CN107392318A - Complex machines learning model means of interpretation and device based on local linearization - Google Patents
Complex machines learning model means of interpretation and device based on local linearization Download PDFInfo
- Publication number
- CN107392318A CN107392318A CN201710620391.0A CN201710620391A CN107392318A CN 107392318 A CN107392318 A CN 107392318A CN 201710620391 A CN201710620391 A CN 201710620391A CN 107392318 A CN107392318 A CN 107392318A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- munder
- msubsup
- 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
- 238000010801 machine learning Methods 0.000 claims abstract description 45
- 230000006870 function Effects 0.000 claims abstract description 36
- 238000005457 optimization Methods 0.000 claims abstract description 29
- 238000012417 linear regression Methods 0.000 claims abstract description 14
- 238000005070 sampling Methods 0.000 claims abstract description 11
- 239000012141 concentrate Substances 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 abstract description 16
- 238000003066 decision tree Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 241000235070 Saccharomyces Species 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000003472 neutralizing effect Effects 0.000 description 1
- 238000011020 pilot scale process Methods 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Pure & Applied Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Algebra (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
Abstract
The invention discloses a kind of complex machines learning model means of interpretation and device based on local linearization, wherein, method includes：Collecting sample concentrates any point as sample point, and stochastical sampling obtains multiple sampled points around sample point；In expression of space, Euler's distance between sample point and each sampled point is obtained, using the weight as each sampled point；Machine learning model to be explained is obtained according to the weight of each sampled point and linear model and explains the gap of the fitting result of function, to obtain optimization problem；The linear regression problem of regularization term penalty factor is used in Optimization Solution optimization problem, and obtains explanation results.This method can explain in each data neighborhood of a point to complex machines learning model, take into full account the local characteristicses of sample space, the dominant characteristics of sample space different zones not only can be effectively found, and it is more directly perceived, convenient, it is applicable to the explanation of a variety of machine learning models.
Description
Technical field
The present invention relates to machine learning application and analysis technical field, more particularly to a kind of complexity based on local linearization
Machine learning model means of interpretation and device.
Background technology
At the beginning of machine learning field is started, researchers begin to inquire into the explanation of machine learning algorithm
(Interpretability/Comprehensibility) problem.Here so-called " explanation ", it is containing for machine learning field
Justice, it is desirable to provide the quantitative relationship (qualitative understanding) between input variable and model output.Researcher
It is believed that the precision of model, complexity and interpretation into opposite relation, i.e. the interpretation of naive model is strong, but
It is that precision is relatively low；And complex model can obtain higher computational accuracy, it can be difficult to visual interpretation.
At present, researcher is more likely to first draw the higher model of precision using complex model, then utilizes naive model
Obtained high-precision model is explained, i.e., goes to be fitted the output valve of complex model using naive model.Instructed in correlation technique
Practice explanation of the decision-tree model as neural network model, tree-model complexity is used to indicate the interpretable energy of the interpretation model
Power.(Validity-Interval Analysis) is analyzed to be carried out to neural network model by valid interval in correlation technique
Explain, it is consistent with the general principle of decision-tree model.Correlation technique pilot scale uses first order logic expression (First-order
Logic Formulate) and two kinds of algorithms such as Bayesian network (Bayesian Network) explain matrix decomposition algorithm.Always
From the point of view of knot, the means of interpretation in correlation technique utilizes such as simple mould of linear model, decision-tree model in all input spaces
The characteristics of type is to explain complex model, but can not consider input space part, which is that means of interpretation is maximum in correlation technique, asks
Topic, has much room for improvement.
The content of the invention
It is contemplated that at least solves one of technical problem in correlation technique to a certain extent.
Therefore, it is an object of the present invention to propose a kind of complex machines learning model explanation based on local linearization
Method, this method not only can effectively find the dominant characteristics of sample space different zones, and more directly perceived, convenient, can fit
Explanation for a variety of machine learning models.
It is another object of the present invention to propose that a kind of complex machines learning model based on local linearization is explained to fill
Put.
To reach above-mentioned purpose, one aspect of the present invention embodiment proposes a kind of complex machines based on local linearization
Model explanation method is practised, including：Collecting sample concentrates any point as sample point, and the stochastical sampling around the sample point
Multiple sampled points are obtained, and obtain according to former machine learning model the machine learning mould of each sampled point of the multiple sampled point
Type prediction result；In expression of space, Euler's distance between the sample point and each sampled point is obtained, using as described
The weight of each sampled point；According to the weight of each sampled point and linear model obtain machine learning model to be explained and
The gap of the fitting result of function is explained, and obtains the complexity of the explanation function, to obtain optimization problem；Optimization Solution institute
The linear regression problem that regularization term penalty factor is used in optimization problem is stated, and obtains explanation results.
The complex machines learning model means of interpretation based on local linearization of the embodiment of the present invention, can be in each data
Complex machines learning model is explained in neighborhood of a point, takes into full account the local characteristicses of sample space, not only can be effective
The dominant characteristics of sample space different zones are found, and it is more directly perceived, convenient, it is applicable to the solution of a variety of machine learning models
Release.
In addition, the complex machines learning model means of interpretation according to the above embodiment of the present invention based on local linearization is also
There can be following additional technical characteristic：
Further, in one embodiment of the invention, the weight of each sampled point is：
Wherein, xiFor the sample point,For the multiple sampled point, σ is the standard deviation of all distances.
Further, in one embodiment of the invention, the gap of the fitting result is：
Wherein, f is the machine learning model to be explained, and g is the explanation function, and Γ is the difference of the fitting result
Away from；It is described explain function complexity be：
Wherein, Ω is the complexity.
Further, in one embodiment of the invention, the optimization problem is：
Alternatively, in one embodiment of the invention, by SGD (Saccharomyces Genome Database,
Stochastic gradient descent) linear regression problem described in Algorithm for Solving.
To reach above-mentioned purpose, another aspect of the present invention embodiment proposes a kind of complex machines based on local linearization
Learning model interpreting means, including：Acquisition module, any point is concentrated as sample point for collecting sample, and in the sample
This point surrounding stochastical sampling obtains multiple sampled points, and obtains each of the multiple sampled point according to former machine learning model and adopt
The machine learning model prediction result of sampling point；First acquisition module, in expression of space, obtain the sample point with it is described
Euler's distance between each sampled point, using the weight as each sampled point；Second acquisition module, for according to described every
The weight and linear model of individual sampled point obtain machine learning model to be explained and explain the gap of the fitting result of function, and
The complexity of the explanation function is obtained, to obtain optimization problem；Explanation module, for making in optimization problem described in Optimization Solution
With the linear regression problem of regularization term penalty factor, and obtain explanation results.
The complex machines learning model interpreting means based on local linearization of the embodiment of the present invention, can be in each data
Complex machines learning model is explained in neighborhood of a point, takes into full account the local characteristicses of sample space, not only can be effective
The dominant characteristics of sample space different zones are found, and it is more directly perceived, convenient, it is applicable to the solution of a variety of machine learning models
Release.
In addition, the complex machines learning model interpreting means according to the above embodiment of the present invention based on local linearization are also
There can be following additional technical characteristic：
Further, in one embodiment of the invention, the weight of each sampled point is：
Wherein, xiFor the sample point,For the multiple sampled point, σ is the standard deviation of all distances.
Further, in one embodiment of the invention, in addition to：The gap of the fitting result is：
Wherein, f is the machine learning model to be explained, and g is the explanation function, and Γ is the difference of the fitting result
Away from；It is described explain function complexity be：
Wherein, Ω is the complexity.
Further, in one embodiment of the invention, the optimization problem is：
Alternatively, in one embodiment of the invention, the explanation module passes through linear regression described in SGD Algorithm for Solving
Problem.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partly become from the following description
Obtain substantially, or recognized by the practice of the present invention.
Brief description of the drawings
Of the invention above-mentioned and/or additional aspect and advantage will become from the following description of the accompanying drawings of embodiments
Substantially and it is readily appreciated that, wherein：
Fig. 1 is the complex machines learning model means of interpretation based on local linearization according to one embodiment of the invention
Flow chart；And
Fig. 2 is the complex machines learning model interpreting means based on local linearization according to one embodiment of the invention
Structural representation.
Embodiment
Embodiments of the invention are described below in detail, the example of the embodiment is shown in the drawings, wherein from beginning to end
Same or similar label represents same or similar element or the element with same or like function.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to for explaining the present invention, and is not considered as limiting the invention.
The study mould of the complex machines based on local linearization proposed according to embodiments of the present invention is described with reference to the accompanying drawings
Type means of interpretation and device, the complexity based on local linearization proposed according to embodiments of the present invention is described with reference to the accompanying drawings first
Machine learning model means of interpretation.
Fig. 1 is the flow of the complex machines learning model means of interpretation based on local linearization of one embodiment of the invention
Figure.
Comprise the following steps as shown in figure 1, being somebody's turn to do the complex machines learning model means of interpretation based on local linearization：
In step S101, collecting sample concentrates any point as sample point, and stochastical sampling obtains around sample point
To multiple sampled points, and obtain according to former machine learning model the machine learning model prediction of each sampled point of multiple sampled points
As a result.
It is understood that in an embodiment of the present invention, any point of this concentration is sampled first as sample point xi,
Stochastical sampling obtains the i.e. multiple sampled points of N number of point around the sample point, is designated as respectivelyAnd utilize former machine
Device learning model, calculate the machine learning model prediction result of each sampled pointIt should be noted that the present invention is implemented
The method of example does local explanation to machine learning model in each data vertex neighborhood, can take into full account the different input spaces
The otherness of middle feature, realize that the local linearization of machine learning model is explained.
In step s 102, in expression of space, Euler's distance between sample point and each sampled point is obtained, using as every
The weight of individual sampled point.
Wherein, in one embodiment of the invention, the weight of each sampled point can be：
Wherein, xiFor sample point,For multiple sampled points, σ is the standard deviation of all distances.
It is understood that in expression of space, Euler's distance between sample point and sampled point is calculated, is designated asWherein, σ is the standard deviation of all distances.It should be noted that sampled point is nearer with being explained point, distance is got over
Small, weight is bigger, plays a part of in the study for explaining expression bigger.It should be noted that the embodiment of the present invention uses Europe
Draw the purpose of distance：First, in order to be confined to explain analysis in neighborhood, second, to strengthening the robustness to sampling noiset.
In step s 103, according to the weight of each sampled point and linear model obtain machine learning model to be explained and
The gap of the fitting result of function is explained, and obtains explaining the complexity of function, to obtain optimization problem.
Alternatively, in one embodiment of the invention, the gap of fitting result is：
Wherein, f is machine learning model to be explained, and for g to explain function, Γ is the gap of fitting result；Explain function
Complexity be：
Wherein, Ω is complexity.
Alternatively, in one embodiment of the invention, optimization problem is：
It is understood that the method for the embodiment of the present invention in order in each data vertex neighborhood to machine learning model
Local explanation is done, takes into full account the otherness of feature in the different input spaces, realizes the local linear neutralizing of machine learning model
Release, it is f to assume initially that the machine learning model to be explained, simple to explain that function is g, is originally inputted as x, then to machine learning
Solution to model, which is released, to be equivalent to solve g to meet formula 1：
g(x)≈f(x)。 (1)
In an embodiment of the present invention, the requirement to explaining function g has at 2 points, and one is accuracy (Fidelity), i.e.,
Explain that the model that function g learns wants faithful to master mould f；Another is interpretation, that is, explains that function g form will to the greatest extent can
Can be simple.Requirement of the embodiment of the present invention to explaining function g can be converted into optimization problem as shown in Equation 2：
Min ξ (x)=Γ (f, g)+Ω (g), (2)
Wherein, Γ represents f and g fitting results gap, and it has reacted g accuracy, and the smaller then g of Γ values is closer to f；Ω
G complexity is represented, is represented as decision-tree model with the number of decision tree leaf node, linear model typically uses non-zero
The number of coefficient represents.
It should be noted that, although the parameter that can explain function is trained on all samples, but the table of simple function
Danone power is general, can not meet the requirement of accuracy.For this defect, the method for the embodiment of the present invention proposes local explanation
The thinking of model, i.e., for arbitrary sample point xi, train to obtain in the sample neighborhood of a point and explain function gi, present invention implementation
Example will be converted into master mould f explanation solves a series of local explanation function g=[gi].The method of the embodiment of the present invention uses
Linear model, which is used as, explains function, referred to as LLI (LocalLinear Interpretability, local linearization are explained), such as
Shown in formula 3；
In step S104, the linear regression problem of regularization term penalty factor is used in Optimization Solution optimization problem, and
Obtain explanation results.
It is understood that in an embodiment of the present invention, define Γ (f, g) and Ω (g) such as formula 4 and public affairs respectively first
Shown in formula 5；
Further, formula 2 can be rewritten as：
As shown in Equation 6, left side item is square error sum term, and the right item is L1Regularization term, therefore the present invention is implemented
Local linear can be explained and is converted into using L by example1The linear regression problem of penalty factor, i.e. Lasso problems.
Alternatively, in one embodiment of the invention, SGD Algorithm for Solving linear regression problems are passed through.
It is understood that being directed to above-mentioned Lasso problems, the method for the embodiment of the present invention can be entered using SGD algorithms
Row Optimization Solution.Meanwhile L1Regularization has the effect for making model coefficient rarefaction so that the explanation mould that the embodiment of the present invention obtains
The number of type nonzero coefficient is effective, effectively excludes the interference of useless feature, adds the solution to model property released.
The complex machines learning model means of interpretation based on local linearization proposed according to embodiments of the present invention, Ke Yi
Complex machines learning model is explained in each data neighborhood of a point, takes into full account the local characteristicses of sample space, not only
The dominant characteristics of sample space different zones can be effectively found, and it is more directly perceived, convenient, it is applicable to a variety of machine learning
Solution to model is released.
The complex machines based on local linearization proposed according to embodiments of the present invention referring next to accompanying drawing description learn mould
Type interpreting means.
Fig. 2 is the structure of the complex machines learning model interpreting means based on local linearization of one embodiment of the invention
Schematic diagram.
As shown in Fig. 2 being somebody's turn to do the complex machines learning model interpreting means 10 based on local linearization includes：Acquisition module
100th, the first acquisition module 200, the second acquisition module 300 and explanation module 400.
Wherein, acquisition module 100 concentrates any point as sample point for collecting sample, and in alternatively sample point week
Enclose stochastical sampling and obtain multiple sampled points, and each sampled point of alternatively multiple sampled points is obtained according to former machine learning model
Machine learning model prediction result.First acquisition module 200 be used in expression of space, obtain alternatively sample point with it is optional
Euler's distance between each sampled point in ground, with the weight of optionally each sampled point.Second acquisition module 300 is used for basis
Alternatively the weight of each sampled point and linear model obtain machine learning model to be explained and explain the fitting result of function
Gap, and the complexity of function is alternatively explained, to obtain optimization problem.Explanation module 400 can for Optimization Solution
The linear regression problem of regularization term penalty factor is used in selection of land optimization problem, and obtains explanation results.The embodiment of the present invention
Device 10 can explain in each data neighborhood of a point to complex machines learning model, take into full account the office of sample space
Portion's characteristic, the dominant characteristics of sample space different zones not only can be effectively found, and it is more directly perceived, convenient, it is applicable to
The explanation of a variety of machine learning models.
Alternatively, in one embodiment of the invention, the weight of each sampled point is：
Wherein, xiFor sample point,For multiple sampled points, σ is the standard deviation of all distances.
Alternatively, in one embodiment of the invention, the gap of fitting result is：
Wherein, f is machine learning model to be explained, and for g to explain function, Γ is the gap of fitting result；Explain function
Complexity be：
Wherein, Ω is complexity.
Alternatively, in one embodiment of the invention, optimization problem is：
Alternatively, in one embodiment of the invention, explanation module passes through SGD Algorithm for Solving linear regression problems.
It should be noted that the foregoing solution to the complex machines learning model means of interpretation embodiment based on local linearization
The complex machines learning model interpreting means based on local linearization that explanation is also applied for the embodiment are released, it is no longer superfluous herein
State.
The complex machines learning model interpreting means based on local linearization proposed according to embodiments of the present invention, the device
Complex machines learning model can be explained in each data neighborhood of a point, take into full account the local special of sample space
Property, the device not only can effectively find the dominant characteristics of sample space different zones, and more directly perceived, convenient, applicable
In the explanation of a variety of machine learning models.
In the description of the invention, it is to be understood that term " " center ", " longitudinal direction ", " transverse direction ", " length ", " width ",
" thickness ", " on ", " under ", "front", "rear", "left", "right", " vertical ", " level ", " top ", " bottom " " interior ", " outer ", " up time
The orientation or position relationship of the instruction such as pin ", " counterclockwise ", " axial direction ", " radial direction ", " circumference " be based on orientation shown in the drawings or
Position relationship, it is for only for ease of and describes the present invention and simplify description, rather than indicates or imply that signified device or element must
There must be specific orientation, with specific azimuth configuration and operation, therefore be not considered as limiting the invention.
In addition, term " first ", " second " are only used for describing purpose, and it is not intended that instruction or hint relative importance
Or the implicit quantity for indicating indicated technical characteristic.Thus, define " first ", the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the invention, " multiple " are meant that at least two, such as two, three
It is individual etc., unless otherwise specifically defined.
In the present invention, unless otherwise clearly defined and limited, term " installation ", " connected ", " connection ", " fixation " etc.
Term should be interpreted broadly, for example, it may be fixedly connected or be detachably connected, or integrally；Can be that machinery connects
Connect or electrically connect；Can be joined directly together, can also be indirectly connected by intermediary, can be in two elements
The connection in portion or the interaction relationship of two elements, limited unless otherwise clear and definite.For one of ordinary skill in the art
For, the concrete meaning of above-mentioned term in the present invention can be understood as the case may be.
In the present invention, unless otherwise clearly defined and limited, fisrt feature can be with "above" or "below" second feature
It is that the first and second features directly contact, or the first and second features pass through intermediary mediate contact.Moreover, fisrt feature exists
Second feature " on ", " top " and " above " but fisrt feature are directly over second feature or oblique upper, or be merely representative of
Fisrt feature level height is higher than second feature.Fisrt feature second feature " under ", " lower section " and " below " can be
One feature is immediately below second feature or obliquely downward, or is merely representative of fisrt feature level height and is less than second feature.
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 the spy for combining the embodiment or example description
Point is contained at least one embodiment or example of the present invention.In this manual, to the schematic representation of above-mentioned term not
Identical embodiment or example must be directed to.Moreover, specific features, structure, material or the feature of description can be with office
Combined in an appropriate manner in one or more embodiments or example.In addition, in the case of not conflicting, the skill of this area
Art personnel can be tied the different embodiments or example and the feature of different embodiments or example described in this specification
Close and combine.
Although embodiments of the invention have been shown and described above, it is to be understood that above-described embodiment is example
Property, it is impossible to limitation of the present invention is interpreted as, one of ordinary skill in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, changed, replacing and modification.
Claims (10)
1. a kind of complex machines learning model means of interpretation based on local linearization, it is characterised in that comprise the following steps：
Collecting sample concentrates any point as sample point, and stochastical sampling obtains multiple sampled points around the sample point,
And the machine learning model prediction result of each sampled point of the multiple sampled point is obtained according to former machine learning model；
In expression of space, Euler's distance between the sample point and each sampled point is obtained, each to be adopted as described
The weight of sampling point；
Machine learning model to be explained is obtained according to the weight of each sampled point and linear model and explains the plan of function
The gap of result is closed, and obtains the complexity of the explanation function, to obtain optimization problem；And
The linear regression problem of regularization term penalty factor is used in optimization problem described in Optimization Solution, and obtains explanation results.
2. the complex machines learning model means of interpretation according to claim 1 based on local linearization, it is characterised in that
The weight of each sampled point is：
<mrow>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>=</mo>
<mi>exp</mi>
<mrow>
<mo>(</mo>
<mo>-</mo>
<mfrac>
<mrow>
<mi>d</mi>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
<msup>
<mi>&sigma;</mi>
<mn>2</mn>
</msup>
</mfrac>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Wherein, xiFor the sample point,For the multiple sampled point, σ is the standard deviation of all distances.
3. the complex machines learning model means of interpretation according to claim 2 based on local linearization, it is characterised in that
The gap of the fitting result is：
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>&Gamma;</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>g</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mi>&Sigma;</mi>
<mi>k</mi>
</munder>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>-</mo>
<mi>g</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<munder>
<mi>&Sigma;</mi>
<mi>k</mi>
</munder>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>-</mo>
<munder>
<mi>&Sigma;</mi>
<mi>j</mi>
</munder>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<msubsup>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>,</mo>
</mrow>
Wherein, f is the machine learning model to be explained, and g is the explanation function, and Γ is the gap of the fitting result；
It is described explain function complexity be：
<mrow>
<mi>&Omega;</mi>
<mrow>
<mo>(</mo>
<mi>g</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mi>j</mi>
</munder>
<mo>|</mo>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
<mo>,</mo>
</mrow>
Wherein, Ω is the complexity.
4. the complex machines learning model means of interpretation according to claim 3 based on local linearization, it is characterised in that
The optimization problem is：
<mrow>
<mtable>
<mtr>
<mtd>
<mi>min</mi>
</mtd>
<mtd>
<mrow>
<munder>
<mi>&xi;</mi>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
</munder>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mi>k</mi>
</munder>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>-</mo>
<munder>
<mo>&Sigma;</mo>
<mi>j</mi>
</munder>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<msubsup>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>&CenterDot;</mo>
<munder>
<mo>&Sigma;</mo>
<mi>j</mi>
</munder>
<mo>|</mo>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>.</mo>
</mrow>
5. the complex machines learning model means of interpretation based on local linearization according to claim any one of 1-4, lead to
Cross linear regression problem described in stochastic gradient descent SGD Algorithm for Solving.
A kind of 6. complex machines learning model interpreting means based on local linearization, it is characterised in that including：
Acquisition module, any point is concentrated as sample point for collecting sample, and stochastical sampling obtains around the sample point
To multiple sampled points, and obtain according to former machine learning model the machine learning model of each sampled point of the multiple sampled point
Prediction result；
First acquisition module, in expression of space, obtaining Euler's distance between the sample point and each sampled point,
Using the weight as each sampled point；
Second acquisition module, machine learning mould to be explained is obtained for the weight according to each sampled point and linear model
The gap of the fitting result of type and explanation function, and the complexity of the explanation function is obtained, to obtain optimization problem；And
Explanation module, for using the linear regression problem of regularization term penalty factor in optimization problem described in Optimization Solution, and
Obtain explanation results.
7. the complex machines learning model interpreting means according to claim 6 based on local linearization, it is characterised in that
The weight of each sampled point is：
<mrow>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>=</mo>
<mi>exp</mi>
<mrow>
<mo>(</mo>
<mo>-</mo>
<mfrac>
<mrow>
<mi>d</mi>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
<msup>
<mi>&sigma;</mi>
<mn>2</mn>
</msup>
</mfrac>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Wherein, xiFor the sample point,For the multiple sampled point, σ is the standard deviation of all distances.
8. the complex machines learning model interpreting means according to claim 7 based on local linearization, it is characterised in that
The gap of the fitting result is：
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>&Gamma;</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>g</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mi>&Sigma;</mi>
<mi>k</mi>
</munder>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>-</mo>
<mi>g</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<munder>
<mi>&Sigma;</mi>
<mi>k</mi>
</munder>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>-</mo>
<munder>
<mi>&Sigma;</mi>
<mi>j</mi>
</munder>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<msubsup>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>,</mo>
</mrow>
Wherein, f is the machine learning model to be explained, and g is the explanation function, and Γ is the gap of the fitting result；
It is described explain function complexity be：
<mrow>
<mi>&Omega;</mi>
<mrow>
<mo>(</mo>
<mi>g</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mi>j</mi>
</munder>
<mo>|</mo>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
<mo>,</mo>
</mrow>
Wherein, Ω is the complexity.
9. the complex machines learning model interpreting means according to claim 8 based on local linearization, it is characterised in that
The optimization problem is：
<mrow>
<mtable>
<mtr>
<mtd>
<mi>min</mi>
</mtd>
<mtd>
<mrow>
<munder>
<mi>&xi;</mi>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
</munder>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mi>k</mi>
</munder>
<msub>
<mi>&Psi;</mi>
<mi>k</mi>
</msub>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
<mo>-</mo>
<munder>
<mo>&Sigma;</mo>
<mi>j</mi>
</munder>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<msubsup>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>&CenterDot;</mo>
<munder>
<mo>&Sigma;</mo>
<mi>j</mi>
</munder>
<mo>|</mo>
<msub>
<mi>&beta;</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>.</mo>
</mrow>
10. the complex machines learning model interpreting means based on local linearization according to claim any one of 6-9, institute
State explanation module and pass through linear regression problem described in SGD Algorithm for Solving.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710620391.0A CN107392318A (en) | 2017-07-26 | 2017-07-26 | Complex machines learning model means of interpretation and device based on local linearization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710620391.0A CN107392318A (en) | 2017-07-26 | 2017-07-26 | Complex machines learning model means of interpretation and device based on local linearization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107392318A true CN107392318A (en) | 2017-11-24 |
Family
ID=60341057
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710620391.0A Pending CN107392318A (en) | 2017-07-26 | 2017-07-26 | Complex machines learning model means of interpretation and device based on local linearization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107392318A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902833A (en) * | 2018-12-05 | 2019-06-18 | 阿里巴巴集团控股有限公司 | Machine learning model means of interpretation and device |
| CN111262887A (en) * | 2020-04-26 | 2020-06-09 | 腾讯科技（深圳）有限公司 | Network risk detection method, device, equipment and medium based on object characteristics |
| CN111639688A (en) * | 2020-05-19 | 2020-09-08 | 华中科技大学 | Local interpretation method of Internet of things intelligent model based on linear kernel SVM |
| CN111754287A (en) * | 2020-02-14 | 2020-10-09 | 北京沃东天骏信息技术有限公司 | Article screening method, apparatus, device and storage medium |
| US11651276B2 (en) | 2019-10-31 | 2023-05-16 | International Business Machines Corporation | Artificial intelligence transparency |
-
2017
- 2017-07-26 CN CN201710620391.0A patent/CN107392318A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902833A (en) * | 2018-12-05 | 2019-06-18 | 阿里巴巴集团控股有限公司 | Machine learning model means of interpretation and device |
| CN109902833B (en) * | 2018-12-05 | 2023-06-27 | 创新先进技术有限公司 | Machine learning model interpretation method and device |
| US11651276B2 (en) | 2019-10-31 | 2023-05-16 | International Business Machines Corporation | Artificial intelligence transparency |
| CN111754287A (en) * | 2020-02-14 | 2020-10-09 | 北京沃东天骏信息技术有限公司 | Article screening method, apparatus, device and storage medium |
| CN111262887A (en) * | 2020-04-26 | 2020-06-09 | 腾讯科技（深圳）有限公司 | Network risk detection method, device, equipment and medium based on object characteristics |
| CN111262887B (en) * | 2020-04-26 | 2020-08-28 | 腾讯科技（深圳）有限公司 | Network risk detection method, device, equipment and medium based on object characteristics |
| CN111639688A (en) * | 2020-05-19 | 2020-09-08 | 华中科技大学 | Local interpretation method of Internet of things intelligent model based on linear kernel SVM |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107392318A (en) | Complex machines learning model means of interpretation and device based on local linearization | |
| DE102005058081B9 (en) | Method for the reconstruction of gusts and structural loads in aircraft, in particular commercial aircraft | |
| DE69321952T2 (en) | SYSTEM AND METHOD FOR CONTROLLING A PLANT | |
| DE102009002392A1 (en) | System and method for determining local accelerations, dynamic load distributions and aerodynamic data in an aircraft | |
| van Gent et al. | Knowledge architecture supporting collaborative MDO in the AGILE paradigm | |
| CN105931153A (en) | Indirect questionnaire assessment method based on neural network prediction analysis model | |
| DE112020004471T5 (en) | Inference device, training device, inference method and training method | |
| CN105808311A (en) | Reservoir simulation fast matching method based on dimension reduction strategy | |
| CN102736616B (en) | Dulmage-Mendelsohn (DM)-decomposition-based measuring point optimal configuration method for closed loop system | |
| CN107481218A (en) | Image aesthetic feeling appraisal procedure and device | |
| Wold | Chemometrics, why, what and where to next? | |
| CN115577436A (en) | Combined deep learning method for solving wind-induced vibration response of uncertain structure | |
| Görtz et al. | Collaborative multi-level MDO process development and application to long-range transport aircraft | |
| Karlsen et al. | An error estimate for the finite difference approximation to degenerate convection–diffusion equations | |
| US20240070222A1 (en) | Nonlinear optimization method for parameters of ocean ecological dynamics model | |
| EP1327959B1 (en) | Neural network for modelling a physical system and method for building the neural network | |
| Tsai et al. | Characteristics method using cubic–spline interpolation for advection–diffusion equation | |
| CN106203618A (en) | A kind of method of the neutral net building band border constraint | |
| Jategaonkar et al. | ESTIMA–an integrated software tool for nonlinear parameter estimation | |
| Wujek et al. | Design flow management and multidisciplinary design optimization in application to aircraft concept sizing | |
| DE102015120095A1 (en) | Analyzer and method | |
| Jiang et al. | Analytical target cascading using ensemble of surrogates for engineering design problems | |
| CN104102756B (en) | A kind of hereditary parallel optimization method of rotary type car sealing system for vehicle door parameter | |
| WO2010055003A2 (en) | Method and device for monitoring the combustion of a power plant by means of a real concentration distribution | |
| CN106101035A (en) | The signal blind checking method of unity-feedback neutral network |
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 |
Application publication date: 20171124 |
|
| RJ01 | Rejection of invention patent application after publication |