CN106156077A - The method and apparatus selected for mixed model - Google Patents
The method and apparatus selected for mixed model Download PDFInfo
- Publication number
- CN106156077A CN106156077A CN201510149617.4A CN201510149617A CN106156077A CN 106156077 A CN106156077 A CN 106156077A CN 201510149617 A CN201510149617 A CN 201510149617A CN 106156077 A CN106156077 A CN 106156077A
- Authority
- CN
- China
- Prior art keywords
- hidden variable
- candidate family
- candidate
- sample group
- unit
- 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
Abstract
Embodiments of the invention relate to the method and apparatus that mixed model selects.The method includes: initializes hidden variable by training multiple first and generates multiple candidate family；The second initialization hidden variable is determined based on multiple candidate families；And determine object module based on the second initialization hidden variable.Embodiments of the invention can perform mixed model quickly and efficiently and select, and have good versatility.
Description
Technical field
Embodiments of the invention relate generally to machine learning field, and more particularly, to
The method and apparatus selected for mixed model.
Background technology
Mixed model (mixture model) is a kind of to use mixed distribution for density Estimation
Probabilistic model, it can represent in a Ge great colony and there is sub-group.Mixed model can include
Multiple model, such as gauss hybrid models, piecewise linearity mixed model etc., these models by
It is widely used in multiple fields, such as document classification, handwriting recognition, broad image segmentation etc..
In practice, the Model Selection of mixed model be one extremely important and challenging
Problem.Industry has been proposed for certain methods to carry out the Model Selection of mixed model, wherein
Variation reasoning (variational inference) is a kind of relatively effective method, its attempt to
Go out the analytic approximation of limit log-likelihood.But, variation reasoning is more sensitive for initializing,
If initializing inappropriate, then the effect of variation reasoning may become very poor, thus cannot be accurate
True ground preference pattern.Therefore, initialize to have become as and affect the precision of variation reasoning and efficiency
Key factor.
At present, the initial method of variation reasoning such as can include random initializtion, based on poly-
Initialization of class etc..But, random initializtion typically requires the substantial amounts of initialization sample of trial
Realize, process time-consumingly the longest, thus the speed of Model Selection can be caused slower.It addition, base
Initializing for the mixed model (such as, gauss hybrid models) with cluster as target in cluster
Be probably relatively effective, but for for returning/be categorized as the mixed model of target not
It is suitable for using, so initializing universal based on cluster is relatively low.
Accordingly, it would be desirable to the fast and more generally applicable initialization scheme of a kind of processing speed realizes mixing
Close Model Selection.
Summary of the invention
Generally, embodiments of the invention propose a kind of technical scheme selected for mixed model.
In one aspect of the invention, it is provided that a kind of method selected for mixed model.Described
Method includes: initializes hidden variable by training multiple first and generates multiple candidate family；Base
The second initialization hidden variable is determined in multiple candidate families；And initialize hidden variable based on second
Determine object module.
In another aspect of this invention, it is provided that a kind of device selected for mixed model.Described
Device includes: candidate family signal generating unit, is configured to train multiple first initialization hidden
Variable generates multiple candidate family；First determines unit, is configured to based on multiple candidate's moulds
Type determines the second initialization hidden variable；And second determine unit, it is configured to based at the beginning of second
Beginningization hidden variable determines object module.
According to embodiments of the invention, new by utilizing the training result of candidate family to create
Initializing hidden variable, then obtain new model based on new initialization hidden variable, this is not
It is confined to certain specific blend types of models.Therefore, have according to the solution of the present invention good
Versatility.Meanwhile, according to embodiments of the invention, obtain based on local optimum model and more may be used
The new initialization hidden variable leaned on, thus obtain object module based on new initialization hidden variable.
Thus obtain the time of object module relative to utilizing random initializtion hidden variable to directly obtain mesh
The time of mark model can significantly reduce, thus has higher process according to the solution of the present invention
Speed.Other features and advantages of the present invention will be easy to understand by being described below.
Accompanying drawing explanation
By combining accompanying drawing, exemplary embodiment of the invention is described in more detail, this
Bright above-mentioned and other purpose, feature and advantage will be apparent from. wherein:
Fig. 1 shows the signal of the method selected for mixed model according to embodiments of the present invention
Property flow chart；
Fig. 2 shows the signal of the method selected for mixed model according to embodiments of the present invention
Property flow chart；
Fig. 3 shows the side for determining the second initialization hidden variable according to embodiments of the present invention
The indicative flowchart of method；
Fig. 4 shows the schematic diagram of a graph-based according to embodiments of the present invention；
Fig. 5 shows the signal of the device selected for mixed model according to embodiments of the present invention
Property block diagram；And
Fig. 6 shows showing of the exemplary computer system be suitable to for realizing the embodiment of the present invention
Meaning property block diagram.
In the accompanying drawings, same or analogous label is used to represent same or analogous element.
Detailed description of the invention
It is more fully described the preferred implementation of the disclosure below with reference to accompanying drawings.Although accompanying drawing
In show the preferred implementation of the disclosure, however, it is to be appreciated that can be real in a variety of manners
Show the disclosure and should not limited by embodiments set forth herein.On the contrary, it is provided that these are implemented
Mode is to make the disclosure more thorough and complete, and can be complete by the scope of the present disclosure
Convey to those skilled in the art.
The mechanism of the embodiment of the present invention described in detail below and principle.Unless specifically stated otherwise,
Hereafter represent " being based at least partially on " with the term "based" of use in claim.Term " bag
Include " represent that opening includes, i.e. " include but not limited to ".Term " multiple " expression " two or more
Many ".Term " embodiment " expression " at least one embodiment ".Term " another embodiment "
Represent " at least one further embodiment ".The definition of other terms is given in will be described below.
Fig. 1 shows the method 100 for mixed model selection according to embodiments of the present invention
Flow chart.Traditionally, the initial method of variation reasoning uses random initializtion, based on cluster
The method such as initialization.But, it is relatively slow, suitable that these initial methods are generally of processing speed
By problems such as property are the most extensive, these are all the problem demanding prompt solutions during mixed model selects.Logical
Crossing and be described below it will be appreciated that utilize method 100 according to an embodiment of the invention, these are asked
Topic can be obtained and efficiently solve.
Method 100 starts from step S110, initializes hidden variable at this by training multiple first
Generate multiple candidate family.
In embodiments of the present invention, " hidden variable " can represent and can not be directly observed and need
The variable that sample data to be passed through is derived from.The variation distribution of hidden variable may be used for describing sample
Notebook data is clustered the probability of corresponding classification.It should be noted that, in embodiments of the present invention,
" hidden variable " is not limited to a kind of variable, but can comprise " the variation distribution of hidden variable " and/
Or other suitable information.In the disclosure, hidden variable can include initializing hidden variable, more
New hidden variable, etc., wherein initialize hidden variable and represent the hidden variable for being trained,
And the hidden variable updated obtains hidden variable after representing training.
In embodiments of the present invention, " model " can include candidate family, mid-module and mesh
Mark model, etc..Model can be by being trained generating to initialization hidden variable.So
The model generated can include the variation distribution of hidden variable of renewal, model parameter, model structure,
Etc..Model parameter can be different, due to mixed model according to the type difference of mixed model
It is typically the general name of model I, and a mixed model can be combined by multiple submodels,
Therefore model parameter is associated with the type of concrete mixed model.For example, for Gauss
For mixed model, model parameter can include the equal of the Gauss distribution that each submodel obeyed
Value and variance.For up model, model parameter then can include the bar of a node
Part controls parameter and the regression coefficient of leaf node and deviation.Model structure also with mixed model
Type is associated.For example, for gauss hybrid models, model structure can include
The number of submodel and submodel merge coefficient, etc..For up model, mould
Type structure then can include the tree construction learning.Should be appreciated that above-mentioned example merely for the sake of
The purpose of discussion, it is not intended to limit the scope of the present invention by any way.
In one embodiment, in step S110, can determine multiple based on training sample set
First initializes hidden variable.For example, it is possible to the sample concentrating training sample carries out random packet
Obtain multiple first and initialize hidden variable, or the sample that training sample is concentrated can be carried out
Cluster obtains multiple first and initializes hidden variable.First initializes hidden variable such as can realize
For the form of matrix, the form of data acquisition system or other any suitable forms.A reality
Executing in example, first initializes the matrix that hidden variable can be k × n dimension, and wherein k is this
The line number of matrix, represents the sample group number of the first initialization hidden variable；N is this matrix column
Number, represents the number of samples that training sample is concentrated.A line in this matrix corresponds to a sample
Group, often the value of each element in row can be 0 or 1.Such as, if in this matrix
I-th element in a line is 1, then it represents that comprise training in the sample group corresponding with this row
I-th sample in sample set；If the jth element in this row is 1, then it represents that with this
The sample group that row is corresponding does not comprise the jth sample that training sample is concentrated.Should be appreciated that
The value of the element in above-mentioned matrix need not be 0 or 1, it is also possible to is that any other is fitted
When numerical value.Above example is only used to convenient purpose is discussed, it is not intended to limit the present invention's
Scope.In another embodiment, the first initialization hidden variable can be a data acquisition system,
Such as can include k sample group, each sample group has one or more sample.Should manage
Solving, above-mentioned example is only used to the purpose of discussion, is not intended to limit the scope of the present invention.This
Skilled person can by any suitable by the way of realize the first initialization hidden variable.
It is then possible to initialize each first hidden change of initialization in hidden variable for multiple first
Amount, learns on training sample set, relative with each first initialization hidden variable to generate
The candidate family answered.For example, initialize hidden variable for each first, can be in training
Exercise supervision on sample set or unsupervised learning, automatically study mixed model model parameter,
The hidden variable variation distribution updated and model structure, thus obtain the candidate family of correspondence.
It follows that method 100 proceeds to step S120, determine based on multiple candidate families at this
Second initializes hidden variable.
Second initialization hidden variable is according to embodiments of the invention, and second initializes hidden variable can
To be determined based on multiple candidate families in several ways.In one embodiment, may be used
To match multiple candidate families, and candidate family based on pairing determines the second initialization
Hidden variable.Such as, when candidate family number is 2, and when being paired into a pair, can be based on
This determines that to candidate family one second initializes hidden variable.When candidate family is paired into many
Pair time, can based on every pair of candidate family therein determine one second initialize hidden variable,
Thus obtain multiple second and initialize hidden variable.
Alternatively, in one embodiment, can performances based on multiple candidate families from multiple
Candidate family selects at least two candidate family.The performance of model can include many factors,
Such as precision, ageing, etc..The performance of model can determine in several ways, example
As calculated root-mean-square error, average absolute value error, likelihood ratio etc..As a example by precision, for
For prediction/classification mixed model, candidate family can be used for testing the prediction/classification of data,
Thus obtain the prediction/nicety of grading of correspondence.For Clustering Model, candidate's mould can be calculated
The clustering precision of type, such as standard mutual information etc..It is determined by the performance of model, can be from many
Individual candidate family selects performance preferable at least two candidate family.It is then possible to selected
At least two candidate family selected matches, it is possible to candidate family based on pairing determines the
Two initialize hidden variable.In an additional embodiment according to the present invention, every pair of candidate family
The first candidate family and the second candidate family can be included, now can be based on every pair of candidate family
In the renewal hidden variable of the first candidate family and the renewal hidden variable of the second candidate family, come really
Fixed one second initializes hidden variable.In this way, can be based on one or more pairs of candidates
Model determines that one or more second initializes hidden variable.
According to embodiments of the invention, the second initialization hidden variable can be determined iteratively, thus
Obtain more excellent result.In one embodiment, the following one or many that operates can be performed:
Initialize hidden variable by training multiple second and obtain multiple mid-module；Based on multiple centres
Model determines the 3rd initialization hidden variable；And utilize the 3rd initialization hidden variable to update at the beginning of second
Beginningization hidden variable.In this embodiment, the 3rd hidden change of initialization is determined based on multiple mid-modules
The method of amount can use and determine that second is initial based on multiple candidate families in step S120
Change the method that hidden variable is similar.In this way, it is possible to determine the second initialization hidden variable iteratively,
Until reach iteration stopping condition, such as, reach predetermined iterations, to reach the second initialization hidden
The predetermined accuracy of variable or other conditions of similarities.
Referring still to Fig. 1, method 100 proceeds to step S130, initializes based on second at this
Hidden variable determines object module.
According to embodiments of the invention, the second initialization hidden variable determined in step S120 can
One or more to have.In one embodiment, the second initialization determined by step S120
The number of hidden variable is 1, then can be by training this hidden change of the second initialization in step S130
Amount obtains a model, and using this model as object module.
In another embodiment, determine that multiple second initializes hidden variable in step S120,
The most in step s 130, hidden variable can be initialized by training the plurality of second and obtain many
Individual mid-module, then may determine that the performance of multiple mid-module, and based on determined by property
One of multiple mid-module can be selected as object module.
It is appreciated by the above description that embodiments of the invention are not related to certain types of mixing
The concrete property of model or requirement, therefore have good versatility, it is not limited to certain kinds
The mixed model of type.It addition, embodiments of the invention can determine local optimum by performance
Model, and further determine that more reliable initialization hidden variable from this model, this can improve
Obtain the processing speed of preferable object module, be therefore a kind of efficient solution.Below
The further advantage of embodiments of the invention will be discussed by the embodiment shown in Fig. 2.
Fig. 2 shows the method 200 for mixed model selection according to embodiments of the present invention
Indicative flowchart.Method 200 is considered the specific embodiment of method 100.In side
In method 200 during determining the second initialization hidden variable from multiple candidate families, first from time
Modeling type selects the candidate family of some better performances, then to these selected candidate's moulds
Type carries out matching to determine the second initialization hidden variable.Should be appreciated that shown embodiment is only
Purpose for discussion, it is not intended to limit the scope of the present invention by any way.
Method 200 starts from step S210, determines multiple first at this based on training sample set
Initialize hidden variable.
In the embodiment shown in Figure 2, training sample set is one and includes multiple sample (also referred to as
For sample data) set.Each sample can be characterized by multiple parameters.One sample
Can be expressed as that (x, y), wherein x represents the feature of this sample, and y represents estimating for this sample
Evaluation.Such as, the x of the sample of power consumption data can include power consumption today, temperature,
Humidity etc., y then can represent estimated power consumption tomorrow, i.e.
X=[ele-consumption_today, temperature, humidity ...],
Y=ele-consumption_tomorrow.In the following discussion, mentioned " sample group "
Represent one group of sample, this sample group potentially includes one or more sample.
According to embodiments of the invention, can be randomly generated at the beginning of multiple first from training sample set
Beginningization hidden variable.Alternately, it is also possible to training sample concentrate sample cluster, from
And generate multiple first and initialize hidden variable.In addition, it is also possible to use people in the art
Member can additive method determine the first initialization hidden variable based on training sample set, above-mentioned show
Example embodiment is not limitation of the present invention.
In step S220, initialize each first in hidden variable for multiple first and initialize hidden
Variable, learns on training sample set, initializes hidden variable phase to generate with each first
Corresponding candidate family.
In one embodiment, initialize hidden variable for each first, can be at training sample
Exercise supervision on collection or unsupervised learning, automatically the study model parameter of mixed model, renewal
Hidden variable variation distribution and model structure, thus obtain correspondence candidate family.By this
The mode of kind, can obtain initializing, with multiple first, multiple candidate families that hidden variables are corresponding.
It follows that method 200 proceeds to step S230, property based on multiple candidate families at this
At least two candidate family can be selected from multiple candidate families.
The performance of model can include precision, ageing etc..As a example by precision, for prediction/
For classification mixed model, candidate family can be used more in the prediction/classification of test data
New hidden variable variation distribution, model parameter and model structure, thus obtain the prediction of correspondence/
Nicety of grading.For Clustering Model, the clustering precision of candidate family can be calculated, such as
Standard mutual information etc..Should be appreciated that in addition to precision, it is also possible to be determined by model time
Effect property etc. are about the performance because usually obtaining model.Above-mentioned example is intended merely to facilitate in this
Discuss, be not limitation of the present invention.
Selection for candidate family can comprise this bigger collection of multiple candidate family
Close and be reduced into a less set, thus performance preferably some candidate families are selected for
Further determine that initialization hidden variable.In one embodiment, can be by based on multiple candidates
The hidden variable variation distribution of the renewal of each candidate family in model, model parameter and model knot
Structure, determines the performance of each candidate family.It is then possible to performance determined by based on is to many
Individual candidate family is ranked up, and such as, arranges the plurality of time according to performance order from high to low
Modeling type.Thus, it is possible to select at least two candidate's mould from multiple candidate families based on sequence
Type.For example, it is possible to two or more candidate families that selected and sorted is forward, thus realize
Performance preferably some candidate families are selected from multiple candidate families.
In step S240, at least two candidate family is matched.
Pairing process can realize in several ways.In one embodiment, can will be somebody's turn to do
Each two model at least two candidate family matches, such as, when this at least two is waited
When the number of modeling type is 4, then can obtain 6 pairs of candidate families by permutation and combination computing.
In another kind of alternative, can be by any two mould in this at least two candidate family
Type matches, and such as, when the number of this at least two candidate family is 4, then can obtain
To 1 pair of candidate family, 2 pairs of candidate families ... or 6 pairs of candidate families.
Should be appreciated that in addition to above-mentioned matching method, those skilled in the art can use and appoint
At least two candidate family is matched by what his suitable matching method.Above-mentioned example is only
It is for only for ease of discussion in this, is not limitation of the present invention.
In step S250, candidate family based on pairing determines the second initialization hidden variable.
Second to initialize hidden variable be also a kind of hidden variable that initializes, but its to be different from first initial
Change hidden variable.According to embodiments of the invention, every couple of candidate in the candidate family of pairing
Model can include the first candidate family and the second candidate family, can be based on every pair of candidate family
In the renewal hidden variable of the first candidate family and the renewal hidden variable of the second candidate family, come really
Fixed one second initializes hidden variable.In this way, one can be determined in step S250
Individual or multiple second initialization hidden variables, this second number initializing hidden variable matches with passing through
The number of pairs of candidate family relevant.
According to embodiments of the invention, the second hidden change of initialization can be determined in several ways
Amount.In one embodiment, the second initialization hidden variable can be by performing following operation once
Or repeatedly determine: the renewal hidden variable of the first candidate family from every pair of candidate family is selected
Select a sample group as the first sample group；Based on the first sample group, from every pair of candidate family
The second candidate family renewal hidden variable in sample group determine the second sample group；Determine first
Sample group and the common factor of the second sample group；Based on this common factor, the first sample components is segmented into two sons
Collection；And build the second initialization hidden variable based on these two subsets.The execution of aforesaid operations time
Number can be predefined, it is also possible to is according to the second structure performance initializing hidden variable
Determine.Fig. 3 show according to embodiments of the present invention for determining the second hidden change of initialization
The indicative flowchart of the method 300 of amount.
In step S310, select the renewal hidden variable of the first candidate family in every pair of candidate family
In a sample group as the first sample group.
According to one embodiment of present invention, the renewal hidden variable of the first candidate family is such as
G1={I can be expressed as1,…,Ig1, wherein I represents the renewal hidden variable of the first candidate family
In sample group, represent one group of sample；G1 represents in the renewal hidden variable of the first candidate family
The number of sample group.Similarly, the renewal hidden variable of the second candidate family such as can be expressed as
G2={I '1,…,I’g2, the sample group during wherein I ' represents the renewal hidden variable of the second candidate family,
G2 represents the number of the sample group in the renewal hidden variable of the second candidate family.For example, exist
Step S310, can select sample group (such as a, I in G11) as the first sample
Group.
In step S320, based on the first sample group, second candidate's mould from every pair of candidate family
Sample group in the renewal hidden variable of type determines the second sample group.
According to embodiments of the invention, the second sample group can determine in several ways.?
In one embodiment, the first sample group and the second candidate family can be updated in hidden variable
Each sample group seeks common ground, and then the sample group occuring simultaneously corresponding with maximum can be defined as
Two sample groups.
Optionally, can be by the renewal hidden variable of the second candidate family and the
One sample group has the sample group of common factor to be defined as the second sample group, and without searching maximum of occuring simultaneously
Sample group is used as the second sample group.Should be appreciated that embodiment shown above is only schematically
, it is not intended to limit the scope of the present invention by any way.
In step S330, determine the common factor of the first sample group and the second sample group.
In the case of the second sample group is corresponding to occuring simultaneously with the maximum of the first sample, the first sample
Group (is expressed as I with the common factor of the second sample groupnew) can be identified below:
Wherein IiRepresent the first sample group, and IiIt is in the renewal hidden variable of the first candidate family
I sample group.IjRepresent the jth sample group in the renewal hidden variable of the second candidate family.G2
Represent the renewal hidden variable of the second candidate family.
In step S340, based on this common factor, the first sample components is segmented into two subsets.
In one embodiment, the first subset of the first sample group can be defined as this common factor,
And part in addition to this common factor in the first sample group is defined as the second subset.Such as, first
Sample group IiI can be divided intonewAnd Ii-InewThe two subset.
In step S350, build the second initialization hidden variable based on two subsets.
According to embodiments of the invention, second initializes hidden variable can include multiple sample group.
In step S350, two subsets obtained can will be cut by the first sample components as
The two sample groups initializing hidden variable, to build the second initialization hidden variable.
For example, second initializes hidden variable can be initialized as empty set, namely does not include appointing
What sample group, such asWherein S represents the second initialization hidden variable.In step S350
In can initialize hidden as second using being cut two subsets obtaining by the first sample components
The sample group of variable, it is hereby achieved that S=S ∪ { Inew}∪{Ii–Inew}。
In step S360, from the renewal hidden variable of the first candidate family, remove the first sample group,
To be not repeated in subsequent treatment to use this first sample group.
In step S370, it is judged that second initializes whether hidden variable structure completes.
In one embodiment, the sample group in hidden variable can be initialized by judging second
Number (such as, being expressed as " | S | ") and the sample in the renewal hidden variable of the first candidate family
Whether number (such as, being expressed as " | G1 | the ") sum of group reaches predetermined number (such as, table
It is shown as " k ") determine whether the second initialization hidden variable structure completes.This predetermined number is permissible
The e.g. first sample group number initializing hidden variable.When sum of the two is not reaching to predetermined number
During mesh, i.e. | S |+| G1 | < k, then method 300 proceeds to step S380.When sum of the two reaches pre-
During fixed number mesh, i.e. | S |+| G1 | >=k, then may determine that the second initialization hidden variable has had been built up.
In step S380, it is judged that the sample group number in the renewal hidden variable of the first candidate family is
No is 0.Continue executing with if it is not, then method 300 returns step S310；If it is,
Show that the sample group number in the second constructed initialization hidden variable is hidden less than the first initialization
In the case of the sample group number of variable, the renewal hidden variable of the first candidate family does not has
Sample group, now performs step S381.
In step S381, select a sample group in the second initialization hidden variable.
Through step S310-S380, second initializes hidden variable may include one or many
Individual sample group.Therefore, in step S381, can be from second these samples initializing hidden variable
This group selects a sample group.In one embodiment, the second hidden change of initialization can be selected
Sample group largest in amount.In an embodiment of the present invention, largest sample is selected
Group represents the following sample group of selection, and this number of samples included by sample group is most.Alternately,
Scale can be selected to reach the sample group of predetermined threshold, namely the sample comprised in step S381
The sample group that number is more, without being largest sample group.Predetermined threshold is permissible
It is configured according to many factors such as system regulation, model classification, processing speed requirements,
This repeats no more.
In step S382, the sample components of selection is segmented into two parts, to initialize as second
Two sample groups of hidden variable.
According to embodiments of the invention, can be by many to the segmentation that the sample group of selection is carried out
The mode of kind is carried out.For example, it is possible to the sample group selected by step S381 is randomly divided into two
Part, it is also possible to this sample group is equally divided into two parts, it is also possible to enter according to particular requirement
Row segmentation.Should be appreciated that what described dividing method was merely exemplary, be not to this
Bright restriction.
In step S383, initialize the sample group deleting selection hidden variable from second, in order to after
It is not repeated to use this most divided sample group in continuous process.
In step S384, it is judged that second initializes whether hidden variable structure completes.
In one embodiment, in step S384, can be by judging the second initialization hidden variable
In the number of sample group whether reach the sample group number of the first initialization hidden variable, realize
This judgement.It is complete if it is, think that the second initialization hidden variable builds；If it does not,
Then method 300 returns step S381 and continues executing with.
It should be noted that, the embodiment of shown Fig. 3 is merely for the sake of the second hidden change of initialization is discussed
The purpose of the determination process of amount rather than limiting the scope of the present invention.According to the present invention's
Embodiment, is not necessarily required to perform step S310 to S384 and determines the second hidden change of initialization
Amount.Such as, in an alternative embodiment, can be for the first time in every pair of candidate family
Each sample group in the renewal hidden variable of modeling type, determines the first subset of this each sample group,
This first subset such as can determine in the following way: second from every pair of candidate family
Sample group in the renewal hidden variable of candidate family determines the hidden change with the renewal of the first candidate family
Each sample group in amount has the sample group of maximum common factor, using this maximum common factor as this first
Subset.It is then possible to part in addition to the first subset in each sample group is defined as
Two subsets.In this way, it is possible to by each sample in the renewal hidden variable of the first candidate family
Group is divided into the first subset and the second subset.Assuming that the number of the renewal hidden variable of the first candidate family
Mesh is g1, then be divided into the first subset and the second subset by each sample group, can obtain
2g1 new sample set.As 2g1 > k time, wherein k represents the sample of the first initialization hidden variable
This group number, then can be by two less for element number in this 2g1 new sample set samples
Collection is merged into a sample set, then retains the sample set after merging and deletes merged two
Sample set, thus obtains 2g1-1 sample set.Subsequently, if (2g1-1) > k, then continue
Two less for element number in this 2g1-1 sample set sample sets are merged into a sample
Collection, then retains the sample set after merging and deletes merged two sample sets, thus obtaining
2g1-2 sample set.Can so repeat, until the number of the sample set obtained is equal to
Till k.In this way, equally based on the first candidate family in candidate family
The renewal hidden variable updating hidden variable and the second candidate family determines the second initialization hidden variable.
Referring still to Fig. 2, method 200 proceeds to step S260, at this by training multiple the
Two initialize hidden variable obtains multiple mid-module.
In one embodiment, each second in hidden variable can be initialized for multiple second
Initialize hidden variable, training sample set learn, such as supervision or unsupervised learning,
Automatically the study model parameter of mixed model, the hidden variable variation distribution of renewal and model structure,
To generate and each second initialization corresponding model of hidden variable, upper in the disclosure of this model
It is hereinafter referred to as mid-module.In this way, can obtain initializing with multiple second
Multiple mid-modules that hidden variable is corresponding.
In step S270, determine the performance of multiple mid-module.
The performance of model can include many factors, such as precision, ageing etc., it is possible to logical
Cross various ways to determine.As a example by precision, for prediction/classification mixed model, permissible
Mid-module is used for testing the prediction/classification of data, thus obtains the prediction/classification essence of correspondence
Degree.For Clustering Model, the clustering precision of mid-module, such as standard can be calculated mutual
Information etc..
In one embodiment, can be become by the hidden variable of renewal based on each mid-module
Point distribution, model parameter and model structure, determine each mid-module precision, ageing or
Other characteristics of person, can obtain the performance of each mid-module then.
In step S280, select one of multiple mid-module as object module based on performance.
In one embodiment, can based in performance determined by step S270 in multiple
Between model be ranked up, such as, arrange the plurality of intermediate die according to performance order from high to low
Type.It is then possible to select a mid-module conduct of best performance from multiple mid-modules
Object module.
As optional step, after step S280, can be by according to embodiments of the invention
The object module obtained is supplied to user, in order to user uses.
Fig. 4 shows of the method selected for mixed model according to embodiments of the present invention
The schematic diagram of graph-based 400.Mixed model selection course shown in Fig. 4 and the step of Fig. 2
Rapid S210-S280 is corresponding.As shown in Figure 4, it is possible, firstly, to determine based on training sample set
Multiple first initializes hidden variable, and it is denoted respectively as Z1,Z2,…,Zk1, wherein Z represents
First initialize hidden variable, k1 represent determined by first initialize hidden variable number.This is right
Based on training sample set, Ying Yu determines that multiple first initializes hidden variable in step S210.
It follows that Fig. 4 shows initializes hidden variable Z for multiple first1,Z2,…,Zk1Raw
Become corresponding multiple (that is, k1) candidate family, its be denoted respectively as candidate family 1,
Candidate family 2 ..., candidate family k1.This is corresponding to generating with each in step S220
First initializes the candidate family that hidden variable is corresponding.
Fig. 4 also show from candidate family 1, candidate family 2 ..., candidate family k1 choosing
Select out candidate family 1, candidate family 5, candidate family 7 ....This is corresponding in step
S230 performance based on multiple candidate families selects at least two candidate from multiple candidate families
Model.
It follows that Fig. 4 shows that candidate family 1 is paired with candidate family 7, candidate family
5 are paired with candidate family 7, and candidate family 1 is paired with candidate family 5.This is right
At least two candidate family is matched by Ying Yu in step S240.Although it addition, at this be
Briefly it is shown without, it should be understood that can also to selected by step S230
Other candidate families match.
It follows that Fig. 4 shows from candidate family 1, candidate family 5, candidate family 7 etc.
Determine the second initialization hidden variable Z '1,Z’2,Z’3,…,Z’k2Process, wherein k2 represents institute
The number of the second initialization hidden variable determined.This is corresponding to based on pairing in step S250
Candidate family determines the second initialization hidden variable.
Subsequently, Fig. 4 shows and initializes hidden variable Z from second '1,Z’2,Z’3,…,Z’k2Obtain
Mid-module 1, mid-module 2, mid-module 3 ..., mid-module k2.This is corresponding
Multiple mid-module is obtained in initializing hidden variable in step S260 by training multiple second.
Fig. 4 finally show from mid-module 1, mid-module 2, mid-module 3 ...,
Mid-module k2 determines object module.This is corresponding to true in step S270 and step S280
Determine the performance of mid-module, and select one of multiple mid-module as target based on performance
Model.
Should be appreciated that what graph-based 400 was merely exemplary, it is not intended to limit by any way
The scope of the present invention processed.According to embodiments of the invention, be not necessarily to from candidate family 1,
Candidate family 2 ..., candidate family k1 are selected candidate family 1, candidate family 5, are waited
Match again after modeling type 7 grade.Such as, in an alternative embodiment, can be direct
Candidate family 1, candidate family 2 ..., candidate family k1 are matched, is then based on
The candidate family of pairing determines the second initialization hidden variable.
It addition, according to embodiments of the invention, be not necessarily to from mid-module 1, intermediate die
Type 2, mid-module 3 ..., mid-module k2 directly determine object module.The most such as,
In an alternative embodiment, can determine that the 3rd initialization is hidden based on k2 mid-module
Variable, and utilize the 3rd initialization hidden variable to update the second initialization hidden variable, it is then based on
The the second initialization hidden variable updated obtains the mid-module updated.This process can perform one
Secondary or repeatedly, such as by the number of times performed is counted and compares with predetermined execution number of times
Relatively, can stop performing this process after reaching predetermined execution number of times.Alternately, it is also possible to
Iteration performs this process until reaching pre-provisioning request.Multiple realization can be there is in pre-provisioning request
Mode, the mid-module performance such as updated is the best, and the number of the mid-module of renewal reaches
Predetermined number, etc..Should be appreciated that what above-described embodiment was merely exemplary, be not right
The restriction of the scope of the present invention.Within the scope of the invention, those skilled in the art can be to this
Inventive embodiment carries out various deformation.
Fig. 5 shows the device 500 for mixed model selection according to embodiments of the present invention
Schematic block diagram.As it can be seen, device 500 includes: candidate family signal generating unit 510, quilt
It is configured to pass training multiple first and initializes hidden variable to generate multiple candidate family；First is true
Cell 520, is configured to determine the second initialization hidden variable based on multiple candidate families；With
And second determine unit 530, it is configured to initialize hidden variable based on second and determines target mould
Type.
In one embodiment, candidate family signal generating unit 510 may include that the first initialization
Hidden variable determines unit, is configured to determine that multiple first initialization is hidden based on training sample set
Variable；And training unit, be configured to for multiple first initialize in hidden variables each
First initializes hidden variable, learns on training sample set, to generate and at the beginning of each first
The candidate family that beginningization hidden variable is corresponding.
In one embodiment, first determines that unit 520 may include that pairing unit, is joined
It is set to multiple candidate families are matched；And second initialize hidden variable determine unit, quilt
The candidate family being configured to pairing determines the second initialization hidden variable.
In one embodiment, first determines that unit 520 may include that the first candidate family choosing
Select unit, be configured to performance based on multiple candidate families select from multiple candidate families to
Few two candidate families；Pairing unit, is configured to match at least two candidate family；
And second initialize hidden variable determine unit, be configured to based on pairing candidate family determine
Second initializes hidden variable.
In one embodiment, first determines that the first candidate family selection unit of unit 520 can
To include: the first capabilities determination unit, it is configured to based on each time in multiple candidate families
The distribution of hidden variable variation, model parameter and the model structure of the renewal of modeling type, determines each
The performance of candidate family；Sequencing unit, be configured to based on determined by performance to multiple candidates
Model is ranked up；And second candidate family select unit, be configured to based on sequence from many
Individual candidate family selects at least two candidate family.
In one embodiment, every pair of candidate family in the candidate family of pairing can include
One candidate family and the second candidate family, first determines the second initialization hidden variable of unit 520
Determine that unit can be configured to: based on the first candidate family in every pair of candidate family
Renewal hidden variable and the renewal hidden variable of the second candidate family, determine one second initialization
Hidden variable.
In one embodiment, first determines that the second initialization hidden variable of unit 520 determines list
Unit may include that the first sample group determines unit, is configured to select in every pair of candidate family
A sample group in the renewal hidden variable of the first candidate family is as the first sample group；Second sample
This group determines unit, is configured to based on the first sample group, second from every pair of candidate family
Sample group in the renewal hidden variable of candidate family determines the second sample group；Occur simultaneously and determine unit,
It is configured to determine that the common factor of the first sample group and the second sample group；Cutting unit, is configured to
Based on occuring simultaneously, the first sample components is segmented into two subsets；And second initialize hidden variable build
Unit, is configured to build the second initialization hidden variable based on two subsets.
In one embodiment, first determines that the second initialization hidden variable of unit 520 determines list
If the second initialization hidden variable construction unit of unit is configured to the second constructed initialization
The sample group number of hidden variable less than first initialization hidden variable sample group number, then perform with
Lower operation is until second initializes the sample group number of hidden variable equal to the first initialization hidden variable
Sample group number: select second initialization hidden variable in a sample group；The sample that will select
This component is segmented into two parts, to initialize two sample groups of hidden variable as second；And from
Second initializes the sample group deleting selection in hidden variable.
In one embodiment, first determines that unit 520 may include that mid-module generates list
Unit, is configured to train multiple second to initialize hidden variable to obtain multiple mid-module；
3rd initializes hidden variable determines unit, is configured to determine at the beginning of the 3rd based on multiple mid-modules
Beginningization hidden variable；And updating block, it is configured to, with the 3rd initialization hidden variable and updates the
Two initialize hidden variable.
In one embodiment, have multiple second and initialize hidden variable, and second determines list
Unit 530 may include that mid-module determines unit, is configured to train at the beginning of multiple second
Beginningization hidden variable obtains multiple mid-module；Second capabilities determination unit, is configured to determine that
The performance of multiple mid-modules；And object module selects unit, it is configured to select based on performance
Select one of multiple mid-module as object module.
For clarity, selectable unit that device 500 comprises and each it is shown without in Figure 5
The subelement that individual unit is comprised.Should be appreciated that device 500 can utilize to be embodied in various ways.
Such as, in certain embodiments, device 500 can utilize software and/or firmware to realize.Example
As, device 500 may be implemented as the computer program product comprised on a computer-readable medium
Product, each unit therein is the program module being realized its function by computer instruction.Standby
Selection of land or additionally, device 500 can partially or fully realize based on hardware.Example
As, device 500 can be implemented as integrated circuit (IC) chip, special IC (ASIC)
Or SOC(system on a chip) (SOC).Other modes currently known or exploitation in the future are also feasible
, the scope of the present invention is not limited in this respect.
Fig. 6 shows the exemplary computer system 600 be suitable to for realizing the embodiment of the present invention
Schematic block diagram.As it can be seen, computer system 600 includes CPU (CPU)
601, it can be according to the program being stored in read only memory (ROM) 602 or from depositing
Storage unit 608 is loaded into the program in random access storage device (RAM) 603 and performs various
Suitable action and process.In RAM 603, also storage has CPU 601 to perform various place
The data that reason etc. are required.CPU 601, ROM 602 and RAM 603 are by bus 604
It is connected with each other.Input/output (I/O) unit 605 is also connected to bus 604.
It is connected to I/O interface 605: include the input block 606 of keyboard, mouse etc. with lower component；
Including such as cathode ray tube (CRT), liquid crystal display (LCD) etc. and speaker etc.
Output unit 607；Memory element 608 including hard disk etc.；And include such as LAN card,
The communication unit 609 of the NIC of modem etc..Communication unit 609 is via such as
The network of the Internet performs communication process.Driver 610 is connected to I/O interface also according to needs
605.Removable medium 611, such as disk, CD, magneto-optic disk, semiconductor memory etc.,
It is arranged on as required in driving 610, in order to the computer program read from it is according to need
Memory element 608 to be mounted into.
Especially, according to embodiments of the invention, each process above-described can be implemented
For computer software programs.Such as, embodiments of the invention include a kind of computer program,
It includes the computer program being tangibly embodied on machine readable media, described computer program
Comprise the program code for performing each method.In such embodiments, this computer program
Can be downloaded and installed from network by communication unit 609, and/or from removable medium
611 are mounted.
It is said that in general, the various example embodiment of the present invention can be at hardware or special circuit, soft
Part, logic, or its any combination are implemented.Some aspect can be implemented within hardware, and its
His aspect can by controller, microprocessor or other calculate firmware that equipment performs or
Software is implemented.When each side of embodiments of the invention is illustrated or described as block diagram, flow process
Figure or when using some other figure to represent, it will be appreciated that square frame described herein, device, system,
Techniques or methods can be as nonrestrictive example at hardware, software, firmware, special circuit
Logic, common hardware or controller or other calculate equipment, or its some combinations is implemented.
And, each frame in flow chart can be counted as method step, and/or computer program
The operation that the operation of code generates, and/or it is interpreted as performing the logic of multiple couplings of correlation function
Component.Such as, embodiments of the invention include computer program, this computer journey
Sequence product includes the computer program visibly realized on a machine-readable medium, this computer journey
Sequence comprises the program code being configured to realize method described above.
In the context of the disclosure, machine readable media can be to comprise or store for or have
Any tangible medium about the program of instruction execution system, device or equipment.Machine readable is situated between
Matter can be machine-readable signal medium or machinable medium.Machine readable media is permissible
Include but not limited to electronics, magnetic, optics, electromagnetism, infrared or semiconductor system,
Device or equipment, or the combination of its any appropriate.The more detailed example of machinable medium
Including with the electrical connection of one or more wire, portable computer diskette, hard disk, with
Machine memory access device (RAM), read only memory (ROM), erasable programmable is read-only deposits
Reservoir (EPROM or flash memory), light storage device, magnetic storage apparatus, or its any appropriate
Combination.
Can compile with one or more for realizing the computer program code of the method for the present invention
Cheng Yuyan writes.These computer program codes can be supplied to general purpose computer, dedicated computing
Machine or the processor of other programmable data processing meanss so that program code is by computer
Or the when of the execution of other programmable data processing meanss, cause in flow chart and/or block diagram
Function/the operation of regulation is carried out.Program code can the most on computers, part calculate
On machine, as independent software kit, part the most on computers and part the most on the remote computer or
Perform on remote computer or server completely.
Although it addition, operation is depicted with particular order, but this and should not be construed and require this
Generic operation with the particular order illustrated or completes with sequential order, or performs the behaviour of all diagrams
Make to obtain expected result.In some cases, multitask or parallel processing can be useful.
Similarly, contain some specific implementation detail although discussed above, but this should not explain
For limiting any invention or the scope of claim, and should be interpreted that can be for specific invention
The description of specific embodiment.In this specification described in the context of separate embodiment
Some feature can also combined implementation in single embodiment.On the contrary, in single embodiment
Various features described in context can also be discretely in multiple embodiments or in any appropriate
Sub-portfolio in implement.
The various amendments of example embodiment, change for the aforementioned present invention will looked into together with accompanying drawing
When seeing described above, those skilled in the technology concerned are become obvious.Any and all amendment
Unrestriced and the present invention example embodiment scope will be still fallen within.Additionally, aforementioned specification and
There is the benefit inspired in accompanying drawing, relates to the technology people of the technical field of embodiments of the invention
Member will appreciate that other embodiments of the present invention herein illustrated.
It will be appreciated that embodiments of the invention are not limited to disclosed specific embodiment, and revise
All should be contained in scope of the appended claims with other embodiments.Although being used here spy
Fixed term, but it only uses in the sense that describing general, is not limited to purpose.
Claims (20)
1. the method selected for mixed model, described method includes:
Initialize hidden variable by training multiple first and generate multiple candidate family；
The second initialization hidden variable is determined based on the plurality of candidate family；And
Initialize hidden variable based on described second and determine object module.
Method the most according to claim 1, wherein initializes hidden by training multiple first
Variable generates multiple candidate family and includes:
Determine that the plurality of first initializes hidden variable based on training sample set；And
The each first initialization hidden variable in hidden variable is initialized for the plurality of first,
Learn on training sample set, corresponding with described each first initialization hidden variable to generate
Candidate family.
Method the most according to claim 1, wherein determines based on the plurality of candidate family
Second initializes hidden variable includes:
The plurality of candidate family is matched；And
Candidate family based on pairing determines that described second initializes hidden variable.
Method the most according to claim 1, wherein determines based on the plurality of candidate family
Second initializes hidden variable includes:
Performance based on the plurality of candidate family selects at least from the plurality of candidate family
Two candidate families；
Described at least two candidate family is matched；And
Candidate family based on pairing determines that described second initializes hidden variable.
Method the most according to claim 4, wherein property based on the plurality of candidate family
At least two candidate family can be selected from the plurality of candidate family to include:
The hidden variable variation of renewal based on each candidate family in the plurality of candidate family
Distribution, model parameter and model structure, determine the performance of described each candidate family；
The plurality of candidate family is ranked up by performance determined by based on；And
From the plurality of candidate family, at least two candidate family is selected based on described sequence.
6. according to the method described in claim 3 or 4, the candidate family of wherein said pairing
In every pair of candidate family include the first candidate family and the second candidate family, wherein based on pairing
Candidate family determine described second initialize hidden variable include:
Renewal hidden variable and second based on the first candidate family in described every pair of candidate family
The renewal hidden variable of candidate family, determines that one second initializes hidden variable.
Method the most according to claim 6, wherein based in described every pair of candidate family
The renewal hidden variable of the first candidate family and the renewal hidden variable of the second candidate family, determine one
Individual second initializes hidden variable includes:
The following one or many that operates of execution:
Select in the renewal hidden variable of the first candidate family in described every pair of candidate family
One sample group is as the first sample group；
Based on described first sample group, the second candidate family from described every pair of candidate family
Renewal hidden variable in sample group determine the second sample group；
Determine the common factor of described first sample group and described second sample group；
Based on described common factor, described first sample components is segmented into two subsets；And
The second initialization hidden variable is built based on said two subset.
Method the most according to claim 7, wherein builds described based on said two subset
Second initializes hidden variable includes:
If the sample group number of the second constructed initialization hidden variable is at the beginning of less than described first
The sample group number of beginningization hidden variable, then perform following operation until described second initializes hidden change
The sample group number of amount is equal to the described first sample group number initializing hidden variable:
Described second is selected to initialize a sample group in hidden variable；
The sample components of selection is segmented into two parts, to initialize hidden variable as described second
Two sample groups；And
The sample group deleting described selection hidden variable is initialized from described second.
Method the most according to claim 1, wherein determines based on the plurality of candidate family
Second initializes hidden variable includes:
The following one or many that operates of execution:
Initialize hidden variable by training the plurality of second and obtain multiple mid-module；
The 3rd initialization hidden variable is determined based on the plurality of mid-module；And
Utilize the described 3rd to initialize hidden variable and update described second initialization hidden variable.
Method the most according to claim 1, wherein has multiple second and initializes hidden change
Amount, and wherein determine that object module includes based on described second initialization hidden variable:
Initialize hidden variable by training the plurality of second and obtain multiple mid-module；
Determine the performance of the plurality of mid-module；And
Select one of the plurality of mid-module as described object module based on described performance.
11. 1 kinds of devices selected for mixed model, described device includes:
Candidate family signal generating unit, is configured to train multiple first to initialize hidden variable and comes
Generate multiple candidate family；
First determines unit, is configured to determine the second initialization based on the plurality of candidate family
Hidden variable；And
Second determines unit, is configured to initialize hidden variable based on described second and determines target
Model.
12. devices according to claim 11, wherein said candidate family signal generating unit bag
Include:
First initializes hidden variable determines unit, is configured to determine institute based on training sample set
State multiple first and initialize hidden variable；And
Training unit, is configured to initialize each the in hidden variable for the plurality of first
One initializes hidden variable, learns on training sample set, to generate and described each first
Initialize the candidate family that hidden variable is corresponding.
13. devices according to claim 11, wherein said first determines that unit includes:
Pairing unit, is configured to match the plurality of candidate family；And
Second initializes hidden variable determines unit, is configured to candidate family based on pairing and determines
Described second initializes hidden variable.
14. devices according to claim 11, wherein said first determines that unit includes:
First candidate family selects unit, is configured to performance based on the plurality of candidate family
At least two candidate family is selected from the plurality of candidate family；
Pairing unit, is configured to match described at least two candidate family；And
Second initializes hidden variable determines unit, is configured to candidate family based on pairing and determines
Described second initializes hidden variable.
15. devices according to claim 14, wherein said first candidate family selects single
Unit includes:
First capabilities determination unit, is configured to based on each time in the plurality of candidate family
The distribution of hidden variable variation, model parameter and the model structure of the renewal of modeling type, determines described
The performance of each candidate family；
Sequencing unit, be configured to based on determined by performance the plurality of candidate family is carried out
Sequence；And
Second candidate family selects unit, is configured to based on described sequence from the plurality of candidate
Model selects at least two candidate family.
16. according to the device described in claim 13 or 14, the candidate of wherein said pairing
Every pair of candidate family in model includes the first candidate family and the second candidate family, wherein said
Second initializes hidden variable determines that unit is configured to:
Renewal hidden variable and second based on the first candidate family in described every pair of candidate family
The renewal hidden variable of candidate family, determines that one second initializes hidden variable.
17. devices according to claim 16, wherein said second initialization hidden variable is true
Cell includes:
First sample group determines unit, is configured to select first in described every pair of candidate family
A sample group in the renewal hidden variable of candidate family is as the first sample group；
Second sample group determines unit, is configured to based on described first sample group, from described often
Sample group in the renewal hidden variable of the second candidate family in candidate family is determined the second sample
This group；
Occur simultaneously and determine unit, be configured to determine that described first sample group and described second sample group
Common factor；
Cutting unit, is configured to, based on described common factor, described first sample components is segmented into two
Subset；And
Second initializes hidden variable construction unit, is configured to build the based on said two subset
Two initialize hidden variable.
18. devices according to claim 17, wherein said second initializes hidden variable structure
If building unit to be configured to the sample group number of constructed second initialization hidden variable and be less than
The described first sample group number initializing hidden variable, then perform following operation until described second
Initialize the sample group number of hidden variable equal to the described first sample group number initializing hidden variable
Mesh:
Described second is selected to initialize a sample group in hidden variable；
The sample components of selection is segmented into two parts, to initialize hidden variable as described second
Two sample groups；And
The sample group deleting described selection hidden variable is initialized from described second.
19. devices according to claim 11, wherein said first determines that unit includes:
Mid-module signal generating unit, is configured to train the plurality of second to initialize hidden change
Amount obtains multiple mid-module；
3rd initializes hidden variable determines unit, is configured to based on the plurality of mid-module true
Fixed 3rd initialization hidden variable；And
Updating block, is configured to, with the described 3rd and initializes at the beginning of hidden variable renewal described second
Beginningization hidden variable.
20. devices according to claim 11, wherein have multiple second and initialize hidden change
Measure, and wherein said second determine that unit includes:
Mid-module determines unit, is configured to train the plurality of second to initialize hidden change
Amount obtains multiple mid-module；
Second capabilities determination unit, is configured to determine that the performance of the plurality of mid-module；With
And
Object module selects unit, is configured to select the plurality of intermediate die based on described performance
One of type is as described object module.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510149617.4A CN106156077A (en) | 2015-03-31 | 2015-03-31 | The method and apparatus selected for mixed model |
| JP2016040843A JP6172315B2 (en) | 2015-03-31 | 2016-03-03 | Method and apparatus for mixed model selection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510149617.4A CN106156077A (en) | 2015-03-31 | 2015-03-31 | The method and apparatus selected for mixed model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106156077A true CN106156077A (en) | 2016-11-23 |
Family
ID=57322763
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510149617.4A Pending CN106156077A (en) | 2015-03-31 | 2015-03-31 | The method and apparatus selected for mixed model |
Country Status (2)
| Country | Link |
|---|---|
| JP (1) | JP6172315B2 (en) |
| CN (1) | CN106156077A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993300A (en) * | 2017-12-29 | 2019-07-09 | 华为技术有限公司 | A kind of training method and device of neural network model |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102002732B1 (en) * | 2018-09-20 | 2019-07-23 | 건국대학교 산학협력단 | Deep neural network based data processing method and apparatus using ensemble model |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000163395A (en) * | 1998-11-30 | 2000-06-16 | Nippon Telegr & Teleph Corp <Ntt> | Maximum likelihood estimation for mixed model and recording medium where maximum likelihood estimation program for mixed model is recorded |
| JP4860265B2 (en) * | 2004-01-16 | 2012-01-25 | 日本電気株式会社 | Text processing method / program / program recording medium / device |
| US20120209880A1 (en) * | 2011-02-15 | 2012-08-16 | General Electric Company | Method of constructing a mixture model |
-
2015
- 2015-03-31 CN CN201510149617.4A patent/CN106156077A/en active Pending
-
2016
- 2016-03-03 JP JP2016040843A patent/JP6172315B2/en active Active
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993300A (en) * | 2017-12-29 | 2019-07-09 | 华为技术有限公司 | A kind of training method and device of neural network model |
| CN109993300B (en) * | 2017-12-29 | 2021-01-29 | 华为技术有限公司 | Training method and device of neural network model |
| US11521012B2 (en) | 2017-12-29 | 2022-12-06 | Huawei Technologies Co., Ltd. | Method for training neural network model and apparatus |
| US11966844B2 (en) | 2017-12-29 | 2024-04-23 | Huawei Technologies Co., Ltd. | Method for training neural network model and apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2016194910A (en) | 2016-11-17 |
| JP6172315B2 (en) | 2017-08-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hamilton et al. | Embedding logical queries on knowledge graphs | |
| Jin et al. | Bayesian symbolic regression | |
| Doppa et al. | HC-Search: A learning framework for search-based structured prediction | |
| CN106845530A (en) | character detection method and device | |
| CN104199818B (en) | Method is recommended in a kind of socialization based on classification | |
| CN101561805A (en) | Document classifier generation method and system | |
| Machina et al. | Singular dynamics in gene network models | |
| CN104102917A (en) | Construction method of domain self-adaptive classifier, construction device for domain self-adaptive classifier, data classification method and data classification device | |
| CN106156857B (en) | The method and apparatus of the data initialization of variation reasoning | |
| Van Mechelen et al. | Benchmarking in cluster analysis: A white paper | |
| CN114418035A (en) | Decision tree model generation method and data recommendation method based on decision tree model | |
| CN114298851A (en) | Network user social behavior analysis method and device based on graph sign learning and storage medium | |
| CN112420125A (en) | Molecular attribute prediction method and device, intelligent equipment and terminal | |
| Przybyła-Kasperek et al. | Dispersed decision-making system with fusion methods from the rank level and the measurement level–A comparative study | |
| CN113330462A (en) | Neural network training using soft nearest neighbor loss | |
| Volkovs et al. | Efficient sampling for bipartite matching problems | |
| CN106156077A (en) | The method and apparatus selected for mixed model | |
| CN104391828B (en) | The method and apparatus for determining short text similarity | |
| CN106156856A (en) | The method and apparatus selected for mixed model | |
| CN109886299A (en) | A kind of user draws a portrait method, apparatus, readable storage medium storing program for executing and terminal device | |
| Shim et al. | Fast and accurate interpretation of workload classification model | |
| CN113792132A (en) | Target answer determination method, device, equipment and medium | |
| EP3460723A1 (en) | Evaluating input data using a deep learning algorithm | |
| Wang et al. | The graph-guided group lasso for genome-wide association studies | |
| CN104298729B (en) | Data classification method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into 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: 20161123 |