194results about How to "Improve training effect" patented technology

The application discloses a differential privacy data publishing method based on risk self-adaption and structured Gaussian noise and relates to the technical field of data mining and privacy protection. The method comprises the following steps: preprocessing original table data and initializing a Rényi differential privacy accountant; calculating sample risk based on self-information and mutual information, and evaluating individual identification and structured correlation risk; constructing a structured dependence relationship between features by using a total variation distance and a spectral decomposition dependent projection algorithm, and generating a noise transformation matrix; constructing a synthetic data pool without real individual mapping based on a noisy edge distribution; and finally performing risk self-adaptive data cleaning and publishing. In the publishing stage, a dynamic camouflage mechanism is implemented according to the distance between the risk score and the sensitive threshold: structured noise is injected into high-risk samples to protect feature dependence, K-nearest neighbor camouflage replacement is performed on risk boundary samples by using a synthetic data pool, and independent noise is injected into low-risk samples. The application effectively retains the statistical correlation of data under the premise of strict privacy protection by fine risk measurement and structured noise injection, and significantly improves the utility of data in machine learning modeling.

The application discloses an age classification model generation method, an age classification method and system, and a training method. i Input into a face feature extraction network to obtain a face feature map f i ; the face feature map f i is input into a feature discrimination module D f for first optimization processing; the face feature map corresponding feature vector is input into a first full connection layer for second optimization processing; and iterative training is performed to obtain an age classification model. In terms of training data, the weight setting of data sampling is performed on face images of different age stages, and the data proportion of different age stages is balanced; in model optimization, a feature discrimination module and a ViT module are introduced to reduce the ambiguity information of face images of different age stages, and accurate and stable prediction is realized.

The utility model discloses a kind of intelligent lower limbs reciprocating type strength training equipment, including training frame, the top of training frame is fixedly connected with cushion, the side of cushion is fixedly connected with movable frame, the side of movable frame is installed with backrest, the top of training frame is away from the side of cushion and is equipped with mounting groove, the top of training frame is movably connected with sliding frame by mounting groove, the front of training board is equipped with sliding slot, the front and back of sliding frame are fixedly connected with displacement sensor, the inner wall of mounting groove is equipped with moving groove, the inner side of mounting groove is fixedly connected with tension sensor, the side of tension sensor is fixedly connected with telescopic link, one end of telescopic link is fixedly connected with tension spring. Through the tension sensor, telescopic link and displacement sensor set, improve training effect, so that lower limbs reciprocating type strength training equipment can be intelligently adjusted according to the power change of trainer, avoid the damage caused to trainer in training process.

The application relates to the technical field of remote sensing image detection, and discloses a hyperspectral remote sensing image anomaly detection method, a system, equipment and a storage medium. The method comprises the following steps: acquiring a hyperspectral remote sensing image to be detected; inputting the hyperspectral remote sensing image to be detected into a pre-trained anomaly detection model to perform anomaly detection, and obtaining a detection result of the hyperspectral remote sensing image to be detected; and the anomaly detection model comprises a first convolution module, a second convolution module, a first Transform module, a first Resize layer and a first full connection layer which are sequentially connected. The application detects background and abnormal samples through clustering inspiration, effectively reduces the dependence of a network model on artificial labeling samples, and not only distinguishes abnormal information and background information during the training of the anomaly detection model, but also utilizes a reconstructed image of the hyperspectral image to improve the training effect of the model, so that the whole model has a local and global receptive field, and the detection precision of the anomaly detection model is effectively improved.

The present application relates to an online batch data processing method and device for training deep neural networks, the method comprising the following steps: sampling a data batch from a given dataset; selecting a subset S from the data batch by using an orthogonalization-based objective function; training a target deep neural network based on the subset S. Compared with the prior art, the present application uses an orthogonalization method to remove the redundancy between subsets, provides an objective function based on orthogonalization de-redundancy correlation to measure the diversity and representativeness of the subsets at the same time, and provides an efficient optimization method based on the objective function and a corresponding data selection algorithm, which can efficiently select diverse and representative subsets.

This invention provides a clustered federated multi-task learning method and apparatus for the Internet of Things (IoT). By clustering IoT terminal devices, the data distribution within the same cluster becomes more approximate. A federated multi-task learning algorithm is executed within each cluster, with global training and personalized training tasks performed on each IoT terminal device. This achieves data sharing within the cluster while fully utilizing local data from each IoT terminal device for training on personalized tasks, thus efficiently utilizing local data and improving training effectiveness. During local training on each IoT terminal device, the number of training rounds is adjusted based on computing power, fully utilizing the computing resources of each IoT terminal device and improving model training efficiency. Regularization constraints applied to personalized training tasks using the global model effectively prevent overfitting, control the degree of personalization, and improve model quality.

The application provides an image processing method and device, a computer readable storage medium and an electronic device, relates to the technical field of image processing, and can cut a first image corresponding to a first color depth into a second image corresponding to a second color depth, and normalize the second image into a sample image of the first color depth to simulate a low dynamic range image. Application of such a sample image in model training can facilitate the model to learn the problem of the low dynamic range image, thereby avoiding the model from generating an image with a highlight edge smear problem.

The application discloses a Spike-BP on-chip learning method and system based on a Ca-LIF neuron model and a processor, wherein a trainable linear leakage parameter and a positive and negative double-channel pulse transmitter mechanism based on calcium gating are adopted; for the Ca-LIF neuron model, a linear leakage operation can be realized only by subtraction, complex leakage compensation operations can be avoided when a gradient matrix is solved, hardware implementation complexity is reduced, a large amount of calculation resources of subsequent hardware design is saved, and the performance of pulse neural network training is improved.

The method and related device for predicting component performance based on data-driven generalization enhancement include determining original model input features for predicting component performance; converting the original model input features into basic factor form of an empirical formula according to the structure of the empirical formula; collecting a measured data set for predicting the performance of the structure component, and converting the measured data set into the basic factor form of the empirical formula to establish a data set of the measured data; selecting a representative sampling strategy to generate a representative data set; based on the representative data set, performing model training on a classic data-driven model to obtain a pre-training model; introducing the measured data set into the pre-training model, and using incremental learning to retrain the model to obtain a structure component performance prediction model based on data-driven generalization enhancement, and performing prediction. The method proposed in the application can be used to assist in structure performance evaluation, residual service life prediction, durability analysis and structure maintenance work.

Embodiments of the present application disclose a text recognition method and device, computer equipment and a computer readable storage medium. The method comprises: obtaining a text image sample; performing image index calculation according to image attribute information of the text image sample, and determining a reference sample index based on a calculation result; performing image feature extraction processing on the text image sample by using a feature extraction model to obtain image feature information; performing attention feature extraction based on the image feature information by using the feature extraction model to obtain attention feature information of a context information of interest; predicting a prediction sample index based on the attention feature information; and training the feature extraction model according to the prediction sample index and the corresponding reference sample index, so as to extract the attention feature information of a to-be-recognized text image by using the trained feature extraction model to perform image text recognition. The method can train the feature extraction model by using a large number of unlabeled text image samples, and can enhance the training effect of the feature extraction model.

The utility model discloses a kind of shock-absorbing martial arts training pile, it is related to martial arts training instrument technical field, including pile body, pile hand and pile leg, the bottom of pile body is provided with rubber pad, the number of rubber pad is four groups, four groups rubber pad enclose circle, the bottom outer wall of four groups rubber pad is all set up with butt joint hole, the inside of butt joint hole is clamped with butt joint block, butt joint block and butt joint hole cross section are regular hexagon, the end of butt joint block away from butt joint hole is fixedly connected with clamping seat, the top of rubber pad is provided with buffer assembly and be used for the locking assembly of the fixed pile body, the utility model can play good shock-absorbing effect in the process that people hit training pile, avoid the situation that fixed bolt appears loosening in the process that wooden dummy is used for a long time, guarantee the stability in the process that training pile is used for a long time.

The application discloses a health baseline construction and fault diagnosis method based on a twin residual network, and comprises the following steps: dividing a diagnosis signal into three parts of a pre-training data set, a training data set and a test data set, and performing normalization processing; constructing a twin residual network, and training a residual network feature extraction part in the twin residual network by using the pre-training data set to form a health baseline model; taking out the residual network feature extraction part in the health baseline model to construct a residual network diagnosis model, and training the residual network diagnosis model by using the training data set; and performing fault diagnosis test on the trained residual network diagnosis model by using the test data set, and evaluating network performance.

This invention discloses a smart production training system and method for facility horticulture based on digital twins, belonging to the field of smart production training technology. It includes: a smart horticulture production training module, a virtual twin module, a smart control module, and a training and intelligent evaluation module. The invention uses the smart horticulture production training module to collect environmental, equipment, and crop growth data and execute control commands; the virtual twin module constructs a one-to-one mapped virtual three-dimensional scene, realizing real-time virtual-real mapping between physical entities and the virtual scene; the smart control module generates control strategies based on simulated data and training conditions and issues control commands; the training and intelligent evaluation module provides operation guidance, visual display, data recording, and automated scoring and comprehensive evaluation, improving the safety of smart production training in facility horticulture and solving the problems of low safety in existing technologies due to long cycles, high costs, slow rotation, and high risks.

This invention discloses a method, apparatus, electronic device, and storage medium for training an intelligent agent policy network. The method includes: determining the linear score of a first training sample; inputting the current first training sample into a state-action value network, a state value network, and a policy network respectively, and determining a first actual output, a second actual output, and a third actual output; determining a first loss value based on the second actual output, the cumulative reward at a first historical time, the linear score, and a first loss function, and adjusting the parameters of the state-action value network; determining a second loss value corresponding to the state value network, and adjusting the parameters of the state value network accordingly; determining a third loss value corresponding to the policy network, and adjusting the parameters of the policy network based on the third loss value, to obtain the target policy network. This technical solution improves the training effect and speed of the policy network, enabling more accurate and convenient training of the target policy network.

The utility model discloses an upward stacking type detachable plant cultivation container which comprises a detachable base pot arranged at the bottom and at least one detachable auxiliary pot which is clamped on the detachable base pot through a plurality of clamping assemblies and is sequentially and upwards stacked and combined. The detachable auxiliary basins are circumferentially and uniformly fixed at the bottom of the detachable auxiliary basin and are clamped with the upper edge of the basin opening of the detachable base basin; when the number of the detachable auxiliary basins is multiple, the detachable auxiliary basins at the upper positions are clamped to the upper edges of the basin openings of the detachable auxiliary basins at the lower positions through the clamping assemblies. The plant cultivation container is designed according to the growth state requirements of plants such as flower candle type rhizomes and monstera deliciosa type climbing plants, the overlaid structure of the base pot and the auxiliary pots is adopted, the plant cultivation container is expanded upwards, vertical growth of the plants is facilitated, attractive scenery is easier to form, the detachable local structural design is adopted, and the plant cultivation container is convenient to use. Installation, expansion, pot replacement and detachment are facilitated, and roots are not clamped or damaged.

This disclosure provides a model training method, an advertising effectiveness evaluation method, an apparatus, and equipment, which can be applied to the fields of artificial intelligence, the Internet, advertising, and model training. The model training method includes: acquiring training samples; using the feature extraction module of the model to be trained to extract features from the user attribute information of the samples to obtain target sample features; using the evaluation module of the model to be trained to evaluate the target sample features to obtain an evaluation result representing the degree of interest of the user samples in the advertisement; based on reference features, identifying feature differences in the target sample features to determine the feature alignment loss value of the model to be trained; and adjusting the parameters of the model to be trained based on the feature alignment loss value and the evaluation loss value to obtain the trained model.

The utility model discloses a kind of front landing gear torsion arm dismounting practical operation rack, it includes fixed support, operating plate, torsion arm practical operation model and torsion arm pin assembly on the torsion arm of aircraft front landing gear, the upper surface of fixed support is fixed with two operating plates, the lower of operating plate is equipped with fixed tray, torsion arm practical operation model is respectively equipped in the front of operating plate, the connection between torsion arm practical operation model and operating plate can be disassembled, the front end of torsion arm practical operation model is equipped with left and right torsion arm, torsion arm pin assembly is connected on each torsion arm practical operation model left and right torsion arm. Through the utility model, the entire dismounting process of torsion arm pin assembly of aircraft front landing gear torsion arm can be carried out practical operation training.

This disclosure relates to the field of computer science, providing a method, apparatus, electronic device, and medium for training a multimodal contrastive learning model. The method includes: acquiring multimodal data including text data and image data; processing the multimodal data to obtain serialized marker data corresponding to the text data and image patch data corresponding to the image data; inputting the serialized marker data and image patch data into a feature encoder to obtain initial text features and initial image features; inputting the initial text features and initial image features into a preset routing-based expert layer to obtain multimodal features; inputting the multimodal features into an encoding layer to obtain encoded text features and encoded image features; and training a contrastive learning network based on the encoded text features and encoded image features to obtain a trained multimodal contrastive learning model. This disclosure allows a single model to handle multiple single / multimodal tasks simultaneously, avoiding catastrophic forgetting problems and improving work efficiency and training effectiveness.

This invention provides a method and system for accurate tumor segmentation in CT images, relating to the field of image processing technology. The method includes: determining the region to be identified in a CT image containing a tumor using an image segmentation model; determining the boundary line, center position, and nearest reference pixels of the region to be identified, along with their probability information; and then selecting correction pixels from the nearest reference pixels to obtain a correction region, thereby updating the trained image segmentation model. According to this invention, after obtaining the region to be identified in the CT image of each slice using the image segmentation model, the boundary line of the region to be identified can be mutually verified based on the nearest pixels on the boundary line and the pixels of adjacent slices. This allows for the selection of pixels that better represent the tumor edge from the nearest pixels on the boundary line, thereby achieving more accurate segmentation of the region containing the tumor.

This application discloses a federated learning method, apparatus, device, and storage medium, relating to the field of privacy computing technology. Applied to a federated system composed of multiple participants, the method includes: acquiring the local model parameters of each participant in the federated system; iterating Round training through a local base network and an autoencoder's encoding / decoding layer, and determining the loss value corresponding to the completion of multiple epochs in each Round training process using a preset loss function; sending the local model parameters and loss values ​​to a secure aggregation server, receiving the global parameters and convergence status returned by the secure aggregation server, updating the local model parameters according to the global parameters, and stopping Round training when the convergence status meets preset conditions. The technical solution of this application can improve the learning effect of federated learning in scenarios where the participant data is not independent and identically distributed, and also improve training speed.

This invention belongs to the field of motion data processing technology and provides a tennis motion posture detection method based on a six-axis sensor. The method includes acquiring data including linear acceleration, angular velocity, and rotation angle using a six-axis sensor fixed to the racket; then, processing the acquired data through an improved Kalman filter, zero-bias compensation, and variance thresholding to remove anomalies, obtaining clean linear acceleration and angle data; subsequently, determining the initial posture using a stationary segment, preserving the original vibration characteristics, and performing a quadratic integration of the linear acceleration to achieve three-dimensional trajectory recovery; finally, extracting signal fluctuation features from the trajectory tangent vector, and inputting the feature sequence into a neural network to classify and identify the completion degree of the swing action. This method recovers the racket's motion trajectory during a tennis player's movement, uses a neural network to analyze the motion posture, and judges the completion degree of the hitting action through feature information extraction, helping trainees improve training effectiveness.

The invention provides an intelligent visual perception training method and system based on a micro-view and a Gabor sighting mark, a medium, a program product and a terminal, and the method can generate personalized initial training regions and training sighting marks for different patients through importing a micro-view inspection result and combining with the positioning of a residual function region and micro-view sensitivity, thereby improving the training efficiency. Therefore, the training effect is obviously improved. Meanwhile, stimulation parameters are dynamically adjusted according to patient performance, so that the training content is always matched with the visual ability level of the patient. The characteristics of controllable spatial frequency, direction, contrast ratio and the like of the Gabor sighting target are utilized to construct a multi-level task from easy to difficult, and the diversity and suitability of training are improved. The data-driven weight mechanism not only enhances the scientificity of training, but also provides visual rehabilitation progress reference for clinicians. According to the method, the precise positioning capability of the micro-view and the multi-dimensional stimulation advantage of the Gabor visual target are fully utilized, and the rehabilitation effect and compliance of the low-vision patient are remarkably improved by combining the precise positioning capability with the multi-dimensional stimulation advantage of the Gabor visual target.

The invention discloses a DR focus segmentation method and system based on a dynamic loss optimization function, and the method comprises the steps: collecting image data in an IDRiD data set, and carrying out the preprocessing of the image data, and obtaining the preprocessing data; based on the preprocessed data, a visual SAM large model is constructed, a loss function weight optimization strategy is designed, and an RTSAM segmentation model is obtained; segmenting the DR eye fundus image based on an RTSAM segmentation model to obtain a DR focus segmentation result; and based on the DR focus segmentation result, providing an automatic identification and segmentation visualization result of the DR focus. Aiming at the problems of non-uniform lesion distribution, non-uniform lesion occurrence frequency and the like existing in a DR color fundus image, the segmentation effect is poor, so that a dynamic loss optimization strategy is used, weight distribution in a loss function is adaptively adjusted according to a model segmentation index, and the attention of a model to rare lesion and small lesion areas is enhanced.

The application discloses a skin type prediction method for small sample data autonomous learning, and relates to the technical fields of deep learning and medical beauty skin type. The initial model is obtained by a convolutional neural network through a small number of labeled samples, and the subsequent new samples are accepted through the initial model, and whether the new samples have labeling value is judged according to the prediction result of the model, so that only the samples determined to be inaccurate by the model are selected for labeling and retraining. Compared with the existing skin type method, a single model can be used to obtain a more perfect result, the model is continuously iteratively optimized by using the enhanced unit in the actual use process, and the idea of active learning is used to select the samples with greater uncertainty for labeling, so that the number of samples needing expert labeling can be greatly reduced, and the labor cost is reduced. According to the reliability module, when the samples are expanded, the skin type with insufficient reliability is preferred, so that the final typing effect is improved.

The application provides a model training method based on skeleton action recognition, a skeleton action recognition method, a skeleton action recognition device and a computer readable storage medium. The model training method comprises: using a global information modeling module of a skeleton action recognition model to perform convolution operation on first local features, second local features and third local features in other feature dimensions respectively to obtain first global features corresponding to the first local features, second global features corresponding to the second local features and third global features corresponding to the third local features; fusing the first global features, the second global features and the third global features to obtain global features, and fusing the local features and the global features to obtain fused features. In the above manner, the skeleton action recognition device can comprehensively and effectively mine the global information of the skeleton data space-time domain through the novel multi-view global information modeling module, and improve the training effect and recognition effect of the skeleton action recognition model.

The application discloses a federal automatic driving differentiated privacy protection method, an electronic device, a storage medium and a program product. The method comprises the following steps: determining a priority interval according to the priority of batch samples, and dividing the priority interval into E segments; sampling a value from each segmented interval range, finding a sample corresponding to the priority closest to each sampled value from the batch samples, finding another sample closest to the priority of the found sample from the same type of samples in which the found sample is located, and combining the two found samples to form a similar sample combination; according to the priority of the sample, the total number of batch samples and the training correction factor, each sample of the batch samples is given a non-normalized weight, and each non-normalized weight is normalized to obtain the maximum weight of each sample in the batch samples. The application realizes fine privacy protection for key gradients, and improves sample utilization and model training performance.

The application belongs to the technical field of medical treatment and can be used for sample data enhancement in the scene of human body, tissue organ modeling and the like in the medical field, and particularly relates to an image data enhancement method, a training method and device of a binocular stereo matching model. The method comprises the following steps: obtaining a target sample image pair, the target sample image pair comprising a first image and a second image; generating a first random mask and a second random mask; obtaining a first disordered image and a second disordered image; obtaining a first data enhancement image based on the first random mask, the second disordered image and the first image; obtaining a second data enhancement image based on the second random mask, the first disordered image and the second image; and combining the first data enhancement image and the second data enhancement image to obtain a data enhancement sample image pair. The above method and device improve the adaptability of the introduced noise to the training sample set, so that the expanded training sample has better training effect.

A target vehicle for irregularly moving shooting training systems, belonging to the field of target vehicle technology, includes a mobile target vehicle, a target plate system, a target plate fixing mechanism, and a target plate mounting base. The target plate mounting base is slidably installed inside the mobile target vehicle. Multiple target plate systems are rotatably installed on the target plate mounting base, and the target plate fixing mechanism is fixedly installed on the target plate mounting base. Each target plate system includes a rotating unit, a fixing unit, a pushing mechanism, a locking mechanism, a telescopic sleeve, a spring, a fixing clamp, a target plate, a drive gear, a ratchet, and a limiting shaft. The rotating unit is rotatably installed on the fixing unit, and the pushing mechanism is installed in the groove of the fixing unit. This angle is the target plate ejection angle. Then, the fixing base plate of the target plate system is slidably installed on a hexagonal prism, which realizes the ejection of the target plate at different angles, achieving different shooting angles, thereby improving the training effect.

The application relates to the field of balance training, and discloses a balance training method, device and system of a balance board and a computer readable storage medium. The balance training method of the balance board comprises the following steps: outputting an instruction instruction indicating that a trainer treads the balance board according to a plurality of different reference orientations; collecting a maximum inclination angle corresponding to each reference orientation; selecting a plurality of training tread directions and setting an inclination angle coefficient corresponding to each training tread direction; determining an angle threshold corresponding to each training tread direction according to the maximum inclination angle corresponding to each reference orientation, so as to determine a set inclination angle; and outputting a corresponding training prompt according to the training tread direction and the corresponding set inclination angle, so that the trainer performs balance training according to the training prompt. In the application, information about the tread ability of the trainer treading the balance board is obtained, so that the training target of the trainer is reasonably set, the effectiveness and rationality of the balance training are ensured, and the training effect and training experience are improved.

The utility model relates to the technical field of medical sickbeds, and particularly discloses a turning-over assisting bed which comprises a bed board frame. The bed body backup plates are arranged on the two sides of the bed board frame; the limiting baffle is mounted at the top end of the bed board frame; through the arrangement of the first rotating rod, the bed board body, the fixing base and the tension spring, when a hemiplegic patient and a paraplegic patient are subjected to turnover rehabilitation training, rotating operation of the bed board can be achieved through the first rotating rod and the second rotating rod which are installed on the bed body backup plate, and in the process, elastic stretching of the tension spring is utilized to enable the patient to turn over. The resistance of the bed board during rotation can be effectively increased, by means of the arrangement, when a patient lies on one side, the limb twisting capacity and resistance resisting capacity of the patient can be fully exercised, then the patient recovers to the prostration state, turning-over training becomes more daily, and by means of the design, the training pertinence is improved, and the training effect of the patient is also improved.