Model training method and device used in machine learning
Model training method and device used in machine learning
 CN 105,809,204 A
 Filed: 03/31/2016
 Published: 07/27/2016
 Est. Priority Date: 03/31/2016
 Status: Active Application
First Claim
1. the model training method in machine learning, including:
 Feature is extracted from training sample；
Utilizing the described feature extracted that the object module built in advance is trained, described object module includes linear segment and nonlinear partial, and the nonlinear of described object module is divided into$\frac{1}{2}{X}^{T}({V}^{T}\mathrm{\λVD)X;}$Wherein, V ∈
R^{k×P}、
λ
∈
R^{k×k}And D ∈
R^{P×P}For the model parameter of nonlinear partial in described object module, X ∈
R^{P×N}Represent the eigenmatrix of described training sample, X=(x_{1},x_{2},...x_{i},...x_{N}), x_{i}∈
R^{P}Represent the characteristic vector of ith training sample and x_{i}∈
R^{P}For the column vector of P dimension, λ
∈
R^{k×k}With D ∈
R^{P×P}For diagonal matrix.
Chinese PRB Reexamination
Abstract
The embodiment of the invention provides a model training method used in machine learning.The method comprises the steps of extracting features from training samples; training a preestablished target model by means of the extracted features, wherein the nonlinear part of the target model is expressed as the formula in the description; V belongs to R<k*P>, lambda belongs to R<k*k>, D belongs to R<P*P>, V, lambda and D are the model parameters of the nonlinear part of the target model, X belongs to R<P*N> and represents the feature matrix of the training samples, X=(x1, x2,...xi,...xN), xi belongs to R<P>, represents the feature vector of the i<th> training sample and is the column vector of P dimension, and lambda and D are diagonal matrixes.The invention further provides a model training device used in machine learning.

3 Citations
No References
System and Method for Automated Recommendation of Advertisement Targeting Attributes  
Patent #
US 20120150626A1
Filed 12/10/2010

Current Assignee
N/A

Layering aviation operation system HM/FM (health monitoring/fault management) modeling and evaluating method based on stochastic Petri net  
Patent #
CN 102,592,011 A
Filed 12/30/2011

Current Assignee

Characteristic recommendation method and device  
Patent #
CN 104,615,790 A
Filed 03/09/2015

Current Assignee

10 Claims

1. the model training method in machine learning, including:

Feature is extracted from training sample； Utilizing the described feature extracted that the object module built in advance is trained, described object module includes linear segment and nonlinear partial, and the nonlinear of described object module is divided into $\frac{1}{2}{X}^{T}({V}^{T}\mathrm{\λVD)X;}$ Wherein, V ∈
R^{k×P}、
λ
∈
R^{k×k}And D ∈
R^{P×P}For the model parameter of nonlinear partial in described object module, X ∈
R^{P×N}Represent the eigenmatrix of described training sample, X=(x_{1},x_{2},...x_{i},...x_{N}), x_{i}∈
R^{P}Represent the characteristic vector of ith training sample and x_{i}∈
R^{P}For the column vector of P dimension, λ
∈
R^{k×k}With D ∈
R^{P×P}For diagonal matrix.


2. method according to claim 1, wherein, described λ
 ∈
R^{k×k}Diagonal entry be 1.
 ∈

3. method according to claim 1, wherein, as (V^{T}λ
 VD) for positive semidefinite matrix time, described λ
∈
R^{k×k}Diagonal entry be 1.
 VD) for positive semidefinite matrix time, described λ

4. method according to claim 1, wherein, described D ∈
 R^{P×P}Diagonal entry be 1.

5. method as claimed in any of claims 1 to 4, wherein, the linear segment of described object module is the linear segment of FM model.

6. the model training equipment in machine learning, including:

Feature extraction unit, for extracting feature from training sample； Model training unit, for utilizing the described feature of extraction that the object module built in advance is trained, described object module includes linear segment and nonlinear partial, and the nonlinear of described object module is divided into Wherein, V ∈
R^{k×P}、
λ
∈
R^{k×k}And D ∈
R^{P×P}For the model parameter of nonlinear partial in described object module, X ∈
R^{P×N}Represent the eigenmatrix of described training sample, X=(x_{1},x_{2},...x_{i},...x_{N}), x_{i}∈
R^{P}Represent the characteristic vector of ith training sample and x_{i}∈
R^{P}For the column vector of P dimension, λ
∈
R^{k×k}With D ∈
R^{P×P}For diagonal matrix.


7. equipment according to claim 6, wherein, described λ
 ∈
R^{k×k}Diagonal entry be 1.
 ∈

8. equipment according to claim 6, wherein, as (V^{T}λ
 VD) for positive semidefinite matrix time, described λ
∈
R^{k×k}Diagonal entry be 1.
 VD) for positive semidefinite matrix time, described λ

9. equipment according to claim 6, wherein, described D ∈
 R^{P×P}Diagonal entry be 1.

10. the equipment according to any one in claim 6 to 9, wherein, the linear segment of described object module is the linear segment of FM model.
Specification(s)