175 results about "Paired Data" patented technology

The invention discloses a protein generation model optimization method based on deep learning, and relates to the technical field of protein generation. The optimization method comprises the following steps: generating a candidate protein electron density distribution diagram based on the structural characteristics of a target spot by adopting a pre-trained diffusion model; converting the candidate protein electron density distribution diagram into a corresponding amino acid sequence; determining a functional index value corresponding to the amino acid sequence, and constructing a feedback data set; setting a reward threshold value, and marking the feedback data set as a positive sample data set and a negative sample data set according to the reward threshold value; constructing a utility function taking the reward threshold as a reference point, and respectively calculating utility values of the positive sample data set and the negative sample data set; and performing iterative optimization on the diffusion model according to the utility value. By adopting the technology provided by the invention, the dependence on preference paired data of large-scale and high-quality protein sequences can be avoided, and the performance of the protein generation model can be effectively and continuously improved.

The invention discloses a laser design system, a laser design method, electronic equipment and a medium, and belongs to the technical field of integrated optics. The laser design system comprises a first model, input parameters of which comprise set grating parameters of a grating structure of a first to-be-designed laser, and which is used for predicting and outputting a predicted output characteristic curve of the first to-be-designed laser based on the set grating parameters; the input parameters of the second model comprise a set output characteristic curve of a second laser to be designed, and the second model is used for predicting and outputting predicted grating parameters of a grating structure of the second laser to be designed based on the set output characteristic curve; the first model and the second model share the same training data set, the input training data of the first model corresponds to the output training data of the second model, and the output training data of the first model corresponds to the input training data of the second model; the training data set includes pairing data between grating parameters of a grating structure of the sample laser and an output characteristic curve of the sample laser.

The invention discloses a security constraint type inverse kinematics solving method and device and a storage medium. The method comprises the following steps: firstly, forming a joint angle-tail end pose pairing data set by students through forward movement of a robot; training an improved RBF (Radial Basis Function) network by using the data set to establish initial mapping from a tail end pose to a joint angle to obtain an initial approximation model; taking the prediction output of the model as an initial population of a PSO algorithm, and updating the speed and position of particles in the initial population through iteration; designing a comprehensive cost function, introducing a joint angle constraint and an obstacle collision constraint, and evaluating the advantages and disadvantages of each particle; and when the PSO algorithm reaches the maximum number of iterations or the comprehensive cost function value is lower than a set threshold value, outputting a global optimal particle position as a final inverse kinematics solution. On the premise of ensuring the solving precision, the joint limit constraint and the obstacle avoidance constraint are explicitly processed, so that an accurate, safe and feasible inverse solution is output, and the reliability and the safety of robot operation are remarkably improved.

The invention discloses a two-arm smart teleoperation method and system based on motion redirection and self-adaption, and relates to the technical field of robots, and the method comprises the steps: determining the posture of a human hand and the posture of a human arm according to a human body image; the robot is redirected according to the human hand posture and the human arm posture through the space attention gating map convolutional network, and the joint angle of a mechanical hand of the robot and the joint angle of a mechanical arm of the robot are correspondingly obtained; the mechanical arm is driven according to the joint angle of the mechanical arm so that the mechanical arm can grab the target object; and target force needed for grabbing the target object is determined, and the mechanical arm is driven to grab the target object according to the target force and the joint angle of the mechanical arm. According to the method, topology modeling and cross-platform action redirection between the human body and the robot are achieved through the space attention gating map convolutional network, pairing data dependence is eliminated, and therefore the high precision of action reproduction and the safety and flexibility of the grabbing process are both considered in the complex dynamic environment.

The invention provides a multi-data combined construction site three-dimensional model construction method and system, and relates to the technical field of data processing, and the method comprises the steps: carrying out the calling of a preset resolution satellite image, and obtaining a full-region stereo image pair data set; after radiation correction is carried out, a region surface model and a region elevation model are generated; grid-level modeling defect detection is carried out, and distributed blind area blocks are positioned; performing obstacle avoidance path fitting to generate an obstacle avoidance optimized flight path; the unmanned aerial vehicle is driven to execute multi-angle surrounding scanning, and distributed blind area dense point clouds are directionally collected; constructing a distributed blind area model; and carrying out microscopic superposition, and outputting a three-dimensional model of the construction site. The technical problems that a construction site modeling method in the prior art is limited in the aspects of coverage, modeling efficiency and data real-time performance, depends on a single data source, cannot provide a comprehensive and accurate three-dimensional model of a construction site, and affects the construction progress and decision are solved.

The invention provides a data desensitization method which can be applied to the technical field of data security. The method comprises the following steps: receiving a data stream from a real-time data source; performing data format identification on continuously arriving data in the data stream to obtain a data format; analyzing the data by using an analyzer corresponding to the data format to obtain key-value pair data, the key-value pair data comprising respective key-value pairs of at least one field, and the key-value pairs comprising keys representing field identifiers of the fields and values representing the data of the fields; for any field in the at least one field, determining a desensitization algorithm corresponding to the field under the condition of determining that the field meets a sensitive data matching condition according to at least one of the key or the value and the sensitive data matching condition indicated by each of the plurality of desensitization rules; and desensitizing the value corresponding to the field by using an execution component corresponding to the desensitization algorithm.

The invention discloses an equipment fault and maintenance knowledge intelligent association method based on a large model, and relates to the technical field of equipment intelligent operation and maintenance and industrial artificial intelligence, and the method comprises the steps: carrying out the structural analysis of historical fault work order data and maintenance record data of target equipment, and constructing a fault-maintenance knowledge pair data set; performing semantic coding on the fault description text and the maintenance scheme text by using a pre-trained large model, generating a corresponding semantic vector, and constructing a fault-maintenance semantic association map; and performing semantic coding on the real-time fault description, screening candidate fault nodes based on semantic similarity, calculating the semantic matching degree of the maintenance scheme in the semantic association map, and outputting a target maintenance scheme sequence after matching degree sorting. According to the invention, the intelligent level of equipment fault maintenance and the reuse efficiency of historical maintenance knowledge are improved.

The invention provides a data generation model training method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence, and the method comprises the steps: constructing a preference data pair which comprises an initial question and answer pair and an optimized question and answer pair; inputting the preference data pair into a to-be-trained large language model, and obtaining an enhanced question and answer pair output by the to-be-trained large language model; updating the preference data pair to obtain an updated preference data pair, including: setting the enhanced question and answer pair as an optimized question and answer pair in the updated preference data pair; and iteratively executing the steps of constructing the preference data pair and updating the preference data pair until a preset termination condition is met, and obtaining a trained large language model. According to the method, the question and answer pair data is optimized by utilizing the ability of the large language model, and the large language model is trained by using the preference data pairs subjected to iterative updating, so that high-quality question and answer pair data can be generated efficiently at low cost.

The application discloses a multi-modal heterogeneous-based unmanned aerial vehicle image super-resolution reconstruction method and system and a medium, relates to the technical field of image processing, and collects multi-channel images of a first resolution and reference panchromatic images of a second resolution, generates a paired data set after preprocessing, extracts multi-channel feature maps and panchromatic feature maps from the paired data set based on a MultiHetSRNet model, generates a multi-modal heterogeneous graph by modeling nodes of a graph structure, and generates a relationship mode matrix describing spatial relationships and cross-modal relationships between nodes; target multi-channel images are reconstructed by aggregating local features and global features according to the relationship mode matrix; the scheme is improved in the method on the basis of the prior art, multi-channel images of an unmanned aerial vehicle and remote sensing reference panchromatic images are fused, complex spatial relationships are modeled by using a multi-modal heterogeneous graph, the robustness of a model is improved in combination with multiple degradation modes, and high-quality super-resolution reconstruction is realized.

Provided herein are computer-implemented methods of generating a disease map of a population. In some embodiments, the methods include applying a clustering algorithm to a set of weight and bias values of a trained electronic neural network to generate the disease map of the population. In some embodiments, the electronic neural network has been trained on training data that comprises representations of peptide sequence and binding value pair data sets obtained from reference subjects in the population in which a given peptide sequence and binding value pair data set comprises peptide sequence information and peptide binding values of antibodies to peptides that comprises the peptide sequence information. In some embodiments, the antibodies are from a sample obtained from a given reference subject in the population and are indicative of one or more disease states. Related systems, computer readable media, and additional methods are also provided.

Systems, apparatuses, methods, and computer program products are disclosed for monitoring and tracking technical engagements between digital resources. An example method includes deploying a machine learning monitoring application to a computing environment comprising one or more of a computing device or a network channel and detecting an engagement metric associated with a target technology of the computing environment and a user account of the computing environment. The example method further includes generating name-value pair data representative of the engagement metric, the user account, and a timestamp token and generating a technical engagement score representative of a probability of a successful utilization of the target technology by the user account. The example method further includes initializing an actionable instruction set based on a comparison of the technical engagement score to a technical engagement threshold.

The invention discloses a text question-answer pair data set generation method and device, a medium and electronic equipment, and relates to the field of natural language processing. The method comprises the steps of performing hierarchical standardization analysis on a to-be-processed text, and determining a chapter logic hierarchical structure; extracting candidate knowledge points based on the chapter logic hierarchical structure, and reversely aggregating the semantic context by taking the candidate knowledge points as a core to generate a structured tetrad; positioning each structured tetrad in the to-be-processed text to obtain a text position corresponding to the structured tetrad; based on the similarity between the knowledge points in different structured tetrads, aggregating the structured tetrads meeting the similarity requirement and the corresponding text positions, and generating an index table corresponding to the knowledge points; and based on the index table, generating a question and answer pair data set corresponding to the to-be-processed text. Through an index table for associating knowledge points dispersed in different hierarchical units, the defect that a traditional method is limited to context and cannot reflect the whole knowledge structure is overcome.

The application provides a multimodal graph table question and answer large model construction method, an electronic device and a storage medium, comprising: training a graph-text alignment model based on a first sample data set to obtain a trained graph-text feature alignment model; wherein the first sample data set comprises image samples and corresponding text content; training a multimodal graph table question and answer large model with the trained graph-text feature alignment model based on a second sample data set to obtain a trained multimodal graph table question and answer large model as a final multimodal graph table question and answer large model, and the second sample data set comprises context representation information of a graph table sample, an image and question and answer pair data. The multimodal graph table question and answer large model obtained by the application can further improve the graph table question and answer capability of the existing multimodal graph table question and answer large model, and has strong Chinese understanding capability.

The invention provides a roadside camera calibration data processing method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a credible region of a roadside camera based on the installation parameters and equipment parameters of the roadside camera, determining a target track of a target vehicle in the credible region, and determining the target track of the target vehicle based on a target time period of the target vehicle running on the target track, determining a video clip shot by the roadside camera in the target time period; and matching the pixel coordinates of the target vehicle in each frame of image in the video clip with the world coordinates of each trajectory point in the target trajectory according to the timestamp, and generating a pairing data set of the pixel coordinates and the world coordinates, the pairing data set being used for calibration of the roadside camera.

The invention discloses a data processing method and device, computer equipment, a medium and a product, and the method comprises the steps: obtaining N target objects in a target business scene, determining an object group to which each target object belongs, the object group comprising a control group or an experimental group, and the control group and the experimental group applying different experimental strategies; determining at least one piece of paired data corresponding to the N target objects according to the interaction data records of the N target objects, wherein each piece of paired data comprises object identifiers of the two target objects and group identifiers to which the two target objects belong; determining influence degree data according to the at least one piece of paired data, and determining associated evaluation data corresponding to the at least one piece of paired data according to the influence degree data and the at least one piece of paired data; and determining an inspection result of the target service scene according to the association evaluation data and the influence degree data, wherein the inspection result is used for indicating whether the target service scene has a network effect. The method can improve the accuracy of network effect inspection based on the paired data.

The application discloses a kind of real-time assessment system of pressure sore risk based on finite element analysis and deep learning, it is related to medical monitoring and biomechanics simulation field;Solve the problem that existing assessment lacks real-time monitoring of deep tissue stress, dynamic early warning is insufficient.The application acquires body surface pressure, temperature and continuous pressure duration in real time by flexible pressure sensing blanket;Offline finite element simulation constructs parameterized human-mattress model, generates the paired data set of body surface pressure and deep tissue equivalent stress;Physical information deep learning model is based on pre-training real-time inference deep tissue equivalent stress distribution;Risk evolution calculation combines temperature and continuous pressure duration, and according to damage evolution law, calculates cumulative damage risk index and residual safety time;Visual early warning generates risk thermodynamic diagram and triggers hierarchical early warning and intervention instruction push.The application realizes real-time, accurate, individualized assessment and early warning of pressure sore risk.

To provide data collection for capturing high-quality evidence that may to lead to formulation of convincing and credible answers to questions that have posed.SOLUTION: Embodiments of the present invention provide computer-implemented methods, computer program products and computer systems. Embodiments of the present invention can identify a plurality of data variables within a multivariate event dataset. Embodiments of the present invention can then formulate a causal inference between at least two identified data variables within the multivariate event dataset and generate a structural framework of an average effect value for the multivariate event dataset based on the formulation of the causal inference of the identified data variables. Embodiments of the present invention can then calculate an inverse propensity score for the generated structural framework of the average effect based on a type of identified variable, a predetermined time associated with the identified variable, and a causal connection strength between the identified variables.SELECTED DRAWING: Figure 1

The present application relates to a kind of small sample conditions under the accurate identification method of heterogeneous SAR image target, comprising: sample expansion is carried out to small sample conditions under the heterogeneous SAR image and constitutes support set;The support set contains the support sample of target labeled of N categories;Target to be labeled SAR image is as query sample and constitutes query set;One query sample in query set and any optional sample in each category in support set constitute sample pair;N sample pairs are input to embedding network module, image pre-processing and feature extraction are carried out, and N feature pair data containing support sample feature and query sample feature are obtained;N feature pair data are input into similarity network module and similarity discrimination scoring is carried out, and the class of support sample in the highest feature pair data of similarity score is regarded as the class of query sample and target labeling is carried out.The present application realizes the accurate identification of heterogeneous SAR image target.

The invention discloses a visual localization and attitude estimation method and device based on prior search in a rocket recovery section and a medium, and belongs to the technical field of spacecraft guidance, navigation and control, and the method comprises the steps: carrying out the systematic sampling of a rocket recovery tail end operation space containing a three-dimensional position and a three-dimensional camera attitude in advance; constructing a matching data set in which the rocket recovery state is matched with the image; learning mapping from an image to a low-dimensional feature vector by using a deep learning model, and enabling the structure of a feature space to be consistent with the structure of a physical state space through a designed loss function; and in the rocket recovery stage, pre-stored samples closest to real-time image features are retrieved, and state labels of the pre-stored samples are fused, so that pose estimation is realized. The method can adapt to complex three-dimensional attitude changes, is high in robustness, and can meet the real-time requirement.

Disclosed in the present invention are a self-supervised image denoising method and system. The method comprises: by means of an input interface, acquiring a noise image and transmitting same to a memory; configuring, in the memory, a search space based on a U-Net framework; a graphics processing unit reading configuration information, executing coarse-grained population initialization on the basis of a decimal modular encoding strategy, and writing population data into the memory as a contiguous memory block; the graphics processing unit executing distance-guided parent selection and modular crossover and mutation operations in parallel; decoding offspring individuals into network structures, then executing self-supervised denoising processing in parallel, and calculating PSNR values; and a central processing unit executing environment selection and controlling an iteration process, finally selecting an optimal network structure, and outputting a denoised image. By means of the optimization of search space design and the collaboration of heterogeneous computing architectures, the present invention simultaneously realizes high-quality image detail restoration and efficient denoising processing, without requiring paired data.

The invention discloses a non-pairing underwater image enhancement method based on structure perception multi-mode fusion, and belongs to the technical field of underwater image processing. Comprising the following steps: constructing a generator network based on a U-shaped Transform; training a generator network by using a non-paired data set, and carrying out combined constraint by using generative adversarial loss and multi-scale Patch NCE comparison loss; and performing channel splicing on a to-be-processed underwater RGB image, the space depth image and the structure prior image, constructing a multi-modal input tensor, then feeding the multi-modal input tensor into the trained generator network, and outputting an enhanced underwater image through forward propagation. According to the method, the depth map and the structure prior map are introduced, the perceptual ability of the network to the geometric contour is enhanced, global modeling and non-pairing comparative learning of Transform are combined, high-fidelity restoration of the high-turbidity underwater image is achieved under the condition that paired data is not needed, and the method can be widely applied to the fields of ocean exploration, underwater monitoring and the like.

A computer-implemented method of processing first metrology data of a patterned portion of a substrate patterned using a wafer processing apparatus to generate predicted second metrology data of the, or another corresponding, patterned portion of the substrate, the first metrology data comprising metrology data from a first metrology domain which obtains metrology data using a first metrology process. The method comprises obtaining the first metrology data of the patterned portion of the substrate, and generating predicted second metrology data of the patterned portion of the substrate, or of the corresponding patterned portion of the substrate, based on the output of a first trained diffusion model conditioned on the first metrology data, the predicted second metrology data comprising predicted metrology data from a second metrology domain which obtains metrology data using a second metrology process.

The invention relates to the technical field of infrared image processing, and discloses an infrared image enhancement method based on radiation priori and a generative adversarial network, which comprises the following steps of: constructing a training data set of non-paired low-quality infrared images and high-quality infrared images, constructing a bidirectional cyclic generative adversarial network, pre-establishing infrared radiation quantity databases of different materials, and constructing a training data set of the non-paired low-quality infrared images and high-quality infrared images; the method comprises the following steps: adding a material coding branch in a generative network, inputting the material coding branch in a mapping database, obtaining radiation coding characteristics, constructing a composite loss function, and carrying out feedback optimization on a discriminant network and the generative network by using a training data set through the composite loss function to obtain a trained overall enhanced network. The infrared image to be enhanced is input into the enhancement network, the enhanced infrared image is output, and the defects that an existing infrared image enhancement method depends on paired data, material radiation characteristics are ignored, enhanced image semantics are inconsistent and the like are overcome.

The invention provides a de-learning method for a decoupled multi-modal language model. The method comprises the following steps: S101, dividing an original data set into a forgetting set and a reservation set; s102, obtaining a pre-trained model weight of the original data set, performing pre-training on the forgetting set and the retention set, and completely retaining knowledge of all training data; s103, preprocessing the forgetting set and the retention set, and constructing a concept pair data set; s104, constructing a visual concept detector, and training the visual concept detector by using the concept pair data set; and S105, integrating the visual concept detector into a reasoning process of the multi-modal language model, and modifying visual features containing different visual concepts or directly inputting the visual features into the multi-modal language model. According to the method, the multi-modal language model forgets the specified training set content, bad answers are prevented from being generated for sensitive input data, and the effects of protecting privacy and preventing information leakage are achieved.

The application provides a lightweight orthodontic efficacy prediction method and system based on semantics and three-dimensional optimization, relates to the technical field of orthodontic efficacy prediction, and comprises the following steps: obtaining facial images and cephalometric data of historical orthodontic patients before and after orthodontic treatment to form a training set, constructing a structured orthodontic semantic prompt, using the pre-training diffusion model as the basis, using the training set images and the semantic prompt, and fine-tuning the prediction model through the LoRA technology; inputting the orthodontic pre-image and data of a target patient into the image to the prediction model, generating a two-dimensional facial image after orthodontic treatment, and performing three-dimensional reconstruction; optimizing the geometric consistency with the side image, and outputting a three-dimensional model for visual display. Through the combination of the LoRA fine-tuning technology, the structured medical semantic prompt and the three-dimensional geometric consistency optimization, the application can train an efficient prediction model under a small amount of paired data, reduce the calculation cost of model training and reasoning, adapt to the deployment of a clinical environment, and realize the controllability and interpretability of the generated results through the medical semantic prompt.

The application discloses a 12-lead electrocardiosignal denoising method and system based on a dual generative adversarial network, which comprises the following steps: obtaining time-adjacent clean electrocardiosignal segments and noisy electrocardiosignal segments; constructing a noise generative adversarial network, wherein the generator takes a random noise vector as input and simulates noise as output, the discriminator takes the noisy electrocardiosignal segment as a positive sample and a synthetic signal composed of the clean electrocardiosignal segment and the simulated noise as a negative sample for adversarial training, and a noise generator snapshot library is generated; performing dynamic noise injection on the clean electrocardiosignal segment to obtain 'noisy-clean' paired data; constructing a denoising network comprising a denoiser; obtaining the 12-lead electrocardiosignal containing noise and inputting it into the denoiser to output the denoised 12-lead electrocardiosignal. The application aims to overcome the defects in the prior art, such as the lack of real 12-lead electrocardiosignal denoising paired data and the poor model generalization ability caused by excessive dependence on the Gaussian noise assumption.

The invention provides an oil field inspection multi-source vision and sensing data fused multi-modal large model construction method and system, and belongs to the field of oil field pipeline inspection. The problem of insufficient identification precision of a single data source in the existing oil field pipeline inspection technology is solved. The method comprises the following steps: loading or generating a pairing data set containing an oil field pipeline visual image and corresponding sensing time sequence data; performing deep semantic feature extraction on the visual image, performing time sequence correlation feature extraction on the sensing time sequence data, and respectively mapping to a unified feature dimension to obtain a visual feature vector and a sensing feature vector; based on a multi-head self-attention mechanism, carrying out bidirectional attention interaction on the visual feature vector and the sensing feature vector to obtain a fused multi-modal feature, inputting the fused multi-modal feature into a multi-modal large model, and carrying out optimization by adopting a lightweight fine tuning strategy; and based on the depth features of the large model, classifying and identifying the operation state of the oil field pipeline, and outputting a classification result. The invention relates to oilfield safety.

The invention relates to the technical field of industrial image processing, and discloses a cloth overexposure light-color line intelligent restoration method and system, and the method comprises the steps: collecting a cloth image, synthesizing a pairwise data set based on a dynamic structure mask and a reverse physical illumination model, and dividing the pairwise data set into a training set and a verification set; constructing a generator and a discriminator; a quintuple joint loss function is adopted; training the model by adopting staged dynamic loss weight scheduling and a multi-stage learning rate attenuation strategy; initializing network parameters; and self-adaptive weighted repair of the overexposure saturation area and the normal exposure area in the input image is realized through the automatically generated space gating mask, and a repaired image adaptive to industrial quality inspection is output. According to the method, strong light interference elimination, high-fidelity texture reconstruction and engineering landing can be considered.

This invention belongs to the field of vision and image processing technology, specifically providing an underwater image enhancement method and system based on latent space diffusion and adaptive style transfer. First, underwater and non-underwater image datasets are constructed and preprocessed using standardization. A pre-trained CNN is used to extract features from non-underwater images at multiple levels, calculating the Gram matrix of the convolutional layers as style features. The underwater images are input into a variational encoder, mapped to a low-dimensional latent space to obtain feature vectors. A diffusion denoising network predicts and removes noise in the latent space. An adaptive style transfer network combines the style features of the non-underwater images to achieve decoupling and fusion of content and style. Finally, the enhanced underwater image is reconstructed by a variational decoder. This invention solves the problems of existing methods relying on paired data, high computational complexity, and lacking an effective decoupling mechanism, achieving efficient underwater image color correction and sharpness enhancement.

This invention provides a closed-loop coupling and self-correcting method and system for power system full-element simulation and control based on digital twins, belonging to the field of power system simulation and control technology. The method includes the following steps: acquiring bidirectional data from the physical power system and a pre-constructed digital twin simulation model based on the physical power system; aligning the bidirectional data using an event detection algorithm; constructing a paired dataset based on the alignment results; calculating a mismatch metric on the digital twin simulation model; and calculating the optimal parameter update amount based on the mismatch metric; and updating the parameters of the digital twin simulation model to obtain the latest digital twin simulation model for prediction and decision-making in the next control cycle. This invention significantly improves the perception accuracy and long-term reliability and adaptability of the scheduling optimization effect for the aggregation capability of all elements such as photovoltaics, energy storage, and diverse loads.