CN101419610A - Information processing device, information processing method, and program - Google Patents

Information processing device, information processing method, and program Download PDF

Info

Publication number
CN101419610A
CN101419610A CNA2008101702877A CN200810170287A CN101419610A CN 101419610 A CN101419610 A CN 101419610A CN A2008101702877 A CNA2008101702877 A CN A2008101702877A CN 200810170287 A CN200810170287 A CN 200810170287A CN 101419610 A CN101419610 A CN 101419610A
Authority
CN
China
Prior art keywords
characteristic extraction
extraction expression
characteristic
target signature
list
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.)
Granted
Application number
CNA2008101702877A
Other languages
Chinese (zh)
Other versions
CN101419610B (en
Inventor
小林由幸
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sony Corp
Original Assignee
Sony Corp
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Priority claimed from JP2007273418A external-priority patent/JP4392622B2/en
Priority claimed from JP2007273416A external-priority patent/JP4392621B2/en
Application filed by Sony Corp filed Critical Sony Corp
Publication of CN101419610A publication Critical patent/CN101419610A/en
Application granted granted Critical
Publication of CN101419610B publication Critical patent/CN101419610B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Machine Translation (AREA)

Abstract

An information processing device for generating a target feature amount computational expression for outputting a target feature amount corresponding to input data, comprising: a feature amount extraction expression list generating unit configured to generate and update a feature amount extraction expression list; a feature amount computing unit configured to input actual data supplied as tutor data to each feature amount extraction expression included in the feature amount extraction expression list to compute multiple feature amounts corresponding to the actual data; a target feature amount computational expression generating unit configured to employ the multiple feature amounts, and an existing feature amount corresponding to the actual data supplied as tutor data for the same rank to generate the target feature amount computational expression by machine learning; and an evaluation value computing unit configured to compute the evaluation value of each feature amount extraction expression included in the feature amount extraction expression list.

Description

Messaging device, information processing method and program
The cross reference of related application
The present invention comprise with by quoting with complete content Japanese patent application JP 2007-273417, the Japanese patent application JP 2007-273416 subject content relevant that be incorporated into this, that all submitted to Jap.P. office on 20 22nd, 2007 with Japanese patent application JP 2007-273418.
Technical field
The present invention relates to a kind of messaging device, information processing method and program, and relate to particularly and a kind ofly realize automatic structural attitude amount computational algorithm, can calculate messaging device, information processing method and the program of content-data thus as the characteristic quantity of for example music data.
Background technology
A kind of method (for example international publication number WO2007/049641) of utilizing genetic searching method and a kind of method (for example U.S. Patent Application Publication No. US 2004/0181401A1) of not utilizing genetic searching method have been proposed up to now as a kind ofly being used for automatic structural attitude amount computational algorithm, can exporting the invention of characteristic quantity such as input data such as music data, view data (being to be speed, brightness, lively degree under the situation of music data in the input data) thus.
Summary of the invention
Yet, by correlation technique automatically the characteristic quantity computational algorithm of structure generally include and compare redundant arithmetical operation with the characteristic quantity computational algorithm of manual construction, thereby the arithmetical operation time necessary that is used to obtain and import data characteristic of correspondence amount prolongs in some cases.
In addition, among the existing characteristic quantity that can obtain under the situation that need not to use the characteristic quantity computational algorithm execution calculating that to develop, effective considerable characteristic quantity has been known among characteristic quantity computational algorithm developer for structural attitude amount computational algorithm, but does not propose to be used for the method for structural attitude amount computational algorithm so far as yet.
Note hereinafter in this manual, need not to use the characteristic quantity computational algorithm execution that to construct automatically to calculate the existing characteristic quantity that can obtain and to be called existing characteristic quantity.On the other hand, will will be called the target signature amount by the characteristic quantity that use characteristic amount computational algorithm obtains,
For automatic structural attitude amount computational algorithm, by utilizing the existing characteristic quantity corresponding also can calculate the target signature amount corresponding generally acknowledged demand has been arranged thus with importing data with importing data.
As one embodiment of the present of invention, a kind of according to an embodiment of the invention messaging device, be used to obtain input data and the existing characteristic quantity corresponding as input with the input data, and generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, this messaging device comprises: Characteristic Extraction expression list generation unit, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using; Feature amount calculation unit is configured to the real data as the instruction data supply is input to each Characteristic Extraction expression formula of comprising to calculate a plurality of characteristic quantities corresponding with real data in the Characteristic Extraction expression list; Target signature amount calculation expression generation unit, be configured to utilize comparably a plurality of characteristic quantities corresponding that calculated with real data and with as the corresponding existing characteristic quantity of the real data of instruction data supply with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression; And the evaluation of estimate computing unit, be configured to calculate the evaluation of estimate of each the Characteristic Extraction expression formula that in the Characteristic Extraction expression list, comprises.
Target signature amount calculation expression generation unit can utilize selectively comparably in a plurality of characteristic quantities corresponding that calculated with real data some characteristic quantities and with as some characteristic quantities in the corresponding a plurality of existing characteristic quantity of real data of instruction data supply with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression.
The evaluation of estimate computing unit can come the calculated characteristics amount to extract the evaluation of estimate of the described Characteristic Extraction expression formula that comprises in the expression list based on the contribution rate of each target signature amount calculation expression of a plurality of characteristic quantities corresponding with real data that calculated.
A kind of according to an embodiment of the invention messaging device, be used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, this messaging device comprises: Characteristic Extraction expression list generation unit, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using; Feature amount calculation unit is configured to the real data as the instruction data supply is input to the average computation time of each Characteristic Extraction expression formula to calculate a plurality of characteristic quantities corresponding with real data and also to measure individual features amount extraction expression formula that comprises in the Characteristic Extraction expression list; Target signature amount calculation expression generation unit is configured to utilize a plurality of characteristic quantities corresponding with real data that calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instruction data supply generates target signature amount calculation expression; And the evaluation of estimate computing unit, be configured to calculate the evaluation of estimate of each the Characteristic Extraction expression formula that in the Characteristic Extraction expression list, comprises and also proofread and correct the evaluation of estimate of being calculated based on the average computation time of individual features amount extraction expression formula.
Target signature amount calculation expression generation unit can utilize some characteristic quantities in a plurality of characteristic quantities corresponding with real data that calculated with by being used to estimate to generate target signature amount calculation expression with machine learning as the corresponding target signature amount of the real data of instruction data supply selectively.
Target signature amount calculation expression generation unit can utilize some characteristic quantities in a plurality of characteristic quantities corresponding with real data that calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression based on the average computation time that the character pair amount is extracted expression formula selectively.
A kind of according to an embodiment of the invention messaging device, be used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, this messaging device comprises: Characteristic Extraction expression list generation unit, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using; Feature amount calculation unit is configured to the real data as the instruction data supply is input to each Characteristic Extraction expression formula of comprising to calculate a plurality of characteristic quantities corresponding with real data in the Characteristic Extraction expression list; Target signature amount calculation expression generation unit is configured to utilize a plurality of characteristic quantities corresponding with real data that calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instruction data supply generates target signature amount calculation expression; The evaluation of estimate computing unit is configured to calculate the evaluation of estimate of each the Characteristic Extraction expression formula that comprises in the Characteristic Extraction expression list; And the optimization unit, be configured to optimize each the Characteristic Extraction expression formula in a plurality of Characteristic Extraction expression formulas that comprise in the generation Characteristic Extraction expression list in the end.
Optimizing the unit can comprise: Characteristic Extraction expression optimization unit is configured to optimize in the following manner each the Characteristic Extraction expression formula that comprises in the generation Characteristic Extraction expression list in the end: promptly extract expression formula according to the individual features amount that comprises in the generation Characteristic Extraction expression list in the end and detect the Combination Optimized pattern of the redundant operation symbol of having represented registration in advance and delete operator or replace with the less operational symbol of arithmetic load.
Optimizing the unit can comprise: Characteristic Extraction expression optimization unit is configured to: be out of shape each Characteristic Extraction expression formula of comprising in the generation Characteristic Extraction expression list in the end to generate a plurality of optimization candidate expression formulas; Following optimization candidate expression formula among the optimization candidate expression formula of a plurality of generations is given good evaluation, and the output of this optimization candidate expression formula of acquisition has with shorter as the computing time of the high degree of correlation of the output of the Characteristic Extraction expression formula of deformation sources and this optimization candidate expression formula; As gene, utilize genetic algorithm based on the evaluation of optimizing candidate's expression formula the optimization candidate expression formula of a plurality of generations to upgrade the optimization candidate expression formula of a plurality of generations; And the optimization candidate expression formula that will have the best evaluation individual features amount that finally is defined as comprising in the generation Characteristic Extraction expression list is in the end extracted the optimization result of expression formula.
Optimizing the unit can comprise: reconfiguration unit is configured to utilize the Characteristic Extraction expression formula of being optimized with the target signature amount calculation expression of reconstruct with the corresponding generation of last generation Characteristic Extraction expression list.
A kind of according to an embodiment of the invention information processing method, be used to obtain input data and the existing characteristic quantity corresponding with the input data as input and generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, this information processing method comprises: generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols at random; To be input to each Characteristic Extraction expression formula of in the Characteristic Extraction expression list, comprising as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding with real data; Utilize comparably a plurality of characteristic quantities corresponding calculated with real data and with as the corresponding existing characteristic quantity of the real data of instruction data supply with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression; The evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in the Characteristic Extraction expression list; And a plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using.
A kind of according to an embodiment of the invention information processing method, be used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, this information processing method comprises: generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols at random; To be input to the average computation time of each Characteristic Extraction expression formula that in the Characteristic Extraction expression list, comprises as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding and also to measure individual features amount extraction expression formula with real data; Utilize a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression with real data; The evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in the Characteristic Extraction expression list and the average computation time of also extracting expression formula based on the individual features amount are proofreaied and correct the evaluation of estimate of being calculated; And a plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using.
A kind of according to an embodiment of the invention information processing method, be used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, this information processing method comprises: generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols at random; To be input to each Characteristic Extraction expression formula of in the Characteristic Extraction expression list, comprising as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding with real data; Utilize a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression with real data; The evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in the Characteristic Extraction expression list; A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using; And optimize each Characteristic Extraction expression formula in a plurality of Characteristic Extraction expression formulas that comprise in the generation Characteristic Extraction expression list in the end.
A kind of according to an embodiment of the invention program that is used for the control information treatment facility and makes the computing machine execution processing of messaging device, this messaging device is used to obtain input data and the existing characteristic quantity corresponding with the input data as input and generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, and this processing may further comprise the steps: generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols at random; To be input to each Characteristic Extraction expression formula of in the Characteristic Extraction expression list, comprising as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding with real data; Utilize comparably a plurality of characteristic quantities corresponding calculated with real data and with as the corresponding existing characteristic quantity of the real data of instruction data supply with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression; The evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in the Characteristic Extraction expression list; And a plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using.
A kind of according to an embodiment of the invention program that is used for the control information treatment facility and makes the computing machine execution processing of messaging device, this messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, and this processing may further comprise the steps: generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols at random; To be input to the average computation time of each Characteristic Extraction expression formula that in the Characteristic Extraction expression list, comprises as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding and also to measure individual features amount extraction expression formula with real data; Utilize a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression with real data; The evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in the Characteristic Extraction expression list and the average computation time of also extracting expression formula based on the individual features amount are proofreaied and correct the evaluation of estimate of being calculated; And a plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using.
A kind of according to an embodiment of the invention program that is used for the control information treatment facility and makes the computing machine execution processing of messaging device, this messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, and this processing may further comprise the steps: generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols at random; To be input to each Characteristic Extraction expression formula of in the Characteristic Extraction expression list, comprising as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding with real data; Utilize a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the real data of instructing data supply generates target signature amount calculation expression with real data; The evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in the Characteristic Extraction expression list; A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using; And optimize each Characteristic Extraction expression formula in a plurality of Characteristic Extraction expression formulas that comprise in the generation Characteristic Extraction expression list in the end.
Utilize one embodiment of the present of invention, generation comprises the Characteristic Extraction expression list of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols, will be input to the individual features amount that comprises as the real data of instruction data supply and extract expression formula to calculate a plurality of characteristic quantities corresponding with real data in the Characteristic Extraction expression list.In addition, generate target signature amount calculation expression by machine learning, wherein utilize comparably a plurality of characteristic quantities corresponding calculated with real data and with as the corresponding existing characteristic quantity of the real data of instruction data supply with estimate with as the corresponding target signature amount of real data of instructing data supply.In addition, the individual features amount that calculating comprises in the Characteristic Extraction expression list is extracted the evaluation of estimate of expression formula, and a plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using.
Utilize one embodiment of the present of invention, generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random, to be input to the individual features amount that in the Characteristic Extraction expression list, comprises as the real data of instruction data supply and extract expression formula, and also measure the average computation time that the individual features amount is extracted expression formula with calculating a plurality of characteristic quantities corresponding with real data.In addition, generate target signature amount calculation expression by machine learning, wherein utilize a plurality of characteristic quantities corresponding calculated with real data with estimate with as the corresponding target signature amount of real data of instructing data supply, the individual features amount that calculating comprises in the Characteristic Extraction expression list is extracted the evaluation of estimate of expression formula, and also proofreaies and correct the evaluation of estimate of being calculated based on the average computation time of individual features amount extraction expression formula.In addition, a plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using.
Utilize one embodiment of the present of invention, generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random, will be input to the individual features amount that in the Characteristic Extraction expression list, comprises as the real data of instruction data supply and extract expression formula to calculate a plurality of characteristic quantities corresponding with real data.In addition, generate target signature amount calculation expression by machine learning, wherein utilize a plurality of characteristic quantities corresponding calculated estimating and the corresponding target signature amount of real data as the instruction data supply with real data, and the evaluation of estimate of the individual features amount extraction expression formula that in the Characteristic Extraction expression list, comprises of calculating.Subsequently, a plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of Characteristic Extraction expression formula and generate Characteristic Extraction expression list of future generation by using.In addition, optimize a plurality of Characteristic Extraction expression formulas that comprise in the generation Characteristic Extraction expression list in the end.
According to one embodiment of present invention, structural attitude amount computational algorithm automatically, thus even can utilize the existing characteristic quantity of input data to calculate the target signature amount corresponding with importing data.
According to one embodiment of present invention, structural attitude amount computational algorithm automatically thus can be by calculating the target signature amount corresponding with importing data to applying restriction computing time.
According to one embodiment of present invention, structural attitude amount computational algorithm can calculate the target signature amount corresponding with importing data thus irredundantly automatically.
Description of drawings
Fig. 1 is the figure that is used to describe the characteristic quantity calculation expression that generates by the target signature amount calculation expression tectonic system of having used one embodiment of the invention;
Fig. 2 is the figure that illustrates the data structure of instruction data;
Fig. 3 is the block diagram that the configuration example of the target signature amount calculation expression tectonic system of having used one embodiment of the invention is shown;
Fig. 4 is the figure that illustrates the example of Characteristic Extraction expression formula;
Fig. 5 is the figure that is used to describe the structure of Characteristic Extraction expression formula;
Fig. 6 is the figure that illustrates the example of Characteristic Extraction expression list;
Fig. 7 is the figure that is used to describe genetic algorithm;
Fig. 8 is the process flow diagram that is used to describe the operation of the target signature amount calculation expression tectonic system of having used one embodiment of the invention;
Fig. 9 is the process flow diagram that is used to specifically describe the S4 of step shown in Fig. 8;
Figure 10 is the figure that illustrates the example of option table group;
Figure 11 is the process flow diagram that is used to specifically describe the S10 of step shown in Fig. 8;
Figure 12 is the process flow diagram that is used to specifically describe the S42 of step shown in Figure 11; And
Figure 13 is the block diagram that illustrates the configuration example of computing machine.
Embodiment
Hereinafter be described particularly about using specific embodiments of the invention with reference to the accompanying drawings.
Target signature amount calculation expression tectonic system 10 (Fig. 3) that used one embodiment of the invention utilize a plurality of instruction data that will supply to generate target signature amount calculation expression 1 by machine learning, wherein as shown in fig. 1, obtain input data C and corresponding with it a plurality of existing characteristic quantity F1 c, F2 cUntil Fn cAs input, and output and relevant each characteristic quantity of importing among the corresponding a plurality of characteristic quantity I of data.
Fig. 2 shows the data structure of instruction data.That is to say instruction data T i(i=1,2 is until L) is by conduct and the real data D of input data C with a kind of data i, with real data D iCorresponding a plurality of existing characteristic quantity F1 iTo Fn i, and with real data D iCorresponding a plurality of target signature amount I1 iTo Ik iForm.
Existing characteristic quantity F1 iTo Fn iBe following value, these value representations use the existing method will be from real data D iThe middle real data D that detects iFeature.Target signature amount I1 iTo Ik iBe following value, these value representations use the existing method can't be from real data D iThe middle real data D that detects iFeature, for example by to making many people monitor real data D iAnd the impression that obtains is carried out the determined value of digitizing.
As shown in Figure 2, under the situation that k class target characteristic quantity is arranged, target signature amount calculation expression tectonic system 10 generates k target signature amount calculation expression.
Notice that the type of importing data C so is exactly arbitrarily as long as input data C is a multidimensional data.For example, have time dimension and sound channel dimension music data, have dimension X, dimension Y and pixel dimension view data, can be by add moving image data that time dimension obtains etc. to view data as input data C.
Note to describe in the following description and utilize music data as the example of importing data C.The example of a plurality of existing characteristic quantities corresponding with music data comprises rhythm, speed and rhythm fluctuation.The example of the target signature amount corresponding with music data also comprises the brightness of music data and the diversity of speed and musical instrument.
Fig. 3 illustrates the configuration example of the target signature amount calculation expression tectonic system 10 of using one embodiment of the invention.Target signature amount calculation expression tectonic system 10 is formed by following configuration of cells: Characteristic Extraction expression list generation unit 11 is used to generate and upgrade the Characteristic Extraction expression list of being made up of a plurality of Characteristic Extraction expression formulas; Feature amount calculation unit 12 is used for instruction data T iReal data D iThe individual features amount that substitution generated is extracted expression formula with the calculated characteristics amount; Target signature amount calculation expression generation unit 13, be used for by machine learning generate target signature amount calculation expression, thus can according to calculate by feature amount calculation unit 12 with instruction data T iCharacteristic of correspondence amount and according to instruction data T iExisting characteristic quantity F1 iTo Fn iEstimate to instruct data T iTarget signature amount I1 iTo Ik i, and the evaluation of estimate that also is used to calculate each Characteristic Extraction expression formula; And optimize unit 15, be used to optimize last Characteristic Extraction expression list and target signature amount calculation expression that has finally upgraded in generation.
Characteristic Extraction expression list generation unit 11 generates a plurality of Characteristic Extraction expression formulas of composition first generation Characteristic Extraction expression list at random and these expression formulas is outputed to feature amount calculation unit 12.
Be described about the Characteristic Extraction expression formula that Characteristic Extraction expression list generation unit 11 will generate now with reference to Fig. 4.Fig. 4 A to 4D illustrates Characteristic Extraction expression formula example respectively.
Utilize the Characteristic Extraction expression formula, on the left side is described the type of input data, and describes a class or multiclass operational symbol according to the order that will calculate on the right of the type of input data.Each operational symbol suitably comprises the axle that will handle and comprises parameter.
The example of operational symbol comprises average (Mean), fast Fourier transform (FFT), standard variance (StDev), occurrence rate (Ratio), low-pass filter (LPF), Hi-pass filter (HPF), absolute value (ABS), square (Sqr), square root (Sqrt), normalization (Normalize), differential (Differential), integration (Integrate), maximal value (MaxIndex), all deviations (UVariance) and down-sampling (DownSampling).Note fixing the axle that to handle according to the operational symbol of determining in some cases, therefore utilize in this case and the fixing axle that will handle of parameter.In addition, determining to be attended by under the operational symbol situation of parameter, this parameter also is confirmed as at random or the value that sets in advance.
For example under the situation of Characteristic Extraction expression formula shown in Fig. 4 A, 12TonesM is the input data, and 32#Differential, 32#MaxIndex, 16#LPF_1; 0.861 and the 16#UVariance operational symbol of respectively doing for oneself.In addition, expressions such as the 32# in corresponding operational symbol, 16# will handle the axle.
Now, 12TonesM represents along time shaft monophony PCM (pulse-code modulation sound source) Wave data to be carried out the music compartment analysis, and 48# represents the sound channel axle, and 32# represents frequency axis and music spacing shaft, and 16# express time axle.0.861 of operational symbol is the parameter handled of low-pass filter and the threshold value of the frequency that sent of expression for example.
The input data type that the individual features amount of attention composition first generation Characteristic Extraction expression list is extracted expression formula is identical with the type of input data C, determine the number of operational symbol and the type of operational symbol at random, but as shown in Figure 5, carry out in the temporal restriction that generates each Characteristic Extraction expression formula, wherein when carrying out the arithmetic operator corresponding successively with a plurality of operational symbols, the number that has dimension of arithmetical operation reduces successively, and the final arithmetic operation results of each Characteristic Extraction expression formula becomes the scalar multiple, and perhaps its dimension number becomes predetermined smaller value (for example 1,2 etc.).
Being appreciated that from example shown in Fig. 4 A to 4D that the characteristic quantity that utilizes the Characteristic Extraction expression formula to calculate does not become according to existing notion is defined as significant value, such as about the rhythm of music data, about the pixel histogram of view data etc.That is to say that in the time will importing data substitution Characteristic Extraction expression formula simply, the characteristic quantity that utilizes the Characteristic Extraction expression formula to calculate may be an arithmetic operation results.
Now as shown in Figure 6, suppose that the Characteristic Extraction expression list that Characteristic Extraction expression list generation unit 11 is generated is made up of m Characteristic Extraction expression formula f1 to fm.WavM as the input data of Characteristic Extraction expression formula f1 to fm is a monophony PCM Wave data, is time shaft and sound channel axle and have dimension.
To get back to description now to Fig. 3.Characteristic Extraction expression list generation unit 11 is by generating the second generation and Characteristic Extraction expression list afterwards thereof according to genetic algorithm (GA) renewal Characteristic Extraction expression list of last generation.
Now " genetic algorithm " mean a kind of be used to use select to handle, intersect handle, variation is handled and generate at random and handle the algorithm that generates gene of future generation from current generation gene.Particularly, the composition characteristic amount is extracted a plurality of individual features amounts of expression list and is extracted expression formulas as gene, carry out according to the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of forming current generation Characteristic Extraction expression list select to handle, intersect handle, variation is handled and generate at random and handle to generate Characteristic Extraction expression list of future generation.
That is to say that for example as shown in Figure 7, the Characteristic Extraction expression formula f2 that selects to have high evaluation value is handled in the selection that utilization is extracted expression formulas to a plurality of individual features amounts of forming current generation Characteristic Extraction expression list.The crossing processing that utilization is extracted expression formula to a plurality of individual features amounts of forming current generation Characteristic Extraction expression list makes a plurality of Characteristic Extraction expression formula f2 and f5 with high evaluation value intersect (combination) with generating feature amount extraction expression formula, and comprises this Characteristic Extraction expression formula in Characteristic Extraction expression list of future generation.
The variation that utilization is extracted expression formulas to a plurality of individual features amounts of forming current generation Characteristic Extraction expression list is handled partly Characteristic Extraction expression formula f2 that make a variation (changes) have high evaluation value with generating feature amount extraction expression formula, and comprises this Characteristic Extraction expression formula in Characteristic Extraction expression list of future generation.Utilize random operation to generate new Characteristic Extraction expression formula at random, and in Characteristic Extraction expression list of future generation, comprise the Characteristic Extraction expression formula that this is new.
To get back to description now to Fig. 3.The instruction data T that feature amount calculation unit 12 will be supplied iReal data D iThe substitution composition extracts expression formula f1 to fm to calculate about instruction data T from the individual features amount of the Characteristic Extraction expression list of Characteristic Extraction expression list generation unit 11 supplies iCharacteristic quantity, also measure the necessary computing time of calculating that the individual features amount is extracted expression formula f1 to fm, and by with different L bar real data D iEach Characteristic Extraction expression formula of substitution is carried out when calculating and is calculated the average computation time.The characteristic quantity that calculates and the average computation time of calculating are fed to target signature amount calculation expression generation unit 13.
As mentioned above, instruction data T iNumber be L, and the composition characteristic amount to extract the number of the Characteristic Extraction expression formula of expression list be m, thereby calculate (the individual characteristic quantity of L * m) in feature amount calculation unit 12.Hereinafter, by instructing data T iThe real data D of (i=1,2 is until L) iSubstitution Characteristic Extraction expression formula fj (j=1,2 is until m) and the characteristic quantity that calculates will be called fj[T i].
Whenever from feature amount calculation unit 12 supply (L * m) the individual characteristic quantity fj[T corresponding with current generation Characteristic Extraction expression list i] time, target signature amount calculation expression generation unit 13 utilizes (the individual characteristic quantity fj[T of L * m) as the result of calculation of feature amount calculation unit 12 i], at instruction data T iIn (the individual existing characteristic quantity F1 of L * n) that comprises iTo Fn i, at instruction data T iIn comprise L target signature amount I1 iGenerate at the target signature amount calculation expression shown in the following formula (1) by machine learning (utilizing the linearity of feature selecting to differentiate or recurrence), this expression formula for example by with the corresponding existing characteristic quantity F1 of input data C cTo Fn cAnd characteristic quantity f1[c] to fm[C] between linearity coupling generate target signature amount I1 c
Target signature amount I1 c=b 0+ b 1F1 c+ b 2F2 c+ ...+b nFn c+ b n+ 1F1[C]+b N+2F2[C]+...+b N+mFm[C] ... (1)
Attention in expression formula [1], b 0Be merogenesis, and b 1, b 2Until b N+mIt is linear coupling coefficient.In addition, at the target signature amount calculation expression generation unit 13 actual target signature amount calculation expressions that generate, do not utilize all existing characteristic quantity F1 cTo Fn cWith characteristic quantity f1[C] to fm[C] but utilize these characteristic quantities selectively.In this case, with the existing characteristic quantity F1 that does not utilize cTo Fn cWith characteristic quantity f1[C] to fm[C] corresponding linear coupling coefficient is set to zero.
Similarly, also generate target signature amount calculation expression, thus can be respectively by with the corresponding existing characteristic quantity F1 of input data C cTo Fn cWith characteristic quantity f1[C] to fm[C] between linearity coupling generate the target signature amount I2 corresponding with importing data cTo Ik c
Thereby, generate k target signature amount calculation expression at target signature amount calculation expression generation unit 13.
Subsequently, arrived under the situation of required degree of accuracy at the target signature amount calculation expression that generates, perhaps under the situation that predetermined instruction has been provided by the user, Characteristic Extraction expression list at this moment as last in generation the Characteristic Extraction expression list be fed to target signature amount calculation expression and optimize unit 15.
In addition, target signature amount calculation expression generation unit 13 utilizes built-in evaluation of estimate computing unit 14 to calculate the evaluation of estimate of the individual features amount extraction expression formula of the current generation Characteristic Extraction expression list of composition.That is to say, evaluation of estimate computing unit 14 calculates the contribution rate of k target signature amount calculation expressions each Characteristic Extraction expression formula separately, and will be defined as forming the evaluation of estimate of the individual features amount extraction expression formula of current generation Characteristic Extraction expression list by the total contribution rate that adds up to k contribution rate being calculated obtain.
Now with reference to following formula (2) the contribution rate computing method are described.Attention expression formula (2) is used X 1, X 2Until X N+mReplace existing characteristic quantity F1 cTo Fn cWith characteristic quantity f1[C] to fm[C].
Target signature amount I1 c=b 0+ b 1X 1+ b 2X 2+ ...+b N+mX N+m... (2)
Utilize following formula (3) to calculate X M(M=1,2 is until n+m) calculates target signature amount I1 to expression formula (2) cContribution rate (X M).
(X M)＝b M/StDev(X M)×StDev(I1)×Correl(X M，I1) ...(3)
Here, StDev (X M) representative has been used for the L X of machine learning M(s) standard variance.
StDev (I1) representative has been used for the instruction data T of machine learning iIn L target signature amount I1 comprising iStandard variance.
Correl (X M, I1) representative has been used for the L X of machine learning M(s) with instruction data T iIn L target signature amount I1 comprising iBetween Pearson's related coefficient.
Note as shown in following formula (4), by with L X M(s) with L target signature amount I1 iBetween covariance divided by L X M(s) standard variance and L target signature amount I1 iStandard variance between product calculate Pearson's related coefficient Correl (X M, I1).
Correl (X M, I1)=(X MAnd I1 iBetween covariance)/(X MStandard variance * I1 iStandard variance) ... (4)
Notice that evaluation of estimate computing unit 14 can extract the evaluation of estimate of expression formula f1 to fn based on the individual features amount that Pearson's related coefficient determine to form current generation Characteristic Extraction expression list, rather than as mentioned above based on characteristic quantity f1[as the output valve of the individual features amount extraction expression formula f1 to fm of target signature amount calculation expression] to fm[] contribution rate determine evaluation of estimate.
For example, can carry out following layout, wherein by the L bar is instructed data T iReal data D iSubstitution Characteristic Extraction expression formula f1 and calculate L characteristic quantity f1[D i] between Pearson's related coefficient, and calculate L bar instruction data T iK class target characteristic quantity I1 iTo ik i, and the mean value of k Pearson's related coefficient calculating is defined as the evaluation of estimate of Characteristic Extraction expression formula f1.
In addition, evaluation of estimate computing unit 14 can not only calculate the evaluation of estimate that the individual features amount is extracted the evaluation of estimate of expression formula f1 to fm but also calculated corresponding existing characteristic quantity F1 to Fm.
In addition, evaluation of estimate computing unit 14 is proofreaied and correct the evaluation of estimate of the individual features amount extraction expression formula f1 to fm of the current generation Characteristic Extraction of the composition expression list of determining like this based on the average computation time of extracting expression formula f1 to fm from the individual features amount of feature amount calculation unit 12 supplies.Particularly, evaluation of estimate computing unit 14 evaluation of estimate that is equal to, or greater than the Characteristic Extraction expression formula of predetermined threshold the average computation time is proofreaied and correct and for it minimum value of scope to be set.Subsequently, evaluation of estimate computing unit 14 is notified the evaluation of estimate of being proofreaied and correct to Characteristic Extraction expression list generation unit 11.
Proofread and correct according to such evaluation of estimate, can prevent that correlated characteristic amount that the average computation time is equal to, or greater than predetermined threshold from extracting that expression formula is continuous to pass to Characteristic Extraction expression list of future generation.Thereby, can reduce the calculated load of the next generation and feature amount calculation unit afterwards 12 thereof subsequently.Attention can be provided with the predetermined threshold that will compare with the average computation time automatically according to the computing power of feature amount calculation unit 12, and perhaps the user can be provided with this predetermined threshold arbitrarily.
Optimizing unit 15 holds with lower unit: Characteristic Extraction expression optimization unit 16 is used to optimize the Characteristic Extraction expression formula f1 to fm of the composition that extracts from target signature amount calculation expression generation unit 13 last Characteristic Extraction expression list in generation; And target signature amount calculation expression reconfiguration unit 17, be used to use the Characteristic Extraction expression formula f1 to fm of optimization to come reconstruct target signature amount calculation expression.
Characteristic Extraction expression optimization unit 16 extracts expression formula f1 to fm according to the individual features amount of forming last Characteristic Extraction expression list in generation and detects the combination (hereinafter referred to as " optimization pattern ") of the redundant arithmetical operation of registration in advance, and replace these redundant arithmetical operations with handling the little arithmetical operation of load, can obtain identical arithmetic operation results thus, optimize thereby carry out first.First example of optimizing hereinafter will be shown.
For two of calculating absolute value or the optimization pattern of multioperation symbol Abs are wherein arranged continuously, second and later operational symbol Abs be redundant, therefore by replace with an operational symbol Abs two or more multioperation accord with Abs and carry out optimization.
For expression two of regular arithmetical operation or the optimization pattern of multioperation symbol Normalize are wherein arranged continuously, second and later operational symbol Normalize be redundant, therefore by replace with an operational symbol Normalize two or more multioperation accord with Normalize and carry out optimization.
For operational symbol Sqr that the expression square operation is wherein arranged continuously and the optimization pattern that is used to calculate subduplicate operational symbol Sqrt, carry out optimization by replacing these two operational symbols, can obtain identical arithmetic operation results thus with the little operational symbol Abs of processing load.
For the optimization pattern of the operational symbol Differential that wherein has expression to differentiate continuously, carry out optimization by eliminating operational symbol Differential and Integrate, because they are unnecessary with the operational symbol Integrate of expression integral operation.
Notice that optimization pattern and optimization method thereof are not limited to above-mentioned example.
In addition, Characteristic Extraction expression optimization unit 16 utilizes genetic algorithm to carry out second optimization so that with obtaining identical result of calculation shorter computing time.
Target signature amount calculation expression reconfiguration unit 17 utilizes Characteristic Extraction expression formula f1 to fm and the instruction data optimized to come reconstruct target signature amount calculation expression by machine learning.
Then be described with reference to the operation of process flow diagram shown in Fig. 8 to target signature amount calculation expression tectonic system 10.
In step S1, Characteristic Extraction expression list generation unit 11 generates m Characteristic Extraction expression formula forming first generation Characteristic Extraction expression list at random, and will be fed to feature amount calculation unit 12 by the Characteristic Extraction expression list that m Characteristic Extraction expression formula formed.
In step S2, target signature amount calculation expression tectonic system 10 obtains instruction data T i(i=1,2 is until L).Gained instruction data Ti is fed to target signature amount calculation expression generation unit 13 and optimizes unit 15.
In step S3, feature amount calculation unit 12 will be instructed data T iIn the real data D that comprises iSubstitution is formed from the individual features amount of the Characteristic Extraction expression list of Characteristic Extraction expression list generation unit 11 supplies and is extracted expression formula f1 to fm to calculate (the individual characteristic quantity fj[T of L * m) i], also measure the necessary computing time of calculating that the individual features amount is extracted expression formula f1 to fm, and passing through different L bar real data D iEach Characteristic Extraction expression formula of substitution is carried out when calculating and is calculated the average computation time.Calculate with Characteristic Extraction expression formula f1 to fm each is self-corresponding (the individual characteristic quantity fj[T of L * m) i] and be fed to target signature amount calculation expression generation unit 13 average computing time.
In step S4, target signature amount calculation expression generation unit 13 by utilizing feature selecting the linearity discriminating or recurrence, according to (the individual characteristic quantity fj[T of L * m) as the result of calculation of feature amount calculation unit 12 i] learn to be used for to estimate instruction data T iIn L target signature amount I1 comprising iTarget signature amount calculation expression.
Processing (study is hereinafter referred to as handled) in step S4 is described particularly to target signature amount calculation expression generation unit 13 now with reference to process flow diagram shown in Fig. 9.
In step S21, when generating target signature amount calculation expression, target signature amount calculation expression generation unit 13 generates a plurality of option table TB at random, these option tables TB represent n existing characteristic quantity F1 to Fn and as the characteristic quantity f1[of the output of m Characteristic Extraction expression formula f1 to fm] to fm[] among use (selected) characteristic quantity and do not use (selecting) characteristic quantity, generate first generation option table group thus.A plurality of option table TB of forming the option table group as gene, are upgraded this option table group among the step S29 that describes hereinafter based on genetic algorithm.
Figure 10 illustrates the example of the option table group of being made up of a plurality of option table TB that will generate.Attention circle mark in Figure 10 is represented what selected and X mark was represented not select.
In step S22, target signature amount calculation expression generation unit 13 pay close attention to form in the corresponding option table TB of current generation option table group one next begin the circulation of option table group successively.Note the number (x in shown in Figure 10 example) of option table group round-robin multiplicity for the option table TB of composition option table group.
In step S23, target signature amount calculation expression generation unit 13 utilizes determining and the selected characteristic quantity fj[of option table TB that pay close attention to each self-corresponding average computation time of Characteristic Extraction expression formula f1 to fm from feature amount calculation unit 12 supplies] the characteristic of correspondence amount extracts the summation of the average computation time of expression formula fj and whether is equal to or less than predetermined threshold.The predetermined threshold that can will compare with the summation of average computation time according to the automatic setting of computing power of target signature amount calculation expression generation unit 13, perhaps the user can be provided with this predetermined threshold arbitrarily.
Be equal to or less than under the situation of predetermined threshold in the summation of determining the average computation time, this is handled and continues step S24.
In step S24, target signature amount calculation expression generation unit 13 utilizes at (the individual characteristic quantity fj[T of L * m) as the result of calculation of feature amount calculation unit 12 i] and at instruction data T iIn (the individual existing characteristic quantity F1 of L * n) that comprises iTo Fn iAmong select by the option table TB that pays close attention to characteristic quantity, differentiate or recurrence is learnt the target signature amount calculation expression that number equates with the number (k) of target signature amount type by linearity.
In step S25, target signature amount calculation expression generation unit 13 is calculated as the Akaike information standard (AIC) of the learning outcome handled among the step S24 evaluation of estimate of the option table TB that is paid close attention to.
Attention is determined in step S23 and the selected characteristic quantity fj[of option table Tb that pays close attention to] the characteristic of correspondence amount extracts under the situation of total greater than predetermined threshold of average computation time of expression formula fj, and this is handled and continues step S26.In step S26, the evaluation of estimate of the option table TB that target signature amount calculation expression generation unit 13 is paid close attention to is set to the minimum value that it is provided with scope.Therefore, the total that prevents the average computation time passes to the next generation greater than the option table of predetermined threshold is continuous, can prevent from thus to prolong to be used to calculate the target signature amount calculation expression time necessary that will generate.
After the evaluation of estimate of the option table TB that determines by the processing among step S25 or the step S26 to be paid close attention to, this is handled and continues step S27.In step S27, target signature amount calculation expression generation unit 13 determines to have paid close attention to all option table TB that form current generation option table group, and having under the option table TB situation of not paying close attention to as yet, this processing turns back to step S22, thus the processing among the repeating step S22 to S27.In step S27, paying close attention under all option table TB situations of forming current generation option table group subsequently, this is handled and continues step S28.
In step S28, target signature amount calculation expression generation unit 13 determines whether to have improved the evaluation of estimate of estimating the option table TB of fullest for predetermined some generations.Determining to have improved under the evaluation of estimate situation of the option table TB that estimates fullest subsequently, perhaps determining that the improvement since evaluation of estimate has stopped not passing by as yet to be scheduled under the situation in some generations, this is handled and continues step S29.
In step S29, target signature amount calculation expression generation unit 13 upgrades current generation option table group by use based on the genetic algorithm of the evaluation of estimate of each option table TB and generates option table group of future generation.This processing turns back to step S22, repeats subsequent treatment thus.
In step S28, improve under the evaluation of estimate situation of the option table TB that estimates fullests at predetermined some generations as yet definite subsequently, this handles step S5 shown in continuation Fig. 8.
Handle according to above-mentioned study, generated the target signature amount calculation expression that is used for calculating k class target characteristic quantity each the class target characteristic quantity corresponding with current generation Characteristic Extraction expression list.
Attention is for description mentioned above, is described under the hypothesis that genetic searching method and ACI are used to learn to handle, and handles but can utilize distinct methods to carry out study.In addition, can utilize Local Search rather than genetic algorithm to determine the selection of existing characteristic quantity or the output valve of choosing or Characteristic Extraction expression formula not.
For example, utilizing under the situation of Local Search, at all n existing characteristic quantity F1 to Fn with as the characteristic quantity f1[of the output of m Characteristic Extraction expression formula f1 to fm] to fm[] begin to learn under the situation of not choosing.Subsequently at n existing characteristic quantity F1 to Fn with as the characteristic quantity f1[of the output of m Characteristic Extraction expression formula f1 to fm] to fm[] in one characteristic quantity is selected and generate (n+m) individual option table under the situation that other does not select, and use AIC that each option table is carried out evaluation.Determine to estimate fullest subsequently, be the little option table of ACI value.In addition, utilize n existing characteristic quantity F1 to Fn and as the characteristic quantity f1[of the output of m Characteristic Extraction expression formula f1 to fm] to fm[] among a characteristic quantity selected and generate (n+m) individual option table under other not selected situation, and use AIC etc. that each option table is carried out evaluation.Preferably repeating above-mentioned processing stops until the improvement of evaluations such as AIC.
To get back to description now to Fig. 8.In step S5, the evaluation of estimate computing unit 13 of target signature amount calculation expression generation unit 13 calculate current generation k respective objects characteristic quantity calculation expression extract the characteristic quantity f1[of the result of calculation of expression formula f1 to fm as the individual features amount] to fm[] and contribution rate, and will be defined as forming the evaluation of estimate of the individual features amount extraction expression formula f1 to fm of current generation Characteristic Extraction expression list by the total contribution rate that adds up to k contribution rate being calculated obtain.
Attention in step S5, can based on Pearson's related coefficient rather than as mentioned above based target characteristic quantity calculation expression extract the characteristic quantity f1[of the output of expression formula f1 to fm as the individual features amount] to fm[] and contribution rate determine to form the evaluation of estimate of the individual features amount extraction expression formula f1 to fm of current generation Characteristic Extraction expression list.
In step S6, the individual features amount that evaluation of estimate computing unit 14 is proofreaied and correct the current generation Characteristic Extraction of the composition expression list of determining in the processing among step S5 based on the average computation time of extracting expression formula f1 to fm from the individual features amount of feature amount calculation unit 12 supply is extracted the evaluation of estimate of expression formula f1 to fm.Particularly, evaluation of estimate computing unit 14 evaluation of estimate that is equal to, or greater than the Characteristic Extraction expression formula of predetermined threshold the average computation time is proofreaied and correct and for it minimum value of scope to be set.Subsequently, evaluation of estimate computing unit 14 is notified the evaluation of estimate of being proofreaied and correct to Characteristic Extraction expression list generation unit 11.
In step S7, target signature amount calculation expression generation unit 13 determines whether the result of calculation of the target signature amount calculation expression of current generation has reached required degree of accuracy or whether the user has carried out end operation.Do not reach required degree of accuracy as yet and the user does not carry out under the situation of end operation as yet yet in definite result of calculation, this is handled and continues step S8.
In step S8, Characteristic Extraction expression list generation unit 11 generates Characteristic Extraction expression list of future generation by upgrade current generation Characteristic Extraction expression list according to genetic algorithm.Subsequently, this processing turns back to step S3, wherein repeating step S3 and the processing in the step afterwards thereof.
Subsequently, determine that in step S7 result of calculation from the target signature amount calculation expression of current generation has reached required degree of accuracy or the user has carried out under the situation of end operation, this is handled and continues step S9.
In step S9, target signature amount calculation expression generation unit 13 with the target signature amount calculation expression of current generation Characteristic Extraction expression list and current generation as last in generation the Characteristic Extraction expression list and k corresponding with it target signature amount calculation expression output to optimization unit 15.
In step S10, optimize unit 15 optimize from 13 inputs of target signature amount calculation expression generation unit last in generation the Characteristic Extraction expression list the individual features amount extract expression formula f1 to fm and also utilize the individual features amount of being optimized to extract expression formula f1 to fm and come reconstruct target signature amount calculation expression.
Be described particularly optimizing the processing of unit 15 in step S10 now with reference to process flow diagram shown in Figure 11.
In step S41, Characteristic Extraction expression optimization unit 16 extracts expression formula f1 to fm according to the individual features amount of forming last Characteristic Extraction expression list in generation and detects the optimization pattern, and carrying out following first optimizes, this first optimization is used for handling the little arithmetical operation replacement of load and handles the big arithmetical operation of loading, and can obtain identical arithmetic operation results thus.
In step S42, Characteristic Extraction expression optimization unit 16 is applied to individual features amount extraction expression formula f1 to fm so that to obtain identical result of calculation shorter computing time, carry out second thus and optimize with genetic algorithm after first optimizes.
Processing in step S42 is described particularly to Characteristic Extraction expression optimization unit 16 now with reference to process flow diagram shown in Figure 12.
In step S51, Characteristic Extraction expression optimization unit 16 after first optimizes pay close attention to the individual features amount of forming last Characteristic Extraction expression list in generation extract in the expression formula f1 to fm one next begin the Characteristic Extraction expression list successively and circulate.The attention characteristics amount is extracted expression list round-robin multiplicity is extracted the Characteristic Extraction expression formula f1 to fm of expression list for the composition characteristic amount number m.
In step S52, the part of the Characteristic Extraction expression formula fj that 16 variations of Characteristic Extraction expression optimization unit are paid close attention to is optimized candidate's expression formula fj ' to generate R r(r=1,2 is until R) and these expression formulas are defined as first generation optimization candidate expression formula group.
In step S53, the Characteristic Extraction expression formula fj that S bar evaluating data (these data have and input data C identical type) substitution is paid close attention in Characteristic Extraction expression optimization unit 16 is to calculate S characteristic quantity fj[].
In step S54, Characteristic Extraction expression optimization unit 16 pay close attention to form a current generation optimize R of candidate's expression formula group optimize in candidate's expression formula one next begin to optimize candidate's expression formula group successively and circulate.Noting optimizing candidate's expression formula group round-robin multiplicity is the number of the optimization candidate expression formula of compositional optimization candidate expression formula group.
In step S55, the optimization candidate expression formula fj ' that the S bar evaluating data substitution that utilizes among the step S53 is paid close attention in Characteristic Extraction expression optimization unit 16 rTo calculate S characteristic quantity fj ' r[], also meter evaluation time and calculating the average computation time when corresponding many evaluating datas of substitution.
In step S56, Characteristic Extraction expression optimization unit 16 calculates S the characteristic quantity fj[that has represented as result among the step S53] and as S characteristic quantity fj ' of result among the step S55 rPearson's related coefficient of a degree of correlation and definite S characteristic quantity fj[between []] and S characteristic quantity fj ' rWhether the degree of correlation is approximately 1.0 between [].Subsequently, at definite S characteristic quantity fj[] and S characteristic quantity fj ' rThe degree of correlation is approximately under 1.0 the situation between [], and this is handled and continues step S57.
In step S57, the evaluation of estimate of the optimization candidate expression formula fj ' that the inverse of the average computation time of calculating in the processing of Characteristic Extraction expression optimization unit 16 step S55 is set to pay close attention to.
S characteristic quantity fj[determined in attention in step S56] and S characteristic quantity fj ' rWhether the degree of correlation is approximately 1.0 between [].Subsequently, at definite S characteristic quantity fj[] and S characteristic quantity fj ' rThe degree of correlation is not approximately under 1.0 the situation between [], and this is handled and continues step S58.
In step S58, the evaluation of estimate of the optimization candidate expression formula fj ' that Characteristic Extraction expression optimization unit 16 is paid close attention to is set to the minimum value of its scope.
After the evaluation of estimate of the optimization candidate expression formula fj ' r that determines by the processing among step S57 or the step S58 to be paid close attention to, this is handled and continues step S69.In step S59, Characteristic Extraction expression optimization unit 16 determines whether to pay close attention to all optimization candidate expression formula fj ' r that form current generation optimization candidate expression formula group, and under the situation that the optimization candidate expression formula fj ' r that does not pay close attention to is as yet arranged, this processing turns back to step S54, wherein the processing among the repeating step S54 to S59.In step S69, under the situation of paying close attention to all optimization candidate expression formula fj ' r that form current generation optimization candidate expression formula group, this is handled and continues step S60 subsequently.
In step S60, Characteristic Extraction expression optimization unit 16 determines whether to have improved the evaluation of estimate of estimating the optimization candidate expression formula of fullest for predetermined some generations.Subsequently, under the situation of the evaluation of estimate of determining to have improved the optimization candidate expression formula of estimating fullest, perhaps determining that the improvement since evaluation of estimate has stopped not pass by as yet to be scheduled under the situation in some generations, this processing continuation step S61.
In step S61, candidate's expression formula fj ' is optimized by using based on each in Characteristic Extraction expression optimization unit 16 rThe genetic algorithm of evaluation of estimate upgrade current generation optimization candidate expression formula group and generate the candidate of optimization expression formula group of future generation.Yet note carrying out following layout, wherein optimize and comprise the optimization candidate expression formula of estimating fullest in candidate's expression formula group, and generate zone (R-1) individual optimization candidate expression formula by the Characteristic Extraction expression formula fj part separately that the last generation that variation has a fullest evaluation is optimized candidate's expression formula and concern the next generation.This processing turns back to step S54 to repeat subsequent treatment.
Subsequently, determine not improve under the situation of the optimization candidate expression formula of estimating fullest at predetermined some generations as yet in step S60, this is handled and continues step S62.
According to the processing among the step S52 to S60, one of all Characteristic Extraction expression formulas of forming last Characteristic Extraction expression list after first optimizes are subjected to second optimization in generation.
In step S62, all Characteristic Extraction expression formula f1 to fm of composition last Characteristic Extraction expression list are determined whether to have paid close attention in Characteristic Extraction expression optimization unit 16 in generation after first optimizes, and under the situation that the Characteristic Extraction expression formula fj that does not pay close attention to is as yet arranged, this processing turns back to step S51, wherein the processing among the repeating step S51 to S62.Subsequently in step S62, after first optimizes, paid close attention under the situation of all Characteristic Extraction expression formula f1 to fm that form last Characteristic Extraction expression list in generation, this means that all Characteristic Extraction expression formula f1 to fm that form last Characteristic Extraction expression list after first optimizes have been subjected to second and have optimized in generation, therefore should handle step S43 shown in continuation Figure 11.
To get back to description now to Figure 11.In step S43, target signature amount calculation expression reconfiguration unit 17 is based on the Characteristic Extraction expression formula f1 to fm that is subjected to second optimization and based on instructing data, coming k the target signature amount calculation expression of reconstruct from 13 supplies of target signature amount calculation expression generation unit by statistical study and machine learning.Subsequently, export final Characteristic Extraction expression list and target signature amount calculation expression, and the operation of target end characteristic quantity calculation expression tectonic system 10.
Explanation to the operation of target signature amount calculation expression tectonic system 10 is so far.
As mentioned above, according to the target signature amount calculation expression tectonic system 10 of using one embodiment of the invention, use existing characteristic quantity to construct target signature amount calculation expression automatically, compare the number of the Characteristic Extraction expression formula that can reduce the redundancy that often becomes thus with the situation that the result of calculation of only utilizing the characteristic quantity calculation expression is constructed target signature amount calculation expression.Therefore when calculating the target signature amount, can reduce the treatment capacity of arithmetical operation according to target signature amount calculation expression.
In addition, according to the target signature amount calculation expression tectonic system 10 of using one embodiment of the invention, prevent by genetic algorithm that the gene of the characteristic quantity calculation expression that computing time is of a specified duration is continuous and pass to the offspring, can construct the target signature amount calculation expression of arithmetical operation The limited time system thus.Also can shorten the structure time of target signature amount calculation expression.
In addition, according to the target signature amount calculation expression tectonic system 10 of using one embodiment of the invention, optimized the Characteristic Extraction expression formula of composition characteristic amount extraction expression list, when calculating the target signature amount, can reduce the treatment capacity of arithmetical operation thus according to target signature amount calculation expression.
Subsidiary carrying a bit, above-mentioned consecutive is handled and can be realized by hardware, also can realize by software.Realizing by software under the situation that consecutive is handled that the program of forming its software for example is installed to and can carries out the computing machine that embeds the specialized hardware of types of functionality or general purpose personal computer etc. by each class method is installed from program recorded medium.
Figure 13 is to use program to carry out the block diagram of the hardware configuration example of the computing machine that above-mentioned consecutive handles.
Utilize this computing machine 100, CPU (CPU (central processing unit)) 101, ROM (ROM (read-only memory)) 102 and RAM (random access memory) 103 interconnect by bus 104.
Input/output interface 105 is also connected to bus 104.Input/output interface 105 is connected with the driver 110 of the input block 106 that comprises keyboard, mouse, microphone etc., the output unit 107 that comprises display, loudspeaker etc., the storage unit 108 that comprises hard disk, nonvolatile memory etc., the communication unit 109 that comprises network interface and detachable recording medium 111 such as driving such as disk, CD, magneto-optic disk, semiconductor memory etc.
Utilize the computing machine of configuration like this, CPU 101 for example loads in the memory cell 108 program stored by input/output interface 105 and bus 104 and carries out this program in RAM 103, carry out above-mentioned consecutive thus and handle.
Notice that the performed program of computing machine can be wherein to carry out the program of handling according to the order of describing in this manual with time sequencing or can be wherein concurrently or according to carrying out the program of handling such as essential sequential such as when being called.
In addition, this program also can or can be handled with distribution mode by a plurality of computing machines by single Computer Processing.In addition, this program can be sent to remote computer and carries out.
In addition in this manual, the entire equipment that is formed by a plurality of equipment disposition represented in term " system ".
Notice that embodiments of the invention are not limited to the foregoing description, and can carry out various modifications and do not break away from essence of the present invention.
Those skilled in the art are to be understood that according to designing requirement at present and other factors can expect that various modifications, combination, secondary in the scope of claims and equivalence thereof make up and change.

Claims (19)

1. messaging device, be used to obtain input data and the existing characteristic quantity corresponding with described input data as input and generate the target signature amount calculation expression that is used to export with the corresponding target signature amount of described input data, described messaging device comprises:
Characteristic Extraction expression list generating apparatus, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using;
The characteristic quantity calculation element is configured to the real data as the instruction data supply is input to each Characteristic Extraction expression formula of comprising to calculate a plurality of characteristic quantities corresponding with described real data in described Characteristic Extraction expression list;
Target signature amount calculation expression generating apparatus, be configured to utilize comparably the described a plurality of characteristic quantities corresponding that calculated with described real data and with the corresponding existing characteristic quantity of described real data as the instruction data supply, by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression; And
The evaluation of estimate calculation element is configured to calculate the described evaluation of estimate of each the Characteristic Extraction expression formula that comprises in described Characteristic Extraction expression list.
2. messaging device according to claim 1, wherein said target signature amount calculation expression generating apparatus utilize comparably and selectively in described a plurality of characteristic quantities corresponding that calculated with described real data some characteristic quantities and with as some characteristic quantities in the corresponding a plurality of existing characteristic quantity of the described real data of instruction data supply, with by being used to estimate that the machine learning with as the corresponding described target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression.
3. messaging device according to claim 1, wherein said evaluation of estimate calculation element calculate the described evaluation of estimate of the described Characteristic Extraction expression formula that comprises in the described Characteristic Extraction expression list based on the contribution rate of each target signature amount calculation expression of described a plurality of characteristic quantities corresponding with described real data that calculated.
4. a messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, and described messaging device comprises:
Characteristic Extraction expression list generating apparatus, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using;
The characteristic quantity calculation element is configured to the real data as the instruction data supply is input to the average computation time of each Characteristic Extraction expression formula to calculate a plurality of characteristic quantities corresponding with described real data and also to measure described individual features amount extraction expression formula that comprises in described Characteristic Extraction expression list;
Target signature amount calculation expression generating apparatus is configured to utilize the described a plurality of characteristic quantities corresponding with described real data that calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instruction data supply generates described target signature amount calculation expression; And
The evaluation of estimate calculation element, the evaluation of estimate that is configured to calculate the described evaluation of estimate of each the Characteristic Extraction expression formula that in described Characteristic Extraction expression list, comprises and also proofreaies and correct described calculating based on the described average computation time of described individual features amount extraction expression formula.
5. messaging device according to claim 4, wherein said target signature amount calculation expression generating apparatus utilize some characteristic quantities in described a plurality of characteristic quantities corresponding with described real data that calculated with by being used to estimate to generate described target signature amount calculation expression with machine learning as the corresponding described target signature amount of the described real data of instruction data supply selectively.
6. messaging device according to claim 5, the described average computation time that wherein said target signature amount calculation expression generating apparatus extracts expression formula based on described character pair amount is utilized some characteristic quantities in described a plurality of characteristic quantities corresponding with described real data that calculated selectively, with by being used to estimate that the machine learning with as the corresponding described target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression.
7. a messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, and described messaging device comprises:
Characteristic Extraction expression list generating apparatus, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using;
The characteristic quantity calculation element is configured to the real data as the instruction data supply is input to each Characteristic Extraction expression formula of comprising to calculate a plurality of characteristic quantities corresponding with described real data in described Characteristic Extraction expression list;
Target signature amount calculation expression generating apparatus is configured to utilize the described a plurality of characteristic quantities corresponding with described real data that calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instruction data supply generates described target signature amount calculation expression;
The evaluation of estimate calculation element is configured to calculate the described evaluation of estimate of each the Characteristic Extraction expression formula that comprises in described Characteristic Extraction expression list; And
Optimization means is configured to optimize each the Characteristic Extraction expression formula in a plurality of Characteristic Extraction expression formulas that comprise in the described Characteristic Extraction expression list of a generation in the end.
8. messaging device according to claim 7, wherein said optimization means comprises:
Characteristic Extraction expression optimization device, be configured to: extract the Combination Optimized pattern that detects the redundant operation symbol of having represented registration in advance in the expression formula and by deleting operator or replacing with the less operational symbol of arithmetic load by the described individual features amount that from last generation described Characteristic Extraction expression list, comprises, optimize each the Characteristic Extraction expression formula that comprises in the described Characteristic Extraction expression list of a generation in the end.
9. messaging device according to claim 7, wherein said optimization means comprises:
Characteristic Extraction expression optimization device is configured to: be out of shape each Characteristic Extraction expression formula of comprising in the described Characteristic Extraction expression list of a generation in the end to generate a plurality of optimization candidate expression formulas; As gene, the following optimization candidate expression formula among the optimization candidate expression formula of described a plurality of generations is given good evaluation with the optimization candidate expression formula of described a plurality of generations: i.e. the output of this optimization candidate expression formula of Huo Deing has and is shorter as the computing time of the high degree of correlation of the output of the described Characteristic Extraction expression formula of deformation sources and this optimization candidate expression formula; Utilization based on the genetic algorithm of the evaluation of described optimization candidate expression formula to upgrade the optimization candidate expression formula of described a plurality of generations; And the described optimization candidate expression formula that will have the best evaluation described individual features amount that finally is defined as comprising in the described Characteristic Extraction expression list of a generation is in the end extracted the optimization result of expression formula.
10. messaging device according to claim 7, wherein said optimization means comprises:
Reconfiguration device is configured to utilize the described target signature amount calculation expression of the Characteristic Extraction expression formula of described optimization with the corresponding generation of described Characteristic Extraction expression list of reconstruct and last generation.
11. information processing method, be used to obtain input data and the existing characteristic quantity corresponding with described input data as input and generate the target signature amount calculation expression that is used to export with the corresponding target signature amount of described input data, described information processing method comprises:
Generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random;
To be input to each the Characteristic Extraction expression formula that in described Characteristic Extraction expression list, comprises as the real data of instruction data supply, to calculate a plurality of characteristic quantities corresponding with described real data;
Utilize comparably the described a plurality of characteristic quantities corresponding calculated with described real data and with the corresponding existing characteristic quantity of described real data as the instruction data supply, with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression;
The described evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in described Characteristic Extraction expression list; And
A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of the Characteristic Extraction expression formula that comprises generate described Characteristic Extraction expression list of future generation in Characteristic Extraction expression list of future generation by using.
12. an information processing method is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, described information processing method comprises:
Generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random;
To be input to the average computation time of each Characteristic Extraction expression formula that in described Characteristic Extraction expression list, comprises as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding and also to measure described individual features amount extraction expression formula with described real data;
Utilize the described a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression with described real data;
The described evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in described Characteristic Extraction expression list and the described average computation time of also extracting expression formula based on described individual features amount are proofreaied and correct the evaluation of estimate of described calculating; And
A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of the Characteristic Extraction expression formula that comprises generate described Characteristic Extraction expression list of future generation in Characteristic Extraction expression list of future generation by using.
13. an information processing method is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, described information processing method comprises:
Generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random;
To be input to each Characteristic Extraction expression formula of in described Characteristic Extraction expression list, comprising as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding with described real data;
Utilize the described a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression with described real data;
The described evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in described Characteristic Extraction expression list;
A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of the Characteristic Extraction expression formula that comprises generate described Characteristic Extraction expression list of future generation in Characteristic Extraction expression list of future generation by using; And
Optimize each the Characteristic Extraction expression formula in a plurality of Characteristic Extraction expression formulas that comprise in the described Characteristic Extraction expression list of a generation in the end.
14. one kind is used for the control information treatment facility and makes the computing machine of described messaging device carry out the program of handling, described messaging device is used to obtain input data and the existing characteristic quantity corresponding with described input data as input and generate the target signature amount calculation expression that is used to export with the corresponding target signature amount of described input data, and described processing may further comprise the steps:
Generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random;
To be input to each the Characteristic Extraction expression formula that in described Characteristic Extraction expression list, comprises as the real data of instruction data supply, to calculate a plurality of characteristic quantities corresponding with described real data;
Utilize comparably the described a plurality of characteristic quantities corresponding calculated with described real data and with the corresponding existing characteristic quantity of described real data as the instruction data supply, with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression;
The described evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in described Characteristic Extraction expression list; And
A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of the Characteristic Extraction expression formula that comprises generate described Characteristic Extraction expression list of future generation in Characteristic Extraction expression list of future generation by using.
15. one kind is used for the control information treatment facility and makes the computing machine of described messaging device carry out the program of handling, described messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, and described processing may further comprise the steps:
Generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random;
To be input to the average computation time of each Characteristic Extraction expression formula that in described Characteristic Extraction expression list, comprises as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding and also to measure described individual features amount extraction expression formula with described real data;
Utilize the described a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression with described real data;
The described evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in described Characteristic Extraction expression list and the described average computation time of also extracting expression formula based on described individual features amount are proofreaied and correct the evaluation of estimate of described calculating; And
A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of the Characteristic Extraction expression formula that comprises generate described Characteristic Extraction expression list of future generation in Characteristic Extraction expression list of future generation by using.
16. one kind is used for the control information treatment facility and makes the computing machine of described messaging device carry out the program of handling, described messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, and described processing may further comprise the steps:
Generate the Characteristic Extraction expression list that comprises a plurality of Characteristic Extraction expression formulas of forming by a plurality of operational symbols at random;
To be input to each Characteristic Extraction expression formula of in described Characteristic Extraction expression list, comprising as the real data of instruction data supply to calculate a plurality of characteristic quantities corresponding with described real data;
Utilize the described a plurality of characteristic quantities corresponding calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression with described real data;
The described evaluation of estimate of each Characteristic Extraction expression formula that calculating comprises in described Characteristic Extraction expression list;
A plurality of Characteristic Extraction expression formulas that will comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of the Characteristic Extraction expression formula that comprises generate described Characteristic Extraction expression list of future generation in Characteristic Extraction expression list of future generation by using; And
Optimize each the Characteristic Extraction expression formula in a plurality of Characteristic Extraction expression formulas that comprise in the described Characteristic Extraction expression list of a generation in the end.
17. messaging device, be used to obtain input data and the existing characteristic quantity corresponding with described input data as input and generate the target signature amount calculation expression that is used to export with the corresponding target signature amount of described input data, described messaging device comprises:
Characteristic Extraction expression list generation unit, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using;
Feature amount calculation unit is configured to the real data as the instruction data supply is input to each Characteristic Extraction expression formula of comprising to calculate a plurality of characteristic quantities corresponding with described real data in described Characteristic Extraction expression list;
Target signature amount calculation expression generation unit, be configured to utilize comparably the described a plurality of characteristic quantities corresponding that calculated with described real data and with the corresponding existing characteristic quantity of described real data as the instruction data supply, with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instructing data supply generates described target signature amount calculation expression; And
The evaluation of estimate computing unit is configured to calculate the described evaluation of estimate of each the Characteristic Extraction expression formula that comprises in described Characteristic Extraction expression list.
18. a messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, described messaging device comprises:
Characteristic Extraction expression list generation unit, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using;
Feature amount calculation unit is configured to the real data as the instruction data supply is input to the average computation time of each Characteristic Extraction expression formula to calculate a plurality of characteristic quantities corresponding with described real data and also to measure described individual features amount extraction expression formula that comprises in described Characteristic Extraction expression list;
Target signature amount calculation expression generation unit is configured to utilize the described a plurality of characteristic quantities corresponding with described real data that calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instruction data supply generates described target signature amount calculation expression; And
The evaluation of estimate computing unit, the evaluation of estimate that is configured to calculate the described evaluation of estimate of each the Characteristic Extraction expression formula that in described Characteristic Extraction expression list, comprises and also proofreaies and correct described calculating based on the described average computation time of described individual features amount extraction expression formula.
19. a messaging device is used to generate the target signature amount calculation expression that is used to export the target signature amount corresponding with importing data, described messaging device comprises:
Characteristic Extraction expression list generation unit, the a plurality of Characteristic Extraction expression formulas that are configured to comprise in last generation Characteristic Extraction expression list are as gene, upgrade described last generation Characteristic Extraction expression list based on the genetic algorithm of the evaluation of estimate of a plurality of Characteristic Extraction expression formulas of being made up of a plurality of operational symbols and generate the Characteristic Extraction expression list that comprises described a plurality of Characteristic Extraction expression formulas by using;
Feature amount calculation unit is configured to the real data as the instruction data supply is input to each Characteristic Extraction expression formula of comprising to calculate a plurality of characteristic quantities corresponding with described real data in described Characteristic Extraction expression list;
Target signature amount calculation expression generation unit is configured to utilize the described a plurality of characteristic quantities corresponding with described real data that calculated with by being used to estimate that the machine learning with as the corresponding target signature amount of the described real data of instruction data supply generates described target signature amount calculation expression;
The evaluation of estimate computing unit is configured to calculate the described evaluation of estimate of each the Characteristic Extraction expression formula that comprises in described Characteristic Extraction expression list; And
Optimize the unit, be configured to optimize each the Characteristic Extraction expression formula in a plurality of Characteristic Extraction expression formulas that comprise in the described Characteristic Extraction expression list of a generation in the end.
CN2008101702877A 2007-10-22 2008-10-20 Information processing device, information processing method Expired - Fee Related CN101419610B (en)

Applications Claiming Priority (9)

Application Number Priority Date Filing Date Title
JP2007273416 2007-10-22
JP2007-273418 2007-10-22
JP2007273418A JP4392622B2 (en) 2007-10-22 2007-10-22 Information processing apparatus, information processing method, and program
JP2007-273417 2007-10-22
JP2007273418 2007-10-22
JP2007-273416 2007-10-22
JP2007273417 2007-10-22
JP2007273417A JP4433323B2 (en) 2007-10-22 2007-10-22 Information processing apparatus, information processing method, and program
JP2007273416A JP4392621B2 (en) 2007-10-22 2007-10-22 Information processing apparatus, information processing method, and program

Publications (2)

Publication Number Publication Date
CN101419610A true CN101419610A (en) 2009-04-29
CN101419610B CN101419610B (en) 2011-11-16

Family

ID=40630402

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2008101702877A Expired - Fee Related CN101419610B (en) 2007-10-22 2008-10-20 Information processing device, information processing method

Country Status (2)

Country Link
JP (1) JP4433323B2 (en)
CN (1) CN101419610B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102339278A (en) * 2010-07-14 2012-02-01 索尼公司 Information processing device, information processing method, and program
CN103678436B (en) * 2012-09-18 2017-04-12 株式会社日立制作所 Information processing system and information processing method
CN106708875A (en) * 2015-11-16 2017-05-24 阿里巴巴集团控股有限公司 Characteristic screening method and system
CN106708609A (en) * 2015-11-16 2017-05-24 阿里巴巴集团控股有限公司 Characteristics generation method and system
CN112598109A (en) * 2019-09-17 2021-04-02 富士通株式会社 Information processing apparatus, non-transitory computer-readable storage medium, and information processing method

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7085370B2 (en) * 2017-03-16 2022-06-16 株式会社リコー Diagnostic equipment, diagnostic systems, diagnostic methods and programs

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102339278A (en) * 2010-07-14 2012-02-01 索尼公司 Information processing device, information processing method, and program
CN102339278B (en) * 2010-07-14 2016-01-20 索尼公司 Signal conditioning package and information processing method
CN103678436B (en) * 2012-09-18 2017-04-12 株式会社日立制作所 Information processing system and information processing method
CN106708875A (en) * 2015-11-16 2017-05-24 阿里巴巴集团控股有限公司 Characteristic screening method and system
CN106708609A (en) * 2015-11-16 2017-05-24 阿里巴巴集团控股有限公司 Characteristics generation method and system
CN106708609B (en) * 2015-11-16 2020-06-26 阿里巴巴集团控股有限公司 Feature generation method and system
CN112598109A (en) * 2019-09-17 2021-04-02 富士通株式会社 Information processing apparatus, non-transitory computer-readable storage medium, and information processing method

Also Published As

Publication number Publication date
CN101419610B (en) 2011-11-16
JP4433323B2 (en) 2010-03-17
JP2009104274A (en) 2009-05-14

Similar Documents

Publication Publication Date Title
CN101419610B (en) Information processing device, information processing method
Marinakis et al. Ant colony and particle swarm optimization for financial classification problems
KR102065801B1 (en) Data processing apparatus, data processing method, and program
Groll et al. LASSO-type penalization in the framework of generalized additive models for location, scale and shape
JP4392620B2 (en) Information processing device, information processing method, arithmetic device, arithmetic method, program, and recording medium
CN107133874A (en) A kind of quantization strategy generation system based on genetic planning
CN117391247A (en) Enterprise risk level prediction method and system based on deep learning
Sovilj et al. OPELM and OPKNN in long-term prediction of time series using projected input data
Mohapatra et al. Indian stock market prediction using differential evolutionary neural network model
JP4392621B2 (en) Information processing apparatus, information processing method, and program
Tam Cho An evolutionary algorithm for subset selection in causal inference models
CN116340726A (en) Energy economy big data cleaning method, system, equipment and storage medium
US8712936B2 (en) Information processing apparatus, information processing method, and program
US20230073973A1 (en) Deep learning based system and method for prediction of alternative polyadenylation site
JP4392622B2 (en) Information processing apparatus, information processing method, and program
US8131657B2 (en) Information processing device, information processing method, and program
CN115130924A (en) Microgrid power equipment asset evaluation method and system under source grid storage background
CN115471009A (en) Predictive optimized power system planning method
CN111027709B (en) Information recommendation method and device, server and storage medium
Chen et al. Application of a 3NN+ 1 based CBR system to segmentation of the notebook computers market
Lin Multiobjective fuzzy competence set expansion problem by multistage decision-based hybrid genetic algorithms
Lin Multicriteria–multistage planning for the optimal path selection using hybrid genetic algorithms
Junior et al. The effect analysis of crossover and selection methods on the performance of genclust++ algorithm
AU2021107109A4 (en) Method and system for intelligent bid price selection for profit accumulation using multiple software agents and machine learning model
Ao Automating stock prediction with neural network and evolutionary computation

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20111116

Termination date: 20181020

CF01 Termination of patent right due to non-payment of annual fee