Information Processing System

Abstract

The information processing system includes a generating unit that generates a plurality of synthetic data that are used for training a calculation model different from the generative AI and do not include information that can identify an individual, by training the generative AI constructed using user data that includes information that can identify an individual.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing systems. Background Art

[0002] As such a system, for example, a system has been proposed in which, when hiding personal information, image data to which personal information is added is copied, and the image data is stored as a fictitious subject while hiding the personal information (Japanese Patent Application Laid-Open No. 2009-118889). Summary of the invention

[0003] According to the provisions related to the right to deletion (the so-called right to be forgotten), if there is a request from an individual to delete personal data, the personal data must be deleted. However, there are cases where personal data is used as training data for calculation models involved in Artificial Intelligence (AI). Training data is also used, for example, to verify the performance of AI. There is a technical problem that if personal data used as training data is deleted due to a request to delete personal data, it becomes difficult to verify the performance of AI.

[0004] The present invention has been achieved in view of the above-mentioned problems, and an object of the present invention is to provide an information processing system capable of achieving both compliance with regulations related to the right of deletion and verification of the performance of AI.

[0005] An information processing system involved in one embodiment of the present invention has a generation unit that generates a plurality of synthetic data (synthetic data) for training a computing model different from the generative AI and which does not contain information that can identify an individual by training a generative AI constructed by training using user data that contains information that can identify an individual. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Hereinafter, features, advantages, and technical and industrial significance of exemplary embodiments of the present invention will be described with reference to the accompanying drawings, in which like reference numerals denote like components, wherein: Figure 1 This is a schematic diagram schematically showing an example of an information processing system involved in the embodiment. Figure 2 This is a flowchart showing the operation of the determination device according to the embodiment. Figure 3 It is a schematic diagram schematically showing another example of the information processing system involved in the embodiment. DETAILED DESCRIPTION

[0007] Reference Figure 1 , describing the implementation methods involved in the information processing system. Figure 1 In the embodiment, the information processing system 1 includes a generating device 10, a determining device 20, and a training device 30. The information processing system 1 includes a database DB1 including user data and a database DB2 including synthetic data.

[0008] Each of the generating device 10, the determining device 20, the training device 30 and the extracting device 40 may include, as hardware, for example, a processing unit, a storage unit and a communication unit. The processing unit may include, for example, at least one of a Central Processing Unit (CPU) and a Graphics Processing Unit (GPU). The storage unit may include, for example, at least one of a Random Access Memory (RAM), a Read Only Memory (ROM), a hard disk device, a magneto-optical disk device, a Solid State Drive (SSD) and an optical disk array.

[0009] The generating device 10 has a generative AI 11. The generative AI 11 is a generative AI constructed by training using user data (typically, a plurality of user data) included in the database DB1. In addition, various existing methods can be applied as a construction method of the generative AI 11. Therefore, the detailed description of the construction method of the generative AI 11 is omitted.

[0010] The user data included in the database DB1 will be described. The user data may be, for example, image data including the face of the user. That is, the user data may be image data related to an image of the face of the user. Such user data may be generated, for example, by a camera (such as a driver monitoring camera) that takes pictures of the interior of the vehicle capturing a person riding in the vehicle.

[0011] In the case where the user data is image data including the user's face, the generative AI 11 constructed by training using the user data may generate new image data including the face. For example, the user data may be image data including the face of the driver of the vehicle captured by a driver monitoring camera. In this case, the generative AI 11 constructed by training using the user data may generate image data having a field of view angle and a head angle corresponding to the field of view angle and the head angle involved in the image data generated by the driver monitoring camera capturing the driver of the vehicle. In addition, the generative AI 11 may also be constructed by fine-tuning a pre-trained model (e.g., a base model) by retraining it using user data.

[0012] The generating device 10 may also receive instructions from an operator of the generating device 10. The instructions from the operator may be expressed by text information. The generating device 10 may input the instructions from the operator to the generating AI 11. The generating AI 11 that has input the instructions from the operator may generate synthetic data corresponding to the instructions. That is, the generating device 10 may generate synthetic data using the generating AI 11.

[0013] For example, when the operator's instruction is "face of a woman in her 70s", the generative AI 11 may generate image data containing a fictitious face of a woman in her 70s as the above-mentioned synthetic data. For example, when the operator's instruction is "face of a man in his 20s", the generative AI 11 may generate image data containing a fictitious face of a man in his 20s as the above-mentioned synthetic data. In addition, the generative AI may generate one piece of synthetic data or a plurality of pieces of synthetic data for one operator's instruction.

[0014] User data often contains information that can identify an individual. For example, image data containing a user's face is personal information, so it can be said that image data containing a user's face contains information that can identify an individual. Image data containing the face of a fictional character generated by the generative AI 11 can be said to be image data that does not contain information that can identify an individual. That is, by using the generative AI 11 to generate synthetic data, a large amount of data that does not contain information that can identify an individual can be generated.

[0015] The determination device 20 determines whether the synthetic data generated by the generation device 10 (specifically, the generative AI 11) is similar to the user data included in the database DB1. The determination device 20 includes a calculation unit 21, a determination unit 22, and a deletion unit 23. In addition, the calculation unit 21, the determination unit 22, and the deletion unit 23 can be implemented as logical function blocks or as physical processing circuits.

[0016] The calculation unit 21 calculates a similarity indicating the degree of similarity between one of the plurality of synthesized data generated by the generation device 10 and each of the plurality of user data included in the database DB1. As a method for calculating the similarity, various existing methods can be applied. An example of a method for calculating the similarity is described below. For example, the user data is set to include image data of the face of an actual person, and the synthesized data is set to include image data of the face of a fictitious person. In addition, the user data is set to include coordinate information indicating the coordinates of one or more feature points related to the face. The synthesized data is also set to include coordinate information indicating the coordinates of one or more feature points related to the face. The feature point related to the face may be, for example, at least one of the right eye, the left eye, the nose, and the mouth. For example, the calculation unit 21 may compare the coordinates of the first feature point indicated by the coordinate information included in the user data with the coordinates of the feature point corresponding to the first feature point indicated by the coordinate information included in the synthesized data. The calculation unit 21 may process that the shorter the distance between the coordinates of the first feature point and the coordinates of the feature point corresponding to the first feature point, the higher the similarity. The calculation unit 21 may, for example, compare the relative relationship between the coordinates of the first feature point and the coordinates of the second feature point with the relative relationship between the coordinates of the feature point corresponding to the first feature point and the coordinates of the feature point corresponding to the second feature point represented by the coordinate information contained in the synthetic data, wherein the coordinates of the first feature point and the coordinates of the second feature point are represented by the coordinate information contained in the user data. The calculation unit 21 may, for example, process that the smaller the difference between the distance between the first feature point and the second feature point as the relative relationship and the distance between the feature point corresponding to the first feature point and the feature point corresponding to the second feature point as the relative relationship, the higher the similarity. The calculation unit 21 may also compare the feature quantity related to the user data with the feature quantity related to the synthetic data. The calculation unit 21 may, for example, calculate the distance between the feature quantity vector corresponding to the feature quantity related to the user data and the feature quantity vector corresponding to the feature quantity related to the synthetic data, and the shorter the calculated distance, the higher the similarity. The calculation unit 21 may, for example, calculate the cosine similarity based on the direction of the feature quantity vector corresponding to the feature quantity related to the user data and the direction of the feature quantity vector corresponding to the feature quantity related to the synthetic data.

[0017] The determination unit 22 determines whether the similarity calculated by the calculation unit 21 is greater than a specified value. The deletion unit 23 deletes a piece of synthetic data whose similarity is determined to be greater than a specified value by the determination unit 22. Here, the "specified value" is a value that determines whether to delete a piece of synthetic data. The specified value may be a predetermined fixed value. In addition, the specified value may also be a variable value corresponding to certain parameters. The determination device 20 registers a piece of synthetic data whose similarity is determined to be less than a specified value by the determination unit 22 in the database DB2.

[0018] For example, it is assumed that the generative AI 11 generates image data including the face of a fictitious person as synthetic data. In this case, synthetic data that allows a specific person (i.e., a real person) to be thought of from the face of the fictitious person is sometimes generated. That is, synthetic data including information equivalent to information that can identify an individual is sometimes generated by the generative AI 11. Therefore, the determination device 20 prevents the situation where synthetic data including information equivalent to information that can identify an individual is stored in the database DB2 by deleting a synthetic data that is determined to have a similarity greater than a specified value as described above. From this perspective, for example, the range of possible values ​​of the similarity that can identify an individual based on the synthetic data can be obtained, and the above-mentioned specified value is set as the lower limit of the obtained range.

[0019] Here, refer to Figure 2 The operation of the determination device 20 is described in detail with reference to the flowchart of FIG. Figure 2 In the example, the calculation unit 21 of the determination device 20 calculates the similarity between the one synthetic data generated by the generation device 10 and each of the plurality of user data included in the database DB1 (S101). As a result of the processing of S101, the calculation unit 21 may calculate a plurality of similarities.

[0020] The determination unit 22 determines whether the similarity calculated by the calculation unit 21 is greater than or equal to a predetermined value (S102). When at least one of the plurality of similarities calculated by the calculation unit 21 is greater than or equal to the predetermined value, the determination unit 22 may determine that the similarity is greater than or equal to the predetermined value in the process of S102. On the other hand, when all of the plurality of similarities calculated by the calculation unit 21 are less than the predetermined value, the determination unit 22 may determine that the similarity is not greater than or equal to the predetermined value (that is, the similarity is less than the predetermined value) in the process of S102.

[0021] When it is determined in the process of S102 that the similarity is greater than the prescribed value (S102: Yes), the deletion unit 23 deletes the above-mentioned one synthetic data (S103). On the other hand, when it is determined in the process of S102 that the similarity is not greater than the prescribed value (S102: No), the determination device 20 registers the above-mentioned one synthetic data in the database DB2 (S104). In addition, Figure 2 The actions shown can be performed each time the generating device 10 generates synthesized data.

[0022] return Figure 1The training device 30 has a computational model 31. The computational model 31 is a model before machine learning (for example, a model whose weights are not optimized). The training device 30 uses synthetic data (typically a plurality of synthetic data) contained in the database DB2 to train the computational model 31. In addition, various existing methods can be applied as a training method for the computational model 31. Therefore, the detailed description of the training method for the computational model 31 is omitted. Technical Effects

[0023] For example, as user data, a business providing services to the user may collect data after obtaining the user's consent. For example, in the case of a business providing a service for detecting at least one of distracted driving and drowsy driving from an image captured inside a vehicle, image data containing the face of the user driving the vehicle may be collected after obtaining the user's consent.

[0024] For example, a trained model for detecting at least one of distracted driving and drowsy driving may be constructed by training the computation model 31 through the training device 30. Assume that the trained model is constructed by training the computation model 31 using user data. Then, if the user data is deleted due to a deletion request from the user, it may be difficult to verify the performance of the trained model.

[0025] In contrast, in the information processing system 1, synthetic data generated by the generating device 10 (specifically, the generative AI 11) is used in the training of the computational model 31. Therefore, even if the user data is deleted due to the above-mentioned deletion request, it will not affect the synthetic data included in the database DB2. Thus, according to the information processing system 1, it is possible to achieve both compliance with regulations related to the right to delete and verification of the performance of the trained model (in other words, AI).

[0026] In addition, the generating device 10, the determining device 20 and the training device 30 are not limited to being independent devices. For example, the generating device 10, the determining device 20 and the training device 30 may be implemented by a single device having multiple functions corresponding to at least two of the generating device 10, the determining device 20 and the training device 30. In addition, the user data is not limited to the above-mentioned image data containing the user's face. The user data may also be, for example, sound data. As shown above, in the information processing system 1, the training model 31 is trained by the training device 30. Therefore, the information processing system 1 may also be referred to as a training system. Modifications

[0027] Reference Figure 3 , a modification example of the implementation method involved in the information processing system is described. Figure 3In the example, the information processing system 2 includes a generating device 10, a determining device 20, a training device 30, and an extracting device 40. The information processing system 2 includes a database DB1 including user data, a database DB2 including synthetic data, and a database DB3 including feature quantities.

[0028] The extraction device 40 extracts the feature quantity of the user data contained in the database DB1. The extraction device 40 can extract all the feature quantities of the multiple user data contained in the database DB1. In addition, as an extraction method for extracting the feature quantity from the user data, various existing methods can be applied. Therefore, the detailed description of the extraction method for extracting the feature quantity is omitted. The extraction device 40 registers the feature quantity extracted from the user data in the database DB3. Here, the user data cannot be restored based on the feature quantity extracted from the user data. In addition, the user data cannot be imagined based on the feature quantity. Therefore, it can be said that the feature quantity is data that does not contain information that can identify an individual.

[0029] The calculation unit 21 of the determination device 20 can calculate the similarity using one of the multiple synthetic data generated by the generation device 10 and the multiple feature quantities included in the database DB3. For example, the calculation unit 21 can extract the feature quantity of one synthetic data. The calculation unit 21 can calculate the similarity by comparing the feature quantity extracted from one synthetic data with each of the multiple feature quantities included in the database DB3. Technical Effects

[0030] There is a possibility that after a piece of user data included in the database DB1 is deleted due to a request for deletion of user data from a user, synthetic data similar to the user data may be generated by the generating device 10. When the calculating unit 21 of the determining device 20 calculates the similarity between a piece of synthetic data and the user data included in the database DB1, the similarity between a piece of synthetic data and the user data deleted from the database DB1 cannot be calculated. Therefore, there is a possibility that the synthetic data similar to the user data may be registered in the database DB3.

[0031] In contrast, in the information processing system 2, the calculation unit 21 can calculate the similarity using one of the multiple synthetic data generated by the generating device 10 and the multiple feature quantities included in the database DB3. As described above, the feature quantity is data that does not include information that can identify an individual, so even if there is the above-mentioned deletion request, a feature quantity corresponding to one user data for which there is a deletion request will not be deleted. Thus, according to the information processing system 2, even if there is a deletion request from a user for user data and one user data included in the database DB1 is deleted, it is possible to prevent synthetic data that is very similar to the deleted user data from being registered in the database DB3.

[0032] Various aspects of the invention derived from the above-described embodiment and modifications will be described below.

[0033] An information processing system according to one embodiment of the invention includes a generating unit that generates a plurality of synthetic data that does not include information that can identify an individual and is used for training a computing model different from the generating AI and is constructed by training using user data that includes information that can identify an individual. In the above-mentioned embodiment, the generating device 10 is equivalent to an example of the generating unit.

[0034] The information processing system may also include: a calculation unit that calculates a similarity indicating the degree of similarity between one of the plurality of synthetic data and the user data; a determination unit that determines whether the similarity is greater than a specified value; and a deletion unit that deletes the one synthetic data when it is determined that the similarity is greater than the specified value. In the above-mentioned embodiment, the calculation unit 21 is equivalent to an example of the calculation unit, the determination unit 22 is equivalent to an example of the determination unit, and the deletion unit 23 is equivalent to an example of the deletion unit.

[0035] Here, the user data and the plurality of synthesized data may each be image data including a face, and the user data and the plurality of synthesized data may each include coordinate information indicating coordinates of one or more feature points related to the face. The calculation unit may calculate the similarity based on the coordinate information included in the user data and the coordinate information included in the one synthesized data.

[0036] The information system may also include: a database including synthetic data having a similarity less than the specified value among the plurality of synthetic data, wherein the similarity indicates the degree of similarity between the synthetic data and the user data; and a training unit for training the calculation model using the synthetic data included in the database. In the above-mentioned embodiment, the database DB2 is equivalent to an example of a database, and the training device 30 is equivalent to an example of a training unit.

[0037] The information system may also include: an extraction unit that extracts a feature value of the user data from the user data; and a storage unit that stores the feature value, and the calculation unit may also calculate the similarity using the feature value and the one synthetic data. In the above-mentioned embodiment, the extraction device 40 is equivalent to an example of the extraction unit, and the database DB3 is equivalent to an example of the storage unit.

[0038] The present invention is not limited to the above-described embodiments, and can be appropriately modified without departing from the gist or idea of ​​the invention as obtained from the claims and the entire specification. Information processing systems accompanying these modifications are also included in the technical scope of the present invention.

Claims

1. An information processing system, characterized in that: A generating unit is provided for generating a plurality of synthetic data that are used for training a computing model different from the generative AI and do not contain information that can identify an individual, by training a generative AI constructed by using user data that contains information that can identify an individual.

2. The information processing system according to claim 1, characterized in that have: a calculation unit that calculates a similarity indicating a degree of similarity between one of the plurality of synthetic data and the user data; A determination unit, determining whether the similarity is greater than a specified value; as well as The deleting unit deletes the one piece of synthesized data when it is determined that the similarity is equal to or greater than the predetermined value.

3. The information processing system according to claim 2, characterized in that: The user data and the plurality of synthesized data are each image data including a face, The user data and the plurality of synthesized data each include coordinate information indicating coordinates of one or more feature points associated with a face, The calculation unit calculates the similarity based on the coordinate information included in the user data and the coordinate information included in the one piece of synthesized data.

4. The information processing system according to claim 2, characterized in that: have: a database comprising synthetic data having a similarity less than the specified value among the plurality of synthetic data, wherein the similarity indicates a degree of similarity between the synthetic data and the user data; as well as A training unit is used to train the computational model using the synthetic data contained in the database.

5. The information processing system according to claim 2, characterized in that: have: an extraction unit, extracting a feature amount of the user data from the user data; and a storage unit for storing the feature quantity, The calculation unit calculates the similarity using the feature amount and the one synthetic data.

Images