51 results about "Meta classification" patented technology

Certain aspects of the disclosure provide techniques for slang usage detection and mitigation. A method generally includes receiving an input sentence comprising a plurality of tokens; processing, with a first machine learning (ML) model trained for slang classification, the input sentence and thereby generating at least one of: a first classification output, for the input sentence, comprising a slang instance classification; or a second classification output for each of the plurality of tokens of the input sentence, at least one second classification output comprising a slang token classification; and based on the first classification output comprising the slang instance classification and / or the at least one second classification output comprising the slang token classification, processing with a second ML model trained for slang mitigation, the slang-containing input sentence to generate a slang-free output sentence, wherein an entailment score between the input sentence and the output sentence satisfies a threshold.

The invention relates to the technical field of network security attack defense, and discloses an automatic network attack classification trapping system based on ensemble learning, comprising a network attack trapping module used for constructing a network attack trapping subsystem; the monitoring module is used for monitoring the network attack trapping subsystem; the control module is used for transmitting the system call log acquired by the monitoring module from the network attack trapping subsystem to the automatic attack classification module and carrying out rollback and recovery on the network attack trapping subsystem damaged by an attacker; the automatic attack classification module is used for converting a system call log into a triple and abstraction, then extracting text word frequency features and performing data enhancement in a numerical dimension; and performing feature selection on the enhanced data, selecting corresponding feature subsets by adopting different classifiers to train sub-classifiers, training element classifiers by utilizing prediction results of the sub-classifiers, and generating classification results by utilizing the element classifiers. The method can significantly improve the classification accuracy.

The invention discloses a mangrove forest recognition method and system based on ensemble learning, and relates to the technical field of remote sensing data processing.A mangrove forest integrated recognition model driven by feature indexes is constructed, and the model takes various heterogeneous random forest models, gradient elevator models, support vector machine models and K neighbor models as base classifiers; the mangrove forest identification method effectively utilizes respective learning capabilities for different data models, reduces single model deviation, and then couples with the meta-classifier comprising the logistic regression model, so that the classification performance of the model can be further improved, the mangrove forest identification efficiency and accuracy are remarkably improved, and the mangrove forest identification efficiency and accuracy are improved. And meanwhile, the adaptability and the recognition capability of the model under a complex background condition are improved.

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 invention relates to the field of machine learning, in particular to a range hood operation state recognition and grading early warning method and system based on integrated learning. Comprising the steps of obtaining operation data of a range hood; training the initial integrated learning model to obtain a meta classifier; calculating a state confusion degree index of each abnormal state, and determining a training sequence of the special classifier corresponding to each abnormal state; copying the meta classifier as an initial model according to the training sequence, training the initial model, and in the training process, adjusting the meta error classification rate of the abnormal data in the abnormal state and the data weight of the sample data in combination with the meta classifier, and performing iterative training to obtain a special classifier for identifying the abnormal state; obtaining an integrated classifier; and inputting the operation data of the range hood to be identified into the integrated classifier, and outputting a state classification result through the integrated classifier. The method can improve the recognition accuracy of the operation states of the range hood, such as the operation energy efficiency of the range hood and the health state of equipment.

The present disclosure discloses a return prediction system and method for predicting a likelihood of return of a product. Static and dynamic data associated with the product is obtained using a data acquisition module, and is preprocessed using a preprocessing module, which involves applying a plurality of base classifiers. Individual prediction from each of the plurality of base classifiers is combined using a meta-classifier. The meta-classifier then processes the combined data to generate a final prediction indicating the likelihood of the product being returned. The meta-classifier generates the prediction at one or more stages such as, before purchase of the product, and post purchase of the product. Finaly, a notification module transmits notification based on the prediction to either a customer or a retailer associated with the product.

The present application relates to the technical field of artificial intelligence, which can be applied to business scenarios such as financial technology and medical health, and discloses a data processing method and device based on adaptive meta-learning, equipment and medium, comprising: acquiring multi-modal input data and extracting initial modal features, generating modal confidence scores using a gating network, and weighting the initial modal features to obtain weighted modal features; generating intra-modal prototypes according to the weighted modal features and weighting and fusing them into cross-modal prototypes; encoding the current task, retrieving meta-parameters as initial values of the model, constructing a task-specific classifier and a meta-classifier, updating the parameters of the task-specific classifier and the meta-classifier based on the cross-modal prototypes, processing the target task using the updated task-specific classifier, and outputting the result. The present application improves the information utilization efficiency between modalities through confidence-driven modal weighting and prototype fusion guided by fusion weight, and improves the speed and accuracy of model adaptation to new tasks by combining task encoding with optimal initialization parameters.

The invention discloses non-entity hotline appeal event argument extraction based on boundary enhancement, which specifically comprises the following steps of: firstly, constructing a model architecture taking BERT as a substrate, and enhancing context semantics and vocabulary features by adopting character-level embedding; secondly, a rotation position coding mechanism is innovatively introduced, a boundary sensing module fused with rotation position coding is designed, and the sensitivity of the model to an entity boundary is enhanced by utilizing explicit modeling of inter-word distance characteristics of a relative position matrix. And finally, aiming at a positive and negative sample and multi-class imbalance problem, designing a layered focus loss function, and effectively solving the class imbalance problem through a learning weight optimization argument classification task of a dynamic weight adjustment difficult sample and a low-frequency class. Experimental verification shows that compared with a traditional deep learning method, the method has remarkable boundary recognition precision improvement and overall performance advantages.

The invention discloses a data classification and grading method and system based on a large language model, and relates to the technical field of data management and processing, and the method comprises the steps: uploading classification guide entries on an informatization platform, and determining a ternary classification mode containing a data end, a system end and a response end through derivative engine ternary reconstruction; locking a multi-party database, coding a data unit and mapping the data unit into a data pool; initializing a large language model according to a ternary classification mode, performing three-thread parallel processing on the data pool to obtain a ternary classification result, and generating hierarchical identification codes in a hierarchical manner; and finally, fusing the ternary classification result and the grading identification code, and establishing mapping with a multi-party database. According to the method and the device, the technical problems of insufficient classification and grading precision and low efficiency of the multi-source heterogeneous data during digital data processing are solved, and the technical effects of accurate and efficient classification and grading of the multi-source heterogeneous data and accurate classification and grading of digital data processing in a complex scene are achieved.

The invention relates to the technical field of data analysis, can be applied to business system platforms of financial science and technology, medical health and the like, and discloses a hierarchical comparison decoding method, device and equipment based on a large language model and a medium. Performing original path decoding on the labeled lexical elements to obtain original probability distribution of the next lexical elements, performing different intervention path decoding on the labeled lexical elements according to hierarchical stages of a preset large language model to obtain different intervention probability distribution of the next lexical elements, determining intervention comparison scores, and performing average processing to obtain decoding logarithmic probabilities; and screening out the lexical element with the highest probability as the next lexical element, adding the next lexical element into the target input text, and continuously classifying and gradually decoding to obtain a target decoded text. According to the method, dynamic collaboration is established for the hierarchical signal and the lexical element signal of the large language model, and the intervention precision and the content accuracy are improved.

This invention proposes a disease risk prediction method and system for unplanned readmission. This method constructs a dynamic classifier selection framework; sets binary variables representing medical diagnoses as rule sets, and uses these rule sets as extraction rules to extract several data subsets from a dataset; trains several base classifiers based on the data subsets to form a classifier pool; defines capability regions and meta-features, which are criteria for measuring the capability of each base classifier in the classifier pool within the capability regions; uses the base classifiers within the capability regions to extract meta-features from the dataset; trains the meta-classifier using the meta-features as the training set for the meta-classifier in the dynamic classifier selection framework; uses the meta-classifier to calculate the capability scores of each base classifier in the classifier pool; selects the base classifier with the highest capability score as the classifier ensemble pool; and predicts the samples to be predicted within the trained dynamic classifier selection framework. This method has high prediction accuracy.

The invention provides a space-based radar sea surface target identification method, and the method comprises the steps: obtaining a sea surface high-resolution range profile HRRP matrix of a ship, carrying out the amplitude normalization of the HRRP matrix, and carrying out the superposition and accumulation of the normalized HRRP matrix in the azimuth direction; extracting the structural characteristics of the accumulated HRRP; respectively using short-time Fourier transform and Fourier coefficient fitting to extract micro-motion features from the accumulated HRRP azimuth; the structural features and the micro-motion features are input into a meta classifier, a sea surface target recognition result is obtained, the meta classifier is formed by stacking a first base classifier and a second base classifier, the first base classifier is obtained through training according to structural feature samples, and the second base classifier is obtained through training according to micro-motion feature samples. According to the method, the structural features and the micro-motion features of the HRRP echo matrix are extracted respectively, feature fusion and feature classification are performed by using the stacking classifier, the problem that high-similarity false targets are difficult to distinguish in one-dimensional HRRP recognition is solved, and the recognition accuracy and reliability are improved.

The invention relates to the technical field of power distribution network fault prediction, and discloses a WOA-Stacking-based power distribution network fault multi-scale prediction method, system and device and a medium, and the method comprises the steps: collecting the multi-source data of power distribution network operation, fault history, weather, geography, social activities and the like; after preprocessing, constructing a Stacking integration model comprising a plurality of heterogeneous base classifiers and a meta classifier; the whale optimization algorithm is adopted to automatically search the optimal hyper-parameter combination of each base classifier, so that the generalization ability of the model is improved; generating base model output through cross validation to serve as element classifier input, and forming a complete WOA-Stacking prediction model; and finally, aiming at different time scales of ultra-short period, short period and middle period, extracting corresponding features and outputting fault prediction results, thereby realizing high-precision and multi-granularity power distribution network fault early warning.

The invention particularly relates to a micro-milling flutter detection method based on stack integration. The method comprises the following steps: forming a training sample by using flutter characteristics and labels representing machining states; the training samples are sent into a plurality of base classifiers at the same time, and corresponding flutter detection base models are obtained; taking the output of each flutter detection base model as the input and sending the input into a meta classifier to obtain a final flutter detection model; and for a to-be-tested sample, the corresponding flutter characteristics of the to-be-tested sample are sent to the flutter detection model, and the processing state corresponding to the current to-be-tested sample is obtained. According to the method, the idea of stack integration is adopted, a classification and regression tree, a k-nearest neighbor method, a feedforward neural network and a gating circulation unit are used as base classifiers, a support vector machine is used as an element classifier, complementarity of different base classifiers and the correction effect of the element classifier are integrated, and the robustness of the algorithm is improved. Therefore, the feature capture capability and generalization capability of the model are improved.

The invention belongs to the field of artificial intelligence, and discloses a multi-modal large language model task adaptation fine tuning method and a related device, and the method comprises the steps: introducing a parameter-trainable adapter layer into a preset weight layer of a multi-modal large language model; obtaining a lexical element sequence of the training sample based on a preset weight layer of the multi-modal large language model, calling a preset lexical element classification model to obtain each lexical element type in the lexical element sequence, and processing each lexical element in the lexical element sequence according to the lexical element type to obtain the output of the training sample, parameters of the adapter layer and parameters of the lexical element classification model are updated according to output of the training sample; and finally, obtaining a multi-modal large language model for task adaptation fine tuning. According to the method, fine-grained and lexical-level task adaptation is achieved in the fine tuning process, the problem of disastrous forgetting in the fine tuning process is remarkably relieved, and the performance of a multi-mode large language model on an original pre-training task is effectively maintained while the adaptability of the multi-mode large language model to a new task is improved.

The application provides an industrial material hierarchical classification method fusing retrieval enhancement and explicit hierarchical constraint, belongs to the field of industrial material classification, and comprises the following steps: vectorizing and coding historical industrial material data to obtain a standard material vector knowledge base; according to a to-be-classified industrial material, standard material samples are retrieved in the standard material vector knowledge base based on similarity, and a composite input text is obtained by splicing; historical industrial material classification samples are acquired, a fine-tuning dataset is constructed based on random undersampling of head categories and generative enhancement of tail categories, a large language model is fine-tuned and trained, and an industrial material classification model is obtained; according to an enterprise standard classification system, a legal word element classification prefix tree is constructed, the industrial material classification model is constrained to sample only from legal word elements, and an industrial material classification result conforming to the enterprise standard classification system is obtained. The application solves the problem of inaccurate industrial material classification caused by a large label space and significant long-tail distribution in industrial material classification.

The application claims a neural network backdoor sample and backdoor model detection method, specifically comprising: performing data enhancement on the obtained original data set, respectively training to obtain an original model and an enhanced model, extracting features of the original data on the original model and features of the corresponding enhanced data on the enhanced model, calculating the feature stability score of the data, and distinguishing and separating the sample as a benign sample or a backdoor sample according to the feature stability score. Based on the benign sample and the backdoor sample, a shadow model is trained, an energy function idea is introduced to optimize the query noise used to extract the shadow model features, the shadow model features are extracted using the optimized query noise, a meta-classifier is trained, and whether the target model is a backdoor model is detected. The application can better separate the backdoor sample and the benign sample; the energy function idea is introduced to reduce the number of shadow models required for training the meta-classifier, meeting the low-cost demand of users.

To reduce the burden of manual work in life planning. [Solution] The information processing device comprises a classification unit that classifies multiple numerical data related to a subject's life plan into data items for generating life planning support information using content information and a language model that indicate the content of each numerical data, and a support information generation unit that generates support information using the numerical data as data for the data items classified by the classification unit. Furthermore, this information processing device can also support the subject's decision-making.

A method for training a multi-class classification model includes receiving training data corresponding to a plurality of classes. For each class of the plurality of classes, the method includes training a binary classification model configured to determine whether a training data observation belongs to the class, and for each training data observation identified as belonging to the class, extracting one or more class identifying features from the training data observation based on activations of an intermediate attention layer in the binary classification model. The multi-class classification model is trained using the class identifying features extracted for each class of the plurality of classes.

The application discloses an Alzheimer's disease picture classification method based on a bimodal iterative cross attention fusion, relates to the technical field of Alzheimer's disease medical image classification, and comprises the following steps: acquiring sMRI and PET image data of AD and preprocessing; cutting and data augmenting the preprocessed data; constructing a bimodal iterative cross attention fusion model as a base classifier, and training the base classifier by using the cut images; in the training, inputting the cut images into a residual network module to extract features, using a spatial feature contraction module to reduce redundant information of the features, and using a bimodal iterative cross attention module to extract complementary information in combination with an iterative learning strategy to enhance the features and fusion; inputting a verification set into the trained base classifier for classification, and selecting a base classifier with a classification accuracy greater than a preset threshold; constructing a meta-classifier, classifying the selected base classifier for the cut images to be classified, inputting the fusion features into the meta-classifier for classification, and obtaining a classification result.

The system according to the embodiment comprises a collection unit, a classification unit, an analysis unit, a generation unit, and a provision unit. The collection unit collects languages of various countries and regions. The classification unit classifies the languages collected by the collection unit. The analysis unit analyzes the languages classified by the classification unit and analyzes differences in meaning, nuance, and usage. The generation unit generates languages based on information analyzed by the analysis unit. The provision unit provides the languages generated by the generation unit.

The invention relates to the technical field of seismic signal processing and mode recognition, and discloses a high-noise-robustness tunnel microseismic signal integrated classification method, which comprises the following steps: acquiring and processing an elastic wave signal to obtain a feature vector; a Stacking integrated model comprising a plurality of base classifiers and a meta classifier is constructed, a focus loss function is used for optimizing the problem of class imbalance, a hierarchical optimization strategy is adopted to perform parameter optimization on the model, outer-layer optimization parameters adjust the focus loss function, and inner-layer optimization adjusts hyper-parameters of the base classifiers; and finally, training and generating a final classification model for signal classification according to the obtained optimal parameters. By combining the Stacking model with the focus loss function with the hierarchical optimization strategy, the technical problems of class imbalance and deep coupling of model parameters are effectively solved, and accurate recognition of key microseismic signals can be realized under the background of strong noise.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes of financial science and technology, medical health and the like, and discloses a data processing method, device, equipment and medium based on adaptive meta-learning. Performing weighting processing on the initial modal features to obtain weighted modal features; generating a modal internal prototype according to the weighted modal features, and performing weighted fusion to form a cross-modal prototype; and encoding the current task, retrieving meta parameters as model initial values, constructing a task exclusive classifier and a meta classifier, updating parameters based on a cross-modal prototype, processing a target task by using the updated task exclusive classifier, and outputting a result. According to the method, through modal weighting driven by confidence and prototype fusion guided by fusion weight, the inter-modal information utilization efficiency is improved; the optimal initialization parameters are retrieved in combination with task codes, and the speed and accuracy of the model for adapting to new tasks are improved.

The application discloses a Chinese joint event extraction method based on dependency analysis, first introduces dependency analysis to construct a syntactic structure and strengthens the depth interaction of information; secondly, three types of edge representations are designed to calculate graph convolution features in order to bridge the inconsistency of words; finally, the cascade error propagation problem of the traditional pipeline method is relieved through joint learning of the event trigger word classification task and the event argument classification task, and the effect of extracting event triggers and arguments from documents is improved. The Chinese joint event extraction model based on dependency analysis integrates syntactic structure information while encoding semantics, enhances the information flow between words, and designs different types of edge representations for the construction of an undirected graph according to the characteristics of Chinese word segmentation. The application enriches semantic feature representation by integrating the syntactic structure knowledge contained in the Chinese text, and effectively improves the effect of sentence-level event extraction by using the joint learning method.

The invention discloses a lithofacies identification method for sea-land transition phase shale, and relates to the technical field of lithofacies identification. Comprising the following steps: acquiring historical logging curve data of sea-land transition phase shale and corresponding historical lithofacies type information as a training data set; the method comprises the steps that at least two different types of machine learning models serve as base learners, a logistic regression model Logistic Regression serves as a meta classifier, and a stacking model is constructed; training the stack model through the training data set to obtain a trained stack model; and in the lithofacies identification process of the sea-land transition phase shale, inputting the sea-land transition phase shale to be identified into the trained stacking model to obtain a lithofacies identification result. According to the method, the lithology of the sea-land transition phase shale can be better captured, and the robustness and stability of lithofacies identification and classification are improved.

The invention relates to the technical field of power system fault detection, in particular to a 10KV power distribution network ground fault line selection method based on ICEEMDAN and Stacking ensemble learning, which comprises the following steps: determining the state of a power distribution network according to a line danger coefficient and a fault transient threshold, and determining a line selection mode according to the state of the power distribution network to obtain a pre-fault line; a zero-sequence current signal corresponding to the pre-fault line is collected, and the zero-sequence current signal is decomposed based on the empirical mode decomposition frequency to obtain an IMF; detecting a sample entropy corresponding to each IMF, and determining a fault feature matrix construction mode according to an entropy dislocation threshold; inputting the fault feature matrix into a pre-trained Stacking ensemble learning model, and determining a base classification mode according to a matrix feature coefficient and a matrix attribution coefficient; splicing output results of the base classifiers into input features of the element classifiers so as to select fault lines; the fault line selection accuracy can be improved.

The present document relates to an artificial intelligence meta agent system based on language processing using computer input and output. The proposed meta agent system comprises: an intention interpretation unit for identifying an entity of a specific type from input data collected from an input device of a computer and allocating the entity to a predefined category to classify the entity; an intention work unit for activating an application installed in the computer and performing work intended by a user in the activated application according to the entity classified by the intention interpretation unit; and a result generation unit for providing the result of the work performed by the intention work unit to an output device of the computer in a preconfigured form.

The invention discloses a federated learning-based model training method and system. The method comprises the following steps: obtaining a sub-training set of basic clients through a central server according to the total category amount of a target data set, the number of the clients and the data category amount of the basic clients; a target client updates and trains the basic neural network according to a target sub-training set segment gradient to obtain to-be-selected extraction layer parameters, a target feature recognition layer, a target class average activation vector and target class Weibull probability distribution; averagely calculating full-amount to-be-selected extraction layer parameters through a central server to obtain target extraction layer parameters, and constructing an initial global model in combination with a preset meta classifier and a full-amount target feature recognition layer; and training the initial global model through the target client based on the target sub-training set, the target class average activation vector and the target class Weibull probability distribution to obtain a target global model. By means of the technical scheme, model training of heterogeneous data scenes can be achieved, and the accuracy of the model is improved.

The present disclosure discloses a return prediction system and method for predicting a likelihood of return of a product. Static and dynamic data associated with the product is obtained using a data acquisition module, and is preprocessed using a preprocessing module, which involves applying a plurality of base classifiers. Individual prediction from each of the plurality of base classifiers is combined using a meta-classifier. The meta-classifier then processes the combined data to generate a final prediction indicating the likelihood of the product being returned. The meta-classifier generates the prediction at one or more stages such as, before purchase of the product, and post purchase of the product. Finaly, a notification module transmits notification based on the prediction to either a customer or a retailer associated with the product.

The application relates to the technical field of data mining, and discloses a SNP site mining method and system based on a Transformer model ensemble learning, which comprises the following steps: acquiring genetic data containing chromosome SNP sites, and performing screening processing on the genetic data; dividing the screened genetic data into a training set, a verification set and a test set; constructing a basic classifier based on a Transformer model by using the training set and the verification set; analyzing the basic classifier to obtain significant SNP sites; performing ensemble learning on the basic classifier by using the test set, constructing and screening an optimal meta-classifier; and performing classification processing on the significant SNP sites by using the optimal meta-classifier, and screening significant SNP sites related to genetic diseases. By using the Transformer model to construct a basic classifier, analyzing and performing ensemble learning to construct an optimal meta-classifier, and integrating feature information, significant SNP sites related to genetic diseases are finally obtained, and the SNP site mining efficiency and accuracy are improved.