Ai learning data and system for improving authenticity of ai model

Abstract

To efficiently improve authenticity of AI learning data and AI models.SOLUTION: A system (1) for improving authenticity of AI learning data and AI models includes: a holding unit (2) which holds a history of AI learning data and a history of AI models in a block chain; an evaluation unit (3) which evaluates authenticity of each AI learning data and each AI model, based on the held histories; and an assignment unit (4) which assigns incentive or penalty to sources that provide each of the AI learning data and each of the AI models, the incentive or penalty including adding or associating ranking data according to the evaluated authenticity to / with each of the AI learning data and each of the AI models using NFTs.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to the technical field of an authenticity improvement system that improves the authenticity of AI learning data and AI models in an AI learning system that creates an AI model based on AI learning data applicable to, for example, autonomous driving.

Background Art

[0002] As this type of AI learning system, in addition to traditional AI learning systems for autonomous 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 in such a system is generally black-boxed from the user's perspective, and the authenticity of AI learning data and AI models basically depends on the providers of those data and models.

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-mentioned background art, although it is said that efficient operation of resources can be achieved, there is a possibility that AI learning data with unclear authenticity, such as the provider or source not being clear, is being used. In addition, it is difficult to evaluate the authenticity itself. Furthermore, there is a possibility that an AI model with unclear authenticity, such as the provider or vendor not being clear, is created as a learning result, and there is also a technical problem that it is difficult to evaluate the authenticity itself.

[0005] Regarding such authenticity, from the perspective of the provider who provides the data and the model, there is not much motivation to invest funds in data collection and model creation to enhance authenticity, nor is there much motivation to disclose the provider of the data and the model, which may include confidential information. As a result, it is not very likely that highly authentic learning results will be provided. In addition, for data and models where the provider or source is not clear, it is difficult to detect forgery, whether it is initial or subsequent forgery, that is, there are also technical problems in that they are easily forged.

[0006] An object of the present invention is to provide a system for improving the authenticity of AI learning data and AI models, which can efficiently enhance the authenticity of AI learning data and AI models.

Means for Solving the Problems

[0007] One aspect of the system for improving the authenticity of AI learning data and AI models according to the present invention includes a holding unit that holds the history of AI learning data and the history of AI models in a blockchain, an evaluation unit that evaluates the authenticity of each of the AI learning data and each of the AI models based on the held history, and adding or associating with each of the AI learning data and each of the AI models, in the form of NFT, ranking data indicating a ranking according to the degree of the evaluated authenticity, and an awarding unit that awards a predetermined type of incentive or penalty according to the degree to each provider of each of the AI learning data and each of the AI models.

Effects of the Invention

[0008] According to one aspect of the authenticity improvement system according to the present invention, by awarding incentives or penalties to the providers of each of the AI learning data and each of the AI models, it becomes possible to efficiently enhance the authenticity of those data and models gradually as AI learning progresses.

[0009] Such an operational effect according to the present invention will be made clearer by the embodiments of the invention described below.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Mode for Carrying Out the Invention

[0011] First, referring to FIG. 1, the overall configuration of the authenticity improvement system according to the embodiment will be described. As shown in FIG. 1, the authenticity improvement system 1 is configured to receive the provision of AI learning data Dm from the AI learning data Dm providing unit 11 via a network such as the Internet. Further, the authenticity improvement system 1 is configured to receive the provision of the AI model Mn from the AI model Mn providing unit 21 via a network such as the Internet. The authenticity improvement system 1 that performs these centralized or distributed processes, as well as the plurality of m AI learning data Dm providing units 11 at each data source and the plurality of n AI model Mn providing units 21 at each model source, are accommodated in a network such as the Internet.

[0012] The AI learning data Dm (D1, D2,..., Dm) is various data necessary for the AI learning, such as for autonomous driving, various matching, various prediction, and various promotion, according to its use. The AI learning data Dm providing unit 11 is various computer-mounted devices and various computer devices, and is configured to provide the AI learning data Dm collected therein to the authenticity improvement system 1 and the AI model Mn providing unit 21 via the network, either as it is or in a data format subjected to a predetermined type of processing. For example, in the case of autonomous driving, vehicle driving data, navigation data, GPS data, etc. will be provided as AI learning data Dm from sources such as vehicle-mounted systems and control systems of traffic information centers.

[0013] The AI model Mn providing unit 21 is configured to include one or more various computer devices that perform centralized processing or distributed processing. It receives the provision of the AI learning data Dm from the AI learning data Dm providing unit 11 and performs AI learning (in other words, machine learning or deep learning), such as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, and generative AI. Furthermore, the AI model Mn obtained as a result of the learning is configured to be provided to the authenticity improvement system 1 sequentially or collectively via a network, either regularly or irregularly. The AI models Mn (M1, M2,..., Mn) generated here are various models learned by the AI learning according to their uses, such as for autonomous driving, various matching, various prediction, and various promotion.

[0014] The AI model Mn providing unit 21 (in other words, the AI learning unit) may be configured using a neural network that performs efficient AI learning through representation learning, transfer learning, feature selection, fine-tuning, ensemble learning, etc., or may be configured as a generative AI that learns the patterns and relationships of the AI learning data Dm and generates content data different from the AI learning data Dm.

[0015] The authenticity improvement system 1 includes a memory, a processor, etc., and is configured as a centralized system that performs centralized processing or a distributed system that performs distributed processing, and includes a holding unit 2, an evaluation unit 3, and an imparting unit 4.

[0016] The holding unit 2 includes the memories of a plurality of computers accommodated in a network, etc., and is configured using existing or future-upgraded blockchain technology. The holding unit 2 holds the history of the AI learning data Dm and the history of the AI model Mn in the blockchain sequentially or at appropriate timings.

[0017] The evaluation unit 3 is configured to evaluate the authenticity of each of the AI learning data Dm and each of the AI models Mn based on the history held by the holding unit 2. The evaluation unit 3 may be configured to perform such evaluation by AI learning that performs authenticity evaluation using the input data as the history. Such an evaluation unit 3 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 models to the providing unit 4 via a network. The evaluation unit 3 may be configured using a neural neural network that performs efficient AI learning through representation learning, transfer learning, feature selection, fine-tuning, ensemble learning, etc., or may be configured as a generative AI that learns patterns and relationships in the history and generates content data different from the history. The evaluation unit 3 may be configured to perform a Gold evaluation when the authenticity is the highest, a Silver evaluation when the authenticity is very high, and a Bronze evaluation when the authenticity is moderately high, for example.

[0018] The awarding unit 4 adds or associates, in the form of NFTs, ranking data DRm (DR1, DR2,..., DRm) indicating the ranking of AI learning data Dm according to the degree of authenticity evaluated by the evaluation unit 3, and ranking data MRn (MR1, MR2,..., MRn) indicating the ranking of AI models Mn, to the corresponding AI learning data Dm and AI models Mn respectively. For example, NFTs such as a Gold NFT indicating the highest authenticity, a Silver NFT indicating a very high authenticity, and a Bronze NFT indicating a relatively high authenticity are added as ranking data. That is, the awarding unit 4 is configured to award incentives or penalties according to the degree of authenticity to the providers of each AI learning data Dm and each AI model Mn in the form of awarding ranking data. Further, instead of or in addition to providing the ranking data DRm and MRn that give such "endorsement" or status to the corresponding providers, the awarding unit 4 may be configured to award convertible points or gift points to the corresponding providers.

[0019] On the other hand, the linking data DRm and MRn thus awarded by the awarding unit 4 are stored in the holding unit 2 in the blockchain together with the history.

[0020] Next, with reference to the sequence chart of FIG. 2, an example of the process of awarding incentives in the authenticity improvement system according to this embodiment will be described.

[0021] In FIG. 2, first, the AI learning data Dm is collected by the AI learning data Dm providing unit 11 (step S1). For example, in the case of AI learning related to autonomous driving, driving data, navigation data, GPS data, road traffic information data, etc. of each vehicle are collected as the AI learning data Dm.

[0022] Subsequently, the AI learning data Dm providing unit 11 provides the AI learning data Dm collected in step S1 to the AI model Mn providing unit 21 and the authenticity improvement system 1 sequentially, periodically, irregularly, or at a predetermined timing (step S2). The data provision here is efficiently performed in an environment where the data provider (i.e., the AI learning data Dm providing unit 11) and the data recipient (the AI model Mn providing unit 21) are accommodated in the same network.

[0023] In response to this, the AI model Mn providing unit 21 functions as an AI learning unit and executes a preset type of AI learning, such as AI learning for autonomous driving, using the AI learning data Dm as an input (step S5). In parallel with or before and after this, the authenticity improvement system 1 stores the AI learning data Dm in the blockchain as part of the data history by its storage unit 2 (step S6). Even if the AI learning data Dm is stored in the database of the storage unit 2 without using the blockchain here, the effect of improving authenticity by granting incentives or penalties according to this embodiment can be obtained accordingly. However, by storing it using blockchain technology here, the reliability of the ranking related to the AI learning data Dm can be enhanced and the possibility of forgery can be eliminated, and the value of the ranking can be increased. Ultimately, the effect of improving authenticity in this embodiment becomes even more prominent.

[0024] Subsequently, the AI model Mn providing unit 21 provides the AI model Mn obtained by AI learning in step S5 to the authenticity improvement system 1 sequentially, periodically, irregularly, or at a predetermined timing (step S7). The data provision here is efficiently performed in an environment where the data provider (i.e., the AI model Mn providing unit 21) and the data recipient (the authenticity improvement system 1) are accommodated in the same network.

[0025] In response to this, the authenticity improvement system 1 has its holding unit 2 hold the AI model Mn on the blockchain as part of the AI model history (step S8). Even if the AI model history Mn is held in the database possessed by the holding unit 2 without using the blockchain here, the effect of improving authenticity by granting the incentives or penalties according to this embodiment can be obtained accordingly. However, by using the blockchain technology to hold it here, the reliability of the ranking related to the AI model can be enhanced and the possibility of forgery can be eliminated, and the value of the ranking can be increased. Ultimately, the effect of improving authenticity in this embodiment becomes even more prominent.

[0026] Subsequently, the authenticity improvement system 1 has its evaluation unit 3 evaluate the authenticity of each of the AI learning data Dm and each of the AI models Mn based on the history held on the blockchain (step S9). The evaluation unit 3 may perform the evaluation using various AI learnings as described above.

[0027] Subsequently, the authenticity improvement system 1 has its granting unit 4 grant the ranking data DRm related to the AI learning data Dm obtained by the AI learning etc. in step S9 to the AI learning data Dm providing unit 11 and the holding unit 2 in the form of NFT sequentially, periodically, irregularly, or at a predetermined timing (step S10). Here, the holding unit 2 holds the granted ranking data DRm on the blockchain. The granting unit 4 makes the ranking data DRm into the form of NFT and grants it to the AI learning data Dm providing unit 11. At this time, preferably as described above, in addition to or instead of granting the ranking data DRm, points that increase as the ranking is higher are granted to its provider.

[0028] In parallel with or before or after this, the authenticity improvement system 1, by the granting unit 4, sequentially or periodically or irregularly or at a predetermined timing, grants the ranking data MRn related to the AI model Mn obtained by AI learning or the like in step S9 to the AI model Mn providing unit 21 and the holding unit 2 (step S10). Here, the holding unit 2 holds the granted ranking data MRn in the blockchain. The granting unit 4 converts the ranking data MRm into the NFT format and grants it to the AI model Mn providing unit 11. At this time, preferably as described above, in addition to or instead of the granting of the ranking data MRm, points that increase as the ranking is higher are granted to the provider thereof.

[0029] Subsequently, the AI model Mn providing unit 21 refers to the ranking data MRm granted in step S10 and reflects it in the subsequent AI learning (step S11). That is, the higher the ranking indicated by the ranking data, the stronger the objectively obtained "endorsement" for the AI model provided by itself. More preferably, as a result of being able to obtain higher points, the motivation to provide a more highly evaluated AI model and not provide a less highly evaluated AI model occurs in each AI model Mn provider 21 according to the situation.

[0030] In parallel with or before or after this, the AI learning data Dm providing unit 11 refers to the ranking data DRm granted in step S10 and reflects it in the processing of data collection to data provision thereafter (step S12). That is, the higher the ranking indicated by the ranking data, the stronger the objectively obtained "endorsement" for the data provided by itself. More preferably, as a result of being able to obtain higher points, the motivation to provide more highly evaluated data and not provide less highly evaluated data occurs in each AI learning data Dm provider 11 according to the situation.

[0031] As described in detail above, according to this embodiment, the history of the AI learning data Dm and the history of the AI model Mn are stored in the blockchain by the holding unit 2, and the authenticity of each of the AI learning data Dm and each of the AI models Mn is evaluated by the evaluation unit 3. Depending on the degree of authenticity evaluated here, a predetermined type of incentive, penalty, or points are given to the provider of each of the AI learning data Dm and each of the AI models Mn. By these means, it becomes possible to generate a motivation for the provider to improve the authenticity of the AI learning data Dm and the AI model Mn. At this time, since it is stored in the blockchain, it is substantially impossible to carry out forgery from the beginning or afterwards, and moreover, it is substantially impossible to forge the NFT-formatted ranking data corresponding to the evaluation result. Since such a high reliability regarding the data Dm, the model Mn, and the evaluation result (that is, the ranking data DRm and MRn) is the basis, the motivation to improve the authenticity of the provider is not only to obtain an incentive or avoid a penalty, but also to obtain "endorsement", "social status", or "business status" based on a highly reliable evaluation result, and is further enhanced. As a result, as AI learning progresses, the authenticity is gradually improved.

[0032] Supplementary Note Regarding the embodiment described above, the following additional remarks are further disclosed.

[0033] [Supplementary Note 1] The authenticity improvement system for AI learning data and AI models according to Supplementary Note 1 of the present invention includes a holding unit that holds the history of AI learning data and the history of AI models in a blockchain, an evaluation unit that evaluates the authenticity of each of the AI learning data and each of the AI models based on the held history, and adding or associating, with respect to each of the AI learning data and each of the AI models, ranking data indicating a ranking according to the evaluated degree of authenticity in the form of NFTs, and an awarding unit that awards a predetermined type of incentive or penalty according to the degree to each of the providers of the AI learning data and each of the AI models.

[0034] According to the authenticity improvement system described in Supplementary Note 1, the history of AI learning data with unclear authenticity, such as the provider or source not being clear, and the history of AI models with unclear authenticity, such as the provider or vendor not being clear, are stored by the storage unit on the blockchain. As a result, it is made practically impossible to tamper with the data or models, either initially or retrospectively. The evaluation unit identifies, for each piece of AI learning data, the degree of its authenticity or classification by degree (in other words, "ranking of the data"), and for each AI model, the degree of its authenticity or classification by degree (in other words, "ranking of the model") as an evaluation. Furthermore, as a result of the evaluation by the evaluation unit of the authenticity of each piece of AI learning data and each AI model, when the authenticity is high, an incentive corresponding to the degree of its highness is given to the provider, or when the authenticity is low, a penalty corresponding to the degree of its lowness is given to the provider. The granting of such incentives or penalties includes adding or associating ranking data indicating a ranking corresponding to the degree of authenticity evaluated by the evaluation unit to each piece of AI learning data and each AI model in the form of an NFT. For example, ranking data with "endorsement" or points, such as AI learning data with high points and AI models with low or negative points, is given to the provider. At this time, it is also substantially impossible to tamper with the ranking data in the form of an NFT corresponding to the evaluation result. As a result, a strong motivation or sense of crisis to improve authenticity always occurs in the provider of each data and model, and as AI learning progresses, the authenticity is gradually improved.

[0035] [Supplementary Note 2] The authenticity improvement system described in Supplementary Note 2 according to the present invention is the authenticity improvement system described in Supplementary Note 1, characterized in that the granting unit grants the provider convertible points or gift points corresponding to the degree as part of the granting of the incentive or penalty.

[0036] According to the authenticity improvement system described in Supplementary Note 2 of the present invention, by the granting unit, for example, convertible points or gift points corresponding to the evaluated degree of authenticity or the classification according to the degree are granted to the provider. Therefore, the motivation to enhance authenticity will more directly arise on the provider side.

[0037] [Supplementary Note 3] The authenticity improvement system described in Supplementary Note 3 of the present invention is the authenticity improvement system according to Supplementary Note 1 or 2, wherein the evaluation unit evaluates the authenticity by AI.

[0038] According to the authenticity improvement system described in Supplementary Note 2 of the present invention, in the evaluation unit, the authenticity of the AI learning data and the AI model as described above is evaluated by AI. Therefore, as the AI learning in the evaluation unit progresses, the number of samples or the scale of learning related to the data and models for evaluating authenticity expands, and more appropriate or faster authenticity evaluation becomes possible, which is extremely advantageous in practice.

[0039] [Supplementary Note 4] The AI learning data described in Supplementary Note 4 of the present invention is the AI learning data for granting a predetermined type of incentive or penalty to the provider of the AI learning data, and the history of the AI learning data is held in a blockchain together with the history of the AI model created based on the AI learning data. Based on the held history, ranking data indicating a ranking according to the degree of authenticity, as a result of evaluating the authenticity of each of the AI learning data, is added or associated with each of the AI learning data by NFT.

[0040] According to the AI learning data described in Supplementary Note 4 of the present invention, by applying the AI learning data to the authenticity improvement system described in Supplementary Note 1, the same operational effects as those of the authenticity improvement system described in Supplementary Note 1 are achieved.

[0041] [Supplementary Note 5] The AI model described in Supplementary Note 5 according to the present invention is the AI model for imparting a predetermined type of incentive or penalty to the provider of the AI model, wherein the history of the AI model is held in a blockchain together with the history of AI learning data used for creating the AI model, and based on the held history, ranking data indicating a ranking according to the degree of authenticity as a result of evaluating the authenticity of each of the AI models is added or associated with each of the AI models by NFT.

[0042] According to the AI model described in Supplementary Note 5 according to the present invention, by applying the AI model to the authenticity improvement system described in Supplementary Note 1, the same operational effects as those of the authenticity improvement system described in Supplementary Note 1 are achieved.

[0043] The present invention can be appropriately modified within a range not contrary to the gist or idea of the invention that can be read from the claims and the entire specification, and the authenticity improvement system involving such modifications, as well as the AI learning data and AI models applied to the authenticity improvement system, are also included in the technical idea of the present invention.

Explanation of Signs

[0044] Authenticity improvement system... 1 Holding unit... 2 Evaluation unit... 3 Granting unit... 4 AI learning data Dm providing unit... 11 AI model Mn providing unit... 21

Claims

1. A holding unit that holds the history of AI learning data and the history of AI models on a blockchain; An evaluation unit that evaluates the authenticity of each piece of the AI learning data and each of the AI models based on the held history; An awarding unit that includes adding or associating, by NFT, ranking data indicating a ranking according to the degree of the evaluated authenticity to each piece of the AI learning data and each of the AI models, and awarding a predetermined type of incentive or penalty according to the degree to each provider of each piece of the AI learning data and each of the AI models An AI learning data and AI model authenticity improvement system comprising the above.

2. The awarding unit awards convertible points or gift points according to the degree to the provider as part of the awarding of the incentive or penalty. The AI learning data and AI model authenticity improvement system according to Claim 1.

3. The evaluation unit evaluates the authenticity by AI. The AI learning data and AI model authenticity improvement system according to Claim 1 or 2.

4. The AI learning data for awarding a predetermined type of incentive or penalty to a provider of AI learning data, wherein the history of the AI learning data is held on a blockchain together with the history of an AI model created based on the AI learning data, and ranking data indicating a ranking according to the degree of authenticity, as a result of evaluating the authenticity of each piece of the AI learning data based on the held history, is added or associated with each piece of the AI learning data by NFT Characterized by the above for the AI learning data.

5. The AI model for awarding a predetermined type of incentive or penalty to a provider of an AI model, wherein the history of the AI model is held on a blockchain together with the history of the AI learning data used for creating the AI model, and ranking data indicating a ranking according to the degree of authenticity, as a result of evaluating the authenticity of each of the AI models based on the held history, is added or associated with each of the AI models by NFT Characterized by the above for the AI model.

Images