-processor that operates as a microcontroller with versatile interface and operating options. The FPIC can also be used as a concurrent processor for a microcontroller or other processor. A tightly coupled Multiple Chip Module design incorporates non-volatile memories, a large field programmable gate array (FPGA), field programmable high precision analog to digital converters, field programmable digital to analog signal generators, and multiple ports of external mass data storage and control processors. The FPIC has an inherently open architecture with in-situ reprogrammability and state preservation capability for discontinuous operations. It is designed to operate in multiple roles, including but not limited to, a high speed parallel digital signal processing; co-processor for precision control feedback during analog or hybrid computing; high speed monitoring for condition based maintenance; and distributed real time process control. The FPIC is characterized by low power with small size and weight.

A hybrid system-on-chip provides a plurality of memory and processor die mounted on a semiconductor carrier chip that contains a fully integrated power management system that switches DC power at speeds that match or approach processor core clock speeds, thereby allowing the efficient transfer of data between off-chip physical memory and processor die.

An integrated hybrid system is provided. The hybrid system includes compute components of different types and architectures that are integrated and managed by a single point of control to provide federation and the presentation of the compute components as a single logical computing platform.

Executing a service program for an accelerator application program in a hybrid computing environment that includes a host computer and an accelerator, the host computer and the accelerator adapted to one another for data communications by a system level message passing module; where the service program includes a host portion and an accelerator portion and executing a service program for an accelerator includes receiving, from the host portion, operating information for the accelerator portion; starting the accelerator portion on the accelerator; providing, to the accelerator portion, operating information for the accelerator application program; establishing direct data communications between the host portion and the accelerator portion; and, responsive to an instruction communicated directly from the host portion, executing the accelerator application program.

Placing an application on a private portion and a public portion of a hybrid computing environment for processing. An application may be received for placement and processing. A primary processing objective and a split preference of the application may be determined. The split preference indicates whether the application can be processed using one or both of the private portion and the public portion of the hybrid computing environment. The application may be placed on one or both of the private portion and the public portion of the hybrid computing environment for processing, based on the primary processing objective and based on the split preference.

The present disclosure relates to a processor for implementing artificial neural networks, for example, convolutional neural networks. The processor includes a memory controller group, an on-chip bus and a processor core, wherein the processor core further includes a register map, an instruction module, a data transferring controller, a data writing scheduling unit, a buffer module, a convolution operation unit and a hybrid computation unit. The processor of the present disclosure may be used for implementing various neural networks with increased computation efficiency.

A computer system architecture and memory controller for close-coupling within a hybrid computing system using an adaptive processor interface port (“APIP”) added to, or in conjunction with, the memory and I / O controller chip of the core logic. Memory accesses to and from this port, as well as the main microprocessor bus, are then arbitrated by the memory control circuitry forming a portion of the controller chip. In this fashion, both the microprocessors and the adaptive processors of the hybrid computing system exhibit equal memory bandwidth and latency. In addition, because it is a separate electrical port from the microprocessor bus, the APIP is not required to comply with, and participate in, all FSB protocol. This results in reduced protocol overhead which results higher yielded payload on the interface.

The invention provides a distributed computing system based on hybrid computing resources, used for reasonable allocation of resources, to meet the requirement for diversity of computing tasks. The system comprises a computing engine layer and a resource scheduling layer. The computing engine layer is composed of a plurality of deep learning frameworks constructed on the same Spark computing engine, and access interfaces of all the deep learning frameworks are packaged in a unified mode for the computing engine layer. The resource scheduling layer comprises a plurality of heterogeneous computing resources, and the heterogeneous computing resources comprise at least one of a CPU, a GPU and an FPGA. In the resource scheduling layer, different task queues are divided according to the task types of the to-be-processed tasks, different logic clusters are divided according to the types of computing resources carried by different physical machines, and the tasks in the task queues are distributed to the corresponding logic clusters for execution according to the task types of the to-be-processed tasks.

The invention discloses a power consumption sensing scheduling system and a power consumption sensing scheduling method for parallel application for hybrid computation environments. The power consumption sensing scheduling system comprises a user layer, a scheduling layer and a resource layer. The user layer transmits user requests to the scheduling layer, the scheduling layer transmits execution tasks and data required by the execution tasks to the resource layer and comprises an analysis module, a task clustering module, a processing unit selection analysis module and a task distribution module, analysis results of the analysis module are transmitted to the task clustering module, clustering results of the task clustering module are transmitted to the processing unit selection analysis module, the processing unit selection analysis module comprises a time computation module and a power consumption computation module, selection analysis results of the processing unit selection analysis module are transmitted to the task distribution module, and the resource layer comprises a plurality of DVS (dynamic voltage scaling) processing units and a plurality of non-DVS processing units. The power consumption sensing scheduling system and the power consumption sensing scheduling method have the advantages that DVS and non-DVS hybrid characteristics of the system are taken into consideration on the premise that application execution time minimization is a scheduling task, and execution power consumption of the application is reduced to the greatest extent.

Systems and methods are described for operating a hybrid computing system using cluster contraction for converting large, dense input to reduced input that can be easily mapped into a quantum processor. The reduced input represents the global structure of the problem. Techniques involve partitioning the input variables into clusters and contracting each cluster. The input variables can be partitioned using an Unweighted Pair Group Method with Arithmetic Mean algorithm. The quantum processor returns samples based on the reduced input and the samples are expanded to correspond to the original input.

The invention provides a network flow analysis method based on a GPU, Hadoop / Spark hybrid computing framework. The method mainly comprises the following steps: constructing a GPU computing analysis framework and a Hadoop / Spark computing analysis framework, selecting the GPU or Hadoop / Spark computing analysis framework to process real time or offline network flow, wherein the GPU computing analysis framework is deployed on a stand-alone node installed with the GPU, the Hadoop / Spark computing framework is a distributed processing system and deployed in a server cluster, preferentially adopting the GPU computing analysis framework to process the real-time or offline network flow when the size of the available memory of the GPU is greater than or equal to twice network flow data. Through the construction of the GPU computing analysis framework and the Hadoop / Spark computing analysis framework, the GPU or Hadoop / Spark computing analysis framework is selected for processing the real-time or offline network flow, thereby effectively responding the real-time or offline statistical analysis processing of the high-speed network flow; and an operator, a maintainer and a manager can conveniently backtrack the analysis data.

In accordance with exemplary implementations, application computation operations and communications between operations on a host processing platform may be adapted to conform to the memory capacity of a parallel accelerator. Computation operations may be split and scheduled such that the computation operations fit within the memory capacity of the accelerator. Further, the operations may be automatically adapted without any modification to the code of an application. In addition, data transfers between a host processing platform and the parallel accelerator may be minimized in accordance with exemplary aspects of the present principles to improve processing performance.

The invention discloses an active power distribution network optimization scheduling method for large-scale electric vehicle access, and particularly relates to the technical field of power distribution network optimization scheduling, and the method specifically comprises the following steps: S1, a first layer is two-stage optimization scheduling implemented by DEMS; S2, a second layer is real-time scheduling executed by the regional EV-EMS; and S3, optimized scheduling models in each stage are established and solving methods are adopted. According to the invention, the scheduling strategy isoptimized through three stages; day-ahead scheduling and intra-day correction of controllable resources of the power distribution network and real-time control of EVs charging are cooperatively realized; the influence of uncertainty factors such as RES power generation and EVs charging behavior randomness is reduced; by adopting a 'centralized+distributed 'hybrid computing framework, the computing time is greatly reduced, the problem of dimensionality disasters is solved, in addition, the adverse effect of local communication failure on overall optimal scheduling is also reduced, and the requirement on the reliability of a communication system is reduced.

The invention discloses a system and a method for classifying network messages on basis of hybrid computation hardware. According to the system and the method, centralized dispatching is carried out on various different computation hardware resources by a central processor to construct a multistage classified processing assembly line, matching rules are reasonably divided according to the characteristics of different computation hardware and configured in each stage of the assembly line, classified processing for simple rules can be finished by a special hardware chip, and complex custom classifying rules can be cooperatively realized by the special hardware chip, a universal parallel processor and a universal central processor, thus enhancing the capacity of processing messages.

The invention provides a DNN task unloading method in an edge-cloud hybrid computing environment and a terminal. According to the types and the number of computing nodes, the number of DNN tasks to be unloaded and the number of layers of each DNN task to be unloaded, calculating the number of the DNN tasks to be unloaded; establishing a target function based on total cost minimization; determining a corresponding constraint condition. The influences of conditions such as computing power and time delay constraints of different types of nodes are considered, the feasibility of the obtained optimal solution is guaranteed, when the optimal solution is solved, crossover operation and mutation operation in the genetic algorithm are introduced into the particle swarm algorithm, a specific algorithm is given, and the problem that the particle swarm algorithm is prone to falling into local optimum in the optimal solution solving process is effectively solved.

The method of predicting gas composition in a multistage separator includes solutions to the regression problem of gas composition prediction that are developed using an ensemble of hybrid computational intelligence (CI) models. Three separate homogeneous and one heterogeneous ensemble of hybrid computational intelligence (EHCI) models are developed using a parallel scheme. The homogeneous models have the same types of CI models used as base learners, and the heterogeneous model has of different types of CI models used as base learners. Various popular CI models, including multi-layer perceptron (MLP), support vector regression (SVR) and adaptive neuro-fuzzy inference system (ANFIS), are used as base learners of ensemble models.

The invention relates to a hybrid computation method for acquiring few-group cross section parameters of a fast neutron reactor. The hybrid computation method comprises the following steps: combining a Monte Carlo method with a certain theory method, finely considering the resonance effect of a fast reactor assembly by utilizing the Monte Carlo method, calculating precise multi-group microcosmic total cross section, fission cross section and elastic scattering cross section of each nuclide, solving various orders of elastic scattering cross sections and various orders of neutron flux moments of the fast reactor assembly by utilizing the certain theory method, and aggregating the multi-group cross section into the few-group cross section. The hybrid computation method provided by the invention is strong in universality and wide in application range, a high-precision few-group cross section of the fast reactor assembly can be produced, and precise and reliable cross section parameters are provided for a task of designing a nucleus of a reactor core.

The invention provides an improved high-order nonlinear spatial discretization method for solving an Euler equation, and the method comprises the following steps: 1, reading initial flow field data, and calculating the positive and negative fluxes of each node at the moment of the Euler equation; 2, performing feature projection on the positive flux and the negative flux on each node to obtain a feature flux, and calculating an intermittent detection factor according to the feature flux on each node; 3, constructing a high-order hybrid calculation method of numerical flux on a half point according to the intermittent detection factor, and completing spatial discretization of the Euler equation; 4, discretizing the time items by adopting a three-order Runge-Kutta method; and 5, pushing thetime to a specified tN to finish calculation to obtain flow field data at the tN moment. According to the improved high-order nonlinear spatial discretization method, the WENN-LC format has higher flow structure resolution than a traditional NND format under the same grid; in addition, the hyWENN-LC mixed format not only has higher resolution, but also has higher calculation efficiency.

The invention discloses a RISC-V-based hybrid mixed computing system and method. The system comprises an instruction control module and a mixed calculation module, an expansion instruction is arrangedin the instruction control module, and the expansion instruction contains operation information; the instruction control module is used for automatically setting operation data and operation information in sequence and then transmitting the operation data and the operation information to the hybrid mixed calculation module through an extension instruction; and the hybrid calculation module is used for selecting a corresponding operation mode according to the operation information, performing hybrid operation by combining the operation data and the selected operation mode, and outputting an operation result. Complex hybrid operation is achieved through a single extension instruction, and some complex operations of the processor based on the RISC-V instruction set are simplified. The calculation process of complex hybrid computation is simplified; the disassembling machine code is simpler and clearer; time consumption caused by multiple times of cyclic operation is reduced, the system performance is improved, and the method can be widely applied to the technical field of communication.

The invention relates to a high-precision hybrid calculation method and device and equipment for shock wave instability, and a storage medium. The method comprises the following steps: initializing flow field unit information, and setting a reconstruction mode and a data format at the current moment as a preset mode; reconstructing the flow field unit information according to the flow field information and the reconstruction mode at the current moment; according to a reconstruction result, calculating a flow field interface flux by adopting a data format at the current moment; obtaining a stability matrix according to the flux of the grid interface; analyzing the feature vector of the stability matrix through a matrix stability analysis method, and further obtaining the position where shock wave instability occurs; using a high-robustness numerical format for calculation at an unstable position, and using a low-dissipation numerical format for calculation at other positions. The shock wave is stably captured by applying targeted treatment to the shock wave instability position. According to the method, shock waves can be stably captured, and then accurate distribution of the surface heat flow of the hypersonic aircraft is obtained.

The embodiment of the invention provides a hybrid computing system and a data processing method and device. The hybrid computing system comprises a classification layer and a computing layer, and thecomputing layer comprises a first computing engine based on batch processing and a second computing engine based on stream processing. The classification layer is used for obtaining features of the calculation task according to the first code corresponding to the calculation task and determining a target calculation engine according to the features. The classification layer is further used for converting the first code into a second code corresponding to a target computing engine and sending the second code to the target computing engine, so that the target computing engine executes a computing task to process the to-be-processed data. The hybrid computing system comprises the first computing engine based on batch processing and the second computing engine based on stream processing, so that the hybrid computing system is suitable for executing both batch processing tasks and stream processing tasks, and the applicability of the hybrid computing system is improved.

The invention discloses an unmanned aerial vehicle group task allocation method and system based on cloud and fog hybrid computing and a readable storage medium. The method comprises the following steps that unmanned aerial vehicle groups are grouped according to different usage scenarios and execution task levels; the entry and exit point is used by a specific type of unmanned aerial vehicle group, and a fog computing node is created; the cloud computing center selects a corresponding standby unmanned aerial vehicle set to send a task preparation instruction; after the instruction is received, the unmanned aerial vehicle set sends a response data packet to the fog computing node of the entry and exit point to which the unmanned aerial vehicle group belongs; the fog computing node performscomprehensive analysis, determines an unmanned aerial vehicle group executing the task, and allocates the task to each unmanned aerial vehicle to be flied in a targeted manner; and the cloud computing center and the unmanned aerial vehicle group keep a communication state all the time, and task redistribution is performed according to task execution conditions. According to the invention, by useof the cloud-fog hybrid computing technology, the task allocation intelligence and high efficiency of the unmanned aerial vehicle group are realized, and the time delay and the computing pressure of the cloud are effectively reduced.

A hybrid system-on-chip provides a plurality of memory and processor die mounted on a semiconductor carrier chip that contains a fully integrated power management system that switches DC power at speeds that match or approach processor core clock speeds, thereby allowing the efficient transfer of data between off-chip physical memory and processor die.

The invention discloses a target grouping method based on a self-organizing feature mapping network. The target grouping method comprises the following steps of reading the data obtained by a sensor at the current moment; cleaning the read sensor data; introducing an SOM to group the processed data, calculating the distance between the neurons and the sensor data by using a hybrid calculation method, and checking the accuracy of the group by using a standardized confidence value; evaluating the target grouping condition, and timely correcting according to the actual condition; and outputting atarget clustering result, and repeating the process. According to the method, by carrying out data cleaning before target grouping, noise interference is effectively filtered, and the accuracy of thetarget grouping is improved, the difference between targets can be effectively reflected, and the accuracy of target grouping is improved. By introducing the SOM, the key problem that the grouping number needs to be specified in advance and the threshold value needs to be set is solved, the accuracy and speed of target grouping are improved, and the requirements of practical application are met.By introducing a CV to check the target grouping condition, the robustness of the algorithm is improved.

The invention discloses a neural network and swarm hybrid calculation method for intelligent environment carrier robot floor identification. After various acquired data are clustered according to weather modes, based on different weather modes, fluctuating pressure sensor readings are subjected to FIR filtering processing, the readings are then transmitted to a data analysis module for neural network learning, and the accuracy and the real-time performance of floor identification are greatly improved. The vibration problem of data acquired by a pressure sensor can be greatly improved, and the height data signal analysis precision is greatly improved; universal adaptability is realized, and the method can handle elevator floor identification under each altitude, each geographical location and each weather condition; and the method is not limited to be used by the carrier robot in the elevator, floor estimation in a passageway can also be carried out, and the method can also be applied to fields such as overhead operation and unmanned aerial vehicles.

A hybrid computing system comprising a quantum computer and a digital computer employs a digital computer to use machine learning methods for post-processing samples drawn from the quantum computer. Post-processing samples can include simulating samples drawn from the quantum computer. Machine learning methods such as generative adversarial networks (GANs) and conditional GANs are applied. Samples drawn from the quantum computer can be a target distribution. A generator of a GAN generates samples based on a noise prior distribution and a discriminator of a GAN measures the distance between the target distribution and a generative distribution. A generator parameter and a discriminator parameter are respectively minimized and maximized.

The invention discloses a method for solving a nuclide migration governing equation on basis of a hybrid computation architecture. The method comprises the following steps of reading parameters, wherein computation parameters, boundary conditions and computational grid files are read in; generating matrixes, wherein a rigidity sparse matrix, a time evolution sparse matrix and loading vectors are generated, corresponding sparse matrixes are generated, and a later rigidity sparse matrix and a later time evolution sparse matrix are generated; carrying out circulation computation for solution, wherein a finite difference method is used for carrying out the circulation computation, a Jacobi iteration method is adopted in a circulation process for solving a matrix equation of AX=B, and the obtained solution vector Z is the nuclide concentration field datum of the time point. According to the method, data matrixes in a higher grid body number scale can be stored, and solving speed is higher.