JP7676945B2 - Management device, management method, and program - Google Patents

Description

This disclosure relates to a management device, a management method, and a program.

In recent years, the use of machine learning for data analysis has become widespread. For example, Patent Literature 1 discloses the configuration of a machine learning system that generates a learning model based on a dataset and implements the generated learning model. A user accesses and uses the machine learning system via a graphical user interface. In the machine learning system disclosed in Patent Literature 1, a learning model is generated by importing a dataset including multiple categories and executing machine learning.

JP 2019-212278 A

The machine learning system disclosed in Patent Literature 1 deletes datasets of categories designated by the user when importing a dataset to generate a learning model. However, although the user designates and deletes categories that are deemed unnecessary for the learning model to be generated, the user does not delete categories from the perspective of whether or not they are permitted to be used to generate the learning model. Even when the user deletes categories that cannot be used to generate the learning model, the user's judgment is subjective, and there is a problem in that data of categories that should be deleted may be mistakenly used to create the learning model.

One of the objectives of the present disclosure is to provide a management device, a management method, and a program that can reduce the possibility of a learning model being generated by using unauthorized data in machine learning.

The management device according to the first aspect of the present disclosure includes an acquisition unit that acquires at least one auxiliary information related to a dataset, a management unit that determines whether the dataset contains sensitive data by applying the auxiliary information to a determination model that determines whether the dataset contains sensitive data, and a monitoring unit that, when the dataset containing the sensitive data is specified when generating a learning model, determines whether to allow use of the dataset in generating the learning model based on the administrator authority of the user generating the learning model.

The management method according to the second aspect of the present disclosure obtains at least one auxiliary information related to a dataset, and determines whether the dataset contains sensitive data by applying the auxiliary information to a determination model that determines whether the dataset contains sensitive data, and when generating a learning model, if the dataset containing the sensitive data is specified, determines whether to allow use of the dataset in generating the learning model based on the administrator authority of the user generating the learning model.

The program according to the third aspect of the present disclosure causes a computer to acquire at least one auxiliary information related to a dataset, determine whether the dataset contains sensitive data by applying the auxiliary information to a determination model that determines whether the dataset contains sensitive data, and, when generating a learning model, if the dataset containing the sensitive data is specified, determine whether to allow use of the dataset in generating the learning model based on the administrator authority of the user generating the learning model.

The present disclosure provides a management device, a management method, and a program that can reduce the possibility of a learning model being generated by using unauthorized data in machine learning.

FIG. 2 is a configuration diagram of a management device according to the first embodiment; FIG. 11 is a configuration diagram of a management device according to a second embodiment. FIG. 11 is a diagram showing metadata management information according to the second embodiment. FIG. 11 is a diagram showing metadata management information according to the second embodiment. FIG. 11 is a diagram showing data processing information according to the second embodiment; A diagram showing learning model management information according to the second embodiment. FIG. 11 is a diagram showing the flow of a determination process as to whether or not handle with care data is included in the second embodiment. FIG. 11 is a diagram showing the flow of a process for storing information regarding processing on a dataset in the second embodiment. A diagram showing the flow of the machine learning model generation process in embodiment 2. A diagram showing the flow of analysis processing of a machine learning model in embodiment 2. FIG. 11 is a diagram showing a flow of a monitoring process according to the second embodiment. FIG. 2 is a configuration diagram of a management device according to each embodiment.

(Embodiment 1)
Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings. A configuration example of a management device 10 according to the first embodiment will be described with reference to FIG. 1. The management device 10 may be a computer device that operates by a processor executing a program stored in a memory. For example, the management device 10 may be a server device or the like.

The management device 10 has an acquisition unit 11, a management unit 12, and a monitoring unit 13. The acquisition unit 11, the management unit 12, and the monitoring unit 13 may be software or modules that perform processing by a processor executing a program stored in a memory. Alternatively, the acquisition unit 11, the management unit 12, and the monitoring unit 13 may be hardware such as a circuit or a chip.

The acquisition unit 11 acquires at least one piece of auxiliary information related to the dataset. The dataset may be, for example, image data, video data, or text data. The dataset may also include two or more pieces of data, such as image data, video data, and text data. The auxiliary information may be information indicating an attribute of the dataset, such as image data, video data, or text data, and may specifically be information indicating the type or size of the dataset. The auxiliary information may be referred to as metadata, for example. The auxiliary information may be information generated by a user who manages the management device 10, or may be information generated autonomously in the management device 10 that manages the dataset.

The management unit 12 determines whether the dataset contains sensitive data by applying the ancillary information to a determination model that determines whether the dataset contains sensitive data. The sensitive data may be personal information that can identify an individual, such as a phone number, address, name, or date of birth. Alternatively, the sensitive data may be data that can be viewed only by specific people within a company.

The judgment model may be, for example, an analytical model or a predictive model that performs an analysis using a decision tree. The analysis method is not limited to analysis using a decision tree, and other algorithms, etc. may be used. For example, the judgment model determines whether or not the data corresponds to data that requires special handling, by setting the objective variable as to whether or not the data corresponds to data that requires special handling, and setting the auxiliary information related to the dataset as an explanatory variable.

When a dataset containing sensitive data is specified when generating a learning model, the monitoring unit 13 determines whether or not to permit use of the dataset in generating the learning model based on the administrator authority of the user generating the learning model.

The learning model may be, for example, a learning model generated by executing machine learning, or may be a learning model generated by performing deep learning or reinforcement learning, etc. The learning model is generated by learning a specified dataset. The accuracy of the results output by the learning model varies depending on the selection of data or dataset used for learning, processing of the dataset or data, the number of times learning is performed, the combination of learning methods, etc.

The administrator authority may be information indicating whether or not the user has the authority to use the sensitive data included in the dataset specified when generating a learning model. For example, the user's identification information may be input to the management device 10 when generating a learning model. In this case, if the dataset specified by the user contains sensitive data but the user does not have the authority to use the sensitive data, the monitoring unit 13 refuses to use the dataset specified by the user. In other words, the monitoring unit 13 monitors whether or not the user uses only data that the user has the authority to use when generating a learning model.

As described above, the management device 10 according to the first embodiment determines whether or not handle-with-handle data is included for each dataset generated in the learning model. This allows the management device 10 to detect that a dataset containing handle-with-handle data is being used when a learning model is generated using various datasets.

(Embodiment 2)
Next, a configuration example of the management device 20 according to the second embodiment will be described with reference to FIG. 2. The management device 20 corresponds to the management device 10 in FIG. 1. The management device 20 has an acquisition unit 21, a metadata management unit 22, a data usage monitoring unit 23, a data processing management unit 24, a learning model management unit 25, a storage unit 26, and a display unit 27. The acquisition unit 21 corresponds to the acquisition unit 11 in FIG. 1. The metadata management unit 22 corresponds to the management unit 12 in FIG. 1. The data usage monitoring unit 23 corresponds to the monitoring unit 13 in FIG. 1. The components of the management device 20 may be software or modules in which processing is performed by a processor executing a program stored in a memory. Alternatively, the components of the management device 20 may be hardware such as a circuit or a chip.

The acquisition unit 21 outputs the identification information of the dataset used to generate the machine learning model and the metadata of the dataset to the metadata management unit 22. The identification information of the dataset is information capable of uniquely identifying the dataset. For example, when the dataset is represented in a file format, it may be a file name. The metadata may be, for example, the name of the registrant of the dataset, the data type, the data length, the column name, and the data location. The dataset used to generate the machine learning model may be stored in a device different from the management device 20, or may be stored in the management device 20. Therefore, the data location may be the name of the device in which the dataset is stored or address information of the device. In addition, when the dataset is stored in the management device 20, it may be information indicating the folder name, memory location, etc. in which the dataset is stored.

The registrant name may be information that identifies a user who registered the dataset in the management device 20 or in a device different from the management device 20. The registrant name may be the user who generated the dataset. The data type may be, for example, a data type used when a programming language handles a dataset. Or, the data type may be information indicating the type of data included in the dataset. The data type may be information indicating, for example, image data, video data, or text data. The data type may be rephrased as the attribute of the data. The data length may be rephrased as the data size, etc. The column name may be, for example, the name of an item managed in each column of data managed in a table format.

Furthermore, the acquisition unit 21 outputs information regarding the processing performed on the dataset to the data processing management unit 24. The information regarding the processing performed on the dataset may be, for example, information or parameters indicating the contents of preprocessing or processing, or information or parameters identifying a program for executing preprocessing or processing. Hereinafter, the term "information" may be replaced with "parameters."

The information indicating the contents of the preprocessing or processing may be, for example, the contents of the arithmetic processing performed on the dataset. The preprocessing or processing may be the correction or deletion of data included in the dataset, or the addition of further data to the dataset. Furthermore, the information regarding the processing performed on the dataset may include identification information for the preprocessing data and the postprocessing data, and may further include information indicating the respective locations of the preprocessing data and the postprocessing data. The preprocessing data is data before the preprocessing or processing is performed, and the postprocessing data is data after the preprocessing or processing is performed.

The metadata management unit 22 uses the metadata received from the acquisition unit 21 to determine whether or not the dataset contains sensitive data. For example, the metadata management unit 22 uses a decision tree as a determination model for determining whether or not the dataset contains sensitive data. The determination process using a decision tree may be referred to as decision tree analysis. For example, the metadata management unit 22 may create a decision tree in advance using sample data. The sample data may be associated with information on whether or not the data corresponds to sensitive data, and metadata. The metadata management unit 22 uses whether or not the data corresponds to sensitive data as the objective variable and generates a decision tree that classifies the sample data using metadata.

For example, the metadata management unit 22 identifies the number of sample data items that are handled with care and the number of sample data items that are not handled with care among sample data items whose data type is A. Furthermore, the metadata management unit 22 identifies the number of sample data items that are handled with care and the number of sample data items that are not handled with care among sample data items whose data type is B. Furthermore, the metadata management unit 22 identifies the number of sample data items that are handled with care and the number of sample data items that are not handled with care among sample data items whose data type is A and whose data length is N bytes or more. Furthermore, the metadata management unit 22 identifies the number of sample data items that are handled with care and the number of sample data items that are not handled with care among sample data whose data type is A and whose data length is less than N bytes. In this way, the metadata management unit 22 can generate a decision tree having multiple hierarchies by classifying sample data using multiple metadata.

When the metadata management unit 22 determines that the dataset contains sensitive data, it associates the metadata and identification information associated with the dataset with flag information indicating that the dataset contains sensitive data. Furthermore, when the metadata management unit 22 determines that the dataset does not contain sensitive data, it associates the metadata and identification information of the dataset with flag information indicating that the dataset does not contain sensitive data. The metadata management unit 22 associates the metadata, identification information, and flag information of the dataset and stores them in the storage unit 26. Storing may also be rephrased as recording, memorizing, saving, etc.

For example, the metadata management unit 22 may associate the identification information of the dataset, the data type, the data length, the column name, the data location, and flag information as shown in FIG. 3, and store them in the storage unit 26 as metadata management information. A 1 set in the flag information indicates that handling data is included, and a 0 indicates that handling data is not included. N1 to N3 set in the data length may each indicate the number of bytes. The information set in the data type and column name is information that identifies the data type and column name, respectively. The information set in the data location may be, for example, a device name.

The data processing management unit 24 compares the identification information of the pre-processing data and the post-processing data contained in the information on the processing performed on the data set received from the acquisition unit 21 with the metadata management information stored in the storage unit 26. For example, the data processing management unit 24 receives information in which the program that executes the processing is P1, the pre-processing data is D1, and the post-processing data is D2. The pre-processing data D1 and the post-processing data D2 are stored in the storage unit 26 as metadata management information, as shown in FIG. 3. In this case, the data processing management unit 24 may further add an item related to related data to the metadata management information shown in FIG. 3, as shown in FIG. 4, and associate the data set D1 and the data set D2. Specifically, D2 may be set in the item of related data of the data set D1, and D1 may be set in the item of related data of the data set D2. Furthermore, the data processing management unit 24 stores in the storage unit 26 the data processing information in which the program that executes the processing, the pre-processing data, and the post-processing data are associated, as shown in FIG. 5. For example, a value indicating the contents of the pre-processing or processing may be set in the item of the program.

Returning to FIG. 2, the learning model management unit 25 selects a dataset to be used to generate a machine learning model from the metadata management information stored in the memory unit 26 via the data usage monitoring unit 23. Furthermore, the learning model management unit 25 downloads the dataset from another device where the selected dataset is located. For example, the learning model management unit 25 downloads the dataset from the other device via the communication interface and network of the management device 20.

The learning model management unit 25 generates a machine learning model using the downloaded dataset. The machine learning model is a learning model generated by performing machine learning using the downloaded dataset as learning data. The learning model management unit 25 may generate machine learning models used in various fields, such as image recognition and prediction of attendees at an event. The machine learning model is a learning model used to find a specific objective variable.

For example, as shown in FIG. 6, the learning model management unit 25 associates the identification information of the generated machine learning model, the objective variable of the machine learning model, and the dataset used to generate the machine learning model, and stores them in the memory unit 26 as learning model management information.

When a machine learning model is generated in the learning model management unit 25, the data usage monitoring unit 23 determines whether a flag indicating that sensitive data is included is set in the dataset selected by the learning model management unit 25. Furthermore, when the learning model management unit 25 selects a dataset in which a flag indicating that sensitive data is included is set, the data usage monitoring unit 23 determines whether the user generating the machine learning model has administrator authority for the dataset.

The data usage monitoring unit 23 may manage the identification information of a dataset in association with the management level at which the dataset can be used. Furthermore, the data usage monitoring unit 23 may acquire information on the management level permitted to a user who generates a machine learning model from the learning model management unit 25. If the management level received from the learning model management unit 25 is higher than the management level at which the dataset can be used, the data usage monitoring unit 23 permits the use of the dataset that includes the data with special handling. If the management level received from the learning model management unit 25 is lower than the management level at which the dataset can be used, the data usage monitoring unit 23 does not permit the learning model management unit 25 to use the dataset that includes the data with special handling.

The learning model management unit 25 selects the identification information of the dataset stored in the storage unit 26 via the data usage monitoring unit 23. Therefore, when the data usage monitoring unit 23 does not permit the use of a dataset that includes the sensitive data, the data usage monitoring unit 23 may not notify the learning model management unit 25 of the location of the dataset. Alternatively, when the data usage monitoring unit 23 does not permit the use of a dataset that includes the sensitive data, the data usage monitoring unit 23 may output a message to the learning model management unit 25 indicating that the use of the selected dataset is not permitted. In addition, the data usage monitoring unit 23 may display on the display unit 27 that the use of the dataset that includes the sensitive data is not permitted. Furthermore, even when the data usage monitoring unit 23 permits the use of a dataset that includes the sensitive data, the data usage monitoring unit 23 may display on the display unit 27 that the dataset includes the sensitive data.

Next, the flow of the process of determining whether or not sensitive data is included, which is executed in the metadata management unit 22, will be described with reference to FIG. 7. First, the metadata management unit 22 receives information on the registrant's name, data type, data length, column name, and data location from the acquisition unit 21 as metadata of the dataset used to generate the machine learning model (S11).

Next, the metadata management unit 22 performs a decision tree analysis using the metadata (S12). The metadata management unit 22 performs a decision tree analysis using at least one of the metadata.

If the metadata management unit 22 determines, as a result of the decision tree analysis, that the dataset contains data that requires special handling, it sets a flag indicating that the dataset contains data that requires special handling in the identification information of the dataset that contains the data that requires special handling (S13).

Next, the metadata management unit 22 stores metadata management information that associates the data set identification information, metadata, and flags in the storage unit 26 (S14).

Next, the metadata management unit 22 reflects the judgment result performed in step S12 in a decision tree, which is a judgment model (S15). For example, if the user recognizes that the judgment result in step S12 is incorrect, the metadata management unit 22 accepts information input by the user indicating that the judgment result is incorrect. For example, when using the metadata used in step S12, the metadata management unit 22 corrects the number of pieces of data that contain sensitive data or the number of pieces of data that do not contain sensitive data.

In step S12, if the metadata management unit 22 determines as a result of the decision tree analysis that the dataset does not contain sensitive data, it executes the processes from step S14 onward. In this case, the flag may be set to, for example, 0.

Next, the process flow for storing information regarding processing of a data set in the data processing management unit 24 in the second embodiment in the storage unit 26 will be described with reference to FIG. 8.

First, the data processing management unit 24 receives the name of the registrant of the dataset and information regarding the processing of the dataset from the acquisition unit 21 (S21).

Next, the data processing management unit 24 determines whether the identification information of the pre-processing data or the post-processing data included in the information on the processing performed on the dataset is managed in the storage unit 26 (S22). Specifically, the data processing management unit 24 determines whether the identification information of the pre-processing data or the post-processing data matches the identification information of the dataset included in the metadata management information stored in the storage unit 26 by the metadata management unit 22 described in FIG. 3.

When the data processing management unit 24 determines that the identification information of the pre-processing or post-processing data is included in the metadata management information, it updates the metadata management information using the pre-processing data or the post-processing data (S23). Specifically, as shown in FIG. 4, when the metadata management information contains identification information identical to that of the pre-processing data, the data processing management unit 24 associates the identification information of the post-processing data with the identification information. Furthermore, when the identification information of the post-processing data is included in the metadata management information, the data processing management unit 24 associates the identification information of the pre-processing data with the identification information.

Next, the data processing management unit 24 stores the metadata management information shown in FIG. 4, which is associated with the identification information of the pre-processing data or the post-processing data, in the storage unit 26 (S24).

Next, the data processing management unit 24 stores information related to the processing of the data set in the storage unit 26. Specifically, as shown in FIG. 5, the data processing management unit 24 stores information in the storage unit 26 that associates pre-processing data, a program for executing pre-processing or processing, and post-processing data. In addition to the program for executing pre-processing or processing, processing content, etc. may also be associated.

Next, the flow of the machine learning model generation process executed by the learning model management unit 25 in the second embodiment will be described with reference to FIG. 9.

First, the learning model management unit 25 selects the identification information of the dataset stored in the memory unit 26 via the data usage monitoring unit 23 (S31). For example, the learning model management unit 25 may obtain a list of the identification information of the dataset stored in the memory unit 26 via the data usage monitoring unit 23 and select the identification information of the dataset.

Next, the learning model management unit 25 accesses the device indicated in the location of the data associated with the identification information of the selected dataset, and downloads the dataset (S32). Next, the learning model management unit 25 generates a machine learning model using the downloaded dataset (S33).

Next, the learning model management unit 25 stores in the memory unit 26 learning model management information that associates the identification information of the generated machine learning model with the identification information of the dataset used to generate the machine learning model (34). Specifically, the learning model management unit 25 stores in the memory unit 26 learning model management information that associates the identification information of the learning model with the objective variable of the learning model and the identification information of the dataset, as shown in FIG. 6.

Next, the flow of the machine learning model analysis process executed by the learning model management unit 25 will be explained using FIG. 10.

First, the learning model management unit 25 determines whether or not there are machine learning models with the same objective variable in the storage unit 26 (S41). Machine learning models with the same objective variable are machine learning models used for the same purpose, and the data sets used to generate each machine learning model may be different.

When the learning model management unit 25 determines that a machine learning model having the same objective variable exists, it determines whether or not there is data processing associated with a dataset used only in a machine learning model with high accuracy of output values (S42). For example, a user or the like may evaluate the accuracy of values output from the machine learning model and determine a score for each machine learning model. The score may be determined based on, for example, the proportion of correct output of object recognition results or prediction results using the machine learning model. The learning model management unit 25 may, for example, extract datasets used only in machine learning models with scores equal to or greater than a threshold, and further identify the contents of preprocessing or processing associated with the extracted dataset using the data processing information shown in FIG. 5.

For example, when the learning model management unit 25 determines that there is data processing associated with a data set used only in a machine learning model with high accuracy of output values, the learning model management unit 25 sets a factor flag in the data processing information shown in FIG. 5 (S43). For example, the learning model management unit 25 may set a flag in the record of the data processing information extracted in step S42 indicating that the factor is a factor that improves the output accuracy of the machine learning model.

Next, the learning model management unit 25 determines whether there is metadata associated with a dataset that is used only in a machine learning model that has a high accuracy of output values (S44). The learning model management unit 25 may, for example, extract a dataset that is used only in a machine learning model whose score is equal to or greater than a threshold, and further identify metadata associated with the extracted dataset using the metadata management information shown in FIG. 4.

For example, when the learning model management unit 25 determines that there is metadata associated with a data set used only in a machine learning model with high accuracy of output values, the learning model management unit 25 sets a factor flag in the metadata management information shown in FIG. 4 (S45). For example, the learning model management unit 25 may set a flag indicating a factor that improves the output accuracy of the machine learning model in the record of the metadata management information extracted in step S44.

In step S42, if the learning model management unit 25 determines that there is no data processing associated with a data set used only in a machine learning model with high accuracy of output values, for example, the learning model management unit 25 executes the process of step S44. In step S41, if the learning model management unit 25 determines that there is no machine learning model having the same objective variable, the processing ends. In step S44, if the learning model management unit 25 determines that there is no metadata associated with a data set used only in a machine learning model with high accuracy of output values, the processing ends.

Next, the flow of the monitoring process executed by the data usage monitoring unit 23 in the second embodiment will be explained with reference to FIG. 11.

First, the data usage monitoring unit 23 determines whether or not the dataset specified by the learning model management unit 25 for generating a machine learning model contains data that requires special handling (S51). For example, the data usage monitoring unit 23 may check flag information in the metadata management information to determine whether or not the dataset specified by the learning model management unit 25 contains data that requires special handling.

When the data usage monitoring unit 23 determines that the specified dataset contains sensitive data, it determines whether or not the user has administrator authority to use the dataset (S52). The data usage monitoring unit 23 may determine whether or not the user has administrator authority, for example, by determining whether or not the management level of the user received from the learning model management unit 25 is higher than the management level at which the specified dataset can be used.

When the data usage monitoring unit 23 determines that the user has administrator authority to use the dataset, it displays on the display unit 27 that the dataset specified by the user contains sensitive data (S53). When the data usage monitoring unit 23 determines that the user does not have administrator authority to use the dataset, it denies the learning model management unit 25 from using the dataset (S54).

In step S51, if the data usage monitoring unit 23 determines that the specified data set does not contain sensitive data, it terminates the process.

As described above, when a dataset specified for generating a machine learning model contains sensitive data, the management device 20 according to the second embodiment can refuse to use the specified dataset for generating the machine learning model. Whether or not the dataset contains sensitive data is determined by decision tree analysis using metadata. This allows the determination of whether or not the dataset contains sensitive data without relying on the subjective judgment of the user, making it possible to avoid using sensitive data in generating the machine learning model.

Furthermore, the management device 20 can identify data processing or metadata that contributes to improving the accuracy of values output by the machine learning model. This allows for efficient accumulation of know-how about the machine learning model.

FIG. 12 is a block diagram showing an example of the configuration of a management device 10 and a management device 20 (hereinafter referred to as the management device 10, etc.). Referring to FIG. 12, the management device 10, etc. includes a network interface 1201, a processor 1202, and a memory 1203. The network interface 1201 may be used to communicate with a network node (e.g., eNB, MME, P-GW, etc.). The network interface 1201 may include, for example, a network interface card (NIC) that complies with the IEEE 802.3 series. Here, eNB stands for evolved Node B, MME stands for Mobility Management Entity, and P-GW stands for Packet Data Network Gateway. IEEE stands for Institute of Electrical and Electronics Engineers.

The processor 1202 reads out and executes software (computer programs) from the memory 1203 to perform the processing of the management device 10 and the like described using the flowcharts in the above-mentioned embodiment. The processor 1202 may be, for example, a microprocessor, an MPU, or a CPU. The processor 1202 may include multiple processors.

Memory 1203 is composed of a combination of volatile memory and non-volatile memory. Memory 1203 may include storage located away from processor 1202. In this case, processor 1202 may access memory 1203 via an I/O (Input/Output) interface not shown.

In the example of FIG. 12, the memory 1203 is used to store a group of software modules. The processor 1202 can read and execute these software modules from the memory 1203 to perform the processing of the management device 10 and the like described in the above embodiment.

As described with reference to FIG. 12, each of the processors of the management device 10 in the above-described embodiment executes one or more programs including a set of instructions for causing a computer to execute the algorithm described with reference to the drawings.

In the above examples, the program includes instructions (or software code) that, when loaded into a computer, cause the computer to perform one or more functions described in the embodiments. The program may be stored on a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable media or tangible storage media include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technology, CD-ROM, digital versatile disc (DVD), Blu-ray® disk or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device. The program may be transmitted on a transitory computer-readable medium or communication medium. By way of example and not limitation, the transitory computer-readable medium or communication medium includes electrical, optical, acoustic, or other forms of propagated signals.

Note that this disclosure is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit and scope of the present disclosure.

REFERENCE SIGNS LIST 10 Management device 11 Acquisition unit 12 Management unit 13 Monitoring unit 20 Management device 21 Acquisition unit 22 Metadata management unit 23 Data usage monitoring unit 24 Data processing management unit 25 Learning model management unit 26 Memory unit 27 Display unit

Claims (9)

An acquisition unit for acquiring at least one auxiliary information related to a data set;
A management unit that determines whether the data set includes sensitive data by applying the auxiliary information to a determination model that determines whether the data set includes sensitive data;
A monitoring unit that, when the data set including the sensitive data is specified when generating a learning model, determines whether or not to permit use of the data set in generating the learning model based on an administrator authority of a user who generates the learning model ,
a data processing management unit that acquires data processing information relating to pre-processing or processing performed on the data set;
Further, a learning model management unit is provided that determines whether or not the pre-processing or the processing associated with the data set exists, the pre-processing or the processing is used only for a learning model whose output result has a higher accuracy than a threshold value;
The data processing management unit includes:
a management device that manages at least one of a pre-processing data set before the pre-processing or the processing is performed and a post-processing data set after the pre-processing or the processing is performed in association with the auxiliary information; An acquisition unit for acquiring at least one auxiliary information related to a data set;
A management unit that determines whether the data set includes sensitive data by applying the auxiliary information to a determination model that determines whether the data set includes sensitive data;
A monitoring unit that, when the data set including the sensitive data is specified when generating a learning model, determines whether or not to permit use of the data set in generating the learning model based on an administrator authority of a user who generates the learning model,
The management device further includes a learning model management unit that determines whether or not there is auxiliary information associated with the dataset that is used only in a learning model whose output result accuracy is higher than a threshold value. The management unit
The management device according to claim 1 , further comprising: a decision tree analysis using the auxiliary information to determine whether or not the data set includes the data with special handling capabilities. The management unit
The management device according to claim 1 , further comprising: a data set management unit that manages identification information of the data set in association with flag information indicating whether the data set includes the sensitive data. The monitoring unit is
The management device according to any one of claims 1 to 4, further comprising: a management device that receives information regarding a management level of a user who generates the learning model; and permits the user to use the dataset if the management level of the user is higher than the management level associated with the specified dataset. The computer
Obtaining at least one ancillary piece of information associated with the dataset;
determining whether the data set includes sensitive data by applying the ancillary information to a decision model that determines whether the data set includes sensitive data;
When generating a learning model, if the dataset including the sensitive data is specified, a determination is made as to whether or not to permit use of the dataset in generating the learning model based on an administrator authority of a user generating the learning model;
obtaining data processing information relating to pre-processing or processing performed on the data set;
managing at least one of a pre-processing data set before the pre-processing or the processing is performed and a post-processing data set after the pre-processing or the processing is performed in association with the auxiliary information;
A management method for determining whether there is pre-processing or processing associated with the dataset that is used only in learning models whose output results have accuracy higher than a threshold. The computer
Obtaining at least one ancillary piece of information associated with the dataset;
determining whether the data set includes sensitive data by applying the ancillary information to a decision model that determines whether the data set includes sensitive data;
When generating a learning model, if the dataset including the sensitive data is specified, a determination is made as to whether or not to permit use of the dataset in generating the learning model based on an administrator authority of a user generating the learning model;
A management method for determining whether or not there is auxiliary information associated with the dataset that is used only in a learning model whose output result accuracy is higher than a threshold. Obtaining at least one ancillary piece of information associated with the dataset;
determining whether the data set includes sensitive data by applying the ancillary information to a decision model that determines whether the data set includes sensitive data;
When generating a learning model, if the dataset including the sensitive data is specified, a determination is made as to whether or not to permit use of the dataset in generating the learning model based on an administrator authority of a user generating the learning model;
obtaining data processing information relating to pre-processing or processing performed on the data set;
managing at least one of a pre-processing data set before the pre-processing or the processing is performed and a post-processing data set after the pre-processing or the processing is performed in association with the auxiliary information;
A program that causes a computer to determine whether or not there is preprocessing or processing associated with the dataset that is used only in a learning model whose output results have accuracy higher than a threshold . Obtaining at least one ancillary piece of information associated with the dataset;
determining whether the data set includes sensitive data by applying the ancillary information to a decision model that determines whether the data set includes sensitive data;
When generating a learning model, if the dataset including the sensitive data is specified, a determination is made as to whether or not to permit use of the dataset in generating the learning model based on an administrator authority of a user generating the learning model;
A program that causes a computer to determine whether or not there is auxiliary information associated with the dataset that is used only in a learning model whose output results have accuracy higher than a threshold.

Images