CN110309869A - Stabilization recognition methods and device towards unknown scene - Google Patents
Stabilization recognition methods and device towards unknown scene Download PDFInfo
- Publication number
- CN110309869A CN110309869A CN201910558189.9A CN201910558189A CN110309869A CN 110309869 A CN110309869 A CN 110309869A CN 201910558189 A CN201910558189 A CN 201910558189A CN 110309869 A CN110309869 A CN 110309869A
- Authority
- CN
- China
- Prior art keywords
- variable
- historical sample
- entered
- sample collection
- historical
- 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
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The invention discloses a kind of stabilization recognition methods towards unknown scene and devices, wherein this method comprises: obtaining historical sample collection；Processing is weighted to historical sample collection and generates historical sample collection to be entered；It calculates historical sample to be entered and concentrates causality between each predictive variable and outcome variable；According to the model learning that preset machine learning method concentrates the weight of each historical sample to be entered, causality to be weighted historical sample to be entered, cause and effect regression coefficient between each predictive variable and outcome variable is concentrated to obtain historical sample to be entered；Test data is identified according to cause and effect regression coefficient.This method may be implemented towards the stability forecast in the case of the specified mistake of model and data Unknown Distribution.
Description
Technical field
The present invention relates to machine learning techniques field, in particular to a kind of stabilization recognition methods and dress towards unknown scene
It sets.
Background technique
The validity of many machine learning methods needs following two to assume to guarantee: 1) test data is with model training number
According to being independent identically distributed；2) specified model is correct.However, we test no future in many practical problems
The priori knowledge of data set and data true model.Based on the specified model of mistake, test data between training data with being distributed
Difference will lead to machine learning method and be distributed the problems such as different test data prediction is unstable to following.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.
For this purpose, this method can an object of the present invention is to provide a kind of stabilization recognition methods towards unknown scene
To realize towards the stability forecast in the case of the specified mistake of model and data Unknown Distribution.
It is another object of the present invention to propose a kind of stabilization identification device towards unknown scene.
In order to achieve the above objectives, one aspect of the present invention embodiment proposes a kind of stable identification side towards unknown scene
Method, comprising:
Obtain historical sample collection；
Processing is weighted to the historical sample collection and generates historical sample collection to be entered；
It calculates the historical sample to be entered and concentrates causality between each predictive variable and outcome variable；
Added according to weight, the causality of the preset machine learning method to the historical sample collection to be entered
The model learning of power, with obtain the historical sample to be entered concentrate between each predictive variable and the outcome variable because
Fruit regression coefficient；
Test data is identified according to the cause and effect regression coefficient.
The stabilization recognition methods towards unknown scene of the embodiment of the present invention, is restored by being weighted to historical data
Causal correlation between variable, removes false related, and the physical meaning of primitive character can be saved by sample weighting, in conjunction with its because
Fruit relationship is able to achieve interpretable stability forecast, and excavates the bound term of causalnexus between variable, can be with various engineerings
Learning method combines, and promotes its stability forecast.
In addition, the stabilization recognition methods according to the above embodiment of the present invention towards unknown scene can also have it is following attached
The technical characteristic added:
Further, in one embodiment of the invention, the data of the historical sample collection are and the test data
Relevant historical data.
Further, in one embodiment of the invention, described that processing generation is weighted to the historical sample collection
Historical sample collection to be entered, comprising:
Wherein, X is the matrix for inputting historical sample, and every row indicates that a sample, each column indicate one-dimensional prediction variable, W
For the weight for inputting historical sample, a is variable X _ j square order, and b is variable X _ k square order, T representing matrix transposition.
Further, in one embodiment of the invention, the predictive variable is multiple, and the outcome variable is one
It is a.
Further, in one embodiment of the invention, the test data is calculated by the cause and effect regression coefficient
The predicted value of middle outcome variable, to be identified to the test data.
In order to achieve the above objectives, another aspect of the present invention embodiment proposes a kind of stable identification dress towards unknown scene
It sets, comprising:
Module is obtained, for obtaining historical sample collection；
Processing module generates historical sample collection to be entered for being weighted processing to the historical sample collection；
First computing module, for calculate the historical sample to be entered concentrate each predictive variable and outcome variable it
Between causality；
Second computing module, for according to preset machine learning method to the weight of the historical sample collection to be entered,
The model learning that the causality is weighted concentrates each predictive variable and institute to obtain the historical sample to be entered
State the cause and effect regression coefficient between outcome variable；
Identification prediction module, for being identified according to the cause and effect regression coefficient to test data.
The stabilization identification device towards unknown scene of the embodiment of the present invention, is restored by being weighted to historical data
Causal correlation between variable, removes false related, and the physical meaning of primitive character can be saved by sample weighting, in conjunction with its because
Fruit relationship is able to achieve interpretable stability forecast, and excavates the bound term of causalnexus between variable, can be with various engineerings
Learning method combines, and promotes its stability forecast.
In addition, the stabilization identification device according to the above embodiment of the present invention towards unknown scene can also have it is following attached
The technical characteristic added:
Further, in one embodiment of the invention, the data of the historical sample collection are and the test data
Relevant historical data.
Further, in one embodiment of the invention, described that processing generation is weighted to the historical sample collection
Historical sample collection to be entered, comprising:
Wherein, X is the matrix for inputting historical sample, and every row indicates that a sample, each column indicate one-dimensional prediction variable, W
For the weight for inputting historical sample, a is variable X _ j square order, and b is variable X _ k square order, T representing matrix transposition.
Further, in one embodiment of the invention, the predictive variable is multiple, and the outcome variable is one
It is a.
Further, in one embodiment of the invention, the test data is calculated by the cause and effect regression coefficient
The predicted value of middle outcome variable, to be identified to the test data.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is the stabilization recognition methods flow chart towards unknown scene according to one embodiment of the invention；
Fig. 2 is the stabilization recognition methods block diagram towards unknown scene according to one embodiment of the invention；
Fig. 3 is the stabilization identification device structural schematic diagram towards unknown scene according to one embodiment of the invention.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
The stabilization recognition methods proposed according to embodiments of the present invention towards unknown scene and dress are described with reference to the accompanying drawings
It sets.
In forecasting problem, towards the test data set that all possible future is unknown, realize that the stabilization of interpretation is pre-
It surveys.For example, how the historical data based on some policlinic patient, predict the state of an illness of following patient, Yi Jiru
The problems such as what predicts the patient in other cities, the present invention can excavate causalnexus, realization pair by de-correlation
Future patient across hospital realizes the stability forecast of the state of an illness.
The stabilization recognition methods towards unknown scene proposed according to embodiments of the present invention is described with reference to the accompanying drawings first.
Fig. 1 is the stabilization recognition methods flow chart towards unknown scene according to one embodiment of the invention.
As shown in Figure 1, should stabilization recognition methods towards unknown scene the following steps are included:
In step s101, historical sample collection is obtained.
Specifically, the prediction to the following unknown data is realized by historical data, the historical sample collection of acquisition is and test
The relevant data of data then need the history based on certain hospital patient for example, predicting the state of an illness of certain hospital future patient
Data.
In step s 102, processing is weighted to historical sample collection and generates historical sample collection to be entered.
Processing is weighted to the historical sample collection of acquisition, obtains historical sample collection to be entered, historical sample collection is carried out
After weighting, the falseness removed between the predictive variable of sample and outcome variable is related, restores between predictive variable and outcome variable
Causality, sufficiently excavate variable between causalnexus bound term, realize towards model formulate mistake stability forecast.
Specifically, processing is weighted to historical sample collection and generates historical sample collection to be entered, comprising:
Wherein, X is the matrix for inputting historical sample, and every row indicates that a sample, each column indicate one-dimensional prediction variable, W
For the weight for inputting historical sample, a is variable X _ j square order, and b is variable X _ k square order, T representing matrix transposition.
In step s 103, it calculates historical sample to be entered and concentrates cause and effect between each predictive variable and outcome variable
Relationship.
After being weighted to historical sample collection, the physical meaning of primitive character can be saved by sample weighting, is calculated
Interpretable stability forecast is able to achieve in conjunction with its causality to the causality between predictive variable and outcome variable.
It is understood that predictive variable is the variable that historical sample to be entered is concentrated, for example, sample set to be entered is more
Scenery picture is opened, then predictive variable is multiple scenery pictures, and each picture is all a predictive variable, is carried out to picture pre-
Survey, whether have dog in predicted pictures, be then outcome variable with the presence or absence of dog, herein, predictive variable be it is multiple, outcome variable is
One, predict whether there is dog in plurality of pictures.
Further, the sample weights W learnt by previous step is weighted sample, can make predictive variable
Between it is mutually indepedent, guarantee the correlation between each predictive variable and outcome variable be derive from its to outcome variable because
Fruit effect, that is, restored the causality between predictive variable and outcome variable.
In step S104, according to preset machine learning method to weight, the causality of historical sample collection to be entered
The model learning being weighted concentrates cause and effect between each predictive variable and outcome variable to obtain historical sample to be entered
Regression coefficient.
Specifically, using the weight of historical sample collection to be entered, in conjunction with any machine learning method, the model that is weighted
Study, assesses the cause and effect regression coefficient of each predictive variable and outcome variable.
It is understood that realizing the machine learning method of sample weighting to sample weighting based on sample weights W.Due to
It can restore the causality between all predictive variables and outcome variable, the machine learning method of weighting by sample weights W
It is that model learning is carried out based on the causality between predictive variable and outcome variable, the regression coefficient for finally learning out can be with
Referred to as cause and effect regression coefficient.
In step s105, test data is identified according to cause and effect regression coefficient.
The predicted value of outcome variable in test data is calculated, by cause and effect regression coefficient to identify to test data.
Specifically, the cause and effect regression coefficient for learning out based on previous step, the predictive variable of binding test data, Ke Yiji
The predicted value of all sample results variables in test data set is calculated, test data is predicted in realization.
In the present invention, mainly for the inconsistent equal challenge of the data distribution under model errors particular cases, needle is proposed
To property measure, the stability forecast that mistake is specified towards model is realized:
1) challenge one: the inconsistent incidence relation meaned between its variable of distribution between data set is different, will lead to
Model prediction unstability.The invention proposes a kind of methods based on Causal model, are closed by the cause and effect between assessment variable
System, and stability forecast is realized using causal stable invariance.
2) challenge two: model assumption mistake can make unrelated predictive variable related to there is falseness between outcome variable.This hair
It is bright to propose variable de-correlation, make any predictive variable all uncorrelated by sample weighting, to eliminate unrelated prediction
Falseness between variable and outcome variable is related, realizes the stability forecast that mistake is specified towards model.
The present invention considers the stability forecast problem under model errors are specified, and passes through causal approach and de-correlation technique proposes
Variable de-correlation in de-correlation restores cause and effect between variable by the variable balancing technique of sample weighting
Correlation removes false correlation, improves machine learning method in the case where specifying error situation towards model to unknown test data
Predict stability.
By attached drawing 2, the present invention is described in detail.
As shown in Fig. 2, including two parts, variable de-correlation modules and stability forecast module.
Specifically, it in variable de-correlation modules, proposes to learn global sample weights W, so as to any one predictive variable
X_i and all other predictive variable X_ {-i }={ X/X_i } are independent or uncorrelated.All predictive variables are all independently of each other or not
The cause-effect that correlation makes the correlation between each predictive variable and outcome variable be only derived from it to outcome variable.Also
To say, the sample weights for learning out by variable decorrelation part, can restore any predictive variable X and outcome variable Y it
Between causality.
Second part is stability forecast module, which is supervised learning, has used the predictive variable and knot of historical sample
Fruit variable.
Specifically, it is proposed in stablizing study module using the sample weights W for learning out based on variable decorrelation model
The machine learning method of sample weighting.By sample weights W to sample weighting, all predictive variables and outcome variable can be restored
Between causality.Therefore, stablizing study module, what we utilized is the causality between variable to instruct prediction mould
The study of type.Due to causal invariance (stability), the stability forecast module of proposition can be real to all unknown scenes
Existing stability forecast.
The stabilization recognition methods towards unknown scene proposed according to embodiments of the present invention, by adding to historical data
It weighs to restore causal correlation between variable, removes false correlation, the physical meaning of primitive character can be saved by sample weighting,
In conjunction with its causality, be able to achieve interpretable stability forecast, and excavate the bound term of causalnexus between variable, can with it is each
Kind machine learning method combines, and promotes its stability forecast.
The stabilization identification device towards unknown scene proposed according to embodiments of the present invention is described referring next to attached drawing.
Fig. 3 is the stabilization identification device structural schematic diagram towards unknown scene according to one embodiment of the invention.
As shown in figure 3, should stabilization identification device towards unknown scene include: to obtain module 100, processing module 200, the
One computing module 300, the second computing module 400 and identification prediction module 500.
Wherein, module 100 is obtained, for obtaining historical sample collection.
Processing module 200 generates historical sample collection to be entered for being weighted processing to historical sample collection.
First computing module 300, for calculate historical sample to be entered concentrate each predictive variable and outcome variable it
Between causality.
Second computing module 400, for according to preset machine learning method to the weight of historical sample collection to be entered, because
The model learning that fruit relationship is weighted is concentrated between each predictive variable and outcome variable with obtaining historical sample to be entered
Cause and effect regression coefficient.
Identification prediction module 500, for being identified according to cause and effect regression coefficient to test data.
Further, in one embodiment of the invention, the data of historical sample collection are go through relevant to test data
History data.
Further, in one embodiment of the invention, processing is weighted to historical sample collection and generates to be entered go through
History sample set, comprising:
Wherein, X is the matrix for inputting historical sample, and every row indicates that a sample, each column indicate one-dimensional prediction variable, W
For the weight for inputting historical sample, a is variable X _ j square order, and b is variable X _ k square order, T representing matrix transposition.
Further, in one embodiment of the invention, predictive variable is multiple, and outcome variable is one.
Further, in one embodiment of the invention, result in test data is calculated by cause and effect regression coefficient to become
The predicted value of amount, to be identified to test data.
It should be noted that the aforementioned explanation to the stabilization recognition methods embodiment towards unknown field scape is also applied for
The device of the embodiment, details are not described herein again.
The stabilization identification device towards unknown scene proposed according to embodiments of the present invention, by adding to historical data
It weighs to restore causal correlation between variable, removes false correlation, the physical meaning of primitive character can be saved by sample weighting,
In conjunction with its causality, be able to achieve interpretable stability forecast, and excavate the bound term of causalnexus between variable, can with it is each
Kind machine learning method combines, and promotes its stability forecast.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiment or examples.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (10)
1. a kind of stabilization recognition methods towards unknown scene, which comprises the following steps:
Obtain historical sample collection；
Processing is weighted to the historical sample collection and generates historical sample collection to be entered；
It calculates the historical sample to be entered and concentrates causality between each predictive variable and outcome variable；
It is weighted according to weight, the causality of the preset machine learning method to the historical sample collection to be entered
Model learning concentrates the cause and effect between each predictive variable and the outcome variable to return to obtain the historical sample to be entered
Return coefficient；
Test data is identified according to the cause and effect regression coefficient.
2. the method according to claim 1, wherein
The data of the historical sample collection are historical data relevant to the test data.
3. the method according to claim 1, wherein described be weighted processing generation to the historical sample collection
Historical sample collection to be entered, comprising:
Wherein, X is the matrix for inputting historical sample, and every row indicates that a sample, each column indicate that one-dimensional prediction variable, W are defeated
Enter the weight of historical sample, a is variable X _ j square order, and b is variable X _ k square order, T representing matrix transposition.
4. the method according to claim 1, wherein the predictive variable be it is multiple, the outcome variable be one
It is a.
5. the method according to claim 1, wherein calculating the test data by the cause and effect regression coefficient
The predicted value of middle outcome variable, to be identified to the test data.
6. a kind of stabilization identification device towards unknown scene characterized by comprising
Module is obtained, for obtaining historical sample collection；
Processing module generates historical sample collection to be entered for being weighted processing to the historical sample collection；
First computing module is concentrated between each predictive variable and outcome variable for calculating the historical sample to be entered
Causality；
Second computing module, for according to preset machine learning method to the weight of the historical sample collection to be entered, described
The model learning that causality is weighted concentrates each predictive variable and the knot to obtain the historical sample to be entered
Cause and effect regression coefficient between fruit variable；
Identification prediction module, for being identified according to the cause and effect regression coefficient to test data.
7. device according to claim 6, which is characterized in that
The data of the historical sample collection are historical data relevant to the test data.
8. device according to claim 6, which is characterized in that described to be weighted processing generation to the historical sample collection
Historical sample collection to be entered, comprising:
Wherein, X is the matrix for inputting historical sample, and every row indicates that a sample, each column indicate that one-dimensional prediction variable, W are defeated
Enter the weight of historical sample, a is variable X _ j square order, and b is variable X _ k square order, T representing matrix transposition.
9. device according to claim 6, which is characterized in that the predictive variable be it is multiple, the outcome variable be one
It is a.
10. device according to claim 6, which is characterized in that calculate the test number by the cause and effect regression coefficient
According to the predicted value of middle outcome variable, to be identified to the test data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910558189.9A CN110309869A (en) | 2019-06-25 | 2019-06-25 | Stabilization recognition methods and device towards unknown scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910558189.9A CN110309869A (en) | 2019-06-25 | 2019-06-25 | Stabilization recognition methods and device towards unknown scene |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110309869A true CN110309869A (en) | 2019-10-08 |
Family
ID=68077355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910558189.9A Pending CN110309869A (en) | 2019-06-25 | 2019-06-25 | Stabilization recognition methods and device towards unknown scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110309869A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113379436A (en) * | 2020-03-09 | 2021-09-10 | 阿里巴巴集团控股有限公司 | Information processing method, device, computing equipment and medium |
-
2019
- 2019-06-25 CN CN201910558189.9A patent/CN110309869A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113379436A (en) * | 2020-03-09 | 2021-09-10 | 阿里巴巴集团控股有限公司 | Information processing method, device, computing equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105912990B (en) | The method and device of Face datection | |
| KR20180130925A (en) | Artificial intelligent device generating a learning image for machine running and control method thereof | |
| CN114331829A (en) | Countermeasure sample generation method, device, equipment and readable storage medium | |
| CN114139637B (en) | Multi-agent information fusion method and device, electronic equipment and readable storage medium | |
| Chatterjee et al. | Software fault prediction using Nonlinear Autoregressive with eXogenous Inputs (NARX) network | |
| CN111881804B (en) | Posture estimation model training method, system, medium and terminal based on joint training | |
| CN114781272A (en) | Carbon emission prediction method, device, equipment and storage medium | |
| Pandit et al. | Prediction of earthquake magnitude using adaptive neuro fuzzy inference system | |
| KR20190078899A (en) | Visual Question Answering Apparatus Using Hierarchical Visual Feature and Method Thereof | |
| CN107273979A (en) | The method and system of machine learning prediction are performed based on service class | |
| CN109685805A (en) | A kind of image partition method and device | |
| CN111950633A (en) | Neural network training method, neural network target detection method, neural network training device, neural network target detection device and storage medium | |
| CN114997036A (en) | Network topology reconstruction method, device and equipment based on deep learning | |
| CN110309869A (en) | Stabilization recognition methods and device towards unknown scene | |
| CN115620122A (en) | Training method of neural network model, image re-recognition method and related equipment | |
| CN112906586A (en) | Time sequence action nomination generating method and related product | |
| CN112966547A (en) | Neural network-based gas field abnormal behavior recognition early warning method, system, terminal and storage medium | |
| CN111126566B (en) | Abnormal furniture layout data detection method based on GAN model | |
| CN109961160A (en) | A kind of power grid future operation trend predictor method and system based on trend parameter | |
| KR102010031B1 (en) | Method and apparatus for predicting game indicator information | |
| CN103763123A (en) | Method and device for evaluating health condition of network | |
| CN116362894A (en) | Multi-objective learning method, multi-objective learning device, electronic equipment and computer readable storage medium | |
| Mohammadi et al. | Machine learning assisted stochastic unit commitment: A feasibility study | |
| CN111931870B (en) | Model prediction method, model prediction device and system based on model multiplexing | |
| CN114358186A (en) | Data processing method and device and computer readable storage medium |
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: 20191008 |