59 results about "Relationship learning" patented technology

The invention discloses an industrial Internet of Things time sequence self-supervision anomaly detection method and a monitoring and early warning system, and relates to the field of industrial Internet of Things, and the method comprises the steps: S1, constructing an anomaly detection model, and S2, obtaining a training data set; s3, training and optimizing an anomaly detection model; s4, acquiring to-be-detected data in real time; s5, performing anomaly detection analysis on the to-be-detected data, and outputting an anomaly detection result; through a time sequence and relation learning module, a dynamic graph topological structure learning module and an enhancement module, internal characteristics of a time sequence in a time domain and a space domain are deeply mined. The time sequence and relation learning module comprehensively captures a multi-scale time pattern, and the dynamic graph topological structure learning module eliminates dependence on a predefined graph structure; the enhancement module enhances the invariant representation under noise, and improves the recognition capability of the model to a normal mode; through wide experiments, the advancement of the method in detection performance is verified, and reliable support is provided for intelligent manufacturing and infrastructure diagnosis.

The invention provides a multi-modal semantic-action alignment method and equipment for end-to-end automatic driving, and belongs to the technical field of automatic driving, and the method comprises the steps: carrying out semantic reasoning through a large language model based on various sensor data and language instruction information of a vehicle, and generating semantic information containing a driving intention; inputting the semantic information into a semantic-action alignment module, and converting the semantic information into corresponding driving action representation through a learned consistency mapping relation from a semantic space to an action space; and generating an executable control track of the vehicle according to the driving action representation. According to the method, the problem of insufficient semantic and action space alignment is solved, information distortion and precision limitation caused by post-processing depending on rules are avoided, the high-level driving intention can be generated based on complex multi-mode information (such as navigation instructions and traffic environments), it is ensured that the final execution action is highly consistent with the intention, and the driving intention is more accurate. And the decision-making rationality of the system in a complex scene is enhanced.

The invention discloses a traffic engineering multi-source monitoring data fusion intelligent management and control system, and the system comprises a multi-source data dynamic access module which is used for the access of multi-source heterogeneous data; a data processing module; the dynamic relation learning module is used for mining dynamic space-time association and potential laws in the multi-source data and establishing a time-varying mapping model among the data; the intelligent fusion decision module is used for generating a traffic control decision based on the processed multi-source data and an output result of the time-varying mapping model; an adaptive adjustment module; a data service module; through cooperation of the multi-source data dynamic access module, the data processing module, the dynamic relation learning module, the intelligent fusion decision-making module, the adaptive adjustment module and the data service module, full-process automation from multi-source data access to intelligent management and control decision-making can be realized, and self-adaption to changes of a complex traffic environment can be realized. And the fine management level of traffic engineering is improved.

The invention provides a multi-modal semantic guided three-dimensional target positioning method, a medium and equipment. The method is realized based on a multi-modal semantic guided three-dimensional target positioning model, and comprises a multi-view semantic prior module, a text coding module, a double-branch point cloud coding and multi-source comparison supervision module, a sparse scene graph construction and graph relationship learning module and a positioning decoding module. The multi-view semantic prior module is used for segmenting the 3D scene point cloud into a 3D object point cloud, generating a multi-view 2D visual representation and encoding the semantics of the multi-view 2D visual representation; the double-branch point cloud coding and multi-source comparison supervision module is used for injecting semantic features into a 3D object point cloud to obtain 3D fusion features and realizing multi-source feature alignment through multi-source comparison supervision; the sparse scene graph construction and graph relation learning module is used for constructing a sparse scene graph and optimizing the sparse scene graph through a graph attention network; and the positioning decoding module is used for decoding and outputting a positioning result. The method can improve the target positioning capability of the model in a complex scene.

The invention provides a text video retrieval method of a fine-grained relation learning network based on energy perception. The method comprises the following steps: giving a query text and a video clip; inputting the query text into a text encoder of the CLIP, inputting the video clip into a text encoder image encoder of the CLIP, and extracting to obtain text embedding and frame embedding; inputting the text embedding and the frame embedding into a fine-grained relation learning network for text enhancement operation to obtain enhanced text embedding; taking the enhanced text embedding as a frame fusion condition to carry out frame fusion operation on the frame embedding to obtain video embedding; calculating the similarity of a text-video pair formed by the query text and the video embedding based on a cosine similarity function; and selecting the text-video pair with the highest similarity as a retrieval output result of text video retrieval. According to the method, the problem of randomness of the random text of single sampling is solved, so that semantic information of text coding is better expanded, and the final retrieval effect is improved.

The application discloses a graphite ore grade identification method and device, equipment and medium, and relates to the technical field of graphite ore grade identification. The application introduces Laplacian wavelet convolution instead of a traditional ordinary convolution kernel, realizes parallel extraction of multi-scale features, combines a multi-head attention mechanism to capture scale features of the ore under a global view, constructs a sparse sample relation graph based on an enhanced feature matrix, and uses a Chebyshev graph convolution network to perform multi-order neighborhood information propagation and aggregation on a graph structure to realize grade identification. The process reconstructs the grade identification problem from'single sample independent judgment' to'sample relation learning' under a small sample on the basis of global multi-scale feature extraction, so that the final prediction of each sample node depends not only on its own features, but also on similar feature information of all neighbor samples, thereby realizing high-precision identification of the grade of the graphite ore under a small sample.

The application discloses an uncertainty knowledge graph reasoning method based on semi-supervised confidence distribution learning, comprising the following steps: converting the triple confidence in the uncertainty knowledge graph training data into a confidence distribution; simultaneously learning the embedding of the uncertainty knowledge graph on the labeled data and the pseudo-labeled data generated by a pseudo-labeled data generator by using a relation learner based on confidence distribution learning; generating high-quality pseudo-confidence distribution labels for unlabeled data by using the pseudo-labeled data generator; iteratively training the relation learner based on confidence distribution learning and the pseudo-labeled data generator by using meta-self-training until the two converge; and inputting the data to be completed into the trained relation learner based on confidence distribution learning to perform reasoning, thereby achieving the completion of the uncertainty knowledge graph. The application can capture the supervision information of a small number of confidence or unseen confidence in the labeled data and is suitable for the scene where the triple confidence distribution of the uncertainty knowledge graph is unbalanced.

This invention provides a learning device, an evaluation device, an evaluation system, a learning method, an evaluation method, and a recording medium. The learning device includes: a correspondence receiving unit that receives correspondences between the quality evaluations of each product of an object process and the quality evaluations of each downstream product produced in a downstream process using the products of the object process; a learning processing unit that uses at least one production parameter relating to the production of each product of the object process and the quality evaluations of each downstream product produced using the products of the object process to generate a predictive model that infers the quality evaluation of the downstream products based on the at least one production parameter; a calculation unit that calculates a model evaluation based on at least one of the accuracy or complexity of the predictive model; and a model evaluation sending unit that sends the model evaluation calculated by the calculation unit to the evaluation device, which evaluates at least one upstream process using model evaluations for each of at least one upstream process that is upstream of the downstream process.

The application belongs to the field of computer vision, and discloses a video description method based on semantic disambiguation structured coding. The application proposes to introduce prior knowledge such as a knowledge graph to construct the relationship between objects in a video (a concept semantic graph), so as to obtain a structured coding of a deeper level of understanding of the semantic relationship of the video. On the basis of the concept semantic graph, according to the guidance of the semantic of the video scene, a relationship most conforming to the current video context is dynamically learned from various relationships of the same pair of objects to eliminate the semantic ambiguity problem existing between the objects, so as to achieve semantic disambiguation structured coding. A cross-domain guidance relationship learning strategy is proposed, which analyzes each object and its relationship in a description sentence to fit the learning of the concept semantic graph in the model, so as to better learn each object and the relationship between the objects in the video. The method of the application can realize more accurate and comprehensive video description.

The application discloses a dangerous driving behavior detection method based on a line-of-sight direction time relationship learning. A convolutional neural network is designed to estimate the head orientation and the binocular orientation of a driver. In view of the possible inconsistency between the head orientation and the binocular orientation, a head orientation and binocular orientation joint network is designed to estimate the line-of-sight direction of the driver. In view of the problem that the line-of-sight direction changes with time during driving and it is difficult to accurately determine the dangerous line-of-sight direction state, a Gaussian time weight-based line-of-sight direction time relationship learning is designed, a line-of-sight direction time positioning network is constructed, and reliable time positioning of the dangerous line-of-sight direction is realized. When the duration of the dangerous line-of-sight direction exceeds a threshold value, a safety warning is given to the driver. The application can handle the inconsistency between the head orientation and the binocular orientation, can robustly handle different line-of-sight direction time change processes, and can effectively realize dangerous driving behavior detection.

The invention provides a spatial transcriptome cell composition inference method based on graph contrast learning. According to the method, a cross-modal low-dimensional feature spatial data simulation module is constructed, single cell and spatial transcriptome data are respectively mapped to a potential embedding space, feature space difference is reduced through aligned distribution, and structural consistency of cross-modal data is enhanced; designing a double heterogeneous graph construction and potential relation learning module, constructing a feature heterogeneous graph of two cells and spatial points based on gene expression similarity, and capturing a high-level potential relation of spatial data in double graphs by adopting a meta-path potential relation reasoning strategy; and introducing a graph representation optimization mechanism based on structure contrast learning, maximizing the consistency of the cross-modal nodes in the embedding space, and obtaining a final cell composition inference result. The method realizes inference of spatial transcriptome data cell composition, can be used for cell space positioning and tissue microenvironment analysis, and provides a reliable computational analysis basis for related biomedical research.

The invention relates to the technical field of graph neural networks, in particular to a graph data imbalance classification method for synthesizing enhanced nodes, which introduces attention-based feature extraction and neighbor aggregation methods, captures a high-order relationship among nodes by integrating a multi-head attention mechanism through attention-based feature extraction, and obtains a high-order relationship among the nodes; according to the method, global features with more representativeness are learned, the problem of insufficient feature learning is solved, and expression of features and topological information between nodes is enhanced through aggregation and splicing of node features and neighbor aggregation, so that generated samples are more uniform and have subclass representativeness, and the problem of insufficient node representativeness is solved. According to the attention-based feature extraction method, the high-order relationship between the nodes is captured by integrating a multi-head attention mechanism, and the expression of the features and topological information between the nodes is enhanced by aggregating and splicing the node features and neighbor aggregation, so that the generated sample is more uniform and has subclass representativeness; and more valuable guidance is provided for node generation.

Systems and methods are described for grapheme-phoneme correspondence learning. In an example, a display of a device is caused to output a grapheme graphical user interface (GUI) that includes a grapheme. Audio data representative of a sound made by the human user is received based on the grapheme shown on the display. A grapheme-phoneme model can determine whether the sound made by the human corresponds to a phoneme for the displayed grapheme based on the audio data. The grapheme-phoneme model is trained based on augmented spectrogram data. A speaker is caused to output a sound representative of the phoneme for the grapheme to provide the human with a correct pronunciation of the grapheme in response to the grapheme-phoneme model determining that the sound made by the human does not correspond to the phoneme for the grapheme.

The application discloses a kind of robust causal relationship learning method, system, equipment and storage medium, large-scale variable set can be effectively handled, and when variable set scale is larger, the problem is decomposed into multiple small-scale problems and is handled, improve the efficiency of processing large-scale variable set;Meanwhile, the application has good scalability, can be used with any causal algorithm, so that the application can adapt to a variety of different application scenarios and needs;And, the application provides a new way to solve complex causal inference problem, the idea of divide and conquer and the concept of causal cut provide a new perspective and tool for subsequent causal inference research;In addition, the directed causal graph finally obtained by the application has high accuracy, and can improve the effect of the application field.

The invention discloses an accounting term automatic identification method and system based on a neural network. The method relates to the technical field of natural language processing, and comprises the following steps: by constructing a scene-based accounting term corpus and introducing a scene perception mechanism, enabling a model to accurately recognize term differences in different accounting scenes, and solving the problem of low cross-scene recognition accuracy of a traditional method; information dilution in a long text is effectively avoided through a multi-attention Encoder-Decoder model, and context association of accounting terms in a long sentence is accurately captured; through the combination of the composite loss function and the CRF layer, the term category prediction precision is optimized, and the dependency relationship learning of the term tag sequence is enhanced.

The present application relates to a kind of large model atlas question answering method and device based on potential unit fine-tuning, the method includes the following steps: based on the knowledge graph of symbolization and relationship path and the variable of using high-dimensional prompt word representation potential unit, potential relationship learning is carried out between large language model and knowledge graph, and relationship reasoning path is generated, wherein the input layer of large language model is additionally trainable potential unit;And potential unit is randomly initialized by normal distribution, and is fine-tuned with large language model in hidden learning process;Relevance evaluation is carried out to relationship reasoning path, and the most relevant relationship reasoning path is selected;Input natural language question, and the answer in knowledge graph is obtained by the most relevant relationship reasoning path.The present application filters out irrelevant and misleading context fragments using these fine-tuned potential units, while improving response efficiency and performance.

The application relates to a recommendation method based on a knowledge graph and an attention mechanism and belongs to the field of personalized recommendation. The method comprises the following steps: defining a knowledge graph perception recommendation problem, establishing a user and item interaction matrix and a knowledge graph as input; setting a seed set of users and items in a knowledge graph propagation layer, propagating the seed set along adjacent entities, and capturing knowledge-based high-order interaction information of the users and items; learning a corresponding latent relationship representation for each relationship through an attention mechanism in a knowledge perception attention embedding layer to obtain a latent associated entity set; processing the entity set through an aggregator to obtain a corresponding aggregation vector, and predicting a preference score of a user for an item according to the aggregation vector; and designing a loss function to train the above process, which is used for user preference recommendation after the training is completed. The application can more accurately capture the latent relationship between the user and the item, and provide more accurate and personalized recommendation services for the user.

The invention discloses a dynamic asymmetric data relation learning and trend prediction method based on environmental state self-adaption, which comprises the following steps of: a) reading multivariable time sequence data, decoupling endogenous and exogenous characteristics through a dual-channel architecture, and aggregating to generate a global state context and a potential factor; b) constructing a directed influence network by using an asymmetric attention mechanism, and introducing a directed acyclic graph constraint to obtain a dynamic adjacency matrix and an external effect; c) generating a gating signal based on the global state, and adaptively adjusting an endogenous feature weight; and d) fusing the multi-source features to carry out full connection processing, and outputting a prediction result. According to the method, the dynamic asymmetric causal relationship between the data is obviously modeled, the adaptive perception of the change of the environment system is realized, and the prediction accuracy and robustness of the multivariable time sequence in the complex non-stationary environment are improved.

The present application relates to the field of safety warning, especially to a punching aluminum veneer processing safety warning method and system. The method comprises the following steps: obtaining a punching aluminum veneer processing video, the punching aluminum veneer processing video contains a plurality of frames of punching aluminum veneer processing images; obtaining a pre-constructed safety monitoring network, and training the safety monitoring network using the punching aluminum veneer processing images to realize punching aluminum veneer processing safety warning; in the training process, the two consecutive frames of punching aluminum veneer processing images are input into the safety monitoring network in turn, the feature maps corresponding to the two frames of punching aluminum veneer processing images are obtained in the last convolution layer of the safety monitoring network, and a loss function for controlling job relationship learning is constructed; the training of the safety monitoring network is supervised by using the loss function for controlling job relationship learning. More accurate safety warning is realized.

The invention discloses a causal-driven honey array attack chain reconstruction and strategy linkage method, and aims to realize high-precision and low-delay causal relationship reasoning on security logs collected by honey points, honey courts and the like, provide structural support for honey array strategy scheduling and attack path guidance, and improve the algorithm based on a pre-training language model. Context contrast learning and Prompt example guidance are fused, the dependence of a traditional method on a rule template and annotation data is broken through, and the method has cross-scene migration ability and complex attack semantic modeling ability, including honey security event pair modeling; constructing a causal example guide sample library; prompt construction of causal event pairs and causal relationship learning and prediction are carried out; causal-driven attack chain reconstruction is linked with a honey matrix strategy; honey array attack path reconstruction and deployment strategy self-adaptive linkage is driven through a causal recognition result, a recognition-reconstruction-decision closed loop is formed, and the response efficiency and trapping efficiency of a honey array system in complex attack scenes such as transverse movement and multi-hop penetration are improved.

The invention discloses a biological sample storage shelter with microenvironment stabilization and vibration suppression functions, comprising: a digital twinborn and multi-mode sensing base layer for constructing a digital twinborn body synchronized with a physical biological sample storage shelter in real time; the multi-modal data fusion and state estimation layer is used for converting the multi-modal data into a state vector comprehensively representing a current storage microenvironment; the adaptive core control layer under the time-varying constraint is used for generating an instantaneous control instruction according to the state vector; the RL top decision-making layer capable of long-periodicity optimization is used for carrying out optimization learning according to the state vector; the prospective disturbance suppression layer based on causal inference and mode recognition is used for carrying out deep mining and causal relationship learning on historical data and recognizing and predicting internal and external disturbances; and a built-in security and collaboration mechanism layer for ensuring the credibility of the system and used for ensuring the authenticity of the data and coordinating each subsystem in the biological sample storage shelter. Through the mode, the super-stable state control of the microenvironment is realized.

The invention belongs to the technical field of teaching informatization, and relates to a knowledge graph-based learning path recommendation method, computer equipment and a storage medium. The method comprises the following steps: generating a basic learning path based on a knowledge graph pre-repair relationship, a learning target and knowledge points mastered by a learner, and collecting and filtering learning behavior data to obtain effective evidence; updating a knowledge point mastering state by adopting a double-threshold hysteresis mechanism, detecting and remedying fallback oscillation by combining double windows with the mastering state, and accurately positioning a bottleneck node according to an oscillation contribution degree; and performing minimum residence control on the bottleneck node, generating a minimum remedy packet path only containing a forward gain node, and checking a mastering state to complete path merging. The method can effectively solve the problems of learning path remedy rollback oscillation, unstable grasp state judgment, remedy range redundancy and the like, and can inhibit path oscillation, reduce invalid learning and improve learning path convergence and learning efficiency.

An information processing device is provided that can calculate an interpretation index for each explanatory variable in the process of calculating a risk score, thereby reducing the calculation load. [Solution] An accounting audit support device 1 detects risks that can be detected from accounting-related data using a machine learning model. The system is equipped with a learning model generation unit 32 that uses a plurality of training records that combine a plurality of explanatory variables and a dependent variable selected from the financial data to generate, as a machine learning model, a shape function that indicates the relationship between each explanatory variable and dependent variable; a learning model storage unit 41 that stores the shape function for each explanatory variable generated by the learning model generation unit; and a risk calculation unit 34 that calculates an interpretation index for each explanatory variable using the shape function for a new record that is the target of risk detection, and calculates a risk score from the plurality of interpretation indexes.

The invention discloses a software development application data processing method based on an AI large model, and the method comprises the following steps: 1, constructing a dependency relationship learning module based on a dynamic dependency modeling mechanism, constructing a semantic fusion module based on a semantic association enhancement mechanism, and constructing a version adaptation module based on a version evolution tracking mechanism; the modules jointly form a data processing model. The invention relates to the technical field of data processing. According to the method, a multi-dimensional dependency fusion mechanism is formulated under the framework of a dynamic dependency modeling strategy and a semantic association enhancement strategy, and the mechanism enables semantic associations of different time spans to be mutually fused and all dependencies of multiple versions to be mutually interacted in a one-time processing process aiming at each version, so that the semantic associations of different time spans can be mutually fused; therefore, feature representation with rich semantics is obtained.

The application discloses an intelligent auditing and financial fraud identification method based on a dynamic heterogeneous graph neural network. The method comprises the following steps: constructing a dynamic enterprise risk heterogeneous graph; generating a first feature vector that fuses node static features and node dynamic features, and a second feature vector that fuses edge static features, edge dynamic features and the interaction results of connected node features; obtaining a standard embedding representation of a node; obtaining a preliminary aggregated representation of the node based on an attention mechanism and the standard embedding representation of the node; modeling the global dependence of the preliminary aggregated representation of the node by using a Transformer structure to obtain an enhanced aggregated representation of the node; fusing the standard embedding representation and the enhanced aggregated representation of the node through a residual connection to obtain a final embedding representation of the node; and finally obtaining a financial fraud identification result. The application effectively solves the problems of insufficient fusion of heterogeneous features, weak global relationship learning ability and insufficient time sequence dynamic capture in the prior art.

This invention discloses an image matching method and system based on hypergraph guidance and high-order relation learning. By introducing a hypergraph structure, it flexibly constructs multi-node high-order relations between keypoints. Combined with fuzzy C-means clustering, it achieves soft membership of keypoints to multiple hyperedges, effectively expressing the multi-level association characteristics of keypoints. Based on this hypergraph structure, a hyperedge convolution operation is designed. Through feature propagation between nodes and hyperedges, it captures long-distance dependencies and global contextual information between keypoints, significantly improving the discriminative ability of feature descriptors. This invention proposes a hypergraph-guided additional attention mechanism, combining self-attention and cross-attention mechanisms to simultaneously mine high-order and low-order relations within and between images, enhancing the accuracy and stability of matching.

The application discloses a kind of anti-noise article and multi-modal recommendation method of interactive behavior modeling, belong to personalized recommendation technical field, including the following steps: (1) isomorphic graph learning: respectively constructs collaborative enhancement's article-article isomorphic graph and user-user isomorphic graph, article original modal feature and user modal perception preference are handled, obtain preliminary article embedding and user embedding;(2) modal attention optimization: after isomorphic graph learning, introduce modal inner attention module and cross-modal attention module, preliminary article embedding is handled, obtain refined article representation;(3) isomorphic graph relationship learning;(4) dynamic multi-view contrast learning;(5) model training.The application, by explicitly modeling the collaborative signal between articles, optimizing multi-modal feature denoising and complementation, distinguishing the importance of user-article interaction in fine granularity, realizes more accurate recommendation article modeling, and improves the effect of personalized recommendation.

The invention discloses an anti-noise article and interactive behavior modeling multi-modal recommendation method, which belongs to the technical field of personalized recommendation, and comprises the following steps: (1) homomorphic composition learning: respectively constructing an article-article homomorphic composition and a user-user homomorphic composition which are synergistically enhanced, processing original modal features of an article and modal perception preference representation of a user, and learning the article-article homomorphic composition and the user-user homomorphic composition; preliminary object embedding and user embedding are obtained; (2) modal attention optimization: introducing an intra-modal attention module and a cross-modal attention module after homomorphic composition learning, and processing the preliminary article embedding to obtain refined article representation; (3) heterogeneous graph relation learning; (4) performing dynamic multi-view contrast learning; and (5) model training. According to the method, more accurate recommended article modeling is realized through explicit modeling of cooperative signals among articles, optimization of multi-modal feature denoising and complementation and fine-grained distinguishing of user-article interaction importance, and the personalized recommendation effect is improved.

The invention discloses a farmland carbon sink data storage and tracing system and method based on a block chain, and the method comprises the following steps: collecting and preprocessing farmland plot data, and generating a standardized farmland carbon sink input feature vector set; inputting the feature vectors into a tree-shaped long and short-term memory network based on a Gumbel sampling mechanism, generating an original path set and executing structured dependency learning; based on the influence score of the path node on the carbon sink output, adjusting the structure sequence to generate a causal guide path set; executing attention aggregation, extracting a high response path, and generating a significant path coding set; generating a hash abstract for the code set and packaging a block chain transaction structure; calling an intelligent contract to verify a transaction structure and then writing the transaction structure into a block chain account book to form a carbon sink evidence storage record; and reconstructing a path diagram and generating a carbon sink traceability map to support result verification and structure traceability. According to the invention, credible evidence storage and structured traceability of the farmland carbon sink result are realized, and the prediction accuracy and supervision verifiability are improved.

The invention discloses a time series prediction method based on general and unique dependency balance modeling, belongs to the technical field of time series prediction, and aims at overcoming the defects of an existing non-stationary time series prediction method, learning general dependency and unique dependency by modeling a stationary sequence and an original non-stationary sequence respectively. And a time slice dependency learning layer and an inter-channel relationship learning layer are designed based on an adaptive stationary-non-stationary attention mechanism, and dynamic balance modeling of general dependency and unique dependency can be realized in two dimensions of time and channels, so that the accuracy and generalization performance of non-stationary time sequence prediction are effectively improved.