166 results about "Gibbs sampling" patented technology

The invention discloses a network text segmenting method based on the genetic algorithm, used for segmenting short network texts. The method comprises the following steps of: evaluating a Latent Dirichlet allocation (LDA) model corresponding to a corpus by using a Gibbs sampling method, inferring latent topic information using the model, representing texts by using the latent topic information; then transforming a text-segmenting process into a multi-target optimum process by using a parallel genetic algorithm, and calculating the coherency of segmented units, the divergence among the segmented units and fitness functions by using deeper semantic information; and carrying out the genetic iteration of the text segmenting process, and determining whether the segmenting process terminates based on the similarity among multi-iteration results or the upper limit of iterations to obtain the global optimal solution for segmenting the texts. Therefore, the invention improves the accuracy for segmenting the short network texts.

In accordance with the principles of the present invention, an apparatus, simulation method, and system for modeling loss in a term structured financial portfolio are provided. An historical date range, time unit specification, maturity duration, evaluation horizon, random effects specification, and set of portfolio covariates are selected. Historical data is then segmented into infinitely many cumulative loss curves according to a selected covariate predictive of risk. The s-shaped curves are modeled according to a nonlinear kernel. Nonlinear kernel parameters are regressed against time units up to the maturity duration and against selected portfolio covariates. The final regression equations represent the central moment models necessary for prior distribution specification in the hierarchical Bayes model to follow. Once the hierarchical Bayes model is executed, the finite samples generated by a Metropolis-Hastings within Gibbs sampling routine enable the inference of net dollar loss estimation and corresponding variance. In turn, the posterior distributions enable the risk analysis corresponding to lifetime loss estimates for routine risk management, the valuation of derivative financial instruments, risk-based pricing for secondary markets or new debt obligations, optimal holdings, and regulatory capital requirements. Posterior distributions and analytical results are dynamically processed and shared with other computers in a global network configuration.

The invention provides a text-subject-model-based data processing method for commodity classification. The method comprises the following steps of: importing Chinese and English vocabulary related to a service into a universal word library of a word segmentation system, and importing white name English words related to the service for brands and common commodity English; further expanding a stop word library of the word segmentation system; segmenting words for a description text part of a commodity, so that each commodity can have a bag of words which is not related to sequence; counting word segmentation results to acquire uncommon vocabulary with high frequency, and thus constructing a preferential word library; and appointing a general classification quantity, setting related parameters, executing quick Gibbs sampling, acquiring potential semantic association, comparing the latent semantic association with the preferential word library, the universal word library and the stop word library respectively, calculating comparison results to obtain the most possible classification of the commodity, and marking the classification by using the bags of words. In consideration of latent semantics, the influence of subjective factors of editorial staff is reduced, so that the commodity classification is accurate.

The invention discloses a news event evolution analysis method based on time sequence distribution information and a topic model and relates to the field of text analysis. The method comprises the following steps: firstly, dividing a corpus into a plurality of sub-corpuses according to time by analyzing distribution characteristics, presented on a time sequence, of a news report, and by using a K-Means clustering algorithm; secondly, sequentially performing topic modeling on each sub-corpus by using the topic model, and learning the model through a Gibbs sampling method to obtain topic distribution information of each sub-corpus; finally, connecting topics between which the distance is minimum in series by calculating a Jensen-Shannon distance between each two topics in the adjacent sub-corpuses, wherein the topics are connected in series to obtain a main topic of an event, and auxiliary topics except the main topic in each sub-corpus are concerns and new developments of the event in each stage. According to the method, the mainline of event development in a news prediction and new concerns burst in each stage can be better described.

The invention relates to the field of ontology learning, in particular to an ontology concept and hierarchical relation generation method. The technical scheme includes that a PAM (probability aim model) is applied to extraction of ontology concept and hierarchical relation to improve effect of ontology concept and hierarchical relation learning so as to generate ontology concept more accurately and effectively. The method includes: establishing a PAM-based ontology concept and hierarchy generation model to effectively convert a domain ontology concept learning problem into a domain document set based statistic inference problem, and adopting a Gibbs sampling method to obtain a probability distribution feature vector; performing Wordnet-based semantic similarity calculation, generating a conception according to similarity correlation, and obtaining a set and hierarchical relation of ontology concepts. By the method, concept sets and hierarchical relation among concepts in domain ontology can be obtained more accurately and effectively.

The invention relates to the field of traditional Chinese medicine information search and discloses a traditional Chinese medicine data mining method based on an LDA (Latent Dirichlet Allocation) topic model. The method comprises the following specific steps: 1) determining two groups of priors, namely prescription-topic and topic-medicament, in the LDA model, and conducting priori assumptions to the two groups of priors by an AS (Asymmetry Symmetry) method, wherein the prescription-topic and the topic-medicament are determined respectively by Alpha and Beta; 2) determining the number of topics in the LDA model; 3) solving the LDA model by a Gibbs sampling method; 4) generating a semantic RDF (Resource Description Framework) document of the LDA model, mapping the result of the LDA model to a tetrad, and expressing the result with the semantic RDF document; 5) associating medicaments with prescriptions to build a prescription-topic-medicament with visual structure network G. The method has the advantages of being suitable for handling and mining a great quantity of traditional Chinese medicine prescriptions and capable of obtaining visual structure models.

The embodiment of the invention discloses a data processing method, a data processing device and a related device, for increasing the parameter solving speed and the parameter solving accuracy of an hLDA (hierarchical latent Dirichlet allocation) model. The method disclosed by the embodiment of the invention comprises the following steps of: sending the global initial global statistical information to various slave nodes, merging the local statistical information received from the slave nodes to obtain the new global statistical information, calculating the probability distribution between documents and themes and the probability distribution between themes and words according to the new global statistical information if the Gibbs sampling performed by the slave nodes is over, establishing and maximizing a likelihood function of a text set according to the calculated probability distributions to obtain new hLDA hyper-parameters, and calculating and outputting the probability distribution between the documents and the themes and the probability distribution between the themes and the words according to the new hLDA hyper-parameters if the iteration for solving the hLDA hyper-parameters is converged.

The invention relates to a microblog transmission group division and account activeness evaluation method based on a theme possibility model, aiming at effectively dividing according to transmission groups, dividing accounts participating in microblog topic transmission into a plurality of groups and quantifying active microblog accounts in each group. The method comprises: taking a keyword of a specific accident as the foundation and acquiring microblog specific text information and an account set participating in the microblog transmission; constructing a participating personnel list by text content of each microblog and account set input of the microblog; generating a model of each microblog in a sample library based on a theme probability; sampling words and participating personnel from a constructed word list and the participating personnel list according to group-theme, theme-word and group-human in the model; calculating by adopting a Gibbs sampling method; and adopting a merge sorting method to sort the words under each theme and people in each group. The method provided by the invention is simple and can be used for grasping microblog hot spots, emotional tendencies and public opinion guidance in time.

A multi-relational data set is represented by a probabilistic multi-relational data model in which each entity of the multi-relational data set is represented by a D-dimensional latent feature vector. The probabilistic multi-relational data model is trained using a collection of observations of relations between entities of the multi-relational data set. The collection of observations includes observations of at least two different relation types. A prediction is generated for an observation of a relation between two or more entities of the multi-relational data set based on a dot product of the optimized D-dimensional latent feature vectors representing the two or more entities. The training may comprise optimizing the D-dimensional latent feature vectors to maximize likelihood of the collection of observations, for example by Bayesian inference performed using Gibbs sampling.

The invention discloses a short text topic modeling method based on word semantic similarity. The method comprises: according to word semantic similarity provided by external, establishing a similar word set of short text centralized words; determining the number of topics used in modeling; randomly distributing the topic of each short text; through a Gibbs sampling process, iteratively determining the topic of each short text and the distribution of the words in the topic; according to a final distribution result of the above variable, feeding back the word distribution under each topic and the topic associated to each short text. The method preferably solves problems of sparse information contents of short texts and unclear semantic expression. According to the model result, short texts can be preferably expressed as topic vectors, and the topic vectors are used as final feature vectors of a short essay. The topic vector-based expression has good semantic interpretability, and can be used as algorithm basis of various applications. The method can be widely applied in various short text data, and has wide actual meaning and commercial values.

The invention discloses a multi-scale dictionary scene image classification method based on latent Dirichlet analysis and mainly aims to solve the problems that the manual marking workload is higher and the classification accuracy is lower by adopting a traditional classification method. The multi-scale dictionary scene image classification method based on the latent Dirichlet analysis comprises the implementation steps of respectively establishing a training set and a test set for natural scene image classification; extracting scale invariant features from the training set to generate a multi-scale dictionary; performing dictionary mapping on images by using the multi-scale dictionary, and generating multi-scale sparse representation vectors by using a BOW (bag of words model); generating a latent semantic topic model of the multi-scale sparse representation vectors by using a Gibbs sampling method to obtain latent semantic topic distribution of the images, and further building a natural scene image classification model; classifying the natural scene images by using the classification model. According to the latent Dirichlet analysis-based method for classifying the scene images by using the multi-scale dictionary disclosed by the invention, by adopting multi-scale features and the latent semantic topic model, the feature information of the images is enriched, a large amount of manual marking work is avoided, and the classification accuracy is improved. The multi-scale dictionary scene image classification method based on the latent Dirichlet analysis can be used for object identification and vehicle and robot navigation.

The invention relates to the field of noumenon learning, in particular to a noumenon concept attribute learning method based on Web information. By means of the technical scheme, Web serves as a language database, a language pattern is built to serve as a query set of a Google search engine, webpage fragment and corresponding source website URL are extracted to build a candidate concept attribute word bank, an URL built text set of candidate works serves as LAD input, training parameters of an LDA model are obtained by adopting a Gibbs sampling method, an attribute candidate bank is trimmed and combined according to operation results of the LDA model, and a final concept attribute work set can be determined. The noumenon concept attribute learning method of Web information can accurately and effectively obtain a concept attribute set in a noumenon, and accordingly, automatic or semiautomatic body building can be possible.

The invention discloses an expansion target tracking method based on GLMB (Generalized labelled multi-bernoulli) filtering and Gibbs sampling. The expansion target tracking method based on GLMB filtering and Gibbs sampling estimates the target number and the shape of the expansion target, provides a multiple expansion target tracking method under a labelled random finite sets (L-RFS) framework, and mainly includes two aspects: dynamic modeling of multiple expansion targets and tracking estimation of multiple expansion targets. The expansion target tracking method based on GLMB filtering and Gibbs sampling includes the steps: combined with a generalized label multi-bernoulli filter, establishing a measurement limit hybrid model of the expansion targets, by means of Gibbs sampling and Bayesian information criterion, deriving the parameters of the limit hybrid model to learn tracking of the state of the multiple expansion targets, using an equivalent measurement method to replace measurement generated from the expansion targets, and performing ellipse approximating modeling on the shape of the expansion targets to realize estimation of the shape of the expansion targets. The simulation experiment shows that the expansion target tracking method based on GLMB filtering and Gibbs sampling can effectively track the multiple expansion targets, can accurately estimate the state and theshape of the expansion targets, and can obtain the track of the targets.

The invention discloses a WSN (wireless sensor network) intra-network data fusion method based on kernel density estimation and non-parameter belief propagation, which comprises data acquisition and data fusion. The data acquisition is that monitoring unions which are respectively composed of no less than three sensor nodes for gathering the monitoring data are constructed in a monitoring region, each monitoring union is corresponding provided with a union header node for collecting the monitoring data, the sensor nodes in each monitoring union are respectively used for gathering the monitoring data of an object entering the monitoring region; and the data fusion is that the gathered monitoring data are subjected to KDE (kool desktop environment) processing by the sensor nodes in the monitoring unions respectively, the processed data are transmitted and collected to the union header nodes through NBP (name bind protocol) processing, the collected data are subjected to gauss mixing by the union header nodes, the data after gauss mixing are subjected to Gibbs sampling fusion, and the fused result is acted as a characteristic of the monitoring data. The accuracy of the monitoring data can be improved under a noisy or an uncertain environment, and the accurate fusion characterization of the monitoring data of the multi-node unions can be realized.

The present invention discloses a stand-alone residual life prediction method for integrating life data and performance degradation data, and relates to the field of product residual life prediction. The method comprises the following steps: (S1) initializing the drift parameter mu and the diffusion parameter sigma; (S2) updating the drift parameter mu and the diffusion parameter sigma; (S3) carrying out Gibbs sampling numerical simulation to solve the parameter estimation value; and (S4) carrying out residual life prediction based on the degradation process. The purpose of the present invention is providing a method, and when a component performance degradation process can be taken as a linear drift Wiener process, the method can be used for integrating life data and performance degradation data, estimating distribution parameters of the component residual life, and further estimating the product residual life, so that the problem of residual life prediction of the complex product with high reliability, long life and small sample is solved, and real-time updating of parameters is realized.

The invention discloses a probability calculation-based multiple input multiple output detector, which comprises a matrix QR composer, the matrix QR composer is respectively connected with two random sequence generators, the two random sequence generators are respectively connected with a probability complex multiplier, one of the probability complex multipliers is connected with a Gibbs sampling update unit, and both the other probability complex multiplier and the Gibbs sampling update unit are connected with a log-likelihood ratio calculation unit. Because the multiple input multiple output detector uses probability calculation to carry out the MCMC (Markov chain Monte Carlo) algorithm, the complexity of operation is greatly decreased, the transition probability of the Markov chain in the MCMC algorithm is increased, and the problem of locking under a high signal-to-noise ratio is solved. A sliding window sequence generation method is utilized to carry out Gibbs sampling update, so the length of a probability sequence is reduced. When the multiple input multiple output detector is applied to construct a full-parallel detector, a high throughput rate can be achieved with the full-parallel mode.

The invention discloses a matrix decomposition recommendation method based on Bayesian probability with social relations and project content. The method includes the steps that a PMF method is used for performing hidden matrix analysis on an observation evaluation matrix to obtain a hidden user characteristic matrix and a hidden project characteristic matrix; a BPMFSR method or a BPMFSRIC method is used for performing Gibbs sampling on the hidden user characteristic matrix and the hidden project characteristic matrix to obtain the hidden user characteristic matrix after sampling and the hidden project characteristic matrix after sampling; according to the hidden user characteristic matrix after sampling and the hidden project characteristic matrix after sampling, a forecast evaluation matrix is calculated, and recommendation is performed based on the forecast evaluation matrix. The method is efficient in calculation, can be applied to a recommendation system which has a large-scale data set and is based on trust or content, has a larger convergence rate, obtains more accurate recommendation results compared with other matrix decomposition methods and solves the problems of data sparseness and cold start better than other methods.

The invention relates to a microblog theme emotion analysis method based on mixed characteristic calculation. The method comprises the following steps of (1) preprocessing microblog data: extracting microblog characteristics by applying Chinese words segmentation, English stemming and emotion extraction technologies, and utilizing priori knowledge to carry out initial assignment on the emotion and the theme of the microblog characteristics; (2) initializing algorithm parameters; (3) utilizing a Gibbs sampling technology to carry out valuation on joint distributions A and H of the parameters of a multi-characteristic theme emotion model MfJST; and (4) judging emotion polarity of each microblog through the appearance probability of the emotion polarity in the microblog; and detecting an implied theme of a microblog message through carrying out marginalization about a microblog emotion variable on the joint distribution A. according to the method, the concealed theme concerned by a user and the concealed real emotion of the user in the microblog message can be effectively excavated, the method is applied to various social media, such as Twitter and Sina microblog, the online public opinion monitoring can be optimized, and the information service quality can be improved.

A classification method and apparatus, and a computer readable storage medium are disclosed. The method includes: feature data is input into a pre-established deep belief neural network model; throughmulti-thread parallel Gibbs sampling, the probability of mutual activation between a hidden layer and an explicit layer is calculated in parallel according to the sampling result, and the weight andbias value between neurons are updated in parallel; the updated weight and bias value between neurons are used as initialization training parameters to supervise the deep belief neural network model;and the trained deep belief neural network model is used for classification and recognition. The present application exerts the advantage of algorithm parallelism and improves the operation speed of the system by multithreaded parallel Gibbs sampling and parallel updating of the weight and bias values between the neurons.

An automatic annotating method for subjects of open source software comprises the steps as follows: open source project data is obtained, project labels are converted to roots of the project labels, then the labels of the identical roots are merged, and project descriptions are converted to word packets; names, the labels and the project descriptions of an open source project are taken as input, annotated LDA (Latent Dirichlet Allocation) models are applied, the input data is trained through a Gibbs sampling process, all labels and counts designated by certain words in the project descriptions are obtained after stabilization, and words are generated in label designation; a label network is constructed in the label designation according to the generated words, and semantic distances and semantic cohesion of points are calculated; in addition, a new project can be annotated automatically according to the constructed label network, the name and the description of any one project p are input, each word in the description is searched in the label network, respective label sets Li of each different word i in the description are obtained, one label 1i is selected from each Li, the semantic cohesion (Cohesion L) can be maximum, and the labels satisfying the conditions are annotated to the new project automatically.

The invention discloses a geostatistics-based wind power station wind speed spatio-temporal data modeling method. The method comprises the following steps: 1, constructing a space structure matrix of a wind power station according to geographical location information of fans, space convariance function of wind power station wind speed and variation function of wind power station wind speed to represent the spatial correlation of input wind speed of each fan; 2, performing layered modeling on the input wind speed of each fan by a universal Kriging method and a Bayesian algorithm and estimating model parameters by adopting Gibbs sampling; and 3, predicting the forward P steps of the wind speed of each fan to acquire simulation samples of P step forward prediction distribution of the wind speed of each fan, and sampling and averaging to acquire the optimal prediction result of the wind speed of each fan, wherein P is more than or equal to 1. By the method, the time and space correlation of wind speed data of different fan positions is comprehensively analyzed based on the physical characteristics of wind, so that a more accurate prediction model is built for the whole wind power station, and the prediction result is better than that of the traditional method.

The invention discloses a radar target identification method based on a label maintaining multitask factor analyzing model and mainly solves the problem that the prior art is poor in target identification performance under a small sample condition. The radar target identification method includes the steps of firstly, performing normalization and alignment pretreatment on a radar high range resolution profile; secondly, using the preprocessed high range resolution profile to build a label maintaining multitask factor analyzing model; thirdly, performing Gibbs sampling on the parameters of the model, and saving sampling average of the model parameters; fourthly, performing normalization and alignment pretreatment on a to-be-tested sample; fifthly, calculating the frame probability density function value of the to-be-tested sample according to the sampling average, learned by the training steps, of the parameters of the label maintaining multitask factor analyzing model; sixthly, judging the category attribute of the to-be-tested sample according to the frame probability density function value. The radar target identification method has the advantages that supervised learning of the model is achieved, the identification performance under the small sample condition is increased, and the method is applicable to the radar target identification under the small sample condition.

The invention discloses a cross-modal searching method based on a topic model. The cross-modal searching method based on the topic model comprises the following steps that (1) feature extracting and label recording are conducted on all types of modal data in a database, (2) a cross-modal searching image model based on the topic model is built, (3) the cross-modal searching image model based on the topic model is solved by using a Gibbs sampling method, (4) a user submits data of a type of modal, and another relative type of modal data are returned by using the cross-modal searching model after feature extracting, and (5) the real corresponding information and the label information of the cross-modal data are used, and the cross-modal searching model is assessed from two aspects of correspondence and differentiation. The cross-modal searching method based on the topic model introduces the cross-modal topic and the different modal topic improving concept, the labor information is used, the interpretability and the flexibility of the topic modeling are improved and the cross-modal searching method based on the topic model has a good expandability and discrimination performance.

Methods of training Boltzmann machines include rejection sampling to approximate a Gibbs distribution associated with layers of the Boltzmann machine. Accepted sample values obtained using a set of training vectors and a set of model values associate with a model distribution are processed to obtain gradients of an objective function so that the Boltzmann machine specification can be updated. In other examples, a Gibbs distribution is estimated or a quantum circuit is specified so at to produce eigenphases of a unitary.

The invention discloses a face recognition method based on structuralized factor analysis. The problem that the face recognition accuracy rate is low due to the fact that local clustering features and a global distribution structure of face image data cannot be kept in the prior art is mainly solved. The face recognition method comprises the achieving steps of 1 dividing a face image data set; 2 performing clustering analysis on all training sample data; 3 calculating an optimal feature projection matrix through Gibbs sampling; 4 extracting low-dimension features of all test sample data and the training sample data; 5 recognizing a face image. The face recognition method synthesizes the local clustering features and the global distribution structure of the face image data and improves the face recognition accuracy rate.

The present invention provides a technique for extracting depth-domain seismic wavelets by using depth-domain seismic data and using the depth-domain seismic wavelets to directly synthesize depth-domain seismic records, in particular, a depth-domain seismic wavelet extraction and seismic record synthesis integrated method. According to the method, the variation characteristics of seismic waveletsin space are considered, and depth-domain seismic records containing effective information in an entire stratum can be obtained. The method of the invention comprises the following main steps that: (1) depth, velocity and density information is obtained on the basis of the logging data of a certain well, and a corresponding reflection coefficient r is calculated according to the depth, velocity and density information; (2) in order to satisfy a requirement for wavelet extraction under a linear time invariant condition, with a constant velocity vc adopted as a reference velocity, constant velocity depth transformation is performed on the well side seismic trace S of the well and the reflection coefficient r, so that a transformed trace S1 and reflection coefficient r1 can be obtained, and depth-domain seismic wavelets w are extracted on the basis of a Gibbs sampling method; and (3) with a logging velocity v, the reflection coefficient r, and the extracted depth-domain wavelets w directly adopted as input, a depth-domain synthetic seismic record can be made.

The invention discloses a state switching prediction method and system based on a Markov state transition model. The method comprises the steps that a time sequence data stream is obtained, a time sequence of a financial data stream is filtered based on a prediction cycle, the autoregression Markov state transition model is established, a conditional probability density function of all variables is solved, Gibbs sampling is carried out to adjust the model, parameters and a state sequence of the model are estimated, the state transition probability is estimated according to the state sequence, and the current state is estimated and market expectation is predicted accordingly. According to the state switching prediction method and system based on the Markov state transition model, a large amount of invalid information in data is removed by filtering the time sequence, a style switching predicting cycle is better matched with a data cycle, different states are fit with different models, state sequence estimation and state transition probability estimation are divided into two steps, the Gibbs sampling method is used for carrying out estimation, the calculation complexity is lowered, responding delay is reduced, and data mining and analyzing can be reasonably, effectively and quickly carried out on the financial data stream.

The invention discloses a multi-mobile-robot high-precision collaborative tracking method based on an ultra-wideband technology. The method comprises the following steps: a multi-mobile-robot collaborative tracking experiment platform is built under an ROS; a multi-node ranging network is built through an ultra-wideband sensor, and information of the distances between robots and between the robots and an anchor point is obtained at the same time; an ultra-wideband ranging error weakening algorithm based on Bayesian filtering is provided for effectively weakening LOS and NLOS errors of obtained ranging values to restore real distance values; position information of the mobile robots is estimated by adopting a cooperative tracking algorithm, namely, a collaborative particle filtering algorithm based on Gibbs sampling; and real tracks of corresponding motions of the robots are obtained by adopting an OptiTrack motion capturing system, and a collaborative tracking algorithm is evaluated. The LOS and NLOS errors in a complex environment can be effectively weakened; the restored distances are real; the position information of each robot at any moment can be accurately determined; and multi-robot collaborative tracking is realized.

The invention discloses a classifying method for aliasing audio events. According to the method, an author-subject model theta and a subject-word model phi are obtained through a Gibbs sampling method in the training stage. In the testing stage, according to the author-subject model theta and the subject-word model phi obtained in the training stage, aliasing audio events are classified by solving uncertainty values of audio documents relative to aliasing audio classes. The classifying method for aliasing audio events can classify and recognize multiple audio events in audio samples, not just a certain audio event, so that content of audio documents is analyzed better. Currently, research on classification of aliasing audio events is little, and the classifying method is a good supplement of the research on classification of aliasing audio events.

The invention belongs to the technical field of radars, and discloses a radar high-resolution range profile target recognition method based on an iMMFA (infinite max-margin factor analysis) model so as to improve classification performance of the radar and reduce solving complexity of the model. The method comprises steps: radar high-resolution range profiles of M categories of targets; feature extraction is carried out on each radar high-resolution range profile; an iMMFA model is built; an initial value of each parameter of the iMMFA model is set, according to a Gibbs sampling rule, after a preheating process of I0 times of iteration, sampling values of each parameter at T0 times are stored respectively; a final FA model, a final TSB-DPM model and a final LVSVM classifier are determined, and a final iMMFA model is formed; a radar high-resolution range profile of a test target is acquired, and feature extraction is carried out on the radar high-resolution range profile of the test target; and according to a clustering label of a test hidden variable as described in the description, a target category to which the test target belongs is determined.