51 results about "Initial training" patented technology

The invention discloses a database question and answer model training method and device, a storage medium and computer equipment, and the method comprises the steps: associating standard structured query language statements, standard execution result answers and standard natural language questions, and generating training annotation data; and collecting simulation derivation problems possibly proposed for the database to obtain non-labeled data for training. Based on a GRPO reinforcement learning framework and a scoring reward function provided by a double-tower model, training is carried out on the scoring reward function by utilizing training labeling data, supervised fine tuning training is carried out on a database question and answer model, and non-labeling data for training, format rewards, executable rewards and scoring rewards of the scoring reward function are combined, so that the scoring reward function of the database question and answer model is obtained. And continuing to train the database question and answer model after supervised fine tuning training. Preliminary training is carried out through a small amount of annotation data, then subsequent training is carried out through non-annotation data, the reasoning ability of the model can be stimulated, the annotation cost is reduced, and the training efficiency is improved.

The invention discloses a robot motion strategy model optimization method and a related device, and relates to the field of artificial intelligence, and the method comprises the steps: carrying out the preliminary training of a robot motion strategy model based on an autonomous interaction playback data memory library, and obtaining a preliminarily trained robot motion strategy model; according to the situation when the second robot executes the task, whether a scene which does not meet the requirement exists or not is judged; if yes, retraining the preliminarily trained robot motion strategy model on the basis of a hybrid interaction playback data memory bank for a scene which does not meet the requirement to obtain a trained robot motion strategy model; according to the method and the device, the demonstration data of the person is collected for a scene in which task execution of the preliminarily trained robot motion strategy model does not meet requirements, so that post-mixed training is performed in a targeted manner, the robot motion strategy model can quickly meet training requirements, the training efficiency of the robot motion strategy model is improved, the training period is shortened, and the training efficiency of the robot motion strategy model is improved. The training cost is reduced.

The invention discloses a fine tuning method and system for precise parameters of a pre-training language model. The method comprises the following steps: constructing a robust adapter; a plurality of constructed robust adapters are connected in series to form a robust adapter chain, and the maximum chain node is determined; preparing a weight of a pre-training language model and a data set of a corresponding task label; firstly, each robust adapter in the robust adapter chain is subjected to preliminary training, optimization and approximate fine tuning of parameters of the robust adapters; after the parameters of the robust adapter are approximately fine-tuned, the parameters are combined with the weight of the preorder frozen pre-training language model to complete calculation of a current robust adapter chain node, residual information used for next iteration is generated, and the parameters of the current robust adapter are updated at the same time; and carrying out iterative training, updating residual information until the number of iterations is equal to the maximum node of the robust adapter chain, and obtaining the optimal weight of the pre-training language model and the optimal parameters of the robust adapters, so that more stable convergence and simpler hyper-parameter adjustment can be provided.

The invention discloses an LLM distributed training method, system and device and a storage medium. The method comprises the steps of collecting data and performing data cleaning and preprocessing; dividing the preprocessed data into a plurality of subsets; a part of data subsets serve as a training set, the remaining data subsets serve as a verification set, and the data subsets are stored on different computing nodes in a distributed mode; each computing node performs distributed training on the LLM by using the training set, and each computing node outputs a preliminarily trained LLM when the maximum training round is reached by taking minimization of a loss function of the LLM as a target; integrating the preliminarily trained LLMs in the computing nodes to obtain an integrated LLM; evaluating the integrated LLM by using the verification set and the evaluation index, and adjusting training parameters of the LLM to obtain an optimal LLM; and performing automatic content creation by using the optimal LLM. The resource utilization rate and the calculation performance are further improved, the problem of high calculation complexity is solved, and the overall training efficiency is improved.

The invention relates to a place name recognition method and device, equipment, a storage medium and a program product, and relates to the technical field of natural language processing. By adopting the method, the problems of overfitting and knowledge forgetting can be relieved. The method comprises the following steps: performing format conversion on obtained initial training data to obtain current training data consisting of a text to be extracted and an extracted place name; querying a plurality of synonyms corresponding to the extracted place names, and performing data enhancement on the current training data based on the plurality of synonyms to obtain a target training data set; constructing an auxiliary classification task according to the plurality of synonyms, the negative examples and the classification loss; based on the target training data set and the auxiliary classification task, training the initial model until model parameters of the initial model are the same as target parameters, and obtaining a preliminary training model; obtaining model fine tuning loss, and adjusting the preliminary training model according to the model fine tuning loss to obtain a pre-training model; the pre-training model is used for place name identification.

The invention discloses a model training method, a humanoid robot dynamic control method based on multi-modal data fusion and related equipment, and the method comprises the steps: carrying out the preliminary training of a neural network model through a first-stage training sample set, obtaining an initial motion instruction, and carrying out the dynamic control of a humanoid robot based on the execution feedback information after the humanoid robot executes the initial motion instruction. Generating a sample optimization result, constructing a corresponding second-stage training set, carrying out the secondary training of the neural network model through the second-stage training set, obtaining a robot control model, and achieving the model training of multi-modal data fusion; the action control instruction is output through the robot control model according to the dynamically collected multi-mode sensing data, the humanoid robot is controlled to execute the target action based on the action control instruction so as to synchronously simulate the action of an operator, and the action simulation efficiency and accuracy of the humanoid robot in a complex scene are remarkably improved. The method can be widely applied to the technical field of robot control.

The invention relates to a production line visual defect detection optimization method, a computing device and a readable storage medium, and the method comprises the steps: carrying out the preliminary training of a large language model through a collected product database, forming a pre-trained large language model, carrying out the preliminary configuration of a visual detection system, and carrying out the defect detection of a product, when a defective product is detected subsequently, the image data of the defective product and the text guidance data can be bound to form secondary training data to carry out fine tuning on the pre-trained large language model, so that model optimization is realized, and the accuracy of product defect detection is improved. When a defect image of a to-be-detected product is input into the big language optimization model, a visual detection process and a parameter configuration guidance text corresponding to the defect image are output through the big language optimization model, the output form is presented in a dialogue mode based on the big language optimization model, and guided configuration of each process and parameter is completed. And the technical requirements of production line personnel are reduced.

The invention provides an optimization method and device for an automobile collision simulation model, a medium and equipment, and the method comprises the steps: building an initial model of a simulation vehicle based on the vehicle parameters of the simulation vehicle; performing preliminary training on the initial model to obtain a preliminary training model; training the initial training model again to obtain a target model of the simulation vehicle; according to the method, an initial model of a simulation vehicle is constructed, the initial model is preliminarily trained by using historical collision simulation sequence data, rapid training is performed by using simulation data with a large data size, so that the whole vehicle model rapidly reaches a certain precision, and the preliminary training model is further trained by using experimental test sequence data, so that the accuracy of the whole vehicle model is improved. According to the method, the precision of the whole vehicle model is improved by adopting the test data with small data volume, and the training precision of the initial model is improved by selecting the historical collision simulation sequence data with large correlation with the experimental test sequence data, so that the efficiency and effect of model training and optimization are improved.

The invention discloses a method for optimizing the grabbing capacity of a musculoskeletal mechanical arm. The method comprises the steps that a rigidity parameter space of the musculoskeletal mechanical arm is constructed; a plurality of initial rigidity parameter sets are generated based on the space, reward functions of the rigidity parameter sets are modeled into a Gaussian process, preliminary training is conducted on a grabbing control strategy based on the initial rigidity parameter sets and energy constraint conditions, initial reward values are obtained, and the rigidity parameter sets and the initial reward values of the rigidity parameter sets are stored in an observation data set; storing the rigidity parameter set into a rigidity parameter pool; selecting a to-be-optimized rigidity parameter set from the rigidity parameter pool through the expectation improvement function, training the grabbing control strategy based on the to-be-optimized rigidity parameter set and the energy constraint condition, updating the Gaussian process, the observation data set and the rigidity parameter pool, and continuously repeating the step until the iteration termination condition is reached, so as to obtain the optimal grabbing control strategy. And an optimal rigidity parameter and a grabbing control strategy are obtained. According to the method, the success rate and generalization ability of the grabbing task can be remarkably improved.

The invention relates to the technical field of computer science, and discloses an intelligent auxiliary system and method for rehabilitation nursing, and the method comprises the steps: obtaining and inputting user data, and forming a personalized rehabilitation threshold; a reinforcement learning model is initialized, real-time physiological feedback of the user is collected, a preliminary training scheme is generated in combination with the personalized rehabilitation threshold, and the training scheme is optimized in combination with micro expressions and gestures of the user; performing rehabilitation progress evaluation on the trained user data, and judging whether a training scheme needs to be adjusted or not; and based on a judgment result, updating a training scheme, and adjusting a personalized rehabilitation target. The intelligent decision-making ability of the training strategy is improved, human intervention requirements are reduced, the rehabilitation experience and training effect of the user are finally improved, and the rehabilitation process is accelerated.

The application discloses a kind of based on blockchain's federated learning method, device and blockchain node, it is related to blockchain technology field, to solve the problem of poor model precision that existing based on blockchain's federated learning framework is trained.The method comprises: training initial model using own sample until target sample residual that satisfies preset condition is obtained, different alliance chain members have different attribute characteristics;Collaborative training request is issued to blockchain, so that the participants in the alliance chain that respond to collaborative training request collaboratively train blockchain model, and target sample residual is carried in collaborative training request;Collaboratively train blockchain model using common sample with participant, and submit the model parameter obtained after training to blockchain.In this way, the requester publishes the residual of preliminary training in blockchain, so that each participant collaborates to train blockchain model using its unique data characteristics, and higher precision model training can be achieved.

The invention relates to the technical field of computers, in particular to a code review method and device, electronic equipment and a storage medium, which are used for efficiently screening out high-quality and high-diversity training data so as to improve the accuracy of code review. The method comprises the following steps: acquiring a training sample set and a test sample set, and dividing the training sample set into a plurality of training sample subsets; the test sample set comprises test samples reaching expected quality; after a to-be-trained code review model is preliminarily trained based on the training sample set, the test influence degree of each training sample subset on the code review model on the test sample set is determined; for at least one target training sample subset of which the test influence degree meets a preset influence condition in the plurality of training sample subsets, performing clustering according to a sample gradient to obtain a plurality of clusters; and sampling is carried out in each cluster, and the code review model after preliminary training is trained again according to the obtained sampling sample set.

The invention discloses an auscultation sound data mining method and device, and the method comprises the steps: obtaining a training data set which comprises labeled training data and unlabeled training data; iterative training is carried out on the main model and the auxiliary model based on the labeled training data to obtain a model pool, and the model pool comprises the main model and the auxiliary model which are preliminarily trained; respectively inputting the unlabeled training data into the primary model and the auxiliary model which are preliminarily trained for classification processing to obtain a classification result; performing data screening on the classification result to obtain high-value data; after labeling the high-value data, combining the labeled high-value data with the labeled training data to serve as new labeled training data; and performing iterative training on the primary model and the auxiliary model which are subjected to preliminary training on the basis of the new annotation training data until a preset convergence condition is met. The invalid labeling is effectively reduced, the data labeling efficiency is improved, and the labeling cost in the aspects of manpower, time, funds and the like is reduced.

The present disclosure provides a medical imaging system and method. The method may include obtaining a machine learning model and preliminary training data of at least one sample subject.The method may also include generating training input data by processing the preliminary training data, the preliminary training data being superior to the training input data with respect to a data quality parameter. The method may further include determining a trained machine learning model by training the machine learning model based on the training input data and the preliminary training data, the preliminary training data being configured as training target data of the machine learning model.

The application discloses a model training method, a humanoid robot dynamic control method based on multi-modal data fusion and related equipment. The scheme performs preliminary training on a neural network model through a first-stage training sample set, obtains an initial action instruction, generates a sample optimization result and constructs a corresponding second-stage training set based on execution feedback information after the humanoid robot executes the initial action instruction, performs secondary training on the neural network model through the second-stage training set, obtains a robot control model, and realizes model training of multi-modal data fusion. Through the robot control model, action control instructions are output according to dynamically collected multi-modal sensing data, and the humanoid robot is controlled to execute target actions based on the action control instructions to synchronously imitate operator actions, so that the action imitation efficiency and accuracy of the humanoid robot in a complex scene are significantly improved. The application can be widely applied to the technical field of robot control.

The invention relates to the technical field of power system operation control, and discloses a training method of a transient voltage stability prevention and control evaluation model considering source load uncertainty, and the method comprises the steps: carrying out the preliminary training of a stability evaluation model through selecting a small number of training samples which are sparsely distributed, and obtaining an initial stability evaluation model; then, the initial stability evaluation model is put into prevention and control optimization solution; according to the change of an optimal solution obtained by the prevention and control optimization model, respectively adding training samples in a training sample sparse region and a region near the optimal solution to avoid missing the optimal solution and improve the precision of the optimal solution, and retraining the evaluation model by using the updated training samples; through several rounds of training, the stability evaluation model is gradually improved, and compared with an existing stability evaluation model training method, the method has the advantages that the number of training samples is effectively reduced on the premise of ensuring the precision, the time consumption of model training is reduced, and the prevention and control optimization efficiency is also improved.

The invention discloses a mouth shape driving method and device of a virtual character and a storage medium. The method comprises the following steps: carrying out preliminary training on a deep learning model by utilizing an original data set to obtain a preliminary model, and carrying out preliminary screening on the original data set by utilizing the preliminary model to obtain a preliminary screening data set; increasing a difference value between the similarity of a positive sample and the similarity of a negative sample in the primary screening data set based on a dynamically adjusted stepping value, retraining the primary model based on the difference value to obtain a retraining model, and rescreening the primary screening data set by using the retraining model to obtain a secondary screening data set; and performing enhanced training on the retraining model by using the secondary screening data set to obtain an enhanced model, and driving the mouth shape of the virtual character by using the enhanced model. The technical problem that in the prior art, mouth shape driving of a virtual character is not vivid is solved.

There are provided methods and systems for generating a training dataset for use in supervised learning, by augmenting synthetic data labels with human-generated data labels. A preliminary training dataset corresponding to a first set of instructions is generated by a large language model (LLM). A contribution score for the preliminary training dataset is generated. Responsive to determining that the contribution score is below a threshold, human-generated data is obtained, based on a second set of instructions. An updated contribution score may be generated, based on the preliminary training dataset and the human-generated data. Responsive to determining that the updated contribution score meets or is above the threshold, the training dataset may be generated based on the preliminary training dataset and the human-generated data. The disclosed methods and systems may enable robust and efficient model performance while minimizing costs associated with labor intensive human labelling.

The present application relates to the technical field of space target detection, and specifically provides a progressive active learning weak supervision training method for dim and weak target detection, comprising: obtaining sample data to form a sample set, dividing the sample set into an initial sample set and a to-be-evaluated group according to the signal-to-noise ratio of the target, training a model using the initial sample set to obtain a preliminarily trained model, and starting in-group iteration: selecting the to-be-evaluated group in turn in descending order of signal-to-noise ratio, dividing the target in the to-be-evaluated group into simple samples and difficult samples, training a model using the simple samples to obtain an updated model, re-dividing the difficult samples into simple samples and difficult samples, repeatedly training and dividing until a preset in-group iteration termination condition is met, completing in-group iteration, and selecting a next to-be-evaluated group in turn to repeatedly execute in-group iteration. The present application progressively improves the training difficulty according to the real-time capability of the model, improves the detection precision and training efficiency, and reduces the data labeling cost.

The invention discloses a visual emotion model pre-training method based on semi-supervised learning, and the method comprises the steps: obtaining a video clip of a to-be-recognized emotion type, and obtaining a prediction result of the emotion type through a trained visual emotion model; wherein the training process of the model comprises the following steps of: firstly, initializing the model, and carrying out preliminary training on a video encoder and a classifier by using a labeled data set; then entering an iterative training stage of semi-supervised continuous learning, and in each training round, performing data partitioning and combination on the unlabeled data set to construct training data; performing feature extraction and data enhancement on each video clip sample by using an encoder to obtain a feature set of the current round; performing deep clustering and cluster label generation on the feature set, and finally performing cluster purity screening and label consistency verification to form a high-purity cluster subset; a joint total loss function is constructed based on the high purity cluster subset to update the visual sentiment model.

The application discloses a model training method and device, electronic equipment and a storage medium; the application can obtain a pre-training model and a target model; obtain an unlabeled sample group and a labeled sample group, and unlabeled probability and labeled probability; based on the unlabeled probability and the labeled probability, linearly combine the unlabeled sample group and the labeled sample group respectively to obtain unlabeled augmented samples corresponding to the unlabeled sample group and labeled augmented samples corresponding to the labeled sample group; based on the unlabeled augmented samples and the pre-training model, preliminarily train the target model; based on the labeled augmented samples and the pre-training model, retrain the preliminarily trained target model to obtain a trained target model. In the application, the target model is trained by augmenting the unlabeled samples and the labeled samples and combining the pre-training model, so that the generalization and robustness of the target model can be improved. Therefore, the application can improve the performance of the target model.

The present invention discloses an eloquence optimization method, device and medium based on instant feedback and tracking, the method comprising: calculating the probability distribution of the feedback strategy adapted by the user in a preset situation to obtain a first probability distribution; combining the first probability distribution with the user's current performance data in the speech data to generate a first feedback plan and train the user to obtain real-time training data; generating an eloquence expression optimization plan according to the user's expected skill level value and the real-time training data. The present invention proposes an eloquence optimization method, device and medium based on instant feedback and tracking, which performs preliminary training on the user through the first feedback plan generated by the user's current performance data, can obtain first-hand feedback real-time training data, timely understand the user's eloquence weaknesses, and then generate a more scientific eloquence expression optimization plan to track the user, which can solve the problem that it is difficult to conduct effective eloquence tracking training for the user based on feedback information.

The invention relates to the technical field of data processing, in particular to a physical training plan intelligent generation and management method. Comprising the steps that physiological data, athletic performance data, external environment data and health state data of an athlete are obtained and serve as original multi-dimensional data, the original multi-dimensional data are preprocessed, and preprocessed multi-dimensional data are obtained; performing feature extraction to obtain preliminary feature data; performing enhancement processing to obtain feature data after enhancement processing, and performing preliminary analysis in combination with a training target to generate a preliminary training plan; in the training process of the preliminary training plan, real-time data of the athletes are obtained, a dynamic adjustment algorithm based on multi-dimensional data analysis and self-adaptive feedback optimization is used for analysis, and quantitative feedback is obtained; and dynamically adjusting and optimizing the preliminary training plan to obtain an optimal training plan. The technical problems that related data processing of athletes is not accurate and comprehensive enough, and the real-time performance of the data is poor are solved.

The invention provides an intelligent logistics vehicle optimization control model generation method and device and a storage medium, and belongs to the technical field of automatic driving, and the method comprises the steps: training a depth deterministic strategy gradient algorithm model based on the dynamic information and risk information of a logistics vehicle in a simulation environment, and obtaining a preliminary training model; and taking the dynamic information and the risk information of the logistics vehicle in the actual environment as the input of the preliminary training model, taking the logistics vehicle control instruction as the output of the preliminary training model, training the preliminary training model again, and adjusting the dynamic state of the logistics vehicle based on the logistics vehicle control instruction. The preliminary training model is optimized according to dynamic information fed back by the logistics vehicle, and a logistics vehicle optimization control model is obtained; and the logistics vehicle optimization control model is used for generating a control instruction of the logistics vehicle based on the dynamic information and the risk information in the environment where the logistics vehicle is located. The problem that the logistics vehicle control method is difficult to meet the efficient and safe operation requirements of the logistics vehicle can be solved.

The invention discloses a large model training method and device for data enhancement based on high-quality data, and the method comprises the steps: inputting basic data into a self-training word vector model, and generating a corresponding high-dimensional vector representation; inputting the high-dimensional vector representation into the embedded large model with the removal position for training and tuning to obtain a preliminarily trained large model; and performing performance evaluation on the preliminarily trained large model by using the verification set, and performing multi-round iterative training on the preliminarily trained large model based on an evaluation result to obtain a finally trained large model. By optimizing data input and enhancing the data processing capability, the effectiveness and precision of the large model are improved, the problem of excessive dependence on large data and the large model in the prior art is solved, the utilization efficiency of the large model on high-quality data is remarkably improved, the generalization capability of the large model and the performance in practical application are enhanced, and the method is suitable for popularization and application. And a technical path with higher operability and higher efficiency is provided for large model training.

This invention provides a missile guidance method based on hierarchical reinforcement learning, belonging to the field of control strategies. The steps are as follows: First, a missile-target kinematic model and a missile velocity decay model are introduced; second, based on the missile-target kinematic model, process rewards and outcome rewards are designed to obtain a total reward function, guiding the missile to pursue the target; third, after completing the reward function design, the Flexible Action-Evaluation Algorithm (SAC) is used to initially train the missile under the condition that the target is moving in a straight line, obtaining an initial guidance strategy; finally, based on the initial guidance strategy, the enemy aircraft adopts an evasion strategy, and the SAC is used to continue training the missile, ultimately obtaining a high-quality guidance strategy. This invention enables missiles to autonomously adjust their flight paths in complex combat situations, overcoming interference and challenges in complex environments, thereby significantly improving guidance accuracy and attack success rate, and enhancing missile strike effectiveness and survivability.

The invention relates to a multi-modal multi-round dialogue agent construction method in the IT operation and maintenance service field, and the method comprises the steps: extracting multi-modal information from each historical IT operation and maintenance case, so as to form a training data set; performing preliminary training on the joint neural network to establish a multi-round dialogue agent, performing behavior judgment, tracking a dialogue state, and preliminarily determining intention and named entity information; based on the dialogue action, a DQN reinforcement learning model is adopted to perform reinforcement training on a multi-round dialogue agent, intention-based training modeling is performed on dialogue action generation, and a dialogue strategy is clarified; based on a vertical domain question and answer data knowledge base, a multi-round dialogue agent generates a natural language in a rule matching mode, supplemental information is fed back to a user, and then perfect dialogue state information is combined to clarify the intention of the user. The method has the beneficial effects that fuzzy information can be preliminarily provided according to the user, the user is gradually guided to focus specific intentions through dialogues, and a solid foundation is provided for the intelligence of the I-T operation and maintenance service.

The application provides a vehicle collision simulation model optimization method, device, medium and equipment, an initial model of a simulation vehicle is constructed based on vehicle parameters of the simulation vehicle; the initial model is preliminarily trained to obtain a preliminary training model; the preliminary training model is further trained to obtain a target model of the simulation vehicle; that is, the initial model of the simulation vehicle is constructed, and the initial model is preliminarily trained by using historical collision simulation sequence data, so that the simulation data with a large amount of data is used for rapid training, the vehicle model quickly reaches a certain precision, and the preliminary training model is further trained by using test sequence data, so that the test data with a small amount of data is used to improve the precision of the vehicle model, and the historical collision simulation sequence data with a large correlation with the test sequence data is selected to improve the training precision of the initial model, thereby improving the efficiency and effect of model training and optimization.

The present invention relates to the field of computer science and technology, and discloses an intelligent assistance system and method for rehabilitation nursing, including: acquiring and entering user data to form a personalized rehabilitation threshold; initializing a reinforcement learning model, collecting real-time physiological feedback from the user and combining it with the personalized rehabilitation threshold to generate a preliminary training plan, and optimizing the training plan by combining the user's micro-expressions and gestures; evaluating the rehabilitation progress of the user data after training to determine whether the training plan needs to be adjusted; and updating the training plan based on the judgment results and adjusting the personalized rehabilitation goals. This system improves the intelligent decision-making ability of the training strategy, reduces the need for human intervention, and ultimately enhances the user's rehabilitation experience and training results, accelerating the rehabilitation process.

The present invention provides a wind power prediction method and system based on data enhancement and hidden layer incremental cascade. The method comprises: collecting an initial sample set and performing preprocessing, and dividing the initial sample set into a training set, a validation set and a test set; based on a traditional RVFL model, introducing a hidden layer node addition mechanism, and constructing an HLICRVFL model; performing preliminary training on the HLICRVFL model using the training set; adding Student's t distribution noise to the training set, generating a new training set, and training the newly added nodes, while keeping the parameters of the original nodes unchanged, continuously performing training until the model converges, and finally using the optimal model to predict wind power; the present invention adopts a hidden layer incremental training mode, does not need to retrain all parameters, only needs to train the newly added nodes, and significantly reduces the calculation cost; in addition, the present invention performs data enhancement based on Student's t distribution, can generate higher quality training samples, and further improves the prediction accuracy of the model.