WO2021059847A1

WO2021059847A1 - Information processing system, information processing method, and information processing program

Info

Publication number: WO2021059847A1
Application number: PCT/JP2020/032345
Authority: WO
Inventors: 択渡久地
Original assignee: ＡＩｉｎｓｉｄｅ株式会社
Priority date: 2019-09-27
Filing date: 2020-08-27
Publication date: 2021-04-01
Also published as: JP6695534B1; JP2021056658A

Abstract

An information processing system 1 comprising: a learning server 100 that obtains target data for data analysis, performs machine learning, generates a trained model, and provides the trained model to an analysis device; and the analysis device 200 that obtains the trained model, obtains the target data, performs data analysis on the basis of the trained model, and performs outputs in accordance with the purpose of the trained model.

Description

Information processing system, information processing method and information processing program

This disclosure relates to an information processing system, an information processing method, and an information processing program that perform machine learning to generate a specific trained model and perform data analysis using the trained model.

In recent years, various devices have been developed in which an AI engine having an artificial intelligence (AI) function is incorporated and machine learning is performed to control the operation. For example, in an industrial waste treatment plant, in the process of sorting from a mixture of various types of waste and appropriately treating each sorted waste, machine learning about waste is performed by a robot equipped with an AI engine. Is performed, and the sorting is performed mechanically by the AI engine.

For example, Patent Document 1 discloses a sorting system capable of efficiently sorting even rubble in which a wide variety of materials are mixed and waste obtained by dismantling a building. In this sorting system, image processing is performed from image data obtained by capturing images of waste, and the types of waste materials such as wood, plastic, glass, and gypsum board are sorted.

In order to use such an AI engine, it is necessary to perform appropriate machine learning. For example, in supervised machine learning, which is a type of machine learning, it is necessary to perform machine learning with a large amount of teacher data to generate model information, and the output result depends on this model information. Therefore, it is necessary for the AI engine to use appropriate model information. For example, Patent Document 2 discloses a system for evaluating a machine learning model using a training data set.

JP-A-2017-109197 JP-A-2017-004509

By the way, the machine learning required to use such an AI engine requires, for example, machine learning using a large amount of teacher data, which requires a great deal of time and effort. In addition, such machine learning model information can also be used by other AI engines. However, a system that provides the model information used in one AI engine as a learned model to another AI engine has not been provided.

Also, so-called SaaS (Software as a Service), which provides such an AI engine from a cloud server via the Internet, is known. By using such a service, it is possible to use model information by machine learning with other AI engines, but in companies, etc., providing various data of their own company on a cloud server is for security reasons. It becomes a problem. Therefore, a system that can use such an AI engine in a local environment has been desired.

Therefore, in the present disclosure, a learning center for acquiring target data to be analyzed and performing machine learning is provided by a server, a learned model used by an AI engine is acquired from the learning center, and a self-AI engine is used. The information processing system, information processing method, and information processing program that can be used to perform analysis processing in.

The information processing system according to one aspect of the present disclosure is an information processing system including a learning server that performs machine learning to generate a specific trained model and an analysis device that performs data analysis using the trained model. , The learning server generates a trained model which is model information for performing machine learning based on the target data and the target data acquisition unit for acquiring the target data to be data-analyzed, or performing data analysis. A learning unit that updates the trained model and a providing unit that provides the trained model to the analysis device are provided, and the analysis device is based on the model acquisition unit that acquires the trained model from the training server and the trained model. It includes an analysis unit that analyzes the data of the target data, and a result output unit that outputs the trained model according to the purpose of the trained model based on the result of the data analysis.

The information processing method in one aspect of the present disclosure is an information processing method that performs machine learning to generate a specific trained model and provides the trained model to an analyzer that performs data analysis using the trained model. Therefore, the target data acquisition step for acquiring the target data to be the target of data analysis performed by the target data acquisition unit, and the model information for performing machine learning based on the target data and performing data analysis performed by the learning department. A training step of generating a trained model or updating a trained model, a providing step of providing a trained model to an analysis device performed by the providing unit, and a training model performed by the analysis device performed by the model acquisition unit. The purpose of the trained model is determined by the model acquisition step of acquiring the model, the analysis step of analyzing the target data based on the trained model performed by the analysis unit, and the result of the data analysis performed by the result output unit. It includes a result output step for outputting.

Further, the information processing program in one aspect of the present disclosure performs machine learning to generate a specific trained model, and provides the trained model to an analyzer that performs data analysis using the trained model. In the program, the target data acquisition step to acquire the target data to be the target of data analysis and the trained model which is the model information for performing machine learning based on the target data and performing the data analysis are generated, or Data analysis of target data is performed based on the training step of updating the trained model, the provision step of providing the trained model to the analysis device, the model acquisition step of acquiring the trained model by the analysis device, and the trained model. The computer is made to execute the analysis step and the result output step that outputs according to the purpose of the trained model based on the result of the data analysis.

According to the present disclosure, a learning server (learning center) that performs machine learning to generate a specific trained model acquires target data to be analyzed and performs machine learning to generate a trained model. , Provided to the analyzer (AI engine). The analysis device analyzes the data based on the trained model and outputs the data according to the purpose of the trained model. Therefore, the result of machine learning performed by the learning server can be used by a plurality of analysis devices. As a result, the labor of machine learning can be reduced, and the trained model, which is the result of machine learning of others, can be used. In addition, the analysis device can be used in a local environment separated from the learning server.

It is a block block diagram which shows the information processing system which concerns on one Embodiment of this disclosure. It is a block block diagram which shows the information processing system 1A which is a modification of the information processing system 1 of FIG. It is a functional block block diagram which shows the learning server 100 of FIG. It is a functional block block diagram which shows the analysis apparatus 200 of FIG. It is a perspective view which shows the appearance of the sorting apparatus 400 controlled by the analysis apparatus 200 of FIG. It is a flowchart which shows the operation of the machine learning process in the information processing system 1 of FIG. It is a flowchart which shows the operation of the data analysis processing in the information processing system 1 of FIG. It is a flowchart which shows the operation of the charge amount determination process in the information processing system 1 of FIG. It is a functional block block diagram which shows the analysis apparatus 200B of the information processing system which concerns on one Embodiment of this disclosure. It is a functional block block diagram which shows the analysis apparatus 200C of the information processing system which concerns on one Embodiment of this disclosure. It is a functional block block diagram which shows the computer 700 which concerns on one Embodiment of this disclosure.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. The embodiments described below do not unreasonably limit the contents of the present disclosure described in the claims. Also, not all of the components shown in the embodiments are essential components of the present disclosure.

(Embodiment 1)
<Structure>
FIG. 1 is a block configuration diagram showing an information processing system 1 according to the first embodiment of the present disclosure. The information processing system 1 is not limited to, but as an example, is a system that performs machine learning to generate a specific trained model and analyzes data using the generated trained model. This trained model is, for example, model information obtained by machine learning from target data to be analyzed for a certain data, and model information that can be used not only by the analysis device that provided the target data but also by other analysis devices. Is. That is, the information processing system 1 is a system that enables horizontal expansion of a trained model generated by performing machine learning.

The information processing system 1 has a learning server 100,

analysis devices

201, 202, ..., And a network NW. The learning server 100 and the

analysis devices

201, 202, ... Are connected to each other via the network NW. The network NW is a communication network for communication, and is not limited to the Internet, an intranet, a LAN (Local Area Network), a WAN (Wide Area Network), a wireless LAN (Wireless LAN: WAN), and a wireless WAN (Wireless LAN). It is configured to be connected via a communication network including a wireless WAN (WAN), a virtual private network (VPN), and the like.

Note that the

analysis devices

201, 202, ... Each have the same configuration, and when the configurations of the

analysis devices

201, 202, ... Are described, they are collectively referred to as the analysis device 200.

The learning server 100 generates a specific trained model by performing machine learning from the target data to be analyzed, and uses the generated trained model as one or a plurality of analysis devices 200 (

analysis devices

201, 202, ...・) Is a device that provides a function as a learning center by providing it to. As an example, not limited to, model information for reading character information by OCR for handwritten characters and sorting of industrial waste are performed. It is configured as a device that generates model information for controlling the device by machine learning. The learning server 100 is not limited, but is composed of, for example, a computer (desktop, laptop, tablet, etc.) that provides various Web services, a server device, and the like. The learning server 100 is not limited to a device that operates independently, and may be a distributed server system or a cloud server in which a plurality of devices are connected to each other via a communication network and cooperate with each other to perform communication.

The analysis device 200 acquires a trained model from the learning server 100, analyzes the target data to be analyzed as an AI engine based on the trained model, and outputs the target data according to the purpose of the trained model. It is a device, and is not limited to an example, and is configured as a device for reading character information by OCR for handwritten characters and a device for controlling a device for separating industrial waste. The analysis device 200 is not limited, but is composed of, for example, a computer (desktop, laptop, tablet, etc.) that controls various devices as described above, a server device, and the like. Further, the analysis device 200 is composed of

analysis devices

201, 202, ... Having one or a plurality of similar configurations, which are used by different companies and the like, and use the same trained model provided in the learning server 100. It is possible. The external device 200 is not limited to a device that operates independently, and may be a distributed server system or a cloud server in which a plurality of devices are connected to each other via a communication network and cooperate with each other to perform communication. In the present embodiment, as an example of various devices controlled by the analysis device 200, a sorting device 400, which will be described later, is connected, or the analysis device 200 is built in the sorting device 400.

FIG. 2 is a block configuration diagram showing an information processing system 1A which is a modification of the information processing system 1 of FIG. The information processing system 1A provides a learning center by performing machine learning from target data to be analyzed to generate a specific trained model and providing the generated trained model to the analysis device 200. It is similar to the information processing system 1 in that it is different from the information processing system 1 according to the first embodiment in that it is provided with the communication means T instead of the network NW. In this modification, the analysis device 200 does not need to be always connected to the learning server 100, and when the trained model is acquired, for example, a predetermined timing set or a user of the analysis device 200 operates the learning server. An example of the case of being connected when accessing 100 is shown. That is, the analysis device 200 can be operated stand-alone except for the timing of acquiring the trained model from the learning server 100, thereby reducing the labor of machine learning and the machine of another person (other company). A trained model that is the result of learning can be used. Other configurations are the same as those in the first embodiment.

The communication means T is a line facility that communicates directly or via a communication network, and is not limited, but is configured to be directly connected by a USB (Universal Serial Bus) cable, a LAN (Local Area Network) cable, or the like. Or, Internet, Intranet, LAN (Local Area Network), WAN (Wide Area Network), Wireless LAN (Wireless LAN: WAN), Wireless WAN (Wireless WAN: WAN), Virtual Private Network (VPN) It is configured to be connected via a communication network including the above.

FIG. 3 is a functional block configuration diagram showing the learning server 100 of FIG. The learning server 100 includes a communication unit 110, a storage unit 120, and a control unit 130.

The communication unit 110 is a communication interface for communicating with the analysis device 200 by wire or wirelessly via the network NW or the communication means T, and any communication protocol may be used as long as mutual communication can be executed. Good. The communication unit 110 is not limited, and for example, communication is performed by a communication protocol such as TCP / IP (Transmission Control Protocol / Internet Protocol).

The storage unit 120 stores programs, input data, and the like for executing various control processes and each function in the control unit 130. The storage unit 120 is not limited, but as an example, a RAM (RandomAccessMemory) and a ROM (ReadOnly). It is composed of a memory including a memory) and a storage including an HDD (Hard Disk Drive), an SSD (Solid State Drive), a flash memory and the like. Further, the storage unit 120 stores the target data DB 121A, the target data DB 121B, and the trained model DB 122. Further, the storage unit 120 temporarily stores the data when communicating with the analysis device 200 and the data generated by each process described later.

The

target data DBs

121A and 121B store target data for machine learning by the learning server 100. This target data is data provided by the analysis device 200 by the user's selection of the analysis device 200, which will be described later, and is not limited but, as an example, data according to the device on which the analysis device 200 is mounted, and is OCR. It is image data for reading text information and image data for separating industrial waste.

Further, the target data DB 121A and the target data DB 121B store the target data provided by different analysis devices 200. For example, the target data DB 121A stores the target data provided by the analysis device 201. The target data provided by the analysis device 202 is stored in the data DB 121B. Then, the users of the

analysis devices

201, 202, ... Are configured so that they cannot refer to data other than the target data provided by themselves. The reason for this configuration is that if there is a risk that other companies may refer to your own data, you may hesitate to provide the data, and such data is for the individual customer of the user. This is because information may be included. Therefore, the target data stored in the

target data DBs

121A and 121B is used only for machine learning and is not used for any other purpose.

The trained model DB 122 stores model information generated by machine learning by the learning server 100. This model information is model information for the learning server 100 to perform data analysis as an AI engine, not limited to, as an example, model information for reading character information by OCR, and model information for separating industrial waste. Is stored.

The control unit 130 controls the overall operation of the learning server 100 by executing a program stored in the storage unit 120. The control unit 130 is not limited to the CPU (Central Processing Unit) and the MPU (Micro). Devices including ProcessingUnit), GPU (GraphicsProcessingUnit), microprocessor (Microprocessor), processor core (Processorcore), multiprocessor (Multiprocessor), ASIC (Application-Specific IntegratedCircuit), FPGA (Field ProgrammableGateArray), etc. Consists of. As the functions of the control unit 130, the target data acquisition unit 131, the learning unit 132, the provision unit 133, and the billing amount determination unit 134 are provided. The target data acquisition unit 131, the learning unit 132, the provision unit 133, and the charge amount determination unit 134 are started by the program stored in the storage unit 120 and executed by the learning server 100.

The target data acquisition unit 131 uses the analysis target data provided by the analysis device 200 by the user of the analysis device 200 as the target data for the learning server 100 to perform machine learning. Acquired via the communication unit 110. In the above example, the image data of the handwritten character when reading the character information by OCR for the handwritten character and the image data of the industrial waste for separating the industrial waste according to the implementation example of the analyzer 200. To get. Further, the target data acquisition unit 131 transfers the acquired target data to the target data DB 121A from the target data provided by the analysis device 201 and the target data provided from the analysis device 202 to the target data DB 121B as described above. Store each in.

The learning unit 132 performs machine learning based on the target data acquired by the target data acquisition unit 131, generates a trained model and stores it in the trained model DB 122, or has trained stored in the trained model DB 122. Update the model. This target data is data including the data to be analyzed by the analysis device 200 and the result of the data analysis. The trained model is updated, for example, by an aggregation process that merges the updated information based on the result of machine learning with the trained model stored in the trained model DB 122. Further, the learning unit 132 has learned a model that is the result of machine learning based on the target data stored in the target data DB 121A and a trained model that is the result of machine learning based on the target data stored in the target data DB 121B. You may perform aggregation processing to merge with the model.

The machine learning by the learning unit 132 is not limited, but as an example, it may be performed by supervised machine learning, it may be performed by unsupervised machine learning, or it may be performed by deep learning.

The providing unit 133 provides the learned model generated by the learning unit 132 by the operation of the user of the analysis device 200 by transmitting the learned model to the analysis device 200 via the communication unit 110. In the present embodiment, the analysis device 200 is configured to be provided by downloading the trained model from the learning server 100, but is provided as a software service of the AI engine in a state of being stored in the learning server 100. , So-called SaaS may be provided.

The billing amount determination unit 134 determines the billing amount, which is the consideration for the provision of the learned model. This billing amount may be set by, for example, the amount of data of the trained model or the amount of money per unit time using the trained model. Further, for example, when the target data acquired by the target data acquisition unit 252, which will be described later, is provided to the learning server 100, the billing amount is set low, and when the target data is not provided to the learning server 100, the billing amount is high. It may be set. By setting in this way, the user is motivated to provide the target data, and the trained model can be made more accurate by performing more learning.

FIG. 4 is a functional block configuration diagram showing the analysis device 200 of FIG. The analysis device 200 includes a communication unit 210, a display unit 220, an operation unit 230, a storage unit 240, and a control unit 250.

The communication unit 210 is a communication interface for communicating with the learning server 100 by wire or wirelessly via the network NW or the communication means T, and any communication protocol can be used as long as mutual communication can be executed. Good. The communication unit 210 is not limited, and for example, communication is performed by a communication protocol such as TCP / IP. Further, in the case of the information processing system 1A shown in FIG. 2, the communication unit 210 operates when the communication means T is connected between the learning server 100 and the analysis device 200.

The display unit 220 is a user interface used to display the operation content input by the user and the transmission content from the learning server 100, and is composed of a liquid crystal display or the like. The display unit 220 displays, for example, the analysis result by the analysis unit 253.

The operation unit 230 is a user interface used for the user to input operation instructions, and is composed of a keyboard, a mouse, a touch panel, and the like. The operation unit 230 is used, for example, for inputting an instruction for acquiring a learned model provided by the learning server 100 and for responding to an instruction from the learning server 100.

The storage unit 240 stores programs for executing various control processes and each function in the control unit 250, input data, and the like. The storage unit 240 is not limited, and as an example, a memory including a RAM, a ROM, and the like, an HDD, and the like. It is composed of storage including SSD, flash memory and the like. Further, the storage unit 240 stores the trained model DB 241 and the target data DB 242. Further, the storage unit 240 temporarily stores the data communicated with the learning server 100 and the data generated by each process described later.

The trained model DB 241 stores model information generated by the learning server 100 by performing machine learning and provided by the learning server 100. This model information is the same model information as the trained model DB 122 of the learning server 100, and is not limited to model information for performing data analysis as an AI engine, but as an example, model information for reading character information by OCR. , Model information for sorting industrial waste is stored. As described above, when the analysis device 200 is provided as a software service of the AI engine, the trained model DB 241 does not have to be provided.

The target data DB 242 stores the target data to be analyzed by the analysis device 200 as an AI engine. This target data is data acquired from a device for reading character information by OCR for handwritten characters and a device for controlling a device for separating industrial waste in the above-mentioned example. In this embodiment, it is the data acquired from the sorting device 400 controlled by the analysis device 200.

The control unit 250 controls the overall operation of the analyzer 200 by executing the program stored in the storage unit 240, and is not limited to, but as an example, a CPU, an MPU, a GPU, a microprocessor, and a processor. It is composed of a core, a multiprocessor, an ASIC, a device including an FPGA, and the like. The functions of the control unit 250 include a model acquisition unit 251, a target data acquisition unit 252, an analysis unit 253, a result output unit 254, and a target data providing unit 255. The model acquisition unit 251, the target data acquisition unit 252, the analysis unit 253, the result output unit 254, and the target data provision unit 255 are started by the program stored in the storage unit 240 and executed by the analysis device 200. ..

The model acquisition unit 251 acquires a learned model provided by the learning server 100 as a learning center. For example, when the trained model is transmitted to the learning server 100 by the operation of the analysis device 200 of the user, the learning server 100 transmits the trained model, and the trained model is accepted via the communication unit 110. This trained model is model information for the analysis device 200 to perform data analysis as an AI engine, and may be a single trained model, in order to reflect the latest training results on the already acquired trained model. It may be the update information of. Further, the model acquisition unit 251 stores the acquired trained model in the trained model DB 241.

The target data acquisition unit 252 acquires the target data to be analyzed by the analysis device 200 as an AI engine. In the above example, the image data of the handwritten character when reading the character information by OCR for the handwritten character and the image data of the industrial waste for separating the industrial waste according to the implementation example of the analyzer 200. To get. In the present embodiment, the data acquired from the sorting device 400 controlled by the analysis device 200 is acquired. Further, the target data acquisition unit 252 stores the acquired target data in the target data DB 2421A.

The analysis unit 253 refers to the target data acquired by the target data acquisition unit 252 and stored in the target data DB 242 based on the trained model acquired by the model acquisition unit 251 and stored in the trained model DB 241. Performs predetermined data analysis as an engine. Specifically, in the above-mentioned example, in the case of reading the character information for the handwritten character by OCR, the image data of the handwritten character is analyzed, and in the case of separating the industrial waste, the image data of the industrial waste is used. This is an analysis, and in the present embodiment, the image data acquired from the sorting device 400 is analyzed.

The result output unit 254 outputs the result data and various control signals according to the purpose of the trained model, that is, the purpose of the AI engine, based on the result of data analysis on the data to be analyzed by the analysis unit 253. In the above example, in order to output the judgment result of the character information when reading the character information by OCR for the handwritten character and to move the handwritten character to the target industrial waste when separating the industrial waste. It is an output of a control signal, and in the present embodiment, the operation control of the sorting device 400 is performed.

When the target data providing unit 255 provides the target data acquired by the target data acquisition unit 252 to the learning server 100 at the user's choice, the target data providing unit 255 transmits the target data to the learning server 100 via the communication unit 210. By performing machine learning using the target data of the data analysis as the AI engine, the trained model of the learning server 100 can perform more learning, so that the target data can be provided to the learning server 100. desirable. However, some users may not want to provide the target data to the learning server 100, which is a learning center that is a service provided by another company. Therefore, the target data is low-deficient only when the user desires.

FIG. 5 is a perspective view showing the appearance of the sorting device 400 controlled by the analysis device 200 of FIG. The sorting device 400 is a device that sorts industrial waste, which is an example of various devices controlled by the analysis device 200, and includes a fixed portion 410, a mounting movable portion 420, an upper arm portion 430, and a joint portion 440. It is composed of a forearm portion 450, a wrist portion 460, a support portion 470, and a holding upper portion 480. The sorting device 400 captures the industrial waste X placed on the lane L of the belt conveyor shown in FIG. 5 with a camera (not shown), analyzes the image data, and analyzes the material. It is a robot arm for sorting by material. The cameras are arranged, for example, on lane L or on the wrist 460.

The fixing portion 410 is a place where the sorting device 400 is fixed to the mounting table D. The mounting table D is fixedly arranged at a predetermined position in the vicinity of the lane L, for example. The mounting movable portion 420 is a movable portion to which one end of the upper arm portion 430 is connected and the angle with the upper surface of the mounting base D can be changed. A servomotor is built in the mounting movable portion 420, and the servomotor is driven by a control signal to rotate the upper arm portion 430. The upper arm portion 430 is a rod-shaped member on the mounting table D side in the sorting device 400.

The joint portion 440 is a portion that rotatably connects the other end of the upper arm portion 430 and one end of the forearm portion 450. A servomotor is built in the joint portion 440, and the servomotor is driven by a control signal in the same manner as the mounted movable portion 420 to rotate the upper arm portion 430. The forearm portion 450 is a rod-shaped member on the distal end side of the sorting device 400.

The wrist portion 460 is a portion that rotatably connects the other end of the forearm portion 450 and one end of the support portion 470. A servomotor is built in the wrist portion 460, and the servomotor is driven by a control signal in the same manner as the mounted movable portion 420 to rotate the support portion 470. The support portion 470 is a portion that supports the holding upper portion 480 at the tip end portion of the sorting device 400. The upper part 480 is a place where the industrial waste X is sandwiched and lifted. A servomotor is built in the holding upper part 480, and the servomotor is driven by a control signal in the same manner as the mounted movable portion 420, so that the holding upper part 480 is driven.

In the sorting device 400, machine learning is performed by image data obtained by capturing the industrial waste X with a camera, and the relationship between the feature amount and the material of the industrial waste X by data analysis of the image data is generated as a learned model. .. By providing this image data to the learning server 100, the learning unit 132 generates a more accurate trained model by performing machine learning of the industrial waste X actually placed on the lane L. However, it is possible to build an AI engine.

<Processing flow>
The processing flow of an example of the information processing method executed by the learning server 100 and the analysis device 200 of the information processing system 1 will be described. First, with reference to FIG. 6, the flow of machine learning processing, which is a part of the information processing method, executed by the learning server 100 will be described. FIG. 6 is a flowchart showing the operation of the machine learning process in the information processing system 1 of FIG.

In the case of the information processing system 1A shown in FIG. 2, the learning server 100 and the analysis device 200 do not have to be connected by the communication means T while the machine learning process is performed.

As the process of step S101, the target data providing unit 255 of the analysis device 200 transmits the analysis target data to be the target of the predetermined data analysis by the user's selection, so that the target data acquisition unit 131 of the learning server 100 communicates. Obtained via unit 110. In the example of the sorting device 400 shown in FIG. 5, the industrial waste X placed on the lane L is imaged by a camera and image data is acquired. The acquired target data is stored in, for example, the target data provided by the analysis device 201 in the target data DB 121A, and the target data provided by the analysis device 202 in the target data DB 121B.

As the process of step S102, the learning unit 132 performs machine learning based on the target data acquired in step S101 and stored in the

target data DBs

121A and 121B.

As the process of step S103, the learning unit 132 generates a trained model based on the result of machine learning performed in step S102 and stores it in the trained model DB 122, or has trained stored in the trained model DB 122. The model is updated. The trained model is updated, for example, by an aggregation process that merges the updated information based on the result of machine learning with the trained model stored in the trained model DB 122.

Next, with reference to FIG. 7, the flow of data analysis processing, which is a part of the information processing method, executed by the analysis device 200 will be described. FIG. 7 is a flowchart showing the operation of the data analysis process in the information processing system 1 of FIG.

In the case of the information processing system 1A shown in FIG. 2, while this data analysis process is being performed, the learning server 100 and the analysis device 200 are connected by the communication means T so that communication is possible.

As the process of step S201, in the providing unit 133 of the learning server 100, the learned model generated in step S103 by the user's operation is transmitted to the analysis device 200, so that the model acquisition unit 251 of the analysis device 200 learns. The learned model provided by the server 100 as a learning center is acquired via the communication unit 210. The acquired trained model is stored in the trained model DB 241.

As the process of step S202, the target data acquisition unit 252 acquires the target data to be analyzed by the analysis device 200 as the AI engine. In the example of the sorting device 400 shown in FIG. 4, the industrial waste X placed on the lane L is imaged by a camera and image data is acquired. The acquired analysis target data is stored in the analysis target data DB 142.

As the process of step S203, the analysis unit 253 refers to the target data acquired in step S201 and stored in the target data DB 242 based on the trained model acquired in step S201 and stored in the trained model DB 241. A predetermined data analysis as an AI engine is performed. In the example of the sorting device 400 shown in FIG. 4, the image data of the industrial waste X is analyzed, and the type of the material of the industrial waste X is determined.

As the process of step S204, the result output unit 254 outputs the result data and various control signals based on the result of the data analysis on the target data performed in step S203. In the example of the sorting device 400 shown in FIG. 4, a control signal for operating the sorting device 400 is output to operate the mounting movable portion 420, the joint portion 440, the wrist portion 460, and the holding upper portion 480, and the industrial waste X is carried. It is lifted by the upper part 480 and moved to an appropriate place according to the type of material of industrial waste X.

Next, with reference to FIG. 8, the flow of the billing amount determination process, which is a part of the information processing method, executed by the learning server 100 will be described. FIG. 8 is a flowchart showing the operation of the billing amount determination process in the information processing system 1 of FIG.

As the process of step S301, the operation unit 230 of the analysis device 200 accepts whether or not the target data is provided by the target data providing unit 255 by the user's operation. If it is instructed to provide the target data, the target data will be provided, and if it is instructed not to provide the target data, the target data will not be provided.

As the process of step S302, the billing amount determination unit 134 of the learning server 100 determines the billing amount, which is the consideration for the provision of the learned model. For example, in step S301, if the acquired target data is provided to the learning server 100, the billing amount is set at a low price, and if the target data is not provided to the learning server 100, the billing amount is expensive. Set.

<Effect>
As described above, the information processing system and the information processing method according to the present embodiment acquire the target data to be analyzed, perform machine learning to generate a trained model, and provide the learning to the analysis device. It is equipped with a server (learning center) and an analysis device (AI engine) that analyzes data based on the trained model and outputs data according to the purpose of the trained model. Therefore, the result of machine learning performed by the learning server can be used by a plurality of analysis devices. As a result, the labor of machine learning using a large amount of teacher data can be reduced, and a trained model that is the result of machine learning performed by another person can be used.

Also, by separating the learning server and the analysis device, the analysis device is separated from the learning server and can be used in the local environment. This makes it possible to use the AI engine in a local environment even if there is resistance to providing various data of the company on the cloud server.

Furthermore, the target data provided by oneself is configured so that it cannot be referred to by others (other companies). Therefore, since there is no possibility that the data owned by the other company will be referred to by other companies, it is considered that the number of people who accept the provision of the data will increase, and it will be possible to make the trained model perform more learning. This makes it possible to build a more accurate AI engine.

(Embodiment 2)
FIG. 9 is a functional block configuration diagram showing an analysis device 200B of the information processing system 1 according to the second embodiment of the present disclosure. The analysis device 200B acquires a trained model from the training server 100, analyzes the target data to be analyzed as an AI engine based on the trained model, and outputs an output according to the purpose of the trained model. It is the same as the analysis device 200 according to the first embodiment in that it is an apparatus for performing the data, but is different from the information processing system 1 according to the first embodiment in that the input reception unit 256 is provided as a function of the control unit 250. ..

In the present embodiment, when the learning server 100 performs machine learning, it is possible to input tag information related to the analysis target data which is the target of machine learning.

The input reception unit 256 receives the input of tag information related to the analysis target data, which is the target of machine learning, by the operation of the operation unit 230 of the user at the time of machine learning by the learning unit 132 of the learning server 100. The tag information is, for example, information on the date and time and conditions when the analysis target data is acquired.

Further, even if the learning unit 132 of the learning server 100 processes the annotation that associates the tag information related to the analysis target data received by the input reception unit 256 with the analysis target data and updates the trained model. Good. Other configurations and processing flows are the same as those in the first embodiment.

According to the present embodiment, in addition to the effect of the first embodiment, an input receiving unit for receiving input of tag information related to analysis target data is provided, and the input tag information is associated with the analysis target data. Processing is done. Therefore, machine learning by the learning unit is performed more appropriately, and more accurate model information can be generated.

(Embodiment 3)
FIG. 10 is a functional block configuration diagram showing an analysis device 200C of the information processing system 1 according to the third embodiment of the present disclosure. The analysis device 200C acquires a trained model from the training server 100, analyzes the target data to be analyzed as an AI engine based on the trained model, and outputs an output according to the purpose of the trained model. It is the same as the analysis device 200 according to the first embodiment in that it is an apparatus for performing the data, but is different from the information processing system 1 according to the first embodiment in that the evaluation unit 257 is provided as a function of the control unit 250.

In this embodiment, the result of data analysis is evaluated, and machine learning is performed based on the evaluation result.

The evaluation unit 257 evaluates the result of data analysis and generates evaluation result data for the evaluation result. Further, the evaluation result data is transmitted to the learning server 100. The evaluation result data may be, for example, data showing the relationship between the target data and the data of the analysis result by the trained model, or the data obtained by evaluating the trained model.

Further, the learning unit 132 of the learning server 100 may select the target data for machine learning based on the evaluation result data. For example, when an abnormal value is detected as a result of data analysis, if the target data is provided to the learning server 100, it becomes a target of machine learning and machine learning based on the abnormal value is performed. Therefore, by showing that it is noise, it is possible to prevent it from affecting machine learning. Other configurations and processing flows are the same as those in the first embodiment.

According to the present embodiment, in addition to the effect of the first embodiment, the provision of the evaluation unit makes it possible to select the target data for machine learning based on the evaluation result data. Therefore, for example, when an abnormal value is detected as a result of data analysis, if the target data is provided to the learning server, it becomes a target of machine learning and machine learning based on the abnormal value is performed. It is also possible to prevent it from affecting machine learning. This enables more accurate machine learning.

(Embodiment 4 (Program))
FIG. 10 is a functional block configuration diagram showing an example of the configuration of the computer (electronic computer) 700. The computer 700 includes a CPU 701, a main storage device 702, an auxiliary storage device 703, and an interface 704.

Here, the target data acquisition unit 131, the learning unit 132, the provision unit 133, the charge amount determination unit 134, the model acquisition unit 251 and the target data acquisition unit 252, and the analysis unit 253 according to the first to third embodiments. The details of the control program (information processing program) for realizing each function constituting the result output unit 254, the target data providing unit 255, the input receiving unit 256, and the evaluation unit 257 will be described. These functional blocks are implemented in the computer 700. The operation of each of these components is stored in the auxiliary storage device 703 in the form of a program. The CPU 701 reads the program from the auxiliary storage device 703, expands it to the main storage device 702, and executes the above processing according to the program. Further, the CPU 701 secures a storage area corresponding to the above-mentioned storage unit in the main storage device 702 according to the program.

Specifically, the program is a target data acquisition step for acquiring the target data to be analyzed in the computer 700, and model information for performing machine learning based on the target data and performing data analysis. Based on the training step of generating a trained model or updating the trained model, the providing step of providing the trained model to the analyzer, the model acquisition step of acquiring the trained model with the analyzer, and the trained model. This is a control program that realizes an analysis step for analyzing the data of the target data and a result output step for outputting the trained model according to the purpose of the trained model based on the result of the data analysis.

The auxiliary storage device 703 is an example of a tangible medium that is not temporary. Other examples of non-temporary tangible media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memories, etc. connected via interface 704. When this program is distributed to the computer 700 via the network, the distributed computer 700 may expand the program to the main storage device 702 and execute the above processing.

Further, the program may be for realizing a part of the above-mentioned functions. Further, the program may be a so-called difference file (difference program) that realizes the above-mentioned function in combination with another program already stored in the auxiliary storage device 703.

Although the embodiments related to the disclosure have been described above, these can be implemented in various other embodiments, and can be implemented by making various omissions, replacements, and changes. These embodiments and modifications, as well as those omitted, replaced and modified, are included in the technical scope of the claims and the equivalent scope thereof.

1,1A Information system, 100 learning server, 110 communication unit, 120 storage unit, 121A, 121B target data DB, 122 learned model DB, 130 control unit, 131 target data acquisition unit, 132 learning unit, 133 providing unit, 134 Billing amount determination unit, 200, 200B, 200C analyzer, 210 communication unit, 220 display unit, 230 operation unit, 240 storage unit, 241 trained model DB, 242 target data DB, 250 control unit, 251 model acquisition unit, 252 Target data acquisition unit, 253 analysis unit, 254 result output unit, 255 target data provision unit, 256 input reception unit, 257 evaluation unit, NW network, T communication means

Claims

An information processing system including a learning server that performs machine learning to generate a specific trained model and an analysis device that analyzes data using the trained model.
The learning server
The target data acquisition unit that acquires the target data to be the target of the data analysis,
A learning unit that performs machine learning based on the target data, generates the trained model that is model information for performing the data analysis, or updates the trained model.
A providing unit that provides the trained model to the analysis device is provided.
The analyzer is
A model acquisition unit that acquires the trained model from the learning server,
An analysis unit that analyzes the target data based on the trained model, and
An information processing system including a result output unit that outputs a result according to the purpose of the trained model based on the result of the data analysis.
The analysis device includes a target data providing unit that provides the target data to the learning server.
The information processing system according to claim 1, wherein the learning server acquires the target data provided by the analysis device by the target data acquisition unit.
The target data acquisition unit acquires and stores the target data provided by the analysis device so that it cannot be referred to by other analysis devices.
The learning unit according to claim 2, wherein the learning unit generates the trained model or updates the trained model so that the target data provided by the analysis device cannot be referred to by another analysis device. Information processing system.
The learning server includes a billing amount determining unit that determines a billing amount when the trained model is provided to the analysis device, and receives the provision of the target data from the analysis device and provides the target data. The information processing system according to claim 2 or 3, wherein the billing amount differs depending on whether or not the data is received.
The analysis device includes an input receiving unit that receives input of tag data related to the target data.
The target data providing unit provides the tag data together with the target data to the learning server.
The target data acquisition unit acquires the tag data from the analysis device, performs annotations relating the tag data to the target data, and then performs annotations.
The information processing system according to any one of claims 2 to 4, wherein the learning unit performs machine learning based on the target data and the tag data.
The analysis device includes an evaluation unit that evaluates the result of the data analysis based on the trained model and generates evaluation result data.
The target data providing unit provides the evaluation result data together with the target data to the learning server.
The target data acquisition unit acquires the evaluation result data from the analysis device, and obtains the evaluation result data.
The information processing system according to any one of claims 2 to 5, wherein the learning unit selects the target data based on the evaluation result data and performs machine learning based on the selected target data.
An information processing method that performs machine learning to generate a specific trained model and provides the trained model to an analysis device that performs data analysis using the trained model.
The target data acquisition step of acquiring the target data to be the target of the data analysis performed by the target data acquisition unit, and the target data acquisition step.
A learning step performed by the learning unit to perform machine learning based on the target data, generate the trained model which is model information for performing the data analysis, or update the trained model.
The providing step of providing the trained model to the analysis device performed by the providing unit, and
A model acquisition step of acquiring the trained model by the analysis device performed by the model acquisition unit, and
An analysis step performed by the analysis unit to analyze the target data based on the trained model, and
An information processing method including a result output step of performing output according to the purpose of the trained model based on the result of the data analysis performed by the result output unit.
An information processing program that performs machine learning to generate a specific trained model and provides the trained model to an analysis device that performs data analysis using the trained model.
The target data acquisition step for acquiring the target data to be the target of the data analysis, and
A learning step of performing machine learning based on the target data, generating the trained model which is model information for performing the data analysis, or updating the trained model.
A provision step of providing the trained model to the analysis device, and
A model acquisition step of acquiring the trained model with the analysis device, and
An analysis step for analyzing the target data based on the trained model, and
An information processing program for causing a computer to execute a result output step that outputs a result according to the purpose of the trained model based on the result of the data analysis.