44 results about "Simultaneous learning" patented technology

The invention discloses a coal mine disaster early warning method based on multi-source data fusion and a time-space diagram network. The method comprises the following steps: (a) collecting multi-source heterogeneous monitoring data; (b) constructing a geologic-physical dual constraint anisotropic space-time diagram; (c) training the graph by using a geologic-physical anisotropic space-time graph neural network, wherein a graph convolutional layer of the network is designed into a propagation mode which is learned and defined at the same time; and (d) predicting a disaster risk by using the trained model and generating an early warning. According to the method, geological and physical priori knowledge is taken as constraints to be incorporated into the topological structure of the GNN model, so that the model can follow physical laws, disaster propagation under anisotropic geological conditions is predicted, and the accuracy and timeliness of early warning can be improved.

The invention provides a micro-service anomaly detection method based on multi-modal feature fusion, and belongs to the technical field of computers. The method comprises the following steps: firstly, performing semantic feature extraction on a log template by combining BERT and a knowledge distillation technology, and learning category features of the log template while enhancing log semantic information; secondly, performing time sequence feature extraction on the log template sequence and the KPIs data by using LSTM and TCN respectively; and then, designing a double-path cross-modal Transform module, respectively extracting collaborative information and redundant shared information between the log and the KPIs, and carrying out weighted fusion through an improved gating unit to obtain a final fusion feature vector. And finally, reconstructing the fusion feature vector by using a diffusion model, calculating a reconstruction error, and comparing the reconstruction error with an abnormal threshold value to realize abnormal detection. Experimental results show that the micro-service anomaly detection method is superior to the existing mainstream method in various indexes of micro-service anomaly detection, and has relatively high detection precision and robustness.

A method for training a student model based on a pre-trained language model (PLM) and a text classification system are disclosed. The method includes: constructing cue-based training samples; adjusting the pre-trained language model using the cue-based training samples to obtain a cue-adjusted teacher model; and training the student model using the processed training samples, wherein during training, the student model simultaneously learns the classification probability vectors output by the cue-adjusted teacher model and the original teacher model. This invention requires the student model to learn from two teacher models simultaneously, thereby alleviating the overfitting problem of the student model in small-sample scenarios by adding a distillation path that learns from the original PLM teacher model with unsupervised data. Furthermore, by transferring the intermediate layer representation of the PLM through knowledge probes and stabilizing the performance of knowledge distillation through comparative learning, the student model can learn higher-order dependencies from the intermediate layer representation of the teacher model, improving the accuracy and efficiency of knowledge distillation.

The application discloses a sintering endpoint intelligent sensing method based on a 3D convolution network and belongs to the field of industrial process soft measurement modeling. The prediction model utilizes a space-time 3D convolution network to simultaneously learn space-time features hidden in data, and adopts an encoding-decoding network to perform multi-step prediction on a present sintering endpoint. Firstly, a random permutation and combination method is used to convert a two-dimensional data structure into a three-dimensional format. Then, a 3D convolution is used to extract time features and space features in the data. Next, a space-time feature calibration module is proposed to learn the importance of different features and finely process the features. Finally, the fine space-time features are input into the encoding-decoding network, and a dynamic multi-step prediction loss function is used to realize multi-step prediction on the sintering endpoint, so that the purpose of intelligent sensing is achieved. Related data collected in a sintering plant in South China verifies the effectiveness and feasibility of the method.

A method of performing target detection during autonomous driving, comprising: performing 3D target detection in a 3D target detection segment; uploading outputs of a plurality of sensors in communication with the 3D target detection segment into a plurality of point clouds; transmitting point cloud data of the plurality of point clouds to a region proposal network (RPN); independently performing 2D target detection in a 2D target detector, the 3D target detection being performed in parallel in the 3D target detection segment; and obtaining a given input image and simultaneously learning bounding box coordinates and class label probabilities in a 2D target detection network that operates to treat target detection as a regression problem.

The application discloses an image classification method based on adaptive local ordinal preserving analytic dictionary learning and belongs to the technical field of computer vision. The method comprises the following steps: performing feature extraction on images in a data set, dividing the images into a training set and a test set, adopting a support adaptive local ordinal preserving analytic dictionary learning model to simultaneously learn an analytic dictionary and a classifier based on the training set, then calculating coding coefficients of the test set based on the learned analytic dictionary, and finally obtaining class labels of the test set through the classifier and the coding coefficients of the test set. The method introduces an adaptive ordinal local preserving term based on a discriminative convolution analytic dictionary learning model, simultaneously preserves neighborhood correlation between dictionary atoms and distance ordering information of atoms in the neighborhood in the learning process, optimizes the analytic dictionary learning model, enhances the discriminativeness of the dictionary, and improves the image classification accuracy of the model in general scenarios such as face recognition, object recognition and scene recognition.

The invention provides a training method and device of a document processing model, electronic equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence, in particular to the fields of large language models, image processing, document processing, text recognition and the like. Comprising a first sub-data set associated with the identification task, a second sub-data set associated with the question and answer task and a third sub-data set associated with the understanding task, and the data proportion among the sub-data sets is dynamically adjusted in multiple stages of training; and performing multi-task collaborative learning on the document processing model based on the corresponding data proportion in each stage according to the sequence of the multiple stages, so that the document processing model learns an identification task, a question and answer task and an understanding task at the same time. The document processing model can process at least one of an identification task, a question and answer task, and an understanding task by generating a layout inference chain associated with the layout information of the target document.

This invention relates to the field of power distribution technology, providing a power distribution edge intelligent agent, an intelligent fusion terminal, energy storage, and load regulation method. The power distribution edge intelligent agent includes an intelligent algorithm model, which comprises a shared layer and a dedicated layer. The shared layer includes multiple gated recurrent units for acquiring common temporal features among multi-dimensional data. The shared layer simultaneously learns the common temporal features of multiple tasks using a multi-task learning algorithm. The dedicated layer includes multiple task branch modules. The prediction task branch module includes a gated recurrent unit for obtaining the predicted value of the corresponding task based on the common temporal features of multiple tasks. The regulation task branch module includes a gated recurrent unit and a reinforcement learning policy network for obtaining the regulation strategy of the corresponding task based on the common temporal features of multiple tasks. The reinforcement learning policy network is updated using a near-end policy optimization algorithm. This invention enables the coordinated operation of power distribution system sources, grid, load, and storage.

The invention discloses a face analysis method based on multi-task learning and transformer, and the method comprises the following steps: dividing an input image into a plurality of non-overlapping windows, and carrying out the feature transformation through a multi-layer perceptron; tasks are classified according to output types, each prediction head has an independent and unique feature fusion module and a decoding module and can learn a plurality of sub-tasks at the same time, finally, corresponding losses are calculated, and a gradient normalization strategy is adopted to ensure that a model does not tend to each task. The invention provides a face analysis architecture based on multi-task learning and transformer, a plurality of related or unrelated tasks are allowed to be processed in a unified model at the same time, and feature representation is shared among different tasks, so that redundancy of feature extraction is reduced, and the resource utilization efficiency is improved. By dynamically calculating the learning weight for different tasks, the multi-task learning framework can synchronize the learning speed of each sub-task to the greatest extent, and an effective convergence effect is achieved.

The invention discloses an intelligent contract vulnerability detection method based on large model reasoning enhancement, and the method comprises the steps: firstly generating the data of a CoT (Chainof Though, CoT) containing a logic derivation path through employing a Teacher Model, and carrying out the consistency verification and cleaning through combining with a static analysis tool (Slitter); then, a multi-task learning framework is constructed, and supervised fine tuning (SFT) is carried out on a Student Model, so that the Student Model learns vulnerability classification and logical reasoning at the same time; and finally, a direct preference optimization (DPO) technology is adopted, and alignment calibration is carried out on the model strategy according to the constructed preference data set. The method is suitable for security automatic auditing of various blockchain smart contracts, and can improve the detection precision of complex logic vulnerabilities while maintaining high detection efficiency.

This invention provides a wind farm power prediction method based on data distribution fitting. When constructing a weighted triple attention mechanism bidirectional GRU neural network model, the method first estimates the mixture Gaussian distribution of the fitting error using maximum likelihood estimation. The error distribution parameter θ is then embedded into the error loss function of the PINN neural network model, enabling the PINN model to simultaneously learn data patterns and error distribution characteristics. This solves the problem of mismatch between distribution assumptions and actual wind farm data characteristics in traditional methods, improving the accuracy of wind farm power prediction. Then, the output layer of the triple attention mechanism model is connected to the bidirectional GRU layer of the PINN neural network. Based on triple attention-bidirectional GRU dynamic feature selection, dynamic feature enhancement of the input wind turbine characteristic data is achieved. The wind farm power prediction accuracy using the weighted triple attention mechanism bidirectional GRU neural network model constructed in this invention is high, and it solves the problem of model response lag during sudden wind speed changes in traditional methods.

The invention relates to the technical field of radar target detection, in particular to an HFSWR target intelligent detection method based on background subspace projection filtering and a deep learning network, and the method comprises the following steps: S1, carrying out the unbiased covariance estimation and feature decomposition on a multi-channel distance-Doppler spectrum through employing a protection-reference window, and obtaining an HFSWR target; constructing a local clutter main subspace and projecting the whitening snapshot to obtain an energy tensor changing along with a subspace rank k; according to the method, array phase information is effectively utilized by adopting background matching subspace projection, the separation degree of target and non-target components is remarkably improved, projection energy of different ranks is stacked into a three-dimensional tensor according to channels, a frame of distance-Doppler spectrum is expanded in a distance-Doppler-rank three-dimensional space, and the distance-Doppler-rank three-dimensional space is expanded. And the deep network can learn the distance-Doppler spectrum structure features and the cross-channel complementary features under the multi-rank hypothesis at the same time.

The application discloses a large model security reinforcement method and system of category adaptive reinforcement learning, and the method comprises the following steps: dividing safety risk data obtained from a large model security evaluation system into multiple safety categories, generating corresponding safety responses through an auxiliary large language model, and constructing training data sets of the safety categories; constructing a loss function by comparing original large model answers with safety responses in the training data sets of the safety categories, training a corresponding reward model by minimizing the loss function; obtaining an advantage function based on the reward model corresponding to each safety category and a reward scaling factor thereof; constructing an overall objective function based on the advantage function, and alternately and iteratively optimizing the objective functions of the safety categories in the large model training process, so that the strategy model can simultaneously learn and align safety preferences of multiple safety categories, and finally obtain a large model optimized by multiple safety categories. The application enhances the safety and robustness of the model.

The invention discloses a hafnium oxide-based ferroelectric material performance prediction method based on machine learning. The method comprises the following steps: acquiring physicochemical property data and process parameters of each doping element of a target hafnium oxide-based ferroelectric material; calculating an average factor and a variance factor of each physicochemical property of each doped element based on the physicochemical property data of each doped element; inputting the average factor, the variance factor and the process parameters into a pre-trained target performance prediction model, and outputting a performance prediction result for the target hafnium oxide-based ferroelectric material; wherein the target performance prediction model is constructed based on an automatic machine learning framework and is obtained by adopting a pre-constructed target data set to carry out supervised training. According to the method, the process parameter characteristics of the hafnium oxide-based ferroelectric material and the physicochemical property characteristics of the doped elements are jointly input into the model, so that the model can learn the complex nonlinear relationship between the physical intrinsic attribute and external process regulation at the same time, and the prediction precision and generalization ability of the model are improved.

The present disclosure provides a training method of an operation and maintenance decision model, which comprises: obtaining first resource scheduling information of a reference application service in a first time period, and performing feature extraction on the first resource scheduling information to obtain a plurality of first time sequence behavior features; performing feature enhancement on the plurality of first time sequence behavior features to obtain first enhanced behavior features; constructing a user portrait of an operation and maintenance operation object performing an operation and maintenance operation according to an operation and maintenance operation history record associated with the reference application service; generating sample input features according to the plurality of first time sequence behavior features, the first enhanced behavior features and the user portrait; and training the operation and maintenance decision model by using the sample input features and labeled operation and maintenance decision information. Thus, the model can learn the objective system state and the user operation preference at the same time, has stronger context adaptation capability in actual deployment, and the output operation and maintenance decision is more consistent with the expectation of the operation and maintenance operation object, thereby improving the accuracy of the operation and maintenance system.

The application provides a global runoff flow prediction method and system based on a state space model, and the method comprises the following steps: S1, obtaining gridded historical meteorological data covering one or more target river basins, and organizing the historical meteorological data into a time-space sequence; S2, forming a feature vector sequence; S3, performing multi-scale space-time coding on the feature vector sequence to generate a multi-scale fused enhanced feature vector sequence; S4, constructing a network based on a state space model; S5, constructing a deep learning model training framework to obtain a trained runoff prediction model; and S6, outputting runoff flow prediction values of one or more stations in a target region at multiple time steps in the future. The application can simultaneously learn multi-scale space and time patterns in hydro-meteorological data, automatically focus on physically relevant regions, improve the ability to capture global-scale complex hydro-meteorological dynamics, and improve the prediction accuracy, interpretability and evaluation consistency of the regional hydrological model.

The invention discloses a data expansion and exoskeleton joint end-to-end torque estimation method based on a diffusion model, and the method comprises the steps: carrying out the normalization processing of time series data from a multi-source sensor, carrying out the fragmentation according to a fixed length, and constructing a training sample with a motion class label; then, training a classifier-free conditional diffusion model by adopting a sample, and simultaneously learning conditional and unconditional denoising mapping relationships in a manner of randomly inactivating category conditions; in the generation stage, based on a classifier-free condition guidance mechanism, multi-modal time series data with specified motion category features are gradually generated from random noise; and finally, fusing the generated synthetic data with real acquired data to train a joint torque end-to-end prediction network, thereby realizing joint torque estimation of input sensor time sequence data. The method improves the prediction precision, generalization ability and stability of the end-to-end torque estimation model in a multi-action and few-sample scene, and has a good engineering application value.

According to the low-voltage distributed photovoltaic power prediction method and device provided by the invention, on one hand, the low-voltage distributed photovoltaic characteristic parameters with insufficient measurement information are extracted, and the feeder line low-voltage distributed photovoltaic installation capacity identification model is constructed through the improved particle swarm optimization, so that the distributed photovoltaic capacity information installed in the feeder line is effectively identified; and on the other hand, the improved bidirectional long-short-term memory neural network model is adopted, so that the forward and backward dependency relationship of the load sequence can be learned at the same time, and a complex time sequence mode can be captured more effectively. And on the other hand, an optimal set of key parameters of the bidirectional long-short-term memory neural network is automatically searched by using an improved particle swarm optimization algorithm, so that the model performance is remarkably improved.

The method for completing missing attributes of users based on a time-series bipartite graph neural network first, pre-processes a time-series user attribute matrix, performs data normalization, and records categorical variables and continuous variables. Then, the time-series user attribute matrix is converted into a time-series bipartite graph structure, and a graph convolutional neural network is used to aggregate and extract node embedding vectors and edge embedding vectors in the graph, and an LSTM is used to fuse node features at different time steps; the graph features of the two end nodes of the predicted edge in the time-series bipartite graph are spliced as the input of an edge prediction model of user nodes-attribute nodes; a plurality of linear layers are used to build the edge prediction model, and a multi-task learning framework is introduced to simultaneously learn classification tasks and regression tasks, so as to flexibly handle categorical variables and continuous variables. Finally, the framework is used to complete missing attributes of users, and the results are written back to the original data. The application also includes a system for completing missing attributes of users based on a time-series bipartite graph neural network.

The application discloses a music feature extraction method, device, equipment and medium, extracts a music label embedding vector from acquired music data, and constructs a KNN neighbor graph according to the music label embedding vector; inputs the music label embedding vector into a preset autoencoder for feature learning to obtain a first learning feature; inputs the KNN neighbor graph and the feature learned by the autoencoder into a preset graph convolutional neural network model for feature learning to obtain a second learning feature; wherein the graph convolutional neural network model is at least used for learning structural information and potential similarity between samples; and obtains a final music feature vector according to the first learning feature and the second learning feature. By using the application, the deep learning network can guarantee the extraction of the characteristics of the data itself, simultaneously learn the structural information and high-dimensional potential features between samples, strengthen the effectiveness of the feature vector, and improve the accuracy of music feature extraction.

This invention provides a dimensional measuring device, a dimensional measuring system, and a dimensional measuring method that enable the measurement of the dimensions of various objects visible in a target image with minimal preparation time. [Solution] This dimensional measuring device 100 includes a learning unit 11 that generates an additional layer that learns the object to be measured 4a using an additional learning dataset L that includes a learning data image 1 showing the object to be learned 1a, a model selection unit 13 that selects a combined model M by combining the additional layer R1 with the segmentation base model B, a boundary line identification unit 14 that identifies the contour of the object to be measured 4a that has undergone segmentation processing as boundary line positions, and a dimension calculation unit 15 that calculates the distance between the boundary line positions of the object to be measured 4a as dimensions. The learning unit 11 is configured to learn the parameters of the additional layers R1 to R3 while fixing the parameters of the segmentation base model B during learning.

The application relates to a clustering method and device for misaligned multi-views, electronic equipment and a storage medium, comprising the following steps: constructing an adjacency graph for each view, and matching the graph structure of the adjacency graph of a non-reference view with the adjacency graph of a reference view through a permutation matrix; inputting each view and the adjacency graph thereof into a first-layer graph convolution network to extract a deep representation of each view; constructing a local adjacency graph of each view based on the adjacency graph and the deep representation of each view, and updating the number of neighbors of the local adjacency graph at least once every training period; inputting the deep representation and the local adjacency graph of each view into a second-layer graph convolution network to extract a clustering indication matrix of each view, training and optimizing the permutation matrix based on a unified loss function, simultaneously learning adaptive weights of the views, and obtaining a clustering result, so that a consistent clustering structure across views can be efficiently learned under the condition that samples are misaligned, and the accuracy and robustness of the misaligned multi-view clustering task are improved.

The invention relates to the technical field of image feature extraction, in particular to an STEM diffraction image self-supervised learning method and system based on multi-loss function fusion, and the method comprises the steps: firstly carrying out the targeted preprocessing of a diffraction image based on physical characteristics; then, a universal base model based on Vision Transform is constructed, a teacher-student framework is adopted, and self-supervised pre-training is carried out by fusing DINO loss, iBOT loss and Gram loss specially used for capturing feature correlation, so that the model learns a global mode and local details at the same time; finally, the pre-training model can be directly used for various downstream tasks such as grain segmentation and orientation recognition. According to the method, a transferable general feature extraction base model is provided for the 4D-STEM field for the first time, and the feature quality and the cross-task generalization ability are remarkably improved.

The application discloses a general commodity sequence representation learning method in a recommendation system, comprising the following steps: S1: using a pre-training language model to encode the associated text of commodities to learn transferable commodity representations; first, using a pre-training language model to learn initial text representations, and converting the semantics of the text to a unified semantic space suitable for a recommendation task through a parameter whitening network and a mixed expert enhanced adapter network; S2: further enhancing the fusion and adaptation between different domain data representations through a sequence-commodity comparison task and a sequence-sequence comparison task; S3: considering two fine-tuning settings, namely induction and transduction, according to whether the commodity labels of the target domain are suitable for use. The commodity sequence representation learning method proposed by the application can simultaneously learn general representations on sequence data in multiple domains and efficiently migrate to new recommendation scenarios such as new domains, new markets and new platforms without sharing users or commodities.

The application discloses a multi-modal mixed fusion method for natural gesture recognition and application thereof, and the method comprises the following steps: 1, acquiring multi-modal data, pre-processing and feature extraction of multi-modal signals; 2, constructing a multi-modal signal mixed fusion model, combining pre-fusion, multi-scale hierarchical feature fusion and post-fusion to enable the model to simultaneously learn specific information of a single mode and interaction information between multi-modes; 3, designing a multi-scale space attention module, fully mining key information of input signals by fusing multi-scale hierarchical features; 4, designing a metric learning loss function, enhancing the discrimination ability of the model to similar gestures; and 5, training the multi-modal mixed fusion model to obtain an optimal natural gesture recognition model. The application can improve the robustness of natural gesture recognition, especially increase the recognition accuracy of the model to a large number of gestures and similar gestures, thereby promoting the popularization and application of a gesture-based human-computer interaction system.

This invention relates to a method and apparatus for training a neural network model on noisy multi-label data. The method includes the following steps: selecting a clean set of samples for each category as a meta-dataset using a sample selection algorithm, and estimating the category-dependent label noise transition matrix; initializing some parameters in the instance feature-dependent label noise transition matrix network using the category-dependent label noise transition matrix; transforming the learning problem into a two-layer optimization problem based on statistically consistent label noise learning loss, and simultaneously learning the instance feature-dependent label noise transition matrix network parameters, data imbalance parameters, and multi-label classification neural network parameters using a meta-learning algorithm. This invention innovatively utilizes a meta-learning algorithm in a data-driven manner to unify the learning of instance feature-dependent label noise transition matrix network parameters, data imbalance parameters, and multi-label classification neural network parameters within a single framework.

The application relates to a robot learning method based on model-based generative adversarial interactive imitation learning, which combines the advantages of model-based reinforcement learning, generative adversarial imitation learning and interactive reinforcement learning to form a model-based generative adversarial interactive imitation learning (MAILDH) method, and solves the problem of slow robot learning speed. Firstly, a forward dynamic model is simultaneously learned in the process of generative adversarial imitation learning, the model is used to generate simulation data to train the generator and discriminator parts in the generative adversarial imitation learning framework, so as to improve the sample utilization rate. In addition, different from the traditional generative adversarial imitation learning which learns from demonstration, the application also combines the judgment of human beings on the behavior of the robot, weakens the restriction of demonstration quality on the learning performance of the robot, achieves the effect of reaching or exceeding the demonstration performance, learns a better control strategy, and can improve the stability of the strategy and be suitable for large and complex task control.

The invention discloses an incomplete alignment multi-view clustering method based on structural consistency comparative learning, and belongs to the technical field of computer vision. The method comprises semantic consistency comparative learning and view consistency comparative learning, the semantic consistency comparative learning realizes implicit class alignment of samples by capturing global semantic consistency information, and the view consistency comparative learning learns consistency information between views while retaining local structure information of the views by hierarchically selecting positive samples. Therefore, an accurate sample corresponding relation is established.

The application discloses a learning index model facing update distribution, comprising the following steps: (1) using an overhead model to construct a key into an RMI structure; (2) uniformly dividing the range of data by using the cumulative distribution function of data at each internal node; (3) constructing a monotonous gapped unary linear regression model at each leaf node; (4) learning the update distribution of data during the construction of the index; (5) point query and insertion; and (6) model expansion and splitting. The application uses a machine learning model to replace a traditional B-tree-based database index structure, simultaneously learns the update distribution of data, utilizes the characteristics of the update distribution of data and the advantages of the machine learning model, and greatly reduces the storage overhead of the index structure and the overhead during data insertion.

The application discloses a partial discharge source positioning method and system based on a physical constraint neural network and belongs to the technical field of electrical equipment state monitoring. The method comprises the following steps: collecting and pre-processing a partial discharge signal; and constructing a neural network model. A key step is to design a joint loss function which is fused with a coordinate prediction loss, a time delay consistency physical constraint loss and a geometric boundary constraint loss. The network is trained by optimizing the joint loss function, and a physical law is directly embedded into the learning process of the model. Finally, the end-to-end coordinate positioning is realized by using the trained model. The application overcomes the defects of a complex process of a traditional method and poor physical consistency of a pure data-driven model. By explicitly introducing a physical constraint into a loss function, the network simultaneously learns data features and physical laws in the training, so that the positioning precision and the model robustness are significantly improved on the premise of ensuring the physical rationality of output results.