105 results about "Swarm computing" patented technology

The invention relates to the technical field of artificial intelligence and environmental information processing, discloses a carbon emission AI intelligent accounting method and system, and aims to solve the problems of weak data fusion capability, poor accounting real-time performance, insufficient model generalization and lack of self-evolution capability of the system in the prior art. The method comprises the steps that multi-source heterogeneous operation data are collected in real time, standardized management and spatial-temporal feature extraction are carried out, direct, indirect and implicit carbon emissions are calculated through a fusion graph neural network, a time sequence attention network and an accounting cluster of an integrated tree model, and online iterative optimization of model parameters is driven based on external calibration signals. According to the scheme, carbon emission accounting is converted into a high-precision minute-level response dynamic deduction mode, the transparency and traceability of accounting are remarkably improved, continuous self-adaptive optimization in a large-scale scene is supported, and reliable data support is provided for precise carbon reduction.

Technologies for automated security incident analysis include a computing device that clusters security incidents and runbooks into multiple clusters based on investigation similarity. For each cluster, the computing device determines a summary of all security incidents in the cluster with a large language model, determines criteria for inclusion of a security incident in the cluster, and determines a suggested investigation step with a retrieval augmented generation pipeline. The suggested investigation step includes a natural language description and a programmatic query. Upon receiving approval from a user, the computing device stores the cluster information in a curated query repository. The computing device may receive a security incident for investigation, assign the security incident to a cluster based on the stored criteria, and retrieve a suggested investigation step from the curated query repository. The computing device may provide the suggested investigation step to a user. Other embodiments are described and claimed.

The invention relates to a transformer area group optimization method and device, computer equipment and a readable storage medium. The method comprises the following steps: obtaining voltage, current and frequency of each edge node in a to-be-optimized transformer area group, photovoltaic output data of the transformer area group, energy storage charge and discharge state data and electric vehicle charging power data, and presetting various operation constraints set for the transformer area group; and inputting the voltage, the current, the frequency, the photovoltaic output data, the energy storage charge and discharge state data, the electric vehicle charging power data and the various operation constraints into a pre-constructed multi-target optimization model, obtaining optimization strategies meeting the various operation constraints, and performing optimization guidance on the operation of the transformer area group according to the optimization strategies. The transformer area group level multi-objective optimization task is calculated through the cloud high-performance computing cluster, so that an optimization strategy which is more adaptive to the current transformer area group is obtained, edge setting is controlled according to the optimization strategy to execute optimization output, and the load balance and operation robustness of the transformer area group are ensured.

Presented herein are systems and methods for using machine learning (ML) models to identify species-biased antimicrobial peptides (SB-AMPs) that target select microbial population. A computing system may retrieve a training dataset comprising AMP sequences and non-AMP sequences. The AMP sequences may target the select microbial populations. The computing system may generate, for each of the AMP and non-AMP sequences, a respective plurality of properties. The computing system may provide as input to a ML model, the AMP sequences, the non-AMP sequences, and the respective plurality of properties for each of the AMP sequences and non-AMP sequences. The computing system may determine, based on providing the input to the ML model, a set of rules defining values for the respective plurality of properties to discriminate between the AMP sequences and the non-AMP sequences.

The invention provides a digital quantum simulation method, which comprises the following steps of: performing symmetry analysis and subspace decomposition on a target Hamiltonian to obtain an effective Hamiltonian and a symmetry constraint condition; constructing a target unitary operator according to the effective Hamiltonian, and initializing a quantum circuit population of the target unitary operator; calculating entanglement characteristics of each quantum line in the quantum line population, performing structure variation on the quantum line population based on the entanglement characteristics, and performing line fragment crossing in combination with a symmetry constraint condition to obtain a candidate line set; performing line parameter optimization on the candidate line set based on the fidelity gradient to obtain an optimized line set, and extracting gate distribution structure features of the optimized line set; and non-uniform sampling is carried out on the optimized line set based on gate distribution structure characteristics to obtain a sampling line subset, an optimal quantum line is screened out from the sampling line subset based on the deviation between an experimental observation value and a theoretical expected value, and quantum simulation is carried out based on the optimal quantum line.

The invention provides a large model calculation method and device oriented to a low GPU resource card group calculation system. The method comprises the steps that parameters of a part of a large language model are unloaded into a CPU memory in the GPU resource card group calculation system; obtaining a reasoning request, wherein the reasoning request comprises a batch processing size and a sequence length; in the cue word processing stage, whether the sequence length is larger than two times of the dimension of a hidden layer of the large language model or not is judged, and if yes, the calculation intensity of all attention layers and all multi-layer perceptron of the large language model is calculated; judging whether the calculation intensity of all the attention layers is greater than that of the multi-layer perceptron or not, if yes, unloading all the attention layers in a GPU video memory and unloading all the multi-layer perceptron in a CPU memory, and if not, unloading all the multi-layer perceptron in the GPU video memory and unloading all the attention layers in the CPU memory; and using the unloaded GPU resource card group computing system to run the large language model for reasoning to obtain a reasoning result of the reasoning request.

Embodiments of the present disclosure provide a data processing method and cluster, a computing device, a computer readable storage medium, and a computer program product. The data processing method is applied to an application program interface service unit of the data processing cluster. The method comprises: receiving a data processing request sent by a client for a target storage unit, wherein the data processing request carries data to be processed and a unit path of the target storage unit; analyzing the data to be processed, and when it is determined on the basis of an analysis result that the data attribute of the data to be processed is a target data attribute, encrypting the data to be processed to obtain ciphertext data and an encryption key; and sending the ciphertext data and the encryption key to the target storage unit on the basis of the unit path, thereby minimizing an attack surface of a malicious user to a node level, and reducing the risk of sensitive data leakage.

The embodiment of the invention provides a multi-node parallel acceleration method, device and equipment based on a high-performance computing platform and an RoCE network and a medium, and relates to the technical field of cluster computing, and the method comprises the following steps: constructing a communication stack based on the RoCE network, and realizing an interconnection transmission network between nodes from a sending end to a receiving end; a computing task and a communication transmission task of the high-performance computing platform are executed through a CPU core binding technology of the high-performance computing platform; based on the transparent large page function, processing a memory allocation request initiated by an application program of an operating system and a request of the application program of the operating system for accessing a memory through a virtual address; before a main computing task of the high-performance computing platform starts, pre-processing is executed through multiple threads, open type multi-processing threads are started, the main computing task is executed through the multiple open type multi-processing threads, and a next round of pre-processing task is started in parallel. According to the scheme, parallel acceleration of multiple nodes in cluster calculation is realized.

The invention provides a cluster computing resource isolation method and device, electronic equipment and a storage medium, and relates to the technical field of cluster computing resource management. And scheduling the containerized task to an adaptive physical graphics processor node based on the information, intercepting a resource access request during task operation, allocating an isolated video memory space and a computing context according to an allocated virtual graphics processor resource limit, and dynamically maintaining an isolated state of multiple tasks, so that the task access efficiency is improved. The problems of inter-task interference, limited hardware compatibility, insufficient resource allocation flexibility and low utilization rate caused by lack of a flexible virtualization scheme for decoupling from hardware and imperfect resource scheduling and isolation mechanisms in the prior art can be solved.

This disclosure provides a node anomaly detection method, apparatus, electronic device, and storage medium. The method is applied to a computing cluster comprising multiple computing nodes. The method includes: performing risk prediction on a target node to be predicted among the multiple computing nodes; determining the predicted anomaly probability of the target node for multiple anomaly types, including multiple types such as faults, slow nodes, and silent data corruption (SDC); identifying risk nodes from the target nodes based on the predicted anomaly probabilities; performing anomaly detection on the risk nodes based on the predicted anomaly probability of each risk node to obtain anomaly detection results for the risk nodes; and identifying anomaly nodes from the risk nodes based on the anomaly detection results. Embodiments of this disclosure can improve the overall performance of the computing cluster and the proportion of computing nodes in the cluster that are in an available state.

The invention relates to the technical field of semantic analysis, in particular to a customer group evolution prediction method based on large model semantic analysis, which comprises the following steps: acquiring historical unstructured data of streaming media platform users, and executing semantic analysis at multiple time points through a large model to generate semantic vectors; processing the semantic vector through a loop module to generate a historical semantic sequence; training a semantic evolution model based on the sequence; the method comprises the following steps: in an offline batch processing stage, pre-calculating future semantic vectors of all users by utilizing an evolution model, and constructing a semantic index; when a seed customer group is received online, calculating a customer group centroid vector representing a future trend for the seed customer group; and performing approximate neighbor search on the centroid vector in a predictive semantic index, and outputting a predicted evolved customer group. According to the method, the prediction accuracy is improved through future matching, and low delay and feasibility of an evolution prediction function are ensured through an off-line pre-calculation framework.

The invention provides a service scheduling strategy determination method, which is used for scheduling and processing jobs in a large-scale cluster. The service scheduling strategy determination method comprises the following steps: inputting job information of a plurality of jobs to be scheduled into a first decision model; the first decision model selects at least one target job from the multiple to-be-scheduled jobs according to the job information of the multiple to-be-scheduled jobs; inputting the job information of the target job and the information of the computing resources of the computing cluster into a second decision model; the second decision model outputs a scheduling strategy of the target job according to the job information of the target job and the information of the computing resources, wherein the scheduling strategy comprises the information of the computing resources for executing the target job; and sending the scheduling strategy to a scheduler, and indicating the scheduler to schedule the target job to the corresponding computing resource for execution according to the scheduling strategy. In addition, the invention also provides a corresponding device, a computing device cluster, a computer readable storage medium and a computer program product.

The application relates to the technical field of attack traceability data security, in particular to an information system APT attack process tracing method and system and a medium. The method comprises the following steps: acquiring log information of an information system, converting the log information into a traceability graph, identifying each community in the traceability graph, calculating community structure singularity of each community, clustering the community structure singularity of all communities, acquiring abnormal behavior significance of each community according to deviation and outlying conditions of the community structure singularity of each community in the cluster, identifying each abnormal community, obtaining behavior credit of each node in each abnormal community in combination with abnormal conditions of difference degrees of out-degree and in-degree of each node in each abnormal community, identifying abnormal nodes under APT network attack, and tracing paths under APT network attack by using an STP steiner tree algorithm. The application can improve the accuracy of APT attack process tracing.

The invention discloses a data processing system, method and device, a storage medium and a computer program product, and relates to the technical field of cooperative computing of artificial intelligence and quantum computing, the system comprises a computer device and a quantum computing device cluster, an artificial intelligence model runs in the computer device, and the computer device is connected with the quantum computing device; the computer device is used for determining a data processing task in the operation process of the artificial intelligence model, selecting an optimal quantum computing device, converting the data processing task into a quantum program conforming to a quantum circuit standard format, and sending the quantum program to the optimal quantum computing device; the optimal quantum computing device is used for executing the quantum program to obtain a quantum computing result, converting the quantum computing result into a data format supported by the artificial intelligence model and returning the data format to the computer device. According to the invention, efficient interaction between the quantum computing device and the artificial intelligence model running in the computer device is realized.

The application relates to the technical field of semantic analysis, in particular to a customer group evolution prediction method based on large model semantic analysis, which comprises the following steps: obtaining historical unstructured data of users of a streaming media platform, performing semantic analysis on multiple time points by a large model to generate semantic vectors; processing the semantic vectors by a loop module to generate a historical semantic sequence; training a semantic evolution model based on the sequence; in an offline batch processing stage, calculating future semantic vectors of all users in advance by using the evolution model, and constructing a semantic index; when a seed customer group is received online, calculating a customer group centroid vector representing a future trend for the seed customer group; performing approximate neighbor search on the centroid vector in the predictive semantic index, and outputting a predicted evolved customer group. The application improves prediction accuracy through future-to-future matching, and ensures low delay and feasibility of the evolution prediction function through an offline precalculation architecture.

The invention provides a cluster privacy set intersection method and device, and is suitable for the technical field of data processing, and the method comprises the steps: in an offline state, employing a data binning parameter and a data redundancy sharing parameter to carry out the preprocessing of a set element held by a specified participant, so as to at least obtain a coding result; carrying out casual transmission with other participants so as to respectively obtain specified random numbers held by the specified participant and the other participants; and in an online state, on each computing node of the computing cluster, according to the coding result and the specified random number, interacting with other participants so as to perform intersection calculation on the set elements distributed to the computing nodes to obtain an intersection result. According to the scheme, the availability and performance improvement of the privacy computing cluster can be considered to a certain extent, and the computing efficiency of the privacy computing cluster is exerted and improved.

The embodiment of the invention provides a gas turbine self-adaptive PID control method for a load shedding working condition, and relates to the technical field of gas turbine control, and the method comprises the following steps: determining a gain optimization variable based on a gas turbine model, constructing an optimization objective function, and determining an input parameter; iterating the gain optimization variable through an intelligent optimization algorithm, constructing an initialized litsea rotundifolia of the intelligent optimization algorithm through a chaotic mapping method based on the search space, the litsea rotundifolia scale and the optimization dimension, calculating an optimal solution of the gain optimization variable based on the initialized litsea rotundifolia, and if the number of iterations reaches the maximum number of iterations, ending the iteration and outputting the optimal solution. If the maximum number of iterations is not reached, continuing iteration until the maximum number of iterations is reached; and taking the optimal solution of the gain optimization variable as a parameter of a cascade PID controller, and controlling the gas turbine in the load shedding working condition. According to the scheme, under the load shedding working condition, the gas turbine can rapidly adjust parameters, and stable output is kept.

The present invention provides a system, device, and method for intelligently coordinating a set of worker nodes within a computing cluster. [Solution] A computing device or service monitors performance metrics of a set of worker nodes in a computing cluster, and when it detects a performance metric below a threshold, it performs a first adjustment to the number of nodes in the cluster, acquires training data at least in part on the first adjustment and uses it together with supervised learning techniques to train a machine learning model to predict future performance changes in the cluster, provides the machine learning model with subsequent performance metrics and / or cluster metadata to obtain an output showing the predicted performance changes, and performs an additional adjustment to the number of worker nodes at least in part on this output.

The embodiment of the invention provides a task scheduling method and device, management equipment and a computing equipment cluster, and relates to the technical field of cloud computing, the task method is applied to the computing cluster, the computing cluster comprises different network levels, and each network level comprises one or more network performance domains. Computing resources with the same network performance are distributed in the same network performance domain, and the method comprises the steps that the computing resource demand quantity of each task in a plurality of tasks and computing resource position constraints of one or more tasks in the plurality of tasks are obtained; according to the computing resource demand quantity of each task and the computing resource position constraint of one or more tasks, respectively dividing one or more affinity groups from computing resources of each network performance domain under each network level; scheduling of a plurality of tasks is performed in one or more affinity groups of each network performance domain under each network level. The method and the device can be compatible with task scheduling of computing clusters in different network forms.

Disclosed in the present invention is a construction system for a big data analysis algorithm library, comprising: an analysis process construction module, an operator platform selection module, a code reverse generation module, a user code verification module, a cluster control host, and a cluster computing host. Cross-platform and cross-language algorithm fusion can be realized, and data analysis is realized in a complete and heterogeneous data analysis flow. Also disclosed in the present invention is a construction method for a big data analysis algorithm library.

The invention is suitable for the technical field of impedance calculation, and provides a new energy station impedance calculation method, which comprises the following steps: inputting acquired broadband impedance characteristic data into a preset clustering algorithm model for clustering analysis, and generating a plurality of unit clusters; calculating an equivalent impedance characteristic of the current cluster, and determining an equivalent frequency domain network equivalent FDNE unit corresponding to the current cluster according to the equivalent impedance characteristic; constructing a function taking maximization of a stability margin and minimization of a harmonic distortion rate as optimization targets, and solving an optimal solution set of the function by taking key parameters of each equivalent FDNE unit of the equivalent frequency domain network as optimization variables; according to the actual electrical topology of the new energy station, the optimized FDNE units are connected, and a full-field equivalent simulation model is constructed; calculating the equivalent impedance characteristic of the grid-connected point of the new energy station through a full-field equivalent simulation model; the calculation efficiency is greatly improved on the premise of ensuring the modeling precision.

The invention relates to the technical field of audit data analysis, in particular to an audit data analysis monitoring method and device, and the method comprises the steps: S1, collecting multi-source data; s2, data cleaning and conversion; s3, data association based on semantics; s4, fusion algorithm optimization; s5, a multi-dimensional data analysis model is established; s6, machine learning auxiliary analysis; s7, real-time data acquisition and transmission; s8, carrying out real-time risk early warning; s9, a distributed storage architecture; and S10, performing data authority management. Through the multi-source data acquisition module and the optimized data acquisition mode, audit data can be quickly and comprehensively acquired. Through application of a distributed storage architecture and a cluster computing technology, the data storage and analysis efficiency is greatly improved, the time cost of data processing is reduced, an advanced data cleaning algorithm and a semantic-based data fusion technology are adopted, the data fusion accuracy is improved, and a reliable data basis is provided for subsequent analysis.

A non-transitory, machine-readable storage medium containing instructions that, when executed, cause a system (400) to: Generating a training dataset (122) based on features of example workloads, wherein the training dataset (122) includes labels associated with the features of the example workloads, the labels being based on load indicators generated in a computing environment and relating to load conditions of the computing environment resulting from the execution of the example workloads; Grouping (502) selected workloads (106) into a plurality of workload clusters (118) based on features of the selected workloads (106), wherein a workload cluster (118) of the plurality of workload clusters (118) includes workloads (106) that are similar to each other according to a similarity criterion; Computation (506) of parameters (124) representing the contributions of each workload cluster (118) of the plurality of workload clusters (118) to a load condition in the computing environment, using a model (120) trained on the basis of the training data set (122), wherein the parameters (124) include a first parameter (124) representing a contribution of a first workload cluster (118) to the load condition and a second parameter (124) representing a contribution of a second workload cluster (118) to the load condition, wherein the first and second workload clusters (118) are part of the plurality of workload clusters (118); Selection of a workload cluster (118) from the multitude of workload clusters (118) based on different values ​​of the calculated parameters (124), wherein a value of a parameter (124) calculated for a selected workload cluster (118) indicates that the selected workload cluster (118) has a workload (106) that negatively affects the work performance of a workload (106) in another workload cluster (118); and Perform workload management (508) in the computer environment based on the calculated parameters (124), wherein workload management (508) includes: Determining a relative priority of the identified workload (106) over other workloads (106), and Restricting the resource utilization of the identified workload (106) in the selected workload cluster (118) in response to the determined relative priority.

Embodiments of the present application provide a method for distributed training of a neural network model in a computing device cluster, a computing node, and a computing device cluster. Computing units of the computing node can obtain weight coefficients of all network layers by means of intra-node communication prior to forward computation and backward computation, without the need for inter-node communication, thereby reducing the number of global communications, and improving model training efficiency. The method comprises: sequentially performing global collection on weight coefficients of network layers in computing units of a node; performing forward computation on input data of the network layers on the basis of a global collection result; upon obtaining a prediction result by means of the forward computation, sequentially performing global collection on the weight coefficients of the network layers in the computing units of the node; performing backward computation on input gradient data of the network layers on the basis of the global collection result; and updating the weight coefficients of the computing units on the basis of a gradient obtained by means of the backward computation, and iteratively executing the described steps until training is finished.

The application provides a method and device for supporting distributed training of a large model across clusters and an electronic device. The method comprises: S1, collecting model structure information of a large model to be trained by a lightweight isomorphic model, and performing topology detection on a cross-cluster computing network to obtain cluster topology information; S2, constructing a cost model based on the model structure information and the cluster topology information, and taking memory constraint and training execution time minimization as an optimization objective; S3, constructing a local hybrid parallel strategy for each computing center, aggregating and splicing the local hybrid parallel strategies that meet the splicing conditions across domains to obtain a global hybrid parallel strategy; S4, for each global hybrid parallel strategy, searching for optimal layer distribution of each stage, tensor parallelism of each stage, and pipeline offset based on the cost model and the optimization objective by using a heuristic Lamarck genetic algorithm; and S5, performing distributed training on the large model according to the optimal parallel strategy obtained by searching.

The invention discloses a set communication method, which is applied to a computing device cluster, the computing device cluster comprises a host and an accelerator cluster, the accelerator cluster comprises a plurality of artificial intelligence (AI) accelerators, and the method comprises the following steps: the host obtains topological positions of available AI accelerators in the accelerator cluster and obtains the number of the AI accelerators participating in set communication; when the number of the AI accelerators included in the maximum continuous topology in the topology of the accelerator cluster is smaller than the number of the AI accelerators participating in the set communication, the host determines a plurality of accelerator groups with continuous topologies according to the topology position and the number of the AI accelerators participating in the set communication; each accelerator group in the plurality of accelerator groups executes the set communication operator to obtain a first execution result; the plurality of accelerator groups perform inter-group communication according to the first execution result to obtain a second execution result. In this way, set communication can be adaptively expanded to the topology discontinuous AI accelerator cluster, and the performance of the AI accelerators is fully utilized.

The invention relates to the technical field of attack traceability data security, in particular to an APT attack process traceability method and system for an information system and a medium, and the method comprises the steps: obtaining log information of the information system, converting the log information into a traceability graph, recognizing each community in the traceability graph, calculating the community structure singularity of each community, and determining the information system APT attack process according to the community structure singularity. Clustering the community structure singularity of all communities, obtaining the abnormal behavior significance of each community according to the deviation outlier condition of the community structure singularity of each community in a clustering cluster so as to identify each abnormal community, and combining the abnormal condition of the difference degree between the out-degree and the in-degree of each node in each abnormal community so as to identify each abnormal community. The method comprises the following steps of: acquiring behavior reliability of each node in each abnormal community, identifying abnormal nodes under APT network attacks, and tracing paths under the APT network attacks by using an STP Steiner tree algorithm. According to the invention, the accuracy of APT attack process tracing can be improved.

The embodiment of the invention provides a data processing method and cluster, computing equipment, a computer readable storage medium and a computer program product, the data processing method is applied to an application program interface service unit of the data processing cluster, and the method comprises the steps that a data processing request sent by a client and aiming at a target storage unit is received, the data processing request carries data to be processed and a unit path of the target storage unit; analyzing the to-be-processed data, and encrypting the to-be-processed data to obtain ciphertext data and an encryption key under the condition that the data attribute of the to-be-processed data is determined to be a target data attribute according to an analysis result; sending the ciphertext data and the encryption key to the target storage unit according to the unit path; the attack surface of a malicious user is minimized to the node dimension, and the risk of sensitive data leakage is reduced.

The invention discloses a job scheduling and resource allocation system in a high-performance computing cluster, and relates to the technical field of high-performance computing. The system comprises a job scheduler, a resource manager, a newly-added job DAG (directed acyclic graph) analyzer, a cluster data distribution monitor, a cache preheating priority calculator, a collaborative decision engine, a distributed cache preheating actuator and a performance optimization module, and the job DAG analyzer constructs a job dependence directed acyclic graph and analyzes a critical path; the cache preheating priority calculator adopts a dynamic priority drift algorithm to determine a data preheating priority, the collaborative decision engine coordinates scheduling and preheating resource allocation, the performance optimization module dynamically adjusts system parameters, the system realizes deep collaboration of job scheduling and cache preheating, the job access delay is reduced, and the system performance is improved. The method improves the cluster computing efficiency and the resource utilization rate, and is suitable for processing a high-performance computing cluster of complex operation workflow.

This application provides a cloud platform storage system, method, device, and medium, relating to the field of cloud platform technology. The cloud platform storage system includes a compute node cluster and a storage node cluster. The compute node cluster includes multiple compute nodes, and multiple disks on these nodes constitute a MinIO object storage cluster for storing image resources and / or cloud disk backup resources. The storage node cluster includes multiple storage nodes, and multiple disk arrays on these nodes constitute an NFS storage cluster for storing cloud disk resources. The compute node cluster stores cloud disk resources in the NFS storage cluster through a cloud disk service, stores cloud disk backup resources in the MinIO object storage cluster, and stores image resources in the MinIO object storage cluster through an image management service. This approach meets the storage needs of users in small- to medium-sized cloud platform usage scenarios while reducing the complexity of system deployment and lowering the operation and maintenance costs of using the cloud platform.