38 results about "Model tuning" patented technology

Database mining, analysis and optimization techniques in conjunction with the model-based functional coverage analysis are used to turn raw verification and coverage data into design intelligence (DI) and verification intelligence (VI). The required data and attributes are automatically extracted from verification, simulation and coverage analysis databases. Design finite state machine extraction, design functional event extraction, and automatic coverage model generation and optimization techniques are applied to the design HDL description. Coverage model tuning and optimization directives, as well as test spec tuning and optimization directives are generated based on the analysis and mining of various verification, simulation, and coverage databases. An integrated web-based interface portlet is used for access, analysis and management of the resulting databases, generated reports and verification directives. Dissemination rules are used to automatically generate and distribute analysis reports and verification directives to engineers at wired or wireless interface devices via Internet or Intranet.

A methodology and system for performing radio network coverage surveys of multiple site candidates concurrently using one reserved channel based on reverse path signal strength measurements of a signal from a test mobile station transmitter (100) to a plurality of prospective base station receiver sites (200, 200′, . . . ) over a geographical area is presented. The instantaneous position, time-stamp and power of the Transmitter (101) is recorded and stored as it moves along a test route(s). Stationary Test Receiver(s) (201) tuned to the selected channel at each prospective site location (200, 200′, . . . ) save received instantaneous signal strength and, it's time-stamp so they can matched up to the position of the Test Transmitter (101) and used to ascertain received signal strength at each site. Results can be useful for ascertaining optimum site locations, antenna configurations, propagation model tuning, frequency planning, pilot planning and extraction of signal correlation statistics.

The invention discloses a multi-classifier training method and a classifying method based on non-deterministic active learning. The method comprises the following steps: 1) selecting or initializing a multi-classifier and calculating the overall information quantity info of each sample in an unlabeled sample set by the utilization of the multi-classifier, wherein the overall information quantity is the sum of model change information quantity and model tuning information quantity; 2) clustering the unlabeled sample set to obtain J subclasses; 3) selecting a plurality of unlabelled samples with minimum overall information quantity Info from each subclass; selecting, labeling and adding K samples from the selected sample into a labeled sample set L; 4) training the multi-classifier again through taking the updated labeled set L as training data; 5) iteratively executing the steps 1)-4) to set the number of times and then classifying the unlabelled set by the utilization of the finally-obtained multi-classifier. According to the multi-classifier training method and the classifying method, the comprehensive evaluation on the sample information quantity is realized, so that the efficient and intelligent classifier is obtained.

A method and apparatus for synchronizing signals. For memory devices, such as SDRAMs, implementing a synchronization device to synchronize one signal, such as an external clock signal with a second signal, such as a data signal, tuning elements may be provided at various points in the signal path of the synchronization device. The tuning elements are designed to be identical, such that a single design may be used to a signal mismatch that is produced in either direction, using a single design. The tuning elements may be implemented to provide uniformity in the access time through a range of conditions, such as drain voltages and temperatures.

The invention discloses a big data acquisition and analysis system based on intelligent image recognition, and an application method, and relates to the technical field of intelligent data analysis. The system comprises an intelligent cloud server; the intelligent cloud server comprises a computing server and a storage server. An image recognition system composed of a data reading module, a videostream data processing module, an AI image recognition module, a data storage module and a model tuning module is carried in the computing server, and a video stream storage database, a video stream management module and a data center database which are in interactive connection are arranged in the storage server. According to the system, non-private real digital behaviors of consumers are restored and higher commercial value is generated; the method is advantaged in that problems of delay, omission, slow speed, large error and high cost are not generated in the process, a consumer real-time digital behavior analysis bottleneck is solved, business analysis is made to be closer to the reality, more valuable analysis results are brought to a brand party, and a brand global optimization consumption path is guided.

The embodiment of the invention provides a cross-protocol communication method and device from Wi-Fi to BLE. The method comprises the following steps: splitting and encoding each bit of a BLE symbol to generate an initial phase sequence; adjusting the initial phase sequence according to a decoding probability model to obtain a phase sequence with the maximum probability that the BLE symbol is correctly decoded, the decoding probability model being determined according to the probability that an error occurs when a sampling point falls into a specified phase; optimizing the phase sequence according to a phase optimization model to obtain an optimal phase sequence of the BLE symbol, the phase optimization model being an objective function constructed by taking minimum simulation error of Wi-Fi as an objective; and simulating a target waveform by using a Wi-Fi signal to perform cross-protocol communication from Wi-Fi to BLE, the target waveform being a time domain waveform corresponding to the optimal phase sequence. According to the embodiment of the invention, cross-protocol communication from Wi-Fi to BLE can be realized, the Wi-Fi simulation error is small, and the decoding probability of the symbol at the receiving end is high.

The invention provides a data processing method of continuous wave test for propagation model tuning, which comprises: step 1, calculating the distance between adjacent sampling sites in an original sampling site aggregate; step 2, deleting a part of the sampling sites according to the distance between the adjacent sampling sites in the original sampling site aggregate, to obtain a first sampling site aggregate, the distance between the adjacent sampling sites in the first sampling site aggregate being larger than or equal to the preset threshold; step 3, deleting a part of the sampling sites from the first sampling site aggregate to obtain a second sampling site aggregate, in the second sampling site aggregate, the number of actual sampling sites within the intrinsic length being larger than the preset threshold. The invention provides also a data processing device of continuous wave test for propagation model tuning, which comprises: a computation module, a first data processing module, a second data processing module and an output module. The invention can realize the judgment of sampling site number within the intrinsic length, which complies with the Lee's criterion, in the test with failure to correctly control the motion speed of the test terminal, and ensure the accuracy of the sampling sites.

The present application discloses a method, system, and computer system for building a model associated with a dataset. The method includes receiving a data set, the dataset comprising a plurality of keys and a plurality of key-value relationships, determining a plurality of models to build based at least in part on the dataset, wherein determining the plurality of models to build comprises using the dataset format information to identify the plurality of models, building the plurality of models, and optimizing at least one of the plurality of models.

A multi-horizon predictor system that predicts a future parameter value for multiple horizons based on time-series data of the parameter, external data, and machine-learning. For a given time horizon, a time series data splitter splits the time into training data corresponding to a training time period, and a validation time period corresponding to a validation time period between the training time period and the given horizon. A model tuner tunes the prediction model of the given horizon fitting an initial prediction model to the parameter using the training data thereby using machine learning. The model tuner also tunes the initial prediction model by adjusting an effect of the external data on the prediction to generate a final prediction model for the given horizon using the validation data. A multi-horizon predictor causes the time series data splitter and the model tuner to operate for each of multiple horizons.

A methodology and system for performing radio network coverage surveys of multiple site candidates concurrently using one reserved channel based on reverse path signal strength measurements of a signal from a test mobile station transmitter (100) to a plurality of prospective base station receiver sites (200, 200′, . . . ) over a geographical area is presented. The instantaneous position, time-stamp and power of the Transmitter (101) is recorded and stored as it moves along a test route(s). Stationary Test Receiver(s) (201) tuned to the selected channel at each prospective site location (200, 200′, . . . ) save received instantaneous signal strength and, it's time-stamp so they can matched up to the position of the Test Transmitter (101) and used to ascertain received signal strength at each site. Results can be useful for ascertaining optimum site locations, antenna configurations, propagation model tuning, frequency planning, pilot planning and extraction of signal correlation statistics.

The invention discloses a turboshaft engine reverse modeling method based on test data, and belongs to the field of aero-engines. The implementation method comprises the following steps: performing state conversion on test data of a plurality of test points of the turboshaft engine to obtain a test point data set in a target test state; screening data in the test point data set to obtain a screened test point data set; according to the plurality of actual measurement parameters, sequentially performing parameter trial and error on the plurality of trial and error iteration parameters to obtaina preliminary turboshaft engine model; adjusting the characteristics of universal parts, and calibrating the preliminary turboshaft engine model in combination with throttling test data; and performing Reynolds number and gap correction on the calibrated turboshaft engine model to obtain a target turboshaft engine model. According to the method, tTest data is directly utilized, dependence on parttest characteristics is not needed, model construction can be completed in a short time, and high precision is kept. The method has the advantages of being easy to operate and high in applicability.

Systems and methods for an accelerated tuning of hyperparameters of a model supported with prior learnings data include assessing subject models associated with a plurality of distinct sources of transfer tuning data, wherein the assessing includes implementing of: [1] a model relatedness assessment for each of a plurality of distinct pairwise subject models, and [2] a model coherence assessment for each of the plurality of distinct pairwise subject models; constructing a plurality of distinct prior mixture models based on the relatedness metric value and the coherence metric value for each of the plurality of distinct pairwise subject models, identifying sources of transfer tuning data based on identifying a distinct prior mixture model having a satisfactory model evidence fraction; and accelerating a tuning of hyperparameters of the target model based on transfer tuning data associated with the distinct prior mixture model having the satisfactory model evidence fraction.

The invention discloses a medium-voltage distribution network user electricity consumption abnormity diagnosis method based on machine learning. The method comprises the steps of generating a potential electricity consumption abnormity user set based on acquired user name data; on the basis of the potential abnormal power utilization user set, acquiring forward active power, voltage and three-phase current data of the industry to which the potential abnormal power utilization user set belongs and near two cycles; carrying out missing value preprocessing on the obtained data; calculating and adding 5 eigenvalues into historical power consumption data anomaly four-level labels of all the users to form a sample set; dividing the sample set, training an artificial intelligence model, and testing a model effect and model tuning; carrying out model training and evaluation by adopting a random forest in a machine learning integration algorithm; and carrying out batch marking processing by using the trained model. The method is simple in calculation, and can help an operator to find and adjust an abnormal line loss line in time.

The invention provides a method for predictive maintenance of electric power production equipment based on recurrent neural network (RNN). A data modeling platform based on Spark is established to support parallel data preprocessing and RNN modeling; Predefined RNN structure supports for multiple data input interfaces (HDFS, NFS, S3); Data pre-processing can standardize the historical data of several main function modules such as coal burner, pump system and fan according to the user-defined data cleaning logic; Iterative row modeling and model tuning, modeling process includes extracting datafeatures in the RNN way and combined with user-tagged fault state to model the diagnosis model, tuning process includes verifying the data set to detect the success rate of pre-judgement and the user-defined neural network correction strategy to reconstruct RNN. The above technical scheme provides an iterative modeling based on the time series data of a recurrent neural network (RNN) power production equipment, and provides fault prediction for a power production operator to perform predictive maintenance by identifying a fault occurrence mode.

The invention relates to an electricity selling company evaluation emotion classification method based on a joint attention mechanism, and belongs to the technical field of information. The method comprises the four steps of construction and division of an electricity selling company evaluation text data set, model training, model tuning and electricity selling company evaluation sentiment classification. The emotion evaluation classification result obtained by the method can reflect the emotional tendency of the power consumer to the power selling company to a certain extent, and can describe the customer satisfaction and favorable comment of the power selling company in the retail market in a certain sense; the classification result can be used as an important index for evaluating the service quality and credit rating of the electricity selling company, the electricity selling company can be promoted to continuously improve the service quality, optimize the electric power package structure and adjust the electric power price, and benign competition and healthy and stable operation of the electric power retail market are guaranteed.

A computing device includes a memory, a scheduling unit, and an acceleration unit, wherein the acceleration unit is used for executing each quantitative model; the memory stores instructions; the scheduling unit reads the instruction to perform the following steps: creating a plurality of configuration combinations for the deep learning model, each configuration combination specifying a value combination of a plurality of quantitative configuration parameters; based on each configuration combination, performing quantization operation on the deep learning model to obtain a plurality of models after quantization operation; sequentially deploying the plurality of quantized models to an acceleration unit, and receiving precision data corresponding to the plurality of quantized models from the acceleration unit; and on the basis of the respective precision data of the plurality of models after the quantification operation, obtaining an optimal model of which the precision loss meets a set condition. According to the embodiment of the invention, the neural network acceleration unit and the scheduling unit are matched with each other, so that the optimal model with small precision loss can be quickly obtained.

The invention relates to a method for recommending system data abstraction and automation characteristic engineering. For recommending system data for any scene, only keywords and specified processingfunctions in fields generated after the recommendation system data are adapted need to be understood, universal data processing and feature engineering codes can be used for completing feature generation, and the method comprises the two steps of data abstraction, schema configuration for standard abstract data and corresponding universal processing function development. According to the method,the development workload of engineers is reduced, so that the engineers have more sufficient time and energy to perform model tuning work.

An improved apparatus and method for detecting fraud is described using a stacked auto encoder with embedding to encode and decode a transaction to determine fraud. The technique includes model tuning software and transaction review software. The model tuning software views the transaction and tunes an artificial neural network model to minimize reconstruction loss. The transaction review software processes the transaction through the artificial neural network model, converting the transaction into a feature vector, encoding the feature vector into a compressed vector, decoding the compressed vector into a reconstructed vector, subtracting the reconstructed vector from the feature vector, and determining a fraud indication and reasoning based on a difference from the reconstructed vector from the feature vector.

The invention discloses a method for tuning and managing a deep learning model, which includes: in response to the number of pictures in a data set that needs to be predicted is greater than a set number threshold, the data set is divided into several sub-data sets on average, and each sub-data set is processed by a deep learning model. The data set is predicted to obtain the first prediction data; the judgment data of each sub-data set is obtained through manual prediction and judgment; the Kalman filter is performed on the first prediction data and judgment data of each sub-data set to obtain the first prediction data of each sub-data set Two prediction data; obtain the error rate of the deep learning model according to the first prediction data and the second prediction data; trigger the retraining of the deep learning model in response to the error rate exceeding the error rate threshold. The invention also discloses a device, equipment and medium. The principle of the invention is easy to understand, easy to operate and realize, and the accuracy is guaranteed, and the function of automatic model retraining and tuning is optimized.

The invention relates to a method and a system for generating and switching high and low modes of Maya. The method comprises the following steps: firstly, loading a model adjusting tool to a YHKT menuof Maya; secondly, setting an outline in Maya, and defining precision parameters of different precision models; then, storing the current three-dimensional view as different precision models according to the precision parameters defined in the outline by utilizing the model adjusting tool; and finally, selecting and calling different precision models for switching display through the model adjusting tool. According to the method, the three-dimensional view is stored as a plurality of models with different precisions, and artists can conveniently switch among the models with different precisions through mode selection, so that the working efficiency is greatly improved.

The invention relates to a method for recommending system data abstraction and automation characteristic engineering. For recommending system data for any scene, only keywords and specified processingfunctions in fields generated after the recommendation system data are adapted need to be understood, universal data processing and feature engineering codes can be used for completing feature generation, and the method comprises the two steps of data abstraction, schema configuration for standard abstract data and corresponding universal processing function development. According to the method,the development workload of engineers is reduced, so that the engineers have more sufficient time and energy to perform model tuning work.

The invention discloses a text classification model tuning hyper-parameter recommendation method and device and a storage medium, and the method comprises the steps: constructing a hyper-parameter set according to the hyper-parameter type of a text classification model; according to the category system and the classification performance index of the text classification model, obtaining a first group of data through calculation, wherein the first group of data comprises category system weight information and overall classification performance index weight information; according to the hyper-parameter set, training and testing the text classification model to obtain a second group of data, the second group of data comprising an overall classification performance result and a category classification performance result set; according to the first group of data, calculating the second group of data to obtain a third group of data, wherein the third group of data comprises an overall classification performance comprehensive result and a category classification performance comprehensive result; and sorting the third group of data to obtain a recommended hyper-parameter group. According to the invention, the efficiency of deep learning text classification model tuning can be improved; The method and device can be widely applied to the field of machine learning.

The fault warning method of the pitch motor temperature sensor of wind turbine based on the LightGBM algorithm includes the following steps: data collection and processing, collecting enough fan operation data before modeling, and performing preliminary processing; training the model, selecting a sensor with a pitch motor Perform model training on the faulty data and save it as a lightGBM model; parameter adjustment optimization, in order to make the algorithm achieve the optimal training effect, adjust the parameter variables of the model; run the model to get the early warning result, after the model is saved, in the new data prediction process , you need to load the saved model first, the data that needs to be predicted, judge the fault point by comparing the predicted value with the actual value, and output the early warning information; then analyze the early warning information, find the root cause of the failure, and give reasonable maintenance suggestions . The invention overcomes the deficiencies of the prior art, utilizes the distributed and high-efficiency characteristics of the LightGBM, and well solves the problems of difficult early warning modeling and low accuracy of pitch motor faults of wind turbines.

According to an aspect of the present invention, a method for controlling access to a plurality of application programs in an electronic device is disclosed. The method includes receiving a voice command from a speaker for accessing a target application among the plurality of applications, and verifying whether the voice command indicates authorized access to the application based on a speaker model of an authorized user. user. In this approach, each application is associated with a security level with a threshold. The method further includes updating the speaker model with the voice command when the voice command is verified to be indicative of the user, and adjusting at least one of the thresholds based on the updated speaker model.

The invention belongs to the technical field of train loading, and relates to a train loading 3D model front-end assembly and display method, which comprises the following steps: 1) loading a pre-generated loading 3D model, correcting the position of the loading 3D model, and adjusting the center of gravity of the loading 3D model to an original point; 2) obtaining an empty car model according to a model file prefabricated in advance, adjusting the size and specification of the empty car model, and matching the empty car model with the car loading 3D model in the step 1); 3) assembling the loading 3D model and the empty car model to obtain an integral model; and 4) mapping the overall model in the step 3) to obtain a visual final model. According to the invention, a loading scene can be completely restored, a real carriage can be simulated, and on-site inspection and loading condition review are facilitated.

In order to reduce oscillations in process variables of an injection molding process, an injection molding machine may be operatively connected to a model database that stores models of injection molding machines and molds. A tuning controller may set initial gain values of a variable-gain proportional-integral-derivative (PID) controller. To set the initial gains, the tuning controller may be configured to obtain, from the model database, a model for a first and second injection molding machines and a model for a mold. The tuning controller may analyze the models to determine a correlation between injection molding machine parameters and mold cycle performance for the mold. Accordingly, the tuning controller may apply the correlation to determine an initial gain value for a least one of the first, second, and third gains of the PID controller. The tuning controller may then set the initial gain values for the PID controller.