24 results about "Computation type" patented technology

Techniques for building and maintaining model webs are described. In some examples, a model web is built by selecting models for the model web from one or more available model types based on at least one or more of availability, tensor information, and compute type, instantiating synapses between the selected models to form the model web and updating information regarding availability of the selected models of the model web to indicate being in use.

The invention discloses a Pandas expression generation system based on visual interaction, and belongs to the technical field of data processing. The front-end interaction layer is used for providing a graphical interface; the logic processing layer is used for converting a Pandas expression character string; and the expression storage layer is used for storing the generated expression in a JSON format. According to the method, a user can intuitively configure a data set, a screening condition, a grouping dimension and a calculation type through a graphical interface, and a system automatically converts configuration into a Pandas expression and stores the Pandas expression; according to the scheme, the problem that a traditional data processing tool depends on the programming capability is solved, the efficiency and autonomy of data processing are improved, and the method is suitable for scenes such as data analysis and business reports.

An aspect of the present disclosure facilitates routing diverse incoming requests to optimal computing options while satisfying requisite performance metrics. In one embodiment, in a computing environment having multiple compute types, a historical data containing characteristics in transport payloads of incoming requests, and characteristics in processing corresponding incoming requests by respective compute types is collected. A system trains, based on the historical data, a machine learning (ML) model to select compute types for incoming requests. Upon receiving a new incoming request sought to be processed, the system extracts from a transport payload of the new incoming request, the first set of characteristics to create a new request context. The system then applies the ML model to the new request context to identify a target compute type and forwards the new incoming request to the target compute type.

The invention provides a server energy efficiency optimization system and method based on heterogeneous computing and a storage medium, belongs to the technical field of data processing, and is used for optimizing the energy efficiency of a server of heterogeneous computing through computing with a computing tool as granularity. The method comprises the following steps: an orchestration agent obtains description information from a consumption entity, wherein the description information is used for describing a consumption entity request to obtain a target service; the orchestration agent splits the target service into N sub-services and determines N calculation tools of which the calculation types are matched with the N sub-services one by one from the M calculation tools, and N is an integer larger than 2 and smaller than or equal to M; the orchestration agent calls N calculation tools to execute calculation of the N sub-services, and results of the N sub-services are obtained; and the orchestration agent combines the results of the N sub-services into the result of the target service, and returns the result of the target service to the consumption entity.

The application discloses an AI generated text detection method, system, device and medium, and relates to the technical field of text processing. The application converts a Chinese text sentence to be detected into a tone category sequence composed of tone categories. In this process, the categories of Chinese tones are used as a quantitative index of phonological structure, and text analysis is converted from a high-dimensional and high-cost semantic space to a low-dimensional and high-efficiency phonological feature space. The application is not dependent on training data of a specific model and is adaptable to text detection of multiple models. Then, N-gram analysis is performed on the generated tone category sequence, and the occurrence frequency of each N-gram combination is calculated to form a feature vector. Finally, an AI generated text is judged through a pre-trained classification model. In this process, the calculation type in processing is mainly string processing and frequency statistics, and the inconsistency problem of the model in multiple model invocations is avoided, so that the AI generated text can be accurately detected.

The application provides a splitting implementation method of a RISC-V vector cryptography instruction, which splits a message scheduling instruction into a plurality of micro-instructions for execution on a processor with a vector data path bit width of 128 bits according to a preset splitting rule in an instruction decoding stage, the splitting rule comprising: obtaining a message scheduling instruction in a SHA-512 algorithm, the message scheduling instruction being used to expand an input message block into a required message sequence; splitting the message scheduling instruction into a 0th type micro-instruction for processing a transformation path of sig0 and a 1st type micro-instruction for processing a transformation path of sig1 according to a calculation type; and the method does not introduce additional storage data space, and can reduce the area overhead of the processor for implementing the instruction while maintaining the unit data path operation throughput.

The application provides a transaction statistics method and device. The method comprises the following steps: obtaining a configuration object of a target statistics variable, the configuration object being generated according to a configuration operation of a user on a configuration page, the configuration object corresponding to a preset data structure; obtaining a calculation type and a condition object from the configuration object according to the preset data structure, the condition object comprising a condition attribute, the condition attribute being used for defining a matching rule of the condition object; obtaining at least one piece of transaction data; for each piece of transaction data, matching the transaction data with the condition object, and when the matching with each condition object is successful, updating a statistics result of the target statistics variable according to the transaction data and the calculation type. The application can generate a configuration object of a statistics variable according to user input. When it is necessary to add, modify and delete the statistics variable, the user only needs to re-input information of the statistics variable, and the software code does not need to be modified, thereby reducing the cost of adding, modifying and deleting.

An architecture of a multi-cloud inspector for any computing device type is provided. According to an embodiment, a method for implementing multi-cloud inspection includes accessing an object list, determining which objects to inspect, determining which inspectors to use, creating object copies, providing and running inspectors for each object copy, receiving inspection report summaries, generating an enriched dataset, and adding the enriched dataset to a security graph database.

The invention discloses a processor performance test method, device and equipment and a storage medium, and relates to the technical field of performance test.The processor performance test method comprises the steps that a performance test program is compiled into a machine instruction, the machine instruction is divided into a plurality of instruction blocks, and all the instruction blocks are marked according to target labels; the target label comprises an instruction type and a memory access mode; the instruction type comprises a floating point calculation type and an integer calculation type; obtaining performance data of each kernel when the processor executes the task; the performance data comprises a core utilization rate, a cache hit rate, memory delay and inter-core communication delay; distributing the instruction block to a target kernel of the processor based on the target label and the performance data, so that the target kernel executes the instruction block to obtain corresponding test data; and recording the test data, and determining a performance test result of the processor according to the test data. Therefore, the effectiveness of the performance test can be improved.

Embodiments of the present disclosure may provide a processing unit and a computing system that divide computations according to the types of computations performed during an inference computation using an artificial intelligence model and perform the divided computations by separate processing units, thereby being capable of reducing an overall time required for the inference computation and improving the performance of the inference computation.

The invention relates to a heterogeneous multi-core task scheduling and heat balance control method and device for an X21E processor. The method comprises the following steps: constructing a dynamic weighing multiprocessor cluster interconnection system based on a CXL protocol; according to the functional positioning and real-time demand difference of the dynamic weighing data, data transmission is realized by adopting a dual-channel transmission logic in which metadata and original data are separated; acquiring thermal state information of each processor endpoint in real time, and judging whether task migration is triggered or not; computing resources of all processor end points are virtualized into shared computing power pools, and adaptive resources are distributed from the corresponding computing power pools according to the computing types and computing power requirements of the dynamic weighing tasks; and when a thermal balance scheduling condition is met, the non-interruption migration of the dynamic weighing task is carried out, and thermal balance control and task scheduling of the multiprocessor cluster are completed. According to the method, the problems of high data interaction time delay, uneven cluster heat distribution, poor CXL technical scene adaptability, task migration data loss and rigid computing power configuration of a multiprocessor cluster in a dynamic weighing scene are solved.

The invention relates to the technical field of data processing, in particular to a data label definition and configuration method and system based on a trusted data space, and the method comprises the following steps: generating a unique label identifier, executing a calculation type decision based on a stream batch integrated architecture, executing metadata conversion, and configuring a label based on a multi-modal label definition system. The method has the beneficial effects that the label definition complexity is reduced: an open rule definition mechanism is adopted, so that business personnel can autonomously define the label without a professional technical background, the use threshold of a label system is greatly reduced, and the label configuration efficiency is improved; through a multi-type label combination mechanism and a multi-dimensional association query function, deep mining of data values is realized, so that a single data source can derive dozens of service labels, and the data reuse rate is remarkably improved.

The application relates to a method and device for implementing reduction calculation, a computer device, a readable storage medium and a program product. The method comprises the following steps: arranging first registers used by a plurality of threads in a thread bundle corresponding to a target reduction calculation into a register block, the numbers of the first registers used by each thread being the same; generating j pooling instructions based on the size of the register block and the calculation type of the target reduction calculation, the pooling instructions carrying a pooling type matched with the calculation type and pooling frame size information, the input registers of the first pooling instruction being the first registers, the input registers of the i-th pooling instruction being the output registers of the (i-1)-th pooling instruction, i being an integer greater than 1 and less than or equal to j, and j being an integer greater than 1; and sequentially executing the j pooling instructions, each thread storing a calculation result of the target reduction calculation in the output registers of the j-th pooling instruction. The method can improve the efficiency of reduction calculation.

The invention relates to the technical field of automatic driving, in particular to an automatic driving task scheduling method, system and equipment based on dynamic priority, and the method comprises the steps: dividing an automatic driving task into a sensing task and a reasoning task based on the calculation type, execution frequency and period of the automatic driving task, constructing a sensing task pool and a reasoning task pool based on the execution information; calculating the scheduling priority of the automatic driving task based on a preset dynamic priority calculation formula; and executing the automatic driving task based on the scheduling priority. In this way, for various heterogeneous computing tasks facing the automatic driving task, the resource utilization rate can be effectively improved, and task blocking delay is avoided.

The invention provides a multi-core-switching fast number-theory transformation system and a task execution method. The system comprises a task identification module, a scheduling module, a control module and a multi-core-particle operation module, wherein the multi-core-particle operation module comprises core particles which are respectively arranged for a plurality of privacy calculation types; wherein the task identification module is used for dynamically identifying a privacy calculation type to which a fast number theory transformation operation task belongs; the scheduling module is used for determining the priority of the operation task according to the privacy calculation type to which the operation task belongs, and scheduling the operation task to the control module according to the priority; the control module is used for distributing the operation task to the core particle of the corresponding type according to the privacy calculation type to which the scheduled operation task belongs so as to serve as the operation task of the core particle; and the multi-core-particle operation module is used for controlling all types of core particles to execute respective operation tasks in parallel so as to realize parallel execution of the operation tasks of various privacy calculation types.

The technology disclosed relates to accessing external data in spreadsheet cells and includes spreadsheet cells that can use formulaically defined external data, allowing users to utilize many cells as if they were one cell and to simplify or expand the type of calculations possible within a cell. The disclosed technology further includes a spreadsheet application that can perform looped calculations within the spreadsheet cells without the need to use an embedded programming language. Additionally the disclosed spreadsheet application can use named variables and formulaic cell designators for cell operations in which formulaically defined non-spreadsheet cell data variables and their values can be exposed and multiple values can exist in a cell, and the user can step through and see those different values.

This invention discloses a simplified dataflow instruction set processor, comprising on-chip memory, a planar arithmetic unit, a cube arithmetic unit, a point arithmetic unit, a tensor arithmetic unit, and an instruction controller. The on-chip memory directly reads and stores computational data from off-chip memory, and this computational data is used for computation by the planar arithmetic unit, the cube arithmetic unit, and the point arithmetic unit. The tensor arithmetic unit directly obtains data from off-chip memory and performs swapping, splitting, aggregation, and reverse aggregation operations on the tensor data according to instruction configuration information from the instruction controller. The instruction controller directly obtains instructions from off-chip memory and sends instruction configuration information to the planar arithmetic unit, the cube arithmetic unit, the point arithmetic unit, or the tensor arithmetic unit according to the instructions. This processor features simplified and limited computational types, maintaining the high efficiency of a dataflow computing architecture while possessing general computing characteristics, thus achieving the advantages of high computing power, high clock speed, and low power consumption.

This invention provides an emerging application workflow classification method based on a combined embedding graph Transformer, relating to the field of cloud computing technology. First, the overall structure of the workflow, the characteristics of task nodes, and the dependency edge features between task nodes are clearly defined. Then, a model is constructed that can fuse and learn node features and dependency edge features. By employing different combined embedding methods, task node features and relationship features are combined. After network learning, rich information about workflow task node features and inter-task dependencies is obtained. Finally, this learned information is applied to downstream classification tasks, accurately classifying different task computation types. This invention enhances the model's learning ability, enabling the classification of unknown workflow nodes and significantly improving the classification accuracy for different task computation types.

The invention provides a low-power-consumption pulse neural network calculation circuit. The low-power-consumption pulse neural network calculation circuit comprises a pulse data coding module, an input cache and routing module, a multiply-add calculation array, an output cache and routing module, a neuron array, a win-win calculation module and a calculation iteration control module. When the calculation type is a general calculation mode, closing the pulse data coding module and the neuron array; and when the calculation type is a neuron calculation mode, the pulse data coding module and the neuron array are started, and the winner-eating calculation module controls the multiply-add calculation array and the neuron array to work in a sparse mode according to the state of the output pulse of the neuron array, and closes part of calculation units and part of neurons. The overall power consumption is reduced by optimizing a bottom layer circuit structure, a simplified neuron circuit is used as a calculation core, and the circuit overhead is small; sparse calculation is realized by utilizing a win-win rule, and the dynamic power consumption is low.

The invention discloses a user interface prototype design method and device, and relates to the field of computers. According to the method, a movable physical element is used for representing a user interface element, a conductive contact, an electronic tag and a magnetic material are arranged in the physical element, and the physical element can be arranged in a design area in a magnetic attraction mode, a slot mode and the like; detecting physical element types and positioning data; fusing calculation types and positioning data, and generating 'HTML / CSS / Javascript' and Flutter user interface program source codes by adopting a grid layout algorithm; and sending the generated source code to external equipment through a wired or wireless network and storing the source code. The device includes a housing, a design area, a power indicator, a function button, a movable physical element, a processor, a sensor, and a communication component. The technology overcomes the defects that a paper prototype is not easy to manage and lacks standards, and a computer prototype tool needs professional knowledge and is tedious to operate, supports offline scene use without a computer, and automatically generates a user interface program source code which can be directly compiled and run.

The invention relates to the technical field of intelligent driving, and discloses a method and device for cooperating heterogeneous computing resources and a vehicle, and the method comprises the steps: carrying out the feature analysis of a plurality of to-be-executed tasks in an intelligent driving scene, and obtaining a feature analysis result; according to a feature analysis result and a system load condition, allocating each to-be-executed task to a corresponding target heterogeneous computing resource; decomposing the to-be-executed task into a plurality of sub-tasks, and dividing the target heterogeneous computing resources into a plurality of target core clusters according to computing types of different sub-tasks; optimizing each target core cluster according to the calculation types of the sub-tasks, and distributing the sub-tasks to the optimized target core clusters according to the calculation types of the sub-tasks; and cooperatively executing the corresponding sub-tasks in different target core clusters. According to the invention, the execution efficiency of the data processing task in a complex driving scene can be improved.

The invention discloses an AI generated text detection method, system and device and a medium, and relates to the technical field of text processing. Chinese text sentences to be detected are converted into a tone category sequence composed of tone categories, the categories of Chinese tones are used as quantitative indexes of a phonetic structure in the process, and the accuracy of the phonetic structure is improved. Text analysis is converted from a high-dimensional and high-cost semantic space to a low-dimensional and high-efficiency phoneme feature space, and the method does not depend on training data of a specific model and can adapt to multi-model text detection; n-gram analysis is carried out on the generated tone category sequence, the occurrence frequency of each N-gram combination is calculated to form a feature vector, finally, AI text generation judgment is carried out through a pre-trained classification model, the calculation types during processing in the process are mainly character string processing and frequency statistics, the problem of inconsistency of models when the models are called for multiple times is avoided, and the accuracy of the AI text generation judgment is improved. The method and the device are used for accurately detecting the AI generated text.

This invention relates to data encapsulation, specifically to a tree-structured encapsulation method for first-principles calculation data. The method includes: creating a file and root group: creating an HDF5 file and establishing the necessary top-level groups in the root directory; encapsulating the physical system: under the `system` group, creating subgroups and datasets according to a predetermined structure to store crystal structure and atomic information; recording the calculation process: for each calculation, creating a subgroup named after the calculation type under the `calculation` group to store the calculation data; and linking workflows: under the `workflow` group, explicitly recording the logical relationships and data flow between different calculation steps by creating subgroups and defining attributes. The technical solution provided by this invention effectively overcomes the shortcomings of existing technologies, such as data fragmentation and loss of correlation, lack of self-description, and low data access efficiency.

The application relates to the technical field of computers, and provides a correctness verification method, device and equipment of a large text sequence BWT and a medium. The method comprises the following steps: blocking a character string X and a suffix array thereof, recursively calling block induction sorting to contract the character string X until the characters in the contracted character string X1 are unique; in the backtracking stage of the recursion 0 layer, the order of S* type suffixes of the character string X is calculated according to the suffix array of X1, and the hash value of the descending S* type suffix sequence is calculated; according to the order of the S* type suffixes of the character string X, the L type and S type suffix predecessor character sequence and the hash value thereof of each block are calculated; the L* type suffix order and the hash value of each block of the L type BWT sequence are calculated by using a small top heap; the S* type suffix sequence and the hash value of each block of the S type BWT sequence are calculated by using a large top heap; and the two types of BWT sequences are combined, and the final verification result is output, so that the time and space overheads of BWT correctness verification are reduced.