Information processing device and information processing method

Abstract

To ensure reliability of learning data.SOLUTION: An information processing device comprises acquisition means to acquire information generated by the generative AI constructed by the machine learning using user data including personally identifiable information and related to the artificial data not including the personally identifiable information, and generation means to generate a transaction including the information related to the artificial data.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to the technical field of information processing devices and information processing methods. [Background technology]

[0002] As this type of device, for example, a device has been proposed that separates data into sensitive data and non-sensitive data and provides the non-sensitive data to a distributed ledger (see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-511393 Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, the behavior of a trained model constructed by machine learning may be verified. This verification is often performed a certain period of time after the trained model is constructed. For the verification, training data used in the machine learning for constructing the trained model is used, but there is a risk that the reliability of the training data is not guaranteed. Patent Document 1 does not consider the verification of the behavior of the trained model.

[0005] The present invention has been made in view of the above circumstances, for example, and aims to provide an information processing device and an information processing method that can ensure the reliability of training data. [Means for solving the problem]

[0006] An information processing device according to one embodiment of the present invention comprises an acquisition means for acquiring information relating to artificial data that does not contain personally identifiable information, which is generated by a generation AI constructed by machine learning using user data that contains personally identifiable information, and a generation means for generating transactions that include information relating to the artificial data.

[0007] An information processing method according to one embodiment of the present invention includes an acquisition step of acquiring information about artificial data that does not contain personally identifiable information, the artificial data being generated by a generation AI constructed by machine learning using user data that contains personally identifiable information; a generation step of generating transactions that include information about the artificial data; and a storage step of storing the transactions in a distributed ledger. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a conceptual diagram illustrating an example of machine learning. [Figure 2] 1 is a block diagram showing a configuration of an information processing system according to an embodiment. [Figure 3] 4 is a flowchart showing the operation of the information processing system according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] An embodiment of an information processing device and an information processing method will be described with reference to FIGS.

[0010] First, an example of machine learning will be described with reference to FIG. 1. In FIG. 1, user data is registered in a database DB. Here, at least a portion of the user data may be data including information that can identify an individual. The user data may be, for example, image data including the user's face. In other words, the user data may be image data related to an image in which the user's face appears. Such user data may be generated, for example, by capturing an image of a person riding in a vehicle using a camera that captures an image of the interior of the vehicle (for example, a driver monitor camera).

[0011] The generating device 10 has a generating AI (Artificial Intelligence) 11. The generating AI 11 is a generating AI constructed by learning using user data (typically, multiple user data) registered in a database DB. Various existing methods can be applied to the construction method of the generating AI 11. Therefore, detailed description of the construction method of the generating AI 11 will be omitted. The generating AI 11 may be constructed by fine-tuning, in which a pre-trained model (for example, a base model) is re-trained using user data.

[0012] The generation device 10 may receive instructions from an operator of the generation device 10. The instructions from the operator may be expressed as text information. The generation device 10 may input the instructions from the operator to the generation AI 11. The generation AI 11, to which the instructions from the operator have been input, may generate artificial data in accordance with the instructions. In other words, the generation device 10 may generate artificial data using the generation AI 11.

[0013] For example, if the user data is image data including the user's face, the generation AI 11, which has learned using the user data, may be able to generate new image data including the face as the artificial data. For example, if the operator's instruction is "the face of a woman in her 70s," the generation AI 11 may generate image data including the face of a fictitious woman in her 70s as the artificial data. For example, if the operator's instruction is "the face of a man in his 20s," the generation AI 11 may generate image data including the face of a fictitious man in his 20s as the artificial data. Note that the generation AI may generate one piece of artificial data or multiple pieces of artificial data in response to one instruction from the operator.

[0014] For example, image data including a user's face corresponds to personal information. Therefore, image data including a user's face as user data can be said to include information that can identify an individual. In contrast, image data including the face of a fictitious person generated by the generation AI 11 does not correspond to information that can identify an individual. Therefore, artificial data can be said to be data that does not include information that can identify an individual. As described above, by generating artificial data using the generation AI 11, it is possible to generate a large amount of data that does not include information that can identify an individual.

[0015] The learning device 20 has a computational model 21. The computational model 21 is a model before machine learning is performed (for example, a model in which weighting is not optimized). The learning device 20 may perform learning of the computational model 21 using artificial data (typically, a plurality of pieces of artificial data) generated by the generation device 10 (specifically, the generation AI 11). Note that various existing aspects can be applied to the learning method of the computational model 21. Therefore, a detailed description of the learning method of the computational model 21 will be omitted.

[0016] Next, an information processing system 1 according to an embodiment will be described with reference to Fig. 2 and Fig. 3. In Fig. 2, the information processing system 1 includes an information processing device 100 and a distributed network 200 having a plurality of nodes. The information processing device 100 may be, for example, a server device such as an application server, or may be, for example, a terminal device such as a personal computer.

[0017] The information processing system 1 registers information about the artificial data used in the machine learning of the computational model 21 in a distributed ledger (e.g., a blockchain) realized by the distributed network 200. For example, the information processing system 1 may function as a database related to the artificial data.

[0018] The information processing device 100 may include an arithmetic device 110 and a storage device 120. The arithmetic device 110 may include, for example, at least one of a central processing unit (CPU) and a graphics processing unit (GPU). The storage device 120 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. In other words, it can be said that the information processing device 100 includes at least one processor and at least one memory.

[0019] The storage device 120 may store a computer program 121. The arithmetic device 110 may execute processing to be performed by the information processing device 100 together with the storage device 120 in which the computer program 121 is stored (in other words, together with the storage device 120 and the computer program 121 stored in the storage device 120). For example, the arithmetic device 110 may execute the computer program 121 to realize logical functional blocks within the arithmetic device 110 for executing processing to be performed by the information processing device 100. For example, the arithmetic device 110 may have an acquisition unit 111 and a generation unit 112 as functional blocks.

[0020] The acquisition unit 111 of the calculation device 110 acquires information about the artificial data from the generation device 10 (see FIG. 1 ). The information about the artificial data may be, for example, the artificial data itself, information for identifying the artificial data, information for specifying the artificial data, or information indicating the location of the artificial data.

[0021] An example of information for identifying artificial data is a hash value generated from the artificial data. An example of information for specifying artificial data is a file name related to the artificial data. An example of information indicating the location of artificial data is information indicating the location where the artificial data is recorded (e.g., Uniform Resource Locator: URL).

[0022] The generation unit 112 of the computing device 110 generates a transaction including information about the artificial data. The computing device 110 may transmit the generated transaction to the distributed network 200. At least one node constituting the distributed network 200 may verify the transaction transmitted from the information processing device 100 and then register the transaction in the distributed ledger.

[0023] If the information about the artificial data is the artificial data itself, the computing device 110 may generate a transaction including the artificial data. In this case, the artificial data will be registered in the distributed ledger. If the information about the artificial data is not the artificial data itself (for example, if the information about the artificial data is a hash value generated from the artificial data), the artificial data may be stored in, for example, the storage device 120 of the information processing device 100, or the information processing system 1 may include a device that stores artificial data, different from the information processing device 100.

[0024] Next, the operation of the information processing system 1 will be described with reference to the flowchart in Fig. 3. In Fig. 3, the acquisition unit 111 of the calculation device 110 of the information processing device 100 acquires information about the artificial data from the generation device 10 (step S101). Next, the generation unit 112 of the calculation device 110 generates a transaction including information about the artificial data (step S102). Thereafter, the calculation device 110 may transmit the transaction to the distributed network 200. At least one node constituting the distributed network 200 may register the transaction in the distributed ledger (step S103).

[0025] (Technical Effects) After a trained model is constructed by performing machine learning of a computational model (e.g., computational model 21), the behavior of the trained model may be verified. For the verification, learning data used in the machine learning of the computational model is used.

[0026] Meanwhile, service providers may collect user data with the user's consent. The service providers may use the collected user data as training data to perform machine learning of a computational model. On the other hand, the user may request the service provider to delete the user data collected by the service provider (in other words, provided by the user). If at least a portion of the user data as training data is deleted in accordance with a request to delete the user data, it may become difficult to verify the behavior of the trained model.

[0027] In contrast to this, in the present embodiment, machine learning of the computational model 21 is performed using artificial data generated by the generation device 10 (specifically, the generation AI 11). That is, user data registered in the database DB is not used in the machine learning of the computational model 21. Therefore, artificial data is used to verify the behavior of the trained model constructed by performing machine learning of the computational model 21. Therefore, even if user data registered in the database DB is deleted due to a request to delete the user data, this does not affect the verification of the behavior of the trained model using the artificial data.

[0028] In this embodiment, in particular, information about the artificial data used in the machine learning of the computational model 21 is registered in a distributed ledger. It is extremely difficult to tamper with information about the artificial data registered in a distributed ledger. Therefore, even if the artificial data is tampered with, by referencing the information about the artificial data registered in the distributed ledger, it is possible to detect the tampered artificial data. For example, if the artificial data itself is registered in a distributed ledger as information about the artificial data, it is possible to prevent the artificial data from being tampered with. In other words, according to this embodiment, it is possible to prevent a decrease in the reliability of the artificial data.

[0029] As described above, according to this embodiment, it is possible to protect the user's right to request data deletion and ensure the reliability of the artificial data used to verify the behavior of the trained model. Since the reliability of the artificial data is ensured, it is expected that the trained model will be verified appropriately.

[0030] Various aspects of the invention derived from the above-described embodiments will be described below.

[0031] An information processing device according to one aspect of the invention includes an acquisition unit that acquires information about artificial data that does not contain personally identifiable information and that is generated by a generation AI constructed by machine learning using user data that contains personally identifiable information, and a generation unit that generates a transaction that includes information about the artificial data. In the above-described embodiment, the acquisition unit 111 corresponds to an example of the acquisition unit, and the generation unit 112 corresponds to an example of the generation unit.

[0032] Here, the transaction may include a hash value generated from the artificial data as information about the artificial data. Alternatively, the transaction may include the artificial data as information about the artificial data. The artificial data may be data used for machine learning of a computational model different from the generation AI.

[0033] An information processing method according to one embodiment of the invention includes an acquisition step of acquiring information about artificial data that does not contain personally identifiable information, the artificial data being generated by a generation AI constructed by machine learning using user data that contains personally identifiable information; a generation step of generating transactions that include information about the artificial data; and a storage step of storing the transactions in a distributed ledger.

[0034] The present invention is not limited to the above-described embodiments, but can be modified as appropriate within the scope that does not contradict the gist or idea of ​​the invention that can be read from the claims and the entire specification, and information processing devices and information processing methods that involve such modifications are also included in the technical scope of the present invention. [Explanation of symbols]

[0035] 1... information processing system, 100... information processing device, 111... acquisition unit, 112... generation unit

Claims

1. An acquisition means for acquiring information about artificial data that does not contain personally identifiable information, which is generated by a generation AI constructed by machine learning using user data that contains personally identifiable information; generating means for generating a transaction including information about the artificial data; An information processing device comprising:

2. The transaction includes, as information about the artificial data, a hash value generated from the artificial data.

2. The information processing apparatus according to claim 1, wherein:

3. The transaction includes the artificial data as information about the artificial data.

2. The information processing apparatus according to claim 1, wherein:

4. The artificial data is data used for machine learning of a computational model different from the generated AI.

2. The information processing apparatus according to claim 1, wherein:

5. An acquisition step of acquiring information about artificial data that does not contain personally identifiable information, which is generated by a generation AI constructed by machine learning using user data that contains personally identifiable information; generating a transaction including information about the artificial data; storing the transaction in a distributed ledger; An information processing method comprising:

Images