JP6412767B2 - Noise generating apparatus, noise generating method and program - Google Patents

Description

The present invention relates to a noise generation device, a noise generation method, and a program.

By adding perturbation to the data to be analyzed, the data to be analyzed may be processed and kept secret. A technique of a noise generating device that perturbs data to be analyzed by generated noise is known (for example, see Patent Document 1).
Even if the data to be analyzed is perturbed by applying the generated noise, the data before perturbation may be readable from the data after perturbation unless sufficient perturbation is added to the data. There is a need to ensure appropriate confidentiality.

JP2013-83801A

However, Patent Document 1 discloses that one component data is targeted and the Pk-anonymity of the data is ensured, but a configuration that targets two or more component data is also suggested. Nor.
In view of the above problems, the present invention provides a noise generating device and a noise generating device capable of obtaining a highly accurate analysis result even in the case of multiple variables or a plurality of components while ensuring anonymity of data to be analyzed. Methods and programs are provided.

[1] The present invention has been made to solve the above-described problem, and the noise generator according to the first aspect of the present invention is a technique for perturbing data to conceal data to be analyzed. A conversion rule determining unit that determines a conversion rule that reduces a correlation between components that constitute data to be analyzed, a conversion unit that converts each data to be analyzed according to the conversion rule, and a predetermined value for each data after conversion And a perturbation unit for adding the perturbation.

[2] Further, the noise generating device according to the above aspect includes a distribution restoration unit that restores the distribution of the data to which the predetermined perturbation is added, and an inverse transformation corresponding to each transformation rule for each piece of the data that has been restored. And an inverse conversion unit for processing.

[3] In addition, the noise generator according to the second aspect of the present invention provides a distribution for restoring a distribution of data to which a predetermined perturbation is added in a technique for perturbing data to conceal data to be analyzed. A restoration unit, and a process corresponding to a conversion rule for reducing a correlation between components constituting data to be analyzed, and an inverse conversion unit that performs an inverse conversion process on the data converted according to the conversion rule. It is characterized by that.

[4] In addition, the distribution restoration unit of the noise generating device according to the aspect described above may be configured so that the distribution of data to which the predetermined perturbation is added approaches the distribution of data before the predetermined perturbation is added. It is characterized in that the data to which the perturbation is added is restored.

[5] Further, the conversion rule determining unit of the noise generating device according to the above aspect estimates a conversion matrix for converting the data to be analyzed according to the conversion rule.

[6] In addition, the perturbation unit of the noise generation device according to the above aspect generates perturbation data to which perturbation is added according to a distribution characteristic capable of adjusting the confidentiality of the data to be analyzed.

[7] In addition, the conversion rule determination unit of the noise generation device according to the above aspect determines the conversion rule according to a result of principal component analysis on the analysis target data.

[8] A noise generation method according to the third aspect of the present invention is a noise generation method in a noise generation device that perturbs data in order to conceal data to be analyzed , and the noise generation device includes: Determining a conversion rule for reducing the correlation between components constituting the analysis target data, converting each data to be analyzed by the conversion rule, adding a predetermined perturbation to each data after the conversion, It is characterized by executing a process including:

[9] Further, the program according to the fourth aspect of the present invention reduces the correlation between the components constituting the analysis target data in the computer of the noise generating device that perturbs the data to conceal the analysis target data. A program for executing a step of determining a conversion rule to be performed, a step of converting each data to be analyzed based on the conversion rule, and a step of adding a predetermined perturbation to each data after the conversion.
[10] A noise generation method according to the fifth aspect of the present invention is a noise generation method in which perturbation is added to data in order to conceal the data to be analyzed, and the computer is added with a predetermined perturbation. A step of restoring the distribution of data, and a process corresponding to a conversion rule for reducing a correlation between components constituting the data to be analyzed, the step of performing an inverse conversion process on the data converted according to the conversion rule And a process including
[11] A program according to a sixth aspect of the present invention includes a step of restoring a distribution of data to which a predetermined perturbation is added to a computer that perturbs data in order to conceal data to be analyzed, A program for executing a process corresponding to a conversion rule for reducing a correlation between components constituting data to be analyzed, and performing an inverse conversion process on the data converted according to the conversion rule. .

  According to the present invention, it is possible to obtain a highly accurate analysis result even in the case of multiple variables or a plurality of components while ensuring anonymity of data to be analyzed.

It is a block diagram which shows the noise generator 1 which concerns on embodiment of this invention. It is a schematic block diagram which shows the structure of the electronic control apparatus which concerns on this embodiment. It is a flowchart which shows the procedure of the process in the noise generator 1 which concerns on this embodiment. It is a block diagram which shows the noise generators 1A and 1B which concern on 2nd Embodiment. It is a flowchart which shows the procedure of the process of the noise generator 1A which concerns on this embodiment. It is a flowchart which shows the procedure of the process of the noise generator 1B which concerns on this embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[Outline of the present invention]
In the case of extracting useful information by analyzing a large amount of personal data, it is important to take measures against leakage of information for identifying an individual such as an individual's action history and purchase history.
Although there is an approach to prevent information leakage by restricting users who can access personal data, the risk of information leakage remains because an authorized user can access the personal data itself. As a method of reducing this risk, there is a method of applying noise to data within a range that does not affect statistical analysis results. According to this approach, the data to be handled is anonymized data by applying noise, and the personal data itself is not handled, so the effect of reducing the risk of data leakage can be expected.
However, when applying noise in this way, if noise with an independent distribution is applied to the multivariable distribution, the value of the correlation coefficient that represents the correlation included in the original data changes. End up. In short, when noise is applied independently to each variable of the multivariable, the correlation between the original variables is not preserved, and an accurate analysis result cannot be obtained for the correlation.
Therefore, the noise generation device described below determines a conversion rule that reduces the correlation between components constituting the data to be analyzed. The noise generator converts the correlation between the components of each data to be analyzed according to the conversion rule. The noise generator adds a predetermined perturbation to each data after the conversion. Hereinafter, an embodiment of a noise generator will be described.

[First Embodiment]
(Outline of noise generator 1)
FIG. 1 is a configuration diagram showing a noise generator 1 according to the present embodiment. The noise generator 1 shown in the figure includes an original data storage unit 10, a conversion rule determination unit 20, a perturbation unit 30, and a conversion data storage unit 40.

The original data storage unit 10 stores data to be analyzed which is original data.
FIG. 2 is an explanatory diagram illustrating an example of data to be analyzed by the noise generator 1 according to the present embodiment. As shown in this figure, the data to be analyzed is composed of a plurality of components. For example, data to be analyzed has data composed of d components as elements. Here, the data to be analyzed are collectively shown as a d-dimensional data vector x _k (k is a natural number equal to or less than n). Note that the number of data to be analyzed (data vector x _k ) is n (n is a natural number). Here, data to be analyzed (data vector x _k ) are collectively referred to as a data matrix X. This data matrix X is configured as n rows, and has n data vectors x corresponding to each row as elements.

  Returning to FIG. 1, the conversion data storage unit 40 stores data after adding a predetermined perturbation to the data to be analyzed which is the original data.

The conversion rule determination unit 20 calculates a conversion rule based on the correlation of the analysis target data that is the original data, and determines a conversion rule that reduces the correlation between the components that constitute the analysis target data. For example, the conversion rule determination unit 20 determines a conversion rule that reduces the correlation between components constituting the data to be analyzed according to the result of principal component analysis on the data to be analyzed. As a more specific example, the following explanation will be given by exemplifying a case where the conversion rule determination unit 20 uses the invariant matrix (eigenvector) of the covariance matrix of the data matrix X as an axis.
The conversion rule determination unit 20 includes a covariance matrix calculation unit 21 and a conversion matrix calculation unit 23. The covariance matrix calculation unit 21 calculates the covariance matrix Σ of the data matrix X to be analyzed. The transformation matrix calculation unit 23 calculates d eigenvectors a _i by solving the eigenvalue problem defined by Equation (6) for the covariance matrix Σ calculated by the covariance matrix calculation unit 21. Further, according to the conversion rule, a conversion matrix A having eigenvectors a _i as elements of each row is calculated. This transformation matrix A is a dxd matrix.

The perturbation unit 30 adds a predetermined perturbation to the data to be analyzed that is the original data. The perturbation unit 30 includes a conversion unit 31, a random number application unit 33, a distribution restoration unit 35, and an inverse conversion unit 37.
The conversion unit 31 converts each data to be analyzed according to the conversion rule determined by the conversion rule determination unit 20. In the present embodiment, the correlation of each data is stored in the conversion rule by using the conversion rule for converting into a space in which the correlation between data is reduced.
The random number application unit 33 (perturbation unit) adds a predetermined perturbation to each data after being converted by the conversion unit 31. A rule for determining a predetermined amount of perturbation to be added to each data is predetermined. For example, the amount of predetermined perturbation added to each data is determined in advance according to the confidentiality of the data to be analyzed. For example, the amount is calculated in advance according to Pk-anonymity. In accordance with the calculation result of the amount of perturbation, the random number application unit 33 generates perturbation data to which perturbation is added according to a distribution characteristic (Laplace distribution) that can adjust the confidentiality of the data to be analyzed.
The distribution restoring unit 35 converts the data to which the predetermined perturbation is added so that the distribution of the data to which the predetermined perturbation is added approaches the distribution of the data before the predetermined perturbation is added.
The inverse conversion unit 37 performs an inverse conversion process corresponding to the conversion rule determined by the conversion rule determination unit 20. For example, the inverse conversion unit 37 performs an inverse conversion process corresponding to the above-described conversion rule on the data whose distribution is restored by the distribution restoration unit 35 after the perturbation is applied by the random number application unit 33 (perturbation unit). .

(About the principle of adding perturbations to the original data while ensuring confidentiality)
When a d-dimensional data matrix X is given, an operation of perturbing with noise according to a Laplace distribution whose distribution probability is given by Pdist (x) is called Pk anonymization. Here, the data matrix X is defined as in Expression (1) having a data vector x _k (k is a natural number up to n) as an element. In the following description, the description of “→” indicating a vector may be omitted. It is known that the Laplace distribution Pdist (x) forms an exponential distribution expressed as Equation (2).

  In the above equation (2), in the case of one variable, the relationship between the anonymity index k and ψ is strictly given, and k that is the anonymity index can be controlled by using ψ. For details, refer to a reference document (Japanese Patent Laid-Open No. 2013-83801).

  However, in the case of multiple variables or a plurality of components, since it depends on the distribution characteristics of the sample values of the variables, the relationship between the anonymity index k and ψ cannot be strictly given. For example, in the case of the multivariable Gaussian distribution PG shown in Equation (3), this equation has a cross term.

  On the other hand, in the case of the multi-variable version PL of the Laplace distribution shown in Expression (4), the expression does not have an intersection term.

  When noise generated according to the distribution characteristics of a unimodal distribution that does not have a correlation structure as described above is applied to perturb data, the applied noise destroys the covariance structure that the data has. End up. Therefore, there arises a problem that the data after noise is applied exhibits different distribution characteristics from the original data.

  For example, height and weight have a strong positive correlation, and there is a tendency that the higher the height, the heavier the weight. However, noise based on the applied Laplace distribution does not have such a tendency. For this reason, the more noise is applied, the weaker the positive correlation that “the higher the height, the heavier the weight”.

Therefore, as a method for avoiding the above problem, a method of performing Pk anonymization on data whose correlation has been reduced (or eliminated) by projecting onto a space in which the correlation between elements constituting the data is reduced Is illustrated. The most common method used to eliminate correlations between data is principal component analysis.
For example, for the data vector x that is an element of the data matrix X, the variance-covariance matrix Σ is calculated by the arithmetic expression shown in Expression (5).

  In the above equation (5), the vector μ is an average value of all the data vectors x. By solving Equation (6) of the eigenvalue problem of the variance-covariance matrix Σ, a transformation matrix A shown in Equation (7) is obtained.

  The data vector x is converted into the vector ξ by the equation (8) using the conversion matrix A.

  A variable vector ξ in which the correlation is reduced (or eliminated) is obtained by the conversion according to the equation (8). Since the correlation between the variables is reduced (or absent) in the vector ξ 1, each variable can be handled as an independent variable.

  In order to anonymize the vector ξ, as shown in the equations (9) and (10), noise based on the Laplace distribution corresponding to each component may be applied.

The vector ξdist obtained after applying noise is different from the distribution of the original data in the distribution state, and thus cannot be analyzed accurately. Therefore, the distribution of the vector ξdist is restored so that the distribution of the vector ξdist approaches the same distribution as the original data distribution before applying noise. For example, a general method such as an iterative Bayesian restoration method can be used to restore the distribution of the vector ξdist. The restored data is written as a vector ξ ^† .

Since the vector ξdist shown in the above equation (10) and the vector ξ ^† restored by Bayes reconstruction are converted into a space in which the correlation between the data is reduced, each data constituting the vector ξdist has no meaning. And it is anonymized by the above processing and cannot be interpreted even if compared with the original data. Therefore, a vector xdist is created by the equation (11) using the inverse matrix A- ¹ of the transformation matrix A, and returned from the vectors ξ, ξdist, ξ ^† to the space of the vector x where each data element is meaningful This is anonymized public data.

As shown in the above equation (11), the vector xdist is data generated based on the vector ξ ^† in which only the distribution is restored from the vector to which noise is applied.

(Processing procedure in the noise generator 1)
With reference to FIG. 3, the process in the noise generator 1 is demonstrated. FIG. 3 is a flowchart showing a processing procedure in the noise generator 1 according to the present embodiment.

The conversion rule determination unit 20 determines a conversion rule for data to be analyzed stored in the original data storage unit 10 (step S20).
The conversion unit 31 performs conversion processing according to the conversion rule on the data to be analyzed stored in the original data storage unit 10 (step S31).
The random number application unit 33 applies perturbation to the converted data (step S33).
The distribution restoration unit 35 restores the data distribution after the perturbation is applied (step S35).
The inverse conversion unit 37 performs an inverse conversion process corresponding to the conversion rule determined by the conversion rule determination unit 20, and writes and stores the data after the inverse conversion process in the conversion data storage unit 40 (step S37).

  As shown in the present embodiment, the noise generator 1 adds perturbation to the analysis target data, but can suppress a decrease in analysis accuracy.

[Second Embodiment]
(Outline of noise generators 1A and 1B)
FIG. 4 is a configuration diagram showing the noise generators 1A and 1B according to the present embodiment. The same reference numerals are given to the same components as those in FIG. Hereinafter, the noise generators 1A and 1B will be described focusing on differences from the noise generator 1 shown in FIG.

A noise generating device 1A shown in FIG. 4 includes an original data storage unit 10, a conversion rule determination unit 20A, a perturbation unit 30A, and a perturbation data storage unit 50.
The conversion rule determination unit 20A in the noise generator 1A calculates a conversion rule based on the correlation of the analysis target data that is the original data, and determines a conversion rule that reduces the correlation between the components constituting the analysis target data. To do. The conversion rule determination unit 20A can make the configuration related to the determination of the conversion rule the same as the configuration of the conversion rule determination unit 20 (FIG. 1). The conversion rule determination unit 20A supplies the determined conversion rule to the perturbation unit 30A and the perturbation unit 30B of the noise generator 1B.
The perturbation data storage unit 50 stores data obtained by adding a predetermined perturbation to the data to be analyzed which is the original data.
The perturbation unit 30A adds a predetermined perturbation to the analysis target data that is the original data. The perturbation unit 30A includes a conversion unit 31 and a random number application unit 33A. The random number application unit 33A converts the correlation between the components of each data to be analyzed by the conversion rule determined by the conversion rule determination unit 20A by the same method as the random number application unit 33 described above, and converts the converted data to And stored in the perturbation data storage unit 50.

The noise generator 1B shown in the figure includes a perturbation unit 30B and a conversion data storage unit 40.
The perturbation unit 30B in the noise generator 1B restores the distribution of data after applying a predetermined perturbation, and reversely transforms the distribution after restoring the distribution. The perturbation unit 30B obtains data obtained by adding a predetermined perturbation to the data to be analyzed which is the original data by the above processing.
The perturbation unit 30B includes a distribution restoration unit 35B and an inverse conversion unit 37B.
The distribution restoration unit 35B refers to the conversion data storage unit 40B, reads the stored data, and converts the data added with the predetermined perturbation by the random number application unit 33 into the data of the correlation reduction space.
The inverse conversion unit 37B performs an inverse conversion process corresponding to the conversion rule determined by the conversion rule determination unit 20A on the data converted into the data of the correlation reduction space by the distribution restoration unit 35B. The inverse conversion unit 37B stores the data subjected to the above-described inverse conversion process in the conversion data storage unit 40.

(Processing procedure in the noise generators 1A and 1B)
With reference to FIG. 5 and FIG. 6, the processing in the noise generators 1A and 1B will be described.
FIG. 5 is a flowchart showing a processing procedure of the noise generator 1A according to the present embodiment. FIG. 6 is a flowchart showing a processing procedure of the noise generator 1B according to the present embodiment.

First, as illustrated in FIG. 5, the noise generation device 1 </ b> A performs the following process of applying a predetermined perturbation to data to be analyzed. The conversion rule determination unit 20 determines a conversion rule for data to be analyzed stored in the original data storage unit 10 (step S20).
The conversion unit 31 performs conversion processing according to the conversion rule on the data to be analyzed stored in the original data storage unit 10 (step S31).
The random number application unit 33 applies a perturbation to the converted data, and stores the data after the perturbation is applied in the perturbation data storage unit 50 (step S33A).

Next, as illustrated in FIG. 6, the noise generation device 1 </ b> B performs the following processing on the data after a predetermined perturbation is applied to the data to be analyzed by the noise generation device 1 </ b> A.
The distribution restoration unit 35B refers to the perturbation data storage unit 50 and acquires data stored in the perturbation data storage unit 50, which is data after the perturbation is applied in the noise generating device 1A. The distribution restoration unit 35B performs a process of restoring the data distribution after the perturbation is applied (step S35B).
The inverse conversion unit 37B acquires the conversion rule determined by the conversion rule determination unit 20 from the conversion rule determination unit 20A of the conversion rule determination unit 20. The inverse conversion unit 37B performs an inverse conversion process corresponding to the conversion rule acquired from the conversion rule determination unit 20A, and writes and stores the data after the inverse conversion process in the conversion data storage unit 40B (step S37B).

As shown in this embodiment, the noise generators 1A and 1B ensure anonymity by adding perturbations after converting the analysis target data into a space in which the correlation between the components constituting the data is reduced. A decrease in analysis accuracy can be suppressed.
The noise generators 1A and 1B perform the processing separately from the above-described processing from step S20 to step S33A and the processing from step S35B to step S37B, so that the distribution before restoration and inverse transformation are performed. It becomes easy to use data. In order to reversely convert this data, it is necessary to perform reverse conversion according to the conversion rule determined by the conversion rule determination unit 20, in other words, the data before reverse conversion is encrypted using the conversion rule. Can be considered to be By using the encrypted data at the time of distribution, the data can be distributed while ensuring the confidentiality of the data.

  The noise generators 1A and 1B perform the processing separately from the above-described processing from step S20 to step S33A and the processing from step S35B to step S37B, so that the distribution before restoration and inverse transformation are performed. It becomes easy to use data. In order to reversely convert this data, it is necessary to perform reverse conversion according to the conversion rule determined by the conversion rule determination unit 20, in other words, the data before reverse conversion is encrypted using the conversion rule. Can be considered to be By using the encrypted data at the time of distribution, the data can be distributed while ensuring the confidentiality of the data.

Further, the embodiment of the present invention is not limited to the above-described example, and can take various forms within the scope of the present invention.
For example, instead of using principal component analysis for correlation extraction, L1-PCA (PCA based on L1-norm maximization) may be used. In that case, the transformation matrix can be obtained using a method such as Newton's method.

  The random number application unit 33 may acquire data from the storage unit in the same manner as the conversion rule determination unit 20 instead of receiving the perturbation target data directly from the conversion rule determination unit 20.

  In addition, the random number application unit 33 has been described as performing the process of perturbing the data sequentially for each piece of data. However, according to the conversion rule determined by the conversion rule determination unit 20, all the data included in the data group May be processed in accordance with a procedure of collectively converting all the converted data, applying noise to all the converted data, and collectively performing reverse conversion on all the data to which the noise is applied.

  The noise generating device 1A outputs a conversion matrix A as a conversion rule, but uses a conversion matrix obtained from a perturbed data group without outputting the conversion matrix A instead of the conversion matrix A. You may do it. After converting to a space that reduces the correlation between variables, when adding a perturbation that is independent of other variables (does not affect other variables), from the covariance matrix of the data group before / after the perturbation The required transformation matrix is approximate. Therefore, a transformation matrix obtained from the covariance matrix of the data group after perturbation can be used instead of the transformation matrix A obtained from the covariance matrix of the data group before perturbation.

  In addition, the original data has no data of some elements, and becomes a singular point in the data group. Such a singular point is preferably supplemented with data for elements that do not have data before adding the above perturbation. For example, there is a case where there is no data of some elements, a case where there is no answer to a question, and a case where measurement of a specific measurement item among a plurality of measurement items is omitted. In such a case, the data may be supplemented by applying a standard value or an average value derived according to the correlation.

  Note that a message display control is performed by recording a program for realizing the function of the processing unit in the present invention on a computer-readable recording medium, causing the computer system to read and execute the program recorded on the recording medium. May be performed. Here, “loading and executing a program recorded on a recording medium into a computer system” includes installing the program in the computer system. The “computer system” here includes an OS and hardware such as peripheral devices. Further, the “computer system” may include a plurality of computer devices connected via a network including a communication line such as the Internet, WAN, LAN, and dedicated line. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. As described above, the recording medium storing the program may be a non-transitory recording medium such as a CD-ROM. The recording medium also includes a recording medium provided inside or outside that is accessible from the distribution server in order to distribute the program. The code of the program stored in the recording medium of the distribution server may be different from the code of the program that can be executed by the noise generator 1 (1A, 1B). That is, the format stored in the distribution server is not limited as long as it can be downloaded from the distribution server and installed in a form that can be executed by the noise generation device 1 (1A, 1B). It should be noted that the program may be divided into a plurality of parts and downloaded at different timings and then combined in the noise generator 1 (1A, 1B), or the distribution server that distributes each of the divided programs may be different. Furthermore, the “computer-readable recording medium” holds a program for a certain period of time, such as a volatile memory (RAM) inside a computer system that becomes a server or a client when the program is transmitted via a network. Including things. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

  In addition, some or all of the functions described above may be realized as an integrated circuit such as an LSI (Large Scale Integration). Each function described above may be individually made into a processor, or a part or all of them may be integrated into a processor. Further, the method of circuit integration is not limited to LSI, and may be realized by a dedicated circuit or a general-purpose processor. In addition, when an integrated circuit technology that replaces LSI appears due to the advancement of semiconductor technology, an integrated circuit based on the technology may be used.

DESCRIPTION OF SYMBOLS 1, 1A, 1B Noise generator, 10 Original data storage part, 20 Conversion rule determination part, 30 Perturbation part, 31 Conversion part, 33 Random number application part, 35 Distribution restoration part, 37 Inverse conversion part, 40 Conversion data storage part

Claims (11)

In the technology that perturbs data to keep the data to be analyzed confidential,
A conversion rule determining unit that determines a conversion rule for reducing the correlation between components constituting the data to be analyzed;
A conversion unit that converts each data to be analyzed according to the conversion rule;
A perturbation unit for applying a predetermined perturbation to each data after the conversion;
A noise generating device comprising: A distribution restoration unit for restoring the distribution of the data to which the predetermined perturbation is added;
The noise generating apparatus according to claim 1, further comprising: an inverse conversion unit that performs an inverse conversion process corresponding to the conversion rule on each data whose distribution is restored. In the technology that perturbs data to keep the data to be analyzed confidential,
A distribution restoration unit for restoring the distribution of data to which a predetermined perturbation is added;
A process corresponding to a conversion rule for reducing a correlation between components constituting data to be analyzed, comprising: an inverse conversion unit for performing an inverse conversion process on the data converted according to the conversion rule; Noise generator. The distribution restoration unit
The data to which the predetermined perturbation is added is restored so that the distribution of the data to which the predetermined perturbation is added approaches the distribution of the data before adding the predetermined perturbation. 3. The noise generator according to 3. The conversion rule determining unit
In accordance with the conversion rule, noise generating apparatus according to claim 1 or 2, characterized in that estimating a transformation matrix for converting the data of the analysis target. The perturbation part is
The noise generating apparatus according to claim 1, wherein perturbation data to which perturbation is added according to a distribution characteristic capable of adjusting the secrecy of the analysis target data is generated. The conversion rule determining unit
The noise generation device according to claim 1 or 2, wherein the conversion rule is determined according to a result of principal component analysis on the data to be analyzed. The data to be analyzed to a noise generation method in a noise generating device that perturb the data in order to conceal,
The noise generator is
Determining a transformation rule that reduces the correlation between the components comprising the data to be analyzed;
Converting each data to be analyzed according to conversion rules;
Adding a predetermined perturbation to each converted data;
A noise generation method characterized by executing a process including: To the computer of the noise generator that perturbs the data to keep the data to be analyzed confidential,
Determining a transformation rule that reduces the correlation between the components comprising the data to be analyzed;
Converting each data to be analyzed according to conversion rules;
Adding a predetermined perturbation to each converted data;
A program for running A noise generation method that adds perturbation to data in order to conceal the data to be analyzed,
Computer
Restoring a distribution of data to which a predetermined perturbation is added;
A process corresponding to a conversion rule for reducing a correlation between components constituting data to be analyzed, the step of performing an inverse conversion process on the data after being converted according to the conversion rule;
A noise generation method characterized by executing a process including: To a computer that perturbs the data to keep the data under analysis confidential
Restoring a distribution of data to which a predetermined perturbation is added;
A process corresponding to a conversion rule for reducing a correlation between components constituting data to be analyzed, the step of performing an inverse conversion process on the data after being converted according to the conversion rule;
A program for running

Images