157 results about "Model representation" patented technology

Methods and systems enabling multimodal inputs for interacting with a live digital object are provided. The system receives the live digital object which includes a digital artifact extracted from a digital model file through a model representation. The system receives a user's security level and determines the user's access permission and modification permission to access and modify the digital artifact. The system accesses a multimodal interface configured to receive a conversational input and a spatial input, outputs the digital artifact to the multimodal interface based on the access permission, receives from the multimodal interface a conversational input and a spatial input from the user, and generates a modified digital artifact from the digital artifact via the digital model representation, based on the modification permission and on the conversational or spatial input. The multimodal interface may include conventional interfaces (GUIs / APIs), conversational interfaces (text / voice), and spatial computing interfaces (VR / AR / MR / gestural).

Techniques are described for automatically generating visual model representations of buildings based at least in part on captured external imagery of the buildings, and using the generated building visual models to generate and present new images and optionally in additional manners, such as to improve navigation of a building and / or its surroundings. The described techniques may include acquiring building data from a plurality of exterior acquisition locations at multiple heights and view angles of an exterior of a building (e.g., using a flying drone and / or other flying device that captures the data), generating visual model representation(s) of the building (e.g., a 3D Gaussian Splat model), and using the generated visual model representation(s) to generate and present a new image with a view of the building exterior from a particular pose along with associated user-manipulatable controls.

The invention discloses a neural network image-text analysis and cross-framework code generation method and system, and the method comprises the steps: analyzing a neural network architecture picture through a visual large model, and extracting a layer type, a connection relation and a topological structure feature; performing semantic understanding on text description by combining a large language model, and extracting layer parameters and configuration information in a standardized manner; utilizing a multi-modal alignment mechanism to fuse vision and text features, and generating unified model representation; and finally, directly generating an executable code supporting a mainstream framework based on a large language model and grammar check. According to the method, the limitation of traditional manual coding is broken through, end-to-end generation from a complex framework to a multi-framework code is achieved, the problems of cross-framework adaptation and semantic understanding are solved, the development efficiency of deep learning is improved, and the method is suitable for scientific research verification and industrial deployment scenes.

This application discloses a model training method, a power prediction method, and a device. A model includes at least a first model and a second model. The method includes: obtaining a dataset including historical power data and historical meteorological data in preset duration; determining a weight of a loss function of each model based on the dataset; constructing a target loss function, where the first model and the second model may be used to represent input data, and data represented by the first model is not completely the same as data represented by the second model; and finally training the models based on the target loss function. More complete information is obtained from different modal data and different time scales based on a currently available data source, to optimize the learning model and achieve higher prediction precision.

The invention relates to a large language model dynamic routing method and device based on context learning model characterization and a readable storage medium. Query is embedded and mapped to a language model input space by using a projection model, semantic alignment is realized, a representative evaluation set covering multi-dimensional capability is automatically screened from a benchmark question bank, and the representative evaluation set is used for evaluating the multi-dimensional capability. Performance characteristics of the model on an evaluation set are efficiently obtained at a time, and high-quality context model capability representation is formed; then real-time query embedding and context model capability representation are combined, a lightweight routing language model is used for supervised learning, so that the fine-grained model distinguishing capability is achieved, an increment embedding updating mechanism is designed, and when a new model is accessed or an old model is upgraded, cold start can be rapidly completed only through a very small number of fixed questions, so that the efficiency is improved. The calculation and maintenance cost is greatly reduced, and the accuracy, real-time performance and flexible expansibility of model routing are effectively improved.

Systems and methods for generating an editable design using an auto-regressive pre-trained large language model (LLM) are disclosed. The method includes: receiving a prompt to generate the editable design; sequentially generating a set of tokens of model representation data for the editable design, each token in the set of tokens defining an attribute of the editable design; for each token in the set of tokens, determining whether the token is a predicted special token associated with a design asset or a non-special token; upon determining that the token is a non-special token, providing the non-special token as an input to the LLM to generate a next token in the set of tokens; upon determining that the token is a predicted special token: replacing the predicted special token with a replacement special token associated with a design asset stored in a design asset library; and providing the replacement special token as the input to the LLM to generate the next token in the set of tokens.

The invention relates to a three-dimensional grid model processing method, computer equipment, a readable storage medium and a program product. The method comprises the steps of performing voxelization on a three-dimensional grid model of a target object to obtain a plurality of voxels corresponding to the target object in a preset reference space; determining the visibility value of the voxel according to the projection information of the voxel under the preset camera visual angle; determining a position relation label of the voxel relative to a reference point according to a communication relation between the voxel and the reference point in a preset reference space; screening out a target voxel from the voxels according to the visibility values and the position relation labels of the voxels; and determining model characterization parameters of the three-dimensional grid model according to the spatial information of the target voxel in the three-dimensional grid model. By adopting the method, the generation precision of the three-dimensional model can be improved.

The invention provides a map detection method and system based on gradient feature extraction and multi-stage fusion, relates to the field of deep learning and image processing, and aims at solving the problems that in the prior art, defects exist when the problems of a fine-grained structure, boundary blur, category imbalance and the like of complex map data are solved. According to the method, gradient information and edge details are extracted from a map through a pure gradient extraction module; a progressive attention fusion module and a double-layer attention mechanism are used for enhancing feature representation by emphasizing key boundary information; through a cross-scale fusion encoder module, multi-stage enhancement and fusion are performed on high-level semantic information and high-resolution edge detail features, so that the performance of the model is improved, and the capabilities of representing semantic information and capturing edge details of the model are remarkably enhanced. The problems in the prior art are solved, the model prediction capability and accuracy are improved, and various challenges in map detection tasks can be coped with.

The invention belongs to the technical field of federated learning, and more specifically relates to a non-data-driven quantum federated learning method based on single communication. The method comprises the following steps: a plurality of clients independently train classical models by depending on respective local data, only perform single communication after training is completed, and upload model representation information to a server; after receiving the model representation information uploaded by each client, the server uniformly fuses the information to construct an integrated teacher model; the server synthesizes a pseudo sample from the random noise by using a pseudo sample generator under the guidance of the integrated teacher model; and migrating the knowledge of the integrated teacher model to the quantum student model through soft label distillation by using the generated pseudo sample, and finally obtaining a unified quantum student model at a server side. The technical problems that in existing quantum federation learning, communication overhead is large, privacy protection and efficiency are difficult to consider, and quantum hardware resources are seriously limited are solved.

The invention discloses a causal decoupling method and device based on multi-scale noise and adversarial supervision, and the method comprises the steps: carrying out the simulation of the causal relationship of variables in a causal graph, so as to generate observation image data, and constructing a training set and a test set according to the observation image data, the corresponding causal label information and the causal graph; constructing a causal decoupling model, and performing adversarial supervision training under multi-scale noise by using the training set; and obtaining anti-fact intervention data by using the test set and the trained causal graph matrix and utilizing the trained observation data coding module and observation data decoding module. According to the method, an auto-encoder and causal acyclic constraints are fully combined, the discrimination module is trained under multi-scale noise, and high-quality adversarial supervision is performed, so that the model representation learning ability is improved, the representation understanding of the model on data with causal relationships is enhanced, the accuracy of implicit causal network prediction is improved, and the prediction efficiency is improved. And the causal decoupling accuracy is improved.

A method may obtain a failure propagation model, wherein the failure propagation model comprises: a model representation of a plurality of hardware components, a set of hardware failure probabilities; and a logic that maps signals to the hardware components. A method may receive an alert signal from the aircraft. A method may map the alert signal to the plurality of hardware components. A method may perform a root-cause diagnosis of the alert signal that has been mapped to the plurality of hardware components comprising: determining via the failure propagation model, one or more combinations of hardware failures associated with the alert signal; and determining a probability of occurrence associated with the combinations of hardware failures. A method may display a report that includes combinations of hardware failures associated with the alert signal and the probability of occurrence associated with the combinations of hardware failures.

The invention provides a performance optimization method and system for a closed cooling tower, and relates to the technical field of data processing.The method comprises the steps that input parameters describing the closed cooling tower are obtained; determining a cumulative distribution function of each input parameter; discretizing each input parameter to form a discretized simulation sample; performing computational fluid mechanics simulation on the closed cooling tower by utilizing the simulation samples, and determining system response of each simulation sample; performing principal component analysis on system response data through singular value decomposition; establishing an incidence relation between the input variable and the score matrix through Kriging interpolation, and forming a complete preliminary prediction model between the input variable and the system response; through a high-dimensional model representation technology, the high-order interaction effect of the preliminary prediction model is optimized, and a performance prediction model of the closed cooling tower is obtained; in order to improve the cooling efficiency and reduce the construction cost, the design parameters of the closed cooling tower are optimized through a butterfly optimization algorithm.

The invention discloses a visual language action model-oriented target injection type fine tuning method, which comprises the following steps of: firstly, constructing any existing visual language action model, introducing a condition image generation model, and generating a target image with consistent semantics and vision according to an initial observation image and a task target instruction; secondly, target image features are injected into observation input through zero-initialization convolution, parameters are gradually increased from zero, it is ensured that interference noise is not introduced in the initial stage of fine adjustment to destroy a pre-training strategy, and in the training process, along with gradual optimization of the parameters, target image feature information is gradually fused into model representation, and the target image feature information is obtained; therefore, the understanding ability and the execution performance of the task target are improved. According to the method, through a lightweight target image injection mechanism and an efficient fine adjustment process, the performance of the model on various reference tasks can be remarkably improved in few training rounds, and the problem that an existing visual language action model cannot systematically introduce target image guidance is effectively solved.

The invention provides a method for performing human-object interaction detection on an image based on visual language model representation. The method comprises the following steps: extracting visual features of a to-be-detected image by using a CLIP visual encoder, and extracting text features of the to-be-detected image by using a CLIP text encoder; performing representation learning on the visual features by using a representation learning module to obtain human-object interaction HOI features; and recognizing a bounding box and an interaction type of a person and an object related to interaction in the to-be-detected image according to the text features and the HOI features through an interaction detection module. According to the method, the generalization ability of a new interaction category is enhanced by utilizing priori knowledge of the pre-trained visual language model, knowledge migration from the pre-trained model is realized by exploring representation learning of the pre-trained image and the text, and the ability of the model in processing complex visual contents related to human-object interaction is enhanced. According to the method, knowledge migration from the pre-trained model is realized, and the capability of the model in processing complex visual contents related to human-object interaction is enhanced.

A computing device includes at least one processor, one or more non-transitory computer-readable storage media, a system for fine-tuning a large language model under low-bit weight-activation quantization. The computing device further comprises a graphics processing unit (GPU), a neural processing unit (NPU), or a tensor processing unit (TPU). The hardware interface module of the system is configured to load a low-bit model representation from the memory module and transmit the model representation to the GPU, NPU, or TPU for inference execution.

Methods and systems for a document review process are provided. The method includes receiving an input digital model representation comprising at least one externally-accessible model endpoint for generating a digital artifact. Then, generating a document splice comprising access to multiple document subunits, with at least one document subunit written in a natural language and comprising the digital artifact; the access to each document subunit is provided through an externally-accessible document endpoint for the subunit. Then, generating a document by combining the document subunits, and generating a view associated with the document, based on an user authorization result including selective access rights to the document subunits. The view comprises access to the digital model representation, the digital artifact, each document subunit, and the document. Finally, receiving a user input and updating, via one of the externally-accessible document endpoints, the document splice based on the user input.

The invention belongs to the field of animal and plant breeding prediction, and discloses a rapid and accurate genome prediction method based on a heritability model, which comprises the following steps: calculating the frequency and linkage imbalance score of a marked secondary allele; refining the SNP heritability model to obtain an optimal SNP heritability model; selecting an optimal layering number; acquiring layer genotype data formed by genotype data of all individuals of the to-be-predicted data set in corresponding layers, calculating a diagonal weight matrix of each layer of SNP estimated heritability assignment to the corresponding layer based on the optimal SNP heritability model, and calculating a genetic relationship matrix between individuals of the layer genotype data corresponding to each layer; and fitting the multi-random hybrid model to obtain the genome estimation breeding value of each phenotype of all individuals in the data set to be tested. The method is based on a linear model framework, the calculation complexity is low, the key steps can be subjected to parallel processing, and the calculation speed is high; according to the method, the character specific SNP heritability model is constructed, and the model representation of a complex character genetic structure is optimized, so that the prediction performance is improved.

According to the smart city 3D map building indoor and outdoor integrated visualization method, based on a Unity rendering engine and a Mapbox map tool, an indoor and outdoor integrated 3D visualization model representation mode divided into a graph layer and a detail layer is researched and developed, 3D visualization is rapidly achieved, a vivid roaming effect is obtained, an environment similar to the real world is formed, and the method is suitable for popularization and application. Therefore, the user can carry out the dynamic interaction operation of the 3D scene. When outdoor roaming is carried out, only terrain, ground feature and building structure models in a visual range are rendered, and the internal structure and details of a building do not need to be loaded. When roaming into a room, a user can see the internal structure, characteristics, indoor real objects and outdoor scenes of the building at the current view angle. The indoor and outdoor integrated 3D visualization not only can carry out dynamic and vivid visual representation on the building on a computer to meet user requirements to the greatest extent, but also promotes research and application of indoor and outdoor integrated positioning and navigation, and promotes indoor intelligent management of the building and development of a digital city.

This application discloses a method and system for multilingual enhancement based on a large language model using model ensemble. The system includes: a pre-trained multilingual translation model, a semantic representation mapping module, and a large language model. The multilingual translation model is used for multilingual semantic modeling and language generation, including a multilingual encoder module and a multilingual decoder module. The semantic representation mapping module is used to transform the latent space representations of different models into an interactive unified semantic space based on a cross-model representation mapping mechanism. The output of the multilingual encoder is mapped to the unified semantic representation space of the large language model, and the mapped semantics are input into the large language model to perform language-independent instruction understanding. The intermediate semantic representation output by the large language model is mapped and transformed to a cross-attention representation space, generating the final output text under the target language distribution. The system of this application outperforms existing technologies in terms of efficiency, stability, and generation quality in multilingual capability extension.

The application discloses a kind of based on double model representation's aspheric surface parameter fitting and surface shape deviation measurement method, comprising 1) collection aspheric surface optical element surface several points, processing obtains the three-dimensional data of each sampling point;2) establish initial optical path model and set its wavefront evaluation function;3) input aspheric element's design parameter in turn;4) by the coordinate data of measured XOY plane is fitted to obtain the coordinate of its center point, conversion to the coordinate system of with optical design software coaxial, carries out Zernike fitting and obtains Zernike coefficient;5) in aspheric surface optical path model, using Zernike coefficient in step 4 is represented to be measured aspheric surface, in the same optical design software, constructs aspheric surface using two kinds of representation mode;6) radius of curvature is set as variable, the rest aspheric surface parameters are unchanged, and optimization obtains actual radius of curvature;7) conic coefficient and high order coefficient are set as variable, and optimization obtains aspheric surface parameter and surface shape deviation.

The application discloses a kind of energy storage cluster diversification modeling characterization method, system, equipment and storage medium, the method includes: obtaining the real-time data of each type of energy storage unit in energy storage cluster;Based on the real-time data obtained, different types of energy storage units are modeled using a diversified modeling method, and a diversified model library considering various types of energy storage units is constructed;According to the modeling results of the energy storage unit, the behavior of the energy storage cluster is characterized, the results of different diversified models are output, and the performance of the energy storage cluster is comprehensively evaluated;Using the diversified model constructed, an optimization algorithm is used to optimize the scheduling of the energy storage cluster;Monitor the operating state of the energy storage cluster, and adjust and update the diversified model in real time according to the operating state data.The application can effectively characterize the performance characteristics of various types of energy storage equipment and optimize the overall scheduling efficiency of the energy storage cluster.

Disclosed is a homologous image retrieval method and system, including dividing a teacher network and a student network into multiple modules, respectively, and performing loss constraints and knowledge distillation on a shallow feature extraction module and a deep feature extraction module; using the pseudo labels output by the teacher network training on shallow features as the labels output by the student network after shallow feature learning, and calculating the soft classification cross entropy loss; performing similarity calculation on the encoding outputs of the teacher network and the student network of deep features, and adding the hard classification cross entropy loss of the teacher network and the student network to the total loss function to complete the heuristic network lightweight training. This application constrains the feature learning capabilities of models at different levels of depth through heuristic learning of the teacher network by the student network, and simultaneously considers the model representation capabilities at the label level and the feature level, so that the student network can achieve image representation capabilities that are not lower than those of the teacher network, thereby achieving model lightweighting.

The application provides an open vocabulary segmentation method and system for multi-modal model representation optimization, and belongs to the technical field of computer vision. Image data to be segmented is acquired; a pre-trained multi-modal model is used to process the acquired image to obtain a segmentation result. The application better optimizes visual-text representation in a multi-modal task, effectively aligns the same visual-text representation space, proposes a mask-sensitive loss to constrain the classification score and mask quality to be consistent in the parameter fine-tuning process, thereby giving the visual encoder local perception ability and improving the effect of the model in the fine-grained downstream task, introduces the original pre-training feature as a representation compensation to ensure the zero-shot ability of the pre-training visual-language model in the optimization process, and interacts the text representation and the visual representation, so that the text representation can be adaptively enhanced for different input images, and the alignment property of the visual-text in the open vocabulary segmentation can be effectively improved.

The invention discloses an analysis method and system for model training and extraction of case causes, and the method comprises the steps: carrying out the joint coding of a case document through fusion of Transform and a graph neural network (GNN), achieving the dual modeling of the semantic and structural relation of a case text sequence, introducing a knowledge graph to embed and enhance the model representation, and achieving the analysis of the case causes. The method has the advantages that by the aid of the multi-task learning framework, key elements can be identified, legal article quotation can be predicted and the like while action cause classification is completed by the model, the integral understanding depth is increased, the integral process automation degree is high, the accuracy and recall rate of action cause extraction can be greatly increased, and manual annotation dependence is reduced.

This application discloses a method and system for interpreting sparse interaction utility modeled by a black-box artificial intelligence model, relating to the field of machine learning technology. The method and system can automatically analyze the interaction distribution modeled by the model. Implementation of the method and system includes the steps of providing data that is essential for evaluation; predicting the data using a black-box model and obtaining the model's prediction results; modeling the interaction effects between sample input units based on the output of the black-box model, calculating the interaction intensity of the combinations formed between the input units, representing the black-box model as "and additive relationships" and "or additive relationships" between combinations of input units; and optimizing the "and additive relationships" and "or additive relationships" to make them more sparse. The advantage of the present invention is that it provides a quantification method for interpreting interactions modeled by a black-box artificial intelligence model, enabling a more sparse and concise interpretation of interactions compared to conventional studies.

Methods and systems for generating digital threads that connect models and software tools for an interconnected digital engineering and certification ecosystem are provided. In one embodiment, a process for generating a software-code-defined digital thread includes receiving a first model representation of a first engineering model, a second model representation of a second engineering model, and an intent input. Determining, using a machine learning (ML) model, one or more model endpoints in the first and the second model representations. Generating, using a script-generating ML model, a platform orchestration script that accomplishes the intent input by connecting the first model representation and the second model representation, using the one or more model endpoints. Finally, storing the platform orchestration script as the software-code-defined digital thread. The digital threading of digital engineering models, with and without AI assistance, leads to exponential increases in efficiency of designing, testing, certifying, manufacturing, and operating physical products, allowing physical products to be made at digital speeds.

The invention discloses a mechanical arm action generation method based on a self-supervised reconstruction task. And training a mechanical arm imitation learning strategy close to an expert strategy on the basis of the obtained expert motion trail for manually controlling the mechanical arm to complete the specified task. According to the mechanical arm imitation learning strategy, a de-noising diffusion probability model with U-Net as a backbone network is used, down-sampling and up-sampling coding is carried out on an action sequence after Gaussian noise is randomly added, a global condition vector is used in each coding layer to modulate features, the action increment is predicted, the control action of a mechanical arm is output, and the control action of the mechanical arm is obtained. And executing and completing the specified task. In the training process, a self-supervised reconstruction task is introduced, and the difference between the action based on middle-layer feature reconstruction, point cloud features and expert motion trails is compared, so that a denoising network is promoted to carry stronger observation semantics, the characterization ability of a model to a three-dimensional hidden space is enhanced, and the success rate and generalization ability of task execution of the mechanical arm are effectively improved.

The present disclosure relates to computer-aided dental restoration system that is configured to estimate crown pose and representing virtual crown in 3D model. The system obtains 3D model of dentition of the patient. The system segments 3D model to obtain segmented tooth data and generates encoded 3D model representation suitable for processing by trained neural network. The method for generating encoded 3D model representation comprises subsampling point cloud representation based on segmented tooth data, determining surface normal representation, retrieving dental notation of restorative site, encoding each of plurality of points with corresponding surface normal representation and with value relative to restorative site to produce encoded 3D model representation, inputting encoded 3D model representation into trained neural network, and producing output, using trained neural network, wherein output includes prediction of translation of each point of encoded 3D model representation into positions corresponding to crown pose.