Multi-view dynamic migration clustering method and system based on gaussian mixture model

Abstract

The application discloses a multi-view dynamic migration clustering method and system based on a Gaussian mixture model, and the method comprises the following steps: obtaining a multi-view data set to be clustered, presetting initial values of view weights of each view and dynamic migration learning factors between the views; clustering data samples in each view respectively, and calculating initial values of Gaussian mixture model parameters of each view and membership degrees of each data in each view to each cluster according to clustering results; updating the Gaussian mixture model parameters, the view weights and the dynamic migration learning factors according to the membership degrees; taking the updated values as the latest values, and iteratively calculating the membership degrees of each view until the iteration ends; and performing clustering learning between the views according to the dynamic migration learning factors obtained through the final iteration, so that clustering results of the views tend to be consistent. Through the dynamic migration learning technology, the application realizes in-depth mining of complementary information between the multi-views, and improves clustering performance of the multi-view data.

Description

Technical Field

[0001] This invention relates to the field of multi-view clustering technology, and in particular to a multi-view dynamic migration clustering method and system based on Gaussian mixture model. Background Technology

[0002] With the rapid development of cloud computing and big data, uncovering the potential information in massive amounts of data has become a crucial means for enhancing the core competitiveness of enterprises. Cluster analysis is a widely used unsupervised machine learning method, extensively applied in fields such as image segmentation, content retrieval, and data mining. Cluster analysis methods divide a set of data into different clusters based on the similarity between data points. Based on the structural characteristics of cluster partitioning, clustering algorithms mainly include prototypical clustering, density clustering, and hierarchical clustering. Gaussian Mixture Model (GMM) is currently the most widely used and representative model clustering algorithm. This model clustering algorithm assumes that each cluster is a single Gaussian distribution, using a model composed of a mixture of multiple finite Gaussian distributions to represent the overall data samples. Due to the simplicity and reliability of the GMM solution algorithm and its rigorous mathematical proof, it has been widely applied in many scientific fields, such as object recognition, image segmentation, signal processing, streaming data processing, and real-time systems.

[0003] With the rise of multimedia technology, information sources and data presentation methods have become increasingly diverse and multi-sourced. For example... Figure 1 As shown, a news story may be reported by multiple countries in multiple languages, accompanied by a large amount of images, videos, and text materials; similarly, an image can be represented by features in multiple ways; and an autonomous vehicle needs to collect information from multiple fields such as radar, video, and sound. This makes single-view data composed of a single feature set no longer suitable for modern multi-source or multi-view data. Therefore, multi-view clustering methods are needed to address this data trend. Currently, common multi-view clustering methods include multi-view clustering based on collaborative training, multi-kernel multi-view clustering, multi-view subspace clustering, and multi-view graph clustering. Among them, collaborative multi-view clustering, as a method that is simple to implement, easy to apply, and has considerable computational complexity, focuses more on mining complementary information between views during the iteration process compared to other methods. The collaborative multi-view clustering algorithm can maximize the consistency of multiple views and achieve clustering by utilizing prior knowledge and information interaction between views. However, most existing collaborative training algorithms are based on partitioning and clustering extensions, and their accuracy is often lower than that of model-based clustering algorithms such as GMM. Moreover, multi-view collaborative algorithms do not consider the differences between different views, cannot make good use of the potential information between multiple views, and often require a large number of parameters to ensure effectiveness.

[0004] With the wide application of transfer learning in the field of deep learning, the training efficiency of the target model can be improved by transferring the parameters of the pre-trained model to the target model. However, the traditional clustering algorithm generally has poor clustering effect when dealing with a data set with insufficient samples. Therefore, researchers introduce the concept of transfer learning and propose a variety of new transfer clustering algorithms. The basic idea of these algorithms is to use the knowledge of source domain clustering to improve the clustering performance of target domain data. Although transfer learning can effectively solve the problem of poor clustering effect of traditional clustering methods in the case of insufficient samples, most transfer learning clustering algorithms based on Gaussian mixture model are for clustering single-view data, and are not suitable for clustering multi-view data. There is no transfer learning clustering algorithm based on Gaussian mixture model for clustering multi-view data in the prior art. SUMMARY

[0005] To solve the above problems of the prior art, the present application provides a multi-view dynamic transfer clustering method based on Gaussian mixture model and system, which can realize in-depth mining of complementary information between multi-views and improve the clustering performance of multi-view data through dynamic transfer learning technology.

[0006] In a first aspect, the present disclosure provides a multi-view dynamic transfer clustering method based on Gaussian mixture model.

[0007] A multi-view dynamic transfer clustering method based on Gaussian mixture model comprises:

[0008] Obtain a multi-view data set to be clustered, and preset the initial values of the view weight of each view and the dynamic transfer learning factor between views;

[0009] Cluster the data samples in each view data set respectively to obtain the clustering results of each view;

[0010] According to the clustering results, calculate the initial values of the Gaussian mixture model parameters of each view;

[0011] According to the Gaussian mixture model parameters, calculate the membership degree of each data sample in each cluster in each view;

[0012] Update the Gaussian mixture model parameters, view weights and dynamic transfer learning factors according to the membership degrees;

[0013] Take the updated Gaussian mixture model parameters, view weights and dynamic transfer learning factors as the latest values, and perform membership degree iterative calculation on each view until the iteration ends;

[0014] According to the dynamic transfer learning factor obtained through the final iteration, perform clustering learning between views to obtain a clustering result consistent among views.

[0015] In a second aspect, the disclosure provides a multi-view dynamic migration clustering system based on a Gaussian mixture model.

[0016] A multi-view dynamic migration clustering system based on a Gaussian mixture model comprises:

[0017] A data acquisition module is configured to acquire a multi-view data set to be clustered.

[0018] A data preprocessing module is configured to preset initial values of view weights of each view and dynamic migration learning factors between views, cluster data samples in each view data set respectively, acquire clustering results of each view, and calculate initial values of Gaussian mixture model parameters of each view according to the clustering results.

[0019] A membership iterative calculation module is configured to calculate, according to the Gaussian mixture model parameters, a membership of each data sample in each cluster in each view, update the Gaussian mixture model parameters, the view weights and the dynamic migration learning factors according to the membership, and perform membership iterative calculation on each view with the updated Gaussian mixture model parameters, the view weights and the dynamic migration learning factors as the latest values until the iteration ends.

[0020] A multi-view dynamic migration clustering module is configured to perform clustering learning between views according to the dynamic migration learning factors obtained through final iteration, and obtain clustering results of the views tending to be consistent.

[0021] In a third aspect, the disclosure further provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein when the computer instructions are run by the processor, the steps of the method of the first aspect are completed.

[0022] In a fourth aspect, the disclosure further provides a computer readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the steps of the method of the first aspect are completed.

[0023] The above one or more technical solutions have the following beneficial effects:

[0024] 1. The present application provides a multi-view dynamic migration clustering method and system based on a Gaussian mixture model, which takes membership as a migration object and introduces a learning factor to control migration learning between multiple views, promotes in-depth mining of potential information between multiple views, solves the problem that traditional multi-view clustering algorithms cannot fully mine complementary information between views, and improves multi-view clustering performance.

[0025] 2. The application discloses a dynamic transfer learning method, designs a parameter self-adjusting strategy, and enables a learning factor to be automatically adjusted in a clustering process, thereby improving universality of the method and solving the problem of poor feasibility of a multi-view clustering algorithm under influence of initial parameters.

[0026] 3. The application automatically optimizes weight distribution of the views by view weighting and a maximum entropy regularization method, suppresses excessive influence of unreliable views on transfer learning and global clustering, and further improves clustering performance of the multi-view data. BRIEF DESCRIPTION OF DRAWINGS

[0027] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application, and are incorporated in and constitute a part of this application. The drawings illustrate one exemplary embodiment of the present application and, together with the description, serve to explain the application.

[0028] Figure 1 FIG. 1 is a schematic diagram of a multi-view data set;

[0029] Figure 2 FIG. 5 is a flowchart of a multi-view dynamic transfer clustering method based on a Gaussian mixture model according to an embodiment of the application;

[0030] Figure 3 FIG. 6 is a schematic diagram of a dynamic transfer learning technique according to an embodiment of the application. DETAILED DESCRIPTION

[0031] It should be noted that the following detailed description is merely exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0032] It is to be understood that the terminology used herein is for the purpose of describing the particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, devices, components and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components and / or combinations thereof.

[0033] Embodiment One

[0034] In view of the problem that the traditional multi-view clustering algorithm cannot sufficiently mine the complementary information between views, the embodiment provides a multi-view dynamic transfer clustering method based on a Gaussian mixture model (Multi-View Adaptive Transfer Clustering Algorithm based on Gaussian Mixture Model, MAT-GMM) with strong universality and high clustering precision based on the characteristics of Gaussian mixture clustering and multi-view data. The method takes membership as a transfer object, introduces a dynamic transfer learning factor to control the transfer learning between multiple views, and then realizes in-depth mining of the potential information between multiple views, thereby improving the precision and performance of multi-view clustering. The method comprises the following steps:

[0035] obtaining a multi-view data set to be clustered, and presetting initial values of view weights of each view and dynamic transfer learning factors between views;

[0036] performing clustering on data samples in each view data set respectively to obtain clustering results of each view;

[0037] obtaining initial values of Gaussian mixture model parameters of each view according to the clustering results;

[0038] obtaining membership of each data sample in each view to each cluster according to the Gaussian mixture model parameters;

[0039] updating the Gaussian mixture model parameters, the view weights and the dynamic transfer learning factors according to the membership;

[0040] performing iterative calculation of the membership for each view by taking the updated Gaussian mixture model parameters, the view weights and the dynamic transfer learning factors as the latest values, until the iteration ends;

[0041] performing clustering learning between views according to the dynamic transfer learning factors obtained through the final iteration to obtain clustering results consistent between views.

[0042] As shown in the following specific steps, the multi-view dynamic transfer clustering method based on the Gaussian mixture model is further illustrated. Figure 2

[0043] Step S1, obtaining a multi-view data set to be clustered, and presetting initial values of view weights of each view and dynamic transfer learning factors between views. Specifically, a set of multi-view data sets containing L views to be clustered X = {X1,..., XL} is read in, and for each view, the lth view data set is Xl = {xl, xl,..., xl}, where xl = {xl, xl,..., xl} represents the lth view data set, and xl represents the lth data sample in the lth view data set. l L l l1 ​​​​,...,x li ,...,x lI}and Where I represents the number of data samples in the current view dataset, x li D represents the i-th data sample in the l-th view. l This represents the data dimension of the l-th view.

[0044] With a preset cluster number K, Φ = {φ lk The set of parameters for the Gaussian mixture model of the MAT-GMM algorithm is defined as l = 1,…,L,k = 1,…,K, and φ is the set of parameters for the Gaussian mixture model. lk ={μ lk ,Σ lk}, where μ lk Σ represents the cluster mean of the k-th class in the l-th view. lk Let w represent the cluster covariance matrix of the k-th class in the l-th view; the initial weights of each view are preset to w. l =1 / L, preset maximum number of iterations T, and set the iteration counter t to 0, preset dynamic transfer learning factor ξ. ll′ The initial value is 1 / (L-1).

[0045] Step S2: Cluster the data in each view dataset to obtain the clustering results for each view. Specifically, perform the K-means++ algorithm on the data samples in each view dataset to obtain the initial clustering results.

[0046] Step S3: Based on the clustering results of each view, calculate the initial values ​​of the Gaussian mixture model parameters for each view, i.e., calculate the mean and covariance matrix of the data samples for each class in each view. Specifically, calculate the mean of the data samples for the k-th class (or cluster, where a cluster is the representation of a "class" in clustering) in the l-th view. Using this as the initial cluster mean in the MAT-GMM algorithm, the covariance matrix of the data samples of the k-th class (or cluster) in the l-th view is calculated. This is used as the initial cluster covariance matrix for the MAT-GMM algorithm, thereby obtaining the initial Gaussian mixture model parameters Φ for the MAT-GMM algorithm. (0) Furthermore, the inverse data sample covariance matrix for each cluster can be calculated.

[0047] Increment the iteration counter t by 1 to start the first iteration. During the t-th iteration, execute step S4.

[0048] Step S4: Based on the clustering results of each view and the initial values ​​of the Gaussian mixture model parameters obtained in step S3, calculate the membership degree of each data sample in each view to the K clusters in the current view. Specifically, taking the l-th view as an example, calculate the membership degree u of the i-th data sample to the k-th cluster in this view one by one. lik Finally, the membership matrix U of the view is obtained. A dynamic transfer learning factor is used to control knowledge transfer between multiple views.

[0049] In the above steps, the membership degree u of the i-th data sample belonging to the k-th cluster is calculated for each view. lik It includes the following two steps:

[0050] Step S4.1: Calculate the membership degree of each data sample in each view to multiple clusters in the current view using the following formula, that is, calculate the membership degree u of the i-th data sample to the k-th cluster. lik The formula is:

[0051]

[0052] In the above formula, ω l Let f be the weight of the l-th view (initially 1 / L), where D represents the dimension of the view, and f is the weight of the l-th view. lk Represents a function. Indicates w l f lk When calculating, first calculate f. lk Then calculate f lk w l to the power of, and thus obtain Similarly, Indicates ξ ll' w l' f l'k ξ ll′ p is the learning factor for knowledge transfer from the l-th view (i.e., the current view) to the l-th view (i.e., other views); l′ represents other views besides the l-th view; ll′ik The Lagrange multipliers necessary for updating membership (which are set to 0 in the initial state) are β, which is the membership entropy parameter with a value range of [0.1,1]; 1≤l≤L, 1≤i≤I, 1≤k≤K, {l′|1≤l′≤L and l′≠l};

[0053] Step S4.2: Based on membership degree, use formula p ll'ik =p ll'ik +ρ(u lik -u l'ik) update the Lagrange multipliers one by one, and update the Lagrange multipliers as the Lagrange multipliers in the next iteration process, where p is an iteration step length, which is set to 10 in this embodiment -6 .

[0054] Step S5, in combination with the membership obtained in step S4, updating the Gaussian mixture model parameters, view weights and dynamic transfer learning factors, specifically including the following steps:

[0055] Step S5.1, according to the calculated membership, calculating and updating the model parameter set φ lk in each view one by one to obtain the clustering center matrix Φ. Specifically, the calculation formula of each cluster mean in each view is:

[0056]

[0057] In the formula, u lik and u l′ik are the membership of the sample points obtained in the step S4, where u lik is the membership of the i-th sample point (i.e. data sample) in the l-th view with respect to the k-th clustering cluster, and u l′ik is the membership of the i-th sample point in the l'-th view with respect to the k-th clustering cluster; 1≤l≤L, 1≤i≤I, 1≤k≤K, {l'|1≤l'≤L and l'≠l};

[0058] The calculation formula of each cluster covariance matrix in each view in the above step S4 is:

[0059]

[0060] In the formula, u lik and u l′ik are the membership of the sample points obtained in the step S4, where u lik is the membership of the i-th sample point in the l-th view with respect to the k-th clustering cluster, and u l′ik is the membership of the i-th sample point in the l'-th view with respect to the k-th clustering cluster; 1≤l≤L, 1≤i≤I, 1≤k≤K, {l'|1≤l'≤L and l'≠l}.

[0061] Step S5.2, according to the calculated membership matrix and clustering center matrix, calculating and updating the view weight w l of each view one by one to obtain the view weight vector Ω. Wherein, the calculation formula of the view weight is:

[0062]

[0063] In the formula, w lLet be the weight of the l-th view, and η be a non-negative regularization parameter, which is set to 10 in this embodiment. -4 ; 1≤l≤L, 1≤i≤I, 1≤k≤K, {l′|1≤l′≤L and l′≠l};

[0064] Step S5.3: Based on the updated view weights, calculate and update the dynamic transfer learning factor between the current view and other views, that is, calculate and update the transfer learning factor ξ of the l-th view and the l′-th view. ll′ Specifically, firstly, based on membership degree and updated view weight, the formula is used... Calculate the approximate global membership degree of multiple views; then, based on the membership degree and global membership degree of the l-th view and the l′-th view, use the formula... and formula Calculate the KL divergence values ​​between the membership degrees of the l-th view and the global membership degree of the l′-th view, respectively; then, use the formula... The ratio τ of the two KL divergences is calculated, serving as an important basis for evaluating the knowledge reference value among multiple views. Finally, based on the KL divergence ratio and the initial dynamic transfer learning factor, the formula ξ is used to... ll′ =ξ ll′ +θξ ll′ [1-2 (1-τ) The dynamic transfer learning factor between the l-th view and the l′-th view is calculated and updated one by one, where θ is the update step size greater than 0, and is taken as 10 in this embodiment. -3 .

[0065] Step S6: Using the updated Gaussian mixture model parameters, view weights, and dynamic transfer learning factor as the latest values, iteratively calculate the membership degree for each view until the iteration ends. In step S6, the updated Gaussian mixture model parameters, view weights, and dynamic transfer learning factor are used as the latest values ​​to perform iterative calculations in steps S4 and S5 until the iteration ends, obtaining the final dynamic transfer learning factor. The criterion for this iteration is: calculating the objective function value F obtained in the t-th iteration. (t) Based on the objective function value F obtained in the t-th iteration (t) The objective function value F obtained in the (t-1)th iteration (t-1) Judge the difference between them. If the difference satisfies |F (t) -F (t-1) If the difference is less than ε, or the number of iterations t > T, then the iteration terminates; otherwise, the iteration continues. Here, ε is the iteration convergence threshold.

[0066] In the above steps, the objective function value F for the t-th iteration is calculated. (t)The formula is:

[0067]

[0068] Step S7, clustering learning between views is performed according to the dynamic transfer learning factor obtained in the final iteration, and clustering results of the views tend to be consistent. As shown in Figure 3 , the three views can be regarded as source domains and target domains in the clustering process, and transfer learning occurs between any two views. As shown in Figure 3 , when the first view view1 clusters, it needs to learn knowledge from the second view view2 and the third view view3 to promote its own clustering, at this time, view1 can be regarded as a target domain, and view2 and view3 are regarded as source domains; when view2 and view3 cluster, they also need to learn complementary knowledge from view1, at this time, view1 can be regarded as a source domain, and view2 and view3 are regarded as target domains. With the execution of clustering, each view will switch between the source domain and the target domain until the clustering task is completed. In this embodiment, the current view is regarded as a source domain, and other views are regarded as target domains. According to the dynamic transfer learning factor between the current view and other views obtained by calculation, the clustering results of other views are learned, and then the results of the views tend to be consistent, improving the accuracy and clustering performance of the clustering results. For example, three kinds of fruits, apples, bananas and strawberries, are data visualized according to shape, color and nutritional value, and divided into three views. In each view, the data in each view is divided into 3 categories (i.e. 3 categories of apples, bananas and strawberries) by using the above method of this embodiment. Since the data in a single view may be close to two categories that are difficult to distinguish during classification, for example, in the "shape" view, the data (such as area data) of apples and bananas may be close and cannot be well distinguished. At this time, the dynamic transfer learning factor between the "shape" view and other views is obtained by the above method of this embodiment, and the classification result of the "shape" view can be learned from the classification result of apples and bananas in the "color" view (the classification of red apples and yellow bananas is clear). The degree of learning depends on the size of the learning factor. In this way, each view learns from each other, and finally the membership degree of all views is consistent.

[0069] In view of the problem that the traditional multi-view clustering algorithm cannot fully mine the complementary information between views, based on the characteristics of Gaussian mixture clustering and multi-view data, a new multi-view transfer learning method is proposed. The method takes membership as the transfer object, introduces a learning factor to control the transfer learning between multiple views, thereby promoting the in-depth mining of the potential information between multiple views.

[0070] The degree of transfer learning should be different due to the diversity of views in multi-view data. In a multi-view transfer clustering algorithm, a learning factor is a global hyperparameter. Once the parameter is determined, the learning rate of each view to other views is fixed, so that the advantage of transfer learning cannot be fully played in the iteration process. In order to solve the problem that the multi-view clustering algorithm has poor feasibility under the influence of initial parameters, the embodiment proposes a dynamic transfer learning method, and designs a parameter self-adjusting strategy, so that the learning factor can be automatically adjusted in the clustering process, and the universality of the method is improved.

[0071] At the same time, the learning rate between views affects the clustering performance of the current view, and the performance of multi-view clustering is usually greatly affected by part of the views. However, the more complex the collection equipment or source of multi-view data is, the more obvious the diversified characteristics of the data are, and there may be unreliable views among them. In this case, the clustering accuracy is often not high due to the influence of unreliable views. Therefore, considering the difference between views, the embodiment further explores the weighting of views, and automatically optimizes the weight distribution of the views by using the maximum entropy regularization method, so as to suppress the excessive influence of unreliable views on transfer learning and global clustering, thereby further improving the clustering performance of multi-view data.

[0072] Embodiment two

[0073] The embodiment provides a multi-view dynamic transfer clustering system based on a Gaussian mixture model, comprising:

[0074] A data acquisition module is configured to acquire a multi-view data set to be clustered.

[0075] A data preprocessing module is configured to preset initial values of view weights of each view and dynamic transfer learning factors between views; cluster data samples in each view data set respectively to obtain clustering results of each view; and calculate initial values of Gaussian mixture model parameters of each view according to the clustering results.

[0076] A membership iteration calculation module is configured to calculate membership degrees of each data sample in each view to each cluster according to the Gaussian mixture model parameters; update the Gaussian mixture model parameters, the view weights and the dynamic transfer learning factors according to the membership degrees; and perform membership iteration calculation on each view by taking the updated Gaussian mixture model parameters, the view weights and the dynamic transfer learning factors as the latest values until the iteration ends.

[0077] A multi-view dynamic transfer clustering module is configured to perform clustering learning between views according to the dynamic transfer learning factors obtained through final iteration, so as to obtain clustering results consistent among the views.

[0078] Embodiment three

[0079] The embodiment provides an electronic device, comprising a memory and a processor and computer instructions stored in the memory and running on the processor, when the computer instructions are run by the processor, steps in the multi-view dynamic migration clustering method based on a Gaussian mixture model are completed.

[0080] Embodiment four

[0081] The embodiment also provides a computer readable storage medium for storing computer instructions, when the computer instructions are executed by a processor, steps in the multi-view dynamic migration clustering method based on a Gaussian mixture model are completed.

[0082] The steps and methods in the above embodiments two to four correspond to the method embodiment one, and the specific embodiments can be understood with reference to the related description part of the embodiment one. The term "computer readable storage medium" should be understood as including a single medium or multiple media of one or more instruction sets; and should also be understood as including any medium capable of storing, encoding or carrying instruction sets for execution by a processor and causing the processor to perform any method in the present application.

[0083] Those skilled in the art should understand that each module or step of the present application described above can be realized by a general computer device, alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device for execution by a computing device, or they can be respectively manufactured into each integrated circuit module, or a plurality of modules or steps among them can be manufactured into a single integrated circuit module to realize. The present application is not limited to any specific combination of hardware and software.

[0084] The above only describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0085] The above describes the specific embodiments of the present application in combination with the accompanying drawings, but is not used to limit the protection scope of the present application, and those skilled in the art should understand that various modifications or changes made on the basis of the technical solutions of the present application without creative labor are still within the protection scope of the present application.

Claims

1. A multi-view dynamic migration clustering method based on Gaussian mixture model, characterized in that, The method comprises the following steps: obtaining a multi-view data set to be clustered, and presetting initial values of view weights of each view and dynamic transfer learning factors between views; respectively clustering data samples in each view data set to obtain clustering results of each view; according to the clustering results, calculating initial values of Gaussian mixture model parameters of each view; According to the Gaussian mixture model parameters, a membership of each data sample in each view belonging to each cluster is calculated; a membership of an i-th data sample in an l-th view belonging to a k-th cluster u lik is calculated, and a calculation formula is: In the above formula, ω l is the view weight of the lth view, and D denotes the dimension of the view, denotes w l f lk , denotes ξ ll' w l' f l'k ; ξ ll′ is the learning factor of the lth view dynamically migrated from the l'th view; l' is other view except the first view; p ll′ik is the Lagrange multiplier, β is the membership entropy item parameter with the value range of [0.1, 1]; 1≤l≤L, 1≤i≤I, 1≤k≤K, {l'|1≤l'≤L and l'≠l}; L is the number of views; I denotes the number of data samples in the current view dataset; K is the number of clusters; x li represents the ith data sample in the lth view; μ lk represents the cluster mean of the kth cluster in the lth view, Σ lk represents the cluster covariance matrix of the kth cluster in the lth view; updating the Gaussian mixture model parameters, the view weights and the dynamic transfer learning factors according to the membership degrees; iteratively calculating the membership degrees of each view by taking the updated Gaussian mixture model parameters, the view weights and the dynamic transfer learning factors as the latest values until iteration is completed; performing clustering learning between the views according to the dynamic transfer learning factors obtained through the final iteration to obtain clustering results consistent among the views.

2. The Gaussian mixture model based multi-view dynamic migration clustering method of claim 1, wherein, The Gaussian mixture model parameters comprise cluster mean values and cluster covariance matrices; the cluster mean values are mean values of data samples of each class after clustering in each view, and the cluster covariance matrices are covariance matrices of data samples of each class after clustering in each view.

3. The Gaussian mixture model based multi-view dynamic migration clustering method of claim 1, wherein, The updating of the Gaussian mixture model parameters, the view weights and the dynamic transfer learning factors according to the membership degrees comprises the following steps: calculating and updating a Gaussian mixture model parameter set in each view according to the calculated membership degrees to obtain a clustering center matrix; calculating and updating the view weights of each view according to the calculated membership degree matrix and the clustering center matrix to obtain a view weight vector; the calculated membership degree matrix is specifically obtained by calculating the membership degree of an ith data sample belonging to a kth cluster in each view, and finally obtaining a membership degree matrix of the view; calculating and updating the dynamic transfer learning factors between the current view and other views according to the updated view weights.

4. The Gaussian mixture model based multi-view dynamic migration clustering method of claim 3, wherein, The calculating and updating of the dynamic transfer learning factors between the current view and other views comprises the following steps: calculating global membership degrees of the multiple views based on the membership degrees and the updated view weights; calculating KL divergence values of the membership degrees and the global membership degrees of an lth view and an l'th view based on the membership degrees and the global membership degrees of the lth view and the l'th view; calculating a ratio of the two KL divergence values; calculating and updating the dynamic transfer learning factors between the lth view and the l'th view based on the ratio of the KL divergence values and the initial dynamic transfer learning factors.

5. The Gaussian mixture model based multi-view dynamic migration clustering method of claim 1, wherein, The iteration criterion comprises the following steps: calculating a target function value obtained through the current iteration; judging the difference between the target function value obtained through the current iteration and a target function value obtained through the last iteration, and terminating the iteration if the difference is less than a set iteration convergence threshold or the current iteration number is greater than a set number; otherwise, the iteration is continued.

6. A multi-view dynamic shift clustering system based on Gaussian mixture model, characterized in that, The method comprises the following steps: a data acquisition module is configured to obtain a multi-view data set to be clustered; a data preprocessing module is configured to preset initial values of view weights of each view and dynamic transfer learning factors between views; respectively clustering data samples in each view data set to obtain clustering results of each view; according to the clustering results, calculating initial values of Gaussian mixture model parameters of each view; The membership iterative calculation module is configured to calculate the membership of each data sample in each view belonging to each cluster according to the Gaussian mixture model parameters; the membership of the i th data sample in the l th view belonging to the k th clustering cluster is u lik The calculation formula is: In the above formula, ω l is the view weight of the lth view, and D denotes the dimension of the view, denotes w l f lk , denotes ξ ll' w l' f l'k ; ξ ll′ is the learning factor of the lth view dynamically migrated from the l'th view; l' is other view except the first view; p ll′ik is the Lagrange multiplier, β is the membership entropy item parameter with the value range of [0.1, 1]; 1≤l≤L, 1≤i≤I, 1≤k≤K, {l'|1≤l'≤L and l'≠l}; L is the number of views; I represents the number of data samples in the current view data set; K is the number of clusters; x li represents the ith data sample in the lth view; μ lk represents the cluster mean of the kth class in the lth view, Σ lk represents the cluster covariance matrix of the kth class in the lth view; the Gaussian mixture model parameters, the view weights and the dynamic transfer learning factor are updated according to the membership; the membership of each view is iteratively calculated with the updated Gaussian mixture model parameters, the view weights and the dynamic transfer learning factor as the latest values, until the iteration ends. The multi-view dynamic transfer clustering module is used for clustering learning between the views according to the dynamic transfer learning factor obtained in the final iteration, so as to obtain a clustering result in which the views tend to be consistent.

7. The Gaussian mixture model based multi-view dynamic migration clustering system of claim 6, wherein, The membership is used to update the Gaussian mixture model parameters, the view weights and the dynamic transfer learning factor, including: According to the calculated membership, the Gaussian mixture model parameter set in each view is calculated and updated one by one to obtain a clustering center matrix; According to the calculated membership matrix and the clustering center matrix, the view weight of each view is calculated and updated one by one to obtain a view weight vector; the calculated membership matrix is specifically: the membership of the i th data sample belonging to the k th cluster in the view is calculated one by one to finally obtain the membership matrix of the view; According to the updated view weight, the dynamic transfer learning factor between the current view and other views is calculated and updated.

8. An electronic device, characterized by comprising: The computer instructions are stored in the memory and run on the processor, and when the computer instructions are run by the processor, the steps of the multi-view dynamic transfer clustering method based on the Gaussian mixture model in any one of claims 1-5 are completed.

9. A computer readable storage medium characterized by, The computer instructions are stored in the memory and run on the processor, and when the computer instructions are run by the processor, the steps of the multi-view dynamic transfer clustering method based on the Gaussian mixture model in any one of claims 1-5 are completed.

Images