WO2020075270A1

WO2020075270A1 - Machine learning model switching system, machine learning model switching method, and program

Info

Publication number: WO2020075270A1
Application number: PCT/JP2018/037955
Authority: WO
Inventors: 修吉桑田; 将仁谷口
Original assignee: 株式会社ウフル
Priority date: 2018-10-11
Filing date: 2018-10-11
Publication date: 2020-04-16

Abstract

According to the present invention, an AI vendor, an owner of an edge device, and an AI user are different from one another. Thus, for example, the AI vendor only has to concentrate on development, creation, update, and acquisition of an AI model, and the owner of the edge device only has to concentrate on creation, update, and acquisition of a data set for machine learning. As a result, as compared to the case where one person or one organization prepares both the AI model and the data set, cost and time required for the preparation thereof can be reduced. In addition, the AI user only has to prepare input data to be inputted to the created AI model, so that the AI user can know a matter that is predicted on the basis of the input data. As a result, as compared to the case where one person or one organization prepares all of the AI model, the data set, and the input data for predicting matters, cost and time required for the preparation thereof can be reduced. Furthermore, the AI user can use the latest AI model provided by the AI vendor.

Description

Machine learning model switching system, machine learning model switching method and program

The present invention relates to a machine learning model switching system, a machine learning model switching method and a program. The present invention relates to IoT (Internet of Things), and the technical field corresponds to G06Q and the like in the IPC classification.

A technology called machine learning is widespread. For example, Patent Document 1 discloses a mechanism for updating a machine learning model.

JP, 2017-120647, A

By the way, with the advancement of the technology called AI (Artificial Intelligence), it is expected that all services desired by users will be realized with the AI model. There are various kinds of services desired by users, and they change day by day. Therefore, various and latest AI models corresponding to these services are required. Therefore, it is convenient if there is a mechanism in which the owner of the edge device, the supplier of the machine learning model, and the user of the machine learning model are different from each other. However, in the technique of Patent Document 1, such different edge devices are used. It is not possible for the owner of the machine, the supplier of the model of machine learning and the user of the model to cooperate with each other to provide a one-stop service for a new machine learning model.

Therefore, an object of the present invention is to provide a technique for causing a user terminal to acquire and switch a new machine learning model generated by using sensor data of an edge device as learning data.

Means for solving problems

The present invention is a machine learning model switching system that allows a user terminal of a user who uses a model of machine learning to switch from a model being used to another model, in which the owner of the edge device changes the edge device. First acquiring means for acquiring the sensor data of the above, providing means for providing the acquired sensor data to a supplier supplying the model, which is different from the owner, and the supplying the sensor data as learning data. A first model of machine learning generated by a person from a supplier, a second acquisition unit, and a user terminal of the user different from the owner and the supplier, and the first model is clouded. Different from the first model, which is used before the first model is acquired in the user terminal and the user terminal, which is controlled so as to acquire the first model. That, from the second model of machine learning, providing a machine learning model switching system comprising a second control means for controlling so as to switch to the first model.

The first acquisition means, the provision means, and the second acquisition means may be realized by a transaction using a block chain.

It may be possible to provide a charging processing means for performing processing for paying money corresponding to at least a part of money paid from the user to the supplier or the owner.

The money paid by the user may be the money according to the first model or the sensor data.

Further, the present invention is a machine learning model switching method for causing a user terminal of a user who uses a model of machine learning to switch from a model being used to another model, wherein the owner of the edge device A first acquisition step of acquiring sensor data of an edge device, a providing step of providing the acquired sensor data to a supplier that supplies the model, which is different from the owner, and the sensor data as learning data. A second acquisition step of acquiring a first model of machine learning generated by the supplier from the supplier, and the first model on a user terminal of the owner and the user different from the supplier. A first control step for controlling to acquire the first model from the cloud, and the user terminal is used before the first model is acquired. Different from the first model, the second model of machine learning, providing a machine learning model switching method and a second control step of controlling so as to switch to the first model.

In addition, the present invention provides a computer included in a machine learning model switching system that causes a user terminal of a user who uses a machine learning model to switch from a model being used to another model, from an owner of an edge device. A first acquisition unit that acquires the sensor data of the edge device, a provision unit that provides the acquired sensor data to a supplier that supplies the model, which is different from the owner, and learns the sensor data. The first model of machine learning generated by the supplier as data is acquired from the supplier by a second acquisition unit, and the user terminal of the user different from the owner and the supplier, First control means for controlling one model to be acquired from the cloud, and the user terminal, which is used before acquiring the first model. Different from that of the first model, to provide a second model of machine learning, a program for functioning as a second control means for controlling so as to switch to the first model.

According to the present invention, in order to provide a one-stop machine learning model, it is possible to cause a user terminal to acquire and switch a new machine learning model generated by using sensor data of an edge device as learning data.

The figure which illustrates the whole structure of the machine learning model switching system 10 which concerns on one Embodiment of this invention. The figure which illustrates the hardware constitutions of AI vendor terminal 1. The figure which illustrates the hardware constitutions of the information processing apparatus 4. The figure which illustrates the function structure of the machine learning model switching system 10. 6 is a flowchart illustrating the operation of the information processing device.

1 ... AI vendor terminal, 2 ... edge device, 3 ... AI user terminal, 4 ... information processing device, 5 ... communication network, 10 ... machine learning model switching system, 101 ... control unit, 102 ... communication unit, 103 ... storage Unit 104 ... UI unit 401 ... control unit 402 ... communication unit 403 ... storage unit 11 ... first acquisition unit 12 ... provision unit 13 ... second acquisition unit 14 ... first control unit 15 ... Second control unit, 16 ... Charge processing unit.

1. Configuration FIG. 1 is a diagram illustrating an outline of a machine learning model switching system 10 according to an embodiment of the present invention. The machine learning model switching system 10 includes an AI vendor terminal 1 used by a vendor (corporation or individual) who is a supplier of an AI model for machine learning, an edge device 2 for performing various kinds of sensing, and an AI model user (corporation). Or, an AI user terminal 3 used by an individual), an information processing device 4, and a communication network 5 for communicatively connecting these terminals or devices 1 to 4. The machine learning model switching system 10 controls the AI user terminal 3 of a user who uses the AI model, from the AI model (second model) used in the AI user terminal 3 to another AI model (second model). This is a system for switching to one model).

The communication network 5 is, for example, a LAN (Local Area Network), a WAN (Wide Area Network), or a combination thereof, and may include a wired section or a wireless section. Each of the AI vendor terminal 1, the edge device 2, the AI user terminal 3, and the information processing device 4 is illustrated in FIG. 1, but each of them may be plural. The confidentiality of communication between the AI vendor terminal 1 and the information processing device 4, the edge device 2 and the information processing device 4, and the AI user terminal 3 and the information processing device 4 is maintained by the block chain technology.

The AI vendor terminal 1 is a computer that creates and updates the AI model by machine learning and provides it to the user of the AI model.

Edge device 2 is a device owned by a corporation or individual (owner) different from the AI model vendor. The edge device 2 is provided with mechanical / electromagnetic / thermal / acoustic / scientific properties of a natural phenomenon or an artificial object or spatial information / temporal information represented by them, such as a camera or a monitoring device of a home electric appliance. , Is a device that applies some scientific principle and replaces it with a signal of another medium that is easy for humans and machines to handle. The data used for machine learning of the AI model generally has an enormous amount of data, and is hereinafter referred to as a data set. The data set used in the present embodiment includes, for example, sensor data detected by the edge device 2 and an event that occurred at the time of detecting each sensor data.

The AI user terminal 3 is a computer used by an AI user different from the AI model vendor and the owner of the edge device. The AI user referred to here is the AI model obtained as a result of the machine learning performed by the information processing apparatus 4 using the above-mentioned data set, and the desired input data prepared by itself is input to the AI model. A person who uses the information output from the model. For example, if the data set is a large number of human face image data and the attributes of the human, and the AI model generated using the data set is a face authentication AI model, the AI user may select any face image data. To the AI model to obtain the authentication result.

As mentioned above, the AI vendor, the owner of the edge device, and the AI user are different from each other. Therefore, for example, the AI vendor only needs to concentrate on the development / generation / update / acquisition of the AI model, and the owner of the edge device only needs to concentrate on the generation / update / acquisition of the data set for machine learning. As a result, compared to the case where one human or one organization prepares both the AI model and the data set, the cost and time required for preparing them can be reduced. Further, some AI vendors are good / not good at the genre, purpose, and algorithm of the AI model, but by developing / generating / updating / obtaining a good AI model, an excellent AI model can be provided. It will be possible. Similarly, although edge devices have different performances and costs, by using edge devices that are superior to other edge devices or specialized in a certain field, it is possible to provide a useful data set to AI vendors. Is possible.

Also, AI users will be able to know the predicted events based on the input data by preparing the input data for input to the generated AI model. As a result, the cost and time required for the preparation can be reduced as compared with the case where one human or one organization prepares the AI model and the data set and all the input data for predicting the event. Becomes In addition, AI users will be able to utilize the latest AI models provided by AI vendors.

The algorithm for generating the above AI model by machine learning is, for example, a machine learning algorithm such as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, expression learning, etc. Other algorithms such as mining and deep learning may be included. Further, it also includes those using various techniques or technologies such as decision tree learning, association rule learning, neural networks, genetic programming, induction logic programming, support vector machines, clustering, Bayesian networks and the like.

Next, the configuration of terminals that make up the machine learning model switching system 10 will be described. FIG. 2 is a diagram illustrating a hardware configuration of the AI vendor terminal 1. The AI vendor terminal 1 includes a control unit 101, a communication unit 102, a storage unit 103, and a UI (User Interface) unit 104. The control unit 101 includes a computing device such as a CPU (Central Processing Unit) and a storage device such as a ROM (Read Only Memory) and a RAM (Random Access Memory). The CPU controls the operation of each unit of the AI vendor terminal 1 by using the RAM as a work area and executing the program stored in the ROM or the storage unit 103.

The communication unit 102 is hardware (transmission / reception device) for performing communication between computers via a wired or wireless communication network, and is also called, for example, a network device, network controller, network card, communication module, or the like. The communication unit 102 is connected to the communication network 5.

The storage unit 103 is a computer-readable recording medium, for example, an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disc, a magneto-optical disc (for example, a compact disc, a digital versatile disc, a Blu- It may be composed of at least one of a ray (registered trademark) disk, a smart card, a flash memory (for example, a card, a stick, a key drive), a floppy (registered trademark) disk, a magnetic strip, and the like. The storage unit 103 may be called an auxiliary storage device. The storage unit 103 stores a data group and a program group used by the control unit 101.

The UI unit 104 includes, for example, a liquid crystal panel and a liquid crystal drive circuit, includes a display unit that displays an image corresponding to image data, and an operator such as a key or a touch sensor, and receives a user's operation to perform the operation. The control unit 101 is provided with a corresponding signal.

The hardware configuration of the AI user terminal 3 is the same as that of the AI vendor terminal 1.

FIG. 3 is a diagram illustrating a hardware configuration of the information processing device 4. The information processing device 4 includes a control unit 401, a communication unit 402, and a storage unit 403. The control unit 401 includes an arithmetic device such as a CPU and a storage device such as a ROM and a RAM. The CPU controls the operation of each unit of the information processing device 4 by using the RAM as a work area and executing a program stored in the ROM or the storage unit 403.

The communication unit 402 is hardware (transmission / reception device) for performing communication between computers via a wired or wireless communication network, and is also called, for example, a network device, a network controller, a network card, a communication module, or the like. The communication unit 402 is connected to the communication network 5.

The storage unit 403 is a computer-readable recording medium, and is, for example, an optical disk such as a CD-ROM, a hard disk drive, a flexible disk, a magneto-optical disk (for example, a compact disk, a digital versatile disk, Blu-ray (registered trademark)). Disk), smart card, flash memory (eg, card, stick, key drive), floppy disk, magnetic strip, and / or the like. The storage unit 403 may be called an auxiliary storage device. The storage unit 403 stores a data group and a program group used by the control unit 401.

FIG. 4 is a diagram illustrating a functional configuration of the machine learning model switching system 10. The machine learning model switching system 10 includes a first acquisition unit 11 that acquires a data set including sensor data from the edge device 2, a providing unit 12 that provides the acquired data set to the AI vendor terminal 1, and a provided data set. The first machine-learned model acquired by the AI vendor terminal 1 from the AI machine terminal 1 and the first machine-learned model generated by the AI vendor using machine learning From the first control unit 14 that controls the model to be acquired from the cloud and the second machine-learned AI model that was used before the acquisition of the first machine-learned model in the AI user terminal 3, The second control unit 15 for switching to the machine-learned model of No. 1 and the money corresponding to at least a part of the money paid by the AI user is AI Ben. Or and a charging processing unit 16 that performs processing for paying the owner of the edge devices. A transaction (transaction) by the first acquisition unit 11, the provision unit 12, the second acquisition unit 13, the first control unit 14, the second control unit 15, and the billing processing unit 16 has a confidentiality due to the mechanism using the block chain. Has been maintained.

Each of the first acquisition unit 11, the provision unit 12, the second acquisition unit 13, the first control unit 14, the second control unit 15, and the billing processing unit 16 is a device that constitutes the machine learning model switching system 10. Is realized by executing the program. For example, the first acquisition unit 11 is realized by agent software executed in the edge device 2 or software executed in the information processing device 4, and the providing unit 12 is executed by a so-called cloud computer (here, the information processing device 4). It is realized by cloud software, the second acquisition unit 13 is realized by cloud software executed by a so-called cloud computer (here, the information processing device 4), and the first control unit 14 is agent software executed by the AI user terminal 3. The second control unit 15 is realized by the agent software executed in the AI user terminal 3, and the billing processing unit 16 is executed by a so-called cloud computer (here, the information processing device 4). It is realized by loud software.

2. Operation FIG. 5 is a flowchart illustrating an operation of the machine learning model switching system 10. In FIG. 5, the first acquisition unit 11 acquires the data set of the edge device 2 from the edge device 2 (step S11).

The providing unit 12 provides the acquired data set to the AI vendor terminal 1 (step S12).

The AI vendor terminal 1 uses this data set as learning data to generate and update the first AI model (latest machine-learned AI model) by machine learning. The second acquisition unit 13 acquires the first machine-learned AI model from the AI vendor terminal 1 (step S13).

The first controller 14 and the second controller 15 provide the AI user terminal 3 with the first machine-learned model (step S14). That is, the first control unit 14 controls the AI user terminal 3 so that the AI user terminal 3 acquires the first machine-learned model, and the second control unit 15 causes the AI user terminal 3 to acquire the first machine-learned model. From the second machine-learned AI model (old machine-learned AI model) that was used before obtaining the first machine-learned model (latest machine-learned AI model), the first machine-learned model Switch to (Latest machine learning AI model).

The charging processing unit 16 performs processing for paying money corresponding to at least a part of the money paid by the user to the AI vendor or the owner of the edge device (step S15). Specifically, first, the user operates, for example, the AI user terminal 3 to perform a procedure (bank transfer process or the like) for paying the money to the administrator of the information processing device 4. When the user pays money, the billing processing unit 16 performs processing (bank transfer processing or the like) for paying money corresponding to at least a part of the money to the AI vendor or the owner of the edge device. At this time, the amount of money paid by the AI user is the amount of money according to the type and amount of the AI model or data set, or the number of man-hours required for its generation, update, acquisition and the like. For example, the more valuable the AI model or the type of data set, or the greater the amount, the higher the money.

According to this embodiment described above, a new AI model can be provided in one stop.

3. Modifications The present invention is not limited to the above-described embodiments, and various modifications can be made. For example, the configuration and operation regarding payment of money are not limited to the example of the embodiment. Specifically, the information processing apparatus 4 does not pay money to the AI vendor or the owner of the edge device after collecting money from the user, but pays the money to the owner of the AI vendor or the edge device first. May collect money from the user. Further, an arbitrary method can be adopted for determining the amount of money.

The present invention is not limited to the above embodiment, and various modifications can be made. For example, part of the functional configuration illustrated in FIG. 4 may be omitted, or another function may be added. The functions included in the information processing device 4 illustrated in FIG. 4 may be implemented by any device or terminal belonging to the machine learning model switching system 10. Further, a computer device group physically composed of a plurality of devices may cooperate with each other to implement a function equivalent to that of the information processing device 4 shown in FIG.

The steps of processing performed in the machine learning model switching system 10 are not limited to the examples described in the above embodiments. The steps of this process may be interchanged as long as there is no contradiction. Further, the present invention may be provided as a machine-learned AI model providing method including steps of processing performed in the information processing device 4.

The program executed by the control unit of each device and each terminal may be provided by a storage medium such as an optical disk, a magnetic disk, a semiconductor memory, or may be downloaded via a communication line such as the Internet. Further, these programs do not have to execute all the steps described in the embodiments.

Claims

A machine learning model switching system that causes a user terminal of a user who uses a model of machine learning to switch from a model being used to another model,
First acquisition means for acquiring sensor data of the edge device from an owner of the edge device,
Providing means for providing the acquired sensor data to a supplier supplying the model, which is different from the owner.
Second acquisition means for acquiring from the supplier a first model of machine learning generated by the supplier using the sensor data as learning data;
First control means for controlling a user terminal of the user different from the owner and the supplier so as to obtain the first model from the cloud;
Second control for controlling the user terminal to switch from the second model of machine learning, which is different from the first model used before acquiring the first model, to the first model A machine learning model switching system including means.
The machine learning model switching system according to claim 1, wherein the first acquisition unit, the provision unit, and the second acquisition unit are realized by a transaction using a block chain.
2. The machine learning model switching system according to claim 1, further comprising a billing processing unit that performs processing for paying money corresponding to at least a part of money paid by the user to the supplier or the owner. .
The machine learning model switching system according to claim 2 or 3, wherein the money paid by the user is the money corresponding to the first model or the sensor data.
A machine learning model switching method for causing a user terminal of a user who uses a machine learning model to switch from a model being used to another model,
A first acquisition step of acquiring sensor data of the edge device from the owner of the edge device;
A providing step of providing the acquired sensor data to a supplier that supplies the model, which is different from the owner.
A second acquisition step of acquiring from the supplier a first model of machine learning generated by the supplier using the sensor data as learning data;
A first control step of controlling a user terminal of the user different from the owner and the supplier so as to obtain the first model from the cloud;
Second control for controlling the user terminal to switch from the second model of machine learning, which is different from the first model used before the acquisition of the first model, to the first model A machine learning model switching method comprising:
A computer included in the machine learning model switching system that causes the user terminal of the user who uses the machine learning model to switch from the model being used to another model,
First acquisition means for acquiring sensor data of the edge device from an owner of the edge device,
Providing means for providing the acquired sensor data to a supplier supplying the model, which is different from the owner.
Second acquisition means for acquiring from the supplier a first model of machine learning generated by the supplier using the sensor data as learning data;
First control means for controlling a user terminal of the user different from the owner and the supplier so as to obtain the first model from the cloud;
Second control for controlling the user terminal to switch from the second model of machine learning, which is different from the first model used before the acquisition of the first model, to the first model A program to function as a means.