Information management system and information management method

Abstract

To provide an information management system and an information management method for managing the authenticity of a learning model.SOLUTION: An information management system A1 includes a computer connected to a distributed ledger B1 of a block chain network BN1, and manages a learning model. An execution system 20 acquires first authenticity information of teacher data, acquires second authenticity information of a learning algorithm, and acquires third authenticity information of the learning model generated by inputting the teacher data to the learning algorithm. A smart contract SC1 then registers a first non-fungible token including the first authentic information, the second authentic information, and the third authentic information in the distributed ledger B1.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to an information management system and an information management method for managing learning models. [Background technology]

[0002] AI (Artificial Intelligence) is used in various fields. However, there are cases where the learning model used in AI does not output an appropriate response. For this reason, as AI becomes more widespread, technology for determining problems with newly generated AI is also being considered in order to evaluate the generated learning model (see, for example, Patent Document 1). The server described in Patent Document 1 acquires learning conditions for generating a new learning model. Next, it identifies the AI ​​ethics that the learning model must satisfy. Then, it is determined whether or not the newly generated learning model satisfies the AI ​​ethics based on the learning conditions and the AI ​​ethics. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2023-64636 A Summary of the Invention [Problem to be solved by the invention]

[0004] The technology described in Patent Document 1 can determine the appropriateness of a learning model. However, when using a learning model, it is not possible to know what kind of teacher data or learning algorithm was used by machine learning to generate the learning model. If it is not possible to know whether the learning model was generated under legitimate circumstances, it may be difficult to use the learning model appropriately. [Means for solving the problem]

[0005] An information management system for solving the above problem includes a computer connected to a distributed ledger of a blockchain network, and manages a learning model. The computer acquires first authentic information of training data, acquires second authentic information of a learning algorithm, acquires third authentic information of a learning model generated by inputting the training data into the learning algorithm, and registers a first non-fungible token including the first authentic information, the second authentic information, and the third authentic information in the distributed ledger. Effect of the Invention

[0006] According to the present invention, it is possible to manage the accurate generation of learning models and ensure their authenticity. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is an explanatory diagram of an information management system according to an embodiment. [Diagram 2] FIG. 2 is an explanatory diagram of a hardware configuration of the embodiment. [Diagram 3] FIG. 2 is an explanatory diagram of information recorded in an NFT according to an embodiment. [Figure 4] FIG. 4 is an explanatory diagram of generation of each data item according to the embodiment. [Diagram 5] FIG. 2 is an explanatory diagram of a processing procedure according to an embodiment. [Figure 6] FIG. 2 is an explanatory diagram of a processing procedure according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Hereinafter, an embodiment of an information management system and an information management method will be described with reference to Figures 1 to 6. In this embodiment, information on the authenticity of AI is published on a blockchain network, thereby enabling accurate AI to be used.

[0009] As shown in Fig. 1, the information management system A1 that manages this AI uses a utilization system 10, an execution system 20, a data system 30, a learning system 40, etc., which are interconnected via a network such as the Internet. The utilization system 10 and the execution system 20 are connected to a blockchain network BN1 via an API.

[0010] (Hardware configuration) 2, the hardware configuration of the information processing device H10 constituting the utilization system 10, execution system 20, data system 30, and learning system 40 will be described. The information processing device H10 includes a communication device H11, an input device H12, a display device H13, a storage device H14, and a processor H15. Note that this hardware configuration is an example, and it is also possible to realize the information processing device H10 using other hardware.

[0011] The communication device H11 is an interface that establishes a communication path with other devices and transmits and receives data. The input device H12 is a device that accepts input from a user, etc. The display device H13 is a display or the like that displays various information.

[0012] The storage device H14 is a storage device that stores data and various programs for executing various functions of the utilization system 10, the execution system 20, the data system 30, and the learning system 40. Examples of the storage device H14 include a ROM, a RAM, and a hard disk.

[0013] The processor H15 uses the programs and data stored in the storage device H14 to control each process in the utilization system 10, the execution system 20, the data system 30, and the learning system 40. Examples of the processor H15 include a CPU and an MPU. The processor H15 expands the programs stored in the ROM, etc., into the RAM and executes various processes for each service.

[0014] The processor H15 is not limited to a processor that performs software processing for all the processes it executes. For example, the processor H15 may include a dedicated hardware circuit (e.g., an application specific integrated circuit (ASIC)) that performs hardware processing for at least a part of the processes it executes. That is, the processor H15 may be configured as follows:

[0015] [1] One or more processors that operate according to a computer program (software). [2] One or more dedicated hardware circuits that perform at least some of the various processes; or [3] Circuits that include combinations of these The processor includes a CPU and memory, such as RAM and ROM, that stores program codes or instructions configured to cause the CPU to execute processes. Memory, i.e., computer-readable media, includes any available media that can be accessed by a general-purpose or special-purpose computer.

[0016] (System Configuration) Next, the functions of the utilization system 10, execution system 20, data system 30, learning system 40, etc. will be described with reference to FIGS.

[0017] The utilization system 10, execution system 20, data system 30, and learning system 40 each include an application execution unit and executes an application stored in a storage unit. The utilization system 10 is a computer device used by a user who utilizes the learning model.

[0018] The execution system 20 is a computer device that executes machine learning using a learning algorithm and teacher data. The execution system 20 generates a learning model by performing machine learning using the teacher data in the learning algorithm.

[0019] The data system 30 is a computer device that accumulates training data used in machine learning. The learning system 40 is a computing device that manages the learning algorithms that implement machine learning.

[0020] The blockchain network BN1 includes a distributed ledger B1 that stores information in a distributed manner. In this embodiment, the distributed ledger B1 includes a utilization system 10 and an execution system 20. By using this blockchain network BN1, data tamper resistance and availability are achieved. The address of each user on the blockchain network BN1 is generated from an ECC (Elliptic Curve Cryptography) signature key. This signature key is a private key for signing transactions on the blockchain network BN1.

[0021] A smart contract SC1 that generates an NFT is registered on this blockchain network BN1. This smart contract SC1 generates an NFT (Non-Fungible Token) when registering a learning model and registers it in the distributed ledger B1 of the blockchain network BN1. In this embodiment, the smart contract SC1 generates an NFT related to the generation of the learning model and registers it in the blockchain network BN1.

[0022] NFTs are digital assets that are issued and traded based on blockchain technology (e.g., ERC-721) and are used to prove ownership.

[0023] As shown in Figure 3, this NFT 500 has index data and metadata recorded on it. Index data is information necessary to prove the uniqueness of an NFT, and includes the NFT's token identifier, issuer, owner, transaction history, etc.

[0024] The token identifier is an identifier for identifying this NFT. The issuer information is information for identifying the issuer of this NFT and is information of the administrator of the execution system 20.

[0025] The owner information is information used to identify the owner of this NFT, and initially matches the issuer information. The transaction history information is information on the history of transactions of this NFT.

[0026] Metadata is information that indicates the characteristics and value of an NFT, and includes the NFT's name, location of the learning model, amount issued, description, etc. The name information is information that indicates the name of this NFT. The learning model location information is information regarding the storage location (address) where the learning model is stored.

[0027] The issuance amount information is information regarding the issuance amount of this NFT. The description information includes a base, a condition, a hash value, and a signature. The base information is information that indicates the preceding learning model on which the learning model is based. The condition information is information regarding the branching conditions and usage conditions of this NFT. The hash value information is an irregular character string generated from the original data by a hash function. In this embodiment, hash values ​​calculated from the teacher data, the learning algorithm, and the learning model are used. The signature information is information for verifying the authenticity of each administrator using a public key system. In this embodiment, the signatures of the administrators of the data system 30, the learning system 40, and the execution system 20 are used.

[0028] As shown in FIG. 4, in this embodiment, hash values ​​h1, h2, and h3 are calculated from the teacher data D1, the learning algorithm D2, and the learning model D3, respectively. Then, using the encryption keys of the administrators of the data system 30, the learning system 40, and the execution system 20, a teacher administrator signature SG1, a learning administrator signature SG2, and a runner signature SG3 are generated from each of the hash values ​​h1, h2, and h3. Then, as described later, an NFT500 including the hash values ​​(h1 to h3) and the signatures (SG1 to SG3) is issued by the smart contract SC1. Using this NFT500, the authenticity of the teacher data D1, the learning algorithm D2, and the learning model D3 can be confirmed by the public keys of the administrators of each signature.

[0029] (NFT issuing process) Next, the NFT issuance process will be explained using Figure 5. First, the execution system 20 executes a condition designation process (step S01). Specifically, the execution system 20 acquires a condition designated by an administrator of the execution system 20 (execution administrator).

[0030] Next, the execution system 20 executes a hashing process on the teacher data (step S02). Specifically, the execution system 20 identifies the teacher data specified by the user. In this case, the execution system 20 transmits a teacher data request to the data system 30. 4, the data system 30 that has received the teacher data request inputs the teacher data D1 corresponding to the teacher data request into a hash function to calculate a hash value h1. The execution system 20 then obtains the hash value h1 from the data system 30.

[0031] Next, the execution system 20 executes a process to obtain the administrator signature of the teacher data (step S03). Specifically, the data system 30 applies the private key of the administrator of the teacher data (teacher administrator) to the calculated hash value to generate the teacher administrator signature (first authentic information).

[0032] 4, a teacher administrator signature SG1 is generated for the hash value h1. Then, the execution system 20 acquires the teacher administrator signature SG1 from the data system 30.

[0033] Next, the execution system 20 executes a hashing process of the learning algorithm (step S04). Specifically, the execution system 20 identifies the learning algorithm specified by the user. In this case, the execution system 20 transmits a learning request to the learning system 40. 4, the learning system 40 that has received the learning request inputs the learning algorithm D2 into a hash function to calculate a hash value h2. The execution system 20 then obtains the hash value h2 from the learning system 40.

[0034] Next, the execution system 20 executes a process for acquiring the administrator signature of the learning system (step S05). Specifically, the learning system 40 applies the private key of the administrator (learning administrator) of the learning algorithm to the calculated hash value to generate the learning administrator signature (second authentic information).

[0035] 4, a learning administrator signature SG2 is generated for the hash value h2. Then, the execution system 20 acquires the learning administrator signature SG2 from the learning system 40.

[0036] Next, the execution system 20 executes a learning process (step S06). Specifically, the execution system 20 uses the teacher data of the data system 30 to send an instruction to the learning system 40 to execute machine learning by the learning algorithm of the learning system 40. In this case, the learning system 40 acquires the teacher data from the data system 30. Then, the learning system 40 performs machine learning by the learning algorithm using the teacher data acquired from the data system 30. Then, the execution system 20 acquires a learning model generated by machine learning from the learning system 40. Then, the execution system 20 records the learning model in a memory area at a specified address.

[0037] Next, the execution system 20 executes a hashing process for the learning model (step S07). Specifically, the execution system 20 inputs the acquired learning model into a hash function to calculate a hash value. In this case, as shown in FIG. 4, a hash value h3 is calculated from the learning model D3.

[0038] Next, the execution system 20 executes a process of acquiring a runner signature of the machine learning (step S08). Specifically, the execution system 20 applies the private key of the administrator (execution administrator) who executes the learning to the calculated hash value to generate a runner signature (third authentic information). In this case, as shown in FIG. 4, a maker signature SG3 is generated from the hash value h3.

[0039] Next, the smart contract SC1 executes the NFT registration process (step S09). Specifically, as shown in FIG. 4, the smart contract SC1 generates an NFT500 (first non-fungible token) that records index data and metadata, and registers it in the distributed ledger B1 of the blockchain network BN1.

[0040] Next, the utilization system 10 executes the utilization process of the learning model (step S10). Specifically, the utilization system 10 acquires the location of the learning model using the NFT. In this case, the utilization system 10 acquires the public keys of the teacher administrator, the learning administrator, and the execution administrator, and verifies the teacher administrator signature, the learning administrator signature, and the execution administrator signature to confirm authenticity. If the authenticity is confirmed, the utilization system 10 utilizes the learning model located in the learning model location.

[0041] (Learning model update process) The learning model update process will be explained using Figure 6. Here, the learning model is updated by re-learning. This re-learning includes updating the learning algorithm and updating the teacher data. In this case, a new NFT issuance process is performed.

[0042] As shown in Fig. 6, in response to the learning model M1, an NFT 501 is registered in the distributed ledger B1 of the blockchain network BN1. In this NFT 501, a condition R1 of the learning model M1 is recorded. Here, assume that a new learning model M2 (relearning model) is generated using second teacher data in which at least one of the learning algorithm or teacher data used to generate the learning model M1 is updated. In this case, an NFT502 (second non-fungible token) corresponding to the learning model M2 is issued in association with an NFT501 corresponding to the learning model M1 (first learning model). In this case, it is recorded in the NFT502 that the NFT501 is the base. And, the condition R2 of the NFT502 inherits the condition R1 of the NFT501. Furthermore, when at least one of the learning algorithm and the teacher data used to generate the learning model M2 is updated and a new learning model M3 is generated, an NFT503 corresponding to the learning model M3 is issued in association with the NFT502 corresponding to the learning model M2. In this case, it is recorded in the NFT503 that the NFT502 is the base. And, the condition R3 of the NFT503 inherits the condition R2 of the NFT502, and thereby inherits the condition R1 of the NFT501.

[0043] In addition to the learning model M2, at least one of the learning algorithm and the teacher data used to generate the learning model M1 may be updated to generate a new learning model M4. In this case, an NFT 504 corresponding to the learning model M4 is issued in association with an NFT 501 corresponding to the learning model M1.

[0044] As described above, according to this embodiment, the following effects can be obtained. (1) In this embodiment, the execution system 20 executes a hashing process for the teacher data (step S02). This makes it possible to identify the teacher data. Furthermore, the execution system 20 executes a process for acquiring the administrator signature for the teacher data (step S03). This makes it possible to identify the administrator of the teacher data.

[0045] (2) In this embodiment, the execution system 20 executes a hashing process for the learning algorithm (step S04). This makes it possible to identify the learning algorithm. Furthermore, the execution system 20 executes a process for acquiring the signature of the administrator of the learning system (step S05). This makes it possible to identify the administrator of the learning algorithm.

[0046] (3) In this embodiment, the execution system 20 executes a learning process (step S06), which allows a learning model to be generated using desired teacher data and learning algorithms.

[0047] (4) In this embodiment, the execution system 20 executes a hashing process for the learning model (step S07). This makes it possible to identify the learning model. Furthermore, the execution system 20 executes a process for acquiring the signature of the person who executes the machine learning (step S08). This makes it possible to identify the administrator of the learning model.

[0048] (5) In this embodiment, the smart contract SC1 executes an NFT registration process (step S09). This allows the authenticity of the learning model and the authenticity of the training data and learning algorithm used to generate the learning model to be confirmed using the NFT. In addition, the issuance of tradable NFTs can promote machine learning and re-learning.

[0049] (6) In this embodiment, a learning model update process is performed. When a new learning model is generated, the NFT of the new learning model inherits the conditions of the NFT of the base learning model. This allows re-learning to be performed within the range that does not deviate from the conditions of the NFT of the base learning model.

[0050] This embodiment can be modified as follows: This embodiment and the following modifications can be combined with each other to the extent that there is no technical contradiction. In the above embodiment, the metadata of the NFT includes the name of the NFT, the location of the learning model, the amount of issuance, a description, etc. The information recorded in the NFT is not limited to these. For example, the location of the training data and the location of the learning algorithm may be recorded.

[0051] In the above embodiment, the location (address) of the learning model is used as the location of the learning model. However, as long as the learning model can be acquired, it is not limited to the address. In the above embodiment, the smart contract SC1 executes the NFT registration process (step S09). Here, the smart contract SC1 may search for an existing NFT that matches the training data and learning algorithm used to generate the learning model. When an existing NFT that matches the training data and learning algorithm is identified, the smart contract SC1 notifies the execution system 20 that the NFT has been issued.

[0052] In the above embodiment, the smart contract SC1 executes the NFT registration process (step S09). Then, the utilization system 10 executes the utilization process of the learning model (step S10). Here, the NFT is made tradable by the executor becoming the owner and being able to be transferred to another account by the owner. The executor may be given a privilege by this transaction.

[0053] The owner of the NFT may have the right to receive incentives (benefits) at regular intervals. The smart contract SC1 accumulates the provision of incentives to the owner of the NFT in the blockchain network BN1, which can be confirmed by multiple parties who own each node. As a result, it is possible to promote the registration of learning models. [Explanation of symbols]

[0054] A1...information management system, BN1...blockchain network, B1...distributed ledger, SC1...smart contract, 10...usage system, 20...execution system, 30...data system, 40...learning system.

Claims

1. An information management system for managing a learning model, comprising a computer connected to a distributed ledger of a blockchain network, The computer, Obtain first authentic information of the teacher data; Obtaining second authentic information of the learning algorithm; Obtaining third authentic information of a learning model generated by inputting the teacher data into the learning algorithm; An information management system characterized by registering a first non-fungible token including the first authenticity information, the second authenticity information, and the third authenticity information in the distributed ledger.

2. The information management system according to claim 1 , wherein the first non-fungible token includes information regarding a storage destination of the learning model.

3. 3. The information management system according to claim 1, wherein the first non-fungible token includes information regarding the conditions of use of the learning model.

4. When a re-learned model is generated from the learning model using second teacher data obtained by updating the teacher data, a second non-fungible token corresponding to the re-learned model is registered in the distributed ledger; 3. The information management system according to claim 1, wherein the second non-fungible token includes information about the learning model on which the re-learning model is based.

5. 5. The information management system according to claim 4, characterized in that the second non-fungible token inherits information regarding the conditions of use of the learning model.

6. A method for managing a learning model using an information management system that manages a learning model, the information management system comprising a computer connected to a distributed ledger of a blockchain network, the method comprising: The computer, Obtain first authentic information of the teacher data; Obtaining second authentic information of the learning algorithm; Obtaining third authentic information of a learning model generated by inputting the teacher data into the learning algorithm; An information management method characterized by registering a first non-fungible token including the first authenticity information, the second authenticity information, and the third authenticity information in the distributed ledger.

Images