AI Learning System
The AI learning system addresses compliance with legal requirements by selecting, encrypting, and securely computing personal information on a blockchain, facilitating AI learning and preventing information leakage.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- TOYOTA JIDOSHA KK
- Filing Date
- 2023-12-20
- Publication Date
- 2026-07-07
AI Technical Summary
Existing AI learning systems face challenges in performing AI learning with personal information while complying with legal requirements such as the Personal Information Protection Law and GDPR, leading to limited AI utilization of personal information.
An AI learning system that includes a selection unit for confidential personal information, an encryption unit for secure storage on a blockchain, and an AI learning unit for secure computation on encrypted data, setting criteria to reduce computational load based on encryption impact and confidentiality needs.
Enables AI learning with personal information compliance, preventing information leakage, and providing customized value while ensuring legal compliance.
Smart Images
Figure 0007885787000001 
Figure 0007885787000002 
Figure 0007885787000003
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of an AI learning system that performs AI learning based on AI learning data related to personal information while anonymizing the personal information.
Background Art
[0002] As such an AI learning system, in addition to traditional AI learning systems for automatic driving such as the so-called supervised learning method, unsupervised learning method, or reinforcement learning method, recently, systems for various applications such as generative AI have been developed and already put into practical use (see Patent Document 1). AI learning data used in such systems may contain personal information depending on the field. Under the Personal Information Protection Law and the GDPR (General Data Protection Regulation: General Data Protection Regulation in the EU), personal information needs to be encrypted or anonymized so that an individual cannot be easily identified.
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] However, according to the above-described background art, although it is possible to improve the resource-driven efficiency of maintaining the blockchain network, it is difficult to perform AI learning while complying with the requirements of the Personal Information Protection Law and the GDPR based on AI learning data including personal information in a state where the personal information is anonymized or concealed. For this reason, the AI utilization of personal information has not advanced much.
[0005] The objective of this invention is to provide an AI learning system that enables AI learning using AI learning data containing personal information, while complying with legal requirements regarding personal information. [Means for solving the problem]
[0006] One aspect of the AI learning system according to the present invention, in order to solve the above problems, comprises: a selection unit that selects personal information to be kept confidential when creating a desired AI model from AI learning data according to predetermined criteria; an encryption unit that encrypts the selected personal information; a storage unit that stores the AI learning data containing the encrypted personal information on a blockchain; and an AI learning unit that learns the AI model by performing secure computation on the stored AI learning data, at least for the data portion relating to the encrypted personal information, wherein the selection unit sets the predetermined criteria so as to reduce the amount of computation that is performed, depending on the degree to which the amount of computation of the secure computation increases or decreases due to the encryption of the personal information and the degree to which the personal information should be kept confidential. [Effects of the Invention]
[0007] According to one aspect of the AI learning system of the present invention, it becomes possible to perform AI learning using AI learning data that includes personal information, while complying with legal requirements such as the GDPR regarding personal information.
[0008] The effects and benefits of the present invention will be further clarified by the embodiments of the invention described below. [Brief explanation of the drawing]
[0009] [Figure 1] This is a block diagram showing the overall configuration of the AI learning system according to the embodiment. [Figure 2] This is a flowchart showing an example of processing in the AI learning system according to the embodiment. [Figure 3] This flowchart shows an example of the AI learning process in the AI learning system according to the embodiment. [Figure 4] This flowchart shows another example of the AI learning process in the AI learning system according to the embodiment. [Modes for carrying out the invention]
[0010] First, with reference to Figure 1, the overall configuration of the AI learning system according to this embodiment will be described. This embodiment is constructed as a system that performs AI analysis based on AI learning data that includes personal information relating to, for example, elderly drivers, disabled persons, and patients. Specifically, it is constructed as a system that provides customized analysis results for these individuals, which can be useful, for example, in determining whether or not to hire them at a company, selecting a place of employment, assignment, and working hours, determining treatment and compensation, and determining automobile accident insurance premiums. More generally, this embodiment is applicable to AI learning that needs to be conducted in accordance with legal requirements such as the Personal Information Protection Act and GDPR, and in any case, it is possible to enhance the protection of personal information.
[0011] As shown in Figure 1, the AI learning system 10 is constructed as a system that performs centralized or distributed processing, receiving "AI learning data" from data providers (not shown) connected to the network, and outputting or providing the "AI model" obtained as a result of AI learning or AI analysis to model providers (not shown) connected to the network.
[0012] The providers (not shown) are various computer-equipped devices and computer equipment that perform centralized or distributed processing, and are configured to provide the AI learning data input or collected therein, either as is or in a data format after a predetermined type of processing, to the AI learning system 10 via the network. The recipients (not shown) are various computer-equipped devices and computer equipment that perform centralized or distributed processing, and are configured to use the AI models output or provided therein for various tasks such as recruitment, human resources, and insurance.
[0013] Such AI training data related to personal information includes a wide range of information, from general to specialized personal information, such as an individual's age, gender, date of birth, height and weight, blood type, various vital data, medical history of the individual or their relatives, family structure, educational background, work history, job experience, special skills, awards, arrest record, address, place of origin, registered domicile, asset status, identification photos, driver's license number, and My Number number, provided that the individual's consent is obtained. In all cases, the effects and benefits of this embodiment, as described below, will be appropriately achieved. Such personal information may be in the form of text or code in a predetermined format, or it may be handwritten or on a mark sheet, or it may even be in the form of images or videos, as long as existing or future AI analysis is possible.
[0014] The AI learning system 10 includes memory, a processor, etc., and is configured as either a centralized system for centralized processing or a distributed system for distributed processing. It comprises a sorting unit 12, an encryption unit 13, a storage unit 14, and an AI learning unit 15.
[0015] The selection unit 12 is configured to select personal information data Dp from the AI training data that should be kept confidential when creating a desired AI model (for example, an AI model for personnel evaluation) according to predetermined criteria. Specifically, the selection unit 12 sets predetermined criteria to reduce the amount of computation required, depending on the degree to which the amount of computation required for secure computation increases or decreases due to the encryption of personal information and the degree to which the personal information should be kept confidential.
[0016] In this case, the setting of predetermined standards by the standard setting unit 12a may be done, for example, by human input by an operator or by input of a function determined in advance empirically (i.e., input of a function that takes the degree of increase or decrease in the computational processing volume of secure computation and the degree to which personal information should be concealed as input and outputs the computational processing volume), or by referring to a map that defines the relationship between the degree of increase or decrease in the computational processing volume of secure computation, the degree to which personal information should be concealed, and the computational processing volume, or by AI learning.
[0017] When setting these parameters using AI learning, for example, the operator may first input one or more hyperparameters related to the computational processing load of secure computation and the confidentiality of personal information in order to set the predetermined criteria. The AI may then be configured to tune the previously set hyperparameters during its learning process, i.e., adjust the types and combinations of parameters. In particular, the computational processing load of secure computation increases or decreases in complex ways depending on the specific method and content of the computation. On the other hand, the degree to which personal information should be kept confidential also changes in complex ways, such as whether or not an individual can be identified depending on the combination of individual elements that constitute the personal information. Furthermore, the acceptable range that each individual is willing to use for AI analysis may also differ from person to person. Therefore, hypertuning through AI learning is more effective than simply inputting a function that outputs computational processing load or inputting hyperparameters.
[0018] The AI learning in such a reference setting unit 12a may be configured to update the set criteria by performing AI learning such as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, or generative AI. Alternatively, the reference setting unit 12a that performs AI learning may be configured using a neural network that performs efficient AI learning such as representation learning, transfer learning, feature selection, fine tuning or hyperparameter tuning, or ensemble learning, or it may be configured as a generative AI that generates reference data.
[0019] The encryption unit 13 is configured to encrypt the personal information data Dp selected by the selection unit 12 (in other words, the data that should be encrypted) using various existing or future encryption technologies (for example, homomorphic encryption, secret sharing schemes, and other secure computation methods), and output it as encrypted personal information data Dpe.
[0020] The holding unit 14 includes the memories of a plurality of computers housed in the network, etc., and is configured using existing or future-upgraded blockchain technology. The holding unit 14 holds, sequentially or at appropriate timings, AI learning data that contains encrypted personal information data Dpe, i.e., two types of data: encrypted personal information data Dpe and unencrypted personal information data Dn (i.e., data that may be treated as "non-personal information data") in the blockchain.
[0021] The AI learning unit 15 is configured to learn an AI model by performing secret calculations on at least the data portion related to the encrypted personal information Dpe of the AI learning data (i.e., encrypted personal information data Dpe and unencrypted personal information data Dn) held in the blockchain by the holding unit 14, using its secret calculation unit 15a.
[0022] Such an AI learning unit 15 may be configured to perform AI learning such as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, generative AI, etc., and sequentially or collectively provide the learned AI model to a destination via a network. The AI learning unit 15 may be configured using a neural neural network that performs efficient AI learning through representation learning, transfer learning, feature selection, fine-tuning or hyperparameter tuning, ensemble learning, etc., or may be configured as a generative AI that learns patterns and relationships in AI learning data and generates content data different from the AI learning data. In any case, since the criterion setting by the criterion setting unit 12a is performed in the selection unit 12, the amount of calculation processing in the secret calculation unit 15a can be less compared to the case where such criterion setting is not performed. Here, the secret calculation is performed as a secret calculation such as, for example, fully homomorphic encryption or homomorphic encryption method, secret sharing method, etc. In particular, when the criterion setting unit 12a in the selection unit 12 sets the criterion by AI learning, the amount of calculation processing for the secret calculation decreases as the learning progresses, which is practically advantageous.
[0023] Next, in addition to the block diagram of FIG. 1, an example of the processing in the AI learning system according to this embodiment will be described with reference to the flowcharts of FIGS. 2, 3, and 4.
[0024] In FIG. 2, first, in the AI learning system 10 (see FIG. 1), AI learning data related to personal information in text or coded form in a predetermined format is input from a data provider accommodated in the network (step S11). Such AI learning data may include image data or video data.
[0025] Next, in the selection unit 12 (see FIG. 1), personal information data Dp to be anonymized when creating a desired AI model is selected from the AI learning data according to a predetermined criterion (step S12). Here, in particular, the predetermined criterion is set so that the amount of calculation processing is reduced according to the degree of increase or decrease in the amount of calculation processing of the secret calculation by encrypting personal information and the degree to which personal information should be anonymized. The setting of such a criterion is preferably set by AI learning as described later (see FIGS. 3 and 4).
[0026] Next, in the encryption unit 13 (see FIG. 1), the selected personal information data Dp to be encrypted is encrypted by various encryption techniques and output as encrypted personal information data Dpe (step S13).
[0027] Next, in the holding unit 14 (see FIG. 1), the AI learning data including the encrypted personal information data Dpe and the non-encrypted personal information data Dn is held in the blockchain (step S14).
[0028] Next, the AI learning unit 15 (see Figure 1) learns an AI model by performing a secure computation on the data portion of the AI learning data held on the blockchain, at least for the data portion relating to encrypted personal information Dpe, using its secure computation unit 15a (see Figure 1) (step S15). The secure computation here may employ a fully homomorphic encryption or homomorphic encryption scheme that performs operations on the encrypted personal information data Dpe while it remains encrypted, or it may employ a secret sharing scheme that hides the encrypted personal information data Dpe by dividing it into several random number fragments (shares) that have no meaning in themselves. Furthermore, this processing by the AI learning unit 15 is preferably performed by AI learning as described later (see Figures 3 and 4).
[0029] Next, the predicted or created AI model is output to the recipient (step S16), and the series of processes ends.
[0030] The above-mentioned selection process (step S12) and AI learning (step S15) are performed, for example, by traditional AI learning that is not a generative AI, as shown in Figure 3, or by AI learning using a generative AI, as shown in Figure 4.
[0031] Specifically, as shown in Figure 3, in an example of the selection process (step S12), various AI learning data relating to personal information are input (step S17), storage of this data (step S18) and knowledge conversion (step S19) are performed, an AI model representing the appropriate answer is predicted (step S20), and existing content is output. In particular, a predetermined standard is predicted as an AI model so as to reduce the amount of computation required, depending on the degree to which the computational processing load of secure computation increases or decreases due to the encryption of personal information and the degree to which the personal information should be kept confidential. In this case, rules may be established when setting the predetermined standard, such as selecting the data to be encrypted so as not to result in combinations of personal information that can identify an individual, inputting personal information that can be tolerated without encryption for the purpose of the AI learning system 10 as training data, avoiding encryption that would result in an extremely large amount of computational processing load for secure computation as much as possible, and conversely, prioritizing and allowing encryption that would result in a small amount of computational processing load for secure computation.
[0032] On the other hand, as shown in Figure 3, in an example of AI learning (step S15), encrypted personal information data Dpe and unencrypted personal information data Dn are input as AI learning data (step S17), storage (step S18) and knowledge creation (step S19) of this data are performed, an AI model that is the appropriate answer is predicted (step S20), and the output of existing content is performed in step S16 in Figure 2. The calculations related to data storage, knowledge creation and AI model prediction in steps S18 to S20 are processed by secure computation with respect to the encrypted personal information data Dpe while it remains encrypted.
[0033] Alternatively, as shown in Figure 4, in another example of the selection process (step S12), various AI learning data relating to personal information are input (step S17), and storage of this data (step S18), as well as knowledge creation and self-learning through deep learning (step S29), are performed to create an AI model that is an appropriate answer (step S30), and original content that is not existing content is output. In particular, a predetermined standard is created as an AI model so as to reduce the amount of computational processing required, depending on the degree to which the amount of computational processing required for secure computation increases or decreases due to the encryption of personal information and the degree to which the personal information should be kept confidential. For example, deep learning may be performed to avoid combinations of personal information that can identify an individual, or to prioritize and allow encryption that leads to a reduction in the amount of computational processing required for secure computation.
[0034] On the other hand, as shown in Figure 4, in another example of AI learning (step S15), encrypted personal information data Dpe and unencrypted personal information data Dn are input as AI learning data (step S17), and storage of this data (step S18), as well as knowledge creation and self-learning through deep learning (step S29) are performed to create an AI model that is an appropriate answer (step S30), and the output of original content that is not existing content is performed in step S16 of Figure 2. The calculations related to data storage, knowledge creation and deep learning, and AI model creation in steps S18 to S30 are processed by secure computation with respect to the encrypted personal information data Dpe while it remains encrypted.
[0035] As explained in detail above, personal information that should be kept confidential from the AI training data is selected and encrypted, data retention is carried out using blockchain technology, and AI training is carried out using secure computation technology. Therefore, even for personal information data that is inherently difficult to keep confidential because it is retained for a relatively long period of time, it is prevented from being viewed by staff of corporations, etc., using the AI training system 10 as information that can identify the individual concerned. Furthermore, it becomes extremely difficult to view the personal information in a form that can identify the individual concerned. As a result, it is possible to effectively prevent the leakage of personal information to third parties, while providing the owner of the personal information with customized value through AI training that utilizes that personal information.
[0036] Note The following additional information is disclosed regarding the embodiments described above.
[0037] [Note 1] The AI learning system described in Appendix 1 of the present invention comprises: a selection unit that selects personal information to be kept confidential when creating a desired AI model from AI learning data according to predetermined criteria; an encryption unit that encrypts the selected personal information; a storage unit that stores the AI learning data containing the encrypted personal information on a blockchain; and an AI learning unit that learns the AI model by performing secure computation on the stored AI learning data, at least for the data portion relating to the encrypted personal information. The selection unit sets the predetermined criteria such that the amount of computation is reduced according to the degree of increase or decrease in the amount of computation of the secure computation due to the encryption of the personal information and the degree to which the personal information should be kept confidential.
[0038] According to the AI learning system described in Appendix 1, personal information that should be kept confidential is selected from the AI learning data and encrypted. Furthermore, the AI learning data containing the encrypted personal information is stored on the blockchain. The AI model is trained by performing secure computation on the data portion of the AI learning data stored in this way, at least for the data portion related to the encrypted personal information. The above selection is performed by setting predetermined criteria so as to minimize the computational load, depending on the degree to which the computational load of secure computation increases or decreases due to the encryption of personal information and the degree to which the personal information should be kept confidential. In this way, it is possible to effectively prevent the leakage of personal information to third parties, while providing the owner of the personal information with customized value through AI learning utilizing that personal information.
[0039] [Note 2] The AI learning system described in Appendix 2 of the present invention is an AI learning system described in Appendix 1, characterized in that the selection unit sets predetermined criteria and selects personal information by AI learning using the degree of increase or decrease in the amount of computation processing and the degree to which personal information should be kept confidential as AI learning data.
[0040] According to the AI learning system described in Appendix 2 of the present invention, the selection unit sets the predetermined criteria as described above through AI learning. Therefore, the more the AI learning progresses in the selection unit, the larger the number of samples or the scale of learning for the AI learning data related to the amount of computation in secure computation and the AI learning data related to the confidentiality of personal information becomes, making it possible to set more appropriate predetermined criteria. This is extremely advantageous in reducing the amount of computation while ensuring confidentiality.
[0041] The present invention may be modified as appropriate, without contradicting the gist or spirit of the invention as can be inferred from the claims and the specification as a whole, and an AI learning system with such modifications is also included in the technical concept of the present invention. [Explanation of symbols]
[0042] AI learning system...10 Sorting section...12 Encryption section...13 Holding part...14 AI Learning Department...15
Claims
1. A sorting unit that, in accordance with prescribed standards, sorts out personal information from the AI training data that should be kept confidential when creating the desired AI model. An encryption unit for encrypting the selected personal information, A storage unit that stores the AI learning data, which contains the encrypted personal information, on a blockchain, The AI learning unit learns the AI model by performing secure computation on at least the encrypted personal information portion of the retained AI learning data. Equipped with, The sorting unit sets the predetermined criteria such that the amount of computation is reduced, depending on the degree to which the amount of computation of the secure computation increases or decreases due to the encryption of the personal information and the degree to which the personal information should be kept confidential. Its distinguishing feature is its AI learning system.
2. The AI learning system according to claim 1, characterized in that the selection unit sets predetermined criteria and selects personal information by AI learning, using the degree of increase or decrease in the amount of computation processing and the degree to which the personal information should be kept confidential as AI learning data.