102 results about "Graph sampling" patented technology

The invention provides an interpretable node classification prediction method based on adversarial causal graph learning. The method comprises the steps that a constructed prediction model comprises a redundancy filtering module and an adversarial causal graph learning module; a redundancy filtering module and an adversarial causal graph learning module realize a graph information bottleneck mechanism; the redundancy filtering module adopts a two-layer graph attention network GAT structure to carry out information aggregation, and node embedding is obtained; the confrontation causal graph learning module adopts a learnable sub-graph sampler based on an attention mechanism to generate a causal interpretation sub-graph for node embedding, performs gradient disturbance optimization on interpretation sub-graph embedding based on a PGD confrontation training strategy of a causal enhancement mechanism, generates confrontation embedding, and obtains final disturbance interpretation sub-graph embedding through multiple rounds of disturbance iteration; performing end-to-end prediction model training through multi-target loss joint optimization; and after training is completed, embedding of the nodes is input into a classifier, and a prediction result is output. According to the method, the structural transparency and interpretability of the model are remarkably improved.

The invention discloses an automatic standard term recommendation method, which comprises the following steps of: collecting industry data to be evaluated, constructing a knowledge graph, carrying out discretized semantic alignment on entity vectors and path codes, constructing a matrix-graph joint index, processing user input data by using a gated attention fusion mechanism to obtain an input intention, and recommending the input intention to the industry data to be evaluated. Candidate recommendation terms associated with input intentions are searched in the knowledge graph based on the matrix-graph joint index and a graph sampling algorithm and screened to obtain recommendation standard terms, the recommendation standard terms are clustered, the standard terms in clusters are clustered by adopting a rime algorithm, representative terms of each cluster are obtained and are incrementally updated to the matrix-graph joint index, and the candidate recommendation terms and the candidate recommendation terms are combined. And if the recommended standard terms are not searched, recommending the synonyms. According to the method, the information retrieval and screening efficiency is improved through matrix-graph joint indexing, and the knowledge graph is searched through a graph sampling algorithm, so that the recommendation result fits the intention of the user.

The embodiment of the invention provides a power distribution network dynamic reconstruction method based on knowledge-guided large model enhanced fine tuning and related equipment, and belongs to the technical field of smart power grids and artificial intelligence. The method comprises the following steps: constructing a power distribution network dynamic knowledge graph, and providing structured knowledge guidance for model training; subgraph sampling is carried out based on the timestamp and converted into a fine tuning sample, and a training data set is generated; utilizing the data set to supervise, finely adjust and preheat the large language model; designing a multi-dimensional reward function of fusion format accuracy, economy and security based on mechanism knowledge in the knowledge graph and expert experience; a group relative strategy optimization mechanism is adopted to carry out reinforced fine tuning on the large language model, and the large language model is guided to output a safe, reliable and economical dynamic reconstruction strategy in interaction with the environment. According to the method, the problems of lack of training data, lack of physical knowledge guidance and insufficient decision reliability of a large language model in the power grid field are solved, and the intelligent level and decision quality of dynamic reconstruction of the power distribution network are remarkably improved.

The invention discloses an APT (Advanced Persistent Threat) network attack detection method based on meta-path learning and subgraph sampling, which relates to the technical field of network security, and comprises the following steps of: constructing a network entity relation graph, collecting and standardizing various log data and network topology information, and introducing an external knowledge base to excavate a potential relation; the attribute graph model is used for endowing rich attributes for the nodes and the edges; a meta-path is dynamically generated through an adaptive algorithm, the meta-path is optimized in combination with frequent sub-graph mining and historical attack data, a model in which an attention mechanism and a noise reduction auto-encoder are combined with a graph convolutional network is introduced, and node features are further extracted; defining a node influence propagation index to perform sub-graph sampling, and generating a hierarchical sub-graph; and finally, fusing sub-graph topology, nodes of meta-path learning and attribute features, and adopting multi-classifier ensemble learning to optimize the performance of the attack detection model. According to the method, the accuracy, timeliness and adaptability of APT attack detection are effectively improved, and the method is suitable for a complex dynamic network environment.

The invention discloses a spatio-temporal data driven consumption behavior prediction method and system, and relates to the technical field of text travel data intelligent analysis, and the method comprises the steps: respectively defining a user node, a business district node and an event node according to user attribute data, business district attribute data and a historical external event, performing analysis in combination with user historical consumption data and business district POI data to determine a corresponding user-business district consumption edge, a business district-event influence edge and an event-user trigger edge so as to construct a heterogeneous ternary graph; performing multi-scale Motif sub-graph sampling on the heterogeneous ternary graph to extract a multi-granularity sub-graph, and extracting multi-scale spatial features of corresponding user nodes from the multi-granularity sub-graph; and the two are organically fused into a highly expressed space-time fusion feature, and the business district consumption behaviors of the user in different business districts are predicted according to the space-time fusion feature. Therefore, the prediction precision and robustness of the commercial district consumption behavior of the user are greatly improved.

The invention relates to the technical field of government affair entrepreneurship guidance service intelligent dialogue systems, in particular to a dialogue intention recognition and correction method based on a knowledge graph, and the method comprises the steps: obtaining a voice signal and text information inputted by a user, and carrying out the cross-modal fusion processing to generate a semantic vector; carrying out evolution modeling on the dialogue state by adopting a graph neural network, and outputting a dynamic dialogue state graph containing a time sequence relationship between a user entity and a business object; taking the semantic vector and the focus information as query conditions, executing dynamic subgraph sampling from the government affair knowledge graph, and generating an intention correction strategy vector; and finally, utilizing reinforcement learning to generate personalized guide statements and outputting the guide statements to the user. According to the method, through deep coupling of semantic understanding, state tracking, knowledge reasoning and decision optimization, real-time perception and active completion of dynamic updating of hidden nodes and policies of the government affair process are achieved, and the intention recognition accuracy and systematic adaptive capacity are effectively improved.

The present invention discloses a method for automatically recommending standard terms, including: collecting industry data to be evaluated, constructing a knowledge graph, discretizing semantic alignment between entity vectors and path encodings, constructing a matrix-graph joint index, using a gated attention fusion mechanism to process user input data to obtain input intentions, searching for candidate recommended terms associated with the input intentions in the knowledge graph based on the matrix-graph joint index and a graph sampling algorithm and screening to obtain recommended standard terms, clustering the recommended standard terms, using a rime algorithm to aggregate the standard terms within the cluster, obtaining representative terms for each cluster and incrementally updating them to the matrix-graph joint index, and if no recommended standard terms are searched, recommending synonyms. This method improves the information retrieval and screening efficiency through the matrix-graph joint index, and searches the knowledge graph through the graph sampling algorithm, making the recommended results conform to the user's intentions.

Embodiments of this specification provide a shard storage method and apparatus for a graph and a subgraph sampling method and apparatus for a graph. In a distributed storage process of a graph, local identifiers of a vertex and an edge are implicitly stored, and data is stored in an ordered manner, so that the local identifiers of the vertex and the edge can be implicitly calculated. A connecting edge is stored in a CSR format, to ensure that a first-order neighbor of a node is contiguously stored in a memory. In this way, there can be a higher data loading speed and lower memory occupation.

The invention discloses an anti-parasitic drug target affinity prediction method based on multi-modal feature fusion. The method comprises the following steps: 1, acquiring an anti-parasitic drug related data set; 2, constructing a multi-scale convolutional neural network, and obtaining sequence features of drugs and proteins; 3, constructing a graph convolutional neural network, and obtaining structural features of the medicine; 4, constructing a graph sampling and aggregation network, and obtaining structural features of the protein; 5, constructing a self-adaptive gating network, fusing two modal characteristics of the drug and the protein, and generating drug-protein joint expression; 6, carrying out feature alignment on the two modal features of the drug and the protein by utilizing comparative learning; and 6, updating model parameters according to an output result of the prediction layer. According to the method, multiple neural networks are adopted to learn sequence features and structural features of drugs and proteins, and information of two modal features is balanced and integrated by utilizing comparative learning, so that drug-protein interaction expression is enhanced, and the accuracy of predicting the affinity of anti-parasitic drugs and proteins is improved.

The invention discloses a model multiplexing-based image-text retrieval graph network method, which comprises the following steps of: firstly, constructing a multi-field image-text pair data set from an internet public data source according to a user demand, performing multi-stage data cleaning by utilizing a pre-training model to ensure the data quality, and secondly, multiplexing a large-scale pre-trained multi-modal model as a feature extractor to obtain a multi-domain image-text pair data set; the deep semantic representation of the image text pair is efficiently obtained, and the model training cost is remarkably reduced. Then, a heterogeneous topological structure is designed, image text nodes respectively form same-proton graphs, and cross-modal edges are dynamically generated through learnable attention weights; and finally, by utilizing the constructed image text semantic relation graph, completing context information supplementation of a retrieval target through graph sampling and aggregation, further generating retrieval features, and completing a retrieval task of multi-modal combination. According to the method, the pre-training model features are multiplexed, so that the process is more efficient, and the model convergence speed is greatly improved.

The invention belongs to the technical field of computers and electronics, and discloses a graph neural network hardware Trojan detection method and system fused with an attention mechanism, and the method comprises the steps: S1, mapping a netlist circuit code carrying a hardware Trojan into a directed graph; s2, performing graph embedding, graph segmentation and graph sampling on the directed graph; s3, extracting inherent characteristics of the nodes through an encoder; s4, extracting node local features through a graph neural network; s5, extracting global features of the nodes through an attention mechanism model; and S6, carrying out feature fusion on the data inherent features extracted by the auto-encoder, the local structure features extracted by the bidirectional GCN convergence model and the global structure features extracted by the self-attention mechanism, and obtaining an HT detection result through a classification model. According to the method, the Trojan horse detection accuracy and efficiency are greatly improved; compared with an existing method, the detection range or type is expanded, and the method can be applied to hardware Trojan horse detection of a large-scale integrated circuit actually.

The invention provides a workpiece surface target merging and partitioning method, system and device, and relates to the technical field of image processing, and the workpiece surface target merging and partitioning method comprises the steps: obtaining a surface image of a target region of a target maintenance workpiece; performing graph sampling and visual processing on the surface image to generate a binary image of the target area, and analyzing the binary image to obtain a plurality of connected domains of the target area; performing convex hull generation according to the connected domains to obtain a convex hull corresponding to each connected domain; generating a rotation enclosing rectangle set according to the leading edge vector of each convex hull; and through a variant ant colony optimization algorithm, carrying out global optimization on the rotating enclosing rectangle set, and generating a combined partition of the target area. According to the invention, through accurate capturing of the original image and subsequent layer-by-layer optimization, the redundancy of the laser maintenance path is reduced, and the operation cost is reduced; through construction and global optimization of a high-adaptation form carrier, precise coverage of a complex form target is realized, and the maintenance precision is improved.

The invention particularly relates to a CAD part feature recognition method based on a graph neural network, and the method comprises the steps: extracting geometric entity information and topological relation information based on C # and NX Open API, and generating standardized AAGJSON format data; converting the geometric data into a graph structure which comprises a topological adjacent matrix and a multi-dimensional node feature vector comprising geometric, topological and shape features; a graph neural network model comprising a graph convolution layer, a graph attention layer and a graph sampling aggregation layer is adopted for training, and feature categories and confidence degrees of holes, grooves, bosses and the like are output; and the prediction result is corrected based on geometric consistency, topological connectivity and context rationality constraint. The method has the characteristics of high recognition precision, strong adaptability and good robustness of complex parts, can be integrated in UG NX to realize real-time recognition and visualization, provides important technical support for design automation and manufacturing industry digitization, and solves the problems of low recognition precision, poor adaptability and insufficient topological information utilization in the existing CAD part feature recognition method.

The invention relates to the technical field of brain anomaly detection and artificial intelligence auxiliary diagnosis, in particular to a graph convolutional network brain disease diagnosis method based on sub-graph sampling and multi-feature fusion, and aims to improve the accuracy of brain disease diagnosis. The method comprises the following steps: obtaining resting state functional magnetic resonance imaging data, preprocessing the data, and constructing a brain function connection diagram; and the brain function connection graph represents a brain interval collaborative activation relationship in a graph structure. Subgraph sampling is carried out based on function module division and node degree sorting, and an initial subgraph set is generated; and performing optimization selection on the initial sub-graph set by utilizing reinforcement learning to obtain an optimal sub-graph, introducing a node attention mechanism into the optimal sub-graph, screening key nodes based on attention scores, and generating a discriminant sub-graph. And extracting and fusing position features, neighborhood features and structural features of the discriminant subgraphs, and performing brain disease diagnosis based on the fused features.

The invention discloses a federated learning-based functional magnetic resonance imaging data classification method. The method comprises the following steps of obtaining multi-site resting state functional magnetic resonance imaging data and performing preprocessing; dividing a brain region and extracting a time sequence; constructing a brain function connection network by adopting a Pearson correlation coefficient; a graph sampling aggregation neural network fusing a residual connection structure and a multi-head attention mechanism is trained at each site, and shallow brain region features are reserved; adopting a linear kernel maximum mean value difference loss function to align the brain region node feature distribution of each site, and minimizing the data distribution difference between the sites; and carrying out classification training by adopting cross validation, aggregating parameters of each station through federal weighting, and evaluating classification performance indexes to obtain a final classification prediction result. According to the method, the graph sampling aggregation neural network and the cross-network layer feature alignment method are combined, the heterogeneity problem of multi-site functional magnetic resonance imaging data can be solved, and therefore generalization and classification performance of a global model are improved.

The invention discloses an ultra-short-term photovoltaic power prediction method and system based on an image derived graph structure, and belongs to the technical field of new energy development and utilization. The method comprises the following steps: taking a foundation cloud picture as input, performing image element segmentation by using a superpixel segmentation method, and extracting block high-dimensional features by using a ResNet neural network; calculating nodes and edges of a graph structure by taking a historical data time sequence feature and an image partitioning feature as input, realizing embedded construction of the graph structure, and aggregating coupling features of data and an image by using a graph sampling aggregation network; based on the photovoltaic power value, taking the time sequence feature, the image local feature and the global coupling feature as input, and performing multi-modal feature fusion by using a multi-head attention mechanism to obtain a fusion feature vector; and a multi-step prediction result is obtained through an encoder-decoder. The photovoltaic power ultra-short-term prediction precision can be improved, support is provided for safe and stable operation of a power grid, and the method has certain engineering practical value.

The invention provides an information cascading prediction method and system based on a subgraph, and belongs to the technical field of social network analysis and neural networks, and the method comprises the steps: S1, constructing a deep learning information cascading prediction model CasSubTS, enabling collected user published information to pass through an input layer, and constructing an information cascading graph G; s2, inputting the G into a sub-graph sampling layer, dividing the G into a plurality of information cascade sub-graphs according to different time steps, converting the information cascade sub-graphs into adjacency matrixes, and performing node feature aggregation on the adjacency matrixes to obtain a feature representation matrix B; s3, inputting the B into a feature learning layer to obtain a feature vector # imgabs0 # with a structural feature and a time sequence feature; s4, inputting the # imgabs1 # into a feature weighting layer, and performing weighted fusion on the # imgabs2 # by using a channel attention mechanism to obtain a weighted feature vector # imgabs3 #; and S5, inputting # imgabs4 # into a prediction layer to predict a final macroscopic cascade increment. According to the method, the information cascading in the social network is effectively predicted.

The invention belongs to the technical field of financial industry transaction fraud detection, and relates to a residual aggregation enhanced neighborhood graph sampling credit card fraud detection method, which comprises the following steps: acquiring credit card transaction data and inputting the credit card transaction data into a credit card fraud detection model to obtain a fraud detection result; the training process of the credit card fraud detection model comprises the steps of obtaining credit card transaction data and constructing a multi-relation heterogeneous graph; inputting the multi-relation heterogeneous graph into a plurality of graph aggregation layers which are sequentially connected in series to obtain a multi-relation heterogeneous graph output by each graph aggregation layer; the node features of the multi-relation heterogeneous graph output by each layer are input into a shallow layer residual error aggregation module, and final node features are obtained; inputting the final node feature representation into a classification module to obtain a fraud detection result; updating model parameters according to a fraud detection result until a trained fraud detection model is obtained; according to the method, the node features are updated under each relation type in the graph aggregation layer, and the neighbor information of the node features in various relation types is fully utilized.

The invention provides a global optimization method and device for a refinery enterprise supply chain, and the method comprises the steps: obtaining a supplier set, a sales company set, a refinery plant set, a transportation set, a product set and a spatial position set of a refinery enterprise, and constructing a supply chain model, represented by a network graph, of the refinery enterprise; simplifying the supply chain model represented by the network graph by using a graph sampling algorithm to obtain a sampled supply chain model, constructing an effective lower bound function of a supply chain optimization problem, and solving a first optimal solution according to the effective lower bound function and the sampled supply chain model; reducing the supply chain model represented by the network graph by using a graph reduction algorithm to obtain a reduced supply chain model, constructing an effective upper bound function of a supply chain optimization problem, and solving a second optimal solution according to the effective upper bound function and the reduced supply chain model; and generating a global optimal solution of the supply chain model according to the first optimal solution and the second optimal solution.

The application relates to a kind of smart grid dynamic graph data generation methods based on deep adversarial training, comprising the following steps: obtaining smart grid time series graph data, inputting the smart grid dynamic graph data generation model based on deep adversarial training, and generating smart grid dynamic data graph;The smart grid dynamic graph data generation model based on deep adversarial training includes sampling module, generative adversarial network and reconstruction module, wherein the sampling module is used to generate the center graph using the center graph sampling method;The generative adversarial network includes generator and discriminator, the generator includes decoder and encoder, the encoder is used to encode the center graph through the time series graph self-attention network constructed based on the multi-head self-attention mechanism, and obtain the hidden variable of the center graph;The reconstruction module is used to generate the classification distribution of each time series edge according to the center graph score matrix, and then generate the smart grid dynamic data graph.Compared with the prior art, the application can efficiently and reliably generate smart grid dynamic data graph.

The invention relates to an unsupervised graph anomaly detection model for government affairs, finance and public data governance. The invention discloses an unsupervised graph anomaly detection model for government affairs, finance and public data governance. The unsupervised graph anomaly detection model comprises a graph sampling module, an attribute graph anomaly detection module and a graph anomaly detection module, the neighborhood subtraction identification module is used for representing positive and negative examples and comprises a GNN encoder sub-module, a neighborhood subtraction sub-module and a low-frequency reconstruction sub-module; and the abnormal value scoring module is used for calculating a final abnormal value score. According to the unsupervised graph anomaly detection model for government affairs, finance and public data governance, high-frequency components are amplified without changing data distribution through neighborhood subtraction operation, and the identifiability of abnormal nodes is further improved; low-frequency information of the graph is effectively reserved through low-frequency reconstruction. And finally, determining an abnormal state of the node by an abnormal scoring mechanism.

The invention provides a large-scale supply chain solving method and system based on graph sampling and graph coarsening, and the method comprises the steps: building a target supply chain model with the maximization of total welfare as a target according to the basic element parameters of an electric truck in a supply chain network; constructing a first supply chain model based on an active transportation path in the target supply chain model and a graph sampling technology, and solving the first supply chain model to obtain an optimal lower bound and an optimal feasible candidate solution; constructing a second supply chain model based on active supply chain nodes in the target supply chain model and a graph coarsening technology, and solving the second supply chain model to obtain an optimal upper bound; and according to the optimal lower bound and the optimal upper bound, solving to obtain an optimal gap, and based on the optimal gap and the optimal feasible candidate solution, determining a target large-scale supply chain solving result. According to the invention, the solving efficiency and accuracy are improved.

The present invention relates to a method for generating static graph data of electric power data with structural feature preservation, comprising the following steps: obtaining electric power data, inputting a static graph data generation model into the model, and generating corresponding static graph data; the static graph data generation model comprises a sampling module, a generative adversarial network, and a reconstruction module connected in sequence; wherein the sampling module is used to sample the electric power data using a central graph sampling method to generate a central graph; the generative adversarial network comprises a generator and a discriminator, the generator comprises a decoder and an encoder, the encoder is used to encode the central graph using a graph self-attention network constructed based on a multi-head self-attention mechanism; the decoder is used to decode the output of the encoder to obtain a central graph score matrix; and the reconstruction module is used to generate static graph data based on the central graph score matrix. Compared with the prior art, the present invention can preserve the structural features of the original electric power data as much as possible while improving the efficiency of generating static graph data.

A heterogeneous optimization method and device based on subgraph link prediction, the method comprising the steps of: based on the behavior or relationship between users, commodities and shop entities, abstractly constructing a no-weight and no-direction graph required by a model with node labels; implementing closed subgraph sampling extraction of target node pairs on the no-weight and no-direction graph; simply controlling the upper limit of the size of the subgraph and reasonably filtering the samples; calculating the node heterogeneity index of each closed subgraph; generating a new heterogeneous feature vector according to the calculated index data; replacing the obtained heterogeneous feature vector with the original node feature to become a new node feature; extracting subgraph features and performing model training, and finally realizing the prediction of whether an edge exists between the target node pairs; obtaining a recommended candidate list of each user according to the obtained prediction result of the existing edge; and according to the prediction score, completing the sorting of the recommended candidate list obtained by each user to obtain the final recommended result displayed to the user.

The invention discloses a power system load flow analysis training data enhancement method and related equipment, and the method comprises the following steps: taking all buses in a power grid as nodes in a topological graph, taking all connection equipment as middle edges of the topological graph, and obtaining a structure graph of the power grid; the sample enhancement method is recorded as a sample enhancement method based on subgraph sampling and / or a sample enhancement method based on subgraph grafting based on the structure graph, and samples are generated to enhance the training data. According to the power system load flow analysis training data enhancement method based on subgraph sampling and subgraph grafting, node scale change is carried out on the actual operation scene of a power grid, the number of training samples is increased, the number of nodes of the samples is increased, and the richness of a topological wiring mode is improved. The invention provides a sample expansion method based on sub-graph sampling and sub-graph grafting, and the adaptability of the model to a node number reduction scene caused by starting and stopping of a generator and a node number increase scene caused by power grid expansion can be improved respectively.

The invention discloses a paper classification method based on graph matching and self-supervised graph learning, and relates to the technical field of document classification based on deep learning. According to the method, literature data is represented by adopting a literature relation graph, a graph learning model ConGM based on the literature relation graph is constructed, and reference and theme association between literatures are mined through sub-graph sampling and data enhancement, linear node matching, secondary edge alignment and double-layer negative sample selection, so that precise classification of fields to which papers belong is realized.

The invention provides a method and a system for constructing a knowledge representation learning model based on a periodic perception contrast graph attention network. The method comprises the steps of constructing a sub-graph sampler, constructing a graph attention network and performing time sequence perception contrast learning. According to the method, firstly, adjacent neighbor entities are screened and queried through a period time sensing sub-graph sampler in a period dynamic weighting mode, consumption of computing resources is reduced, and key time sequence context and period information are reserved; then, the graph attention network with the time perception function generates dynamic representation of time perception by using graph attention in the constructed query related sub-graph, and entity embedding representation fused with neighbor information is obtained; then, time sequence perception contrast learning improves robustness in time knowledge graph reasoning through dynamic and static representation contrast learning of an entity, and entity prediction is carried out through embedded representation obtained through learning. Finally, the entity representation is decoded using a decoder.

This application discloses a method and system for predicting consumer behavior driven by spatiotemporal data, which relates to the field of digital and intelligent analysis technology for cultural tourism. The method includes: defining user nodes, business district nodes, and event nodes based on user attribute data, business district attribute data, and historical external events, respectively; analyzing historical user consumption data and business district POI data to determine corresponding user-business district consumption edges, business district-event influence edges, and event-user trigger edges, thereby constructing a heterogeneous ternary graph; performing multi-scale Motif subgraph sampling on the heterogeneous ternary graph to extract multi-granularity subgraphs, and extracting multi-scale spatial features of corresponding user nodes from the multi-granularity subgraphs; and then organically fusing the two into a highly expressive spatiotemporal fusion feature. Based on this spatiotemporal fusion feature, the user's business district consumption behavior in different business districts is predicted. This significantly improves the accuracy and robustness of predicting the user's business district consumption behavior.

The embodiment of the invention provides a database query statement generation method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out graph sampling on a database model graph, and selecting a target data column from a sampling result to construct a target node; detecting the validity of the connection edge of the target node to adjust the data column in the target node; constructing a database model sub-graph based on the target node; and based on the database model sub-graph, generating test data through a large language model. According to the embodiment of the invention, the method comprises the steps: carrying out the sampling of a database model graph, obtaining candidate nodes, selecting a target data column from the candidate nodes, and constructing a target node to reduce the subsequent database structure information needing to be processed; meanwhile, the validity of the connection edge of the target node is detected, and the target node is adjusted based on the validity, so that the reliability of the database model subgraph is improved, and the semantic integrity and rationality of the subsequently generated test data are improved; and finally, based on the database model sub-graph, generating test data on a large scale.

The invention relates to a random subgraph sampling method and system for machine or deep learning training and reasoning. The method comprises the following steps: step 1, estimating a budget; 2, scale solving is carried out; 3, at least one sampling rule is selected from a preset sampling rule base to generate sub-image batches meeting constraints; step 4, obtaining sub-graph labels; 5, implementing a boundary retention mechanism and / or virtual neighbor convergence on the cut edges of the cross-sub-graph to compensate cross-sub-graph information; 6, in the training stage, importance weighting and / or reweighting loss are / is carried out on estimation deviation introduced by non-uniform sampling; step 7, training and / or reasoning the machine / deep learning model based on the subgraph batch, and performing budget adaptive adjustment on the subgraph scale and the sampling rule according to actual occupation and time delay during operation; and executing a corresponding task through the machine / deep learning model. According to the method, data which cannot be trained / inferred originally runs on limited equipment in a sub-graph batch form, and the problem of memory overflow is avoided.