CN109165664A - A kind of attribute missing data collection completion and prediction technique based on generation confrontation network - Google Patents
A kind of attribute missing data collection completion and prediction technique based on generation confrontation network Download PDFInfo
- Publication number
- CN109165664A CN109165664A CN201810722774.3A CN201810722774A CN109165664A CN 109165664 A CN109165664 A CN 109165664A CN 201810722774 A CN201810722774 A CN 201810722774A CN 109165664 A CN109165664 A CN 109165664A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- prediction
- attribute
- filling
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention discloses a kind of based on the attribute missing data collection completion and prediction technique that generate confrontation network, comprising steps of 1) normalizing to data minmax, while being encoded to the attribute of discrete type using one hot, missing values are labeled as 0；2) the deletion sites coding vector about sample is established using data set；3) building production confrontation network and auxiliary prediction network carry out the prediction of data filling and label；4) result before minmax normalization is reduced to according to maximin in attribute；5) suitable hyper parameter is chosen by test；The present invention makes full use of Data distribution information and label information in data set, effective data filling can be carried out to high-dimensional missing data collection, simultaneously after training is completed, another auxiliary prediction network for including in this method is capable of prediction result of the attribute missing data to outgoing label of direct team's input, and process is simple and direct, has higher predictablity rate.
Description
Technical field
The present invention relates to the technical fields of data prediction, refer in particular to a kind of based on the attribute missing for generating confrontation network
Data set completion and prediction technique.
Background technique
Data set attribute lacks this phenomenon and is widely present in Various types of data concentration, usually acquires or transmits in data
During information lose caused by.Sample in data set, which loses one, can make subsequent foundation prediction, classification with multiple attributes
Model prediction accuracy decline.How completion is carried out to these missing datas, and contained using the sample with attribute missing
Information constructs high-precision prediction model, is the critical issue that data prediction faces.
Most statistical tools, which are taken, deletes the problem of missing sample corresponds to the mode processing attribute missing of row, column, or makes
Deletion sites are filled with the column median, average；This kind of mode is although efficient, convenient, but fails that sample is fully utilized
Notebook data distributed intelligence causes the inaccuracy of calculated result.It is past between data different attribute during multidimensional data processing
Toward there are many relevances, the relevance between these attributes can provide more information for the filling of data, it is contemplated that this
The data filling method of class relevance has smaller deviation when estimating missing values, excavates and lacks so as to depth
Lose the information that sample contains.
On this basis, further data filling method fills up missing values by modeling.Such as returning enthesis will
Missing attribute establishes regression equation as dependent variable and realizes prediction, and EM algorithm first initializes missing values, walks iteration with M by E step
To obtain final fill up as a result, k nearest neighbor algorithm (KNN) then calculates Euclidean distance matched sample concentration according to the attribute not lacked
K most like sample obtains filling up result by weighted average.These algorithms often in the enough situations of data volume, take
It obtains and is more accurately filled up than mean value, median as a result, being then also typically present some problems: returning in enthesis, need each category
Property between have a significant ground wire sexual intercourse, and the fill method based on EM algorithm, computation complexity are high, and are easily trapped into part most
It is excellent；Fill method based on k neighbour realizes that simply, but when facing big data quantity computationally intensive complexity is high to be caused to calculate
It is difficult.
In addition, the main purpose of data filling is to provide for more complete data for subsequent modeling and forecasting.With
The process of modeling is not directed in upper method, the data of filling often often have some associations with the label of prediction, will be pre-
The data that surveying model and fill method can combine to that filling is obtained play better prediction effect.For traditional number
There are computation complexity height when handling high dimensional data according to fill method, fail sufficiently to excavate label information to correct filling result
Both of these problems；The present invention will based on production confrontation e-learning data distribution carry out data filling, while establish one it is auxiliary
Being associated between the prediction abundant mining data of network and label is helped, so that its mutual information reaches maximum.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, propose a kind of scarce based on the attribute for generating confrontation network
Data set completion and prediction technique are lost, Data distribution information and label information in data set are made full use of, can be lacked to high-dimensional
It loses data set and carries out effective data filling, while after training is completed, another auxiliary prediction network for including in this method
Can be directly to the attribute missing data of input to the prediction result of outgoing label, process is simple and direct, has higher predictablity rate.
In order to achieve the above object, technical solution provided by the present invention are as follows: a kind of to be lacked based on the attribute for generating confrontation network
Data set completion and prediction technique are lost, mainly includes minmax firstly, carrying out data prediction for the data set of attribute missing
The one hot code conversion of normalization and discrete numerical variable；Then for the sample with attribute missing, building missing position
The coding vector set, thus the location information of expression deletion；Then the filling network and auxiliary prediction network of missing data are constructed
Synchronously complete filling and the Tag Estimation of missing data；After network training completion, the defeated of network is generated in network to fill
Result is filling as a result, the column maximin recorded when being normalized according to minmax carries out scale reduction out；Finally, passing through
Constantly modification hyper parameter observes it and completes the setting of hyper parameter in the loss of the prediction result of verifying collection；It includes the following steps
1) data prediction；
2) deletion sites coding vector is constructed；
3) building missing data filling network and auxiliary prediction network；
4) filling data scale reduction；
5) test is arranged with hyper parameter.
In step 1), different pretreatments is carried out to different types of data, the main data types being related to are divided into continuously
Continuous type numerical value is directly normalized in type numerical value and discrete type numerical value using minmax；For discrete type numerical value, turn
After turning to one hot coding, is normalized using minmax, 0 is uniformly filled for deletion sites；In addition, by data set whether
It is divided into two parts: the data that data and attribute with attribute missing do not lack.
In step 2), deletion sites coding vector is constructed, situation is: in data filling, the attribute of sample missing
Position is also a kind of important information, when being filled using neural network, it is only necessary to be filled out to the position of these missings
It fills, when constructing deletion sites coding vector, each column of all samples is traversed, if the attribute lacks, be denoted as
" 1 " is otherwise denoted as " 0 ", executes by this process, and each sample can have a deletion sites coding vector corresponding.
In step 3), building missing data filling network and auxiliary prediction network, situation is: the network is original
Production confrontation network has done following improvement: 1. removing noise obtained through stochastical sampling in the input for generating network；2. making
The data of filling are formed with the data of generation and deletion sites vector coding；In addition, the introducing of auxiliary prediction network is more abundant
The considerations of contacting between attribute and label, predicted simultaneously, to use auxiliary prediction network using attribute missing data
Loss carries out feedback calculating by BP algorithm and has updated generation network between prediction label and true tag, so that generate
Filling data has better effect when constructing prediction model；Joint production is fought the loss function in network and is assisted pre-
Loss function in survey grid network controls its weight ratio by hyper parameter, to determine the filling data distribution generated closer to complete
The distribution of data either enables to prediction model prediction more acurrate；Wherein, data filling network and auxiliary predict network
Structure includes to generate network, differentiate network, auxiliary prediction network；The structure of these three networks is described in detail below:
It generates network: importation structure being spliced by the corresponding deletion sites coding vector of the data lacked with attribute
At；Different according to the structure of data, hidden layer is able to use full articulamentum or warp lamination to constitute, especially in the number of input
When according to being picture type data, the filling data generated are operated using deconvolution；It is assumed herein that the data of input are denoted as I,
The vector of 100 dimensions, thus corresponding deletion sites coding vector is denoted as the dimension of E and 100, spliced obtained input to
Measuring dimension is 200；Hidden layer is made of full articulamentum, and activation primitive uses relu；Final output layer has 100 outputs
Unit, is denoted as O, and the activation primitive of output layer uses sigmoid；The data of filling I (1-E)+OE finally by being made of；
Differentiate network: the data of input have two parts, and first part is the filler obtained based on the output for generating network
According to as a result, second part is the sample data that attribute does not lack, output result is the decimal between 0~1, represents and differentiates that network is recognized
For the probability for the data whether received input data does not lack from attribute；According to the difference of input data type, network knot
The setting of structure is also different, when input data is image type data, is constructed by convolutional neural networks；It is assumed herein that input data
It is 100 dimensional vectors, then hidden layer can select to be made of full articulamentum, activation primitive is set as relu；Output layer only includes
One unit, activation primitive are selected as sigmoid, characterize probability；
Auxiliary prediction network: input and differentiation network are completely the same, and output is then the prediction to input sample about label
Value, using cross entropy as loss function, when forecasting problem is regression problem, is used when forecasting problem is classification problem
L2 norm or L1 norm are as loss function；Network structure is identical as the setting method of network is differentiated；It is assumed herein that input number
According to being 100 dimensional vectors, then hidden layer can select to be made of full articulamentum, activation primitive is set as relu；Output layer only wraps
Containing a unit, activation primitive is arranged in a manner described.
In step 4), scale reduction is carried out to the filling data of generation, due to pretreatment stage used minmax into
Data normalization of having gone can restore to obtain the knot of final filling according to the maxima and minima of each attribute of record
Fruit.
In step 5), test is arranged with hyper parameter, and situation is: for network during training, loss derives from two
Part: the prediction that production fights loss and auxiliary prediction network in network is lost；This two parts loses λ in different proportions
Combination obtains comprehensive loss；Different λ will affect the training of model；In operation, cutting data set be training set and
Test set, the λ of selection different scale, respectively 0.1 on training set, 0.3,0.5,0.7,0.9 is trained, meanwhile, it uses
Test set is tested, selection standard of the loss reduction of auxiliary prediction network as hyper parameter using on test set.
Compared with prior art, the present invention have the following advantages that with the utility model has the advantages that
1, traditional fill method such as median, mean value filling etc., method is simple, and filling effect is not good enough, and is based on
Often time complexity is big for the method for KNN, EM, and when handling high dimensional data collection, time complexity is very big, or even occurring can not
The case where processing.And production confrontation network has fabulous effect on the Distributed learning of high dimensional data, thus can solve
Certainly high dimensional data collection bring trouble；In addition the sample for not having usually attribute to lack is to obey with the missing sample with attribute
Same distribution, the data set for allowing filled data to approach no attribute missing from distribution enables to the result of filling not
Meeting bias data distribution, gives prediction model to carry out band negative effect.
2, traditional fill method does not consider filled data to the subsequent prediction result for establishing prediction model
It influences, step is usually that the data completed first are filled to missing data, recycles filled data to establish pre-
Model is surveyed, thus the effect of prediction cannot used to go the filling of guide data.The present invention is by introducing auxiliary prediction network meter
The loss progress backpropagation calculated between the value and true tag for the data prediction filled every time instructs the data for generating network to fill out
It fills, selects prediction effect so as to observe that the data of filling show quality on prediction model, in conjunction with differentiation network
The data of loss limitation filling and the difference of truthful data distribution, reach while having preferable filling effect with good prediction
As a result.Furthermore it completes after training, what is obtained is that a network end to end can directly obtain after the data is entered
The prediction result of auxiliary prediction network.
Detailed description of the invention
Fig. 1 is the flow chart that missing data is filled and predicted.
Fig. 2 is the production confrontation network and prediction network data flow graph for filling data.
Specific embodiment
The present invention is further explained in the light of specific embodiments.
As shown in Figure 1, based on the attribute missing data collection completion and prediction side for generating confrontation network provided by this example
Method, concrete condition are as follows:
1) data prediction: the data type of different attribute is different, and corresponding processing mode is also different.The main number being related to
It is divided into continuous type numerical value and discrete type numerical value according to type, for continuous type numerical value, is directly normalized using minmax；For
Discrete type numerical value is converted into after one hot coding, is normalized using minmax, uniformly fill 0 for deletion sites.Furthermore
Data set is divided into two parts: the data that data and attribute with attribute missing do not lack.
2) construct deletion sites coding vector: in data filling, the property location of sample missing is also a kind of important
Information, when being filled using neural network, it is only necessary to which the position of these missings is filled.It is compiled in building deletion sites
When code vector, each column of all samples are traversed, if the attribute lacks, " 1 " is denoted as, is otherwise denoted as " 0 ".By this stream
Cheng Zhihang, each sample can have a deletion sites coding vector corresponding.
3) building missing data filling network and auxiliary prediction network: the invention proposes one kind to fight net based on production
Network simultaneously combines auxiliary prediction network to carry out data filling while be able to carry out the integrated network of prediction.The network is in original life
Accepted way of doing sth confrontation network has done following improvement: the noise that sampling obtains 1. is removed in the input for generating network；2. using generating
Data and deletion sites vector coding form the data of filling.Furthermore the introducing of auxiliary prediction network more fully considers
Contacting between attribute and label, is predicted simultaneously using attribute missing data, predicts neural network forecast label using auxiliary
Loss carries out feedback calculating by BP algorithm and has updated generation network between true tag, so that the filling data generated
There is better effect when constructing prediction model.Joint production is fought in loss function and auxiliary prediction network in network
Loss function, its weight ratio is controlled by hyper parameter, come determine generate filling data distribution closer to partial data dividing
Cloth either enables to prediction model prediction more acurrate.Fig. 2 is that most important data filling network and auxiliary are pre- in the present invention
The structure chart of survey grid network, comprising generating network, differentiating network, auxiliary prediction network；The structure of these three networks is carried out below
Detailed introduction:
It generates network: importation structure being spliced by the corresponding deletion sites coding vector of the data lacked with attribute
At.Different according to the structure of data, hide can be used full articulamentum or warp lamination to constitute, especially in input layer by layer
When data are picture type data, the filling data that are generated using warp lamination.It is assumed herein that the data (being denoted as I) of input
The vector of 100 dimensions, thus corresponding deletion sites coding vector (being denoted as E) is also 100 dimensions, spliced obtained input to
Measuring dimension is 200；Hidden layer is made of full articulamentum, and activation primitive uses relu；Final output layer has 100 outputs
Unit (is denoted as O), and the activation primitive of output layer uses sigmoid.The data of filling I (1-E)+OE finally by being made of.
Differentiate network: the data of input have two parts, and first part is the filler obtained based on the output for generating network
According to as a result, second part is the sample data that attribute does not lack, output result is the decimal between 0~1, represents and differentiates that network is recognized
For the probability for the data whether received input data does not lack from attribute.According to the difference of input data type, network knot
The setting of structure is also different, when input data is image type data, can be constructed by convolutional neural networks.It is assumed herein that input number
According to being 100 dimensional vectors, constituted then hidden layer may be selected to be full articulamentum, activation primitive is set as relu；Output layer only includes
One unit, activation primitive are selected as sigmoid, characterize probability.
Auxiliary prediction network: input and differentiation network are completely the same, and output is then the prediction to input sample about label
Value, using cross entropy as loss function, when forecasting problem is regression problem, is used when forecasting problem is classification problem
L2 norm or L1 norm are as loss function.Network structure is identical as the setting method of network is differentiated.It is assumed herein that input number
According to being 100 dimensional vectors, constituted then hidden layer may be selected to be full articulamentum, activation primitive is set as relu；Output layer only includes
One unit, activation primitive are arranged in a manner described.
4) filling data scale reduction: since pretreatment stage has used minmax to carry out data normalization, according to note
The maxima and minima of each attribute of record, can restore to obtain the result of final filling.
5) test is arranged with hyper parameter: for network during training, loss fights net by production from two parts
The prediction of loss and auxiliary prediction network in network is lost；λ combines to obtain comprehensive damage the loss of this two parts in different proportions
It loses.Different λ will affect the training of model.In operation, cutting data set is training set and test set, on training set
The λ of selection different scale, respectively 0.1,0.3,0.5,0.7,0.9 is trained, meanwhile, it is tested using test set, with
Selection standard of the loss reduction of auxiliary prediction network as hyper parameter on test set.
Embodiment described above is only the preferred embodiments of the invention, and implementation model of the invention is not limited with this
It encloses, therefore all shapes according to the present invention, changes made by principle, should all be included within the scope of protection of the present invention.
Claims (6)
1. a kind of based on the attribute missing data collection completion and prediction technique that generate confrontation network, it is characterised in that: firstly, being directed to
The data set of attribute missing carries out data prediction, the main one hot including minmax normalization and discrete numerical variable
Code conversion；Then for the sample with attribute missing, the coding vector of deletion sites is constructed, thus the position of expression deletion
Information；Then it is pre- that filling and label of the filling network of missing data with auxiliary prediction Network Synchronization completion missing data are constructed
It surveys；It is filling as a result, being returned according to minmax to fill the output result of generation network in network after network training completion
The one column maximin recorded when changing carries out scale reduction；Finally, observing it in verifying collection by constantly modifying hyper parameter
Prediction result is lost to complete the setting of hyper parameter；It includes the following steps
1) data prediction；
2) deletion sites coding vector is constructed；
3) building missing data filling network and auxiliary prediction network；
4) filling data scale reduction；
5) test is arranged with hyper parameter.
2. a kind of attribute missing data collection completion and prediction technique based on generation confrontation network according to claim 1,
It is characterized by: carrying out different pretreatments to different types of data in step 1), the main data types being related to are divided into company
Continuous type numerical value is directly normalized in ideotype numerical value and discrete type numerical value using minmax；For discrete type numerical value,
It is converted into after one hot coding, is normalized using minmax, 0 is uniformly filled for deletion sites；In addition, by data set root
According to whether thering is attribute missing to be divided into two parts: data and the data that do not lack of attribute with attribute missing.
3. a kind of attribute missing data collection completion and prediction technique based on generation confrontation network according to claim 1,
It is characterized by: constructing deletion sites coding vector, situation is in step 2): in data filling, the category of sample missing
Property position be also a kind of important information, when being filled using neural network, it is only necessary to be carried out to the position of these missings
Filling, when constructing deletion sites coding vector, traverses each column of all samples, if the attribute lacks, is denoted as
" 1 " is otherwise denoted as " 0 ", executes by this process, and each sample can have a deletion sites coding vector corresponding.
4. a kind of attribute missing data collection completion and prediction technique based on generation confrontation network according to claim 1,
It is characterized by: building missing data filling network and auxiliary prediction network, situation is: the network is in original in step 3)
The production confrontation network of beginning has done following improvement: 1. removing noise in the input for generating network；2. using the number generated
According to the data for forming filling with deletion sites vector coding；In addition, the introducing of auxiliary prediction network more fully considers category
Property and label between contact, using attribute missing data predicted simultaneously, using auxiliary prediction neural network forecast label with
Loss carries out feedback calculating by BP algorithm and has updated generation network between true tag, so that the filling data generated exist
There is better effect when constructing prediction model；Joint production is fought in loss function and auxiliary prediction network in network
Loss function controls its weight ratio by hyper parameter, to determine the filling data distribution generated closer to the distribution of partial data
Either enable to prediction model prediction more acurrate；Wherein, the structure of data filling network and auxiliary prediction network includes life
At network, differentiate network, auxiliary prediction network；The structure of these three networks is described in detail below:
Generate network: importation is spliced to form by the corresponding deletion sites coding vector of the data lacked with attribute；
Different according to the structure of data, hidden layer is able to use full articulamentum or warp lamination to constitute, especially in the data of input
When being picture type data, the filling data generated are operated using deconvolution；It is assumed herein that the data of input are denoted as I, it is
The vector of 100 dimensions, thus corresponding deletion sites coding vector is denoted as the dimension of E and 100, through splicing obtained input vector
Dimension is 200；Hidden layer is made of full articulamentum, and activation primitive uses relu；Final output layer has 100 outputs single
Member, is denoted as O, and the activation primitive of output layer uses sigmoid；The data of filling I (1-E)+OE finally by being made of；
Differentiate network: the data of input have two parts, and first part is the filling data knot obtained based on the output for generating network
Fruit, second part are the sample datas that attribute does not lack, and output result is the decimal between 0~1, represent and differentiate that network is thought to connect
The probability for the data whether input data of receipts does not lack from attribute；According to the difference of input data type, network structure
Setting is also different, when input data is image type data, is constructed by convolutional neural networks；It is assumed herein that input data is
100 dimensional vectors, then hidden layer can select to be made of full articulamentum, activation primitive is set as relu；Output layer only includes one
A unit, activation primitive are selected as sigmoid, characterize probability；
Auxiliary prediction network: input and differentiation network are completely the same, and output is then the predicted value to input sample about label, when
When forecasting problem is classification problem, using cross entropy as loss function, when forecasting problem is regression problem, using L2 norm
Or L1 norm is as loss function；Network structure is identical as the setting method of network is differentiated；It is assumed herein that input data is 100
Dimensional vector, then hidden layer can select to be made of full articulamentum, activation primitive is set as relu；Output layer only includes a list
Member, activation primitive are arranged in a manner described.
5. a kind of attribute missing data collection completion and prediction technique based on generation confrontation network according to claim 1,
It is characterized by: scale reduction is carried out to the filling data of generation, since pretreatment stage has used minmax in step 4)
Data normalization has been carried out, according to the maxima and minima of each attribute of record, can restore to obtain final filling
As a result.
6. a kind of attribute missing data collection completion and prediction technique based on generation confrontation network according to claim 1,
It is characterized by: test is arranged with hyper parameter, and situation is: during training, loss derives from network in step 5)
Two parts: the prediction that production fights loss and auxiliary prediction network in network is lost；This two parts loses with different ratios
Example λ combines to obtain comprehensive loss；Different λ will affect the training of model；In operation, cutting data set is training set
And test set, the λ of selection different scale, respectively 0.1 on training set, 0.3,0.5,0.7,0.9 is trained, meanwhile, make
It is tested with test set, selection standard of the loss reduction of auxiliary prediction network as hyper parameter using on test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810722774.3A CN109165664B (en) | 2018-07-04 | 2018-07-04 | Attribute-missing data set completion and prediction method based on generation of countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810722774.3A CN109165664B (en) | 2018-07-04 | 2018-07-04 | Attribute-missing data set completion and prediction method based on generation of countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109165664A true CN109165664A (en) | 2019-01-08 |
| CN109165664B CN109165664B (en) | 2020-09-22 |
Family
ID=64897277
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810722774.3A Active CN109165664B (en) | 2018-07-04 | 2018-07-04 | Attribute-missing data set completion and prediction method based on generation of countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109165664B (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522973A (en) * | 2019-01-17 | 2019-03-26 | 云南大学 | Medical big data classification method and system based on production confrontation network and semi-supervised learning |
| CN109978257A (en) * | 2019-03-25 | 2019-07-05 | 上海赢科信息技术有限公司 | The continuation of insurance prediction technique and system of vehicle insurance |
| CN110046706A (en) * | 2019-04-18 | 2019-07-23 | 腾讯科技（深圳）有限公司 | Model generating method, device and server |
| CN110175168A (en) * | 2019-05-28 | 2019-08-27 | 山东大学 | A kind of time series data complementing method and system based on generation confrontation network |
| CN110647519A (en) * | 2019-08-30 | 2020-01-03 | 中国平安人寿保险股份有限公司 | Method and device for predicting missing attribute value in test sample |
| CN110728297A (en) * | 2019-09-04 | 2020-01-24 | 电子科技大学 | Low-cost antagonistic network attack sample generation method based on GAN |
| CN111046027A (en) * | 2019-11-25 | 2020-04-21 | 北京百度网讯科技有限公司 | Missing value filling method and device for time series data |
| CN111037365A (en) * | 2019-12-26 | 2020-04-21 | 大连理工大学 | Cutter state monitoring data set enhancing method based on generative countermeasure network |
| CN111177135A (en) * | 2019-12-27 | 2020-05-19 | 清华大学 | Landmark-based data filling method and device |
| CN111259953A (en) * | 2020-01-15 | 2020-06-09 | 云南电网有限责任公司电力科学研究院 | Equipment defect time prediction method based on capacitive equipment defect data |
| CN111259916A (en) * | 2020-02-12 | 2020-06-09 | 东华大学 | Low-rank projection feature extraction method under condition of label missing |
| CN111429605A (en) * | 2020-04-10 | 2020-07-17 | 郑州大学 | Missing value filling method based on generation type countermeasure network |
| CN111738007A (en) * | 2020-07-03 | 2020-10-02 | 北京邮电大学 | Chinese named entity identification data enhancement algorithm based on sequence generation countermeasure network |
| CN111737463A (en) * | 2020-06-04 | 2020-10-02 | 江苏名通信息科技有限公司 | Big data missing value filling method, device and computer program |
| CN112036955A (en) * | 2020-09-07 | 2020-12-04 | 贝壳技术有限公司 | User identification method and device, computer readable storage medium and electronic equipment |
| CN112183723A (en) * | 2020-09-17 | 2021-01-05 | 西北工业大学 | Data processing method for clinical detection data missing problem |
| CN112465150A (en) * | 2020-12-02 | 2021-03-09 | 南开大学 | Real data enhancement-based multi-element time sequence data filling method |
| CN112712855A (en) * | 2020-12-28 | 2021-04-27 | 华南理工大学 | Joint training-based clustering method for gene microarray containing deletion value |
| CN113010500A (en) * | 2019-12-18 | 2021-06-22 | 中国电信股份有限公司 | Processing method and processing system for DPI data |
| CN113515896A (en) * | 2021-08-06 | 2021-10-19 | 红云红河烟草(集团)有限责任公司 | Data missing value filling method for real-time cigarette acquisition |
| CN114826988A (en) * | 2021-01-29 | 2022-07-29 | 中国电信股份有限公司 | Method and device for anomaly detection and parameter filling of time sequence data |
| CN114936530A (en) * | 2022-06-22 | 2022-08-23 | 郑州大学 | Multi-element air quality data missing value filling model based on TAM and construction method thereof |
| CN115145906A (en) * | 2022-09-02 | 2022-10-04 | 之江实验室 | Preprocessing and completion method for structured data |
| WO2022222026A1 (en) * | 2021-04-19 | 2022-10-27 | 浙江大学 | Medical diagnosis missing data completion method and completion apparatus, and electronic device and medium |
| CN115829162A (en) * | 2023-01-29 | 2023-03-21 | 北京市农林科学院信息技术研究中心 | Crop yield prediction method, device, electronic device and medium |
| CN115883016A (en) * | 2022-10-28 | 2023-03-31 | 南京航空航天大学 | Method and device for enhancing flow data based on federal generation countermeasure network |
| CN117034142A (en) * | 2023-10-07 | 2023-11-10 | 之江实验室 | Unbalanced medical data missing value filling method and system |
| CN117150231A (en) * | 2023-10-27 | 2023-12-01 | 国网江苏省电力有限公司苏州供电分公司 | Measurement data filling method and system based on correlation and generation countermeasure network |
| CN117421548A (en) * | 2023-12-18 | 2024-01-19 | 四川互慧软件有限公司 | Method and system for treating loss of physiological index data based on convolutional neural network |
| CN117524318A (en) * | 2024-01-05 | 2024-02-06 | 深圳新合睿恩生物医疗科技有限公司 | New antigen heterogeneous data integration method and device, equipment and storage medium |
| CN117556267A (en) * | 2024-01-12 | 2024-02-13 | 闪捷信息科技有限公司 | Missing sample data filling method and device, storage medium and electronic equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106952239A (en) * | 2017-03-28 | 2017-07-14 | 厦门幻世网络科技有限公司 | image generating method and device |
| CN107133934A (en) * | 2017-05-18 | 2017-09-05 | 北京小米移动软件有限公司 | Image completion method and device |
| AU2017101166A4 (en) * | 2017-08-25 | 2017-11-02 | Lai, Haodong MR | A Method For Real-Time Image Style Transfer Based On Conditional Generative Adversarial Networks |
| KR20170137350A (en) * | 2016-06-03 | 2017-12-13 | (주)싸이언테크 | Apparatus and method for studying pattern of moving objects using adversarial deep generative model |
| CN107945118A (en) * | 2017-10-30 | 2018-04-20 | 南京邮电大学 | A kind of facial image restorative procedure based on production confrontation network |
| CN107945140A (en) * | 2017-12-20 | 2018-04-20 | 中国科学院深圳先进技术研究院 | A kind of image repair method, device and equipment |
-
2018
- 2018-07-04 CN CN201810722774.3A patent/CN109165664B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20170137350A (en) * | 2016-06-03 | 2017-12-13 | (주)싸이언테크 | Apparatus and method for studying pattern of moving objects using adversarial deep generative model |
| CN106952239A (en) * | 2017-03-28 | 2017-07-14 | 厦门幻世网络科技有限公司 | image generating method and device |
| CN107133934A (en) * | 2017-05-18 | 2017-09-05 | 北京小米移动软件有限公司 | Image completion method and device |
| AU2017101166A4 (en) * | 2017-08-25 | 2017-11-02 | Lai, Haodong MR | A Method For Real-Time Image Style Transfer Based On Conditional Generative Adversarial Networks |
| CN107945118A (en) * | 2017-10-30 | 2018-04-20 | 南京邮电大学 | A kind of facial image restorative procedure based on production confrontation network |
| CN107945140A (en) * | 2017-12-20 | 2018-04-20 | 中国科学院深圳先进技术研究院 | A kind of image repair method, device and equipment |
Non-Patent Citations (1)
| Title |
|---|
| JINSUNG YOON ET AL.: "GAIN: Missing Data Imputation using Generative Adversarial Nets", 《PROCEEDINGS OF THE 35 TH INTERNATIONAL CONFERENCE ON MACHINE》 * |
Cited By (49)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522973A (en) * | 2019-01-17 | 2019-03-26 | 云南大学 | Medical big data classification method and system based on production confrontation network and semi-supervised learning |
| CN109978257A (en) * | 2019-03-25 | 2019-07-05 | 上海赢科信息技术有限公司 | The continuation of insurance prediction technique and system of vehicle insurance |
| CN110046706A (en) * | 2019-04-18 | 2019-07-23 | 腾讯科技（深圳）有限公司 | Model generating method, device and server |
| CN110046706B (en) * | 2019-04-18 | 2022-12-20 | 腾讯科技（深圳）有限公司 | Model generation method and device and server |
| CN110175168B (en) * | 2019-05-28 | 2021-06-01 | 山东大学 | Time sequence data filling method and system based on generation of countermeasure network |
| CN110175168A (en) * | 2019-05-28 | 2019-08-27 | 山东大学 | A kind of time series data complementing method and system based on generation confrontation network |
| CN110647519A (en) * | 2019-08-30 | 2020-01-03 | 中国平安人寿保险股份有限公司 | Method and device for predicting missing attribute value in test sample |
| CN110647519B (en) * | 2019-08-30 | 2023-10-03 | 中国平安人寿保险股份有限公司 | Method and device for predicting missing attribute value in test sample |
| CN110728297A (en) * | 2019-09-04 | 2020-01-24 | 电子科技大学 | Low-cost antagonistic network attack sample generation method based on GAN |
| CN110728297B (en) * | 2019-09-04 | 2021-08-06 | 电子科技大学 | Low-cost antagonistic network attack sample generation method based on GAN |
| CN111046027A (en) * | 2019-11-25 | 2020-04-21 | 北京百度网讯科技有限公司 | Missing value filling method and device for time series data |
| CN113010500A (en) * | 2019-12-18 | 2021-06-22 | 中国电信股份有限公司 | Processing method and processing system for DPI data |
| CN111037365A (en) * | 2019-12-26 | 2020-04-21 | 大连理工大学 | Cutter state monitoring data set enhancing method based on generative countermeasure network |
| CN111037365B (en) * | 2019-12-26 | 2021-08-20 | 大连理工大学 | Cutter state monitoring data set enhancing method based on generative countermeasure network |
| CN111177135B (en) * | 2019-12-27 | 2020-11-10 | 清华大学 | Landmark-based data filling method and device |
| CN111177135A (en) * | 2019-12-27 | 2020-05-19 | 清华大学 | Landmark-based data filling method and device |
| CN111259953B (en) * | 2020-01-15 | 2023-10-20 | 云南电网有限责任公司电力科学研究院 | Equipment defect time prediction method based on capacitive equipment defect data |
| CN111259953A (en) * | 2020-01-15 | 2020-06-09 | 云南电网有限责任公司电力科学研究院 | Equipment defect time prediction method based on capacitive equipment defect data |
| CN111259916A (en) * | 2020-02-12 | 2020-06-09 | 东华大学 | Low-rank projection feature extraction method under condition of label missing |
| CN111429605A (en) * | 2020-04-10 | 2020-07-17 | 郑州大学 | Missing value filling method based on generation type countermeasure network |
| CN111737463B (en) * | 2020-06-04 | 2024-02-09 | 江苏名通信息科技有限公司 | Big data missing value filling method, device and computer readable memory |
| CN111737463A (en) * | 2020-06-04 | 2020-10-02 | 江苏名通信息科技有限公司 | Big data missing value filling method, device and computer program |
| CN111738007A (en) * | 2020-07-03 | 2020-10-02 | 北京邮电大学 | Chinese named entity identification data enhancement algorithm based on sequence generation countermeasure network |
| CN112036955A (en) * | 2020-09-07 | 2020-12-04 | 贝壳技术有限公司 | User identification method and device, computer readable storage medium and electronic equipment |
| CN112183723A (en) * | 2020-09-17 | 2021-01-05 | 西北工业大学 | Data processing method for clinical detection data missing problem |
| CN112465150A (en) * | 2020-12-02 | 2021-03-09 | 南开大学 | Real data enhancement-based multi-element time sequence data filling method |
| CN112712855A (en) * | 2020-12-28 | 2021-04-27 | 华南理工大学 | Joint training-based clustering method for gene microarray containing deletion value |
| CN112712855B (en) * | 2020-12-28 | 2022-09-20 | 华南理工大学 | Joint training-based clustering method for gene microarray containing deletion value |
| CN114826988A (en) * | 2021-01-29 | 2022-07-29 | 中国电信股份有限公司 | Method and device for anomaly detection and parameter filling of time sequence data |
| WO2022222026A1 (en) * | 2021-04-19 | 2022-10-27 | 浙江大学 | Medical diagnosis missing data completion method and completion apparatus, and electronic device and medium |
| CN113515896A (en) * | 2021-08-06 | 2021-10-19 | 红云红河烟草(集团)有限责任公司 | Data missing value filling method for real-time cigarette acquisition |
| CN113515896B (en) * | 2021-08-06 | 2022-08-09 | 红云红河烟草(集团)有限责任公司 | Data missing value filling method for real-time cigarette acquisition |
| CN114936530A (en) * | 2022-06-22 | 2022-08-23 | 郑州大学 | Multi-element air quality data missing value filling model based on TAM and construction method thereof |
| CN115145906B (en) * | 2022-09-02 | 2023-01-03 | 之江实验室 | Preprocessing and completion method for structured data |
| US11841839B1 (en) | 2022-09-02 | 2023-12-12 | Zhejiang Lab | Preprocessing and imputing method for structural data |
| CN115145906A (en) * | 2022-09-02 | 2022-10-04 | 之江实验室 | Preprocessing and completion method for structured data |
| CN115883016B (en) * | 2022-10-28 | 2024-02-02 | 南京航空航天大学 | Flow data enhancement method and device based on federal generation countermeasure network |
| CN115883016A (en) * | 2022-10-28 | 2023-03-31 | 南京航空航天大学 | Method and device for enhancing flow data based on federal generation countermeasure network |
| CN115829162A (en) * | 2023-01-29 | 2023-03-21 | 北京市农林科学院信息技术研究中心 | Crop yield prediction method, device, electronic device and medium |
| CN117034142A (en) * | 2023-10-07 | 2023-11-10 | 之江实验室 | Unbalanced medical data missing value filling method and system |
| CN117034142B (en) * | 2023-10-07 | 2024-02-09 | 之江实验室 | Unbalanced medical data missing value filling method and system |
| CN117150231A (en) * | 2023-10-27 | 2023-12-01 | 国网江苏省电力有限公司苏州供电分公司 | Measurement data filling method and system based on correlation and generation countermeasure network |
| CN117150231B (en) * | 2023-10-27 | 2024-01-26 | 国网江苏省电力有限公司苏州供电分公司 | Measurement data filling method and system based on correlation and generation countermeasure network |
| CN117421548A (en) * | 2023-12-18 | 2024-01-19 | 四川互慧软件有限公司 | Method and system for treating loss of physiological index data based on convolutional neural network |
| CN117421548B (en) * | 2023-12-18 | 2024-03-12 | 四川互慧软件有限公司 | Method and system for treating loss of physiological index data based on convolutional neural network |
| CN117524318A (en) * | 2024-01-05 | 2024-02-06 | 深圳新合睿恩生物医疗科技有限公司 | New antigen heterogeneous data integration method and device, equipment and storage medium |
| CN117524318B (en) * | 2024-01-05 | 2024-03-22 | 深圳新合睿恩生物医疗科技有限公司 | New antigen heterogeneous data integration method and device, equipment and storage medium |
| CN117556267A (en) * | 2024-01-12 | 2024-02-13 | 闪捷信息科技有限公司 | Missing sample data filling method and device, storage medium and electronic equipment |
| CN117556267B (en) * | 2024-01-12 | 2024-04-02 | 闪捷信息科技有限公司 | Missing sample data filling method and device, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109165664B (en) | 2020-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109165664A (en) | A kind of attribute missing data collection completion and prediction technique based on generation confrontation network | |
| CN109492822B (en) | Air pollutant concentration time-space domain correlation prediction method | |
| CN109508360B (en) | Geographical multivariate stream data space-time autocorrelation analysis method based on cellular automaton | |
| Vladislavleva et al. | Predicting the energy output of wind farms based on weather data: Important variables and their correlation | |
| CN112684379A (en) | Transformer fault diagnosis system and method based on digital twinning | |
| AU2021240155A1 (en) | Control Pulse Generation Method, Apparatus, System, Device And Storage Medium | |
| CN111079977A (en) | Heterogeneous federated learning mine electromagnetic radiation trend tracking method based on SVD algorithm | |
| CN101093559A (en) | Method for constructing expert system based on knowledge discovery | |
| CN106503035A (en) | A kind of data processing method of knowledge mapping and device | |
| CN113486190B (en) | Multi-mode knowledge representation method integrating entity image information and entity category information | |
| CN113255895A (en) | Graph neural network representation learning-based structure graph alignment method and multi-graph joint data mining method | |
| CN110705178A (en) | Tunnel/subway construction overall process surrounding rock deformation dynamic prediction method based on machine learning | |
| CN103885867B (en) | Online evaluation method of performance of analog circuit | |
| CN114707754A (en) | Intelligent ammeter fault prediction method and system based on BiLSTM-CNN model | |
| CN109614896A (en) | A method of the video content semantic understanding based on recursive convolution neural network | |
| CN104732067A (en) | Industrial process modeling forecasting method oriented at flow object | |
| CN115049124A (en) | Deep and long tunnel water inrush prediction method based on Bayesian network | |
| Wang et al. | Roof pressure prediction in coal mine based on grey neural network | |
| Ballı et al. | An application of artificial neural networks for prediction and comparison with statistical methods | |
| CN107729942A (en) | A kind of sorting technique of structured view missing data | |
| CN111898746A (en) | Deep learning method for association of interrupted flight path continuation | |
| Yang et al. | Using genetic algorithms for time series prediction | |
| Zhang et al. | RSVRs based on feature extraction: a novel method for prediction of construction projects’ costs | |
| CN110210523A (en) | A kind of model based on shape constraint diagram wears clothing image generating method and device | |
| CN113887471A (en) | Video time sequence positioning method based on feature decoupling and cross comparison |
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 |