148 results about "Structure learning" patented technology

The invention discloses a wireless cognitive network knowledge base constructing method based on a Bayesian network. A Bayesian network model is constructed by using perceived network performance parameters, so that the dependence among the parameters is obtained, and the conditional probability dependence is transformed into knowledge in a wireless cognitive network to be stored in a knowledge base for guiding a process of intelligent decision making. The process of constructing the knowledge base mainly comprises the steps of: perceiving the network performance parameters, discretizing and analyzing the parameters, constructing the Bayesian network model and transforming the knowledge to construct the knowledge base. The network performance parameters are collected through a perceiving process, and data is discretized and analyzed. Through a structure learning and parameter learning process, the Bayesian network model is constructed by using historical data. The conditional dependence among the parameters in the Bayesian network is transformed into the knowledge so that the knowledge base is constructed and updated.

The invention discloses a coke furnace temperature predicting method based on dynamic working conditions in a coke furnace heating and burning process. The method is characterized by comprising the following steps: 1, acquiring gas flow and coke furnace temperature data in various historical working conditions, and establishing a sample bank; 2, calculating the similarity of a current working point and a sample in the sample bank on the basis of an included angle of Euclidean distance and variation trend of the current working point and the sample in the sample bank; and 3, selecting a plurality of sample structural learning sets with maximum similarity; establishing a learning-set-based local linear model by adopting an iterative least square method; and calculating an output value corresponding to the current working point as a coke furnace temperature predicted value in the coke furnace heating and burning process. The coke furnace temperature predicting method based on dynamic working conditions in the coke furnace heating and burning process has high prediction accuracy, and has the online adaptive capability.

The invention provides a recommendation method based on situation fusion sensing. The recommendation method comprises the following steps that 1, the situation is divided into physical situations and user preference situations according to the definition and the requirements of the situations; 2, a Bayes network is built through parameter learning and structure learning, and the physical situation matching degree in a certain environment is ratiocinated and calculated; 3, through considering the dynamics of hobbies and interests of users along with the time change, a time function is merged into a recommendation algorithm based on the content, and the matching degree of the user preference situation is calculated; 4, the situation matching degree is comprehensively considered, all candidate information resources are graded, and in addition, information ranking in first Top-N is recommended to target users. Compared with the prior art, the recommendation method provided by the invention has the advantages that the considered recommendation factors are more comprehensive, the method can better adapt to changeful environment, the recommendation accuracy is improved, in addition, the condition that the interest of the users is changed along with the time change is considered, the time function is combined with the recommendation based on the resource content, and the user satisfaction degree is improved.

Provided is a semantic image segmentation method based on conditional random field graph structure learning. The method includes the following steps: 1) training a fully convolutional neural network or adopting an existing fully convolutional neural network to perform semantic image coarse segmentation; 2) using an rcf neural network to learn a conditional random field graph structure; 3) trainingconditional random field model parameters through the graph structure obtained by learning and using a conditional random field model obtained through training to perform semantic image segmentation,wherein the step of semantic image segmentation is as follows: solving a Maximum A Posteriori inference problem and finding an optimal label of x by calling an alpha-beta extension routine. The invention provides the semantic image segmentation method with a good segmentation effect.

The invention discloses an SAR image segmentation method based on ridgelet filters and a convolution structure model. The SAR image segmentation method based on ridgelet filters and a convolution structure model mainly solves the problem that in the prior art, segmentation of SAR images is not accurate. The SAR image segmentation method based on ridgelet filters and a convolution structure model includes the following steps: 1) sketching an SAR image, and obtaining a sketch image; 2) according to an area image of the SAR image, dividing the pixel subspace of the SAR image; 3) constructing a ridgelet filter set; 4) constructing a convolution structure learning model; 5) utilizing the SAR image segmentation method based on the ridgelet filters and the convolution structure model to segment the pixel subspace of a hybrid aggregation structure natural object; 6) based on the gathering feature of sketch lines, performing segmentation of an independent object; 7) based on visual sense semantic rules, performing segmentation of line object; 8) based on polynomial logic regression prior model, segmenting the pixel subspace of a formal area; and 9) combining the segmentation results. The SAR image segmentation method based on ridgelet filters and a convolution structure model can acquire good segmentation effect of SAR images, and can be used for semantic segmentation of the SAR images.

The invention provides a structure extended polynomial naive Bayes text classification method. Firstly, a one-dependence polynomial estimator is established by using each word that occurs in a test document as a father node and then all the one-dependence polynomial estimators are subjected to weighted averaging to predict a category of the test document, wherein the weight is an information gain ratio of each word. According to the method, the structure learning phase of a Bayesian network is avoided, thereby reducing time spending brought by high dimensionality of text data; and meanwhile, the estimation process of a dual conditional probability is postponed to the classification stage, thereby ingeniously saving large space cost. According to the method, not only is classification accuracy of a polynomial naive Bayes text classifier improved, but also time spending and space cost of structure learning of the Bayesian network are avoided.

An artificial immune method for constructing a brain effect connection network from fMRI data. On the basis of a biological immune system, an artificial immune system combined with the fMRI data is disclosed and can be used for construction of the brain effect connection network. The artificial immune method particularly comprises the following steps of: carrying out experimental design, i.e. performing functional magnetic resonance scanning by using a resting-state experiment; carrying out fMRI data acquisition, i.e. under the condition of reducing a head movement and other errors as further as possible, carrying out scanning to obtain fMRI image data; carrying out pre-processing, i.e. performing pre-processing on the data by using a statistical method, and removing errors and noise which are caused by partial outside factors; selecting a region in which the user is interested, and selecting a brain region related to the study; constructing the effect connection network by a method of optimizing Bayesian network structure learning by using the artificial immune system, and searching the effect connection network matched with an fMRI data set by means of the network structure learning; and carrying out analysis, i.e. analyzing the constructed network and mining biological characteristics exposed by a network structure.

The invention discloses an object recognition method of a vision mechanism based robust object structural learning method. The method includes a training process and a recognition process and includes the steps: subjecting target objects with marked types and positions in images to information feedback of a vision mechanism, and training to obtain a feedback model; and preliminarily predicating object types and object positions of objects in a to-be-recognized image, and using the feedback model obtained by training to robustly learning structural information of the target objects. Since robust object structures and the vision mechanism have invariance in object recognition, the vision mechanism based robust object structural learning method is adopted for improving object recognition precision, and types and positions of targets in recognition scenes can be accurately recognized by the method. In addition, the object recognition method can be widely applied to safety inspection, internet search, digital entertainment and the like.

The invention discloses a short text sentiment analysis method based on a sum product network depth autocoder. The method comprises the following steps of 1, preprocessing short text data; 2, utilizing a doc2vec model to train sentence vectors; 3, utilizing the sum product network depth coder to code the sentence vectors, and obtaining layered abstract characteristics of the sentence vectors; 4, utilizing a maximum product network depth decoder to decode the layered abstract characteristics, comparing the decoded characteristics with the primary sentence vector characteristics, calculating a reconstruction error, adjusting parameters of the sum product network depth autocoder to make the reconstruction error smallest, obtaining an optimal sum product network depth coder, and obtaining an optimal layered abstract characteristic by the optimal sum product network depth coder; 5, utilizing the optimal layered abstract characteristic to conduct online structure learning to generate a sum product network structure, using a small amount of short text data with tags to finely adjust a sum product network, using an online parameter learning algorithm to continuously adjust network parameters, inputting a test set, and obtaining sentiment classifications through the trained sum product network.

The present disclosure provides an improved approach to implement structure learning of neural networks by exploiting correlations in the data/problem the networks aim to solve. A greedy approach is described that finds bottlenecks of information gain from the bottom convolutional layers all the way to the fully connected layers. Rather than simply making the architecture deeper, additional computation and capacitance is only added where it is required.

The invention discloses an equipment failure Bayesian network prediction method based on a K2 algorithm, and is used for solving the technical problem of low searching efficiency of a conventional equipment failure Bayesian network prediction method. The equipment failure Bayesian network prediction method has the technical scheme that an FPBN (Failure Prediction Bayesian Network) structure learning algorithm based on a K2 searching algorithm is adopted for building an FPBN structure capable of really reflecting each variable incidence relation in a failure data set, so that an FPBN model is built. Finally, the actual operation state of equipment is predicted by uitlizing a parameter learning algorithm on the basis of a built failure prediction model. The method uses the K2 searching algorithm as the basis; the failure knowledge, the expertise and the failure data are effectively fused; and the problem of modeling difficulty in system to FPBN conversion in the equipment prediction process is solved. In addition, the FPBN-K2 algorithm calculation process totally adopts deterministic searching algorithms, and repeated searching for many times is not needed; the searching space is reduced; the number of scoring function calculation times is reduced; and the searching efficiency of the FPBN structure learning algorithm is improved.

The invention discloses an on-line time series data prediction method based on fuzzy inference, a system and a storage medium. Training process: extracting the time series data of the training samples, inputting the time series data of the training samples into the LSTM fuzzy neural network model based on interval type II, outputting each sample, first carrying out the structure learning, and thencarrying out the parameter learning on the basis of the structure learning; the trained LSTM fuzzy neural network model based on interval type 2 being obtained. Forecasting process: extracting the time series data of the sample to be forecasted, inputting the time series data of the sample to be forecasted into the trained LSTM fuzzy neural network model based on interval type II, and outputtingthe forecasting result.

The invention discloses a personal data analysis method based on a Bayesian network and a computer storage medium, and the method comprises the following steps: (1) enabling personal life behavior data to be embodied as a one-dimensional vector of behaviors and behavior attributes, enabling the behavior attributes to at least comprise a time attribute, and obtaining a life behavior data record through data preprocessing; (2) learning the data through a hybrid structure learning algorithm, and constructing a life data Bayesian network; (3) parameter learning is carried out according to the lifedata Bayesian network, and a conditional probability distribution table of each network node is obtained through learning; and (4) calculating the probability of occurrence of other behaviors based on the probability of the specific behavior by using a joint tree reasoning algorithm according to the life data Bayesian network, and completing the analysis and prediction of the personal life behavior. According to the method, the Bayesian network is applied to personal behavior data analysis, and the network construction method is improved, so that the learning accuracy and the convergence of the algorithm are effectively improved, and the operation performance is improved.

The present invention provides an intergenic interaction relation excavation method based on Bayesian network reasoning. The method comprises the following steps of: 1, employing a method of estimation of entropy by employing a Gaussian kernel probability density estimation quantity to calculate interaction information between genes, between genes and phenotypic characters and between phenotypes and the phenotypic characters; 2, employing a three-stage dependence analysis Bayesian network structure learning method to construct a Bayesian network including genes and phenotypic character nodes;3, employing the Bayesian estimation parameter learning method to perform parameter learning to obtain a contingent probability form between nodes; and 4, employing a Gibbs sampling Bayesian network approximate reasoning method to calculate the contingent probabilities of genes with different quantities and the phenotypic characters, and obtaining an intergenic interaction relation influencing thespecial phenotypic characters according to the calculation result. The intergenic interaction relation excavation method based on Bayesian network reasoning can help biology researchers of obtainingof epistasis genetic locuses influencing the special phenotypic characters to assist in gene function excavation and provide reference for hereditary basis analysis of complex quantitative charactersof different species.

The invention discloses an image adaptive denoising method based on an attention mechanism, and the method comprises the construction of image denoising and a convolutional neural network based on theattention mechanism, and the network is mainly composed of two parts which respectively complete the noise image extraction and noise image weight analysis. A natural image is used as the input of the network, wherein the noise image extraction part completes extraction of a noise image, noise image weight analysis is of an action structure, the weight of noise distribution is learned to obtain aweight map of noise, and finally, the noise image extraction part and the noise image analysis part are combined to obtain a denoised image by combining the noise image and the input noise image. Thedepth image convolutional neural network is based on an attention mechanism, and the learning efficiency of the network is adaptively improved. According to the method, denoising work can be carriedout on the image containing noise, and a good visual effect is achieved.

The invention discloses a dark network clue detection method based on a heterogeneous graph attention neural network. The method comprises the steps 1, performing text collection on a dark network; step 2, extracting event titles, keywords and entities for the acquired dark network text information, and constructing a dynamic heterogeneous information network; step 3, performing embedding processing on the nodes in the constructed heterogeneous information network, and feature vectors of the nodes are obtained; 4, learning the graph structure of the heterogeneous information network; and step5, according to a result obtained by learning the graph structure of the heterogeneous information network, performing clue category classification on nodes in the heterogeneous information network, so that clue detection of the dark network information is completed. According to the method, an external knowledge base is used as a support, and two sets of methods are adopted to learn the graph structure of the constructed heterogeneous information network, so that a good clue detection effect is achieved.

The invention relates to an electroencephalogram (EEG) signal online identification method with data structure information being fused. The method comprises the following steps: S1) establishing a classification model based on an online sequential extreme learning machine (OS-ELM) algorithm by utilizing a small training set formed by a small number of labeled EEG samples to serve as an initial classification model in semi-supervised learning; S2) establishing a structure learning model by utilizing an on-line fuzzy clustering method, and estimating a global structure of data distribution afterbatch increase of EEG samples collected online based on prior information of the labeled EEG samples; S3)carrying out labeling on the EEG samples collected online by utilizing the classification model, and through a batch learning mode and based on the structure information estimated by a structural learning model, selecting a batch of EEG samples collected online and meeting a certain conditionsto add to a training set, and re-training the classification model by utilizing the updated training set; and S4) carrying out online identification on the collected EEG signals through the updated classification model.

The embodiment of the invention provides a Bayesian structure learning method and device of a deep neural network. The method comprises the steps that a deep neural network comprising a plurality of learning units with the same internal structure is constructed, each learning unit comprises a plurality of hidden layers, a plurality of calculation units are included between the hidden layers, the network structure is the relative weight of each calculation unit, and parameterized variational distribution is adopted to model the network structure; a training subset is extracted, and a network structure is sampled by adopting a re-parameterization process; an evidence lower bound is calculated; and if the change of the evidence lower bound exceeds the loss threshold, the network structure andthe network weight are optimized, and new training is started. According to the embodiment of the invention, a deep neural network comprising a plurality of learning units with the same internal structure is constructed; and the relative weight of each calculation unit between each hidden layer in the learning unit is trained through the training set to obtain an optimized network structure, thereby bringing comprehensive improvement to the prediction performance and prediction uncertainty of the deep neural network.

The invention discloses an edge cloud anomaly detection method based on network structure learning. The method comprises the following steps of cloud computing center data acquisition, network structure learning, and edge cloud anomaly detection and early warning. The cloud computing center data acquisition refers to performing network topology structure construction and feature extraction on edgeclouds; the network structure learning refers to learning and training the constructed network structure; and the edge cloud anomaly detection and early warning means that anomaly prediction is performed on the edge cloud by using the learned network structure, and the cloud system is notified of the node which predicts anomaly for early warning. According to the method, in the aspect of predicting the edge clouds with abnormal behaviors, the independence hypothesis of a traditional method is broken through, and the possible relevance between the edge clouds is considered by learning a network structure, so that the purpose of improving the edge cloud anomaly detection accuracy is achieved. The method is of great help to edge cloud anomaly detection and security assurance in a cloud computing system, and has a very high application value.

The invention discloses a reservoir flood regulation multi-dimensional uncertainty risk analysis method based on a Bayesian network. The reservoir flood regulation multi-dimensional uncertainty risk analysis method comprises the following steps: identifying risk factors of regulation starting water level uncertainty and flood forecasting uncertainty; carrying out bayesian network structure learning based on an expert experience method; carrying out bayesian network parameter learning to acquire a conditional probability table (CPT) of each node; performing Bayesian network probability reasoning; carrying out risk calculation and analysis. In order to couple the influence of regulation starting uncertainty and reservoir flood forecasting errors on reservoir flood control scheduling risks, areservoir flood control risk analysis model based on the Bayesian network is established, and coordinated conversion of reservoir interest benefits and flood control risks in the flood season can beachieved; the bidirectional reasoning of the Bayesian network can establish a reservoir scheduling risk bidirectional analysis and evaluation mode, and has a good application prospect.

A depth combined structuring and structuring learning method for mankind behavior identification comprises the following steps: 1, forming combined structure and structure formulation; 2, using a space network to extract deep convolution nerve network characteristics from a human body area in an image, using fc6 layer output of the space network as depth characteristics, using gradient histogram and optical flow histogram characteristics to further enhance characteristic expression, connecting CNN, HOG and HOF characteristics so as to express personal behavior or interaction relations in the image, using said characteristics to train two linearity support vector machine classifiers for each data set, and using combined characteristics to calculate combined characteristics in the formula (1); 3, training model parameters; 4, training and related inference in prediction, aiming at each training case to solve loss and enhance inference in each iteration period of the training process. The method is suitable for images of various behavior types, and can identify interaction behaviors.

The invention provides a high-resolution remote sensing image semantic segmentation method sharing multi-scale adversarial features, a multi-scale adversarial network model is introduced, and multi-scale detail information of a remote sensing image is well described by using multi-scale structure learning of the adversarial features; meanwhile, a discriminator of the adversarial network model is improved, is used as a relationship enhancement module; the correlation and boundary information of the target ground object are further described; on one hand, correlation between pixels in the same ground object can be expressed, and on the other hand, edge pixels of each image are associated with pixels of other two or more ground object categories around the image, so that better spatial continuity and boundary accuracy of a target ground object are obtained, and the boundary and semantic accuracy of a remote sensing image prediction result is improved; besides, the adversarial features ofthe method can be flexibly embedded into different semantic segmentation reference models, have good migration application capability, and can correspondingly improve the performance.

The invention belongs to the technical field of fault diagnosis of hydrometallurgical leaching processes, and in particular relates to a DCD-based hydrometallurgical leaching process fault diagnosis method. The DCD-based leaching process fault diagnosis method is mainly used for a hydrometallurgical leaching process and is characterized by extracting information in expert knowledge and process data as the prior information to establish a dynamic causality diagram knowledge base; activating an inference diagnosis mechanism after an abnormal situation is observed; calculating the posterior probability of each possible fault cause by using the abnormal situation as an evidence; and obtaining a diagnostic result by comparing the posterior probabilities. The algorithm mainly includes the stepsof leaching process DCD event determination, DCD structure learning, DCD parameter learning and DCD online process fault diagnosis. The method processes the uncertainty of information in the leachingprocess by using the DCD fault diagnosis technology, reduces the dependence of the diagnostic technology on a large amount of data to a certain extent, can bring more accurate diagnosis results, and ensures the economic benefit and the production benefit of enterprises.

The invention discloses a cause-and-effect structure learning method based on flow characteristics. The method comprises following steps of 1: generating and distributing any new characteristic in a flow manner; 2: carrying out a correlation analysis on each of newly generated characteristics; 3: carrying out a redundancy verification analysis on a characteristic set; 4: carrying out searching orientation based on each of the characteristics; 5: repeating steps of 1-4 until the numbers of the generated characteristics exceed a limit value, finally obtaining a corresponding cause-and-effect structure. According to the invention, a cause-and-effect structure relation can be found in linearly randomly distributed data with flow characteristics and time complexity of learning is reduced, thereby satisfying timeliness requirements of online learning.