90 results about "Class model" patented technology

The invention provides a rheumatoid arthritis patient low muscle quality risk prediction method based on an uncertainty perception stacked meta-learning structure, and belongs to the technical field of machine learning. The method comprises the following steps: acquiring a rheumatoid arthritis data set; constructing a base learner set comprising a table Transform class network and a gradient boosting tree class model; splicing the out-of-fold prediction probabilities of all the base learners and the statistics thereof to form a meta-feature matrix; taking the meta feature matrix as input, and constructing and training a meta learner comprising a spectrum normalization multilayer perceptron and a random feature Gaussian process output layer; collecting to-be-detected data, inputting the to-be-detected data into the trained base learner set and the trained meta learner in sequence, and performing temperature scaling and beta-calibration on a prediction result; and taking the calibrated prediction probability as a final prediction result. Through multi-source data fusion, the defects of an existing model in the aspects of practicability, probability reliability and the like are overcome.

The invention provides a coding method and system of a Transform class model for channel state information prediction. The coding method comprises the following steps: S1, acquiring and processing channel state time sequence data; s2, inputting the processed data into an encoder to extract features; s3, CFR data needing to be predicted are embedded, and position codes of the model are added; decoding the relevant parameters using a decoder; s4, processing the output of the decoder, and predicting channel state information; and S5, taking the CFR data and the CSI data obtained in the actual scene as the input of the model, and predicting channel frequency response information. According to the method, the time sequence characteristics of the channel state information are calculated and coded, and the time information is brought into the model, so that the relative stability time of the channel at the future moment is considered, the problem of weak time sequence information embedding capability in a Transform model is solved, and the accuracy of channel state prediction is improved.

The invention relates to the field of spine detection, and provides a scoliosis classification model training method, a scoliosis classification method and a scoliosis classification system in order to reduce computing resources and improve detection precision, and the scoliosis classification model is trained by constructing a class-balanced metadata set and a class-unbalanced basic data set. The model is guided to learn features of each category, the problem of category imbalance in a scoliosis data set is solved, and the generalization ability of the model under different scoliosis angle categories is effectively improved, so that the detection precision is improved; a Mama structure and an attention mechanism are introduced into the scoliosis classification model, and by optimizing a calculation process and a feature extraction mode, on the premise that the detection precision is guaranteed, the requirement for calculation resources is reduced, and the detection method can be effectively operated on common equipment.

The present application relates to the field of artificial intelligence technology, and discloses a training method for a classification model, a search method for hyperparameters, and a device. The training method includes obtaining target hyperparameters of a classification model to be trained, where the target hyperparameters are used to control the gradient update step size of the classification model to be trained, and the classification model to be trained includes a scale-invariant linear layer, and the scale-invariant linear layer enables the prediction classification result output when the weight parameters of the classification model to be trained are multiplied by any scaling factor to remain unchanged; updating the weight parameters of the classification model to be trained according to the target hyperparameters and a target training method to obtain a trained classification model, and the target training method enables the magnitudes of the weight parameters of the classification model to be trained before and after updating to be the same. Based on the technical solution of the present application, it is possible to reduce the computing resources consumed for training the classification model while ensuring the accuracy of the classification model.

Certain example embodiments relate to meta-learning based error detection. Base classifiers are provided for historical attributes in historical datasets. Each is trained to indicate dirtiness of a value for the associated historical attribute. Clusters and a clustering model are generated using historical clustering features determined for each historical attribute, which are then associated with the clusters. For each dirty attribute in a dirty dataset, corresponding dirty clustering features are determined. The dirty attributes are assigned to the clusters using the corresponding determined dirty clustering features and the clustering model. The base classifiers associated with the clusters to which the dirty attributes were assigned are retrieved. Dirty features are extracted from the dirty dataset, and selectively modified. The extracted dirty features are applied to the retrieved the base classifiers to determine meta-features. A meta-classifier is trained using labeled meta-features. Predictions about the dirty dataset's dirtiness can be made using the meta-classifier.

The application relates to a classification model training method, an object classification method and related equipment. The classification model training method comprises the following steps: obtaining a training sample set; the training sample set comprises a plurality of sample subsets, and the plurality of sample subsets correspond to a plurality of different kinds of category labels; a preset machine learning model is constructed; the category of each training sample in the training sample set is predicted based on the preset machine learning model, and prediction category information of each training sample is obtained; the prediction category information is determined based on the number of samples in each sample subset in the plurality of sample subsets and a feature output index of the training sample; a loss value is determined based on the prediction category information of each training sample and the category label of each training sample; the preset machine learning model is trained according to the loss value until a preset training end condition is met, and a classification model is obtained. In this way, the recognition accuracy of the classification model can be improved, and the safety of the application of the classification model in the automobile field can be improved.

The present application relates to a model training method, a feature classification method, an apparatus and a computer device. The method includes: the computer device determines the contribution of each symptom feature to each department based on the historical symptom feature data of multiple users, determines a sample data set based on the contribution of the historical symptom features, trains the initial classification model based on the sample data set, and obtains a target classification model; wherein the contribution is the frequency of occurrence of the symptom feature for the department, the initial classification model includes a first-layer classification model and a second-layer classification model, wherein the second-layer classification model is a least squares dual support vector machine meta-classification model, and the target classification model is used to classify departments according to the user's symptom features. The target classification model involved in this solution uses the least squares dual support vector machine as the second-layer meta-classifier of Stacking ensemble learning, which reduces the learning difficulty of the model during training and improves the solution efficiency of model training.

The invention provides a question reply method and device and electronic equipment, and relates to the technical field of artificial intelligence such as human-computer interaction, natural language understanding, large models and light weight. The method comprises the steps of obtaining a to-be-processed question input by a user in a current session; a plurality of alternative models associated with the to-be-processed problem are utilized to process the to-be-processed problem in parallel to obtain a plurality of candidate replies, the alternative models comprise at least one first-class model and at least one second-class model, and the parameter scale and / or single call cost of the first-class model are / is lower than that of the second-class model; determining a target model in the plurality of alternative models according to behavior information of the user for the plurality of candidate replies; and during the duration of the current session, generating corresponding reply contents for the input subsequent questions by using the target model. According to the method, the target model conforming to user preferences can be rapidly and clearly determined by simultaneously presenting the behavior information of the multiple candidate replies to the multiple alternative models by the user, and the interaction turns and the clarification cost are reduced.

This specification discloses a distillation method for the fine-tuning stage of a text-based model, comprising: acquiring training set samples and a pre-trained teacher model, the teacher model including multiple hidden layers; selecting at least two specified layers from the multiple hidden layers of the teacher model and aggregating them to obtain aggregated hidden layers; generating a student model based on the aggregated hidden layers and the remaining unspecified layers in the teacher model; distilling the student model using the teacher model; and reusing the distilled student model as the teacher model until it is impossible to select at least two specified layers from the teacher model, at which point the teacher model at this point is used as the target model. Accordingly, this specification also discloses a distillation apparatus for the fine-tuning stage of a text-based model.

The embodiment of the invention provides a classification model training method and device and a classification detection method and device, and relates to the technical field of artificial intelligence. Inputting the input data into a classification model and a teacher model for classification prediction to obtain a student classification result, a teacher classification result, student intermediate features and teacher intermediate features of each training stage, and for each training stage, calculating a heterogeneous alignment loss value according to the student intermediate features and the teacher intermediate features, calculating a teacher distribution consistency loss value according to the teacher classification result and the teacher intermediate feature, calculating a distribution alignment loss value according to the student classification result and the teacher classification result, and calculating a student task loss value according to the student classification result and the data label corresponding to the input data; and calculating a total loss value based on all the loss values to train a classification model. The heterogeneous semantic gap is reduced by utilizing multi-level feature alignment and dynamic loss optimization, perception and adaptation to the learning track of the student model are realized, and the knowledge distillation accuracy of the heterogeneous model is improved.

Embodiments of the present specification provide a kind of training method of classification model, the method and device of user classification, method includes: the element in the feature representation vector of sample user is randomly set 0 process, obtain missing representation vector;Missing representation vector is input into encoder, obtain the mean vector and variance vector of Gaussian distribution that the hidden vector corresponding to sample user is subject to;Based on mean vector and variance vector, generate first hidden vector, input first hidden vector into decoder, obtain the reconstruction representation vector of sample user;According to the difference between reconstruction representation vector and feature representation vector, determine first loss term;Mean vector is input into classifier, obtain the probability distribution of each preset category to which sample user belongs;According to the category label and probability distribution of sample user, determine second loss term;Total loss function includes first loss term and second loss term.Can improve the performance of classification model when the value of user feature exists missing.

An Artificial Intelligence (AI) based computational system is used to non-invasively estimate the presence of a range of aneuploidies and mosaicism in an image of embryo prior to implantation. Aneuploidies and mosaicism with similar risks of adverse outcomes are grouped and training images are labelled with their group. Separate AI models are trained for each group using the same training dataset and the separate models are then combined, such as by using an Ensemble or Distillation approach to develop a model that can identify a wide range of aneuploidy and mosaicism risks. The AI model for a group is generated by training multiple models including binary models, hierarchical layered models and a multi-class model. In particular the hierarchical layered models are generated by assigning quality labels to images. At each layer the training set is partitioned in the best quality images and other images. The model at that layer is trained on the best quality images, and the other images are passed down to the next layer and the process repeated (so the remaining images are separated into next best quality images and other images). The final model can then be used to non-invasively identify aneuploidy and mosaicism and associated risk of adverse outcomes from an image of an embryo prior to implantation.

The invention belongs to the technical field of open set recognition, and discloses a hyperspectral open set recognition system and method based on class semantic reconstruction. The hyperspectral open set recognition system based on class semantic reconstruction comprises a grouping spectral space reservation transformer module which comprises three stages of hierarchical structures, and each stage of hierarchical structure comprises a grouping pixel embedding module and a space enhancement feature transformation module. The spectrum-spatial feature fusion module is used for fusing the stage spectrum-spatial features into discriminative spectrum-spatial features; the class semantic reconstruction module comprises a plurality of class auto-encoders and is used for performing class semantic reconstruction on the discriminative spectral-spatial features to obtain reconstruction errors, constructing and training a classification model through the reconstruction errors and finally realizing open set recognition in combination with a multi-dimensional scoring function; according to the method, inter-class confusion can be reduced, and interference of background noise information is reduced by reconstructing semantic features instead of original pixels.

Embodiments of the present application provide a text intention classification model training method, a recognition method, a device, an electronic device and a storage medium, belonging to the field of data processing. The training method comprises: obtaining the word group of each training text and the labeled label of each training text, obtaining an initial model comprising a graph neural network and a classifier, taking all labeled labels as the iterative input of the graph neural network to enable the graph neural network to learn the correlation between all intention labels, taking the word group of the training text and the correlation as the iterative input of the classifier, training the initial model, and training a classification model for obtaining the labeled label of the training text, so that the trained classification model can adapt to the application scenario of data distribution change, greatly improving the classification accuracy of the classification model.

The present disclosure relates to a classification method and a training method of a classification model, an apparatus, a device and a medium. The classification method comprises: obtaining a to-be-classified object; obtaining a first category and intermediate features of the to-be-classified object by using a classification branch model in a classification model based on the to-be-classified object; obtaining a second category by using a verification branch model in the classification model based on the first category and the intermediate features; and determining the second category as the category of the to-be-classified object. According to the classification method and the training method of the classification model, the verification branch model is added to the traditional classification model, so that the classification result of the classification branch model is verified again, the precision rate of the classification branch model is further improved, and the classification accuracy is improved.

The invention discloses a class model inspection method, electronic equipment, a storage medium and a product. The method comprises the steps of obtaining context information of a target micro-service application in response to a class model check request from a client; performing class model compliance check based on the context information to generate a corresponding error check result; wherein the class model compliance check comprises at least one of the following items: class member reference validity check, interface and abstract method implementation integrity check, technical suite compatibility check, class name uniqueness check and request path uniqueness check; sending an error checking result to the client for visual display; receiving a correction instruction from the client; the correction instruction is associated with a target error item in the error checking result; and executing a response operation corresponding to the correction instruction. According to the scheme, the multi-dimensional compliance check is executed based on the micro-service context, and the class model design quality and efficiency are improved through visual interaction and intelligent response.

A method for receiving a full training data set including a plurality of individual training data set, dividing the plurality of individual training sets into N classes, where N is an integer greater than three, dividing the N classes into M full data classes and N-M partial data classes, performing training to obtain a trained fixed size machine learning (ML) classification model and a trained in-class confidence model, outputting a first set of prediction value(s) based on the performance of training, distributing each class of the N classes of individual training data sets to a different node of a distributed machine learning system; and outputting, from the nodes of the distributed machine learning system, a second set of prediction value(s) for each class of the N classes.

Provided in the present application are a classification model updating method and apparatus, and a device and a storage medium, which are used for improving the classification precision of a classification model. The method comprises: determining a first new-type sample set from a test sample set on the basis of a first classifier in a first classification model, wherein the first classifier is used for classifying a sample into a first known-type sample which belongs to a recognizable first known type or into a first new-type sample which belongs to an unrecognizable first new type, and one sample comprises the value of at least one statistical feature corresponding to one piece of network traffic data; training a second classifier on the basis of the first new-type sample set, wherein the second classifier is used for outputting the type of a sample, apart from the first new-type sample and the first known-type sample, as an unrecognizable second new type; and adding the second classifier to the first classification model so as to obtain a second classification model, wherein the second classification model is used for classifying a sample into any one of the following types: the first known type, the first new type and the second new type.

The application provides a bimodal scene lightweight vertical model construction method and system and electronic equipment, by acquiring bimodal data and preprocessing, then visual-text cross-modal importance scoring is carried out, a structured pruning strategy is adopted, the pruning proportion is dynamically adjusted according to the importance score, the key visual channel and the text neuron are retained, the model is lightened, after knowledge distillation fine-tuning, directional fine-tuning is implemented, multi-target loss is designed for multi-class text tasks for optimization; through hierarchical extraction and splicing features to form comprehensive representation, fully mining the value of text information, improve the utilization efficiency of text modal, use the loss function containing cross-modal contrast loss to enhance the modal alignment, improve the detection accuracy, improve the adaptability of the model in the specific scene.

The invention provides a model conversion method and device and a storage medium. The method comprises the steps of receiving a model conversion request sent by a client based on a designer canvas; the model conversion request comprises a domain model and a model type to be converted; loading a model conversion engine; calling a model conversion engine to convert the domain model into an application class model under the model type; and if the conversion is completed, storing the application class model into a model database. According to the embodiment of the invention, through a mechanism of converting the application model from the domain model and in cooperation with recording and storage functions of the conversion operation, tracing of the model modification history is realized, the user is guided to complete configuration selection and data input through the built-in and predefined model conversion engine, operations such as manual compiling of complex conversion scripts by the user are reduced, and the user experience is improved. The technical operation threshold is greatly reduced, omission or errors possibly caused by manual synchronous operation are effectively reduced through the mechanism, and the consistency maintenance level between different abstract levels is remarkably improved.

The present invention provides a target user mining method and device based on an improved multi-layer Stacking model, the method comprising: obtaining an original data set and an original Stacking classification model, wherein the original data set is pre-labeled user data, and each layer of the original Stacking classification model is provided with X types of basic classifiers; training each layer of the original Stacking classification model in sequence according to the original data set to obtain a trained target user classification model, wherein the first and third layers of the target user classification model only include one basic classifier with the highest classification accuracy among multiple basic classifiers, and the second layer includes all types of basic classifiers; obtaining a set of users to be classified, and inputting the user data to be classified into the target user classification model to obtain a target user set. By optimizing the traditional Stacking classification model, the present invention can optimize the classification effect of the traditional model on the data set and improve the classification accuracy of the target users.

The invention discloses a classroom event analysis method and an analysis model optimization method and system, and relates to the technical field of edge computing. The method comprises the following steps: acquiring real-time classroom data acquired by classroom sensing equipment by edge equipment, and storing the real-time classroom data into an offline classroom file; detecting a real-time event stream in the real-time classroom data through a preset lightweight detection model; the lightweight detection model is optimized through an off-line event stream of an off-line classroom file after class, and the off-line event stream is obtained by analyzing the off-line classroom file through a high-precision detection model; and sending the real-time event stream to visual equipment, so that the visual equipment displays the real-time event stream in real time. Through the above technical means, the lightweight detection model rapidly analyzes the real-time classroom data to obtain a high-timeliness real-time event stream, the after-class model optimization can continuously improve the detection precision of the lightweight detection model, the timeliness and precision of the real-time event stream are considered, and instant and reliable decision support is provided for current teaching.

The application relates to a remote sensing image active learning method based on a class-level graph embedding representation. The method comprises the following steps: randomly taking a labeled set and an unlabeled set from labeled data of a hyperspectral remote sensing image, taking the remaining samples as a test set, setting the iteration number and sample budget of active learning, training a classification model using the labeled set, obtaining spectral feature representation of the samples through network parameters, dividing the labeled set into K classes according to real labels, constructing graph nodes and an adjacency matrix for each class, training a class-level graph convolution network model, obtaining inter-class minimum uncertainty of the unlabeled samples, selecting B uncertain samples as a query set, giving the query set real labels and adding the query set to the labeled set, updating parameters of the classification model and the class-level graph convolution network model using the new labeled set, and when the iteration number is I, training the classification model using the updated labeled set and classifying the test set to obtain a classification result. Therefore, the accuracy of a deep learning model classification is improved.

The invention relates to the technical field of data processing, in particular to a multi-dimensional automobile accessory big data processing method and system based on a vertical class model, and the method comprises the steps: obtaining multi-dimensional data information of automobile accessories at regular time; according to key text information in the multi-dimensional data information, constructing a text feature vector of the automobile accessory; determining a first similarity between each attribute of each automobile accessory and the automobile accessory with the same name; obtaining the attribute importance degree of each attribute of each automobile accessory; based on a vertical class model, evaluating the similarity degree of the text feature vectors between any two automobile accessories obtained in a single time, and classifying the automobile accessories; and determining a first similarity between each attribute of any automobile accessory in the same kind of automobile accessories and the remaining automobile accessories, and calculating the automobile accessory similarity between the any automobile accessory and the remaining automobile accessories by combining the attribute importance of each attribute of the any automobile accessory so as to perform duplicate removal processing on the same kind of automobile accessories. Therefore, the duplicate removal effect of the automobile accessory data is improved.

This invention discloses a semi-supervised adaptive EEG signal classification method, comprising: 1. acquiring labeled data from different individuals, acquiring data from the target individual and selecting a portion for labeling, followed by unified preprocessing; 2. designing a data augmentation mechanism to enhance the representation of the samples; 3. building student and teacher models; 4. recording the weights of the student model after each update, and performing a moving average with the historical weights to serve as the weights of the teacher model; 5. calculating the feature alignment loss between existing individuals and the target individual in the feature space; 6. calculating the consistency loss between the output probabilities of the student model and the teacher model; 7. combining the above loss function with cross-entropy loss to jointly optimize the model parameters and obtain the optimal classification model. This invention can fully calibrate the classification model even when only a small portion of the target individual's data is labeled, thereby achieving high-accuracy cross-individual EEG signal classification.

The invention provides a modeling method and device in the business field and a storage medium. The method comprises the steps that a webpage is loaded for a user under the same project; the webpage is provided with a canvas of an item; drawing a plurality of graphic elements in the canvas according to the operation of the user, and updating a data model bound with the graphic elements; when a modeling instruction is received, packaging the data model bound with the plurality of graphic elements into a business domain model; and converting the business domain model into a class model. According to the embodiment, on-line multi-person collaborative design and cross-end interaction are supported based on a webpage mode, the collaborative efficiency and data consistency are effectively improved, the graphic layer and the model layer are bound, operation of the user on the graphic element is synchronized to real-time updating of the data model, seamless conversion from visual design to structured data is achieved, and the user experience is improved. And the modeling efficiency of the business field is effectively improved.

The application provides a kind of training method of classification model and fundus image classification method, belong to the field of incremental learning.A kind of training method of classification model, the classification model is used for fundus image classification, the method comprises: obtaining pre-trained classification model as teacher model, it includes feature extraction network and classifier, the fundus class that the teacher model can identify is classified as old class;Student model is obtained, the feature extraction network of which is initialized with the teacher model, and its classifier is set to be able to identify the fundus class corresponding to the old class and new class, the new class is the fundus class outside the old class;For each old class used, obtain the pseudo sample corresponding to the old class;The student model is incrementally trained using the generated pseudo sample of old class and the fundus image belonging to new class, and the parameters of the student model are updated based on the total loss determined by the preset total loss function during training, to obtain the student model after incremental training.The present application can alleviate catastrophic forgetting.

The disclosed systems and methods provide a novel action prediction framework that performs personalized action prediction. According to an embodiment, the disclosed framework is able to dynamically predict which action (if any) a user might perform in response to receiving a given message. In some embodiments, for a given message, the action prediction framework can determine the probability that a user (e.g., sender, recipient) associated with the message may perform an action or set of action actions (e.g., open, forward, delete, reply, archive) related to the message. In some embodiments, the framework may be used to suggest a predicted action to the user. In some embodiments, a computing device may use the predicted actions to automatically perform the action. According to an embodiment, the action prediction framework includes a multi-label or multi-class model using a neural network.

The application discloses a kind of based on SVG technology's class model construction method and device, the method includes: monitoring class model construction information, and according to class model construction information, determine class model construction content;According to class model construction content, judge whether the current situation corresponding to class model construction information meets the set construction demand;When the judgment result is yes, according to class model construction information and class model construction content, target class model is obtained by construction.It can be seen that the SVG rendering engine can be used as the core to realize the class model intelligent construction function suitable for various class model construction scenarios, which is beneficial to improve the software class model construction convenience and construction efficiency, thereby reducing the human and material resources and time cost of class model construction development.In addition, the target class model obtained by construction is a structured data model, which is beneficial to improve the standardization and structuring of class model, and thereby improve the applicability and referability of class model.

The invention discloses an ML-ELM-AE target motion pattern recognition method containing class information, and belongs to the technical field of situation cognition. An ELM-AE model is constructed, and an ML-ELM-AE model is constructed by stacking the ELM-AE model; carrying out the mapping of the motion features of the target through the ML-ELM-AE; a CELM classification model is constructed, and according to a difference vector set formed by inter-class samples, the weight of an input layer of the CELM classification model to a hidden layer and a hidden layer node bias item are optimized, so that the mapping of the samples from a feature space to a class space has regularity, and the precision and generalization ability of target motion pattern classification are improved.