109 results about "Hyperparameter optimization" patented technology

The invention discloses a city-level intelligent traffic simulation system. With multi-source heterogeneous data as input, the vehicle queue length of each lane at an intersection can be predicted, the evolution of the vehicle queue length is further dynamically simulated, and the simulation system comprises a data collection module, a multidimensional database, an algorithm module, a traffic state prediction module, a traffic simulation derivation module and an interaction visualization module. In the data collection module, the collected data comprise traffic dynamic data and static data. The multidimensional database is used for receiving various traffic dynamic information in real time and storing road infrastructure configuration information. In the algorithm module, a generated confrontation network is used for data generation and state prediction on the vehicle queue length, and the processing process comprises data feature extraction, data dimension processing, confrontation network generation, and hyperparameter optimization. The vehicle queue length of each lane at the intersection can be predicted, and the evolution of the vehicle queue length is further dynamically simulated.

An improved system and method is provided for sparse Gaussian process regression using predictive measures. A Gaussian process regressor model may be construction by interleaving basis vector set selection and hyper-parameter optimization until the chosen predictive measure stabilizes. One of various LOO-CV based predictive measures may be used to find an optimal set of active basis vectors for building a sparse Gaussian process regression model by sequentially adding basis vectors selected using a chosen predictive measure. In a given iteration, a predictive measure is computed for each of the basis vectors in a candidate set of basis vectors and the basis vector with the best predictive measure is selected. The iterative addition of basis vectors may stop when predictive performance of the model degrades or no significant performance improvement is seen.

The invention discloses a deep neural network multitask hyper-parameter optimization method. The method comprises: firstly, a data training set of each task being subjected to model training to obtaina multi-task learning network model; secondly, predicting all points in an unknown region, screening candidate points from a prediction result, finally evaluating the screened candidate points, adding the candidate points and target function values of the candidate points into the data training set, and establishing a model, predicting, screening and evaluating again; and so on, until the maximumnumber of iterations is reached, finally selecting a candidate point corresponding to the maximum target function value from the data training set, that is, the hyper-parameter combination of each task in the multi-task learning network model. According to the method, the Gaussian model is replaced by the radial basis function neural network model, and the radial basis function neural network model is combined with multi-task learning and is applied to the Bayesian optimization algorithm to realize hyper-parameter optimization, so that the calculation amount of hyper-parameter optimization isgreatly reduced. The invention further discloses an electronic device and a storage medium.

The invention provides a hyper-parameter automatic optimization method and system of non-supervised machine learning. The hyper-parameter automatic optimization method of non-supervised machine learning includes the steps: according to the non-supervised machine learning algorithm, determining an algorithm performance assessment model of the non-supervised machine learning algorithm, the hyper-parameter of the non-supervised machine learning algorithm, the searching space of the hyper-parameter and the evaluation criterion of hyper-parameter optimization; and according to the algorithm performance assessment model, the searching space and the evaluation criterion, determining the optimal value of the hyper-parameter. The hyper-parameter automatic optimization method and system of non-supervised machine learning can deeply analyze the hyper-parameter problem in the non-supervised machine learning algorithm, can analyze the distribution rules of hyper-parameter in the algorithm, can assess the learning effect under different hyper-parameters, and can apply the rules to model training of machine learning so as to achieve the aim of automatically selecting the suitable hyper-parameter,and then the whole optimization process of hyper-parameter is automatically completed, so that the optimization efficiency is high and the usage complexity of the algorithm can be greatly reduced.

An online sparse matrix Gaussian process (OSMGP) uses online updates to provide an accurate and efficient regression for applications such as pose estimation and object tracking. A regression calculation module calculates a regression on a sequence of input images to generate output predictions based on a learned regression model. The regression model is efficiently updated by representing a covariance matrix of the regression model using a sparse matrix factor (e.g., a Cholesky factor). The sparse matrix factor is maintained and updated in real-time based on the output predictions. Hyperparameter optimization, variable reordering, and matrix downdating techniques can also be applied to further improve the accuracy and / or efficiency of the regression process.

The embodiment of the invention provides a hyper-parameter optimization method in a machine learning model. The method comprises the steps of sending a task configuration file to a server, wherein thetask configuration file comprises a hyper-parameter, a first algorithm type and a parameter search range, and the server runs a hyper-parameter optimization task so as to calculate a candidate valueof the hyper-parameter according to the parameter search range and a first hyper-parameter optimization algorithm corresponding to the first algorithm type; obtaining a candidate value of the hyper-parameter from the server; verifying the candidate value of the hyper-parameter, and updating the task configuration file according to a verification result; sending the updated task configuration fileto the server, wherein the server continues to run the hyper-parameter optimization task according to the updated task configuration file, and a new candidate value is obtained; and after the operation of the hyper-parameter optimization task is finished, determining an optimal value of the hyper-parameter according to a verification result of each candidate value.

The embodiment of the invention provides an interactive automated big data analysis system and method. The interactive automated big data analysis system includes a data receiving module, an operatormanagement module, an analysis model generation module, a process analysis and optimization module and a task scheduling execution module. First, the system supports an interactive function with strong ease of use; second, the system provides users with functions such as batch build process, differentiated multiplexing process, and dynamic planning process through flow control operators, which greatly simplifies the building steps of complex analysis processes. In particular, the process analysis and optimization module is used to perform algorithm selection for each analysis process, users donot have to consider the algorithm implementation problem, and the users' use threshold in the data analysis technical aspect is greatly reduced; in addition, the process analysis and optimization module is used to perform hyperparameter selection for processes needing to undergo hyperparameter optimization. The system automatically selects optimal execution hyperparameters for tasks with uncertain parameters, making the construction and execution of the analysis process more concise and efficient.

The present invention discloses a single frequency point frequency spectrum prediction method based on an optimum long short-term memory model. The method is employed to mine inherent laws and the correlation between frequency spectrum data through historical frequency spectrum data measured on a channel and grasp the developing law of the frequency spectrum state so as to predict the frequency spectrum state of a future moment. The method comprises the steps of: constructing a data set, and sliding a window with a fixed length, wherein the frequency spectrum data of the next slot time is taken as predicted tags; performing hyper-parameter optimization, combining cross validation based on Furuta experiment design method, and selecting good model configuration; establishing a prediction model based on the deep learning from the aspect of the classification, and performing training for the training set; inputting a test set into the trained prediction model, and obtaining a prediction result at the next slot time. The single frequency point frequency spectrum prediction method can predict the frequency spectrum evolvement rule.

The invention discloses a deep neural network hyper-parameter optimization method. The method comprises the steps of using a multi-task conditional neural network model for replacing an existing Gaussian process model to realize a data processing process of hyper-parameter optimization; determining the parameters, sequentially selecting the training sets of a decoder and an encoder according to the set parameters to perform network training, prediction, screening and evaluation; adding the candidate point of each task obtained after evaluation and a target function value into an observation set of the corresponding task; and carrying out the network training, prediction, screening and evaluation again, repeating the above steps until the maximum iterative frequency, and finding out the point, namely the hyper-parameter, when the target function value in the observation set of each task is maximum, thereby solving the problems of complex covariance calculation and the like during the hyper-parameter optimization realized by adopting a Gaussian process in the prior art, and simultaneously realizing the hyper-parameter optimization of multiple tasks. The invention further provides anelectronic device and a storage medium.

The invention discloses a pavement crack rapid extraction method based on a two-step convolutional neural network, and the method comprises the steps: carrying out the classification and judgment of whether there is a crack in a sub-block or not according to the characteristics that a pavement image is large in size and the recognition time is long, discarding the sub-block which is judged to be not damaged, and carrying out the second-step semantic segmentation of the damaged sub-block. In the classification process, a convolutional neural network 1 subjected to hyper-parameter optimization for a fracture continuous topological structure is adopted for training, according to a training result of the convolutional neural network 1, in the semantic segmentation process, a convolutional neural network 2 without downsampling is adopted for training, and a segmentation result with pixel-level accuracy is output. Because the proportion of the crack area in the pavement image is far less than that of the intact area, the two-step extraction method of classification and segmentation can quickly discard a large number of non-target areas before segmentation, avoids wasting the computing power, and greatly accelerates the recognition speed on the basis of very small recall ratio loss compared with a crack extraction algorithm of one-step direct semantic segmentation.

The invention provides a lead zinc ore exploitation process carbon emission short-term prediction method. Grey theories are synthesized to analyze carbon emission sources using motors as research objects and their influence factors in the ore exploitation process. Firstly, the carbon emission sources and their influence factors are analyzed, and the grey theories are adopted to perform clusteringanalysis for like factor merging; secondly, main influence factors are obtained through grey correlation analysis; finally, for solving the super- parameter optimization determining problem, a non-dominated sorting genetic algorithm with an elitist strategy is introduced into a gaussian process regression model, and a gaussian process regression prediction method based on the non-dominated sortinggenetic algorithm with the elitist strategy is proposed. It is proved through an experiment that the proposed prediction method can more accurately predict the carbon emission amount in the lead zincore exploitation process, can be directly achieved on a computer, is good in practicability, low in cost and easy to carry out, and prediction errors are smaller.

The invention discloses a federated learning method and device based on evolutionary computation, a central server and a readable storage medium. The method comprises the following steps of: acquiringa hyper-parameter of each participant group, and sending the hyper-parameters to each participant of each participant group to enable the participants in each participant group to carry out federatedlearning; acquiring sub-target models obtained by trained through federated learning of the participant groups, and performing iterative evolutionary computation based on the performance index of each sub-target model and the hyper-parameter corresponding to each participant group to obtain a target hyper-parameter; and sending the target hyper-parameter to each participant participating in federated learning to enable the participants to determine respective initial model parameters based on the target hyper-parameter and perform federated learning based on the initial model parameters to obtain a target model. The hyper-parameter optimization efficiency in federated learning is improved through evolutionary computation, and meanwhile, the performance of a federated learning model is remarkably improved.

The invention discloses an adaptive threshold channel occupation state prediction method based on an LSTM neural network. The method comprises five parts of determining an adaptive quantization threshold, determining the length of a historical sequence, generating a model input and output set, optimizing the model hyper-parameters and predicting a real-time frequency spectrum, wherein determiningan adaptive quantization threshold is characterized by adaptively setting the quantization thresholds for different channels through probability density estimation; determining the historical sequencelength is to determine the appropriate length of the historical sequence inputted into the model through the autocorrelation function analysis, generating the model input and output set comprises thesub-steps of quantifying and dividing a data set according to an adaptive quantization threshold and a historical sequence length, the hyper-parameter optimization is to optimize the model through agrid search and cross validation combined method, and the real-time frequency spectrum prediction is to predict the real-time occupation state of the actually acquired frequency spectrum data. According to the method, the future spectrum occupation state can be accurately predicted, and a slave user is assisted to adjust the transmitting parameters in advance, and therefore the spectrum resource utilization rate is increased.

The invention discloses a federated learning method and devicebased on evolutionary computation, equipment and a readable storage medium. The method comprises the following steps of: determining hyper-parameters of a first group corresponding to the first equipment in a hyper-parameter set, whereinthe equipment in the first group performs federated learning based on the hyper-parameters of the first group; obtaining a first sub-target model corresponding to the first group, receiving performance indexes of a second sub-target model sent by a plurality of pieces of second equipment, performingiterative evolution calculation based on the performance indexes of the first sub-target model, the performance indexes of the plurality of second sub-target models and the hyper-parameter set, wherein the first equipment and the plurality of second equipment determine respective initial model parameters based on the target hyper-parameters, and the first equipment and the plurality of second equipment perform federated learning based on the respective initial model parameters to obtain a target model. The performance of the federated learning model is remarkably improved while the hyper-parameter optimization efficiency in federated learning is improved.

Systems and methods include receiving a tuning work request for tuning hyperparameters of a third-party model or system; performing, by a machine learning-based tuning service, a first tuning of the hyperparameters in a first tuning region; identifying tuned hyperparameter values for each of the hyperparameters based on results of the first tuning; setting a failure region based on the tuned hyperparameter values of the first tuning; performing, by the machine learning-based tuning service, a second tuning of the hyperparameters in a second tuning region that excludes the failure region; identifying additional tuned hyperparameter values for each of the hyperparameters based on results of the second tuning; and returning the tuned hyperparameter values and the additional hyperparameter values for implementing the third-party model or system with one of the tuned hyperparameter values and the additional hyperparameter values.

A hyperparameter optimization method performed by a hyperparameter optimization apparatus to optimize hyperparameters of a model includes calculating an evaluation score for each of a plurality of hyperparameter samples constituting a first hyperparameter sample set by applying each of the hyperparameter samples to a target model, performing weighted clustering of the hyperparameter samples by using the calculated evaluation scores as weights and constructing a second hyperparameter sample set based on the clustering result.

The invention relates to a kernel function construction method and device, a data prediction method and device, equipment, and a storage medium. The kernel function construction method includes the following steps: obtaining a preset variance parameter and a preset frequency translation parameter; constructing a kernel function according to the preset variance parameter and the preset frequency translation parameter; performing characteristic decomposition on a covariance matrix of the kernel function to obtain a decomposition result; and optimizing a hyper-parameter of the kernel function bya maximized edge log-likelihood function to obtain an optimal kernel function according to the decomposition result. Hyper-parameter optimization prevents from being fallen into a locally optimal solution, so that the accuracy of the optimal kernel function obtained by the optimization is higher, and the accuracy of a gaussian process regression model constructed by the optimal kernel function ishigher, and therefore, when the gaussian process regression model is applied to data prediction, the accuracy of an obtained prediction result is higher.

The invention discloses a steel structure corrosion identification method based on a convolutional neural network. The steel structure corrosion identification method comprises the steps: 1, collecting pictures of a steel structure, preprocessing the pictures, and dividing a training set and a test set; 2, designing a convolutional neural network structure; 3, performing hyper-parameter optimization and model training through cross validation; and 4, inputting the to-be-identified pictures of the steel structure into the model obtained in the step 3 to obtain a rust identification result. According to the steel structure corrosion identification method, the convolutional neural network is utilized to realize automatic extraction of the structure corrosion characteristics; complex and tedious characteristic design work is avoided; the steel structure corrosion identification efficiency is improved; accurate identification of the steel structure corrosion is achieved; an objective identification result can be provided; and a new solution is provided for identification of the steel structure corrosion.

The invention discloses a deep reinforcement learning model training method and device based on hyper-parameter optimization, wherein the method comprises the steps of obtaining a plurality of initial hyper-parameter combinations and a plurality of first deep reinforcement learning models; training a plurality of first deep reinforcement learning models by adopting a plurality of hyper-parameters in the initial hyper-parameter combination to obtain training evaluation indexes respectively corresponding to the plurality of first deep reinforcement learning models; screening out a second deep reinforcement learning model from the plurality of first deep reinforcement learning models according to the training evaluation index; performing optimization processing on the initial hyper-parameter combination by adopting a plurality of target hyper-parameters corresponding to the second deep reinforcement learning model to form a target hyper-parameter combination; and obtaining a target deep reinforcement learning model. Therefore, hyper-parameter optimization and model training are combined to achieve training of the deep reinforcement learning model, the deep reinforcement learning model with higher performance can be trained, and the trained model can adapt to wider application scenes.

The invention discloses a ground radar automatic target classification and recognition method based on a one-dimensional convolutional neural network. The method comprises the following steps: preprocessing radar human and vehicle echo target sample data; obtaining a time domain echo signal, a power spectrum and a power conversion power spectrum, processing the preprocessed feature vector by usingan auto-encoder, constructing a one-dimensional convolutional neural network (1D-CNN) structure, and optimizing hyper-parameters of the convolutional neural network structure by using a Bayesian hyper-parameter optimization method. And inputting the coded data into a one-dimensional convolutional neural network, and performing target classification and recognition through a softmax classifier toobtain a classification and recognition result of personnel and vehicle samples. According to the method, the target classification and recognition function can be efficiently and stably completed, the calculation speed is high, implementation is easy, the network structure is simplified, the parameter calculation scale is reduced, and the method has excellent recognition accuracy for target classification and recognition of the low-resolution ground radar.

The invention discloses a regression model hyperparameter optimization method comprising the first step of sequentially training regression models with a hyperparameter being p1 on an entire data set,and obtaining L well-trained candidate hyperparameter models; the second step of obtaining errors of each candidate hyperparameter model on each sample (xi, yi); the third step of calculating a direction similarity matrix; the fourth step of calculating the symmetry similarity of each parameter; and the fifth step of searching a hyperparameter with the smallest symmetry similarity, and taking thehyperparameter as an optimal parameter for returning. The parameter optimization process provided by the method does not need the artificial setting of other parameters, and the optimization processhas no subjective interference; the optimization process does not require data division and has high efficiency and certainty; and main calculation parts of the method are relatively independent and can be processed in parallel on large data. The efficiency of the method is 6-8 times that of a cross validation method.

The invention relates to a credit scoring method based on hyper-parameter optimization, and the method comprises the steps: S1, collecting scoring main body information data, carrying out the preprocessing and feature selection of the data, and making a training data set and a test data set; S2, establishing a credit scoring model, selecting an XGBoost algorithm for modeling, and optimizing hyper-parameters of the algorithm by combining a Gaussian process with Bayesian; S3, selecting an optimal hyper-parameter set to fix an XGBoost algorithm, and training a credit scoring model by using the training data set; and S4, predicting and evaluating the credit scoring model by adopting the test data set, and calculating a credit score through a formula score = A-B * ln (p / (1-p)). According to theinvention, the hyper-parameters are optimized; when the target function curve cannot be determined, through conjecture hypothesis, it is determined that the target function meets multivariable Gaussian distribution, and the hypothesis evaluation model is further corrected, so that the efficiency and reliability of hyper-parameter optimization are improved, the model generation efficiency is improved, the enterprise model replacement efficiency is improved, and the risk control capability is improved.

The invention discloses a cascade hydropower station generating capacity prediction method based on a long-term and short-term memory network. The method comprises the steps of performing stability test on a hydropower station generating capacity time sequence; performing correlation test on the generating capacity time sequence; converting the generating capacity time sequence data into supervised learning data; establishing a generating capacity prediction model based on the long-term and short-term memory network; performing integrated empirical mode decomposition on the generating capacitydata to obtain a training set and a test set; training the generating capacity prediction model by using the training set, and performing model hyper-parameter optimization by using an improved discrete differential evolution algorithm to obtain optimal model parameters; and predicting the generating capacity of the cascade hydropower station by adopting the generating capacity prediction model.The method is suitable for predicting the generating capacity of large and medium-sized cascade power stations; the power generation quantity prediction model based on the LSTM neural network has moreadvantages for predicting the power generation quantity of the power station adjusted for many years, and the fitting precision of the model is improved through hyper-parameter optimization of the model.

The invention discloses an automatic machine learning system and method for streaming data analysis of concept migration, and the system comprises an obtaining module which is used for obtaining serialized streaming data sets of all batches; a generation module which is used for carrying out automatic feature processing, self-adaptive concept migration and hyper-parameter optimization on the serialized streaming data set of each batch to generate a prediction model; a feedback module which is used for inputting a preset batch of streaming data sets into the prediction model to generate a feedback result; and an online adjustment module which is used for carrying out online adjustment on the prediction model according to the feedback result so as to realize automatic machine learning. The system automatically selects a feature processing technology for different application scenes, self-adapts to concept migration, is good in performance in a real scene, automatically compresses an automatic learning search space through a resource control technology, and ensures that a model operation scale adapts to a data scale and features.

A method for optimizing a hyperparameter of an auto-labeling device performing auto-labeling and auto-evaluating of a training image to be used for learning a neural network is provided for computation reduction and achieving high precision. The method includes steps of: an optimizing device, (a) instructing the auto-labeling device to generate an original image with its auto label and a validation image with its true and auto label, to assort the original image with its auto label into an easy-original and a difficult-original images, and to assort the validation image with its own true and auto labels into an easy-validation and a difficult-validation images; and (b) calculating a current reliability of the auto-labeling device, generating a sample hyperparameter set, calculating a sample reliability of the auto-labeling device, and optimizing the preset hyperparameter set. This method can be performed by a reinforcement learning with policy gradient algorithms.

The invention provides a resource scheduling method and device in a hyper-parameter optimization process, equipment and a medium. A plurality of hyper-parameter tuning strategies are fused, each hyper-parameter tuning strategy is used for selecting a hyper-parameter combination for a machine learning model based on a hyper-parameter selection strategy corresponding to the hyper-parameter tuning strategy, and each round of iteration process comprises the step that currently available resources are determined; allocating currently available resources to the plurality of hyper-parameter tuning strategies; and obtaining one or more hyper-parameter combinations generated by the hyper-parameter tuning strategies allocated to the resources based on the allocated resources. Therefore, according tothe resource scheduling scheme fusing the plurality of hyper-parameter tuning strategies, the risk of poor effect or convergence to local optimum caused by using a single strategy can be eliminated,and the hyper-parameter optimization effect can be further improved by allocating resources to the plurality of hyper-parameter tuning strategies.

A method includes defining model attributes of an organizational machine model that organizes feedback data from one or more sources of the feedback data into plural different topic groups based on similarities in concepts expressed in the feedback data. The model attributes represent criteria for establishment of the organizational machine model and include a topic model number that defines how many of the different topic groups are to be created by the organizational machine model and used to organize the feedback data into, a hyperparameter optimization alpha value that defines how likely a feedback datum in the feedback data is to be included in a single topic group of the different topic groupings or multiple topic groups of the different topic groupings, and a hyperparameter optimization beta value that defines how broadly each of the different topic groups are defined relative to the feedback data.

The invention discloses a federated learning method and device based on evolutionary computation, a central server and a medium. The method comprises the following steps of: grouping the participantsparticipating in federated learning based on a grouping strategy; obtaining hyper-parameter sets, respectively sending each hyper-parameter in the hyper-parameter set to participants in different groups, so that participants in different groups can perform federated learning based on the received hyper-parameters; obtaining a sub-target model corresponding to each group; performing iterative evolution calculation based on the performance index, the grouping strategy and the hyper-parameter set of each sub-target model; and finally, sending the target hyper-parameter to each participant participating in federated learning, so that the participant determines respective initial model parameters based on the target hyper-parameter, and each participant performs federated learning based on theinitial model parameters to obtain a target model. The hyper-parameter optimization efficiency in federated learning is improved through evolutionary computation, and meanwhile, the performance of a federated learning model is remarkably improved.

The invention discloses a BiLSTM voltage deviation prediction method based on Bayesian optimization, and the method comprises the steps of carrying out the standard deviation standardization processing of a voltage deviation time series data set, carrying out the data segmentation according to a proportion, and obtaining a training set and a verification set; training a BiLSTM voltage deviation prediction model by using the preprocessed voltage deviation data training set; inputting the verification set into a trained BiLSTM voltage deviation prediction model, obtaining a voltage deviation prediction value, then carrying out inverse standard deviation processing, taking a root-mean-square error as a target function for the hyper-parameter optimization of the BiLSTM voltage deviation prediction model, optimizing the hyper-parameters of the BiLSTM voltage deviation prediction model by using a Bayesian optimization algorithm, and obtaining an optimal hyper-parameter combination; and taking the optimal hyper-parameter combination as hyper-parameters of a BiLSTM prediction model, constructing a BiLSTM voltage deviation prediction model based on a Bayesian optimization algorithm, and predicting the voltage deviation time sequence data to obtain final prediction data. The invention is high in precision and reliable in prediction effect.