CN112116073A - Gaussian distribution data adjusting method based on improved GAN network - Google Patents
Gaussian distribution data adjusting method based on improved GAN network Download PDFInfo
- Publication number
- CN112116073A CN112116073A CN202010985207.4A CN202010985207A CN112116073A CN 112116073 A CN112116073 A CN 112116073A CN 202010985207 A CN202010985207 A CN 202010985207A CN 112116073 A CN112116073 A CN 112116073A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- gaussian distribution
- generated
- distribution
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a Gaussian distribution data adjustment method based on an improved GAN network. However, the GAN network only changes the distribution of data and does not change the data itself, so that the generated data is consistent with the real data in characteristics. The invention is explained by taking Gaussian distribution as an example, and can also use other data statistical distribution rules to carry out directional normalization so as to meet the expected requirements of data generation.
Description
Technical Field
The invention relates to a Gaussian distribution data adjusting method based on an improved GAN network, and belongs to the technical field of machine learning.
Background
A Generative Adaptive Networks (GAN) is a deep learning model, and is one of the most promising methods for unsupervised learning in complex distribution in recent years. The model passes through (at least) two modules in the framework: the mutual game learning of the Generative Model (Generative Model) and the Discriminative Model (Discriminative Model) yields a reasonably good output.
It is often necessary that the data generated by the GAN network is consistent with the actual data, but at the same time, the data is distributed according to a certain expected rule, that is, a certain adjustment is made to the direction of the generated data. However, no scheme for adjusting the direction of generated data through a GAN network exists at present.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a method for adjusting Gaussian distribution data based on an improved GAN network, wherein the generated data output by the GAN network conforms to the Gaussian distribution rule, and meanwhile, the generated data is consistent with the real data in characteristics.
In order to solve the technical problem, the technical scheme adopted by the invention is as follows: a Gaussian distribution data adjusting method based on an improved GAN network comprises the following steps:
s01), transmitting the natural random number, the artificial setting value or the data generated by other systems to a data generation network, and outputting the generated data by the data generation network according to the input data;
s02), the generated data in the step S01 and the real data for measuring the similarity of the generated data are transmitted to a data judging network, and the Loss1 is obtained by the data judging network according to the similarity of the generated data and the real data;
s03), transmitting the n data generated in batch by the data generation network to a Gaussian distribution deviation judgment network, counting the distribution of the n data generated in batch by the Gaussian distribution deviation judgment network by using a central limit theorem, and calculating the distance between the distribution and normal Gaussian distribution to obtain Loss 2;
s04), Loss2 and Loss1 are weighted and summed to serve as the overall Loss of the data generation network, the image generation network is trained by the overall Loss, and the trained data generation network adjusts input data according with Gaussian distribution.
Furthermore, the image generation network, the image discrimination network and the Gaussian distribution deviation judgment network are trained or counted in sequence, wherein when one of the networks is trained or counted, other networks are kept unchanged.
Further, in step S03, the mean square error is calculated as the Loss2 from the normal gaussian distribution and the distribution of the n pieces of data generated in batch.
Further, the overall Loss of the data generation network is set to Loss, and then Loss = Loss2+ λ Loss1, where λ is an adjustable weighting coefficient.
The invention has the beneficial effects that: the invention is used for improving the Gaussian distribution data adjustment of the GAN network, and the randomly generated data is distributed according to a specified mode no matter how the original real data is distributed. However, the GAN network only changes the distribution of data and does not change the data itself, so that the generated data is consistent with the real data in characteristics. The invention is explained by taking Gaussian distribution as an example, and can also use other data statistical distribution rules to carry out directional normalization so as to meet the expected requirements of data generation.
Drawings
FIG. 1 is a flow chart of the method.
Detailed Description
The invention is further described with reference to the following figures and specific embodiments.
Example 1
The embodiment discloses a method for adjusting gaussian distribution data based on an improved GAN network, as shown in fig. 1, including the following steps:
s01), transmitting the data AZi to a data generation network, and outputting the generated data after the data generation network correspondingly calculates the input data AZi;
in this embodiment, the data AZi represents a natural random number, and may also be data set manually or generated by other systems, such as data or images collected by a transmitter;
s02), the generated data in the step S01 and the real data BZi for measuring the similarity of the generated data are transmitted to a data judging network, and the data judging network obtains Loss1 according to the similarity of the generated data and the real data;
s03), transmitting the n data generated in batch by the data generation network to a Gaussian distribution deviation judgment network, counting the distribution of the n data generated in batch by the Gaussian distribution deviation judgment network by using a central limit theorem, and calculating the distance between the distribution and normal Gaussian distribution to obtain Loss 2;
center limitTheorems show that when N is large, the sum of mutually independent random variables approximately follows a normal distribution N (N μ, N σ)2) Thus, the target normal distribution function is N (N μ, N σ)2) The key problem solved by the method is that under the condition of certain n, the approximation degree of the target positive function is as high as possible.
S04), Loss2 and Loss1 are weighted and summed to serve as the overall Loss of the data generation network, the image generation network is trained by the overall Loss, and the trained data generation network adjusts input data according with Gaussian distribution.
In this embodiment, the Loss2 and the Loss1 are weighted and summed through the gate control unit, specifically, the weighting coefficient λ and the Loss1 are multiplied through the and gate, and the Loss2 and the Loss λ Loss1 are summed through the or gate.
In this embodiment, the image generation network, the image discrimination network, and the gaussian distribution deviation determination network are trained or counted in sequence, and when one of the networks is trained or counted, the other networks remain unchanged.
In step S03, the mean square error is calculated from the normal gaussian distribution and the distribution of the n pieces of batch-generated data, and the mean square error is defined as Loss 2.
The process of judging the Loss obtained by the network by the data is as follows: loss1 is the output of the data discrimination network, i.e. the characteristic distance (such as euclidean distance, cosine distance, etc.) between the generated data and the real data; according to the GAN basic theory, training is carried out in two stages, wherein when a network is judged by training, batch data is generated as negative samples and real data is generated as positive samples on the assumption that generated network parameters are unchanged, and the network is trained; when the generated network is trained, the discrimination network uses the trained model to judge the characteristic distance between the input generated data and the real data (namely Loss 1).
And setting the overall Loss of the data generation network to be less, wherein less = less 2+ λ less 1, and λ is an adjustable weighting coefficient, that is, the weights of less 2 and less 1 are adjustable.
In this embodiment, the GAN network has the same principle as a general GAN network in other aspects, and the image generation network, the image discrimination network, and the gaussian distribution deviation determination network may be common VGG, rescet, or the like or a custom network.
The present embodiment is used to improve the gaussian distribution data adjustment of GAN networks, and the randomly generated data is distributed in a specified manner no matter how the original real data is distributed. However, the GAN network only changes the distribution of data and does not change the data itself, so that the generated data is consistent with the real data in characteristics. The invention is explained by taking Gaussian distribution as an example, and can also use other data statistical distribution rules to carry out directional normalization so as to meet the expected requirements of data generation.
The foregoing description is only for the basic principle and the preferred embodiments of the present invention, and modifications and substitutions by those skilled in the art according to the present invention belong to the protection scope of the present invention.
Claims (4)
1. A Gaussian distribution data adjusting method based on an improved GAN network is characterized in that: the method comprises the following steps:
s01), transmitting the natural random number, the artificial setting value or the data generated by other systems to a data generation network, and outputting the generated data by the data generation network according to the input data;
s02), the generated data in the step S01 and the real data for measuring the similarity of the generated data are transmitted to a data judging network, and the Loss1 is obtained by the data judging network according to the similarity of the generated data and the real data;
s03), transmitting the n data generated in batch by the data generation network to a Gaussian distribution deviation judgment network, counting the distribution of the n data generated in batch by the Gaussian distribution deviation judgment network by using a central limit theorem, and calculating the distance between the distribution and normal Gaussian distribution to obtain Loss 2;
s04), Loss2 and Loss1 are weighted and summed to serve as the overall Loss of the data generation network, the image generation network is trained by the overall Loss, and the trained data generation network adjusts input data according with Gaussian distribution.
2. The method of adjusting gaussian distribution data based on an improved GAN network as claimed in claim 1, wherein: the image generation network, the image discrimination network and the Gaussian distribution deviation judgment network are trained or counted in sequence, wherein when one of the networks is trained or counted, other networks are kept unchanged.
3. The method of adjusting gaussian distribution data based on an improved GAN network as claimed in claim 1, wherein: in step S03, the mean square error is calculated from the normal gaussian distribution and the distribution of the n pieces of batch-generated data, and the mean square error is defined as Loss 2.
4. The method of adjusting gaussian distribution data based on an improved GAN network as claimed in claim 1, wherein: and the overall Loss of the data generation network is set to Loss, then Loss = Loss2+ λ Loss1, where λ is an adjustable weighting coefficient.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010985207.4A CN112116073A (en) | 2020-09-18 | 2020-09-18 | Gaussian distribution data adjusting method based on improved GAN network |
| PCT/CN2021/117026 WO2022057691A1 (en) | 2020-09-18 | 2021-09-07 | Gaussian distribution data adjustment method based on improved gan |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010985207.4A CN112116073A (en) | 2020-09-18 | 2020-09-18 | Gaussian distribution data adjusting method based on improved GAN network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112116073A true CN112116073A (en) | 2020-12-22 |
Family
ID=73800066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010985207.4A Pending CN112116073A (en) | 2020-09-18 | 2020-09-18 | Gaussian distribution data adjusting method based on improved GAN network |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112116073A (en) |
| WO (1) | WO2022057691A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022057691A1 (en) * | 2020-09-18 | 2022-03-24 | 神思电子技术股份有限公司 | Gaussian distribution data adjustment method based on improved gan |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11315231B2 (en) * | 2018-06-08 | 2022-04-26 | Industrial Technology Research Institute | Industrial image inspection method and system and computer readable recording medium |
| CN109146988B (en) * | 2018-06-27 | 2022-08-30 | 南京邮电大学 | Incomplete projection CT image reconstruction method based on VAEGAN |
| CN109446339B (en) * | 2018-10-11 | 2021-08-06 | 广东工业大学 | Knowledge graph representation method based on multi-core Gaussian distribution |
| CN112116073A (en) * | 2020-09-18 | 2020-12-22 | 神思电子技术股份有限公司 | Gaussian distribution data adjusting method based on improved GAN network |
-
2020
- 2020-09-18 CN CN202010985207.4A patent/CN112116073A/en active Pending
-
2021
- 2021-09-07 WO PCT/CN2021/117026 patent/WO2022057691A1/en active Application Filing
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022057691A1 (en) * | 2020-09-18 | 2022-03-24 | 神思电子技术股份有限公司 | Gaussian distribution data adjustment method based on improved gan |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2022057691A1 (en) | 2022-03-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109617888B (en) | Abnormal flow detection method and system based on neural network | |
| CN107607954A (en) | A kind of FNN rainfall particle phase identification methods based on T S models | |
| CN108052968B (en) | QSFLA-SVM perception intrusion detection method | |
| CN112561139A (en) | Short-term photovoltaic power generation power prediction method and system | |
| CN115600119A (en) | Data processing method and system suitable for wind power generation | |
| CN112801332B (en) | Short-term wind speed prediction method based on gray level co-occurrence matrix | |
| CN112116073A (en) | Gaussian distribution data adjusting method based on improved GAN network | |
| CN114977939B (en) | Double-fed fan control parameter identification method based on improved multi-target particle swarm algorithm | |
| CN115907195A (en) | Photovoltaic power generation power prediction method, system, electronic device and medium | |
| CN111027229B (en) | Wind power curve fitting method based on sparse heteroscedastic multi-spline regression | |
| CN108154186B (en) | Pattern recognition method and device | |
| CN115829027A (en) | Comparative learning-based federated learning sparse training method and system | |
| CN112991362A (en) | Color sorter adhesion material image segmentation method based on Gaussian mixture model | |
| CN103824285B (en) | Image segmentation method based on bat optimal fuzzy clustering | |
| CN111080001A (en) | Deep neural network prediction method applied to wind speed of wind power plant | |
| CN111311025B (en) | Load prediction method based on meteorological similar days | |
| CN108665002A (en) | A kind of two classification task label noises tolerance grader learning method | |
| CN109919235B (en) | Deep learning image classification model training method based on manual intervention sample set weight | |
| CN116362328A (en) | Federal learning heterogeneous model aggregation method based on fairness characteristic representation | |
| CN111343115A (en) | 5G communication modulation signal identification method and system | |
| CN111369072A (en) | Nuclear minimum mean square time sequence online prediction model based on sparsification method | |
| CN115659323A (en) | Intrusion detection method based on information entropy theory and convolution neural network | |
| CN113205214A (en) | Photovoltaic power ultra-short-term prediction method based on parallel dendritic neural network | |
| CN113516170A (en) | Image classification method based on Bayesian neural network random addition saturation structure | |
| CN112818152A (en) | Data enhancement method and device of deep clustering model |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201222 |