31 results about "Metaknowledge" patented technology

The invention discloses a method for obtaining user specific metaknowledge interest. The method is specific to a learner in a network learning system, aims at exactly identifying and obtaining learning interest of the network learner, predicts the future interest change of the learner, and improves learning efficiency and performance of the network learner. In the aspect of user model, a user interest model facing the network learner group is adopted. In the interest omission algorithm of dynamic interest perception, an interest omission algorithm featuring automatically discovering and omitting the threshold is adopted. In the interest prediction algorithm, a prediction method featuring user collaborative prediction is adopted. The invention analyzes blogs of the learner to exactly identify the interest change of the user towards metaknowledge, conception and learning resource, reflects interest degree in a quantitative way, predicts the interest change of the learner, improves defects of the existing technologies, and has good application prospect in the network learning system and network individual service.

The invention provides a knowledge inference method and system of a knowledge graph based on meta-knowledge and a medium, the training difficulty does not significantly increase with the increase of the scale of the knowledge graph, the inference accuracy is more advantageous, and the method comprises the following steps: step 1, data set construction: establishing a data set composed of knowledgegraph triples: a head entity, a relationship and a tail entity; step 2, data re-processing: re-learning the initialized embedded vectors of the entities in the data set through a data re-processing model to obtain semantics of the entities and obtain final embedded vectors of the entities; step 3, model training: constructing a knowledge inference model, and inputting triples in the data set subjected to data reprocessing into the knowledge inference model for model training to obtain a trained knowledge inference model; and step 4, knowledge reasoning: scoring the alternative tail entities in the knowledge graph triad through the trained knowledge reasoning model, and selecting the tail entity with the highest score as a reasoning result of the knowledge reasoning model.

The invention discloses a knowledge graph construction method and system and a storage medium. The method comprises the steps: collecting initial learning data, storing the initial learning data in alearning resource database, and carrying out the ontology term extraction and marking of the initial learning data according to a subject teaching outline and teaching materials; inputting the relationship and sequence between each meta-knowledge point, constructing a directed subject knowledge graph through an ontology editor, generating an OWL file, and storing the OWL file in a learning systemin the form of a triple; updating and inputting the learning process state data of the user in real time, and performing multi-knowledge-point modeling on the learning process state data of the user through a DKT algorithm to obtain a learning and mastering state of the user; According to the learning and mastering state of the user, mining the relationship among the knowledge points, and dynamically updating the subject knowledge graph. According to the method, the DKT is tracked based on deep learning, so that a multi-knowledge-point modeling scene is effectively solved, the latest knowledgemastering state of a learner is accurately detected, the knowledge graph is constructed, and the internal relation of knowledge points is obtained.

The invention discloses a meta-knowledge fine tuning method and platform for a multi-task language model, and the method comprises the steps: highly-transferable common knowledge, i.e., meta-knowledge, on different data sets of the same kind of tasks are obtained based on cross-domain typicality score learning; and learning processes of similar tasks on different domains corresponding to differentdata sets are mutually associated and mutually enhanced, so that the fine adjustment effect of similar downstream tasks on different domain data sets in language model application is improved, and the parameter initialization capability and generalization capability of a universal language model of similar tasks are improved. According to the method, fine adjustment is carried out on a downstreamtask cross-domain data set, the effect of a compression model obtained through fine adjustment is not limited by a specific data set of the task, on the basis of a pre-trained language model, fine adjustment is carried out on a downstream task through a meta-knowledge fine adjustment network, and therefore a similar downstream task language model irrelevant to the data set is obtained.

The invention discloses a method for automatically classifying, obtaining and storing complex knowledge of a high-end device. The method comprises: an automatic complex knowledge classification method of performing induction and reorganization on knowledge resources from the following three dimensions of the high-end device: a life cycle dimension, a knowledge manifestation pattern dimension and a knowledge theme dimension, and automatically classifying the knowledge resources by using a naive Bayes classifier; a complex knowledge obtaining method of obtaining a template according to complex knowledge based on a meta-knowledge model and obtaining the complex knowledge resources through semi-automatic obtaining technology based on the obtained template; and an automatic complex knowledge storage method of dividing the complex knowledge resources from the physics through a series of automatic division rules, compressing key information and storing the same in different storage spaces in a distributed manner. The method disclosed by the invention covers the automatic complex knowledge classification method, the complex knowledge obtaining method and the automatic complex knowledge storage method, and provides foundation and support for the high-end device manufacturers to use the complex knowledge resources.

The invention provides a cross-modal document processing method and device and electronic equipment. The method comprises the steps of obtaining text modal data and image modal data of a first document; converting the text modal data into a word and sentence feature embedding vector based on a natural language processing model, and extracting a first text element feature according to the text modal data; based on a computer vision algorithm, through a target detection algorithm and an optical character recognition algorithm, positioning a target position, recognizing text content of the firstdocument, extracting second text element features according to the image modal data, and performing element alignment to obtain structural features of the first document; and then obtaining a meta-knowledge graph model which comprises representation of the first document in combination with the structural features and the embedded representation of the multi-dimensional features of the document. Based on this, the text content of the document is converted into the meta-knowledge graph model, so that the electronic equipment can identify and understand the document content more completely by using the meta-knowledge graph model.

The invention discloses a data set classification learning algorithm automatic selection system and method, and belongs to the technical field of machine learning. The method aims at solving the problems that a selection mode of a learning algorithm involved in existing data processing does not have universality, and if attempts are conducted one by one, the calculated amount is too large. The system comprises a training feature selection module for selecting each classification problem data set and processing each classification problem data set to obtain corresponding classification meta-knowledge; a selector module which is used for selecting effective features from the classified meta-knowledge as meta-features, forming a selector training set and training a meta-knowledge training selector; an algorithm selection module which is used for processing the to-be-processed data set to obtain to-be-processed meta-features, analyzing by adopting a meta-knowledge training selector to obtain an optimal learning algorithm of the to-be-processed data set; and a knowledge base module which is used for obtaining an algorithm selection training set comprising a one-to-one correspondence relationship between different classification problem data sets and corresponding learning algorithms. The method can predict the optimal learning algorithm for the data set.

The invention provides a small sample fault diagnosis method based on task sorting meta-learning, and the method comprises the following steps: S1, collecting fault data from fault equipment, carrying out the working condition-level fine-grained division of the fault data, carrying out the segmentation and Fourier transformation of a vibration signal, obtaining a time-frequency image, and then constructing a fine-grained fault time-frequency image data set; S2, performing clustering in each task of a training task module in the meta-learner to obtain a score of the task, and sorting the fault classification tasks from easy to difficult according to the scores; S3, performing inner and outer two-layer loop training through the learning of meta-knowledge and the learning of each task to obtain initialization parameters; and S4, performing fine tuning by using the initialization parameters and a small number of samples of the test task to obtain the output of fault diagnosis. According to the method, fault tasks can be well found in actual industrial scenes by finding sensitive initialization parameters with strong knowledge adaptability.

The invention discloses a small sample image classification method based on a memory mechanism and a graph neural network, which is characterized in that a small sample model is helped to perform reasoning prediction by means of learned conceptual knowledge, and specifically comprises three stages of pre-training, meta-training and meta-testing, wherein the pre-training takes the trained feature extractor and classifier as initialization weights of an encoder and a memory bank; the meta-training is characterized in that features of samples of a support set and a query set are extracted through an encoder, related information of each class is mined from a memory bank to serve as meta-knowledge, and similarity between task related nodes and the meta-knowledge is propagated through a graph neural network; and the meta-test obtains a classification result through task related nodes and meta-knowledge nodes. Compared with the prior art, the method has the advantages that a human recognition process is used for reference, a memory graph augmentation network based on information bottleneck is used, well-learned conceptual knowledge is used, the model is helped to conduct reasoning prediction, the method is simple and convenient, practicability is high, and certain application and popularization prospects are achieved.

The embodiment of the invention relates to the technical field of knowledge graphs, and discloses a knowledge graph embedding method, system and device based on tensor decomposition. The method comprises the steps of firstly obtaining a tensor corresponding to a current knowledge graph; carrying out Tucker decomposition on the tensor to obtain a tensor after decomposition; determining a fact scorecorresponding to the decomposed tensor; and updating the embedded vector corresponding to the preset knowledge graph embedded model according to the fact score, and performing an embedding operationof the current knowledge graph through the updated embedded vector. Obviously, the embodiment of the invention provides a better embedding mode for the N-element knowledge graph, specifically, the tensor decomposition operation is combined on the knowledge graph embedding operation, and finally, the accuracy of the embedding operation can be improved; meanwhile, the parameter complexity is greatlyreduced, and the operation efficiency is improved. Moreover, the embodiment can perform joint decomposition based on the Tucker and the tensor loop expression.

The invention discloses a personalized knowledge service recommendation system based on a knowledge graph, and the system comprises a blog reading and running accompanying service module which comprises a knowledge graph construction unit, a knowledge query unit, a knowledge learning path planning unit, and a knowledge extension unit. The knowledge graph construction unit is used for a user to construct an own knowledge graph system; the knowledge query unit is used for querying the content of knowledge points in an existing knowledge graph in the system; the knowledge learning path planning unit is used for recommending new knowledge points to the user for learning after the user finishes learning one knowledge point; and the knowledge extension unit is used for pushing extended knowledge points to the user, and the extended knowledge points are knowledge points not in the knowledge graph system of the user. According to the invention, the user can be helped to obtain effective transmission and fusion among various types of learned knowledge information.

The invention belongs to the technical field of computers, and particularly relates to a task-adaptive small sample image classification method based on meta transfer learning. According to the method, by combining meta transfer learning, the problem that the feature extraction of the MAML model by adopting the 4Conv shallow network is insufficient is solved; trainable parameters are added to learn use of balance meta-knowledge in each task, and the problems of task imbalance, category imbalance and distribution imbalance of small sample learning in a real scene are solved. According to the method, the samples with low accuracy in each task are selected, and the data of the samples are recombined, so that the samples become more difficult tasks, and the accuracy of the model is improved in the process that the meta-learner learns the more difficult tasks. According to the difficult task mining algorithm provided by the invention, samples with poor classification effects are collected online to form the difficult tasks, so that a learner can learn more quickly in the difficult tasks, and the effect is better.

The invention provides a physical knowledge point intelligent recognition method and device based on teaching thinking, and relates to the technical field of education. The method comprises the following steps: constructing a knowledge element graph based on keywords and meta-knowledge points of a physics department, combing out an examination point knowledge graph in which knowledge points are associated with examination points based on an examination point logic system of the physics department; extracting a graph logic relation according to the knowledge element graph and the examination point knowledge graph; carrying out structured processing on the physical questions, respectively extracting physical elements of each part of content, and carrying out semantic combination on the physical elements through a pre-trained language model; constructing an examination point recognition algorithm model, combining structured analysis and physical semantics of physical questions, and recognizing examination point intervals with strong question correlation so as to mark physical examination points of the examination point intervals. The technical problems of low knowledge point labeling efficiency and low labeling accuracy generally existing in an automatic examination point labeling method in the prior art can be solved.

The invention discloses a cognitive diagnosis method based on meta-knowledge dictionary learning. The cognitive diagnosis method includes the steps of constructing an over-complete meta-knowledge dictionary is constructed, and expressing the answering situation of students on the dictionary; supposing that knowledge points mastered by students and knowledge points involved in questions are linearly composed of meta-knowledge points, and sparsely representing knowledge structure states of the students through a meta-knowledge dictionary; under the condition that expert knowledge points are known, learning and obtaining a relation matrix of the expert knowledge points and meta-knowledge points; and obtaining the mastering condition of each student for the expert knowledge points by utilizingthe product of the sparse representation matrix of the students and the meta-knowledge points and the relation matrix of the expert knowledge points and the meta-knowledge points. Besides, the students with the same knowledge structure are clustered into one class by using the sparse representation matrix of the students so that the students can be better grouped and the student clustering effectcan be enhanced.

The invention discloses a meta-knowledge fine adjustment method and platform based on domain invariant features, and the method comprises the steps: learning highly-transferable common knowledge, i.e., domain invariant features, on different data sets of similar tasks, and learning common domain features on different domains corresponding to different data sets of similar tasks in a fine adjustment network set; and any different domains can be quickly adapted. According to the method, the parameter initialization capability and generalization capability of the similar task universal language model are improved, and finally, the universal compression architecture of the similar downstream task language model is obtained through fine adjustment. In the meta-knowledge fine adjustment network,a domain invariant feature loss function is designed, and domain-independent general knowledge is learned, that is, a learning target of one domain invariant feature is minimized to drive a languagemodel to have domain invariant feature coding capability.

The embodiment of the invention provides a data processing method and device, equipment and a computer readable storage medium. According to the method provided by the embodiment of the invention, the empirical data of model training can be stored in a structured form by pre-creating and maintaining the meta-knowledge resource library consisting of the data meta-features of the data set and the meta-knowledge of the model meta-features generated when the data set is used for model training; further, during model training, model hyper-parameter initialization is performed according to the meta-knowledge resource library; according to the meta-knowledge resource library and the pre-configured early stop control strategy, the training process of the invalid hyper-parameter combination is stopped in advance, model training can be guided according to empirical data of model training without depending on knowledge and experience of a certain person, the training process without hyper-parameter combination can be reduced, the time consumption of model training is shortened, the model training efficiency is improved, and hardware resources and labor cost of model training can be reduced.

The invention discloses a meta-knowledge federation method for behavior analysis, which relates to the technical field of computers and comprises the following steps: acquiring user behavior data andreceiving first meta-knowledge; performing meta-knowledge learning based on the first meta-knowledge and the user behavior data to obtain meta-knowledge update; sending the meta-knowledge update to aserver, so that the meta-knowledge update is subjected to federated collection to obtain updated first meta-knowledge; receiving the updated first meta-knowledge and judging whether the first meta-knowledge is converged or not, if so, taking the first meta-knowledge as reference meta-knowledge, and performing behavior analysis; and if not, continuing meta-knowledge learning. According to the method, user data privacy is effectively protected, the client can quickly obtain customized meta-knowledge only through a small amount of user data, behavior analysis on any client is achieved, and the method is high in applicability, convenient to apply and good in user experience. The invention furthermore discloses a meta-knowledge federation method for behavior analysis, an electronic device, a computer storage medium and a system.

The invention provides an expert information screening method and system. The method comprises steps of expert information screening conditions being acquired; respectively carrying out keyword extraction on the science and technology project information, the science and technology achievement information and the expert information to obtain a science and technology project keyword set, a scienceand technology achievement keyword set and an expert keyword set; constructing a corresponding matter-element knowledge representation model according to the science and technology project keyword set, the science and technology achievement keyword set and the expert keyword set; utilizing the matter-element knowledge representation model to construct an index database containing the expert information; indexing in an index database according to the expert information screening condition, and generating an expert information index file corresponding to the expert information screening condition; determining the matching degree of each expert information index file and an expert information screening condition; and sorting and displaying the expert information index files according to the sequence of the matching degrees from high to low. The method is advantaged in that selection of evaluation experts in a manual screening mode is avoided, screening accuracy of the evaluation experts is improved, and screening objectivity is improved.

The invention discloses an image classification method and system based on meta-learning and memory network. Firstly, the original feature representation of the image is obtained by learning; a memory network module is set, and each memory block in the memory network module is correspondingly stored with meta-knowledge of a corresponding category. ; Calculate the original feature representation of the image with the memory block to obtain the read parameters, and use the read parameters to obtain the final feature representation of the image from the memory block; map the final feature representation of the image to all memory blocks, and calculate its corresponding value in each memory block The probability value on the category, according to the size of the probability value to judge the category it belongs to. The present invention constitutes a memory network module by designing memory blocks corresponding to categories one by one, each memory block corresponds to the meta-knowledge of the corresponding category, and at the same time learns the meta-knowledge information between categories through the mode of sharing the memory blocks, so that when similar images pair Its operation plays an auxiliary role while suppressing the performance of other classes of images on this class to achieve better predictions.

The invention belongs to the field of military operation and research, in particular to a method for calculating target damage. Its technical solution is to propose a fast target damage calculation method based on the meta-paradigm, which is characterized by: according to the shape characteristics of the target area, decompose the target into multiple independent and non-overlapping meta-targets; use mathematical analysis methods to calculate the meta-targets using the simulation method, using the meta-target analysis damage probability as a simulation input condition, directly simulating and generating the conclusion whether each meta-target is damaged, so as to obtain the target damage probability. The invention directly simulates the damage result of each simulation element target, saves each simulation calculation process in the simulation simulation process, greatly reduces the calculation amount, and greatly shortens the calculation time; the general construction of the element knowledge base improves the calculation efficiency.

The invention provides a construction method of a population distribution prediction model, and belongs to the technical field of Internet. The method comprises the steps of optimizing a learning rate of each learning task and a parameter value of each model parameter based on a parameter value of each intermediate state in an optimization process of each model parameter, and obtaining an optimized learning rate of each learning task and an optimized parameter value of each model parameter; and adjusting the optimized parameter values of the model parameters based on the optimized learning rate of the target learning task to obtain a population distribution prediction model. According to the invention, meta-training is carried out based on the initial value of each model parameter and the initial learning rate of each learning task, the parameter value of the optimized model parameter containing meta-knowledge and the learning rate of the optimized learning task are obtained, and then the parameter value of the optimized model parameter is finely adjusted according to the learning rate of the optimized target learning task, so that the learning efficiency of the target learning task is improved. Therefore, a relatively accurate population distribution prediction model is constructed based on a small number of population prediction samples.

The invention discloses a hyperspectral image target detection method based on meta learning and a conjoined network. The method comprises the following steps: firstly, designing a one-dimensional deep residual convolutional network to construct a three-channel deep residual convolutional conjoined network; secondly, designing a spectrum triplet loss function in order to enable the intra-class distance of the spectrums of the same type of pixels to be small and the inter-class distance of the spectrums of different types of pixels to be large; metatraining is carried out on a designed three-channel deep residual convolutional conjoined network by using a task constructed on a known labeled source domain hyperspectral data set, and the similarity and diversity between spectrums are learned in an Euclidean feature space. And updating the learned meta-knowledge by using a priori target pixel spectrum through a designed two-channel deep residual convolutional conjoined network so as to quickly adapt to a new detection task. Wherein the structure and the parameter of the one-dimensional deep residual convolutional network of each channel in the conjoined network are the same. And finally, processing the detection graph of the two-channel depth residual convolutional conjoined network by using guide graph filtering and morphological closed operation in combination with spatial information to obtain a final detection result graph.

The embodiment of the invention provides a neural network training method and device and a computer readable storage medium, and the method comprises the steps of carrying out the iterative training of an initial network through employing a sample of each task in a first data set based on initial meta-knowledge, the initial task knowledge of each task, and a task loss function, and obtaining the task knowledge of each task, wherein the first data set is a source domain data set containing an initial task category; based on the task knowledge, the initial meta-knowledge and the meta-loss function of each task, performing iterative training on the initial network by adopting samples corresponding to various tasks in the first data set to obtain optimal meta-knowledge; and based on the optimal meta-knowledge, the task knowledge of each task and a task loss function, performing iterative training on a network corresponding to the optimal meta-knowledge for a sample of each task in the second data set to obtain the optimal task knowledge of each task and a target network corresponding to the optimal task knowledge. According to the invention, generalization of training of a small number of samples is improved.

The invention relates to the technical field of adaptive learning, and discloses an adaptive learning path recommendation method and system based on a D-S evidence theory, and the method comprises thesteps: selecting meta-knowledge points with an entry degree being zero from a unit knowledge point group from a domain knowledge base, building a recommendation queue, and enabling the meta-knowledgepoints to serve as first knowledge points to be stored in the recommendation queue; judging whether the direct subsequent node set of the meta-knowledge points in the recommendation queue is empty ornot; obtaining the knowledge point with the maximum contribution value in the direct subsequent node set; judging whether knowledge points are reachable or not through a D-S evidence theory; and outputting the knowledge points as a recommended path according to the sequence of the knowledge points in the recommendation queue. The technical problems that in the prior art, a learner is usually matched with learning content only based on one-dimensional attributes, or static attribute value matching is adopted, so that the learning path recommendation accuracy of an adaptive learning system is low, and the real-time dynamic learning path recommendation function cannot be achieved are solved.

A small sample Wi-Fi forgery attack detection method based on meta-learning comprises the following steps: 1) data preprocessing: converting network traffic features into images, wherein the number of image channels is consistent with the hierarchical features of a protocol framework; and performing thermal coding processing on data labels; 2) deep feature extraction: processing data by using a convolutional neural network, and extracting spatial structure features; inputting the original features and the extracted features into a plurality of fully connected layers, and finally outputting a prediction task label; 3) meta-knowledge transfer of multiple auxiliary networks: constructing multiple auxiliary networks to learn meta-knowledge from historical WAID tasks to quickly adapt to new tasks; 4) method evaluation: randomly selecting two types from the three types to form a historical WAID task, and selecting a normal sample and a remaining new attack type to construct a new WAID task. The invention also comprises a system for detecting the small sample Wi-Fi forgery attack based on meta-learning. The method is suitable for the situation of few samples, and meets the requirements of resource constraint and real-time performance of the Internet of Things.

The invention discloses a method for automatically classifying, obtaining and storing complex knowledge of a high-end device. The method comprises: an automatic complex knowledge classification method of performing induction and reorganization on knowledge resources from the following three dimensions of the high-end device: a life cycle dimension, a knowledge manifestation pattern dimension and a knowledge theme dimension, and automatically classifying the knowledge resources by using a naive Bayes classifier; a complex knowledge obtaining method of obtaining a template according to complex knowledge based on a meta-knowledge model and obtaining the complex knowledge resources through semi-automatic obtaining technology based on the obtained template; and an automatic complex knowledge storage method of dividing the complex knowledge resources from the physics through a series of automatic division rules, compressing key information and storing the same in different storage spaces in a distributed manner. The method disclosed by the invention covers the automatic complex knowledge classification method, the complex knowledge obtaining method and the automatic complex knowledge storage method, and provides foundation and support for the high-end device manufacturers to use the complex knowledge resources.