CN113127806A - Regression analysis model selection method based on machine learning - Google Patents
Regression analysis model selection method based on machine learning Download PDFInfo
- Publication number
- CN113127806A CN113127806A CN202110417540.XA CN202110417540A CN113127806A CN 113127806 A CN113127806 A CN 113127806A CN 202110417540 A CN202110417540 A CN 202110417540A CN 113127806 A CN113127806 A CN 113127806A
- Authority
- CN
- China
- Prior art keywords
- data
- function
- regression analysis
- machine learning
- model
- 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
- 238000000611 regression analysis Methods 0.000 title claims abstract description 29
- 238000010801 machine learning Methods 0.000 title claims abstract description 21
- 238000010187 selection method Methods 0.000 title claims abstract description 12
- 230000006870 function Effects 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 24
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000013527 convolutional neural network Methods 0.000 claims description 21
- 238000012887 quadratic function Methods 0.000 claims description 13
- 238000012886 linear function Methods 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 9
- 238000012888 cubic function Methods 0.000 claims description 6
- 238000003384 imaging method Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 description 8
- 230000014509 gene expression Effects 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000004304 visual acuity Effects 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 241000234314 Zingiber Species 0.000 description 1
- 235000006886 Zingiber officinale Nutrition 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000008397 ginger Nutrition 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
Abstract
The invention discloses a regression analysis model selection method based on machine learning, which solves the problems that the traditional regression analysis model selection process is complex and the application range of the existing machine learning for data regression analysis is limited, the technical scheme is that a candidate continuous function model set which can be used for regression is selected, a machine training data image is generated by using a data enhancement strategy according to the selected continuous function model set, the regression analysis model selection method based on machine learning can provide a regression model with clear logic at high accuracy rate and realize the judgment capability of the neural network on the regression model of real data at low cost.
Description
Technical Field
The invention relates to a regression analysis technology, in particular to a regression analysis model selection method based on machine learning.
Background
As an important data analysis means, regression analysis is widely used in the fields of monitoring economic activities, construction engineering, medicine, industrial manufacturing, and the like. The objective is to find a functional relationship between the various variables using a data fitting method. However, there is always a problem of choice of analytical model in the data fitting process. In the conventional regression analysis, the selection process of the analysis model is complicated.
In recent years, with the rapid development of advanced computing technologies, regression analysis methods based on machine learning have gained wide attention. The data regression method based on the BP (Back-Propagation) neural network can explain and predict data with extremely high accuracy. Although a regression model does not need to be selected, the complex network structure and the numerous network parameters in machine learning hide the logical relationship between data. So that the analysis results thereof cannot be directly understood by humans. In addition, a great amount of training data needs to be stored in advance to construct a neural network with an accurate analysis result, so that the application range of performing data regression analysis by using a machine learning neural network is limited.
Disclosure of Invention
The invention aims to provide a regression analysis model selection method based on machine learning, which can provide a regression model with clear logic at high accuracy and realize the judgment capability of a neural network on the regression model of real data at low cost.
The technical purpose of the invention is realized by the following technical scheme:
a regression analysis model selection method based on machine learning comprises the following steps:
s1, selecting a candidate continuous function model set which can be used for regression;
s2, generating a machine training data image by using a data enhancement strategy according to the selected continuous function model set;
s3, establishing a convolution neural network, inputting training data images and training;
and S4, analyzing the data to be analyzed by using the trained convolutional neural network, and giving an optimal regression model.
Preferably, the set of continuous function models includes linear functions, quadratic functions, higher-order polynomial functions, exponential functions, and inverse proportional functions.
Preferably, the generating of the training data image according to the data enhancement strategy specifically includes:
s21, assuming that the data group to be analyzed is { X: X1,x2,…,xnY and { Y: Y }1,y2,…,ynH, interval [ x ] according to data to be analyzed1,xn]Calculating an accurate function value set on the interval according to the function relation of the alternative continuous function model;
s22, adding noise based on the accurate value to generate noise data Y'1,y′2,…,y′n}；
And S23, calculating the central difference value of the noise data, imaging the central difference value, and establishing a training and verification data image set.
Preferably, the imaging of the central difference value of each candidate function data by the data enhancement strategy specifically comprises: the linear function is expressed in the form of a horizontal straight line image, the quadratic function is expressed in the form of an inclined straight line image, and the cubic function is expressed in the form of a curved image.
Preferably, the established convolutional neural network sequentially comprises a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and three full-connection layers; the third full-link layer is set to output in N categories according to the number N of models of the input image data.
In conclusion, the invention has the following beneficial effects:
the training data image is generated by selectively using the alternative regression model and is used for training the neural network, so that the judgment capability of the neural network on the regression model of the real data can be realized at low cost, a distinguishable data quality standard can be given according to the resolution capability of the convolutional neural network, no conclusion is given under the condition that the data quality does not meet the requirement, and the overfitting phenomenon in the traditional analysis method is prevented.
Drawings
FIG. 1 is a flow chart of the present method;
FIG. 2 is a schematic view of the process flow structure of the present method;
FIG. 3 is a difference curve of four different function models with a function parameter selected as 1;
FIG. 4 is a schematic diagram of a convolutional neural network structure;
FIG. 5 is a graph of neural network resolving power under different values of parameter b/c;
FIG. 6 is a graph of experimental neural network resolving power;
fig. 7 is a graph showing actual data collected in the experiment in this example.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
In the conventional regression analysis, the selection process of the analysis model is complicated. Firstly, pre-judging is carried out in various fitting functions according to experience, then, fitting operation is carried out by adopting various functions, and finally, the best fitting model is judged according to the relevant indexes of fitting results. Although professional statistical software has provided various common model fitting functions, the implementation is still time-consuming and labor-consuming and depends on manual judgment, and the possibility of over-fitting cannot be eliminated. To overcome this difficulty, a general regression analysis model based on functional families was proposed for ginger Happo and Liyanru. The model determines parameters of the family of functions in a fitting process to determine a fitted model. However, due to the limitation of the expression form of the function family, the types of functions provided by the function family are limited, and the problem of model selection in regression analysis cannot be solved well.
According to one or more embodiments, a regression analysis model selection method based on machine learning is disclosed, as shown in fig. 1 and 2, including the following steps:
s1, selecting a candidate continuous function model set which can be used for regression;
s2, generating a machine training data image by using a data enhancement strategy according to the selected continuous function model set;
s3, establishing a convolution neural network, inputting training data images and training;
and S4, analyzing the data to be analyzed by using the trained convolutional neural network, and giving an optimal regression model.
Wherein the content of the first and second substances,
in step S1, the operation of selecting the candidate continuous function model set has a high degree of freedom, and the candidate continuous function model set can be screened empirically, which can effectively reduce the computation workload. The selection of the functional form may include various common functions and complex functions. A set including various common continuous functions may include: the higher-order polynomial function, quadratic function, linear function, exponential function, inverse proportional function, etc. can be expressed as the following formulas
y＝ax3+bx2+cx+d (1)
y＝bx2+cx+d (2)
y＝cx+d (3)
y＝aekx+c (4)
In step S2, a training data image is generated by using a data enhancement strategy, and an effective data enhancement strategy needs to be able to convert each candidate function into images of different classes without being affected by the function parameters (a, b, c, d, k) so as to train a convolutional neural network that obtains a recognizable candidate function model.
Two properties of the data to be analyzed, namely the data interval and the relative uncertainty, determine whether a deterministic regression analysis model can be given. First, the behavior of various nonlinear models varies across different data intervals. In addition, large relative uncertainties result in signals that are not resolvable.
Suppose the data set to be analyzed is { X: X1,x2,…,xnY and { Y: Y }1,y2,…,yn}. When generating training data, firstly, according to the interval of data to be analyzed, i.e. [ x ]1,xn]Calculating an accurate function value set on the interval according to the alternative function relationship; then, based on the accurate value, noise, i.e., relative uncertainty, is added to generate noise data { Y ': Y'1,y′2,…,y′n}; and finally, calculating the central difference value of the noise data, imaging the central difference value, and establishing a training and verification data image set.
This data enhancement algorithm focuses on the center difference of each candidate function data. Under acceptable noise conditions, the noise is typically less than 20%, i.e., the noise data is at a relative standard deviation σ<0.1, a linear function is expressed as an approximate horizontal straight line image, a quadratic function is expressed as an approximate oblique straight line image, a cubic function is expressed as an approximate curved line image, and the like. On the basis of different images, the convolutional neural network can separate various functional relationships according to the extraction capability of complex data features. In the expression process, the mapping of the function and the image features only depends on the value range of the function parameters, but not on the specific function constants (a, b, c, d, k), so that the training data image can be generated to train the neural network under the condition of presetting the value range of the function constants, and the trained neural network is used for real data. As shown in FIG. 2, the graphs (a), (b), (c) and (d) show 15 differential curves of linear function, quadratic function, cubic polynomial and exponential function, respectively. In the image, noise data of 11 points is generated under the condition that the relative standard deviation σ is 0.02, and the generation interval of the data is [ x [ ]1＝0.1,xn＝10.1]. The function parameters a, b, c, d, k are all selected to be 1, and when the function parameters are greater than 1, the differential curve characteristics are similar to those of fig. 2. As can be seen from FIG. 3, the difference curves of the different functions are clearDifferent image features are displayed. It can be identified and classified by a convolutional neural network.
In step S3, after the training and verification data image set is established using the candidate function model, a convolutional neural network is established, and the training data image is input into the established convolutional neural network for training.
As shown in fig. 4, the convolutional neural network used in the method has 7 layers, which includes two convolutional layers, two pooling layers, and three fully-connected layers. The data input sequentially passes through a first convolutional layer Conv1, a first pooling layer Pool1, a second convolutional layer Conv2, a second pooling layer Pool2, three full-connection layers IP1Relu1, IP2Relu2 and IP3Relu3, and the last full-connection layer sets output N according to the number of models to be classified, and is responsible for carrying out N classification output on the input image data.
Preferably, the parameters of the convolutional neural network are set as: c1And C2The convolution kernel size of the two convolution layers is 5 x 5, P1And P2The two pooling layers adopt a maximum pooling algorithm, the size of a sampling window is 3 multiplied by 3, the first layer of full-connection layer generates 384 neuron outputs, the second layer of full-connection layer has 192 neurons, and the last layer of full-connection layer has 2 neurons.
In step S4, the trained convolutional neural network analyzes and judges the data to be analyzed.
Generating data to be analyzed { X: X) according to the data to be analyzed1,x2,…,xnY and { Y: Y }1,y2,…,ynAnd (4) inputting the central difference image into the trained convolutional neural network for judgment, and selecting a regression analysis model most suitable for the data by the neural network.
Although different function models have different mathematical expression forms, the mathematical expressions of the function models have high similarity under a certain function parameter value condition. In the comparison of the quadratic function model with the linear function model, if the parameters b/c of the equations (2) and (3) are a small quantity, they have highly similar data expressions. Similarly, if the parameter ratio a/b in formula (1) and formula (2) is a small quantity or k in formula (4) is a small quantity, then the cubic function and the quadratic function, or the exponential and the linear function also have highly similar data expressions. By means of the ultra-strong data feature extraction capability of the convolutional neural network, the model selection method can distinguish slight differences among models.
Performing a limit test of model resolving power on the trained neural network, as shown in fig. 5, where the abscissa is the relative standard deviation σ of the noise data and the ordinate is the resolving accuracy, the larger the parameter b/c is, the easier it is to resolve and the influence of noise is negligible in the comparison between the quadratic function model and the linear function model, and the limit of the parameter b/c is
I.e. the parameter b/c as a limit value for a small quantity, as shown in fig. 6, for different parameter ratios
And under the condition that a/b is 0.1, the neural network distinguishes the capacity curve of the linear model line and the quadratic model curve and distinguishes the capacity curve of the quadratic function curve and the cubic function model cubic curve. When the parameter k is a small parameter, the network judgment rate of the exponential function model and the linear function model is always kept high, and is not shown here. The analysis results show that the convolutional neural network can distinguish tiny nonlinear parameters. This is a capability that conventional regression analysis does not have.
For clarity, the actual data collected in the laboratory was analyzed using the present method according to the regression analysis model selection method.
And selecting an experimental data regression analysis model by a machine learning method. The experimental data were derived from classical attwood linear uniform acceleration object experiments. The speed and position data of the object during acceleration are measured by the Pasco sensor. Table 1 gives the time (seconds s) and position (meters m) data collected in one experiment.
| Time | 5.123 | 5.364 | 5.557 | 5.718 | 5.86 | 5.99 | 6.11 | 6.222 | 6.327 | 6.426 |
| Position of | 0.05 | 0.08 | 0.11 | 0.14 | 0.16 | 0.2 | 0.22 | 0.26 | 0.29 | 0.32 |
| Time | 6.521 | 6.612 | 6.699 | 6.783 | 6.863 | 6.941 | 7.017 | 7.09 | 7.162 | / |
| Position of | 0.35 | 0.37 | 0.4 | 0.44 | 0.46 | 0.5 | 0.53 | 0.55 | 0.58 | / |
TABLE 1
Since the experimental data exhibit a significant increasing function morphology, linear, quadratic, cubic and exponential functions were chosen as candidate models before construction of the recognition network. The mathematical formula for the alternative four functions is shown in (1-4), with the function parameters (a, b, c, d, k) set to be greater than zero. The observation interval for experimental data is only [5,7], distinguishing different functions in such a small interval requires high quality data. In creating the training data pictures, the doping noise relative uncertainty was set to 0.3%. More specifically, 2000 training pictures and 500 verification pictures are generated for each candidate model. When a quadratic function picture is generated, the value range of the quadratic term coefficient is set to be one to ten times of the slope of the linear function. In the coefficient setting of the cubic function, the cubic-term coefficient is set to one to ten times the quadratic-term coefficient. The trained network can identify different functions in one interval. The setting interval of the parameter of the exponential function K is [1,2], and only obvious exponential relation is judged here.
The parameters of the convolutional neural network are set as described above. After 10 rounds of network training, the network had 99.8% resolution for the four functions. Finally, the data given in table 1 is pictured, and as shown in fig. 7, sent to the network for determination. The network gives the optimal model as a quadratic function, and the result proves that the result completely conforms to the theoretical model.
Based on the method, the problem of selecting the traditional regression analysis model is solved by using a machine learning method. Depending on the strong data characteristic analysis capability of machine learning, the method converts the regression model selection problem into the classification problem of the convolutional neural network, and compared with the traditional regression model selection method, the method has three advantages:
firstly, the method uses the alternative regression model to generate training data to train the neural network, and can realize the judgment capability of the neural network on the regression model of the real data with low cost;
secondly, the method provides a regression model with clear logic with high accuracy of machine learning;
finally, the method provides a quality standard of distinguishable data according to the resolution capability of the convolutional neural network. Under the condition that the data quality is not qualified, no conclusion is given, so that the overfitting phenomenon in the traditional analysis method is prevented.
The present embodiment is only for explaining the present invention, and it is not limited to the present invention, and those skilled in the art can make modifications of the present embodiment without inventive contribution as needed after reading the present specification, but all of them are protected by patent law within the scope of the claims of the present invention.
Claims (5)
1. A regression analysis model selection method based on machine learning is characterized by comprising the following steps:
s1, selecting a candidate continuous function model set which can be used for regression;
s2, generating a machine training data image by using a data enhancement strategy according to the selected continuous function model set;
s3, establishing a convolution neural network, inputting training data images and training;
and S4, analyzing the data to be analyzed by using the trained convolutional neural network, and giving an optimal regression model.
2. The method of claim 1 for selecting a regression analysis model based on machine learning, wherein: the continuous function model set comprises a linear function, a quadratic function, a high-order polynomial function, an exponential function and an inverse proportional function.
3. The method of selecting a regression analysis model based on machine learning of claim 2, wherein the generating of the training data image according to the data enhancement strategy is specifically:
s21, assuming that the data group to be analyzed is { X: X1,x2,…,xnY and { Y: Y }1,y2,…,ynH, interval [ x ] according to data to be analyzed1,xn]Calculating an accurate function value set on the interval according to the function relation of the alternative continuous function model;
s22, adding noise based on the accurate value to generate noise data Y'1,y′2,…,y′n}；
And S23, calculating the central difference value of the noise data, imaging the central difference value, and establishing a training and verification data image set.
4. The method for selecting a regression analysis model based on machine learning according to claim 3, wherein the imaging of the central difference value of each candidate function data by the data enhancement strategy is specifically as follows: the linear function is expressed in the form of a horizontal straight line image, the quadratic function is expressed in the form of an inclined straight line image, and the cubic function is expressed in the form of a curved image.
5. The method of claim 1 for selecting a regression analysis model based on machine learning, wherein: the established convolutional neural network sequentially comprises a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and three full-connection layers; the third full-link layer is set to output in N categories according to the number N of models of the input image data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110417540.XA CN113127806A (en) | 2021-04-19 | 2021-04-19 | Regression analysis model selection method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110417540.XA CN113127806A (en) | 2021-04-19 | 2021-04-19 | Regression analysis model selection method based on machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113127806A true CN113127806A (en) | 2021-07-16 |
Family
ID=76777647
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110417540.XA Pending CN113127806A (en) | 2021-04-19 | 2021-04-19 | Regression analysis model selection method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113127806A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103974311A (en) * | 2014-05-21 | 2014-08-06 | 哈尔滨工业大学 | Condition monitoring data stream anomaly detection method based on improved gaussian process regression model |
| CN108023876A (en) * | 2017-11-20 | 2018-05-11 | 西安电子科技大学 | Intrusion detection method and intruding detection system based on sustainability integrated study |
| CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
| CN109508746A (en) * | 2018-11-16 | 2019-03-22 | 西安电子科技大学 | Pulsar candidate's body recognition methods based on convolutional neural networks |
| CN110163278A (en) * | 2019-05-16 | 2019-08-23 | 东南大学 | A kind of flame holding monitoring method based on image recognition |
| CN110321844A (en) * | 2019-07-04 | 2019-10-11 | 北京万里红科技股份有限公司 | A kind of quick iris detection method based on convolutional neural networks |
| CN112418266A (en) * | 2020-10-15 | 2021-02-26 | 南昌大学 | Pile foundation integrity classification and identification method based on convolutional neural network |
| CN112434732A (en) * | 2020-11-17 | 2021-03-02 | 西安交通大学 | Deep learning classification method based on feature screening |
| CN112613617A (en) * | 2020-12-30 | 2021-04-06 | 清华大学 | Uncertainty estimation method and device based on regression model |
-
2021
- 2021-04-19 CN CN202110417540.XA patent/CN113127806A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103974311A (en) * | 2014-05-21 | 2014-08-06 | 哈尔滨工业大学 | Condition monitoring data stream anomaly detection method based on improved gaussian process regression model |
| CN108023876A (en) * | 2017-11-20 | 2018-05-11 | 西安电子科技大学 | Intrusion detection method and intruding detection system based on sustainability integrated study |
| CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
| CN109508746A (en) * | 2018-11-16 | 2019-03-22 | 西安电子科技大学 | Pulsar candidate's body recognition methods based on convolutional neural networks |
| CN110163278A (en) * | 2019-05-16 | 2019-08-23 | 东南大学 | A kind of flame holding monitoring method based on image recognition |
| CN110321844A (en) * | 2019-07-04 | 2019-10-11 | 北京万里红科技股份有限公司 | A kind of quick iris detection method based on convolutional neural networks |
| CN112418266A (en) * | 2020-10-15 | 2021-02-26 | 南昌大学 | Pile foundation integrity classification and identification method based on convolutional neural network |
| CN112434732A (en) * | 2020-11-17 | 2021-03-02 | 西安交通大学 | Deep learning classification method based on feature screening |
| CN112613617A (en) * | 2020-12-30 | 2021-04-06 | 清华大学 | Uncertainty estimation method and device based on regression model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109765053B (en) | Rolling bearing fault diagnosis method using convolutional neural network and kurtosis index | |
| CN110048827B (en) | Class template attack method based on deep learning convolutional neural network | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN109389171B (en) | Medical image classification method based on multi-granularity convolution noise reduction automatic encoder technology | |
| CN108399434B (en) | Analysis and prediction method of high-dimensional time series data based on feature extraction | |
| CN111145145B (en) | Image surface defect detection method based on MobileNet | |
| CN107832789B (en) | Feature weighting K nearest neighbor fault diagnosis method based on average influence value data transformation | |
| CN112820322B (en) | Semi-supervised audio event labeling method based on self-supervised contrast learning | |
| CN113869208A (en) | Rolling bearing fault diagnosis method based on SA-ACWGAN-GP | |
| CN115062272A (en) | Water quality monitoring data abnormity identification and early warning method | |
| CN112699941A (en) | Plant disease severity image classification method and device, computer equipment and storage medium | |
| CN113887342A (en) | Equipment fault diagnosis method based on multi-source signals and deep learning | |
| CN108846128B (en) | Cross-domain text classification method based on adaptive noise reduction encoder | |
| CN114357870A (en) | Metering equipment operation performance prediction analysis method based on local weighted partial least squares | |
| CN110688484B (en) | Microblog sensitive event speech detection method based on unbalanced Bayesian classification | |
| CN112149355A (en) | Soft measurement method based on semi-supervised dynamic feedback stack noise reduction self-encoder model | |
| CN113127806A (en) | Regression analysis model selection method based on machine learning | |
| Padhy et al. | Image classification in artificial neural network using fractal dimension | |
| CN110837787A (en) | Multispectral remote sensing image detection method and system for three-party generated countermeasure network | |
| WO2022077345A1 (en) | Method and apparatus for neural network based on energy-based latent variable models | |
| Kiritoshi et al. | L1-Norm Gradient Penalty for Noise Reduction of Attribution Maps. | |
| CN114692507A (en) | Counting data soft measurement modeling method based on stacking Poisson self-encoder network | |
| CN114723998A (en) | Small sample image classification method and device based on large-boundary Bayes prototype learning | |
| CN113011748A (en) | Recommendation effect evaluation method and device, electronic equipment and readable storage medium | |
| CN112733903A (en) | Air quality monitoring and alarming method, system, device and medium based on SVM-RF-DT combination |
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: 20210716 |