Incomplete multi-view clustering method, storage medium and processing device

Abstract

The present application discloses an incomplete multi-view clustering method, storage medium and processing device, including: constructing an incomplete multi-view dataset, a multi-view model for mining view consistency features and a single-view model for mining view specific features, wherein the incomplete multi-view dataset includes multiple complete samples and multiple incomplete samples; using complete samples to train the multi-view model and mine view consistency features, and using the view consistency features mined by the multi-view model to train the single-view model; using the trained single-view model to obtain the clustering results of the incomplete samples, and using the trained multi-view model to obtain the clustering results of the complete samples. The present application can effectively utilize incomplete information by constructing multiple models, while avoiding the instability of interpolation quality that brings noise to the original data. In addition, clustering effective features can be extracted, thereby improving the applicability and generalization ability of multi-view clustering.

Description

Technical Field

[0001] The present application relates to the technical field of unsupervised clustering, and in particular to an incomplete multi-view clustering method, a storage medium, and a processing device. Background Art

[0002] With the development of multimedia technology, people can observe and describe the same object or problem from different angles or methods, resulting in the emergence of a large amount of multi-view data. For example, in medical images, the same organ may be viewed in different views through different imaging technologies such as CT scans, magnetic resonance imaging, and ultrasound; in behavior recognition, the same action can be captured by cameras at different angles. The emergence of these multi-view data provides new opportunities for a more comprehensive and accurate understanding of complex objects. As an important unsupervised learning method, multi-view clustering aims to explore common clustering patterns between multi-view data and has been widely used in practical applications such as multimedia analysis, recommendation systems, and bioinformatics. In real scenarios, due to factors such as the instability of acquisition storage devices and human omissions, the data collection of multiple views is easily incomplete, resulting in incomplete multi-view data, that is, some samples do not have data from all views. When some views are missing, the natural alignment properties of multiple views will be destroyed, which is not conducive to mining complementary and consistent information across views. In addition, the lack of views will lead to serious information loss, which in turn aggravates the information imbalance between different views. Therefore, the problem of incomplete multi-view clustering or incomplete multi-view clustering has gradually attracted attention.

[0003] For incomplete multi-view clustering, the earliest method used was to fill the samples with missing features with the mean, but this method may introduce noise information or change the data distribution. With the continuous deepening of research, traditional incomplete multi-view clustering methods can be further divided into matrix decomposition-based methods, kernel learning-based methods, and graph learning-based methods. The above methods focus on interpolating, restoring, and inferring missing information in incomplete multi-view data, and also introduce regularization terms on the pre-constructed graph to obtain better interpolation features, so as to better explore clustering information. However, the above methods are limited in feature representation capabilities and complexity, and are not enough to effectively mine the deep knowledge hidden in the data.

[0004] In recent years, deep incomplete multi-view clustering has attracted attention due to its good performance by combining clustering with the representation learning ability of deep models. It uses the generalization ability of deep models to interpolate or predict missing data. However, the above methods have unstable interpolation quality and do not fully consider the differences and importance between different views, which makes it difficult to divide consistency, thus limiting the effectiveness of the model. Summary of the invention

[0005] The purpose of this application is to provide an incomplete multi-view clustering method, storage medium and processing device to solve the technical problems in the prior art that the quality of the existing multi-view clustering interpolation is unstable and the differences and importance between different views are not fully considered. The preferred technical solutions among the many technical solutions provided in this application can produce many technical effects as described below.

[0006] To achieve the above objectives, this application provides the following technical solutions:

[0007] The first aspect of the present application provides an incomplete multi-view clustering method, comprising: constructing an incomplete multi-view dataset, Figure 1 A multi-view model for mining consistent features and a single-view model for mining view-specific features, wherein the incomplete multi-view dataset includes a plurality of complete samples and a plurality of incomplete samples; the multi-view model is trained using the complete samples and the view-specific features are mined; Figure 1 Consistency features are obtained by using the multi-view model to mine the Figure 1 The single-view model is trained based on the consistent features; the clustering results of the incomplete samples are obtained by using the trained single-view model, and the clustering results of the complete samples are obtained by using the trained multi-view model.

[0008] In some embodiments, the method of using the complete sample to train the multi-view model and mine the view Figure 1 The method comprises: extracting view-specific features of each view in the complete sample and calculating a loss value based on the reconstruction loss; performing consistency optimization on the view-specific features extracted by the multi-view model to obtain the view Figure 1 The multi-view model is trained by minimizing the loss value until the multi-view model converges.

[0009] In some embodiments, the view mined using the multi-view model Figure 1 The method comprises: optimizing the view-specific features extracted by the single view model, and optimizing the view-specific features extracted by the single view model based on the view Figure 1 The single view model is trained by minimizing the loss value until the single view model converges.

[0010] In some embodiments, the step of obtaining the clustering result of the incomplete sample using the trained single-view model and obtaining the clustering result of the complete sample using the trained multi-view model includes: using a K-means clustering algorithm to cluster the views mined from the complete sample by the trained multi-view model; Figure 1The consistent features are processed to obtain the clustering results of the complete sample; and the feature-optimized view-specific features obtained from the incomplete sample by the trained single-view model are processed using a K-means clustering algorithm to obtain the clustering results of the incomplete sample.

[0011] In some embodiments, the method further includes: if the incomplete sample includes at least two views, clustering is performed using a single view model corresponding to each view separately, and a clustering result with the highest confidence is obtained as the clustering result of the incomplete sample.

[0012] In some embodiments, the single-view model includes a single-view feature extraction part and a feature optimization part, wherein the single-view feature extraction part performs feature extraction on the view of the complete sample and / or the view of the incomplete sample, and the feature optimization part performs feature optimization on the features extracted by the single-view feature extraction part; the multi-view model includes a multi-view feature extraction part and a consistency optimization part, wherein the multi-view feature extraction part performs feature extraction on the view of the complete sample, and the consistency optimization part performs consistency optimization on the features extracted by the multi-view feature extraction part.

[0013] In some embodiments, constructing an incomplete multi-view dataset includes: constructing a multi-view dataset, wherein the multi-view dataset includes multiple samples; randomly selecting a number of the samples and randomly discarding at least one view of the samples according to a preset missing rate to obtain the incomplete multi-view dataset.

[0014] In some embodiments, the method further comprises: recording available information and missing information of the view in an indication matrix, wherein the missing information is information of the randomly discarded view, and the available information is information of the remaining views in the sample.

[0015] A second aspect of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed, the incomplete multi-view clustering method as described above is implemented.

[0016] The third aspect of the present application provides a processing device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors are used to execute the one or more computer programs stored in the memory, so that the one or more processors perform the incomplete multi-view clustering method as described above.

[0017] Implementing one of the above technical solutions of the present application has the following advantages or beneficial effects: In the present application, a multi-view model is trained using complete samples and visual Figure 1 Consistent features, while using the multi-view model to mine the Figure 1 The single-view model is trained based on the consistent features, and finally the clustering results of complete samples and incomplete samples are obtained using the multi-view model and the single-view model respectively. In this case, multiple models can be constructed to effectively utilize incomplete information while avoiding the instability of interpolation quality that may bring noise to the original data. In addition, in view of the differences between different views, a single-view model that mines view-specific features and a single-view model that mines view-specific features can be constructed. Figure 1 The multi-view model with consistent features can extract effective features for clustering, thereby improving the applicability and generalization ability of multi-view clustering. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. It is obvious that the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0019] Figure 1 is a flow chart of an incomplete multi-view clustering method according to an embodiment of the present application;

[0020] Figure 2 It is a structural block diagram of the processing device of an embodiment of the present application.

[0021] In the figure: 1. processing device; 10. memory; 11. processor. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present application clearer, the various exemplary embodiments to be described below will refer to the corresponding drawings, which constitute a part of the exemplary embodiments, wherein various exemplary embodiments that may be used to implement the present application are described. Unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation methods described in the following exemplary embodiments do not represent all implementation methods consistent with the present disclosure. It should be understood that they are only examples of processes, methods, and devices that are consistent with some aspects disclosed in the present application as detailed in the attached claims, and other embodiments may also be used, or the embodiments listed herein may be modified in structure and function without departing from the scope and essence of the present application.

[0023] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", etc. indicate the orientation or positional relationship based on the drawings, which is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the elements referred to must have a specific orientation, be constructed and operated in a specific orientation. The terms "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. The term "multiple" means two or more. The terms "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, an integral connection, a mechanical connection, an electrical connection, a communication connection, a direct connection, an indirect connection through an intermediate medium, and can be the internal connection of two elements or the interaction relationship between two elements. The term "and / or" includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0024] In order to illustrate the technical solution described in the present application, a specific embodiment is provided below, and only the parts related to the embodiment of the present application are shown.

[0025] The embodiments of the present application can be applied to various application scenarios that require cluster analysis of multi-view data.

[0026] like Figure 1 As shown, the present application provides an incomplete multi-view clustering method, comprising the following steps (step S1 to step S3):

[0027] S1. Construct an incomplete multi-view dataset for mining visual Figure 1 A multi-view model for mining consistent features and a single-view model for mining view-specific features, where the incomplete multi-view dataset includes multiple complete samples and multiple incomplete samples.

[0028] In some embodiments, view-specific features may refer to features that are meaningful only for a specific view or perspective; Figure 1 A consistent feature may be a feature that remains consistent across multiple views or perspectives.

[0029] In some embodiments, constructing an incomplete multi-view dataset may include: constructing a multi-view dataset, the multi-view dataset including multiple samples; randomly selecting a number of samples and randomly discarding at least one view of the number of samples according to a preset missing rate to obtain an incomplete multi-view dataset.

[0030] In some embodiments, each sample may include at least one view.

[0031] In some embodiments, the multi-view dataset may be represented by the following formula:

[0032] ,

[0033] in, Can indicate the number of views.

[0034] In some embodiments, The data of a view can be expressed by the following formula:

[0035] ,

[0036] in, Can indicate the The number of samples for each view, Can indicate the The dimension of the samples in a view.

[0037] In some embodiments, the preset missing rate can be set according to actual needs. For example, the preset missing rate can be 0.4, 0.5, 0.6 or 0.7, etc. When the preset missing rate is 0.5, 50% of the samples can be randomly selected from the multi-view dataset, and at least one view of these samples can be randomly deleted.

[0038] In some embodiments, a complete sample may refer to a sample containing all view information; accordingly, an incomplete sample may refer to a sample with missing view information. In some embodiments, at least one view in a complete sample may be deleted to obtain an incomplete sample. That is, for a complete sample containing V views, the number of views contained in the incomplete sample may be less than V.

[0039] In some embodiments, the incomplete multi-view clustering method may further include: recording the available information and missing information of the view in an indicator matrix, wherein the missing information is the information of the randomly discarded view, and the available information is the information of the remaining views in the sample. Specifically, in the indicator matrix In The sample in is available in a view, then , otherwise The record is 0. This makes it easier to distinguish views later.

[0040] In some embodiments, the single-view model may include a single-view feature extraction part and a feature optimization part, wherein the single-view feature extraction part may extract features from the views of the complete sample and / or the views of the incomplete sample, and the feature optimization part may optimize the features extracted by the single-view feature extraction part. The single-view feature extraction part may adopt an autoencoder architecture, which may be . Among them, Fc can represent a fully connected layer, and the subscript number of Fc can represent the network mapping dimension of this layer. In this architecture, data can pass through four fully connected layers in sequence, and finally get a 32-dimensional vector output. Rectified Linear Unit (ReLU) can be introduced between each fully connected layer to learn more complex features.

[0041] In some embodiments, a corresponding single view model may be constructed for each view, for example, a corresponding single view model may be constructed for each view of an incomplete multi-view dataset. The view corresponding to the single view model may be a view of a complete sample and / or a view of an incomplete sample, that is, the input of the single view model may be a view of a complete sample or a view of an incomplete sample. In some embodiments, the decoder of the single view model may be symmetrical with the encoder of the view corresponding to the single view model.

[0042] In some embodiments, the feature optimization part may adopt a multilayer perceptron (MLP) network architecture. The multilayer perceptron network may include a three-layer structure. The multilayer perceptron network may be Specifically, the network architecture can be composed of three fully connected layers, the input of the network undergoes three linear transformations, and the output dimensions are 32, 256, and 32 respectively.

[0043] By building a single-view model through the above architecture, it is possible to reduce the misleading parts of the features extracted by the single-view model, so that the single-view model can focus on features that are beneficial to clustering.

[0044] In some embodiments, the multi-view model may include a multi-view feature extraction part and a consistency optimization part, wherein the multi-view feature extraction part may perform feature extraction on the views of the complete sample, and the consistency optimization part may perform consistency optimization on the features extracted by the multi-view feature extraction part. The architecture of the multi-view feature extraction part may be the same as that of the single-view feature extraction part. Since the sample dimensions of each view in the multi-view model are different, different network structures may be used to extract features from multiple views respectively.

[0045] In some embodiments, the input of the multi-view model may be all views of the complete sample, that is, the multi-view model may process V views simultaneously. In some embodiments, the decoder of the single-view model may be symmetrical with the encoder of the view corresponding to the single-view model.

[0046] In some embodiments, the architecture of the consistency optimization part can be the same as that of the feature optimization part. The consistency optimization part can integrate and optimize the features extracted from different views by the multi-view feature extraction part. Thus, the consistency and complementary information between different views can be explored through the multi-view model to extract cluster-friendly features.

[0047] S2. Use complete samples to train multi-view models and mine visual Figure 1 Consistent features, using the multi-view model to mine the Figure 1 Consistent features for training single view models.

[0048] In some embodiments, the multi-view model is trained using complete samples and the view Figure 1 The consistency features may include: extracting view-specific features of each view in the complete sample and calculating the loss value based on the reconstruction loss; performing consistency optimization on the view-specific features extracted by the multi-view model to obtain the view Figure 1 The loss value is calculated based on feature contrast learning; the multi-view model is trained by minimizing the loss value until the multi-view model converges.

[0049] In some embodiments, for a multi-view model, a multi-view extraction part may be used to extract view-specific features. Accordingly, for a single-view model, a single-view extraction part may be used to extract view-specific features.

[0050] In some embodiments, for a single view model, the reconstruction loss of the Vth view may be expressed as follows:

[0051] ,

[0052] in, Can indicate the view-specific encoders, can be its corresponding decoder, and are learnable parameters.

[0053] In some embodiments, view-specific features may be represented as:

[0054] .

[0055] In some embodiments, the reconstruction loss of the multi-view model may be a superposition of the reconstruction losses of multiple views processed by the multi-view model.

[0056] In some embodiments, the reconstruction loss of the multi-view model can be expressed by the following formula:

[0057] .

[0058] In some embodiments, the consistency optimization portion of the multi-view model may be utilized to perform consistency optimization on view-specific features.

[0059] In some embodiments, consistency optimization may include regularization or ensemble learning.

[0060] In some embodiments, Figure 1 The consistency characteristics can be expressed by the following formula:

[0061] ,

[0062] in, , which can represent the extracted view-specific features, , which can represent the extracted visual Figure 1 Consistent characteristics, is a trainable parameter in the multilayer perceptron network.

[0063] In some embodiments, the calculation formula for feature contrast learning may be:

[0064]

[0065] in, The similarity of each feature pair can be measured. can be the temperature coefficient, Can be positive feature pairs, and the rest Can be Negative feature pairs.

[0066] In some embodiments, the view mined using the multi-view model Figure 1 The single-view model can be trained based on consistent features, which may include: optimizing the view-specific features extracted by the single-view model, and Figure 1 The consistent features are used for model comparison learning to calculate the loss value; the single-view model is trained by minimizing the loss value until the single-view model converges.

[0067] In some embodiments, the training process of the multi-view model can be parallel to the training process of the single-view model.

[0068] In some embodiments, the feature optimization portion of the single view model may be utilized to perform feature optimization on view-specific features.

[0069] In some embodiments, feature optimization may include feature selection or feature transformation.

[0070] In some embodiments, model contrast learning may refer to the comparison between the multi-view model output and the single-view model output. Specifically, the multi-view model output may refer to the output of the multi-view feature extraction part and the output of the consistency optimization part, that is, the view-specific features extracted by the multi-view model and the view-specific features extracted by the single-view model. Figure 1 Consistent features; the single view model output may refer to the output of the single view feature extraction part and the output of the feature optimization part, that is, the view-specific features extracted by the single view model and the view-specific features optimized by the features.

[0071] In some embodiments, the formula for model contrast learning may be:

[0072] ,

[0073] in, Can be viewed Figure 1 Consistent characteristics, View-specific features that can be extracted for multi-view models, view-specific features that can be extracted for a single-view model, Features can be specified for feature-optimized views.

[0074] Due to the uncertainty of view quality, the view-specific features mined by the single-view model itself may contain view noise, resulting in poor clustering effect. Figure 1 Therefore, it can focus on the common information reflected in the view, which is more conducive to clustering.

[0075] Therefore, in the feature optimization part of the single-view model, this application uses the training results of the multi-view model, that is, the view mined by the multi-view model. Figure 1 Consistent features help single-view models discard view noise and tend to mine and other view Figure 1 The single-view model can generate consistent universal features, which can better represent the characteristics of data samples and are more conducive to clustering. The feature optimization part of the single-view model is based on the consistency optimization part of the multi-view model, that is, it can discover the clustering structure of incomplete samples based on complete samples. In this application, through the organic combination of the multi-view model and the single-view model, data with missing views, that is, incomplete samples, can be clearly clustered.

[0076] S3. Use the trained single-view model to obtain the clustering results of the incomplete samples, and use the trained multi-view model to obtain the clustering results of the complete samples.

[0077] In some embodiments, obtaining clustering results of incomplete samples using the trained single-view model and obtaining clustering results of complete samples using the trained multi-view model may include: using a K-means clustering algorithm to cluster the views mined from the complete samples by the trained multi-view model; Figure 1 The consistent features are processed to obtain the clustering results of the complete samples; the K-means clustering algorithm is used to process the feature-optimized view-specific features obtained from the incomplete samples by the trained single-view model to obtain the clustering results of the incomplete samples.

[0078] The K-means clustering algorithm is an unsupervised learning algorithm that can be used to divide complete samples and incomplete samples into multiple different clusters. Specifically, the K-means clustering algorithm can assign data points to multiple different clusters so that each data point belongs to the cluster center closest to it.

[0079] In some embodiments, the incomplete multi-view clustering method may further include: if the incomplete sample includes at least two views, clustering is performed using the single view model corresponding to each view separately, and the clustering result with the highest confidence is obtained as the clustering result of the incomplete sample. Specifically, each clustering result may be summarized, and then the clustering result with the highest confidence may be selected.

[0080] In this application, the multi-view model is trained using complete samples and the visual Figure 1 Consistent features, while using the multi-view model to mine the Figure 1 The single-view model is trained based on the consistent features, and finally the clustering results of complete samples and incomplete samples are obtained using the multi-view model and the single-view model respectively. In this case, multiple models can be constructed to effectively utilize incomplete information while avoiding the instability of interpolation quality that may bring noise to the original data. In addition, in view of the differences between different views, a single-view model that mines view-specific features and a single-view model that mines view-specific features can be constructed. Figure 1 The multi-view model with consistent features can extract effective features for clustering, thereby improving the applicability and generalization ability of multi-view clustering.

[0081] Those skilled in the art can understand that all or part of the features / steps of the above-mentioned method embodiments can be implemented by methods, data processing systems or computer programs, and these features can be implemented without hardware, entirely by software, or by a combination of hardware and software. The aforementioned computer program can be stored in one or more computer-readable storage media, and the storage medium stores the computer program. When the computer program is executed (such as by a processor), the steps of the above-mentioned incomplete multi-view clustering method embodiment are executed.

[0082] The aforementioned storage media that can store program codes include: static hard disks, solid-state hard disks, random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), optical storage devices, magnetic storage devices, flash memory, magnetic disks or optical disks and / or combinations of the above devices, that is, they can be implemented by any type of volatile or non-volatile storage device or a combination thereof.

[0083] like Figure 2 As shown, the present application also provides an embodiment of a processing device 1, comprising one or more processors 11 and a memory 10; wherein the memory 10 is used to store one or more computer programs, and the one or more processors 11 are used to execute one or more computer programs stored in the memory 10, so that the processor 11 performs the features / steps of the above-mentioned incomplete multi-view clustering method embodiment.

[0084] The above is only the preferred embodiment of the present application. Those skilled in the art will appreciate that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present application. In addition, under the guidance of the present application, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present application. Therefore, the present application is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the present application belong to the protection scope of the present application.

Claims

1. An incomplete multi-view clustering method, characterized in that: include: Constructing an incomplete multi-view dataset, a multi-view model for mining view consistency features, and a single-view model for mining view specific features, wherein the incomplete multi-view dataset includes a plurality of complete samples and a plurality of incomplete samples, and the complete sample is a sample containing all view information; Using the complete samples to train the multi-view model and mine the view consistency features, and using the view consistency features mined by the multi-view model to train the single-view model; Using the trained single-view model to obtain clustering results of the incomplete samples, and using the trained multi-view model to obtain clustering results of the complete samples; The method of training the multi-view model and mining the view consistency features using the complete sample includes: extracting view-specific features of each view in the complete sample, and calculating a loss value based on a reconstruction loss; performing consistency optimization on the view-specific features extracted from the multi-view model to obtain the view consistency features, and calculating a loss value based on feature contrast learning; training the multi-view model by minimizing the loss value until the multi-view model converges; the method of training the single-view model using the view consistency features mined from the multi-view model includes: performing feature optimization on the view-specific features extracted from the single-view model, and calculating a loss value by performing model contrast learning based on the view consistency features; and training the single-view model by minimizing the loss value until the single-view model converges.

2. The incomplete multi-view clustering method according to claim 1, characterized in that: The method of using the trained single-view model to obtain the clustering results of the incomplete samples and using the trained multi-view model to obtain the clustering results of the complete samples includes: using the K-means clustering algorithm to process the view consistency features mined from the complete samples by the trained multi-view model to obtain the clustering results of the complete samples; and using the K-means clustering algorithm to process the feature-optimized view specific features obtained from the incomplete samples by the trained single-view model to obtain the clustering results of the incomplete samples.

3. The incomplete multi-view clustering method according to claim 1, characterized in that: The method further includes: if the incomplete sample includes at least two views, clustering is performed using a single view model corresponding to each view, and a clustering result with the highest confidence is obtained as the clustering result of the incomplete sample.

4. The incomplete multi-view clustering method according to claim 1, characterized in that: The single view model comprises a single view feature extraction part and a feature optimization part, wherein the single view feature extraction part extracts features from the view of the complete sample and / or the view of the incomplete sample, and the feature optimization part optimizes features extracted by the single view feature extraction part; The multi-view model includes a multi-view feature extraction part and a consistency optimization part, wherein the multi-view feature extraction part extracts features from the views of the complete sample, and the consistency optimization part performs consistency optimization on the features extracted by the multi-view feature extraction part.

5. The incomplete multi-view clustering method according to claim 1, characterized in that: The constructing of the incomplete multi-view dataset includes: constructing a multi-view dataset, the multi-view dataset including a plurality of samples; randomly selecting a number of the samples and randomly discarding at least one view of the samples according to a preset missing rate, to obtain the incomplete multi-view dataset.

6. The incomplete multi-view clustering method according to claim 5, characterized in that: The method further includes: recording available information and missing information of the view in an indicator matrix, wherein the missing information is information of the randomly discarded view, and the available information is information of the remaining views in the sample.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed, the incomplete multi-view clustering method according to any one of claims 1 to 6 is implemented.

8. A processing device, characterized in that: include: one or more processors; A memory is used to store one or more computer programs, and one or more of the processors are used to execute the one or more computer programs stored in the memory, so that the one or more processors perform the incomplete multi-view clustering method according to any one of claims 1 to 6.

Images