CN111243045A - Image generation method based on Gaussian mixture model prior variation self-encoder - Google Patents
Image generation method based on Gaussian mixture model prior variation self-encoder Download PDFInfo
- Publication number
- CN111243045A CN111243045A CN202010024870.8A CN202010024870A CN111243045A CN 111243045 A CN111243045 A CN 111243045A CN 202010024870 A CN202010024870 A CN 202010024870A CN 111243045 A CN111243045 A CN 111243045A
- Authority
- CN
- China
- Prior art keywords
- encoder
- gaussian mixture
- variational self
- mixture model
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses an image generation method based on a Gaussian mixture model prior variational self-encoder, which comprises the following steps: s11, presetting a generated image training data set; wherein the training data set consists of a plurality of batches of training data; s12, building a prior variational self-encoder network of a Gaussian mixture model; s13, uploading a plurality of preset batches of training data to a variational self-encoder network, and determining posterior distribution and prior distribution of the variational self-encoder network; s14, determining the relation between Gaussian components in the Gaussian mixture model to obtain a mapping function; s15, obtaining a reconstruction loss function and a KL divergence function by using the variational self-encoder network and the obtained mapping function, calculating the loss functions of posterior distribution and prior distribution of the variational self-encoder network, and updating parameters of the variational self-encoder network to generate an image; and S16, when the image is generated, uploading the pseudo input serving as an input image to a variational self-coder network to obtain a finally generated image.
Description
Technical Field
The invention relates to the technical field of deep learning, in particular to an image generation method based on a Gaussian mixture model prior variational self-encoder.
Background
In the internet era, machine learning develops rapidly, and great achievements are achieved, wherein an image generation technology is taken as a branch of machine learning and plays an important role in understanding images. The image generation model is a probability model used for carrying out probability modeling on the image, and the deep neural network can be regarded as a very complex nonlinear function with very strong fitting capability and can be used for building the generation model to estimate parameters of a probability density function. The image generation model can be used for generating more different picture samples, recovering image information, converting pictures of different modalities or between the pictures and characters, voice and the like, and predicting the future, for example, predicting the future frame according to the past frame and the current frame in the video.
The variational self-encoder is a well-known image generation model based on deep learning, is a natural development of variational inference, combines the advantages of ELBO and neural network, solves the inference problem in a general scene, and simultaneously solves the generation problem of continuous data. It has many advantages including fast training, stability, etc., and thus has wide application in both theoretical models and industry. However, standard variational auto-coders generate blurred pictures a priori due to the under-fitting problem.
Disclosure of Invention
The invention aims to provide an image generation method based on a Gaussian mixture model prior variational self-encoder aiming at the defects of the prior art, which can be used for modeling a complex image and generating a high-quality picture, thereby greatly improving the generation capability of the model.
In order to achieve the purpose, the invention adopts the following technical scheme:
an image generation method based on a Gaussian mixture model prior variation self-encoder comprises the following steps:
s1, presetting a generated image training data set; wherein the training data set consists of several batches of training data;
s2, building a variational self-encoder network based on Gaussian mixture model prior;
s3, uploading the preset training data of a plurality of batches to a constructed variational self-encoder network, and determining posterior distribution and prior distribution of the variational self-encoder network;
s4, determining the relation between Gaussian components in the Gaussian mixture model to obtain a mapping function;
s5, obtaining a reconstruction loss function and a KL divergence function by using the variational self-encoder network and the obtained mapping function, calculating the loss functions of posterior distribution and prior distribution of the variational self-encoder network according to the obtained reconstruction loss function and the KL divergence function, and updating the parameters of the variational self-encoder network to generate an image;
and S6, when the image is generated, uploading a pseudo input serving as an input image to the variational self-encoder network to obtain a finally generated image.
Further, the step S2 further includes constructing a posterior distribution of hidden variables in the variational autoencoder network.
Further, the parameters in the variational self-encoder network constructed in step S2 include a network input image size C × H × W, a batch size B, a hidden variable dimension D, a hidden variable z, a gaussian mixture number M, a pseudo input α, and a pseudo input number K.
Further, in step S3, the preset batches of training data are uploaded to a constructed variational self-encoder network, where the uploaded training data includes image sample X ═ { X ═ X1,x2,…,xNIn which xiFor the ith sample in the current batch, i is 1, 2, … B, and the pseudo input α is α1,α2,...,αKTherein αjRepresenting the jth dummy input, j ═ 1, 2, … K.
Further, the step S3 determines the posterior distribution of the hidden variables and the prior distribution of the hidden variables in the form of the aggregate posterior;
the posterior distribution of the hidden variables is as follows:
the hidden variable prior is:
wherein the content of the first and second substances,m denotes the number of Gaussian mixtures, K denotes the number of pseudo inputs, αkA pseudo-input is represented by a number of input,πmrepresenting the coefficients of a gaussian mixture model.
Further, the relationship between the gaussian components in the gaussian mixture model determined in step S4 is determined by a greedy algorithm.
Further, the step S4 is specifically to sequentially construct a mapping function according to the following functions:
where a ═ β (t) | t ═ 1., m-1}, β (·) denotes a mapping function.
Further, the reconstruction loss function obtained in step S5 is:
wherein n represents the dimension of the input picture; x is the number ofiA value representing the ith dimension of the input sample picture;a value representing the ith dimension of the output picture; l isRERepresenting the reconstruction loss for each sample.
Further, the KL divergence function obtained in step S5 is:
wherein L isKLThe KL distance for each sample is indicated.
Further, the calculated loss function is:
wherein the content of the first and second substances,representing the reconstruction error of the ith sample;indicating the KL divergence for the ith sample.
Compared with the prior art, the method has the advantages that the variational self-coder network based on the optimized Gaussian mixture model prior is built, the training efficiency is high, the convergence is strong, the network can be used for modeling a complex image to generate a high-quality picture, and the generation capability of the model is greatly improved.
Drawings
FIG. 1 is a flowchart of an image generation method based on a Gaussian mixture model prior variation auto-encoder according to an embodiment;
fig. 2 is a schematic diagram of a variational auto-encoder network based on an optimized gaussian mixture model prior according to an embodiment.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
The invention aims to provide an image generation method based on a Gaussian mixture model prior variation self-encoder aiming at the defects of the prior art.
Example one
The embodiment provides an image generation method based on a gaussian mixture model prior variation self-encoder, as shown in fig. 1-2, including the steps of:
s11, presetting a generated image training data set; wherein the training data set consists of several batches of training data;
s12, building a variational self-encoder network based on Gaussian mixture model prior;
s13, uploading the preset training data of a plurality of batches to a constructed variational self-encoder network, and determining posterior distribution and prior distribution of the variational self-encoder network;
s14, determining the relation between Gaussian components in the Gaussian mixture model to obtain a mapping function;
s15, obtaining a reconstruction loss function and a KL divergence function by using the variational self-encoder network and the obtained mapping function, calculating the loss functions of posterior distribution and prior distribution of the variational self-encoder network according to the obtained reconstruction loss function and the KL divergence function, and updating parameters of the variational self-encoder network to generate an image;
and S16, when the image is generated, uploading a pseudo input serving as an input image to the variational self-encoder network to obtain a finally generated image.
In step S11, generating an image training data set is preset; wherein the training data set consists of several batches of training data.
Preparing a qualified training data set of generated images, defining the training data with the size of B in each batch, and forming a training batch { x ] by B image samples1,x2,…,xN}。
In step S12, a variational self-coder network based on a gaussian mixture model prior is built.
The parameters in the constructed variational autoencoder network comprise network input image size C multiplied by H multiplied by W, batch size B, hidden variable dimension D, hidden variable z, Gaussian mixture number M, pseudo input α and pseudo input number K, wherein in the embodiment, D is 40, M is 3, and K is 500.
It should be noted that, in the present embodiment, a gaussian mixture model is established based on the variational self-encoder. Unlike the variational autocoder, the posterior distribution of the hidden variables is constructed by a gaussian mixture model rather than a single gaussian model, which is diagonal in order to simplify the computation of the covariance matrix for each component.
In step S13, the preset batches of training data are uploaded to a constructed variational self-encoder network, and a posterior distribution and a prior distribution of the variational self-encoder network are determined.
Uploading a plurality of preset batches of training data to a constructed variational self-encoder network, wherein each batch of training data sent to the variational self-encoder network comprises image samples expressed as X ═ X1,x2,…,xNIn which xiFor the ith sample in the batch, i ═ 1, 2, … B, and learnable pseudo-input { α }1,α2,...,αKTherein αjRepresents the jth dummy input, j ═ 1, 2, … K; the posterior distribution q (z | x) of the hidden variables of the network and the prior distribution p (z) of the hidden variables in aggregated posterior form are determined.
The posterior distribution of hidden variables is:
hidden variable prior is:
wherein the content of the first and second substances,m denotes the number of Gaussian mixtures, K denotes the number of pseudo inputs, αkA pseudo-input is represented by a number of input,πmrepresenting the coefficients of a gaussian mixture model.
In step S14, the relationship between gaussian components in the gaussian mixture model is determined, and a mapping function is obtained.
A greedy algorithm is used to determine the correspondence between the gaussian components in the two gaussian models, resulting in mapping function β (·).
In this embodiment, according to the Gaussian weight pimAll Gaussian components in the Gaussian mixture model are referenced in a large-to-small order such that c1≥c2≥…≥cMWhen i is 1, the ratio of the total of the two,
Wherein, a is a ∪ { β (M) }, if M < M, M is M +1, returning to the step of constructing the mapping function in sequence, otherwise ending, and continuing to execute step S15.
In step S15, a reconstruction loss function and a KL divergence function are obtained by using the variational self-encoder network and the obtained mapping function, a loss function of posterior distribution and prior distribution of the variational self-encoder network is calculated according to the obtained reconstruction loss function and KL divergence function, and parameters of the variational self-encoder network are updated to generate an image.
In the present embodiment, forThe variational self-encoder network respectively reconstructs a loss function and a KL divergence function to the input X and the output of the variational self-encoder networkAnd calculating a loss function by the posterior distribution q (z | x) and the prior distribution p (z), and updating parameters of the variational self-encoder network by a back propagation algorithm until the network converges.
The reconstruction loss function for each sample is calculated as:
wherein n represents the dimension of the input picture; x is the number ofiA value representing the ith dimension of the input sample picture;a value representing the ith dimension of the output picture; l isRERepresenting the reconstruction loss for each sample.
Using mapping function β (), the KL divergence function for each sample is calculated as:
wherein L isKLThe KL distance for each sample is indicated.
Calculating input X and output of the variational autoencoder network through the calculated reconstruction loss function of each sample and the KL divergence function of each sampleAnd the loss function of the posterior distribution q (z | x) and the prior distribution p (z) is:
wherein the content of the first and second substances,representing the reconstruction error of the ith sample;indicating the KL divergence for the ith sample.
In step S16, when an image is generated, a dummy input is uploaded as an input image to the variational self-encoder network, resulting in a finally generated picture.
When an image is generated, a pseudo input is input into the network as an input image, and a high-quality generated picture can be output.
In the present embodiment, the terms are explained as follows:
the gaussian mixture model is a model that accurately quantifies objects by using a gaussian probability density function (normal distribution curve), and is formed by decomposing objects into a plurality of objects based on the gaussian probability density function (normal distribution curve).
Greedy algorithm (also called greedy algorithm) means that when solving a problem, always the choice that seems best at the present time is made. That is, rather than being considered globally optimal, he makes a locally optimal solution in some sense. A greedy algorithm is an improved hierarchical approach. The core of the method is to select a measurement standard according to the theme. The multiple inputs are then arranged in the order required by the metrology standards, in which order the quantities are input one at a time. If the addition of this input to the part of the best solution that is currently already formed in this quantitative sense does not result in a feasible solution, then this input is not added to this part of the solution. This hierarchical approach to obtaining an optimal solution in some metric sense is known as a greedy algorithm.
Compared with the prior art, the method has the advantages that the variational self-encoder network based on the optimization Gaussian mixture model prior is built, the training efficiency is high, the convergence is strong, the network can be used for modeling a complex image to generate a high-quality picture, and the generation capability of the model is greatly improved.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (10)
1. An image generation method based on a Gaussian mixture model prior variation self-encoder is characterized by comprising the following steps:
s1, presetting a generated image training data set; wherein the training data set consists of several batches of training data;
s2, building a variational self-encoder network based on Gaussian mixture model prior;
s3, uploading the preset training data of a plurality of batches to a constructed variational self-encoder network, and determining posterior distribution and prior distribution of the variational self-encoder network;
s4, determining the relation between Gaussian components in the Gaussian mixture model to obtain a mapping function;
s5, obtaining a reconstruction loss function and a KL divergence function by using the variational self-encoder network and the obtained mapping function, calculating the loss functions of posterior distribution and prior distribution of the variational self-encoder network according to the obtained reconstruction loss function and the KL divergence function, and updating the parameters of the variational self-encoder network to generate an image;
and S6, when the image is generated, uploading a pseudo input serving as an input image to the variational self-encoder network to obtain a finally generated image.
2. The method of claim 1, wherein the step S2 further includes constructing a posterior distribution of hidden variables in the variational autoencoder network.
3. The image generation method based on the Gaussian mixture model prior variational self-encoder as claimed in claim 2, wherein the parameters in the variational self-encoder network constructed in the step S2 include network input image size C x H x W, batch size B, hidden variable dimension D, hidden variable z, Gaussian mixture number M, pseudo input α, and pseudo input number K.
4. The image generation method based on the Gaussian mixture model prior variational self-encoder as claimed in claim 3, wherein in step S3, the pre-set batches of training data are uploaded to the constructed variational self-encoder network, wherein the uploaded training data comprise image samples X ═ { X ═ X1,x2,…,xNIn which xiFor the ith sample in the current batch, i is 1, 2, … B, and the pseudo input α is α1,α2,...,αKTherein αjRepresenting the jth dummy input, j ═ 1, 2, … K.
5. The image generation method based on the Gaussian mixture model prior variation auto-encoder as claimed in claim 4, wherein the step S3 determines the posterior distribution of the hidden variables and the prior distribution of the hidden variables in the form of the aggregation posterior;
the posterior distribution of the hidden variables is as follows:
the hidden variable prior is:
6. The image generation method based on the Gaussian mixture model prior variational self-encoder as claimed in claim 5, wherein the relation between Gaussian components in the Gaussian mixture model determined in step S4 is determined by a greedy algorithm.
7. The image generation method based on the gaussian mixture model prior variational self-encoder as claimed in claim 6, wherein said step S4 is specifically to construct the mapping function according to the following functions in turn:
where a ═ β (t) | t ═ 1., m-1}, β (·) denotes a mapping function.
8. The image generation method based on the gaussian mixture model prior variation self-encoder according to claim 7, wherein the reconstruction loss function obtained in step S5 is:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010024870.8A CN111243045B (en) | 2020-01-10 | 2020-01-10 | Image generation method based on Gaussian mixture model prior variation self-encoder |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010024870.8A CN111243045B (en) | 2020-01-10 | 2020-01-10 | Image generation method based on Gaussian mixture model prior variation self-encoder |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111243045A true CN111243045A (en) | 2020-06-05 |
CN111243045B CN111243045B (en) | 2023-04-07 |
Family
ID=70874471
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010024870.8A Active CN111243045B (en) | 2020-01-10 | 2020-01-10 | Image generation method based on Gaussian mixture model prior variation self-encoder |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111243045B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111860660A (en) * | 2020-07-24 | 2020-10-30 | 辽宁工程技术大学 | Small sample learning garbage classification method based on improved Gaussian network |
CN113255830A (en) * | 2021-06-21 | 2021-08-13 | 上海交通大学 | Unsupervised target detection method and system based on variational self-encoder and Gaussian mixture model |
CN113822437A (en) * | 2020-06-18 | 2021-12-21 | 辉达公司 | Deep layered variational automatic encoder |
CN114501034A (en) * | 2021-12-11 | 2022-05-13 | 同济大学 | Image compression method and medium based on discrete Gaussian mixture super-prior and Mask |
CN114638905A (en) * | 2022-01-30 | 2022-06-17 | 中国科学院自动化研究所 | Image generation method, device, equipment, storage medium and computer program product |
CN115131347A (en) * | 2022-08-29 | 2022-09-30 | 江苏茂融智能科技有限公司 | Intelligent control method for processing zinc alloy parts |
CN115797216A (en) * | 2022-12-14 | 2023-03-14 | 齐鲁工业大学 | Inscription character restoration model and restoration method based on self-coding network |
CN116958712A (en) * | 2023-09-20 | 2023-10-27 | 山东建筑大学 | Image generation method, system, medium and device based on prior probability distribution |
CN117036862A (en) * | 2023-08-21 | 2023-11-10 | 武汉纺织大学 | Image generation method based on Gaussian mixture variation self-encoder |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013061732A (en) * | 2011-09-12 | 2013-04-04 | Fuji Xerox Co Ltd | Image identification information provision program and image identification information provision device |
CN106952240A (en) * | 2017-03-29 | 2017-07-14 | 成都信息工程大学 | A kind of image goes motion blur method |
CN108171324A (en) * | 2017-12-26 | 2018-06-15 | 天津科技大学 | A kind of variation own coding mixed model |
CN108491925A (en) * | 2018-01-25 | 2018-09-04 | 杭州电子科技大学 | The extensive method of deep learning feature based on latent variable model |
CN108875818A (en) * | 2018-06-06 | 2018-11-23 | 西安交通大学 | Based on variation from code machine and confrontation network integration zero sample image classification method |
CN109447098A (en) * | 2018-08-27 | 2019-03-08 | 西北大学 | A kind of image clustering algorithm based on deep semantic insertion |
CN110309853A (en) * | 2019-05-20 | 2019-10-08 | 湖南大学 | Medical image clustering method based on variation self-encoding encoder |
-
2020
- 2020-01-10 CN CN202010024870.8A patent/CN111243045B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013061732A (en) * | 2011-09-12 | 2013-04-04 | Fuji Xerox Co Ltd | Image identification information provision program and image identification information provision device |
CN106952240A (en) * | 2017-03-29 | 2017-07-14 | 成都信息工程大学 | A kind of image goes motion blur method |
CN108171324A (en) * | 2017-12-26 | 2018-06-15 | 天津科技大学 | A kind of variation own coding mixed model |
CN108491925A (en) * | 2018-01-25 | 2018-09-04 | 杭州电子科技大学 | The extensive method of deep learning feature based on latent variable model |
CN108875818A (en) * | 2018-06-06 | 2018-11-23 | 西安交通大学 | Based on variation from code machine and confrontation network integration zero sample image classification method |
CN109447098A (en) * | 2018-08-27 | 2019-03-08 | 西北大学 | A kind of image clustering algorithm based on deep semantic insertion |
CN110309853A (en) * | 2019-05-20 | 2019-10-08 | 湖南大学 | Medical image clustering method based on variation self-encoding encoder |
Non-Patent Citations (2)
Title |
---|
张芳;郭春生: "基于非局部贝叶斯的泊松图像去噪算法" * |
鲍宗袍;陈华华;: "基于稀疏先验和边缘约束的图像盲去模糊算法" * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113822437A (en) * | 2020-06-18 | 2021-12-21 | 辉达公司 | Deep layered variational automatic encoder |
CN111860660A (en) * | 2020-07-24 | 2020-10-30 | 辽宁工程技术大学 | Small sample learning garbage classification method based on improved Gaussian network |
CN113255830A (en) * | 2021-06-21 | 2021-08-13 | 上海交通大学 | Unsupervised target detection method and system based on variational self-encoder and Gaussian mixture model |
CN114501034A (en) * | 2021-12-11 | 2022-05-13 | 同济大学 | Image compression method and medium based on discrete Gaussian mixture super-prior and Mask |
CN114501034B (en) * | 2021-12-11 | 2023-08-04 | 同济大学 | Image compression method and medium based on discrete Gaussian mixture super prior and Mask |
CN114638905B (en) * | 2022-01-30 | 2023-02-21 | 中国科学院自动化研究所 | Image generation method, device, equipment and storage medium |
CN114638905A (en) * | 2022-01-30 | 2022-06-17 | 中国科学院自动化研究所 | Image generation method, device, equipment, storage medium and computer program product |
CN115131347A (en) * | 2022-08-29 | 2022-09-30 | 江苏茂融智能科技有限公司 | Intelligent control method for processing zinc alloy parts |
CN115797216A (en) * | 2022-12-14 | 2023-03-14 | 齐鲁工业大学 | Inscription character restoration model and restoration method based on self-coding network |
CN117036862A (en) * | 2023-08-21 | 2023-11-10 | 武汉纺织大学 | Image generation method based on Gaussian mixture variation self-encoder |
CN117036862B (en) * | 2023-08-21 | 2024-03-22 | 武汉纺织大学 | Image generation method based on Gaussian mixture variation self-encoder |
CN116958712A (en) * | 2023-09-20 | 2023-10-27 | 山东建筑大学 | Image generation method, system, medium and device based on prior probability distribution |
CN116958712B (en) * | 2023-09-20 | 2023-12-15 | 山东建筑大学 | Image generation method, system, medium and device based on prior probability distribution |
Also Published As
Publication number | Publication date |
---|---|
CN111243045B (en) | 2023-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111243045B (en) | Image generation method based on Gaussian mixture model prior variation self-encoder | |
CN111242282B (en) | Deep learning model training acceleration method based on end edge cloud cooperation | |
CN110062934A (en) | The structure and movement in image are determined using neural network | |
CN113065974A (en) | Link prediction method based on dynamic network representation learning | |
CN111986105A (en) | Video time sequence consistency enhancing method based on time domain denoising mask | |
CN112884236B (en) | Short-term load prediction method and system based on VDM decomposition and LSTM improvement | |
CN116523079A (en) | Reinforced learning-based federal learning optimization method and system | |
CN112308961A (en) | Robot rapid robust three-dimensional reconstruction method based on layered Gaussian mixture model | |
CN113947133A (en) | Task importance perception element learning method for small sample image recognition | |
CN110570034A (en) | Bus load prediction method based on multi-XGboost model fusion | |
CN113743474A (en) | Digital picture classification method and system based on cooperative semi-supervised convolutional neural network | |
CN109540089B (en) | Bridge deck elevation fitting method based on Bayes-Kriging model | |
CN114723037A (en) | Heterogeneous graph neural network computing method for aggregating high-order neighbor nodes | |
CN113194493B (en) | Wireless network data missing attribute recovery method and device based on graph neural network | |
CN114880527B (en) | Multi-modal knowledge graph representation method based on multi-prediction task | |
CN116258923A (en) | Image recognition model training method, device, computer equipment and storage medium | |
CN113537613B (en) | Temporal network prediction method for die body perception | |
CN115941871A (en) | Video frame insertion method and device, computer equipment and storage medium | |
CN115759291A (en) | Space nonlinear regression method and system based on ensemble learning | |
CN115908600A (en) | Massive image reconstruction method based on prior regularization | |
CN115358485A (en) | Traffic flow prediction method based on graph self-attention mechanism and Hox process | |
CN114595890A (en) | Ship spare part demand prediction method and system based on BP-SVR combined model | |
CN113065321B (en) | User behavior prediction method and system based on LSTM model and hypergraph | |
Zhao et al. | Combining influence and sensitivity to factorize matrix for multi-context recommendation | |
Davis et al. | Residual Multi-Fidelity Neural Network Computing |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |