25 results about "Structural representation" patented technology

The invention provides a multivariable time series prediction method and system based on structural representation learning, and belongs to the technical field of multivariable time series prediction.The method comprises the steps that a similarity matrix between variables is constructed according to multivariable time series data, and variable community tags are obtained through community detection; carrying out weighted binarization to obtain a binary community sensing adjacency matrix; inputting the initial variable feature matrix and the adjacent matrix into a graph auto-encoder to obtain a variable structure embedded matrix; variable value embedding, time step position embedding and structure embedding are fused to obtain a unified input matrix, a time sequence prediction model based on a structure information guiding attention mechanism is input for modeling, and a prediction result is obtained; wherein structure embedding serves as an auxiliary feature to participate in modeling, and explicit structure constraint is not applied. According to the method, variable association is learned and mined through structural representation, and the accuracy and generalization ability of multivariable time sequence prediction are improved.

The invention requests to protect a defect severity prediction method based on multi-modal comparative learning, which comprises the following steps of: firstly, performing standardization processing on a defect report through Prompt and a guide large language model, generating explainable defect level description, and then extracting semantic features by using a text encoder of CLIP. Secondly, constructing a symbol-level hypergraph structure corresponding to the source code, and designing an improved hypergraph neural network to capture an advanced structure dependency relationship in the source code so as to obtain structural modal features; then, source codes related to the defects are visualized into images, and visual semantic information of the source codes is extracted through a CLIP visual encoder; and finally, fusing the structure representation of the code with the visual features, carrying out comparative learning alignment with text modal features, modeling multi-modal association through a shared semantic space, and realizing accurate prediction of the severity level of the defect report. According to the method, the intelligence level and accuracy of defect severity prediction are remarkably improved.

The invention provides a Riemannian graph word segmentation device for structure knowledge migration, which belongs to the field of graph basic models, and comprises a geometric vocabulary sampling module used for sampling input graph data to obtain geometric vocabularies; the geometric vocabulary encoding module is used for respectively mapping geometric vocabularies into corresponding constant curvature Riemannian spaces for coordinate encoding according to different structural modes of the geometric vocabularies to obtain space coordinates corresponding to the structural modes of the geometric vocabularies, discretization is carried out by using the Riemannian quantization module to obtain quantization marks, and finally, the geometric alignment decoding module is used for decoding the geometric vocabularies. And fusing the quantitative marks in different Riemannian spaces to obtain a unified graph structure representation capable of supporting structure knowledge migration. The problems that in the prior art, due to the fact that a curvature selection mechanism of a local form of a graph structure is lacked, a word segmentation device cannot dynamically adapt to geometric characteristics of the structure, and structural representation aliasing exists, the graph structure coding accuracy is insufficient, and the cross-domain knowledge migration generalization ability is low are solved.

Disclosed are techniques that generate a structural representation of a plurality of documents, the structural representation including a plurality of nodes and a plurality of edges, with the plurality of nodes being representations of the plurality of documents and the plurality of edges representing a feature in common between nodes of the plurality of nodes, with each node holding a vector of confidence values for weak models on a current optimization step and a weighted prediction for each of the weak models, generate a local ensemble model from the structural representation of the plurality of documents combined with the weighted prediction of the weak models, with the generated local ensemble model having a higher predictive power than any weak model individually, and generate a label for each node based on the local ensemble model.

A method may include identifying an input molecule. A structural representation indicative of a conformation of the input molecule may be generated. In some cases, the structural representation may be a canonical conformation signature or a multimodal representation encoding the three-dimensional structural features of the input molecule. An output molecule may be generated by applying a molecule design computation model to modify the amino acid residue sequence of the input molecule. The modifying of the ammino acid residue sequence may be guided by the structural representations of the input molecule in order to preserve the conformation of the input molecule. In some cases, one or more properties of the output molecule may be determined based on the structural representation of the output molecule. Related systems and computer program products are also provided.

This invention relates to the field of cheminformatics. It provides a multi-scale reaction representation learning method based on reaction structure constraints. The method includes the following steps: acquiring chemical reaction data; identifying structural change regions based on the structural identification information and mapping these regions to corresponding structural unit locations to generate structural unit-level annotation information; performing a local structure modeling task on a perceptual model based on the structural representation sequence to enable the perceptual model to perceive structural change regions; and training the perceptual model based on a global structure mapping task to obtain a multi-scale reaction representation model. This invention can introduce reaction structure change constraints during model training and model the mapping relationship between local structural change information and the overall structure at different structural scales, while simultaneously achieving effective synergy between the two, thereby improving the model's performance in organic synthesis reaction modeling tasks.

The present application relates to the technical field of image processing, and discloses a gait recognition method, which captures appearance contour information from different angle positions by polar coordinate sampling at R different sampling angles, explicitly models contour structure in a polar coordinate system, makes up for the deficiency of traditional appearance-based methods in capturing structural features, and thus improves the robustness of gait recognition to appearance changes. Furthermore, the present application also performs multi-scale fusion on structural features of different sampling angles to suppress redundant information and enhance the stability and distinguishability of structural representation. In addition, the present application combines gait features with global information and local information on the final structural features obtained by polar coordinate sampling, maintains the robustness to appearance disturbance, and improves the sensitivity of the model to shape changes. The present application effectively alleviates the influence of view angle change, clothing change and carrying articles, and greatly improves the robustness of appearance contour modeling.

The application provides an agent tool call optimization method, comprising the following steps: when an agent successfully completes a task for the first time, recording the agent tool call sequence, input and output parameters and execution results, generating a structured call log, and converting the call log into a standardized call path knowledge unit; structurally and semantically representing the call path knowledge unit, storing the structural representation in a graph database, and storing the semantic representation in a vector database; taking the task intent, tool entity and call path as heterogeneous nodes, constructing an experience path knowledge graph, the experience path knowledge graph is used for recording the multi-dimensional relationship among the task, path and tool, and the execution performance attribute and feedback attribute are attached to the path node in the graph, and the agent tool call knowledge optimization is completed. The purpose of improving the tool call efficiency and robustness of the agent in a multi-task environment is achieved.

This invention belongs to the field of code vulnerability detection and provides a vulnerability detection method and system based on the fusion of multi-scale graph representation and instance awareness. The method includes: constructing a code attribute graph; constructing local dependency views and global dependency views based on the code attribute graph; processing the local dependency views and global dependency views using a trained dual-branch structure to obtain a first structural representation and a second structural representation; fusing the first structural representation and the second structural representation to obtain a target structural representation; performing code sequence semantic analysis on function-level code fragments to extract function-level semantic representations; dynamically aligning the function-level semantic representations with the target structural representations to obtain function-level fused feature representations; and performing vulnerability classification and detection based on the function-level fused feature representations to obtain vulnerability detection results. This invention can take into account local microstructural information, global long-range dependency information, and code sequence semantic information, improving the accuracy and robustness of vulnerability detection.

The invention provides an information diffusion prediction method and system based on a denoising diffusion model, a medium, equipment and a terminal. According to the method, node-level anisotropic mixed directional noise jointly driven by structural features and preference features is introduced in the forward diffusion process, noise adding disturbance is carried out on low-preference homogeneity structural representation, and a residual condition denoising network is utilized in the reverse denoising process. In combination with preference embedding and homogeneity graph topology self-adaptive recovery structure representation, the technical defect that representation redundancy is caused due to the fact that in the prior art, only a social structure is relied on, and low homogeneity relation noise is difficult to restrain is overcome. In addition, a comparative learning target of preference perception is designed, and the technical prejudice that the structural space and the preference space are separated and are difficult to represent uniformly is effectively relieved. According to the hybrid directional denoising diffusion model provided by the invention, the preference homogeneity of user structure features can be remarkably improved, and the characterization capability and prediction precision of an information diffusion prediction model are enhanced.

This disclosure provides a data processing method, comprising: acquiring interactive behavior data of a target user under reference indicators; determining a behavior representation vector based on the interactive behavior data; determining isolated anomaly-type representations and structural representations corresponding to the target user based on the behavior representation vectors; determining event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly-type representations and structural representations based on pre-configured custom rules and rule matching functions; determining the correlation strength attribute between the event chain attribute, sentiment evaluation attribute, and a pre-created graph structure and candidate essential issues; and determining the attribution analysis result of the target user based on the correlation strength attribute. The technical solution of this disclosure, based on the behavior representation vectors corresponding to interactive behavior data, determines the attribution analysis result associated with the target user, improving the comprehensiveness and accuracy of anomaly detection, and achieving a more accurate attribution analysis result for the target user.

The invention provides an OCT (Optical Coherence Tomography) denoising method based on self-supervised structural representation learning. The method comprises the following steps: forming a self-supervised OCT denoising model by using an encoder and two parallel coupled decoders; the encoder and the two parallel coupled decoders are stacked into an eight-layer structure in an iteration form; the two parallel coupled decoders are respectively a proxy task decoder and a downstream task decoder; and the proxy task decoder and the downstream task decoder share features output by the encoder and are coupled in a feature splicing mode. According to the method, through an innovative multi-view synthesis degradation strategy, training samples with different noise characteristics can be automatically generated only by utilizing a noiseless OCT image which is easy to obtain, and an input-target pair required by self-supervised learning is constructed; the model carries out cooperative training through parallel structural representation learning and an image reconstruction task, and any manually-labeled noise-clear image pair or a tedious parameter tuning process is not needed.

This invention provides a structure-aware adaptive Gaussian splashing method, belonging to the field of 3D scene reconstruction and rendering technology. First, multi-view images of the scene and camera parameters are acquired, and 3D Gaussian primitives are initialized. Then, a structure-aware mask is constructed using morphological processing and surface normal consistency constraints to enhance the geometric representation of thin-structure regions and occlusion boundary regions. Next, a cascaded multi-resolution optimization strategy is employed to progressively update the Gaussian primitive parameters. Then, the orientation and anisotropic shape of the Gaussian primitives are adaptively adjusted based on rendering feedback. Finally, a multi-index evaluation strategy is constructed, combining uncertainty, color entropy, gradient intensity, and view coverage to complete the importance assessment and pruning of the Gaussian primitives, obtaining an optimized set of 3D Gaussian primitives for target image rendering. This invention can effectively improve the structural representation and detail restoration effects in complex scenes, rendering high-quality scene images.

This invention discloses a glass process parameter simulation method based on image generation, belonging to the field of industrial visual analysis technology. The method comprises the following steps: S1, establishing a set of structural representations of glass process images; S2, constructing a spatial distribution of structures using the image structural representation set; S3, mapping process physical parameters using a structural spatial parameter field; S4, constructing a simulation parameter mapping structure using the set of process physical parameters; and S5, generating and calculating simulation parameters using the simulation parameter mapping set. By setting a structural spatial parameter field, this invention can convert structural representation elements extracted from images into continuous spatial distribution parameters through spatial mapping and local structural density calculation. This forms a structural spatial parameter expression that reflects the changing trend of the glass melt structure in a spatial region, enabling image structural information to provide continuous spatial foundation data for subsequent glass process parameter mapping and simulation calculations.

The invention relates to the technical field of graph neural networks, in particular to a high-order structure perception hypergraph convolutional network construction method and device, a medium and equipment, and the method comprises the steps: generating candidate sub-graphs through frequent sub-graph mining based on an original graph structure, introducing a sequential embedding mechanism to perform structure representation learning and screening on the candidate sub-graphs to obtain frequent sub-graphs, dynamically generating a frequent sub-graph set, constructing a high-order hyperedge set based on the frequent sub-graph set, and constructing a hyperedge structure based on the high-order hyperedge set; and performing feature aggregation and representation learning of high-order structure semantics on node features in the hypergraph structure by adopting a hypergraph convolutional network, and generating enhanced node representation. According to the method, the reasonability, representativeness and task correlation of high-order structure selection can be improved, then the expressivity, interpretability and robustness of the model are improved, and the method has wide application prospects and industrialization value.

The invention discloses a semantic continuity guarantee method and device, a terminal and a storage medium, and the method comprises the steps: carrying out the structural representation of a natural language demand, and determining the structural demand information; converting the structured demand information into a formalized protocol in a linear tense logic form; performing semantic representation and semantic alignment on the formalized protocol according to the first image data, and determining aligned semantic representation; decoding the aligned semantic representation, and determining executable data; determining a target mode according to the available CPU resources of the system and the edge equipment information, performing runtime guarantee based on the target mode and the executable data, and determining a runtime guarantee result; and generating a device control instruction based on the runtime guarantee result through the operating system security center. Therefore, the problems that the existing method in the prior art cannot fundamentally solve the problem of lack of full-link semantic continuity and is difficult to meet the high-reliability operation requirement of the safety key AI system in a dynamic environment can be effectively solved.

This invention proposes a collaborative filtering recommendation method that integrates explicit structural constraints and adaptive feature gating, comprising the following steps: S1, modeling user-item interactions using lightweight neighborhood aggregation and introducing explicit one-hop structural constraints to obtain a fused structural representation matrix of nodes; S2, designing an adaptive feature gating mechanism to calculate the activation probability of structural information in each feature dimension and obtain the updated representation of nodes; S3, constructing a structurally consistent lightweight contrastive learning objective to obtain contrastive representations; S4, constructing a joint optimization objective function to obtain the final model parameters; S5, based on the model parameters optimized in step S4, calculating the preference scores between users and items and outputting the recommendation results. This invention can reduce the risk of noise accumulation caused by multi-hop propagation and stabilize the representation learning process, improve the discriminativeness and robustness of embedded representations, and alleviate the semantic aliasing problem caused by the overall injection of structural signals.

In one embodiment, a method includes receiving training utterances associated with a domain, receiving ontology labels for the domain, wherein the ontology labels comprise one or more of an intent or a slot, generating an inventory for the domain, wherein the inventory comprises at least a respective index and respective span for each intent or slot, wherein the respective span comprises a respective descriptive label associated with the intent or slot, and wherein the respective descriptive label comprises a natural-language description of the intent or slot, generating frames for training utterances based on the training utterances and the inventory by a natural-language understanding (NLU) model, wherein each frame comprises a structural representation of the respective training utterance, wherein the structural representation is generated based on a comparison between the corresponding training utterance and the inventory, and updating the NLU model based on the frames.

The application discloses a cryo-EM density map self-supervised learning method, system, storage medium and equipment, and belongs to the technical field of computational structural biology and deep learning. The method comprises the following steps: acquiring a cryo-EM density map, constructing a CryoLVM model based on a self-supervised learning framework, pre-training the CryoLVM model according to the cryo-EM density map, and learning the structural representation of the density map; using the pre-trained CryoLVM model to perform a downstream cryo-EM task, outputting a predicted density map, comparing the predicted density map with a target density map, and fine-tuning the CryoLVM model. The cryo-EM density map self-supervised learning method improves the training efficiency and universality of the cryo-EM density map processing model.

A method can include acquiring resistivity measurements using a downhole tool of a drillstring disposed in a borehole in a subsurface environment; performing a resistivity measurement-based inversion to generate a structural representation of a portion of the subsurface environment that includes an end of the borehole; generating a control instruction using an artificial intelligence framework and the structural representation, where the control instruction is for lengthening the borehole along a current borehole trajectory or a different borehole trajectory; and controlling the drillstring to lengthen the borehole based on the control instruction.

Provided herein is a method for reviewing code. The method can comprise parsing the code to generate a structural representation of the code, wherein the structural representation comprises a graph representative of the code. The method can comprise processing the code and the structural representation to generate a context for the code based at least in part on the graph. The method can comprise processing the context and a set of prompts to generate a set of contextualized prompts, wherein at least two prompts in the set of prompts are specific for different categories of issues. The method can comprise prompting a first set of language models with the set of contextualized prompts to generate a set of issue reports. The method can comprise prompting a second set of language models to generate a set of validated issue reports comprising a set of non-hallucinated issue reports.

The application relates to the fields of biological information processing, computational biology and intelligent computing technology, and discloses a candidate intervention generation method and system based on multi-modal biological data unified representation and structural anomaly inversion. The method acquires multi-modal biological data of a target object, wherein the multi-modal biological data comprises at least two types of genomic data, transcriptomic data, protein interaction data, metabolomic data and phenotype data; the multi-modal biological data is subjected to standardization processing and modality alignment, and a unified structural representation is constructed; based on the unified structural representation, a structural anomaly index of the target object relative to a reference steady-state object is calculated, and an abnormal structural representation unit is determined, or a target structural constraint set is extracted based on a target functional state, a target spatial configuration or a target structure input; a candidate intervention structure or a candidate functional structure output object is generated based on the abnormal structural representation unit or the target structural constraint set; consistency verification, error correction, constraint compliance checking, scoring and sorting are performed on the candidate result through a multi-scale audit module, and the candidate result is output. The application can realize unified fusion of multi-modal biological data, system-level abnormal structure positioning, target-driven reverse generation and multi-scale audit verification, reduce the candidate search space, and improve the explainability, reproducibility and priority sorting efficiency of the candidate output.

The application provides a multivariate time series prediction method and system based on structure representation learning, and belongs to the technical field of multivariate time series prediction. The method comprises the following steps: constructing a similarity matrix between variables according to multivariate time series data, obtaining a variable community label through community detection, and obtaining a binary community perception adjacency matrix through weighted binary coding; inputting an initial variable feature matrix and the adjacency matrix into a graph autoencoder to obtain a variable structure embedding matrix; obtaining a unified input matrix by fusing variable value embedding, time step position embedding and structure embedding; inputting a time series prediction model based on structure information guided attention mechanism modeling to obtain a prediction result; wherein the structure embedding serves as an auxiliary feature for modeling without applying an explicit structure constraint. The application mines variable correlation through structure representation learning, and improves the precision and generalization ability of multivariate time series prediction.

This invention provides a method for generating remote sensing images of targets with few samples based on structure awareness and detail enhancement, relating to the fields of computer vision and deep learning. It addresses the limitations of current diffusion models, such as training instability and structural information degradation. The method only allows modules related to conditional modeling and structural representation in the diffusion generation model to participate in parameter updates, while all other parameters are frozen. The method employs a structure-aware edge supervision module to constrain the predicted results at structural edges; a structure-isolated image detail enhancement module to enhance the details of the output pixel image, and a gradient isolation mechanism prevents the perceptual constraints of the image detail enhancement stage from being propagated back to the diffusion backbone network; and an inference quality self-evaluation control module to evaluate the quality of the generated results and control the output, triggering regeneration when conditions are not met. This invention achieves the advantages of stable and controllable generation of remote sensing images under few-sample conditions and enhanced realism.

The application discloses a brain electrical signal denoising model training method, a denoising method and a data enhancement method, wherein the training method comprises: training a denoising model using a training set, the denoising model comprising a generator and a discriminator, the generator comprising an encoder, a latent space alignment block and a decoder, the discriminator comprising a feature extractor and a classifier, and the classifier comprising a domain classifier and a category classifier. The application introduces an adversarial training mechanism between the generator and the discriminator to constrain the authenticity of the generated signal. In the model training process, a latent space alignment constraint is designed to make the generated signal closer to the real signal in the high-level semantic feature or structural representation level, thereby effectively alleviating the training degradation problem and improving the adaptability of the model to different noise distributions. In addition, the latent space alignment constraint can also enhance the stability of the model in the feature expression layer, so that the denoising result still maintains good robustness when facing distribution shift or complex noise environment.