75 results about "Si model" patented technology

A model deployment tuning system (MDTS) receives a trained ML model, specified constraints, and model evaluation data and applies a plurality of model utilization techniques to the trained ML model to produce a plurality of useable model versions of the trained ML model. The MDTS executes each of the plurality of useable model versions of the trained ML models on a plurality of different compute instance types using the model evaluation data to produce model evaluation results for a plurality of different combinations. The MDTS filters the model evaluation results based on the specified constraints to indicate one or more of the different combinations satisfying the specified constraints. The MDTS deploys one of the plurality of useable model versions of the trained ML model to a compute instance types according to a selected combination satisfying the specified constraints.

The invention discloses an interactive multi-model aircraft filtering algorithm and system driven by reinforcement learning, and the algorithm comprises the steps: firstly constructing a multi-model state prediction set composed of a Singer model, a current statistical model and a Jerk model, and presetting a state transition probability matrix according to the state evolution correlation between the models, so as to define the interaction relation between the models; in the filtering stage, state estimation preliminary fusion of the three maneuvering models is achieved according to a traditional interactive multi-model algorithm, the fusion proportion of each model is adjusted in real time in combination with a weight correction vector output by a reinforcement learning strategy function, and strategy iteration optimization is conducted through a reward function containing an error improvement item, a matching item and a balance item. The method can significantly improve the filtering precision and fusion stability of the aircraft maneuvering target.

A system enables agile model development to speed up innovation by data scientists. Model training and deployment are coordinated and standardized to reduce redundancy. Data is obtained for feature generation and reformatted and de-sensitized for storage. The features are stored in locations available to all models and training modules of a system so data does not need to be adjusted for new models. To generate a machine learning model, the system establishes a cohort for evaluation by the model. A model template and features for use by the model are identified. The selected template and features are used for experimentation and evaluation. Model training artifacts, such as model weights are subsequently recorded in a model store and the model scripts and settings can then be registered in a centralized database where it can be accessed for execution.

The invention discloses a sample automatic collection and model iteration method for electrical equipment defect identification, belongs to the technical field of electrical equipment defect identification, and aims to solve the problems that a difficult sample discovery and collection mechanism is missing, model short plates are difficult to supplement and a model performance evaluation and feedback mechanism is imperfect. The method comprises the following steps: collecting multi-source power equipment inspection data, carrying out standardization processing, automatically discovering and collecting difficult samples, identifying the difficult samples from preprocessed samples based on a preset rule, establishing a difficult sample library, carrying out model iteration training, constructing a training set based on the difficult sample library and a basic sample library, and carrying out model training by adopting a weighted sampling strategy. Generating an iterative model; according to the method, key samples for model optimization are accurately screened through a difficult sample automatic discovery mechanism, storage and computing resources are prevented from being occupied by invalid samples without defects, with simple backgrounds and the like, and limited sample resources serve for model performance improvement in a centralized manner.

One or more embodiments of the invention provide a three-dimensional model generation method, apparatus and device, a medium and a program product. The method is applied to CAD software and comprises the following steps: acquiring a plane image and a model thickness of a modeling object; generating a plane grid for the modeling object according to the plane image; generating a front grid, a back grid and a side grid for the modeling object according to the plane grid and the model thickness; and combining the front grid, the back grid and the side grid into a three-dimensional grid, and mapping the texture of the plane image to the three-dimensional grid to obtain a three-dimensional model of the modeling object.

The invention discloses a MaaS platform construction method and system based on evaluation-driven closed-loop optimization, and the method comprises the steps: enabling a model square to serve as a unified entrance of a model, and providing rich model resources for a user to select; the computing platform provides model development, training and deployment for a user; evaluating the performance of the model; through feedback, close cooperation among the model square, the calculation platform and the model evaluation is realized, and an automatic optimization closed loop is formed. According to the closed-loop optimization mechanism based on evaluation driving, close cooperation among the model square, the calculation platform and the model evaluation module is achieved through feedback of the model evaluation module, an automatic optimization closed loop is formed, and the problem that an existing platform cannot achieve the automatic process from model selection, evaluation to retraining is solved.

The invention provides a test bed deformation field reconstruction and safety early warning method and system based on physical model driving data fusion. According to the invention, an intelligent closed-loop system composed of a sensing layer, a data layer, a model layer and a decision-making layer is constructed. The sensing layer is constructed through a deep bedrock reference network and a sparse multi-element sensor network; the data layer is responsible for synchronization, preprocessing and fusion of multi-source heterogeneous data; the model layer takes a parameterized physical model (such as a finite element model) as a core and performs bidirectional feedback with the data layer; and the decision-making layer carries out diagnosis, early warning and operation and maintenance decision-making based on model output. According to the system, high-precision reconstruction of a deformation field and intelligent operation and maintenance of a safety state are finally realized through an iterative process of data-driven model correction and model-guided data fusion.

The application discloses a kind of training after sparse benchmark evaluation method and device.The method constructs a precise evaluation system by five indexes: sparse distribution mode, reconstruction technology, model architecture accuracy, size robustness and different task accuracy. After the original model is sparsified using the standardized sparse paradigm, the sparsification performance of the model is evaluated using these indexes. This method can compare different post-training sparse models horizontally, comprehensively evaluate the performance of the model from both the algorithm and the model, and distinguish the impact of sparsification and reconstruction on performance, as well as the impact of architecture, size or task on sparsification performance. Using the present application, not only the model compression efficiency is improved, but also an important reference is provided for model selection and optimization.

The application provides a bearing multi-scale lightweight RUL prediction method and device based on a dynamic sparse space-time graph, and belongs to the field of residual life prediction.The method comprises the following steps: adopting an adaptive multi-scale identifier to extract the potential periodicity of a time sequence, and creating a multi-scale representation robust to noise; using a ProbSparse attention mechanism to construct a dynamic and sparse connection ST graph for each scale, and updating the weight of an edge by using a decay matrix, and realizing multi-hop feature propagation by combining a Mixhop GCN, which overcomes the limitation of isolated modeling of traditional time networks and graph networks, and realizes integration of high-precision ST features; a lightweight method is proposed, which adopts two pruning types, one for full connection pruning and the other for hierarchical propagation pruning; the method solves the problems of ignoring key cross-time sensor correlation, the influence of noise on prediction and model complexity, and improves the accuracy and efficiency of RUL prediction.

A computer-implemented method for model building with explainability is provided. The method includes receiving, by a hardware processor, a first metric and a second metric of minimum model performance. The first metric relates to data modeling quality and the second metric relates to model to business rule correlations. The method further includes performing, by the hardware processor, auto Artificial Intelligence model generation responsive to training data and a combination of the first and the second metrics of minimum model performance to obtain a model that is trained and meets model prediction accuracy and model prediction explainability requirements represented by the combination of the first and the second metrics of minimum model performance.

The application relates to the technical field of federated learning, in particular to a personalized federated learning method and device for incremental data, which can solve the problem of the deficiency of traditional federated learning in processing incremental data and personalization to a certain extent. The method comprises the following steps: introducing a weight decay regularization term in the model training process, applying a regularization constraint to the model parameters to prevent model overfitting and improve the generalization ability of the model; fusing the idea of meta-learning and using an inner-outer loop mechanism to train the model; in the outer loop, the personalized algorithm is iterated, so that the model can adapt to new incremental data more quickly, improve the training efficiency and model performance, improve the real-time performance and adaptability of the model, and through the iteration of the personalized algorithm, the model of each client can be optimized according to the characteristics of the client, thereby improving the adaptability and accuracy of the model on different clients.

This invention discloses a load-balanced automatic parallel modeling method, device, and storage medium. First, it analyzes the key factors affecting the execution performance of operators and models (operator in-degree, tensor shape, and tensor data type), and proposes a method for constructing a performance evaluation model based on operator characteristics to assess the computational, communication, and overall costs of operators, as well as the training performance costs of the model. Then, with the goal of load balancing across devices, a layer-by-layer partitioning method based on topological sorting is used to quickly divide the neural network model into multiple sub-models with balanced overall costs, achieving coarse-grained partitioning. Finally, based on the model's training performance evaluation model, and according to the communication characteristics between operators, a fine-grained model partitioning and scheduling method based on communication optimization is used to fine-grainedly adjust the coarse-grained sub-models, reducing the amount of cross-device communication tensor transmission to achieve optimal global model scheduling.

The invention provides an online high-precision map construction method and a model construction method thereof, and the model construction method comprises the steps: sequentially constructing a backbone network, a view conversion network, a hierarchical query embedded network and a model optimization network based on a MapTr model architecture, and obtaining an initial model; training the initial model to obtain a map construction model; and carrying out model integration on the plurality of map construction models to obtain an online high-precision map construction model. According to the method, efficient and accurate high-precision map construction can be realized through a stronger backbone network, an effective view conversion and data enhancement technology and a model integration strategy. The method has excellent performance in the aspect of multi-category semantic information extraction, and is suitable for real-time map construction tasks in an automatic driving system.

The invention discloses a model training integrity verification method based on a trusted execution environment. The method comprises the following steps: acquiring a model offline training data set; constructing an operator training set to train an operator operation time prediction model; obtaining the overall expected time of each training round of the corresponding to-be-trained model; quantifying model parameters after single-round training of each model in the GPU equipment to obtain a parameter group of the corresponding model; corresponding model training is carried out in the target GPU, TEE training and timing are started according to the pseudo-random seeds, and quantitative check points and complete check points are generated; and repeatedly verifying the integrity of model training in the TEE according to the quantitative check points and the complete check points, and regarding the model with the integrity as a qualified model. According to the method, abnormity can be found in time in the training process, waste of computing resources and reduction of model performance are avoided, the storage burden can be effectively relieved, robustness and expandability are improved, and correctness and safety of training are guaranteed.

A system and methods for OCD metrology are provided including receiving reference parameters, receiving multiple sets of measured scatterometric data, and receiving an optical model designed to generate one or more sets of model scatterometric data according to a set of pattern parameters, and training a machine learning model by applying, during the training, target features including the reference parameters, and by applying input features including the sets of measured scatterometric data and the sets of model scatterometric data, such that the trained machine learning model estimates new wafer pattern parameters from subsequently sets of measured scatterometric data.

The application provides a switch cabinet internal temperature mapping transient inversion method, comprising the following steps: step one, obtaining switch cabinet internal and external temperature data streams to obtain a model training group and a model verification group; step two, setting temperature calculation tracing time periods, time period division segment numbers, initial weights and pre-designed calculation accuracies for each data stream in the model training group and the model verification group formed in step one to obtain training samples; step three, inputting the four groups of variables in the training samples obtained in step two into a regression algorithm for training to obtain a convergence formula; step four, when the algorithm is completely converged and the accuracy meets the requirements, using the obtained convergence formula to test the non-training sample accuracy interval, if the requirements are met, directly giving a temperature full value formula at the current time, and if the requirements are not met, performing feedback adjustment and resetting the length of the tracing time period and the segment number. The application can eliminate the error and time lag of the current time inversion technology and realizes real-time temperature inversion.

Result quality metrics of a set of machine learning tasks conducted on various record groups using a plurality of machine learning models are obtained. Based on applying an algorithm to the record groups, respective sets of intermediary results corresponding to records of the groups are obtained. A meta-model for predicting result quality metrics for respective record-group-and-model combinations is trained using a training data set which includes statistical features obtained from the intermediary results. The trained meta-model is stored.

The invention discloses a live-action three-dimensional geographic entity construction system fusing multi-source data, and relates to the technical field of computer vision and geographic information, the live-action three-dimensional geographic entity construction system comprises a data acquisition module, a data processing module and a model output module, and is characterized in that the data processing module comprises a multi-scale consistency constraint engine; and the engine is connected with the data acquisition module and the model output module, and is used for establishing an incidence relation among different detail level models and controlling a model generation process. According to the live-action three-dimensional geographic entity construction system fusing the multi-source data, by establishing a multi-scale consistency constraint mechanism and a hierarchical semantic topology relationship network, the problems of geometric cracks and texture jumping of the same geographic entity among different detail level models are effectively solved. The system can ensure the visual coherence and consistency of multi-scale expression while keeping the geometric accuracy of the model, so that the model can realize smooth transition among different scaling levels.

The invention discloses a two-dimensional multidirectional crustal stress calculation and application method suitable for a quasi-static pushing and covering device, and the method specifically comprises the steps: firstly, arranging the quasi-static pushing and covering device, installing a horizontal actuator on one side of a model box, and supporting the horizontal actuator on an assembly type reaction wall, so as to achieve the quasi-static pushing and covering action of the model box; the stress application is formed by superposing a horizontal tectonic stress and a vertical equivalent self-weight stress; stress cables acting on cable holes in the two sides of the model box provide a stress input source for horizontal tectonic stress, and two sets of vertical actuators acting on the upper portion of the model box apply loads to vertical equivalent self-weight stress; and calculating the ground stress applied by the transverse stress cable and the vertical actuator by combining a test site and the strength of a model similar material. The method is high in practicability, the ground stress application control parameters selected in the formula are visual and effective, and the stress field mapping relation from the engineering field area to the model test is clearer. And the method has wide practicability in the field of tunnel model tests in a high ground stress environment.

The application relates to the technical field of large language models, and discloses a large model optimization method and device, equipment and a storage medium. The method comprises the following steps: acquiring a sampling corpus set; performing data compression on corpus samples in the sampling corpus set based on data similarity to obtain a data set; and training a large model based on the data set to obtain an optimized large language model. According to the application, the redundant data in the training data of the large language model is removed through data compression according to the data similarity, the data quality is effectively improved, and then the model is trained through the compressed data, so that the model training efficiency and the model performance are effectively improved.

The application provides a pyramid knowledge distillation framework model compression limit analysis method, comprising the following steps: constructing N groups of online deep mutual learning models in a pyramid structure; performing online deep mutual learning on each group of online deep mutual learning models, and recording the parameter quantity and model performance of two models in each group of online deep mutual learning models; wherein, starting from the second group of online deep mutual learning models from bottom to top, while performing online deep mutual learning, the previous group of online deep mutual learning models is accepted for offline knowledge distillation; the potential representation of all models from the first group to the N-1th group is extracted and sent to an adapter to generate teacher importance weight soft labels; the Nth group of online deep mutual learning models is subjected to online deep mutual learning, and the parameter quantity and model performance of the Nth group of models are recorded; and the balance point of the model compression ratio and accuracy is analyzed according to the parameter quantity and model performance of two models in each group of online deep mutual learning models and the parameter quantity and model performance of the Nth group of models.

The invention discloses an injection molding machine action and modeling picture synchronization method and system, and the method comprises the steps: building a model library, drawing a high-precision model and a common model of an injection molding machine part in advance based on a preset proportion, importing the models into the model library, rubbing the structure of the high-precision model to a picture sticker, and carrying out the modeling of the high-precision model. The model is attached to the common model; configuring a physical engine, acquiring an action signal of a core part of the injection molding machine through the physical engine, and outputting the action signal to the model library; and receiving an action signal, calling the model in the model library, and realizing synchronous movement of the physical equipment and the model. According to the method, through dynamic modeling parameters, the workload of manual model making is reduced, dynamic animation adjustment parameters are achieved, the virtual world speed is matched according to the physical world speed, the time for personnel to reconstruct animations is shortened, a system configures a visual interface, data are visually adjusted on the interface, and meanwhile the deviation between the virtual world and the real world is calibrated through a sensor.

The invention discloses a multi-modal task unified modeling method and processing method based on flow matching, and relates to the technical field of task unified frameworks. For all visual tasks of an image mode, after a token and model representation of a sample of each target visual task are obtained, a continuous evolution process from the token to the model representation is simulated by constructing an intermediate state token, and a task unified model is obtained through training by adopting a flow matching method. The method comprises the following steps: performing word segmentation on to-be-processed information to obtain an initialized token, gradually evolving a target model representation of any target task from the initialized token through m times of iteration according to a task unification model, and inputting the target model representation into a decoder corresponding to the target task for decoding to obtain a target task result. Due to the fact that the representation forms of the model representations of the different visual tasks are the same, the task unified model only needs to determine the model representations and then decodes the model representations through decoders of the different visual tasks to obtain the task results, and therefore unified modeling and processing of all the visual tasks are achieved.

The application discloses a bridge risk identification method and system fusing integrated learning and an attention mechanism, and the method comprises the following steps: processing an i-th input feature in an input data set to obtain a reconstructed feature vector; constructing XGBoost and RF-MLP models to perform deep feature extraction, and acquiring fusion features by using an attention mechanism; inputting the fusion features into a back-end multi-task perception decoder, and using parallel decoding branches to decouple and identify the fusion features, wherein a first full connection layer and a first Softmax activation function are used as a risk type decoding branch to output a risk type probability distribution; a second full connection layer and a second Softmax activation function are used as a position decoding branch to output a position probability distribution; a third full connection layer and a Sigmoid activation function are used as a damage degree decoding branch in cooperation with linear mapping to output a damage degree quantitative index; and the application has the advantage of high identification precision.

The invention discloses a model training hotspot identification method combining cross-thread dependence and multi-critical path analysis, and the method comprises the steps: 1, collecting an event execution track in a model training process, and carrying out the analysis and standardization of collected original performance data into a unified format; 2, aggregating the original events through hierarchical modeling, and constructing edge connection based on a dependency relationship to form a calculation dependency graph; step 3, identifying a decisive event sequence and a critical path in a model training process based on the constructed calculation dependency graph; and 4, through overlap analysis and statistical processing, extracting hot events with actual optimization values from the key paths, and generating a sorting list. The invention further discloses a system for implementing the method, and the system has a wide application prospect.

The invention discloses a class increment classification method based on union search set graph modeling and fusion space, which comprises the following steps: configuring an independent adaptation module for each increment stage, extracting class characteristics of each stage by freezing a pre-training visual backbone network, and constructing a stage class prototype set. And calculating cross-stage similarity among global category prototypes, establishing a category confusion relationship, and identifying confusion connected branches by using a union-check set structure. And for each connected branch, fusing the features of the adaptive module in the corresponding increment stage and optimizing the fusion weight to form a fusion feature space. In the inference stage, the preliminary classification result is used for judging whether the candidate category belongs to a confusion connected branch or not, if yes, fine judgment is carried out in the fusion feature space to output a final result, and if not, the preliminary classification result is directly output. The method effectively solves the identification problem of cross-stage easily-confused categories, improves the classification precision, and maintains the low incremental training cost and the anti-forgetting capability of the model at the same time.

The present application relates to the field of intelligent prediction, and particularly relates to a weak adhesion hydrogel adhesion performance prediction method and system, comprising the following steps: S1: obtaining hydrogel material parameters and interface state information and preprocessing; S2: identifying state labels through a classification model to form an extended data set with state labels; S3: constructing a segmented physical model system driven by the interface state to obtain state-adaptive physical prediction results; S4: embedding the state-adaptive physical prediction results into a physically-constrained neural network structure to construct a prediction model and obtain model prediction results; S5: using a multi-model collaborative fusion mechanism, comprehensively considering the state-driven physical model output and the model prediction results, and obtaining the final prediction value through a weighted fusion manner; S6: taking the final prediction results and model internal parameters as inputs, performing feature contribution degree analysis, identifying key influencing factors, and outputting parameter optimization suggestions. The present application realizes high-precision and high-generalization-capability prediction.

A sample data acquisition and model training method, medium, device and program product, the method comprising: acquiring a thinking chain template, the thinking chain template being used for defining a target information point, the target information point being used for describing information extracted from an input question for thinking chain reasoning; in response to the acquired current input question, generating question solving step information corresponding to the current input question based on the thinking chain template; the problem solving step information is used for describing a reasoning step adopted when thinking chain reasoning is carried out on the current input problem based on the information described by the target information point; generating sample data based on the current input question and the question solving step information; the sample data is used for training a first question generation model for reasoning based on a thinking chain.

The embodiment of the invention discloses a prediction result generation method and device, equipment, a medium and a program product. A specific embodiment of the method comprises the steps of obtaining target input information and a pre-training language model set; a model output information set for the target input information is generated by utilizing the pre-training language model set, and each piece of model output information comprises a model prediction result group and a model prediction probability group; according to the obtained model prediction probability set, performing prediction result fusion on each model prediction result in the model prediction result set to generate a fusion prediction result set and a corresponding fusion prediction probability set; and according to the fusion prediction probability set, generating at least one actual prediction result. The implementation mode is related to artificial intelligence, and at least one actual prediction result for the target input information can be accurately and efficiently generated by utilizing the pre-training language model set.