82 results about "Training performance" patented technology

The invention discloses a federated learning method and system based on batch size and gradient compression ratio adjustment, which are used for improving model training performance. The method comprise the following steps: in a federated learning scene, enabling a plurality of terminals to share uplink wireless channel resources; completing the training of a neural network model together with anedge server based on training data of a local terminal; in the model training process, enabling the terminal to calculate the gradient by adopting a batch method in local calculation, and in the uplink transmission process, compressing the gradient before transmission; adjusting the batch size and the gradient compression rate according to the computing power of each terminal and the channel stateof each terminal, so as to improve the convergence rate of model training while ensuring the training time and not reducing the accuracy of the model.

The invention discloses a federal model training method and device, electronic equipment and a storage medium, the method is executed by a server in a federal model training system, and the method comprises the following steps: issuing a global model corresponding to a target task to each edge working node; based on a preset algorithm, determining a specified number of local models participating in global model updating according to the current network resources and the number of the target tasks; and when a specified number of local model update data is received, performing federal aggregation to obtain an updated global model. According to the technical scheme provided by the embodiment of the invention, the number of local models participating in global model training is dynamically determined by combining the preset algorithm with the network resources of each iteration, so that the network resource utilization rate of model training in the federated learning process is effectively improved; the problem that the federated model cannot converge due to the problem that network resources are limited is avoided, and the training performance of the federated model is greatly improved.

The invention relates to an adaptive image denoising method based on depth learning, comprising the following steps: step 1) establishing an image set; 2) constructing an adaptive depth convolution neural network; 3) training the adaptive depth convolution neural network: setting the learn rate and momentum parameters of the depth convolution neural network, training the adaptive depth convolutionneural network through the depth learning framework until the train reaches the maximum number of iterations, and generating the trained adaptive depth convolution neural network model; 4) image denoising: inputting the image to be denoised into the trained adaptive depth convolution neural network model to obtain the corresponding residual image, and subtracting the image to be denoised from theresidual image to obtain the denoised image. The method further improves and stabilizes the training performance of the convolution neural network, and ensures the denoising performance, saves time,does not need to manually adjust parameters, and can realize blind denoising of the unknown noise level image.

The invention discloses a computer-aided pulmonary nodule classification method based on migratable multi-model integration to solve the technical problem of low accuracy of an existing pulmonary nodule classification method. The technical scheme includes the steps of training three pre-trained deep convolutional neutral networks (Pre-trained DCNN), describing the textural heterogeneity, shape heterogeneity and global characteristics of pulmonary nodules respectively, performing weighted average for results of the deep convolutional neutral networks, and subjecting the weight of each network to adaptive learning through an error back propagation mechanism so as to improve the accuracy of the pulmonary nodule classification method, wherein, the Pre-trained DCNN migrates the good image characterization capability of the deep convolutional neutral network on a large data set onto a pulmonary nodule classification task so that the training performance of the deep convolutional neutral networks on small pulmonary nodule data is effectively improved. The invention overcomes the technical problem of low accuracy of a classification method based on single type of information, and the accuracy rate can be up to 93%.

The invention discloses an adaptive distributed parallel training method for a neural network based on reinforcement learning, and provides an optimal solution for segmentation and scheduling of a large-scale complex neural network. Firstly, the influence of a neural network model structure and calculation attributes on execution performance is analyzed, on this basis, performance factors including calculation cost, communication cost, memory utilization rate and the like are extracted, a multi-dimensional performance evaluation model capable of comprehensively reflecting distributed training performance is constructed, and comprehensive performance of a parallel strategy is improved; secondly, self-adaptive grouping of operators is realized according to attribute characteristics of the operators by utilizing a feed-forward network, the degree of parallelism is determined, and end-to-end strategy search is realized while the search space is reduced; and finally, based on importance sampling, a near-end strategy gradient iteration optimization reinforcement learning model is adopted, an optimal segmentation and scheduling strategy is searched, the strategy network offline learning capability is expanded, and algorithm stability, convergence rate and strategy search performance are improved.

Systems and methods for encouraging regular participation in training and collaborative activities by employees of an organization are provided through a gamified training platform. Preferably, these systems and methods are implemented in an Internet based environment accessed by the employees from various types of mobile devices. A training activity is determined based on training modules pertaining to the organization and made accessible to the employees for a specified period. When a first employee performs the training activity, the platform generates a collaborative review activity that allows other employees to review the first employee's training performance. When a second employee performs the collaborative review activity, the platform generates a collaborative feedback activity that allows yet other employees to rate the second employee's review. In certain embodiments, the platform may generate a leaderboard displaying the relative ranking of the employees based on the employees' performance of the training and collaborative activities.

A system for and method of weight training performance recordation and analysis are provided. In accordance with the present invention, a user records workout data such as but not limited to type of workout, muscle group, wait time between sets, amount of weight and number of sets in an acquisition component and records the performance data to a database. The stored information can later be retrieved from the database by the user for visualization and analysis. This invention is a portable electronic device that helps gym goers track and organize their exercise progress, offers easy viewing on a computer, and give advice.

Mesh networks of radio node pods for measuring performance of teams and individual athletes by first stimulating and then by measuring a parameter of motion selected from velocity, vector, acceleration, force, and rebound. The system includes one or more radio node pods, each radio node pod having a microprocessor, supporting circuitry, a machine layer with one or more sensors and actuators, firmware for essential functions, and a soft socket for receiving “codelets” on the fly, each codelet containing a soft mini-script and attendant variables for iteration of a stimulus-response-sequence (SRS) customized to a previous iteration or training goal. Radio node pods may work in clusters and are typically multipotent, each pod performing specialized functions as dictated by a resident codelet but otherwise all pods having the same or similar hardware. Shared resources, either internal or external to the mesh network, are used for data analysis. Thus a single pod may trigger a stimulus to a user, and another pod may record a response, but are interchangeable, and each response may be a stimulus to trigger another SRS. Communications with an external administrative network or cloud host is generally delegated to a bridge pod dedicated as a gateway or portal. Each individual subject or team is assessed for performance metrics by which a stimulus results in a qualitative or parametric response. According to current best practice, radio pod nodes are synchronized in each mesh network and wirelessly report raw data and/or derived data to an administrative module for reporting, display and recordation. Relative performance of individuals or teams can be tracked or trended to detect weaknesses and improve workout, sports, military training performance, and contests can also be scored using these systems.

The invention provides an eye-movement tracking method based on recalling and annotation. The eye-movement tracking method comprises the following steps: (1) defining a recalling and annotation manner of a user and setting a corresponding task; (2) representing a task instruction on a display screen and publishing the task to the user; (3) presenting an image stimulating source on the display screen; (4) allowing the user to observe the image stimulating source on the display screen according to task requirements; (5) allowing the user to recall fixation points generated in a task executing process according to the task requirements and annotate coordinate positions and a sequence of the fixation points on the display screen; if entering a training mode, skipping to step (6); if entering a normal mode, skipping to step (8); (6) not representing the image stimulating source, which is represented to the user by the training mode, to the user in the normal mode; (7) evaluating training performances of the user; and (8) when the user enters the normal testing mode, recording andstoring fixation point positions annotated by the user and sequence data thereof. The invention further provides a system utilizing the method.

The invention discloses a power load prediction method based on a Bayesian regularization neural network, and the method comprises the following steps: obtaining electric quantity historical data, andanalyzing a key factor influencing the increase of electric quantity; determining a BP neural network structure; Training the network by using a BP algorithm to minimize the total error F (W); calculating the number of effective parameters; calculating a new estimated value of the hyper-parameter sum by using a Bayesian method; Repeatedly executing the above steps until the required precision isachieved, thereby completing the establishment of the Bayesian regularization optimization neural network; and inputting a new key factor influencing the increase of the power consumption to obtain the whole-society power load condition of the time period. The method has the advantages that the Bayesian method is applied to the modeling process of the neural network, the regularization method is used for correcting the training performance function of the neural network to improve the generalization ability of the neural network, the convergence speed is high, and a smaller training error canbe obtained.

The invention discloses a radar jamming equipment simulation training system, and belongs to the technical field of simulated training. The system comprises a radar jamming device and a simulated training system controlling the radar jamming device to work normally. The radar jamming device comprises a charge station and a hamming station. A supporting platform is installed at the side wall of theradar jamming device. A first seat, a second seat and a jamming seat are installed at the surface of the supporting platform in parallel; the simulated training system comprises an equipment simulated training system and an electronic blue force and evaluating system. An electronic blue force module inside the electronic blue force and evaluating system is used for generating the complex electromagnetic environment, the charge station and the jamming station simulates real equipment operation training, the equipment simulated training system enables training workers to perform actual operation training in the simulated environment, and after training is finished, the training performance is evaluated through the evaluating system, so that the training cost is lowered, the training time isprolonged, and the fighting capacity of a radar jamming team is improved.

The invention discloses a fire extinguishing training performance evaluation system and method based on shipboard aircraft fire in the field of fire fighting. The objective of the invention is to solve the problems of low efficiency and easy error existing in a conventional fire-fighting training performance evaluation method and poor accuracy caused by incapability of truly reflecting the importance of an evaluation subject. The system comprises an index system construction module used for establishing an index system according to indexes of all levels input by a user; a discriminant matrix construction module used for generating a discriminant matrix; a consistency verification module used for determining the maximum characteristic value and the characteristic vector and carrying out consistency verification; wherein the normalization module is used for performing normalization processing, an evaluation grade standard construction module used for constructing an evaluation grade standard library and determining evaluation values, and an index scoring module used for determining an evaluation decision matrix of a primary index, a primary evaluation result vector, a primary index evaluation result and an overall evaluation result. The invention is used for fire extinguishing training performance evaluation of shipboard aircraft fire.

The invention provides an autism spectrum disorder mixed reality rehabilitation training system and method, the clinical rehabilitation training course content of an autism child is combined with theadvantages of the mixed reality technology, and a set of training course is designed under the guidance of a rehabilitator. The curriculums aim at three aspects of cognitive training, understanding training and imitation training, traditional training contents are fused into the system, and an interesting, immersive and game type interactive training system is constructed. According to the system,expensive mixed reality equipment is not needed, and more interests, higher concentration and better training performance of autistic children can be caused.

The invention discloses a method for migrating a generative adversarial network by using adversarial learning and discriminative learning. The method comprises the following steps: S1, preparing a picture data set; s2, constructing a pre-trained GAN model; s3, constructing an ADT-GAN model through parameter migration; and S4, training the ADT-GANc. The ADT-GAN model initializes a generator and a discriminator through parameter transfer on the basis of a pre-trained GAN model trained by a source domain image data set by using transfer learning. A domain discriminator is added, a total objective function composed of an adversarial objective function and a domain discrimination objective function is optimized to drive a generator to generate image data of a target domain, and negative migration is avoided. Therefore, the training performance on a small target domain data set is improved, the number of iterations is reduced, and the image generation quality is improved.

The invention discloses a pure end-to-end deep reinforcement learning for training car racing game AI bot that uses only the velocity information extracted from screen for both training and testing phases without using any internal state from game environment, such as the car facing angle. The learned AI bot can play better than the average performance of human players. In addition, the reward function is designed to consist only the velocity value, and use Ape-X distributed training framework combined with a variant of Deep Q Network to solve the sparse training signal problem caused by the reward function of an original design. Moreover, limit learner rate method is designed that improves the training efficiency and training performance. The AI bot trained in this way can achieve performance beyond the average human level and reach a level close to professional players.

The invention provides an exercise training method and system, intelligent glasses and an exercise training device. The exercise training device comprises the exercise training system and the intelligent glasses used for exercise training. The exercise training system is arranged in an intelligent mobile terminal and comprises a GPS positioning module, an exercise data obtaining module of an exercise user, a storage module, a calling module, a track positioning module and a data sending module. The intelligent glasses are used for comparing and displaying movement of the exercise user and a simulated opponent according to the exercise data of the exercise user and exercise data of the simulated opponent. The user wears the intelligent glasses and a smart mobile phone, and the position of the user can be located; meanwhile, a figure can be simulated and displayed in the glasses through the imported data, exercise speed difference, ranks and positions of simulated figures are displayed according to the data, one or more simulated opponents can be introduced during exercise of the user for assisting the use, and the user is helped to improve exercise training performance and exercise pleasure.

Embodiments of the disclosure may provide a method, system, and computer program configured to simulate a teaching environment for educational training. The method may include displaying an empty simulated classroom environment for an orientation period, displaying an occupied classroom environment for at least one classroom period, generating and displaying a plurality of events, both disruptive and non-disruptive during the at least one classroom period, identifying at least one of the plurality disruptive or non-disruptive events on a display, addressing the at least one identified event, scoring training performance for the simulated teaching environment based on the identifying and addressing actions, and using a score representative of the training performance to add enhancements to a subsequent teaching simulation.

The invention relates to a training data analysis and acquisition system which comprises at least one data acquisition instrument, at least one tablet computer which is used separately used with the data acquisition instrument, and a data processing center. The data acquisition instrument comprises a data acquisition module, a data processing module and a WIFI communication module. The tablet computer receives a digital signal sent by the data acquisition instrument, and then transmits the digital signal to the data processing center through a wireless network. The data processing center is composed of a central processing unit, a data acquisition and storage module, a training task transmission module, a communication module, an encoder, a power supply interface unit, a data analysis module, a training plan simulation module and a touch control panel. According to the invention, the training data analysis and acquisition system can accurately and efficiently acquire the change of the training performance of soldiers; the training efficiency can be improved; data mining and analysis are of high value; and the system is highly targeted.

The invention relates to the field of artificial intelligence, and discloses a method and a device for generating sample data. The method comprises the steps of acquiring an initial predictor model, andconducting the iterative optimization on the predictor model through multiple rounds of iterative operation, wherein the predictor model represents the relation between training sample data and theperformance of a neural network model trained based on the training sample data; in response to determining that the iteratively optimized predictor model reaches a preset convergence condition, generating target sample data by using the iteratively optimized predictor model; wherein the iterative operation comprises the following steps: predicting training sample data corresponding to a neural network model with preset performance as current sample data by adopting a current predictor model; training a preset neural network model based on the current sample data, and obtaining the actual performance of the trained preset neural network model; and updating the parameters of the predictor model according to the deviation between the preset performance and the actual performance. According to the method, the sample data of the neural network model with good training performance can be obtained.

The invention relates to the technical field of artificial intelligence, and provides a target identification method and device based on artificial intelligence, electronic equipment and a medium. The method comprises the following steps: acquiring the incidence relation between customers, the label relation between products, and the affiliation relation between customers and preset enterprises; and establishing an initial knowledge graph with the clients as the nodes on the basis of the incidence relation, the label relation and the affiliation relation, wherein the information of the initial knowledge graph is rich, each node in the initial knowledge graph comprises a plurality of attribute features, and therefore the expression ability of the nodes is high; and then performing weight optimization on the initial knowledge graph according to a preset weight optimization strategy to obtain a target knowledge graph. When a graph convolutional neural network is trained based on the target knowledge graph, the training performance of the graph convolutional neural network can be improved, so that the prediction probability of the graph convolutional neural network is improved. The plurality of to-be-identified objects are identified through the trained graph convolutional neural network, so that the target object with relatively high identification accuracy is obtained.