Learning system and training method

Abstract

Provided is a learning system comprising a plurality of pieces of decentralized infrastructure arranged on a core network, and a management device for managing the plurality of pieces of decentralized infrastructure. Each piece of the plurality of pieces of decentralized infrastructure has: a first reception unit for receiving training data from a plurality of communication terminals; a training data storage unit for storing a plurality of pieces of training data; a first execution unit that includes a decentralized training function for performing machine learning of a training model by using the plurality of pieces of training data stored in the training data storage unit, and a RAN control function for executing AI processing (RAN Intelligent Controller (RIC)) or the like to control the function of the RAN; and a transmission unit for transmitting the training result to the management device. The management device has a distribution unit for distributing the training model to each piece of decentralized infrastructure, a second reception unit for receiving a training result from each piece of decentralized infrastructure, a training result storage unit for storing a plurality of training results, and a second execution unit that has a central training function for performing machine learning on a training model by using the plurality of training results stored in the training result storage unit.

Description

Learning system and learning method

[0001] The present invention relates to a learning system and a learning method.

[0002] Patent Document 1 describes a technique related to federated learning for deep learning models. Patent Document 2 describes a program related to federated learning. [Prior Art Literature] [Patent Document] [Patent Document 1] JP-A-2023-541649 [Patent Document 2] JP-A-2023-128056

[0003] In recent years, research on federated learning, a machine learning technique, has been actively conducted. In federated learning, an AI (Artificial Intelligence) model server distributes an AI model to be learned to each edge device. Each edge device performs machine learning of the AI ​​model using training data that exists only on that edge device, and reports only the update differences of the AI ​​model to the AI ​​model server. The AI ​​model server then comprehensively updates the AI ​​model to be learned based on the update differences of the AI ​​model acquired from each edge device. Federated learning is used, for example, for training predictive text conversion on smartphones. One advantage of federated learning is that it allows the AI ​​model to be learned by machine while maintaining the confidentiality of the learning data used in the AI ​​model's machine learning. Another advantage of federated learning is that it can reduce the amount of data communication required for the AI ​​model's machine learning. On the other hand, in federated learning, since each edge device learns the AI ​​model by machine learning, the computational resources available for the AI ​​model's machine learning are limited. Therefore, it is not possible to use federated learning to machine-learn large-scale AI models such as LLMs (Large Language Models), which have a huge number of parameters or require huge amounts of training data for the training process.

[0004] In the system according to this embodiment, for example, by running the functions of a RAN (Radio Access Network) on a high-performance GPU (Graphics Processing Unit) server rather than on a general-purpose server, the surplus computing resources can be utilized for AI processing. Types of AI processing include AI processing related to RAN control (sometimes referred to as RAN control AI processing) and AI processing not related to RAN control (sometimes referred to as non-RAN control AI processing).

[0005] An example of RAN control AI processing is RIC (RAN Intelligent Controller). RIC is a technology that uses AI to optimize RAN radio resources and automate RAN operations. RIC includes Non-RT RIC and Near-RT RIC (Near-Real Time RIC). Non-RT RIC is sometimes called Centralized RIC. Non-RT RIC is located inside SMO (Service Management and Orchestration), which manages and orchestrates the RAN. Non-RT RIC generates and notifies policies related to RAN control and sends information to Near-RT RIC. For example, the Non-RT RIC performs machine learning using data collected from the RAN to generate a trained model for RAN control and transmits it to the Near-RT RIC. The Near-RT RIC is sometimes called a Distributed RIC. Compared to the Non-RT RIC, the Near-RT RIC is located closer to the RAN nodes (RU (Radio Unit), DU (Distributed Unit), CU (Central Unit)) and controls the RAN nodes, resources, etc. The Near-RT RIC performs processing with higher real-time performance than the Non-RT RIC. The Near-RT RIC performs inference processing related to RAN control using, for example, a trained model acquired from the Non-RT RIC. RAN control AI processing is not limited to the RIC.

[0006] The non-RAN control AI processing may correspond to a so-called MEC (Multi-access Edge Computing) application. Examples of non-RAN control AI processing include a predictive conversion AI execution process that outputs predictive conversion for input character data, a monitoring AI execution process that determines the situation within the imaging range of an input captured image, and a response AI execution process that outputs a response to an input user inquiry, but are not limited to these.

[0007] In the system according to this embodiment, for example, a distributed data center architecture is configured, and a mechanism for performing federated learning is adopted by making each distributed data center the scope of training data sharing. As a result, the system according to this embodiment can perform machine learning of an AI model using a distributed data center that has more abundant computing resources than an edge device. Furthermore, since the system according to this embodiment does not transfer training data to layers above the MEC, it can enjoy the advantage of federated learning, that is, it can perform machine learning of an AI model while maintaining the confidentiality of the training data used in the machine learning of the AI ​​model.

[0008] According to one embodiment of the present invention, there is provided a learning system for machine learning a learning model. The learning system may include a plurality of distributed platforms arranged on a core network. The learning system may include a management device that manages the plurality of distributed platforms. Each distributed platform of the plurality of distributed platforms may include a first receiving unit that receives learning data used in machine learning of the learning model from a plurality of communication terminals, a learning data storage unit that stores the plurality of learning data, a first execution unit including a distributed learning function that machine-learns the learning model using the plurality of learning data stored in the learning data storage unit, and a transmitting unit that transmits learning results obtained by machine learning the learning model using the plurality of learning data to the management device. The management device may include a distribution unit that distributes the learning model to each of the distributed platforms, a second receiving unit that receives the learning results from each of the distributed platforms, a learning result storage unit that stores the plurality of learning results, and a second execution unit having a central learning function that machine-learns the learning model using the plurality of learning results stored in the learning result storage unit.

[0009] In the learning system, the distribution unit may distribute the learned learning model, which has been machine-learned using the multiple learning results, to each of the distributed platforms.

[0010] In any of the learning systems, the learning data storage unit of each of the distributed platforms may store the multiple learning data by storing the learning data for each communication terminal of the multiple communication terminals, and the distributed learning function of each of the distributed platforms may machine-learn the learning model using the multiple learning data by machine-learning the learning model using the learning data for each communication terminal.

[0011] In any of the learning systems, each of the distributed platforms may further have a classification unit that groups the multiple communication terminals into groups that can share the learning data, and the learning data storage unit of each of the distributed platforms may store the multiple learning data by storing the learning data for at least one communication terminal among the multiple communication terminals that has been grouped into the same group by the classification unit, and the distributed learning function of each of the distributed platforms may machine-learn the learning model using the multiple learning data by machine-learning the learning model using the learning data for each of the at least one communication terminal.

[0012] In any of the learning systems, the learning system may perform machine learning to learn a plurality of the learning models, and the management device and at least two distributed platforms located in a first area of ​​the plurality of distributed platforms may perform machine learning to learn a first learning model of the plurality of learning models, and the management device and at least two distributed platforms located in a second area different from the first area of ​​the plurality of distributed platforms may perform machine learning to learn a second learning model of the plurality of learning models.

[0013] In any of the learning systems, the learning system may machine-learn a plurality of the learning models, and the learning system may include a plurality of the management devices, and a management device among the plurality of management devices that is located in a first area and at least two distributed platforms among the plurality of distributed platforms that are located in the first area may machine-learn a first learning model among the plurality of learning models, and a management device among the plurality of management devices that is located in a second area different from the first area and at least two distributed platforms among the plurality of distributed platforms that are located in the second area may machine-learn a second learning model among the plurality of learning models.

[0014] In any of the learning systems, the first execution unit of each of the distribution infrastructures may further include a RAN control function that controls the function of a RAN.

[0015] In any of the learning systems, the distributed learning function of each of the distributed platforms may determine the timing to machine-learn the learning model using the multiple learning data stored in the learning data storage unit based on the prediction results of the RAN control function predicting demand for the RAN.

[0016] In any of the learning systems described above, the management device may be located on the Internet.

[0017] In any of the learning systems, the management device may be implemented in a management infrastructure that is located on the core network and manages the multiple distributed infrastructures.

[0018] In any of the above learning systems, the learning model may be an LLM.

[0019] According to one embodiment of the present invention, there is provided a learning method for machine learning a learning model. The learning method may include a distribution step in which a management device distributes the learning model to each of a plurality of distributed infrastructures arranged on a core network. The learning method may include a first reception step in which each of the distributed infrastructures receives, from a plurality of communication terminals, learning data used for machine learning of the learning model distributed in the distribution step. The learning method may include a first storage step in which each of the distributed infrastructures stores, in a learning data storage unit included in each of the distributed infrastructures, the plurality of learning data received in the first reception step. The learning method may include a distributed learning step in which each of the distributed infrastructures machine-learns the learning model using the plurality of learning data stored in the learning data storage unit in the first storage step. The learning method may include a transmission step in which each of the distributed infrastructures transmits to the management device learning results obtained by machine learning the learning model using the plurality of learning data in the distributed learning step. The learning method may include a second reception step in which the management device receives the learning results from each of the distributed infrastructures. The learning method may include a second storage step in which the management device stores the plurality of learning results received in the second receiving step in a learning result storage unit included in the management device. The learning method may include a central learning step in which the management device machine-learns the learning model using the plurality of learning results stored in the learning result storage unit in the second storing step.

[0020] The above summary of the invention does not list all of the necessary features of the present invention, and subcombinations of these features may also constitute inventions.

[0021] FIG. 1 shows an example of a system 10. FIG. 2 shows another example of a system 10. FIG. 3 shows an explanatory diagram for explaining an example of a processing flow of the system 10. FIG. 4 shows an explanatory diagram for explaining an example of a processing flow of the system 10. FIG. 5 shows an explanatory diagram for explaining an example of a processing flow of the system 10. FIG. 6 shows an explanatory diagram for explaining an example of a processing flow of the system 10. FIG. 7 shows an explanatory diagram for explaining an example of a processing flow of the system 10. FIG. 8 shows another example of a system 10. FIG. 9 shows an explanatory diagram for explaining an example of a processing flow of the system 10. FIG. 10 shows an example of a processing flow of the system 10.

[0022] The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the scope of the invention as claimed. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0023] 1 schematically illustrates an example of a system 10. The system 10 may include a distributed infrastructure 200. The system 10 may include a plurality of distributed infrastructures 200. The system 10 may include a management device 100 that manages the plurality of distributed infrastructures 200. In the system 10 according to this embodiment, for example, the management device 100 and the plurality of distributed infrastructures 200 may cooperate to control the RAN 300 and perform AI processing.

[0024] The RAN 300 may be a virtualized vRAN (Virtual RAN), and the system 10 may control the vRAN. The RAN 300 may be a physical RAN, and the system 10 may control the physical RAN. In this embodiment, a case where the RAN 300 is a vRAN will be mainly described as an example.

[0025] The AI ​​processing performed by the system 10 may include RAN-controlled AI processing (sometimes referred to as RAN_AI). The AI ​​processing performed by the system 10 may include non-RAN-controlled AI processing (sometimes referred to as non-RAN_AI).

[0026] The distributed infrastructure 200 may be a data center located in various locations. The distributed infrastructure 200 may be configured with multiple devices. The distributed infrastructure 200 may be realized on a virtualization platform made up of multiple devices. The distributed infrastructure 200 may also be realized by a single device. In other words, the distributed infrastructure 200 may be a distributed device.

[0027] The distributed infrastructure 200 may be located on a core network, which may include both inside and outside the core network.

[0028] The core network may be compliant with any mobile communication system. For example, the core network may be compliant with a 5G (5th Generation) communication system. The core network may be compliant with a 6G (6th Generation) communication system or later mobile communication system. The core network may be compliant with a 3G (3rd Generation) communication system or an LTE (Long Term Evolution) communication system.

[0029] For example, when the core network conforms to a 5G communication system and a telecommunications carrier that manages the core network provides a 5G network to companies, local governments, etc., the RAN 300 may include a so-called private 5G network. When the core network conforms to a 5G communication system, the RAN 300 may include a so-called public 5G network.

[0030] The distributed infrastructure 200 may be provided with one or more central processing units (CPUs). The distributed infrastructure 200 may be provided with one or more GPUs. The distributed infrastructure 200 may be provided with multiple super chips, each of which has a CPU and a GPU connected via an interconnect. The interconnect may have memory consistency and may be capable of achieving high bandwidth and low latency. In this way, the distributed infrastructure 200 may have CPU resources and GPU resources as computational resources.

[0031] The distributed infrastructure 200 includes, for example, an execution unit including a distributed learning function that performs machine learning of a learning model using learning data. The execution unit further includes, for example, a RAN control function that controls the functions of the RAN 300. The distributed learning function and the RAN control function may use the same computational resources.

[0032] The RAN control function controls, for example, the radio base station 30 that constitutes the RAN 300. The RAN control function controls, for example, the radio base station 30 so as to form a radio communication area using an antenna and provide mobile communication services to communication terminals 50 within the radio communication area.

[0033] The communication terminal 50 is a mobile phone such as a smartphone. The communication terminal 50 may be a tablet terminal, a PC (Personal Computer), or the like. The communication terminal 50 may be a so-called IoT (Internet of Things) device. The communication terminal 50 may include anything that falls under the so-called IoE (Internet of Everything) category.

[0034] The management device 100 is located, for example, on the Internet 40. The management device 100 and the distribution infrastructure 200 communicate with each other, for example, via the Internet 40.

[0035] The system 10 according to the present embodiment may be a learning system that performs machine learning to generate a learning model. In the system 10 according to the present embodiment, for example, the management device 100 and multiple distributed infrastructures 200 cooperate to generate a learning model by machine learning. In the system 10 according to the present embodiment, for example, the management device 100 and multiple distributed infrastructures 200 cooperate to perform distributed learning. In the system 10 according to the present embodiment, for example, the management device 100 and multiple distributed infrastructures 200 cooperate to perform federated learning. Details of the process of the management device 100 and multiple distributed infrastructures 200 cooperating to generate a learning model by machine learning will be described later.

[0036] The learning model may be, for example, an LLM, an image recognition model, or any other learning model.

[0037] 2 is a schematic diagram of another example of the system 10. Here, differences from the system 10 shown in FIG. 1 will be mainly described.

[0038] The management device 100 is placed on, for example, the core network 60. In this case, the management device 100 may be implemented in a management infrastructure that manages a plurality of distribution infrastructures 200.

[0039] The management infrastructure may be a data center that manages multiple distributed infrastructures 200. The management infrastructure may be configured on multiple devices. The management infrastructure may be realized on a virtualization infrastructure made up of multiple devices. The management infrastructure may also be realized by a single device.

[0040] The management infrastructure may be called a Core Brain, and the distributed infrastructure 200 may be called a Regional Brain. For example, a single-level distributed infrastructure 200 may be placed under the management infrastructure. For example, multiple levels of distributed infrastructures 200 may be placed under the management infrastructure. For example, when a two-level distributed infrastructure 200 is placed under the management infrastructure, the management infrastructure may be called a Core Brain, the distributed infrastructure 200 in the level below that may be called a Regional Brain, and the distributed infrastructure 200 in the level below that may be called a Sub-Regional Brain.

[0041] 3 to 8 are explanatory diagrams illustrating an example of the processing flow of the system 10. Here, an example in which the management device 100 and multiple distribution platforms 200 cooperate to perform federated learning will be mainly described.

[0042] 3 , in Step 1, the management device 100 distributes the learning model to be learned to each of the multiple distributed infrastructures 200. The management device 100 distributes the learning model to be learned to each of the distributed infrastructures 200, for example, via the Internet 40 and the core network 60.

[0043] Each distributed infrastructure 200 receives the learning model to be learned from the management device 100. Each distributed infrastructure 200 receives the learning model to be learned from the management device 100, for example, via the Internet 40 and the core network 60.

[0044] 4 , in Step 2, each distributed infrastructure 200 receives, from a plurality of communication terminals 50 within the RAN 300, learning data to be used for machine learning of the learning model received from the management device 100 in Step 1. Each distributed infrastructure 200 receives the learning data from a plurality of communication terminals 50 within the RAN 300 via a radio base station 30, for example.

[0045] The learning data is, for example, data including personal information of the user of the communication terminal 50. The learning data is, for example, data including confidential information of the organization to which the user of the communication terminal 50 belongs.

[0046] An organization may be, for example, a business. An organization may be, for example, a local government. An organization may be, for example, a school. An organization may be, for example, a non-profit organization. An organization may be, for example, a family. An organization may be any other group of people working together to achieve a particular goal.

[0047] The training data may include, for example, data in text format. The training data may include, for example, data in image format. The training data may be data in any other format. The training data may include data in multiple formats, such as data in text format and data in image format.

[0048] 5, in Step 3, each distributed infrastructure 200 performs machine learning to generate a learning model using the multiple pieces of learning data received from the multiple communication terminals 50 in the RAN 300 in Step 2. Each distributed infrastructure 200 performs machine learning to generate a learning model, for example, by supervised learning. Each distributed infrastructure 200 performs machine learning to generate a learning model, for example, by unsupervised learning. Each distributed infrastructure 200 may also perform machine learning to generate a learning model, for example, by reinforcement learning.

[0049] 6 , in Step 4, each distributed infrastructure 200 transmits the learning results obtained by machine learning the learning model using multiple pieces of learning data in Step 3 to the management device 100. Each distributed infrastructure 200 transmits the learning results to the management device 100, for example, via the core network 60 and the Internet 40.

[0050] The management device 100 receives the learning results from each distributed infrastructure 200. The management device 100 receives the learning results from each distributed infrastructure 200 via the core network 60 and the Internet 40, for example.

[0051] The learning result includes, for example, difference information indicating the difference between a learning model before machine learning by the distributed infrastructure 200 and a learning model after machine learning by the distributed infrastructure 200. The learning result may also include data amount information indicating the amount of learning data used by the distributed infrastructure 200 for machine learning of the learning model.

[0052] 7 , in Step 5, the management device 100 performs machine learning to generate a learning model using the multiple learning results received from each distributed infrastructure 200 in Step 4. The management device 100 performs machine learning to generate a learning model, for example, based on difference information included in the multiple learning results received from each distributed infrastructure 200. The management device 100 performs machine learning to generate a learning model, for example, based further on data volume information included in the multiple learning results received from each distributed infrastructure 200.

[0053] 8, in Step 6, the management device 100 distributes the trained learning model trained in Step 5 to each distributed infrastructure 200. Each distributed infrastructure 200 receives the trained learning model trained by the management device 100 from the management device 100. Thereafter, the accuracy of the learning model is improved by repeatedly executing Steps 2 to 6.

[0054] In conventional federated learning, a learning model server distributes a learning model to be learned to each edge device, such as a smartphone, and each edge device machine-trains the learning model using the training data and reports the training results to the learning model server. The learning model server then machine-trains the learning model based on the training results received from each edge device. However, the performance of the computational resources installed in each edge device, such as a smartphone, is lower than that of the computational resources installed in a server. Therefore, conventional federated learning cannot machine-train learning models that require high-performance computational resources, such as LLMs, which machine-train a learning model using a large amount of text-format training data, or image recognition models, which machine-train a learning model using image-format training data, which has a larger data volume than text-format training data. On the other hand, from the perspective of further reducing the risk of leaking personal information or confidential information, there is a high demand for machine-training LLMs, image recognition models, etc., while maintaining the confidentiality of the training data used in the machine learning of the learning model, as in federated learning.

[0055] Furthermore, the radio base stations that make up the RAN must meet very high SLAs (Service Level Agreements), and it is necessary to create mechanisms, such as redundancy, to prevent service disruptions. As a result, only a few tenths of the computational resources on the execution platform that runs the RAN service can be used. In other words, there are many unused computational resources. Furthermore, at night, even though the demand for RAN services decreases, computational resources are wasted in an attempt to maintain the same level of service as during the daytime. In other words, the system is designed to accommodate peak demand.

[0056] In contrast, according to the system 10 of this embodiment, multiple distributed infrastructures 200 having computational resources with higher performance than those installed in smartphones and the like perform machine learning of a learning model. Therefore, the system 10 of this embodiment can perform machine learning of a learning model that requires high-performance computational resources. Furthermore, since the multiple distributed infrastructures 200 are arranged on a core network, the multiple distributed infrastructures 200 can receive training data used for machine learning of the learning model from multiple communication terminals 50 in the RAN 300 via a mobile communication network, which is a more secure communication network than the Internet. Therefore, the system 10 of this embodiment can perform machine learning of a learning model while maintaining the training data used for machine learning of the learning model in a confidential state. Therefore, the system 10 of this embodiment can perform machine learning of a learning model that requires high-performance computational resources while maintaining the training data used for machine learning of the learning model in a confidential state.

[0057] Furthermore, according to the system 10 of this embodiment, the multiple distributed infrastructures 200 use the same computational resources to control the functions of the RAN 300 and to perform machine learning of the learning model. Therefore, for example, it is possible to allocate the computational resources of the distributed infrastructure 200 to the machine learning processing of the learning model within a range in which the SLAs of the radio base stations 30 constituting the RAN 300 are maintained. Therefore, the system 10 of this embodiment can improve the utilization efficiency of computational resources while maintaining the high SLAs of the radio base stations constituting the RAN.

[0058] 9 is a schematic diagram of another example of the system 10. Here, differences from the system 10 described above will be mainly described.

[0059] The system 10, for example, performs machine learning to generate multiple learning models. In the system 10 according to the embodiment shown in FIG. 9 , for example, the management device 100 and at least two distributed infrastructures 200 arranged in a first area among the multiple distributed infrastructures 200 included in the system 10 perform machine learning to generate a first learning model among the multiple learning models to be learned by the system 10. In the system 10 according to the embodiment shown in FIG. 9 , for example, the management device 100 and at least two distributed infrastructures 200 arranged in a second area different from the first area among the multiple distributed infrastructures 200 included in the system 10 perform machine learning to generate a second learning model among the multiple learning models to be learned by the system 10.

[0060] The management device 100 and at least two distributed infrastructures 200 arranged in the first area may cooperate to perform machine learning of a first learning model. The management device 100 and at least two distributed infrastructures 200 arranged in the second area may cooperate to perform machine learning of a second learning model.

[0061] The first learning model and the second learning model may be the same learning model, for example, both the first learning model and the second learning model are LLMs.

[0062] The first learning model and the second learning model may be different learning models, for example, the first learning model is an LLM and the second learning model is an image recognition model.

[0063] The first area and the second area may be any combination of different areas. For example, the first area may be the eastern Japan area and the second area may be the western Japan area. For example, the first area may be the Kanto area and the second area may be the Kansai area. The first area may be the Tokyo area and the second area may be the Osaka Prefecture area.

[0064] When machine learning a learning model is performed in the system 10 shown in Fig. 9, Step 1 in the example of the processing flow of the system 10 shown in Figs. 3 to 8 may be replaced with the following Step 1'. In Step 1', the management device 100 distributes the first learning model to at least two distributed platforms 200 located in a first area, and distributes the second learning model to at least two distributed platforms 200 located in a second area.

[0065] 9, by machine learning a learning model for each area, the learning model can be machine-trained with greater consideration given to regional characteristics. For example, if the learning model machine-trained for each area is an LLM, the LLM can be machine-trained with greater consideration given to language systems that may differ between regions, such as dialects and accents.

[0066] 10 is a schematic diagram of another example of the system 10. Here, differences from the system 10 described above will be mainly described.

[0067] The system 10 includes, for example, a plurality of management devices 100. Fig. 10 shows an example in which the system 10 includes two management devices 100. The system 10 may include three or more management devices 100.

[0068] 10 , for example, a management device 100 arranged in a first area among the multiple management devices 100 included in the system 10 and at least two distributed infrastructures 200 arranged in the first area among the multiple distributed infrastructures 200 included in the system 10 perform machine learning to learn a first learning model among the multiple learning models to be learned by the system 10. In the system 10 according to the embodiment shown in FIG. 10 , for example, a management device 100 arranged in a second area different from the first area among the multiple management devices 100 included in the system 10 and at least two distributed infrastructures 200 arranged in the second area among the multiple distributed infrastructures 200 included in the system 10 perform machine learning to learn a second learning model among the multiple learning models to be learned by the system 10.

[0069] The management device 100 located in the first area and at least two distributed infrastructures 200 located in the first area may cooperate to perform machine learning of a first learning model. The management device 100 located in the second area and at least two distributed infrastructures 200 located in the second area may cooperate to perform machine learning of a second learning model.

[0070] When machine learning a learning model is performed in the system 10 shown in FIG. 10, Step 1 in the example of the processing flow of the system 10 shown in FIGS. 3 to 8 may be replaced with the following Step 1″. In Step 1″, the management device 100 located in the first area distributes the first learning model to at least two distributed infrastructures 200 located in the first area, and the management device 100 located in the second area distributes the second learning model to at least two distributed infrastructures 200 located in the second area.

[0071] 11 shows an example of the functional configuration of the management device 100. The management device 100 includes a model storage unit 102, a distribution unit 104, a receiving unit 106, a learning result storage unit 108, and an execution unit 110. Note that it is not essential for the management device 100 to include all of these components.

[0072] The model storage unit 102 stores a learning model. The model storage unit 102 stores, for example, a plurality of learning models.

[0073] The model storage unit 102 stores, for example, a plurality of learning models for each area. For example, the model storage unit 102 stores a first learning model as a learning model that is machine-learned by at least two distributed infrastructures 200 arranged in a first area, and stores a second learning model as a learning model that is machine-learned by at least two distributed infrastructures 200 arranged in a second area.

[0074] The distribution unit 104 distributes the learning model to each distributed infrastructure 200 of the multiple distributed infrastructures 200. The distribution unit 104 distributes the learning model to each distributed infrastructure 200, for example, via the Internet 40. The distribution unit 104 distributes the learning model to each distributed infrastructure 200, for example, via the core network 60. The distribution unit 104 distributes the learning model to each distributed infrastructure 200, for example, via the Internet 40 and the core network 60.

[0075] The distribution unit 104 distributes the learning model to each distributed infrastructure 200, for example, according to the area in which each distributed infrastructure 200 is located. For example, the distribution unit 104 distributes the first learning model to at least two distributed infrastructures 200 that are located in a first area among the plurality of distributed infrastructures 200, and distributes the second learning model to at least two distributed infrastructures 200 that are located in a second area among the plurality of distributed infrastructures 200.

[0076] The receiving unit 106 receives a learning result obtained by machine learning a learning model using a plurality of pieces of learning data from each of the plurality of distributed infrastructures 200. The receiving unit 106 may store the plurality of learning results received from each of the distributed infrastructures 200 in the learning result storage unit 108.

[0077] The receiving unit 106 receives the learning results from, for example, each distributed infrastructure 200 via the Internet 40. The receiving unit 106 receives the learning results from, for example, each distributed infrastructure 200 via the core network 60. The receiving unit 106 receives the learning results from, for example, each distributed infrastructure 200 via the Internet 40 and the core network 60.

[0078] The receiving unit 106 receives, for example, from each of at least two distributed platforms 200 located in the first area, a learning result obtained by machine learning a first learning model using a plurality of pieces of learning data. The receiving unit 106 receives, for example, from each of at least two distributed platforms 200 located in the second area, a learning result obtained by machine learning a second learning model using a plurality of pieces of learning data.

[0079] The execution unit 110 executes various processes. The execution unit 110 includes, for example, a central learning function 111 that performs machine learning of a learning model. The execution unit 110 includes, for example, a RAN control function 113 that controls the functions of the RAN 300. The central learning function 111 and the RAN control function 113 may use the same computational resources.

[0080] The central learning function 111 performs machine learning of the learning model, for example, by executing an AI application. The central learning function 111 performs machine learning of the learning model, for example, by executing a non-RAN AI application.

[0081] The central learning function 111 performs machine learning on a learning model using, for example, multiple learning results obtained by the distributed platform 200 using multiple pieces of learning data, which are stored in the learning result storage unit 108. The central learning function 111 may store in the model storage unit 102 the trained learning model that has been machine-learned using the multiple learning results.

[0082] The central learning function 111 performs machine learning on a learning model, for example, based on difference information included in each of the multiple learning results. For example, the central learning function 111 calculates the average of the differences indicated by each difference information between the learning model before the distributed infrastructure 200 performed machine learning and the learning model after the distributed infrastructure 200 performed machine learning. The central learning function 111 may then perform machine learning on the learning model by updating the learning model based on the calculated average. Note that the learning model before the distributed infrastructure 200 performed machine learning is the learning model distributed by the distribution unit 104.

[0083] The central learning function 111 performs machine learning on the learning model, for example, further based on data volume information included in each of the multiple learning results. For example, the central learning function 111 calculates a weighted average of the differences between the learning model before machine learning by the distributed infrastructure 200 and the learning model after machine learning by the distributed infrastructure 200, as indicated by each piece of difference information, based on each piece of data volume information. The central learning function 111 weights each learning result, for example, so that a learning result with a larger amount of learning data used in machine learning of the learning model by the distributed infrastructure 200, as indicated by each piece of data volume information, is weighted more heavily. Thereafter, the central learning function 111 may perform machine learning on the learning model by updating the learning model based on the calculated weighted average.

[0084] The central learning function 111 performs machine learning on the first learning model, for example, using a plurality of learning results obtained by machine learning the first learning model using a plurality of pieces of learning data, which are stored in the learning result storage unit 108. The central learning function 111 performs machine learning on the second learning model, for example, using a plurality of learning results obtained by machine learning the second learning model using a plurality of pieces of learning data, which are stored in the learning result storage unit 108.

[0085] The distribution unit 104 distributes, for example, a trained learning model that the central learning function 111 has trained through machine learning using multiple learning results to each of the multiple distributed infrastructures 200. The distribution unit 104 distributes, for example, a trained first learning model that the central learning function 111 has trained through machine learning using multiple learning results to at least two distributed infrastructures 200 located in a first area. The distribution unit 104 distributes, for example, a trained second learning model that the central learning function 111 has trained through machine learning using multiple learning results to at least two distributed infrastructures 200 located in a second area.

[0086] The RAN control function 113 controls the functions of the RAN 300, for example, by executing an AI application. The RAN control function 113 controls the functions of the RAN 300, for example, by executing a RAN_AI application. The RAN control function 113 may control the functions of the RAN 300 by executing any other processing.

[0087] The RAN control function 113 controls the functions of the RAN 300, for example, based on the results of predicting demand for the RAN 300. For example, if the usage of the RAN 300 is on an increasing trend, the RAN control function 113 predicts that the demand for the RAN 300 will increase, and if the usage of the RAN 300 is on a decreasing trend, the RAN control function 113 predicts that the demand for the RAN 300 will decrease.

[0088] The central learning function 111 determines the timing for machine learning a learning model using multiple learning results stored in the learning result storage unit 108, for example, based on the prediction result obtained by the RAN control function 113 predicting demand for the RAN 300. For example, when the prediction result indicates that demand for the RAN 300 will decrease, the central learning function 111 machine-learns a learning model using multiple learning results. On the other hand, when the prediction result indicates that demand for the RAN 300 will increase, the central learning function 111 does not machine-learn a learning model using multiple learning results.

[0089] 12 shows an example of the functional configuration of the distributed infrastructure 200. The distributed infrastructure 200 includes a receiving unit 202, a model storage unit 204, a training data storage unit 206, an executing unit 208, a transmitting unit 212, and a classifying unit 214. Note that it is not essential that the distributed infrastructure 200 include all of these components.

[0090] The receiving unit 202 receives various data. The receiving unit 202 receives various data, for example, via the Internet 40. The receiving unit 202 receives various data, for example, via the core network 60. The receiving unit 202 receives various data, for example, via the Internet 40 and the core network 60. The receiving unit 202 receives various data, for example, via the radio base station 30. The receiving unit 202 receives various data, for example, via the radio base station 30 and the core network 60.

[0091] The receiving unit 202 receives the learning model, for example, from the management device 100. For example, when the distributed infrastructure 200 is located in a first area, the receiving unit 202 receives a first learning model from the management device 100. For example, when the distributed infrastructure 200 is located in a second area, the receiving unit 202 receives a second learning model from the management device 100. The receiving unit 202 may store the received learning model in the model storage unit 204.

[0092] The receiving unit 202 receives learning data used for machine learning of the learning model, for example, from a plurality of communication terminals 50. The receiving unit 202 receives learning data, for example, from a plurality of communication terminals 50 in the RAN 300. The receiving unit 202 may store the received plurality of learning data in the learning data storage unit 206.

[0093] The execution unit 208 executes various processes. The execution unit 208 includes, for example, a distributed learning function 209 that performs machine learning of a learning model. The execution unit 208 includes, for example, a RAN control function 215 that controls the functions of the RAN 300. The distributed learning function 209 and the RAN control function 215 may use the same computational resources.

[0094] The distributed learning function 209 performs machine learning of a learning model, for example, by executing an AI application. The distributed learning function 209 performs machine learning of a learning model, for example, by executing a non-RAN AI application.

[0095] The distributed learning function 209 performs machine learning of a learning model using, for example, a plurality of pieces of learning data stored in the learning data storage unit 206. For example, the distributed learning function 209 performs machine learning of a learning model by fine-tuning the learning model using a plurality of pieces of learning data.

[0096] The transmission unit 212 transmits the learning result, obtained by the distributed learning function 209 using a plurality of pieces of learning data to the management device 100. The transmission unit 212 transmits the learning result to the management device 100, for example, via the Internet 40. The transmission unit 212 transmits the learning result to the management device 100, for example, via the core network 60. The transmission unit 212 transmits the learning result to the management device 100, for example, via the Internet 40 and the core network 60.

[0097] The learning data storage unit 206 stores multiple pieces of learning data, for example, by storing learning data for each communication terminal 50 of the multiple communication terminals 50. The distributed learning function 209 may perform machine learning on a learning model using multiple pieces of learning data by performing machine learning on a learning model using the learning data for each communication terminal 50.

[0098] The classification unit 214 groups the plurality of communication terminals 50. For example, the classification unit 214 groups the plurality of communication terminals 50 into groups that can share learning data.

[0099] For example, if users of the communication terminals 50 belong to the same company, the classification unit 214 groups the communication terminals 50 of the users into the same group. For example, if users of the communication terminals 50 belong to the same group of companies, the classification unit 214 groups the communication terminals 50 of the users into the same group. Group companies include, for example, a parent company and a subsidiary company of the parent company. For example, if users of the communication terminals 50 belong to the same local government, the classification unit 214 groups the communication terminals 50 of the users into the same group. For example, if users of the communication terminals 50 belong to the same school, the classification unit 214 groups the communication terminals 50 of the users into the same group. For example, if users of the communication terminals 50 belong to the same non-profit organization, the classification unit 214 groups the communication terminals 50 of the users into the same group. For example, if users of the communication terminals 50 belong to the same family, the classification unit 214 groups the communication terminals 50 of the users into the same group. If the user of the communication terminal 50 belongs to any other organization that can share personal information or confidential information, the classification unit 214 may group the communication terminals 50 of the user into the same group. Information about the organization to which the user of the communication terminal 50 belongs may be registered in advance in the distributed infrastructure 200.

[0100] The learning data storage unit 206 stores multiple pieces of learning data, for example, by storing learning data for at least one communication terminal 50 out of the multiple communication terminals 50 that has been grouped into the same group by the classification unit 214. The distributed learning function 209 may machine-learn a learning model using multiple pieces of learning data by using the learning data for each of the at least one communication terminal 50.

[0101] The RAN control function 215 controls the functions of the RAN 300, for example, by executing an AI application. The RAN control function 215 controls the functions of the RAN 300, for example, by executing a RAN_AI application. The RAN control function 215 may also control the functions of the RAN 300 by executing any other processing.

[0102] The RAN control function 215 controls the functions of the RAN 300, for example, based on the results of predicting demand for the RAN 300. For example, if the usage of the RAN 300 is on an increasing trend, the RAN control function 215 predicts that the demand for the RAN 300 will increase, and if the usage of the RAN 300 is on a decreasing trend, the RAN control function 215 predicts that the demand for the RAN 300 will decrease.

[0103] The distributed learning function 209 determines the timing for machine learning a learning model using multiple pieces of learning data stored in the learning data storage unit 206, for example, based on the prediction result obtained by the RAN control function 215 predicting demand for the RAN 300. For example, when the prediction result indicates that demand for the RAN 300 will decrease, the distributed learning function 209 machine-learns a learning model using multiple learning results. On the other hand, when the prediction result indicates that demand for the RAN 300 will increase, the distributed learning function 209 does not machine-learn a learning model using multiple learning results.

[0104] 13 schematically illustrates an example of the hardware configuration of a computer 1200 that functions as the management device 100 or the distributed infrastructure 200. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of the device according to the above embodiment, or can cause the computer 1200 to perform operations associated with the device according to the above embodiment or one or more "parts," and / or can cause the computer 1200 to perform a process or steps of the process according to the above embodiment. Such a program can be executed by the CPU 1212 to cause the computer 1200 to perform specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0105] The computer 1200 according to this embodiment includes a CPU 1212, a RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive 1226, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive 1226 may be a DVD-ROM drive, a DVD-RAM drive, or the like. The storage device 1224 may be a hard disk drive, a solid-state drive, or the like. The computer 1200 also includes legacy input / output units such as a ROM 1230 and a keyboard 1242, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0106] The CPU 1212 operates according to programs stored in the ROM 1230 and the RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires image data generated by the CPU 1212 into a frame buffer or the like provided in the RAM 1214 or into the graphics controller 1216 itself, and causes the image data to be displayed on the display device 1218.

[0107] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD drive 1226 reads programs or data from a DVD-ROM 1227 or the like and provides them to the storage device 1224. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0108] The ROM 1230 stores therein a boot program or the like that is executed by the computer 1200 upon activation, and / or programs that depend on the hardware of the computer 1200. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via a USB port, a parallel port, a serial port, a keyboard port, a mouse port, etc.

[0109] The programs are provided by a computer-readable storage medium such as a DVD-ROM 1227 or an IC card. The programs are read from the computer-readable storage medium, installed in the storage device 1224, RAM 1214, or ROM 1230, which are also examples of computer-readable storage media, and executed by the CPU 1212. Information processing described in these programs is read by the computer 1200, and brings about cooperation between the programs and the various types of hardware resources described above. An apparatus or a method may be configured by implementing operations or processing of information in accordance with the use of the computer 1200.

[0110] For example, when communication is performed between computer 1200 and an external device, CPU 1212 may execute a communication program loaded into RAM 1214 and instruct communication interface 1222 to perform communication processing based on the processing described in the communication program. Under the control of CPU 1212, communication interface 1222 reads transmission data stored in a transmission buffer area provided in RAM 1214, storage device 1224, DVD-ROM 1227, or a recording medium such as an IC card, and transmits the read transmission data to a network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0111] Furthermore, the CPU 1212 may cause all or a necessary portion of a file or database stored in an external recording medium such as the storage device 1224, the DVD drive 1226 (DVD-ROM 1227), an IC card, etc. to be read into the RAM 1214, and may perform various types of processing on the data on the RAM 1214. The CPU 1212 may then write back the processed data to the external recording medium.

[0112] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. The CPU 1212 may perform various types of processing on data read from the RAM 1214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequences of the programs, and write the results back to the RAM 1214. The CPU 1212 may also search for information in a file, database, etc. on the recording medium. For example, if multiple entries, each having an attribute value of a first attribute associated with an attribute value of a second attribute, are stored on the recording medium, the CPU 1212 may search for an entry whose attribute value of the first attribute matches a specified condition from among the multiple entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0113] The above-described programs or software modules may be stored in a computer-readable storage medium on or near the computer 1200. A recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can also be used as a computer-readable storage medium, thereby providing the programs to the computer 1200 via the network.

[0114] The blocks in the flowcharts and block diagrams in the present embodiments may represent stages of a process in which an operation is performed or "parts" of a device responsible for performing the operation. Particular stages and "parts" may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable storage medium, and / or a processor provided with computer-readable instructions stored on a computer-readable storage medium. The dedicated circuitry may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. The programmable circuitry may include reconfigurable hardware circuits, such as field programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), including AND, OR, XOR, NAND, NOR, and other logical operations, flip-flops, registers, and memory elements.

[0115] A computer-readable storage medium may include any tangible device capable of storing instructions that are executed by an appropriate device, such that a computer-readable storage medium having instructions stored thereon comprises an article of manufacture, including instructions that can be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of computer-readable storage media may include electronic, magnetic, optical, electromagnetic, and semiconductor storage media. More specific examples of computer-readable storage media may include floppy disks, diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), electrically erasable programmable read-only memories (EEPROMs), static random access memories (SRAMs), compact disc read-only memories (CD-ROMs), digital versatile discs (DVDs), Blu-ray discs, memory sticks, integrated circuit cards, and the like.

[0116] The computer readable instructions may include either assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and conventional procedural programming languages ​​such as the “C” programming language or similar programming languages.

[0117] Computer-readable instructions may be provided locally or over a local area network (LAN), a wide area network (WAN) such as the Internet, to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, or programmable circuitry, such that the processor or programmable circuitry executes the computer-readable instructions to generate means for performing the operations specified in the flowcharts or block diagrams. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0118] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc. A computer may have one processor or multiple processors. In a multiprocessor system with multiple processors, each processor executes a portion of a program and passes data between processors as needed during program execution, allowing the multiple processors to collectively execute the program. For example, in multitasking, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at each time slice. In this case, which portion of a program each processor executes changes dynamically. Which portion of a program each of the multiple processors executes may also be statically determined by multiprocessor-aware programming.

[0119] Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention.

[0120] It should be noted that the execution order of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and that the processes can be performed in any order unless the output of a previous process is used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is described using "first," "next," etc. for convenience, this does not mean that the processes must be performed in this order.

[0121] 10 System, 30 Radio base station, 40 Internet, 50 Communication terminal, 60 Core network, 100 Management device, 102 Model storage unit, 104 Distribution unit, 106 Receiving unit, 108 Learning result storage unit, 110 Execution unit, 111 Central learning function, 113 RAN control function, 200 Distribution base, 202 Receiving unit, 204 Model storage unit, 206 Learning data storage unit, 208 Execution unit, 209 Distributed learning function, 212 Transmission unit, 214 Classification unit, 215 RAN control function, 300 RAN, 1200 Computer, 1210 Host controller, 1212 CPU, 1214 RAM, 1216 Graphics controller, 1218 Display device, 1220 Input / output controller, 1222 Communication interface, 1224 Storage device, 1226 DVD drive, 1227 DVD-ROM, 1230 ROM, 1240 I / O chip

Claims

1. A learning system for machine learning a learning model, comprising: a plurality of distributed platforms arranged on a core network; and a management device that manages the plurality of distributed platforms, wherein each distributed platform of the plurality of distributed platforms has: a first receiving unit that receives learning data used for machine learning of the learning model from a plurality of communication terminals; a learning data storage unit that stores the plurality of learning data; a first execution unit that includes a distributed learning function that machine-learns the learning model using the plurality of learning data stored in the learning data storage unit; and a transmitting unit that transmits learning results obtained by machine learning the learning model using the plurality of learning data to the management device, wherein the management device has: a distribution unit that distributes the learning model to each of the distributed platforms; a second receiving unit that receives the learning results from each of the distributed platforms; a learning result storage unit that stores the plurality of learning results; and a second execution unit that has a central learning function that machine-learns the learning model using the plurality of learning results stored in the learning result storage unit.

2. The learning system described in claim 1, wherein the distribution unit distributes the learned learning model, which has been machine-learned using the multiple learning results, to each of the distributed platforms.

3. The learning system described in claim 1 or 2, wherein the learning data storage unit of each of the distributed platforms stores the multiple learning data by storing the learning data for each communication terminal of the multiple communication terminals, and the distributed learning function of each of the distributed platforms machine-learns the learning model using the multiple learning data by machine-learning the learning model using the learning data for each communication terminal.

4. The learning system described in claim 1 or 2, wherein each of the distributed platforms further has a classification unit that groups the multiple communication terminals into groups that can share the learning data, the learning data storage unit of each of the distributed platforms stores the multiple learning data by storing the learning data for at least one communication terminal among the multiple communication terminals that has been grouped into the same group by the classification unit, and the distributed learning function of each of the distributed platforms machine-learns the learning model using the multiple learning data by machine-learning the learning model using the learning data for each of the at least one communication terminal.

5. The learning system described in claim 1 or 2, wherein the learning system machine-learns a plurality of the learning models, the management device and at least two distributed platforms located in a first area of ​​the plurality of distributed platforms machine-learn a first learning model of the plurality of learning models, and the management device and at least two distributed platforms located in a second area different from the first area of ​​the plurality of distributed platforms machine-learn a second learning model of the plurality of learning models.

6. The learning system according to claim 1 or 2, wherein the learning system learns a plurality of the learning models by machine learning, the learning system comprises a plurality of the management devices, a management device among the plurality of management devices that is located in a first area and at least two distributed platforms among the plurality of distributed platforms that are located in the first area learns a first learning model among the plurality of learning models by machine learning, and a management device among the plurality of management devices that is located in a second area different from the first area and at least two distributed platforms among the plurality of distributed platforms that are located in the second area learns a second learning model among the plurality of learning models by machine learning.

7. The learning system according to claim 1 or 2, wherein the first execution unit of each of the distribution platforms further includes a RAN control function for controlling the function of a RAN (Radio Access Network).

8. The learning system described in claim 7, wherein the distributed learning function of each of the distributed platforms determines the timing of machine learning the learning model using the multiple learning data stored in the learning data storage unit based on the prediction results of the RAN control function predicting demand for the RAN.

9. The learning system according to claim 1 or 2, wherein the management device is located on the Internet.

10. A learning system as described in claim 1 or 2, wherein the management device is installed on a management platform that is located on the core network and manages the multiple distributed platforms.

11. The learning system according to claim 1 or 2, wherein the learning model is an LLM (Large Language Models).

12. A learning method for machine learning a learning model, comprising: a distribution step in which a management device distributes the learning model to each of a plurality of distributed infrastructures arranged on a core network; a first reception step in which each of the distributed infrastructures receives, from a plurality of communication terminals, learning data to be used for machine learning of the learning model distributed in the distribution step; a first storage step in which each of the distributed infrastructures stores the plurality of learning data received in the first reception step in a learning data storage unit included in each of the distributed infrastructures; a distributed learning step in which each of the distributed infrastructures machine learns the learning model using the plurality of learning data stored in the learning data storage unit in the first storage step; a transmission step in which each of the distributed infrastructures transmits to the management device learning results obtained by machine learning the learning model using the plurality of learning data in the distributed learning step; a second reception step in which the management device receives the learning results from each of the distributed infrastructures; and a second storage step in which the management device stores the plurality of learning results received in the second reception step in a learning result storage unit included in the management device. a central learning stage in which the management device machine-learns the learning model using the plurality of learning results stored in the learning result storage unit in the second storage stage.

Images