Machine learning model provision system and machine learning model provision method

Abstract

To propose a machine learning model providing system capable of preventing a machine learning model with insufficient personal information protection from being provided while providing a machine learning model utilizing personal information data without disclosing the personal information data to users.SOLUTION: A machine learning model providing system according to the present disclosure includes a storage device for storing a collection of personal information data and a data catalog, and an information processing apparatus. The information processing apparatus executes: processing of designating metadata to be used for generating a machine learning model, designating a first data range, and obtaining a machine learning logic; processing of generating the machine learning model according to the logic, based on personal information data corresponding to the designated metadata and first data range; processing of calculating a specific risk concerning the generated machine learning model; and processing of suspending provision of the generated machine learning model when the specific risk becomes equal to or greater than a prescribed threshold.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] The present disclosure relates to a machine learning model providing system and a machine learning model providing method that generate a machine learning model and provide the generated machine learning model. [Background technology]

[0002] Patent Document 1 discloses a privacy-preserving data providing system that obtains an anonymous model by performing an anonymization process on a deep learning model obtained by applying a deep learning algorithm. This privacy-preserving data providing system provides an error based on a Laplace distribution to each parameter value for a weight parameter and a bias parameter included in a deep learning model, and limits each parameter to a threshold range indicated by a maximum value and a minimum value when the parameter value to which an error based on the Laplace distribution is provided exceeds the threshold range. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2018-97467 A Summary of the Invention [Problem to be solved by the invention]

[0004] As the use of various types of data advances with the development of machine learning technology, the usefulness of using personal information data is attracting attention. In the future, it is expected that there will be an increasing number of cases where the data collector and the user are different, such as in smart cities, and it is expected that there will be an increasing demand for users to use the personal information data collected by data collectors.

[0005] However, from the viewpoint of personal information protection, there is a risk of personal information data being leaked and it is difficult to provide it to a third party, so personal information data is limited to use within a limited organization or as statistical information. As such, the use of personal information data in machine learning has not progressed so far.

[0006] The privacy protection data providing system disclosed in Patent Document 1 generates an anonymous model (machine learning model) that has been subjected to a predetermined anonymization process. However, since the generated machine learning model is not judged in accordance with personal information protection, there is a risk that a machine learning model that does not adequately protect personal information may be provided.

[0007] The present disclosure has been made in consideration of the above-mentioned problems, and aims to provide a machine learning model providing system and a machine learning model providing method that can provide a machine learning model that uses personal information data without disclosing the personal information data to users, while preventing the provision of a machine learning model that does not adequately protect personal information. [Means for solving the problem]

[0008] The first disclosure relates to a machine learning model providing system that generates a machine learning model and provides the generated machine learning model. This machine learning model providing system includes a storage device that stores a collection of personal information data and a data catalog for metadata related to the personal information data, and an information processing device. The information processing device executes an acquisition process for specifying the metadata to be used in generating the machine learning model, specifying a first data range that provides a data range related to the personal information, and acquiring machine learning logic, a model generation process for generating the machine learning model according to the logic based on the specified metadata and the personal information data that falls within the first data range, an identification risk calculation process for calculating a specific risk for the generated machine learning model that indicates the risk that an individual may be identified from the output, and a provision management process for stopping provision of the generated machine learning model if the identification risk is equal to or greater than a predetermined threshold.

[0009] The second disclosure relates to a machine learning model providing system that further includes the following features in relation to the machine learning model providing system according to the first disclosure. Each of the personal information data is associated with one or more pieces of ID information that indicate a specific individual. In the specific risk calculation process, the information processing device selects the personal information data to be input data to the generated machine learning model, obtains output data that is the output of the machine learning model for the input data, generates associated output information indicating the correspondence between the ID information and a set of output data related to the input data associated with the ID information, and calculates the specific risk based on the associated output information.

[0010] The third disclosure relates to a machine learning model providing system that further includes the following features with respect to the machine learning model providing system according to the second disclosure. The output data is divided into categories. In the specific risk calculation process, the information processing device calculates, for each combination (i, j) of the ID information and the category, the number of elements u(i, j) of the set of output data included in the category and corresponding to the ID information in the related output information, and calculates the specific risk IR using the following equations (1) and (2).

number

number

[0011] The fourth disclosure relates to a machine learning model providing system further including the following features with respect to the machine learning model providing system according to the second or third disclosure. In the specific risk calculation process, the information processing device generates a second data range that gives the data range, and selects the personal information data that falls within the second data range as the input data.

[0012] The fifth disclosure further includes the following features with respect to the machine learning model providing system according to the fourth disclosure. The second data range provides a plurality of the data ranges according to a plurality of divisions having different scales.

[0013] The sixth disclosure relates to a machine learning model providing method for generating a machine learning model and providing the generated machine learning model. This method for providing a machine learning model includes collecting and storing a data catalog of a collection of personal information data and metadata related to the personal information data, acquiring designation of the metadata to be used in generating the machine learning model, designation of a first data range that provides a data range related to the personal information data, and machine learning logic, generating the machine learning model according to the logic based on the designated metadata and the personal information data that falls within the first data range, calculating an identification risk for the generated machine learning model that indicates the risk that an individual may be identified from the output, providing the generated machine learning model if the identification risk is less than a predetermined threshold, and stopping provision of the generated machine learning model if the identification risk is equal to or greater than the threshold.

[0014] The seventh disclosure further includes the following features with respect to the machine learning model providing method according to the sixth disclosure. Each of the personal information data is associated with one or more ID information indicating a specific individual, and the personal information data is selected to be input data for the generated machine learning model, output data that is the output of the machine learning model for the input data is obtained, associated output information is generated that indicates the correspondence between the ID information and a set of output data related to the input data associated with the ID information, and the specific risk is calculated based on the associated output information.

[0015] The eighth disclosure further includes the following features with respect to the machine learning model providing method according to the seventh disclosure. The output data is divided into categories, and for each combination (i, j) of the ID information and the category, the number of elements u(i, j) of the set of output data included in the category and corresponding to the ID information in the related output information is calculated, and the specific risk IR is calculated using the following equations (1) and (2).

number

number

[0016] According to the machine learning model providing system and machine learning model providing method of the present disclosure, a machine learning model using personal information data can be provided. In addition, when the identification risk is equal to or greater than a predetermined threshold, provision of the generated machine learning model is stopped. This makes it possible to prevent the provision of a machine learning model whose output is highly likely to identify an individual and whose personal information protection is insufficient. In turn, it is possible to reduce the risk of personal information data being leaked.

[0017] Furthermore, the metadata and the personal information data corresponding to the first data range are used to generate a machine learning model. This eliminates the need to disclose personal information data to users. As a result, it is possible to provide a machine learning model that is generated by directly using personal information data, rather than statistical information, etc. [Brief description of the drawings]

[0018] [Figure 1] This is a conceptual diagram for explaining an overview of the platform constructed by the machine learning model providing system. [Diagram 2] FIG. 1 is a block diagram showing a configuration of a machine learning model providing system according to an embodiment of the present invention. [Diagram 3] FIG. 4 is a diagram showing an example of personal information data stored in a storage device. [Figure 4] 1 is a block diagram showing a configuration relating to a process executed when an information processing device acquires request information for providing a data catalog. FIG. [Diagram 5] 1 is a flowchart showing a data catalog providing method realized by the machine learning model providing system according to this embodiment. [Figure 6] A block diagram showing a configuration related to a process executed when an information processing device acquires request information for providing a machine learning model. [Figure 7] 1 is a flowchart showing a machine learning model providing method realized by the machine learning model providing system according to this embodiment. [Figure 8] 1A and 1B are diagrams showing an example of input data, output data corresponding to the input data, and related output information that is generated. [Figure 9] FIG. 9 is a diagram showing u(i, j) for each combination (i, j) of ID information and category when the associated output information shown in FIG. 8 is generated. [Figure 10] FIG. 13 is a diagram showing an example of u(i,j) and r(i,j) calculated for that u(i,j). [Figure 11]13 is a flowchart showing a process executed by a specific risk calculation processing unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] Hereinafter, the embodiments of the present disclosure will be described with reference to the drawings. However, when the number, quantity, amount, range, etc. of each element is mentioned in the embodiments shown below, the idea of ​​the present disclosure is not limited to the mentioned number unless it is specifically stated or the number is clearly specified in principle. Furthermore, the configurations etc. described in the embodiments shown below are not necessarily essential to the idea of ​​the present disclosure unless it is specifically stated or the number is clearly specified in principle. In addition, the same reference numerals are attached to the same or corresponding parts in each drawing, and the duplicated explanation is appropriately simplified or omitted.

[0020] 1. Overview The machine learning model providing system according to the present embodiment builds a platform for providing a machine learning model in response to a user request. Fig. 1 is a conceptual diagram for explaining an overview of a platform 1 built by the machine learning model providing system 10.

[0021] The platform 1 stores data including personal information data provided by a data provider 2 (hereinafter also simply referred to as "personal information data"). A user 3 receives a machine learning model by specifying the data to be used and the machine learning logic. Here, the machine learning model provided is generated based on the specified data and in accordance with the specified logic.

[0022] The data provider 2 refers to an entity that collects and provides data, and is not limited to a person. For example, examples of the data provider 2 include a person, a company, a retail store, a surveillance camera, a user terminal, and an automobile. Examples of personal information data collected and provided by these data providers 2 include personal attributes (age, height, weight, etc.), customer data, purchase information (purchased items, purchase date and time, etc.), surveillance video data, terminal usage information, movement data, etc. Other examples of personal information data include location information, behavioral information, related media information, and physical information, etc.

[0023] The data may be provided actively or automatically, for example, by operating a specific device, or automatically, either continuously or periodically, via communication over the Internet or a cable.

[0024] The platform 1 is constructed by a machine learning model providing system 10. The machine learning model providing system 10 includes an information processing device 100 and a storage device 200.

[0025] The information processing device 100 is connected to the storage device 200 and executes machine learning using data stored in the storage device 200. Then, the information processing device 100 outputs a machine learning model generated by the machine learning. However, the data to be used and the machine learning logic follow the specifications of the user 3.

[0026] The information processing device 100 is, for example, a server configured on the Internet. In this case, the user 3 specifies the data and logic to be used through communication on the Internet, and is provided with the machine learning model. Typically, the user 3 specifies the data and logic to be used through an application or a browser, and is provided with the machine learning model.

[0027] As another example, the information processing device 100 may be a dedicated terminal or a general-purpose computer. In this case, typically, the user 3 specifies the data and logic to be used by operating an HMI device (such as a touch panel, keyboard, or display) connected to the dedicated terminal or general-purpose computer, and receives the machine learning model.

[0028] The information processing device 100 may be a system configured with a plurality of computers. For example, the information processing device 100 may be configured to execute a specific process in cooperation with a plurality of servers.

[0029] The storage device 200 stores personal information data provided by the data provider 2. The storage device 200 is, for example, a data server configured on the Internet. In this case, the storage device 200 acquires and stores personal information data from the data provider 2 via communication on the Internet.

[0030] The information processing device 100 and the storage device 200 are connected so as to be able to transmit information to each other. For example, the information processing device 100 and the storage device 200 are connected via the Internet. Alternatively, the information processing device 100 and the storage device 200 are electrically or optically connected via a cable. Alternatively, the information processing device 100 and the storage device 200 each have a wireless device and are connected via wireless communication.

[0031] The information processing device 100 and the storage device 200 may be integrated. For example, the storage device 200 may be an auxiliary storage device of the information processing device 100.

[0032] In the platform 1 constructed by such a machine learning model providing system 10, from the viewpoint of protecting personal information, it is required that personal information data not be disclosed to third parties. Furthermore, if the output of a generated machine learning model is highly likely to identify an individual, the machine learning model cannot be said to provide sufficient personal information protection. For example, if a specific output of a machine learning model is biased when personal information data relating to a specific individual is input, there is a risk that the specific output will have the same meaning as the specific individual.

[0033] Therefore, the machine learning model providing system 10 according to this embodiment generates a machine learning model using personal information data without disclosing the personal information data to the user 3. Furthermore, for the generated machine learning model, a risk that an individual will be identified from the output (hereinafter also referred to as "identification risk") is calculated, and if the identification risk is equal to or exceeds a predetermined threshold, provision of the generated machine learning model is stopped. An embodiment of the machine learning model providing system 10 and a machine learning model providing method will be described below.

[0034] 2. Machine learning model provision system 2-1.Configuration 2 is a block diagram showing the configuration of the machine learning model providing system 10 according to this embodiment. The machine learning model providing system 10 includes an information processing device 100 and a storage device 200.

[0035] The information processing device 100 acquires information related to the request (request information) and executes processing according to the request information. Then, the information processing device 100 outputs the execution result of the processing. Here, in the machine learning model providing system 10 according to the present embodiment, the request to the information processing device 100 includes the provision of a data catalog 220 and the provision of a machine learning model.

[0036] The data catalog 220 is a data catalog for metadata related to the personal information data 210, and is stored in the storage device 200. The metadata includes, for example, table names ("personal attributes", "purchase information", "location information", "behavioral information", "related media", "physical information", etc.), data storage destinations (URLs, etc.), data items ("gender", "age", "address", "product name", "price", "location", etc.), data item types (character strings, numbers, time, etc.), and data value ranges (minimum value, maximum value, etc.).

[0037] The data catalog 220 is updated by the information processing device 100. The information processing device 100 acquires the personal information data 210 from the storage device 200, and extracts metadata related to the personal information data 210. Then, the information processing device 100 updates the data catalog 220 based on the extracted metadata.

[0038] The data catalog 220 is updated, for example, at a predetermined interval. In this case, the information processing device 100 acquires information on the time when the data catalog 220 was updated (hereinafter also referred to as the "update time") from the storage device 200, and determines whether or not the data catalog 220 needs to be updated based on the update information.

[0039] The information processing device 100 includes a catalog confirmation processing unit 110, a model generation processing unit 120, a specific risk calculation processing unit 130, and a provision management processing unit 140. These may be realized as parts of a program that describes the processing executed by the information processing device 100, or may be realized by multiple computers that constitute the information processing device 100.

[0040] The catalog confirmation processing unit 110 executes processing related to updating the data catalog 220 and providing the data catalog 220. The model generation processing unit 120, the specific risk calculation processing unit 130, and the provision management processing unit 140 execute processing related to generating a machine learning model and providing the machine learning model.

[0041] When the information processing device 100 receives request information for providing the data catalog 220, it outputs the data catalog 220 as an execution result. When the information processing device 100 receives request information for providing a machine learning model, it outputs the machine learning model as an execution result. Details of the process executed by the information processing device 100 when each piece of request information is received will be described later.

[0042] The personal information data 210 stored in the storage device 200 is associated with one or more pieces of ID information indicating a specific individual. Fig. 3 is a diagram showing an example of the personal information data 210. (A), (B), and (C) of Fig. 3 show the personal information data 210 relating to personal attributes, purchase information, and surveillance video in the form of a table, respectively. The table shows items in the first line, and each of the following lines shows one piece of personal information data 210.

[0043] As shown in Fig. 3, each personal information data 210 is associated with one or more pieces of ID information. In the example shown in Fig. 3, the ID information is a character string corresponding to a specific individual. The ID information in the personal attributes shown in Fig. 3 (A) indicates, for example, an individual having each personal attribute. The ID information in the purchasing information shown in Fig. 3 (B) indicates, for example, a purchaser. The ID information in the surveillance video shown in Fig. 3 (C) indicates, for example, a person captured in the video file.

[0044] The data for each item may be quantified. For example, the data for the item "gender" may be quantified by setting Male to 0 and Female to 1.

[0045] 2-2. Providing a data catalog Hereinafter, a description will be given of a process (hereinafter also referred to as a "data catalog providing process") that the information processing device 100 executes when it acquires request information for providing the data catalog 220. Fig. 4 is a block diagram showing a configuration related to the data catalog providing process.

[0046] When the catalog confirmation processing unit 110 acquires the request information, it acquires the update time of the data catalog 220 and checks whether a certain period of time has passed since the update time to the present time. Here, the certain period of time may be suitably determined according to the environment in which the machine learning model providing system 10 is employed, the personal information data 210 to be handled, etc.

[0047] Then, if a certain period of time has passed since the update time to the present time, the data catalog 220 is updated. At this time, the catalog confirmation processing unit 110 acquires the personal information data 210 and extracts metadata of the personal information data 210. The catalog confirmation processing unit 110 updates the data catalog 220 based on the extracted metadata. Thereafter, the catalog confirmation processing unit 110 outputs the updated data catalog 220.

[0048] If a certain period of time has not elapsed since the update time to the present time, the catalog confirmation processing unit 110 acquires and outputs the data catalog 220 without updating the data catalog 220.

[0049] Fig. 5 is a flowchart showing a data catalog providing method realized by the machine learning model providing system 10 according to this embodiment. The flowchart shown in Fig. 5 starts when the information processing device 100 acquires request information for providing the data catalog 220, and the process is executed at predetermined intervals.

[0050] In step S100, the catalog confirmation processing unit 110 acquires the update time of the data catalog 220. After step S100, the process proceeds to step S110.

[0051] In step S110, the catalog confirmation processing unit 110 determines whether a certain period of time has passed since the update time acquired in step S100 until the present time. If a certain period of time has passed since the update time until the present time (step S110; Yes), the process proceeds to step S120. If a certain period of time has not passed since the update time until the present time (step S110; No), the process proceeds to step S140.

[0052] In step S120, the catalog confirmation processor 110 acquires the personal information data 210 and extracts metadata from the personal information data 210. After step S120, the process proceeds to step S130.

[0053] In step S130, the catalog confirmation processor 110 updates the data catalog 220 based on the metadata extracted in step S120. After step S130, the process proceeds to step S140.

[0054] In step S140, the catalog confirmation processor 110 obtains and outputs the data catalog 220. After step S140, the process ends.

[0055] As described above, in the data catalog providing process, if a certain amount of time has elapsed since the update of the data catalog 220 to the present time, the data catalog 220 is updated. This makes it possible to keep the information in the provided data catalog 220 up to date while suppressing the update load of the data catalog 220.

[0056] 2-3. Providing machine learning models Hereinafter, a process (hereinafter also referred to as a "machine learning model providing process") that the information processing device 100 executes when it acquires request information for providing a machine learning model will be described. Fig. 6 is a block diagram showing a configuration related to the machine learning model providing process.

[0057] The request information specifies a data range (hereinafter also referred to as a "first data range") related to the metadata and personal information data 210 used to generate the machine learning model, and the machine learning logic.

[0058] Here, the specified metadata and the personal information data 210 corresponding to the first data range are the data used to generate the machine learning model. In other words, the request information does not directly specify the data to be used to generate the machine learning model. By extension, the machine learning model providing system 10 does not need to disclose the personal information data 210 to the user 3 for the convenience of the user 3's request.

[0059] On the other hand, by being provided with the data catalog 220, the user 3 can specify the metadata and the first data range from the data catalog 220.

[0060] The metadata designation is typically a designation of an item of data. The first data range designation is typically a designation of a period related to the time of data acquisition. For example, the personal attributes "gender" and "age" are designated as metadata designation, and the acquisition time period of 2020 / 10 / 01-2020 / 11 / 31 is designated as the first data range designation. In this case, the data used to generate the machine learning model will be data on "gender" and "age" acquired within the range of the period of 2020 / 10 / 01-2020 / 11 / 31.

[0061] Other examples include a case where the metadata is designated by designating a table name, and a case where the first data range is designated by designating a range of points from which the data was acquired.

[0062] The machine learning logic typically includes a learning logic and an evaluation logic. In addition, other machine learning logic such as an input conversion logic for making a synthetic parameter of data into learning data may be included. The machine learning logic may be specified by directly inputting a source code, or may be selected from a library provided by the machine learning model providing system 10.

[0063] Typically, the output content (e.g., labels, etc.) of the machine learning model is determined by the learning logic.

[0064] In addition, a plurality of first data ranges may be specified according to the logic of machine learning. For example, a data range related to data for learning and a data range related to data for verification may be separately provided.

[0065] The model generation processing unit 120 acquires the metadata specified by the request information and the personal information data 210 corresponding to the first data range. Then, the model generation processing unit 120 generates a machine learning model according to the machine learning logic specified by the request information. The generated machine learning model is transmitted to the specific risk calculation processing unit 130 and the provision management processing unit 140.

[0066] The model generation processing unit 120 typically executes processes related to interpretation, learning, verification, and evaluation of source code. However, when machine learning logic is selected from a library, it is not necessary to execute processes related to interpretation of source code.

[0067] The specific risk calculation processing unit 130 executes a process (specific risk calculation process) for calculating a specific risk for the machine learning model generated by the model generation processing unit 120. The specific risk calculation process is performed by using personal information data 210 as input data and verifying the output of the generated machine learning model. The specific risk calculation process will be described in detail later. The calculated specific risk is transmitted to the provision management processing unit 140.

[0068] The provision management processing unit 140 determines whether or not to provide the machine learning model generated by the model generation processing unit 120, based on the specific risk calculated by the specific risk calculation processing unit 130. The provision management processing unit 140 provides the machine learning model when the specific risk is less than a predetermined threshold, and stops providing the machine learning model when the specific risk is equal to or greater than the predetermined threshold. In addition, when stopping the provision of the machine learning model, the provision management processing unit 140 issues an alert related to the machine learning model whose provision has been stopped.

[0069] In addition, when provision of the machine learning model is stopped, the generated machine learning model may be configured to be deleted.

[0070] Fig. 7 is a flowchart showing a machine learning model providing method realized by the machine learning model providing system 10 according to this embodiment. The flowchart shown in Fig. 7 starts when the information processing device 100 acquires request information for providing a machine learning model, and processing is executed at predetermined intervals.

[0071] In step S200, the model generation processing unit 120 acquires the metadata specified by the request information and the personal information data 210 corresponding to the first data range. After step S200, the process proceeds to step S210.

[0072] In step S210, the model generation processing unit 120 generates a machine learning model according to the machine learning logic specified by the request information, based on the personal information data 210 acquired in step S200. After step S210, the process proceeds to step S220.

[0073] In step S220, the specific risk calculation processing unit 130 executes a specific risk calculation process for the machine learning model generated in step S210 to calculate a specific risk. The specific risk calculation process will be described in detail later. After step S220, the process proceeds to step S230.

[0074] In step S230, the provision management processing unit 140 judges whether the specific risk calculated in step S220 is equal to or greater than a predetermined threshold. If the specific risk is equal to or greater than the predetermined threshold (step S230; Yes), the process proceeds to step S240. If the specific risk is less than the predetermined threshold (step S230; No), the process proceeds to step S250.

[0075] In step S240, the provision management processing unit 140 stops providing the machine learning model generated in step S210. Also, an alert is issued regarding the machine learning model whose provision has been stopped. After step S240, the process ends.

[0076] In step S250, the provision management processing unit 140 provides the machine learning model generated in step S210. After step S250, the process ends.

[0077] As described above, in the machine learning model provision process, if the identification risk is equal to or greater than a predetermined threshold, provision of the generated machine learning model is stopped. This makes it possible to prevent the provision of machine learning models whose output is highly likely to identify individuals and whose personal information protection is insufficient.

[0078] 3. Specific risk calculation process The specific risk calculation process executed by the specific risk calculation processing unit 130 will be described below.

[0079] First, the specific risk calculation processing unit 130 selects the personal information data 210 to be used as input data for the machine learning model generated by the model generation processing unit 120.

[0080] Here, the specific risk calculation processing unit 130 generates a second data range that provides a data range related to the personal information data 210, and selects the personal information data 210 that corresponds to the second data range as input data. Since the input data for the generated machine learning model is selected, the second data range provides a data range with the same content as the first data range. For example, when the first data range is a data range for a period related to the acquisition time, the second data range is also a data range for a period related to the acquisition time. Moreover, the selected personal information data 210 is data on metadata related to the request information.

[0081] The second data range is typically generated to be relatively prime to the first data range in order to avoid overlap with the data used for training, validation, and evaluation.

[0082] Furthermore, the second data range may provide multiple data ranges according to multiple divisions of different scales. For example, if the second data range is a data range for a period related to the acquisition time point, the second data range may provide data ranges by minute, hour, day, and month.

[0083] In this way, by providing a plurality of data ranges according to a plurality of divisions with different scales as the second data range, it is possible to obtain the output of the machine learning model generated for each of the input data with different scales. This makes it possible to reduce the occurrence of the specific risk being calculated as being excessively small due to the difference in scale.

[0084] Next, the specific risk calculation processing unit 130 acquires output data for the input data for the generated machine learning model. Note that, if an input conversion logic or the like is given as request information, it may be applied to the input data as appropriate.

[0085] Next, the specific risk calculation processing unit 130 generates related output information indicating the correspondence between the ID information and a set of output data related to the input data associated with the ID information. Fig. 8 is a conceptual diagram for explaining the related output information. Fig. 8 shows an example of input data, output data for the input data, and the generated related output information. Note that numbers are used to indicate the correspondence between the input data and the output data.

[0086] In FIG. 8, the related output information shows the correspondence between each of the three ID information (jfda093jt, gq409jf32, 65k262f) and a set of output data related to the input data associated with each ID information. Here, it can be seen from the related output information that output 4 is outputted especially when the ID information is jfda093jt. In other words, there is a risk that output 4 has the same meaning as a specific individual whose ID information is jfda093jt. By extension, there is a high probability that the output will identify an individual. In this way, the related output information provides information about the identification risk.

[0087] Therefore, the specific risk calculation processing unit 130 calculates the specific risk based on the related output information. Here, the calculation of the specific risk is executed, for example, as follows.

[0088] First, the output data is divided into categories. The categories may be assigned appropriately depending on the output of the machine learning model.

[0089] For example, if the output is a one-dimensional continuous value, the number of categories is determined by Sturges's rule, and the class width is determined from the number of categories and the maximum and minimum values ​​of the output to give a histogram. Then, each class can be treated as a category.

[0090] Alternatively, if the output is an image, a distance between two images can be defined, and then agglomerative hierarchical clustering can be performed using the distance to classify each output into a certain number of groups, with each group being a category.

[0091] Alternatively, if the output is given by discrete values ​​or labels, each value or label may be considered a category.

[0092] Next, for each combination (i, j) of ID information and category, the number of elements u(i, j) of the set of output data included in the category and corresponding to the ID information in the related output information is calculated, where i indicates each ID information and j indicates each category.

[0093] Figure 9 shows u(i,j) for each combination (i,j) of ID information and category when the related output information shown in Figure 8 is generated. Note that in Figure 9, each output value is regarded as a category.

[0094] Then, the specific risk IR is calculated using the following formulas (1) and (2).

number

number

[0095] Figure 10 shows an example of u(i,j) and r(i,j) calculated for that u(i,j). In this case, the specific risk IR is 4 / 5, which is the maximum value of r(i,j).

[0096] If the second data range provides multiple data ranges (for example, multiple data ranges are provided according to multiple divisions with different scales), the specific risk calculation processing unit 130 may be configured to calculate the specific risk IR for each data range. Then, for each specific risk IR, a weighted average value based on the number of ID information related to each data range may be calculated as the final specific risk.

[0097] FIG. 11 is a flowchart showing the process executed by the specific risk calculation processing unit 130 (step S220 in FIG. 7).

[0098] In step S221, the specific risk calculation processing unit 130 generates a second data range. After step S221, the process proceeds to step S222.

[0099] In step S222, the specific risk calculation processing unit 130 selects, as input data, the personal information data 210 that falls within the second data range generated in step S221. After step S222, the process proceeds to step S223.

[0100] In step S223, the specific risk calculation processing unit 130 acquires output data for the input data selected in step S222, for the generated machine learning model. After step S223, the process proceeds to step S224.

[0101] In step S224, the specific risk calculation processing unit 130 generates related output information based on the output data acquired in step S223. After step S224, the process proceeds to step S225.

[0102] In step S225, the specific risk calculation processing unit 130 calculates the specific risk IR based on the related output information generated in step S224. After step S225, the process ends.

[0103] 4. Feedback to users When the specific risk IR reaches or exceeds a predetermined threshold and the provision of the machine learning model is stopped, the provision management processing unit 140 may be configured to issue an alert and return the feedback shown below to the user 3.

[0104] One is to feed back the category where r(i,j) is equal to or greater than a threshold. Taking FIG. 10 as an example, when the threshold is set to 0.5, r(i,j) for the combination of C and the third category is equal to or greater than the threshold. Therefore, the third category is fed back. This is expected to be resolved by changing the output label or category. Therefore, as further feedback, it may be recommended to change the output label or category in the machine learning logic.

[0105] One is to provide feedback that there are individuals (ID information) concentrated in a single category. In FIG. 10, this corresponds to the combination of D and the second category. This is expected to be resolved by expanding the data range related to the first data range. Therefore, the feedback may further include a recommendation to expand the data range related to the first data range.

[0106] 5.Effects As described above, according to the machine learning model providing system 10 and the machine learning model providing method of the present embodiment, a machine learning model using personal information data 210 can be provided. Furthermore, if the identification risk is equal to or greater than a predetermined threshold, provision of the generated machine learning model is stopped. This makes it possible to prevent the provision of a machine learning model whose output is highly likely to identify an individual and whose personal information protection is insufficient. In turn, it is possible to reduce the risk of personal information data being leaked.

[0107] Furthermore, the metadata and the personal information data 210 corresponding to the first data range are data used to generate a machine learning model. This eliminates the need to disclose the personal information data 210 to the user 3. As a result, it is possible to provide a machine learning model that is generated by directly using the personal information data 210, rather than statistical information or the like.

[0108] Furthermore, the second data range may provide a plurality of data ranges according to a plurality of divisions with different scales, thereby making it possible to reduce the specific risk being calculated as being excessively small due to differences in scale. [Explanation of symbols]

[0109] 1. Platform 2. Data providers 3 Users 10 Machine learning model provision system 100 Information processing device 110 Catalog confirmation processing section 120 Model generation processing unit 130 Specific risk calculation processing unit 140 Provision Management Department 200 Storage device 210 Personal information data 220 Data Catalog IR Specific Risks

Claims

1. A machine learning model providing system that generates a machine learning model and provides the generated machine learning model to a user, a storage device that stores a data catalog of a set of personal information data and metadata related to the personal information data; An information processing device; Including, The information processing device includes: providing the data catalog to a user in response to a request from the user; An acquisition process of acquiring request information including designation of the metadata to be used for generating the machine learning model, designation of a first data range that provides a data range related to the personal information data, and machine learning logic from the user who has been provided with the data catalog; A model generation process that generates the machine learning model according to the logic of the request information based on the metadata specified by the request information and the personal information data corresponding to the first data range; A specific risk calculation process for calculating a specific risk indicating a risk that an individual may be identified from the output of the generated machine learning model; A provision management process that stops provision of the generated machine learning model when the specified risk is equal to or greater than a predetermined threshold; A machine learning model providing system characterized by executing the above.

2. The machine learning model providing system according to claim 1 , Each of the personal information data is associated with one or more pieces of ID information indicating a specific individual, The information processing device, in the specific risk calculation process, Selecting the personal information data to be input data to the generated machine learning model; obtaining output data that is an output of the machine learning model for the input data; generating associated output information indicating a correspondence between the ID information and a set of the output data related to the input data associated with the ID information; Calculating the specific risk based on the related output information A machine learning model providing system characterized by:

3. The machine learning model providing system according to claim 2, The output data is divided into categories; The information processing device, in the specific risk calculation process, For each combination (i, j) of the ID information and the category, a number u(i, j) of elements of a set of the output data included in the category and corresponding to the ID information in the related output information is calculated; The specific risk IR is calculated by the following formulas (1) and (2). A machine learning model providing system characterized by: [0010] [0025]

4. The machine learning model providing system according to claim 2 or 3, The information processing device, in the specific risk calculation process, A second data range is generated to give the data range, and the personal information data corresponding to the second data range is selected as the input data. A machine learning model providing system characterized by:

5. The machine learning model providing system according to claim 4, The second data range provides a plurality of the data ranges according to a plurality of divisions with different scales. A machine learning model providing system characterized by:

6. A machine learning model providing method for generating a machine learning model and providing the generated machine learning model to a user, comprising: Collecting and storing a data catalog of a set of personal information data and metadata related to the personal information data; providing the data catalog to the user in response to a request from the user; acquiring request information from the user who has been provided with the data catalog, the request information including a designation of the metadata to be used for generating the machine learning model, a designation of a first data range that provides a data range related to the personal information data, and a machine learning logic; generating the machine learning model according to the logic of the request information based on the metadata specified by the request information and the personal information data corresponding to the first data range; Calculating an identification risk indicating the risk that an individual will be identified from the output of the generated machine learning model; When the specific risk is less than a predetermined threshold, the generated machine learning model is provided, and when the specific risk is equal to or greater than the threshold, provision of the generated machine learning model is stopped. A method for providing a machine learning model, comprising:

7. The machine learning model providing method according to claim 6, Each of the personal information data is associated with one or more pieces of ID information indicating a specific individual, Selecting the personal information data to be input data to the generated machine learning model; obtaining output data that is an output of the machine learning model for the input data; generating associated output information indicating a correspondence between the ID information and a set of the output data related to the input data associated with the ID information; Calculating the specific risk based on the related output information A method for providing a machine learning model, comprising:

8. The machine learning model providing method according to claim 7, The output data is divided into categories; For each combination (i, j) of the ID information and the category, a number u(i, j) of elements of a set of the output data included in the category and corresponding to the ID information in the related output information is calculated; The specific risk IR is calculated by the following formulas (1) and (2). A method for providing a machine learning model, comprising: [0030] [0045]

Images