11 results about "Marked graph" patented technology

A computer-implemented system analyzes program and data flows in a software system comprising code written in multiple programming languages using a generative large language model (LLM) directed by programming-language-specific prompts. The LLM identifies functional components within the code, generating labeled graph nodes that include a node type, a node name, and dependency information. A graph construction computer system processes the labeled graph nodes to generate a directed graph, where nodes represent functional components and directed edges represent dependencies. The system stores the graph in a database and provides a web-based interface for visualization, allowing users to explore, query, and analyze program and data flows across the software system. The system enables automated, language-agnostic dependency mapping, facilitating software analysis, debugging, and modernization.

The invention relates to the technical field of supply chain decision, and discloses a supply chain intelligent decision optimization method and system based on a knowledge graph, and the method comprises the steps: collecting the multi-source heterogeneous data of a supply chain, and constructing the knowledge graph of the supply chain through the multi-source heterogeneous data; identifying order fluctuation characteristics in the logistics dynamic data, performing emotion quantification processing on the market feedback information to obtain a supply chain emotion index, and performing knowledge enhancement on the knowledge graph to obtain an enhanced knowledge graph; carrying out supply toughness marking on the enhanced knowledge graph to obtain a toughness marking graph, analyzing a chain scission risk conduction path of the toughness marking graph, and carrying out link risk grade marking on the toughness marking graph to obtain a link marking graph; and analyzing the elastic coefficient of the supply chain, carrying out compression resistance level quantification on the toughness marker map to obtain a target knowledge map, and carrying out intelligent decision making on the supply chain to obtain an intelligent decision report. According to the invention, the risk pre-judgment capability of the supply chain can be improved.

The invention discloses a copper alloy surface wiredrawing quality detection method and system, and the method comprises the steps: carrying out the multi-source defect feature collaborative analysis of hyperspectral reflectivity data, linear array image data and surface three-dimensional point cloud data, and obtaining an oxidation defect marking graph, a scratch distribution graph and a texture anomaly graph; inputting the oxidation defect mark graph, the scratch distribution graph and the texture anomaly graph into a pre-trained convolutional neural network, wherein the convolutional neural network outputs and obtains defect space coordinates; and calculating the unit area defect density, the maximum defect area and the deepest scratch depth according to the defect space coordinates, and determining the surface wire drawing quality of the copper alloy according to the unit area defect density, the maximum defect area and the deepest scratch depth. The industrial problems of multi-modal data splitting, missed judgment of tiny defects and misalignment of grading standards can be well solved.

The invention provides a mark pattern, a mask pattern and a measurement method, and the mark pattern capable of simultaneously carrying out critical dimension measurement and overlay error measurement is formed by designing a plurality of alternately arranged L-shaped front-layer patterns and current patterns. The problem that the measurement accuracy and repeatability are affected by low contrast images and unstable IBO parameters due to graph damage caused by the technology and the light source problem in the prior art is solved, the front-layer graph and the current graph are integrally designed together, the graph marking structure is simplified, the layout occupied area is reduced, and the measurement accuracy and repeatability are improved. According to the method, the cutting channel space is saved, the alignment errors in the X direction and the Y direction in the mask plate plane can be measured, the CD and the alignment error measuring method are simpler, more convenient and faster, and therefore the overall precision and efficiency of integrated circuit manufacturing are improved.

This graph natural language processing device (1) comprises: a node extraction unit (12) that extracts, from graph data, a node including a node type and a node attribute; and a processing unit (14) that outputs a text in which a node token obtained by combining the node type and the node attribute extracted by the node extraction unit (12), a node token of a node of interest in the graph data, and a node token of a node adjacent to the node of interest are listed.

A method (502; 514; 530) is provided for managing, under the control of a computing system, a plurality of ledgers (Lk). Each ledger comprises a corresponding persistent sequence of data blocks (DB), each ledger (Lk) being participated by a corresponding set of participant nodes (PNi) of the computing system, the method comprising: —for at least one of said sets of participant nodes (PNi), said at least one participant node (PNi) of said set appends (512), to a ledger (Lk) corresponding to said set of participant nodes (PNi), at least one data block (DB) each one comprising a pointer (p(k′)) pointing to a different ledger (Lk′); —in response to a ledger relationship request by a requesting node (RE) about a selected ledger (Lk) and at least one selected pointer (p(k′)) comprised in the selected ledger (Lk), generating (518; 534) a graph structure (GS), comprising a set of graph nodes each one labelled with an identifier of a corresponding ledger (Lk), by: —adding a root graph node labelled with an identifier of the selected ledger (Lk), and —performing a recursive procedure for an iterative pointer initialized to each of the selected pointers (p(k′)), the recursive procedure comprising: a) adding a child graph node, labelled with an identifier of a ledger pointed by the iterative pointer, depending on a graph node of the graph structure labelled with an identifier of the ledger comprising the iterative pointer, and b) iterating the recursive procedure for the iterative pointer set to each pointer of a set of the pointers comprised in the ledger pointed by the iterative pointer; —providing (520; 536) an indication of the generated graph structure to the requesting node (RE)—providing (520; 536) to the requesting node (RE) at least one of: —pointers (p(k′)) comprised in the ledgers (Lk) identified by identifiers labelling the graph nodes of the graph structure (GS), and a proof that said pointers (p(k′)) have been included in data blocks (DB) of said ledgers.

The invention discloses an intelligent segmentation method for an adhesive tape area. An adhesive tape image is input into a segmentation network model; the down-sampling module performs multi-scale feature extraction on the image and outputs a feature map; the feature matching module compares the extracted features with basic features, and if the extracted features are newly added features, the newly added features are supplemented and stored in a basic feature library; calling a plurality of basic features from the basic feature library, carrying out feature matching in the feature map, marking a matching region, and carrying out similarity score to obtain a marked map; an up-sampling module carries out up-sampling on the marked graph and records the marked graph as a graph I; the feature region adjusting module is used for zooming and translating a matching region in each image I to obtain an adjusted image; the output module combines the adjustment images, and performs weighted addition on the similarity scores at the pixel points to obtain a comprehensive score; outputting a segmentation prediction map based on the comprehensive score; according to the method, region segmentation is carried out on the rubber strip image by using the deep learning model, the accuracy and efficiency of rubber strip region segmentation are improved, and the labor cost is reduced.

The application provides a flexible sheet two-dimensional reconstruction and local segmentation method and system, first, line scanning spectral pixel data is collected, foreground masks are generated line by line, and real foreground line segments containing coordinate, pixel and other information are extracted in real time. Then, the current and historical line segments are taken as nodes to update the time sequence line segment graph, and active object blocks are constructed; when the object blocks are not updated for a long time or move out of the detection area, the coordinate and mapping table are established based on the buffered line segments, and the two-dimensional object mask reconstruction is completed. Then, relying on the mask and the time sequence graph, narrow neck, overlapping and other local abnormal areas are identified, stable candidate main bodies are screened out, and seed regions are generated; the sub-object division of the foreground pixels is completed in combination with the preset constraints, the attribution marking graph is obtained, the sub-object line segment sequence is converted and the index mapping is established, and finally the information such as the outline, position and area of each sub-object is output. The application retains the real background gap in the foreground interval of the scanning line, and solves the problems of area distortion and object mismerger caused by the mixing of background pixels.

Embodiments of the present application provide a UI interaction test method and device based on a large model, a medium and a program product, relating to the technical field of automatic testing. The method comprises: obtaining multi-dimensional component information of a user interface to be tested; the multi-dimensional component information comprises an interface component mark graph and corresponding component description information; constructing a structured prompt word based on the multi-dimensional component information; determining the expected response information corresponding to each component to be tested based on the structured prompt word using a first large model; obtaining the actual response information of each component to be tested after executing a preset component interaction operation; and performing semantic verification based on the expected response information and the actual response information to determine the interaction test result of each component to be tested. According to the embodiments of the present application, the expected response result of each component is predicted by using a large model, and the semantic verification of the expected response and the actual response is performed to determine the interaction test result, which can effectively improve the efficiency and the automatic test coverage of UI testing.

Disclosed is a machine learning model architecture that leverages existing large language models to analyze log files for security vulnerabilities. In some configurations, log files are processed by an encoder machine learning model to generate embeddings. Embeddings generated by the encoder model are used to construct graphs. The graphs are in turn used to train a graph classifier model for identifying security vulnerabilities. The encoder model may be an existing general-purpose large language model. In some configurations, the nodes of the graphs are the embedding vectors generated by the encoder model while edges represent similarities between nodes. Graphs constructed in this way may be pruned to highlight more meaningful node topologies. The graphs may then be labeled based on a security analysis of the corresponding log files. A graph classifier model trained on the labeled graphs may be used to identify security vulnerabilities.

A threat detection system in a distributed computing environment, consisting of: a processing unit for creating behavior graphs, configured to receive a continuous stream of event data associated with a variety of digital entities such as users, processes, devices, and services; wherein the processing unit for constructing behavioral graphs creates dynamic directed labeled graphs (DDLGs), each DDLG comprising nodes representing discrete behavioral events and edges representing temporal or causal relationships between these events; a fingerprinting module that is operationally coupled with the processing unit for the construction of behavior graphs and is configured to extract canonical subgraphs from the aforementioned DDLGs using temporal sliding windows and to apply graph normalization operations to generate a behavioral fingerprint that corresponds to the most recent activity of the digital entity; a cryptographic hash processor configured to compute a hash digest of the said behavioral fingerprint, the hash digest being digitally signed with a hardware-secured private key; a distributed ledger interface unit configured to transmit the aforementioned signed hash digest to a distributed ledger network via a consensus-based smart contract, so that the behavioral fingerprint is immutably anchored and timestamped; an anomaly detection control unit configured to receive a live DDLG associated with a monitored digital entity and compare it to one or more previously anchored behavioral fingerprints using graph isomorphism or graph editing distance metrics to generate a threat anomaly score; and A response controller configured to trigger a warning or containment action when the threat deviation value exceeds a predefined risk threshold.