Deep variational auto-encoder construction method, device and equipment based on cloud transformation
By introducing cloud transformation technology into the depth variational autoencoder, the problems of low quality of generated samples, insufficient diversity and poor training stability are solved, and higher quality and diversity of data generation and more stable training process are achieved.
Patent Information
- Application Number
- CN202510372416.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-27
- Publication Date
- 2025-06-27
AI Technical Summary
The existing depth variation autoencoders have problems with low quality of generated samples, insufficient diversity and poor stability in the training process.
The deep variational autoencoder construction method based on cloud transformation is adopted, and the original data set is obtained for data cleaning and normalization. The encoder is used to extract data features, combined with cloud model analysis and processing, the target feature vector is generated and the target data set is output through the decoder.
The stability of the variational autoencoder is enhanced, the quality and diversity of generated samples are improved, and the modeling ability of data distribution is optimized.
Smart Images
Figure CN120218130A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of machine learning, and in particular, to a method, device, and equipment for constructing a deep variational autoencoder based on cloud transformation. Background Art
[0002] The application of generative models in daily learning and life has become increasingly popular, such as the widespread use of large language models, AI drawing, and other technologies. However, the current research on generative models based on variational autoencoders still faces many challenges, including low quality of generated samples, insufficient diversity, and poor stability in the training process.
[0003] To address these technical bottlenecks, the directions for improvement include deepening the network structure of the variational autoencoder and using more complex network structures to extract data features and generate higher-quality samples. However, this method reduces the training stability and increases the difficulty; or by improving the regularization penalty term to enable the model to learn a distribution that better fits the real data. However, the standard Gaussian distribution as the prior distribution is still too simple and insufficient for fitting real data. Summary of the Invention
[0004] This application provides a method, device, and equipment for constructing a deep variational autoencoder based on cloud transformation to solve the problems of low quality of generated samples, insufficient diversity, and poor stability in the training process of existing deep variational autoencoders.
[0005] In a first aspect, this application provides a method for constructing a deep variational autoencoder based on cloud transformation. The variational autoencoder includes: an encoder, a cloud model, and a decoder. The method includes:
[0006] Obtain the original data set, perform data cleaning and normalization processing on the original data set to obtain preprocessed data;
[0007] Input the preprocessed data into the encoder for feature extraction to obtain the first data feature, where the first data feature is used to indicate the data distribution of the original data;
[0008] Use the cloud model to analyze and process the first data feature to obtain the second data feature, where the second data feature is used to indicate the data distribution learned by the cloud model based on the first data feature;
[0009] Perform sampling processing on the second data feature, output the target feature vector, and the target feature vector is decoded by the decoder to output the target data set.
[0010] Optionally, the encoder adopts a multi-layer neural network structure. The step of inputting the preprocessed data into the encoder for feature extraction to obtain the first data feature includes:
[0011] Receive the preprocessed original data through the encoder, and perform feature extraction and transformation on the input original data through a multi-layer neural network structure to obtain the first data feature corresponding to the original data;
[0012] Output the first data feature through the encoder, and the first data feature includes: the expected value Ex, entropy En, and hyperentropy He of the original data set.
[0013] Optionally, analyzing and processing the first data feature using the cloud model to obtain a second data feature, including:
[0014] Input the first data feature and the number of samples to be generated into the cloud model, and generate a sample set according to the first data feature and the number of samples to be generated through the forward cloud algorithm of the cloud model;
[0015] Determine the mean of the sample set, group the generated sample set, determine the mean and variance of each group of samples, and determine the entropy and hyperentropy of the sample set according to the mean and variance of each group of samples;
[0016] Use the mean, entropy, and hyperentropy of the sample set as the second data feature.
[0017] Optionally, sampling the second data feature to output a target feature vector, satisfying the following formula:
[0018] z = Ex' + En'×ε2 + He'×ε1×ε2
[0019] where z is the target feature vector, Ex' is the expected value of the sample set, En' is the entropy of the sample set, He' is the hyperentropy of the sample set, and ε1, ε2 are random variables generated from the standard normal distribution.
[0020] Optionally, the method further includes:
[0021] Perform similarity measurement on the second data feature and the data feature of the preset real data to obtain the difference degree between the second data feature and the data feature of the preset real data;
[0022] Judge whether the difference degree between the second data feature and the data feature of the preset real data is less than a preset difference degree threshold;
[0023] If not, adjust the weight parameters of the variational autoencoder until the difference degree between the second data feature and the data feature of the preset real data is less than the preset difference degree threshold.
[0024] Optionally, the similarity measurement of the second data feature and the data feature of the preset true data to obtain the difference degree between the second data feature and the data feature of the preset true data includes:
[0025] Based on the second data feature and the data feature of the preset true data, determine the feature curves of the second data feature and the data feature of the preset true data, where the feature curves include: an expected curve, an outer envelope curve, and an inner envelope curve;
[0026] Respectively determine the difference degrees of the expected curve, the outer envelope curve, and the inner envelope curve of the second data feature and the data feature of the preset true data, and use the sum of the difference degrees of the expected curve, the outer envelope curve, and the inner envelope curve of the second data feature and the data feature of the preset true data as the difference degree between the second data feature and the data feature of the preset true data.
[0027] In a second aspect, the present application provides a device for constructing a deep variational autoencoder based on cloud transformation. The variational autoencoder includes: an encoder, a cloud model, and a decoder. The device includes:
[0028] An acquisition module, configured to acquire an original data set, perform data cleaning and normalization processing on the original data set to obtain preprocessed data;
[0029] A processing module, configured to input the preprocessed data into the encoder for feature extraction to obtain a first data feature, where the first data feature is used to indicate the data distribution of the original data;
[0030] The processing module is further configured to use the cloud model to analyze and process the first data feature to obtain a second data feature, where the second data feature is used to indicate the data distribution learned by the cloud model based on the first data feature;
[0031] The processing module is further configured to perform sampling processing on the second data feature and output a target feature vector, and the target feature vector is decoded by the decoder to output a target data set.
[0032] In a third aspect, the present application provides a device for constructing a deep variational autoencoder based on cloud transformation, including:
[0033] A memory;
[0034] A processor;
[0035] Wherein, the memory stores computer execution instructions;
[0036] The processor executes the computer-executable instructions stored in the memory to implement the method for constructing a deep variational autoencoder based on cloud transformation as described in the first aspect and various possible implementation manners of the first aspect above.
[0037] In a fourth aspect, the present application provides a computer storage medium, on which a computer program is stored. The computer program is executed by a processor to implement the method for constructing a deep variational autoencoder based on cloud transformation as described in the first aspect and various possible implementation manners of the first aspect above.
[0038] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for constructing a deep variational autoencoder based on cloud transformation as described in the first aspect and various possible implementation manners of the first aspect above.
[0039] The present application provides a method, apparatus, and device for constructing a deep variational autoencoder based on cloud transformation. The method includes obtaining an original data set, performing data cleaning and normalization processing on the original data set to obtain preprocessed data; inputting the preprocessed data into an encoder for feature extraction to obtain first data features, where the first data features are used to indicate the data distribution of the original data; using a cloud model to analyze and process the first data features to obtain second data features, where the second data features are used to indicate the data distribution learned by the cloud model based on the first data features; performing sampling processing on the second data features to output target feature vectors, and outputting a target data set through a decoder according to the target feature vectors, which enhances the stability of the variational autoencoder and improves the quality and diversity of the generated samples. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.
[0041] Figure 1 It is a schematic flowchart of the method for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application Figure 1 ;
[0042] Figure 2 It is a schematic flowchart of the method for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application Figure 2 ;
[0043] Figure 3 It is a schematic structural diagram of the apparatus for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application;
[0044] Figure 4 It is a schematic structural diagram of the device for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application.
[0045] Through the above-mentioned accompanying drawings, specific embodiments of the present application have been shown, and will be described in more detail hereinafter. These drawings and the textual description are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by reference to specific embodiments. Detailed implementation manners
[0046] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below with reference to the accompanying drawings in the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.
[0047] Terms such as "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned accompanying drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein.
[0048] In the embodiments of the present application, words such as "exemplary" or "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.
[0049] Among various types of generative models, the generative model based on variational autoencoders faces problems such as low quality of generated samples, insufficient diversity, and poor stability during the training process in the research and application process.
[0050] The prior art extracts features of data by introducing more complex network architectures. However, as the network structure becomes more complex, the number of parameters during the training process increases sharply, which not only requires more computing resources and time, but also leads to a decrease in the stability of training.
[0051] The prior art also constrains the learning process of the model by introducing regularization penalty terms, enabling the model to learn a distribution that better fits the real data. However, currently in the generative model based on variational autoencoders, a standard Gaussian distribution is usually used as the prior distribution, which cannot fully capture the complex features and distribution laws of real data.
[0052] In view of the above problems, the present application proposes a method for constructing a deep variational autoencoder based on cloud transformation. By introducing cloud transformation technology, this method enhances the model's ability to extract multi-level features from complex data, optimizes the stability and convergence efficiency of the training process, and improves the modeling ability of data distribution, thereby generating more realistic and diverse data, providing an efficient and stable technical solution for data generation, dimensionality reduction, and feature learning.
[0053] The following uses specific embodiments to elaborate in detail on the technical solution of the present application and how the technical solution of the present application solves the above technical problems. These several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.
[0054] Figure 1 Flow schematic of a method for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application Figure 1 As Figure 1 shown, the method for constructing a deep variational autoencoder based on cloud transformation provided in this embodiment includes:
[0055] S101: Obtain the original data set, perform data cleaning and normalization processing on the original data set to obtain the preprocessed data.
[0056] Among them, the original data set refers to a data set directly collected from various data sources without any processing.
[0057] It can be understood that data cleaning refers to the process of processing problems such as errors, missing values, duplicate values, and outliers in the original data set to improve data quality. Normalization processing is to scale the feature values in the original data set to a specific range. Normalization can eliminate the influence of dimensions and scales between different features, making different features have the same importance in model training and improving the stability and convergence speed of the model.
[0058] The data set obtained after data cleaning and normalization processing is conducive to tasks such as data analysis and machine learning model training.
[0059] S102: Input the preprocessed data into the encoder for feature extraction to obtain the first data feature.
[0060] Among them, the first data feature is used to indicate the data distribution of the original data.
[0061] It is understandable that through feature extraction by the encoder, high-dimensional data can be mapped to a low-dimensional space, reducing the data dimension, computational complexity, and storage requirements, while enhancing the generalization ability of the model. The encoder can automatically extract the most representative and discriminative features from the data, removing noise and redundancy, enabling the model to focus more on the essential features of the data.
[0062] Specifically, the preprocessed data is input into the encoder of the deep variational autoencoder, and the encoder processes the input data and outputs the first data feature.
[0063] S103: Analyze and process the first data feature using the cloud model to obtain the second data feature.
[0064] Among them, the second data feature is used to indicate the data distribution learned by the cloud model based on the first data feature.
[0065] It is understandable that the cloud model uses three numerical features, namely the expected value Ex, entropy En, and hyperentropy He, to comprehensively represent a qualitative concept. The expected value is the expectation of the distribution of cloud droplets in the universe of discourse space, representing the point that best represents the qualitative concept; the entropy represents the measurable granularity of the qualitative concept, reflecting the degree of fuzziness of the concept; the hyperentropy is the uncertainty measure of the entropy, that is, the entropy of the entropy, reflecting the degree of dispersion of the entropy.
[0066] Specifically, use the forward cloud generator to generate cloud droplets according to the determined cloud model parameters and the first data feature. Further, analyze and process the generated cloud droplets to extract the second data feature that can reflect the data distribution learned by the cloud model.
[0067] S104: Sample the second data feature to output the target feature vector, and the target feature vector is decoded by the decoder to output the target data set.
[0068] It is understandable that the decoder is a model component that can convert the feature vector into the representation of the original data space. In deep learning, the role of the decoder is to reconstruct a data set with the same format as the original data according to the input feature vector. The target data set is the data set output by the decoder according to the target feature vector, and its format and properties are similar to the original data set, which is a reconstruction or generation of the original data distribution.
[0069] Specifically, sample the second data feature to output the target feature vector, satisfying the following formula:
[0070] z = Ex' + En'×ε2 + He'×ε1×ε2
[0071] Among them, z is the target feature vector, Ex' is the expected value of the sample set, En' is the entropy of the sample set, He' is the hyperentropy of the sample set, and ε1, ε2 are random variables generated from the standard normal distribution.
[0072] Further, input the target feature vector obtained through sampling processing into the trained decoder. The decoder calculates and transforms the input target feature vector according to its own structure, and outputs the target data set.
[0073] A method for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application. This method obtains the original data set, performs data cleaning and normalization processing on the original data set to obtain preprocessed data; inputs the preprocessed data into the encoder for feature extraction to obtain the first data feature; uses the cloud model to analyze and process the first data feature to obtain the second data feature; performs sampling processing on the second data feature and outputs the target feature vector. The target feature vector is decoded by the decoder to output the target data set, which enhances the stability of the variational autoencoder and improves the quality and diversity of the generated samples.
[0074] Figure 2 It is a flow chart of a method for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application. Figure 2 This embodiment is based on Figure 1 the embodiment, and a possible implementation manner of the method for constructing a deep variational autoencoder based on cloud transformation is described in detail. As Figure 2 shown, the method includes:
[0075] S201: Obtain the original data set, perform data cleaning and normalization processing on the original data set to obtain preprocessed data.
[0076] Among them, step S201 is similar to the above step S101 and will not be elaborated here.
[0077] S202: Receive the preprocessed original data through the encoder, and perform feature extraction and transformation on the input original data through a multi-layer neural network structure to obtain the first data feature corresponding to the original data.
[0078] It can be understood that, according to the type of data and the requirements of the task, a suitable neural network architecture is selected as the encoder. The selection of the number of network layers and neurons needs to be adjusted according to the complexity of the data and the difficulty of the task.
[0079] Specifically, input the preprocessed original data into the encoder, and the data will pass through a multi-layer neural network in sequence. In each layer, the neurons will perform weighted summation on the input and perform a non-linear transformation through an activation function to obtain the output of that layer. Further, after multiple feature extractions and transformations, the first data feature corresponding to the original data is obtained.
[0080] S203: Output the first data feature through the encoder.
[0081] Among them, the first data feature includes the expectation Ex, entropy En, and hyperentropy He of the original data set.
[0082] S204: Input the first data feature and the number of samples to be generated into the cloud model, and generate a sample set according to the first data feature and the number of samples to be generated through the forward cloud algorithm of the cloud model.
[0083] It can be understood that the forward cloud algorithm can generate quantitative cloud droplets as samples from a qualitative concept according to the three numerical features of the cloud model: expectation, entropy, and hyperentropy. A cloud droplet is a specific quantitative representation of a qualitative concept in the cloud model. Each cloud droplet represents a specific value of the qualitative concept in the universe of discourse. A large number of cloud droplets aggregated together form a cloud, which can intuitively reflect the uncertainty distribution of the qualitative concept.
[0084] Specifically, the specific steps of the forward cloud algorithm include:
[0085] Randomly generate a normal random number with En as the expected value and He as the standard deviation, denoted as En';
[0086] According to the absolute value of En', generate a normal random number with Ex as the expected value and |En'| as the standard deviation as a sample, denoted as x;
[0087] Calculate the certainty degree of the sample x for the qualitative concept; among them, the difference between x and Ex determines the size of the certainty degree, and the certainty degree reflects the degree to which the sample x belongs to this qualitative concept;
[0088] Repeat the above steps until the required number of samples is generated.
[0089] Generating cloud droplets according to the first data feature through the forward cloud algorithm can achieve data expansion and simulation. In some cases where the data volume is small or the data distribution is uneven, the generated cloud droplets can be used as new data samples, increasing the diversity and richness of the data, thereby improving the generalization ability and stability of the model.
[0090] S205: Determine the mean of the sample set, group the generated sample set, determine the mean and variance of each group of samples, and determine the entropy and hyperentropy of the sample set according to the mean and variance of each group of samples.
[0091] Specifically, determining the mean of the sample set, grouping the generated sample set, determining the mean and variance of each group of samples, and determining the entropy and hyperentropy of the sample set according to the mean and variance of each group of samples includes:
[0092] Step1: Calculate the mean, where x is the sample and N is the number of samples;
[0093] Step 2: Randomly divide the samples into m groups, with n samples in each group, ensuring that the samples in each group are non - overlapping;
[0094] Step 3: For the i - th group X i ={x i1 , x i2 , …, x in}, calculate its mean,
[0095] and variance,
[0096] Finally, obtain the variance of each group
[0097] Step 4: Calculate the entropy
[0098] and the hyper - entropy, He' 2 = EY 2 - En' 2 ,
[0099] where is the mean of Y 2 , is the variance of Y 2 .
[0100] S206: Use the mean, entropy, and hyper - entropy of the sample set as the second data feature.
[0101] S207: Sample the second data feature to output the target feature vector, and use the decoder to output the target data set according to the target feature vector.
[0102] Among them, step S207 is similar to the above - mentioned step S104 and will not be elaborated here.
[0103] In an alternative embodiment, a method for constructing a deep variational auto - encoder based on cloud transformation further includes:
[0104] S301: Measure the similarity between the second data feature and the data feature of the preset real data to obtain the difference degree between the second data feature and the data feature of the preset real data.
[0105] Specifically, measuring the similarity between the second data feature and the data feature of the preset real data to obtain the difference degree between the second data feature and the data feature of the preset real data includes:
[0106] Based on the second data feature and the data feature of the preset real data, determine the feature curves of the second data feature and the data feature of the preset real data. The feature curves include: the expected curve, the outer envelope curve, and the inner envelope curve;
[0107] Determine the difference degrees of the expected curve, outer envelope curve, and inner envelope curve between the second data feature and the data feature of the preset true data respectively, and use the sum of the difference degrees of the expected curve, outer envelope curve, and inner envelope curve between the second data feature and the data feature of the preset true data as the difference degree between the second data feature and the data feature of the preset true data.
[0108] It can be understood that in this embodiment, the difference degrees of the expected curve, outer envelope curve, and inner envelope curve between the second data feature and the data feature of the preset true data are determined by the KL divergence formula. Specifically, for two probability distributions P and Q of continuous random variables, their KL divergence is defined as:
[0109]
[0110] For Gaussian cloud distributions C i (Ex i , En i , He i ) and C j (Ex j , En j , He j ):
[0111] The KL divergence of the expected curve is:
[0112]
[0113] Among them, σ 1i = En i , σ 1j = En j .
[0114] The KL divergence of the outer envelope curve:
[0115]
[0116] Among them, σ 2i = En i + 3He i , σ 1j = En j + 3He j .
[0117] The KL divergence of the inner envelope curve:
[0118]
[0119] Among them, σ 3i = En i - 3He i , σ 3j= En j -3He j 。
[0120] Therefore, the similarity measure between the two cloud models is defined as:
[0121] D KL (P i ||Q j ) =
[0122] D KL (P 1i ||Q 1j ) + D KL (P 2i ||Q 2j ) + D KL (P 3i ||Q 3j ), where D KL is greater than or equal to zero, and the closer it is to zero, the smaller the difference degree is considered.
[0123] S302: Determine whether the difference degree between the second data feature and the data feature of the preset true data is less than the preset difference degree threshold.
[0124] S303: When the difference degree between the second data feature and the data feature of the preset true data is not less than the preset difference degree threshold, adjust the weight parameters of the variational autoencoder until the difference degree between the second data feature and the data feature of the preset true data is less than the preset difference degree threshold.
[0125] A method for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application receives, through an encoder, preprocessed original data, extracts and transforms the input original data through a multi-layer neural network structure to obtain a first data feature corresponding to the original data; outputs the first data feature through the encoder; inputs the first data feature into the cloud model, and generates a certain number of cloud droplets according to the first data feature through the forward cloud algorithm of the cloud model; outputs a second data feature according to the generated cloud droplets through the grouped inverse cloud algorithm of the cloud model, improving the stability of the variational autoencoder training process and the quality of the generated samples.
[0126] Figure 3 It is a schematic structural diagram of a device for constructing a deep variational autoencoder based on cloud transformation provided by an embodiment of the present application. As Figure 3 shown, the device 300 for constructing a deep variational autoencoder based on cloud transformation provided in this embodiment, the variational autoencoder includes: an encoder, a cloud model, and a decoder, and the device includes:
[0127] An acquisition module 301, configured to acquire an original data set, perform data cleaning and normalization processing on the original data set to obtain preprocessed data;
[0128] The processing module 302 is configured to input the preprocessed data into the encoder for feature extraction to obtain first data features, where the first data features are used to indicate the data distribution of the original data;
[0129] The processing module 302 is further configured to analyze and process the first data features by using the cloud model to obtain second data features, where the second data features are used to indicate the data distribution learned by the cloud model according to the first data features;
[0130] The processing module 302 further performs sampling processing on the second data features and outputs a target feature vector, and the target feature vector is decoded by the decoder to output a target data set.
[0131] A device for constructing a deep variational autoencoder based on cloud transformation provided in this embodiment can execute the method provided in the above method embodiment, and its implementation principle and technical effects are similar, which will not be elaborated here in this embodiment.
[0132] Figure 4 It is a schematic structural diagram of a device for constructing a deep variational autoencoder based on cloud transformation provided in an embodiment of the present application. As Figure 4 shown, the device for constructing a deep variational autoencoder based on cloud transformation provided in the present application, the device 400 for constructing a deep variational autoencoder based on cloud transformation includes: a receiver 401, a transmitter 402, a processor 403, and a memory 404.
[0133] The receiver 401 is configured to receive instructions and data;
[0134] The transmitter 402 is configured to send instructions and data;
[0135] The memory 404 is configured to store computer-executable instructions;
[0136] The processor 403 is configured to execute the computer-executable instructions stored in the memory 404 to implement each step performed by the method for constructing a deep variational autoencoder based on cloud transformation in the above embodiment. For details, reference can be made to the relevant descriptions in the foregoing embodiment of the method for constructing a deep variational autoencoder based on cloud transformation.
[0137] Optionally, the above memory 404 can be either independent or integrated with the processor 403.
[0138] When the memory 404 is independently provided, the electronic device further includes a bus for connecting the memory 404 and the processor 403.
[0139] The present application also provides a computer storage medium storing computer-executable instructions, which, when executed by a processor, implement the method for constructing a deep variational autoencoder based on cloud transformation as performed by the above-described device for constructing a deep variational autoencoder based on cloud transformation.
[0140] The present application also provides a computer program product including a computer program, which, when executed by a processor, implements the above-described method for constructing a deep variational autoencoder based on cloud transformation.
[0141] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, and appropriate combinations thereof. In the hardware implementation, the division of the functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be executed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically contains computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and may include any information delivery medium.
[0142] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present application. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include the common general knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and examples are only to be considered as exemplary, and the true scope and spirit of the present application are pointed out by the following claims.
[0143] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.
Claims
1. A method for constructing a deep variational autoencoder based on cloud transformation, characterized in that: The variational autoencoder comprises: an encoder, a cloud model and a decoder, and the method comprises: Acquire an original data set, and perform data cleaning and normalization processing on the original data set to obtain preprocessed data; Inputting the preprocessed data into the encoder for feature extraction to obtain a first data feature, where the first data feature is used to indicate data distribution of the original data; Analyzing and processing the first data feature using the cloud model to obtain a second data feature, where the second data feature is used to indicate a data distribution learned by the cloud model based on the first data feature; The second data feature is sampled and processed to output a target feature vector, and the target feature vector is decoded by the decoder to output a target data set.
2. The method according to claim 1, characterized in that The encoder adopts a multi-layer neural network structure, and the pre-processed data is input into the encoder for feature extraction to obtain a first data feature, including: The encoder receives the preprocessed raw data, and performs feature extraction and transformation on the input raw data through a multi-layer neural network structure to obtain a first data feature corresponding to the raw data; The first data feature is outputted by the encoder, and the first data feature includes: expectation Ex, entropy En and super entropy He of the original data set.
3. The method according to claim 1, characterized in that The using the cloud model to analyze and process the first data feature to obtain the second data feature includes: Inputting the first data feature and the number of samples to be generated into the cloud model, and generating a sample set according to the first data feature and the number of samples to be generated by a forward cloud algorithm of the cloud model; Determine the mean of the sample set, and group the generated sample set to determine the mean and variance of each group of samples, and determine the entropy and super entropy of the sample set according to the mean and variance of each group of samples; The mean, entropy and super entropy of the sample set are used as the second data feature.
4. The method according to claim 3, characterized in that The second data feature is sampled and processed to output a target feature vector, which satisfies the following formula: z=Ex′+En′×ε2+He′×ε1×ε2 Among them, z is the target feature vector, Ex' is the expectation of the sample set, En' is the entropy of the sample set, He' is the hyperentropy of the sample set, and ε1 and ε2 are random variables generated from the standard normal distribution.
5. The method according to claim 1, characterized in that The method further comprises: Performing similarity measurement on the second data feature and the data feature of the preset real data to obtain the difference between the second data feature and the data feature of the preset real data; Determine whether the difference between the second data feature and the data feature of the preset real data is less than a preset difference threshold; If not, the weight parameters of the variational autoencoder are adjusted until the difference between the second data feature and the data feature of the preset real data is less than a preset difference threshold.
6. The method according to claim 5, characterized in that The performing similarity measurement on the second data feature and the data feature of the preset real data to obtain the difference between the second data feature and the data feature of the preset real data includes: Based on the second data feature and the data feature of the preset real data, determining a characteristic curve of the second data feature and the data feature of the preset real data, the characteristic curve comprising: an expected curve, an outer envelope curve and an inner envelope curve; Determine the differences between the expected curve, outer envelope curve and inner envelope curve of the second data feature and the data feature of the preset real data respectively, and take the sum of the differences between the expected curve, outer envelope curve and inner envelope curve of the second data feature and the data feature of the preset real data as the difference between the second data feature and the data feature of the preset real data.
7. A device for constructing a deep variational autoencoder based on cloud transformation, characterized in that: The variational autoencoder comprises: an encoder, a cloud model and a decoder, and the device comprises: An acquisition module is used to acquire an original data set, perform data cleaning and normalization processing on the original data set, and obtain pre-processed data; A processing module, used for inputting the preprocessed data into the encoder to perform feature extraction to obtain a first data feature, where the first data feature is used to indicate data distribution of the original data; The processing module is further used to analyze and process the first data feature using the cloud model to obtain a second data feature, where the second data feature is used to indicate the data distribution learned by the cloud model based on the first data feature; The processing module is further used to perform sampling processing on the second data feature and output a target feature vector, and the target feature vector is decoded by the decoder to output a target data set.
8. A device for constructing a deep variational autoencoder based on cloud transformation, characterized in that: The device comprises: Memory; processor; Wherein, the memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement a method for constructing a deep variational autoencoder based on cloud transformation as described in any one of claims 1-6.
9. A computer storage medium, characterized in that The computer storage medium stores computer execution instructions, which, when executed by a processor, are used to implement a method for constructing a deep variational autoencoder based on cloud transformation as described in any one of claims 1 to 6.
10. A computer program product, characterized in that It includes a computer program, which, when executed by a processor, is used to implement a method for constructing a deep variational autoencoder based on cloud transformation as described in any one of claims 1 to 6.