Federated learning system and federated learning method

Abstract

The purpose of the present invention is to suppress a communication amount between a base station and a higher-level device in RIC processing in communication between a base station and a terminal. A federated learning system comprising a plurality of base stations, and a higher-level device capable of communicating with each of the plurality of base stations. The base stations include: a learning unit that learns communication information related to communication with a terminal and generates a local learning model related to communication by a real-time RIC (Radio access network Intelligent Controller); and a transmission unit that transmits information on the local learning model to the higher-level device. The higher-level device includes: a reception unit that receives information on the local learning model; a generation unit that generates an integrated learning model by the non-real-time RIC on the basis of a local learning model of each of the plurality of base stations; and a feedback unit that feeds back information on the integrated learning model generated by the generation unit to the base station.

Description

Federated learning system and federated learning method

[0001] The present invention relates to a federated learning system that utilizes learning by base stations, and a federated learning method therefor.

[0002] In recent years, the use of a Radio Access Network Intelligent Controller (RIC) has been considered for communication between base stations and terminals. Conventional RIC learning generally involves collecting a large amount of information from base stations and performing non-real-time learning at a central device. Patent Document 1 discloses an invention that uses base stations for learning and selects participants to participate in federated learning.

[0003] Special Publication No. 2023-550806

[0004] However, when learning is performed by a central device and control is performed by learning in real time, the base station sends a large amount of data used for learning to the central device, which increases the amount of traffic between the base station and the central device and causes communication congestion, resulting in a problem of a shortage of resources necessary for learning in the central device performing the learning.

[0005] Therefore, the present invention has been made in consideration of the above problems, and has as its object to provide a federated learning system and a federated learning method that can perform learning without congesting communication between a base station and a central device.

[0006] In order to solve the above problem, one embodiment of the present invention provides a federated learning system comprising a plurality of base stations and an upper level device capable of communicating with each of the plurality of base stations, wherein the base station comprises a learning unit that learns communication information regarding communication between a terminal and the base station and generates a local learning model related to the communication using real-time RIC (Radio access network Intelligent Controller), and a transmitting unit that transmits information regarding the local learning model to the upper level device, and the upper level device comprises a receiving unit that receives information regarding the local learning model, a generating unit that generates an integrated learning model using non-real-time RIC based on the local learning model of each of the plurality of base stations, and a feedback unit that feeds back information regarding the integrated learning model generated by the generating unit to the base station.

[0007] In addition, in the above-mentioned federated learning system, the information regarding the local learning model is parameter information of a portion of the local learning model that has been specified in advance, and the generation of the integrated learning model is performed by averaging each of the parameter information transmitted from each of the multiple base stations, the feedback unit feeds back the averaged parameter information to the base station, and the learning unit may replace parameters corresponding to the parameter information of the local learning model with the averaged parameters.

[0008] In the above-described federated learning system, the generation unit may reduce the frequency of generating the integrated learning model using the non-real-time RIC as time passes.

[0009] In addition, in order to solve the above problem, a federated learning method according to one embodiment of the present invention is a federated learning method in a federated learning system consisting of a plurality of base stations and an upper device capable of communicating with each of the plurality of base stations, and includes the following steps: a learning step in which the base station learns communication information regarding communication with a terminal and generates a local learning model related to the communication using real-time RIC (Radio access network Intelligent Controller); a transmission step in which the base station transmits information regarding the local learning model to the upper device; a reception step in which the upper device receives information regarding the local learning model; a generation step in which the upper device generates an integrated learning model using non-real-time RIC based on the local learning models of each of the plurality of base stations; and a feedback step in which information regarding the integrated learning model generated in the generation step is fed back to the base station.

[0010] In the federated learning system of the present invention, each base station learns in real time based on information obtained through communication with terminals, and only information regarding the local learning model obtained through learning from the base station is transmitted to the upper device, thereby eliminating the need to transmit information from many terminals, thereby preventing traffic congestion between the base station and the upper device.

[0011] 1 is a system diagram showing an example of the configuration of a communication system; FIG. 2 is a block diagram showing an example of the configuration of a base station; FIG. 3 is a block diagram showing an example of the configuration of a service management device; FIG. 4 is a flowchart showing an example of the operation of a base station; and FIG. 5 is a flowchart showing an example of the operation of a service management device.

[0012] A base station according to the present invention will be described below with reference to the drawings.

[0013] <Overview> Fig. 1 is a schematic diagram showing an overview of a federated learning system 1. The federated learning system 1 may be a communication system in which base stations 100 (200a, 200b) relay communications between terminals 300 (300a to 300d), and the base stations 100 may be managed by a service management device 200, which is a higher-level device. The federated learning system 1 may be a telephone communication network formed by a virtualized wireless communication network. The service management device 200 may be a CU (Central Unit) in 5g communication, and the base stations 100 may be a DU (Distributed Unit) or RU (Radio Unit) in 5g communication.

[0014] In this embodiment, each base station 100 performs predetermined learning in real time (some delay is allowed) based on a received signal received from a terminal 300 with which the base station is communicating. The learning content in the predetermined learning may be determined by an operator of a carrier in the federated learning system 1, or the like, and may be arbitrary. As an example, the predetermined learning may be learning the relationship between a signal received from a terminal 300 (for example, a Sounding Reference Signal (SRS), but is not limited to this) and channel information for the terminal 300, but is not limited to this. When the learning in this example is performed, if there is a received signal from the terminal 300, it becomes possible to estimate the channel information to be applied to the terminal 300.

[0015] It is desirable to generate a learning model that is as versatile as possible based on as much information as possible for the learning model generated by such learning, and normally, it is considered that each base station 100 collects information on the terminals 300, uploads and accumulates it to the service management device 200, and the service management device 200 performs learning. However, with this method, a huge amount of data is sent from the base station 100 to the service management device 200, which may increase the traffic volume in the service management device 200 and cause communication congestion.

[0016] Therefore, in the federated learning system 1 according to this embodiment, each base station 100 performs predetermined learning locally using only information acquired through communication with the terminals 300 connected to the base station, and generates a local learning model for each base station 100. In other words, in the example of FIG. 1 , base station 100a performs learning based on data acquired through communication with terminals 300a and 300b connected to base station 100a, and base station 100b performs learning based on data acquired through communication with terminals 300c and 300d connected to base station 100b. Each base station 100 then transmits information about the local learning model it has generated to the service management device 200. This eliminates the need for the base station 100 to transmit a large amount of information about the terminals 300 to the service management device 200, thereby preventing communication congestion between the base station 100 and the service management device 200.

[0017] Then, the service management device 200 generates a global learning model based on the information about the local learning model received from each base station, and feeds it back to each base station 100. This prevents a base station 100 with a small number of connected terminals 300 from having an insufficient amount of information for learning, and as a result, prevents the generated local learning model from becoming a model specialized for specific information.

[0018] The terminal 300 is a portable terminal held by a user, and may be, for example, a smartphone, a tablet terminal, a portable communication module of a mobile phone, or the like, but is not limited to these.

[0019] The associative learning system 1 according to this embodiment will be described in detail below.

[0020] <Configuration> <Configuration of Base Station 100> FIG. 2 is a block diagram showing an example configuration of the base station 100. The base station 100 functions as a base station that relays communications between terrestrial terminals 300 and is a computer system that operates according to a predetermined program. The base station 100 may be configured to construct a virtual radio access network (vRAN) implemented by a CPU and a GPU. In the base station 100, the communication unit 110 described below may be an RU (Radio Unit) in 5G communication. The function implemented by the control unit 130 may be implemented by a DU (Distribution Unit) or a CU (Central Unit) in 5G communication, or may be implemented by an RU. The base station 100 may be configured to implement a radio access network intelligent controller (RIC) or may operate under the control of an RIC.

[0021] 2, the base station 100 includes a communication unit 110, a control unit 130, and a storage unit 140. The base station 100 may also include an input unit 120 and an output unit 150.

[0022] The communication unit 110 is a communication interface having a function of communicating with devices external to the base station 100. The communication unit 110 has a function of communicating with the terminal 300 and the service management device 200 as external devices. The communication unit 110 receives a signal from the terminal 300 and transmits the signal to the control unit 130, and transmits a signal to the terminal 300 according to instructions from the control unit 130. The communication unit 110 also transmits information related to the local learning model to the service management device 200 according to instructions from the control unit 130, and receives information related to the global learning model from the service management device 200 and transmits the information to the control unit 130.

[0023] The input unit 120 has a function of accepting input from an operator of the base station 100 or the like and transmitting the input content to the control unit 130. The input unit 120 may be realized by an input device such as a mouse, keyboard, or touch panel, or in the case of voice input, by a microphone. The input unit 120 may, for example, accept designation of parameters from the local learning model 141 that should be transmitted to the service management device 200 as information related to the local learning model 141, and transmit the designation to the control unit 130.

[0024] The control unit 130 is a processor having the function of controlling each unit of the base station 100. The control unit 130 may be realized by a single core or a multi-core. The control unit 130 executes various programs stored in the storage unit 140 and uses various data to realize the functions of the base station 100.

[0025] The control unit 130 relays communications between the terminals 300 in the same manner as a normal base station 100 .

[0026] The control unit 130 includes a local learning unit 131, a transmission unit 132, and an AI processing unit 133 as functions realized by the control unit 130.

[0027] The local learning unit 131 performs learning based on information obtained by the communication unit 110 communicating with each terminal 300. The local learning unit 131 may perform real-time RIC. The learning content of the local learning unit 131 may be set in advance by an operator managing the federated learning system 1, and the local learning unit 131 may perform learning to improve communication performance between the base station 100 and the terminal 300. As an example, the local learning unit 131 may learn the relationship between a signal received from the terminal 300 (e.g., a Sounding Reference Signal (SRS)) and channel information between the terminal 300 at that time. That is, the local learning unit 131 generates training data using the signal received from the terminal 300 as an explanatory variable and the channel information between the terminal 300 at that time as a target variable, performs learning, and generates a local learning model 141. The local learning unit 131 stores the generated local learning model 141 in the storage unit 140.

[0028] Furthermore, the local learning unit 131 receives information about the global learning model from the service management device 200, which is a higher-level device, and reflects the information in the local learning model 141. The information about the global learning model may be information about some of the parameters of the local learning model 141, and the local learning unit 131 replaces the parameters in the corresponding locations of the local learning model 141 with the parameters transmitted from the service management device 200. This allows the local learning model 141 to be modified into a generalized model.

[0029] Local learning unit 131 may communicate with terminal 300, receive signals from terminal 300, and perform learning in real time, but is not limited to real time. By performing learning in real time, local learning unit 131 can perform reinforcement learning of local learning model 141, and with each learning process, it can grow into a model that is optimal for the location where base station 100 is located.

[0030] The transmitting unit 132 transmits information about the local learning model 141 generated by the local learning unit 131 to the service management device 200 via the communication unit 110. Here, the information about the local learning model 141 may be some of the parameters of the local learning model 141. The parameters may be pre-specified parameters. By transmitting only some of the parameters of the local learning model 141, it is possible to prevent communication congestion between the base station 100 and the service management device 200. The information about the local learning model 141 may be information that associates information indicating which parameters constitute the local learning model 141 are associated with their values, and the number of parameters may be any as long as one or more parameters are included.

[0031] The AI ​​processing unit 133 executes AI processing using the local learning model 141 stored in the storage unit 140. The content of the AI ​​processing differs depending on the learned content. As described above, if the local learning model 141 is a model that has learned the relationship between the signal received from the terminal 300 and the channel information between the local station and the terminal 300, the AI ​​processing unit 133 can perform processing to estimate the channel with the terminal 300 based on the transmitted signal received from the terminal 300. If the channel with the terminal 300 can be estimated, it is possible to uniquely identify a precoding matrix for the terminal 300 and perform beamforming, and as a result, it is possible to improve communication performance with the terminal 300.

[0032] The storage unit 140 has a function of storing various programs and data required for the operation of the base station 100. The storage unit 140 can be realized, for example, by a hard disk drive (HDD), a solid state drive (SSD), a flash memory, etc., but is not limited to these. The storage unit 140 may store various programs and various data for realizing the functions to be performed by the base station 100. The storage unit 140 may store a program for learning based on information obtained by communication with the terminal 300, etc. The storage unit 140 may store the local learning model 141 generated by the local learning unit 131.

[0033] The output unit 150 has a function of outputting designated information in accordance with an instruction from the control unit 130. The output unit 150 may, for example, output text information or image information, in which case the output unit 150 is realized by a monitor provided in or connected to the base station 100. The output unit 150 may, for example, output audio information, in which case the output unit 150 is realized by a speaker provided in or connected to the information processing device.

[0034] The above is an example of the configuration of the base station 100.

[0035] 3 is a block diagram showing an example of the configuration of the service management device 200. The service management device 200 is an information processing device (computer system) for realizing globalization of the local learning model 141 generated by each service management device 200, and may also have a function as a base station.

[0036] 3, the service management device 200 includes a communication unit 210, a control unit 230, and a storage unit 240. The service management device 200 may also include an input unit 220 and an output unit 250.

[0037] The communication unit 210 is a communication interface having a function of communicating with devices external to the service management device 200. The communication unit 210 has a function of communicating with the terminal 300 and the service management device 200 as external devices. The communication unit 210 receives a signal from the terminal 300 and transmits the signal to the control unit 230, and transmits a signal to the terminal 300 according to instructions from the control unit 230. The communication unit 210 also transmits information related to the local learning model to the service management device 200 according to instructions from the control unit 230, and receives information related to the global learning model from the service management device 200 and transmits the information to the control unit 230.

[0038] The input unit 220 has a function of receiving input from an operator of the service management device 200 or the like and transmitting the input content to the control unit 230. The input unit 220 may be realized by an input device such as a mouse, keyboard, or touch panel, or in the case of voice input, by a microphone.

[0039] The control unit 230 is a processor that has the function of controlling each unit of the service management device 200. The control unit 230 may be realized by a single core or a multi-core. The control unit 230 realizes the functions of the service management device 200 by executing various programs stored in the storage unit 240 and using various data.

[0040] The control unit 230 relays communications between the terminals 300 in the same manner as in a normal service management device 200 .

[0041] The control unit 230 has a global learning unit 231, a transmission unit 132, and an AI processing unit 133 as functions realized by the control unit 230.

[0042] The global learning unit 231 generates information 241 about the global learning model based on information about the local learning model 141 of the base station 100 transmitted from the communication unit 210. The global learning unit 231 may perform non-real-time RIC. The information 241 about the global learning model may be information obtained by averaging each parameter as information about the local learning model 141 transmitted from each base station 100. The global learning unit 231 transmits the generated information 241 about the global learning model to the feedback unit 232.

[0043] The feedback unit 232 transmits (feeds back) the transmitted information 241 about the global learning model to each base station 100 via the communication unit 210. The information 241 about the global learning model may be information that associates information indicating which parameter among the parameters that make up the local learning model 141 the information corresponds to with its value (averaged value), and the number of parameters may be arbitrary as long as it includes one or more parameters.

[0044] The storage unit 240 has a function of storing various programs and data required for the operation of the service management device 200. The storage unit 240 can be realized, for example, by a hard disk drive (HDD), a solid state drive (SSD), a flash memory, etc., but is not limited to these. The storage unit 240 may store various programs and various data for realizing the functions to be performed by the service management device 200. The storage unit 240 may store a program for generating information 241 related to the global learning model, etc. The storage unit 240 may also store information 241 related to the local learning model 141 received from each base station 100 and information 242 related to the global learning model.

[0045] The output unit 250 has a function of outputting designated information in accordance with instructions from the control unit 230. The output unit 250 may, for example, output text information or image information, in which case the output unit 250 is realized by a monitor provided in or connected to the service management device 200. The output unit 250 may, for example, output audio information, in which case the output unit 250 is realized by a speaker provided in or connected to the information processing device.

[0046] The above is an example of the configuration of the service management device 200.

[0047] The terminal 300 is similar to a typical information processing device having a configuration similar to that of a communication terminal such as a smartphone or tablet terminal, and therefore a detailed description using a block diagram will be omitted.

[0048] <Operation> Next, the operation of the base station 100 and the service management device 200 in the federated learning system 1 will be described.

[0049] FIG. 4 is a flowchart showing an example of an operation related to learning of the base station 100.

[0050] 4, the communication unit 110 of the base station 100 receives a signal from the terminal 300 (step S401). The communication unit 110 transfers the received signal to the control unit .

[0051] The local learning unit 131 performs learning based on the transmitted signal (step S402). The local learning unit 131 generates a local learning model 141 (step S403) and stores it in the storage unit 140.

[0052] The transmitting unit 132 transmits information about the local learning model 141 to the service management device 200 via the communication unit 110 (step S404). That is, the transmitting unit 132 extracts pre-specified parameters that constitute the local learning model 141 and transmits them to the service management device 200. As a result, information 241 about the global learning model is generated in the service management device 200.

[0053] The communication unit 110 receives information 241 related to the global learning model from the service management device 200 (step S405). The communication unit 110 transmits the received information 241 related to the global learning model to the control unit 130. The information 241 related to the global learning model corresponds to the information related to the local learning model 141 transmitted by the own station, and is information averaged by the service management device 200 across multiple base stations 100.

[0054] The local learning unit 131 reflects the parameters indicated by the transmitted information 241 regarding the global learning model in the local learning model 141 (step S406). That is, the local learning unit 131 replaces (overwrites) the parameters in the corresponding locations of the local learning model 141 with the transmitted parameters. This allows the local learning model 141 to be generalized even when the number of terminals 300 with which the base station 100 communicates is minimal, thereby improving the accuracy of AI processing.

[0055] The AI ​​processing unit 133 receives as input information about the terminal 300 transmitted from the communication unit 110 and executes AI processing using the updated local learning model 141 (step S407). As described above, if the local learning model 141 is a model that has learned the relationship between the signal received from the terminal 300 and the channel information for the terminal 300 at that time, the AI ​​processing unit 133 can execute processing to estimate channel information between the terminal 300 as AI processing. The estimated channel information may be used in subsequent communications with the terminal 300, and as an example, may be used to generate a precoding matrix for beamforming for the terminal 300.

[0056] The above is an example of the operation of the base station 100 according to the embodiment.

[0057] FIG. 5 is a flowchart showing an example of an operation related to learning by the service management device 200.

[0058] 5 , the communication unit 210 of the service management device 200 receives, from each base station 100, information about the local learning model 141 generated therein (step S501). The information about the local learning model 141 may be information about some of the parameters that constitute the local learning model 141. The communication unit 210 transmits the received information about the local learning model 141 to the control unit 230.

[0059] The global learning unit 231 of the control unit 230 averages the corresponding parameters of the information about the local learning model 141 transmitted from the plurality of base stations 100 to generate information about the global learning model 241 (step S502).

[0060] The feedback unit 232 feeds back the information 241 about the global learning model generated by the global learning unit 231 to each base station 100 via the communication unit 210 (step S503), and ends the process.

[0061] The above is an example of the operation of the service management device 200 according to the embodiment.

[0062] <Summary> As described above, the base stations 100 perform learning in real time using signals received from terminals connected to each base station 100 to generate local learning models. This eliminates the need for the base stations 100 to transmit a large amount of information used for learning to the service management device 200, thereby reducing the possibility of communication congestion between the base stations 100 and the service management device 200. Furthermore, if each base station 100 were to learn using only information that only it holds, it would not be able to obtain a sufficient amount of data, which could result in a long learning time, or the local learning model 141 would be generated without sufficient learning, which could result in a decrease in estimation accuracy during operation. However, by feeding back information 241 related to the global learning model from the service management device 200 and reflecting it in each local learning model 141, it is possible to prevent a decrease in the accuracy of the estimation results obtained by the local learning model 141. Furthermore, in each base station 100, by continuing to learn in real time while being generalized using information 241 regarding the global learning model, it is possible to generate a local learning model 141 that is suited to the location where the base station 100 is installed, and since it is possible to perform AI processing using the local learning model 141 that is suited to the local environment, it is possible to improve the possibility of performing processing that is optimal for the environment in which the base station 100 is located.

[0063] <Modifications> The service management device 200 and the base station 100 according to the above embodiment are not limited to the above embodiment, and may be realized by other methods. Various modifications will be described below.

[0064] (1) In the above embodiment, the base station 100 transmits information on pre-specified parameters as information on the local learning model 141, and the service management device 200 transmits the average value of the parameters to each base station 100 as information on the global learning model, to be reflected in the local learning model 141. However, the base station 100 may be configured to transmit the generated local learning model 141 itself to the service management device 200. The service management device 200 may then average the entire local learning model 141 to generate information 242 on the global learning model. In this case, the information 242 on the global learning model becomes the global learning model itself. The feedback unit 232 may then feed back the generated global learning model itself to each base station 100 as information 242 on the global learning model.

[0065] (2) In the above embodiment, the local learning unit 131 may be realized by an external information processing device that can communicate with the base station 100 and receive information obtained through communication with the terminal 300 .

[0066] (3) In the above embodiment, the global learning unit 231 of the service management device 200 performs learning at a fixed timing in a non-real-time manner. However, the timing of learning by the global learning unit 231 may be changed as appropriate. For example, when learning by the federated learning system 1 begins, it is considered that sufficient information for learning may not be accumulated. Therefore, the global learning unit 231 may perform learning at a high frequency. On the other hand, after a certain amount of time has passed since the start of system operation, it is likely that each base station 100 will be able to perform communication with a certain degree of stability. In such a case, it is expected that no significant changes will occur even if information related to the global learning model is not reflected in each base station 100. Therefore, the global learning unit 231 may perform learning less frequently. In other words, the global learning unit 231 may perform a process of relatively decreasing the frequency of performing learning (the frequency of performing a process to average specific parameters of the information group related to the local learning model 141) over time. As a result, the processing load on the global learning unit 231 of the service management device 200 and the local learning unit 131 of the base station 100 can be reduced over time.

[0067] (4) In the above embodiment, in the operation of base station 100 described with reference to FIG. 4, the processes of steps S401 to S406 and the process of step S407 may be processes executed in parallel with each other, and the process of step S407 does not necessarily have to be executed after the process of step S406.

[0068] (5) In the above embodiment, the local learning model 141 was described as a model that learned the relationship between information obtained through communication with the terminal 300 and the channel between the terminal 300 at that time. As described above, the local learning model 141 learns different things and processes differently depending on the content of the AI ​​processing to be performed. As another example, the local learning model 141 may be a model that learned the correspondence between a signal from the terminal 300 and the communication settings (communication frequency, communication timing, etc.) applied to the terminal 300 at that time. In this case, the AI ​​processing can use the signal from the terminal 300 as input and estimate communication settings suitable for the terminal 300.

[0069] Furthermore, the local learning model 141 does not need to learn teacher data using the signal of one terminal 300 as an explanatory variable, but may also learn teacher data using the signals of multiple terminals 300 as explanatory variables.

[0070] In this way, local learning model 141 may be any model as long as it is configured to realize the estimation process required for the application. In this case, the data that will be the explanatory variables used as training data and the data that will be the target variables may be data that base station 100 can acquire by itself.

[0071] (6) The program by which the base station 100 of the present disclosure creates the local learning model 141 based on information from the terminal 300 and the program by which the service management device 200 generates information about the global learning model may be provided in a state stored in a computer-readable storage medium. The storage medium can store the program in a "non-transitory tangible medium." The storage medium can include any suitable storage medium, such as an HDD or SSD, or an appropriate combination of two or more of these. The storage medium may be volatile, non-volatile, or a combination of volatile and non-volatile. Note that the storage medium is not limited to these examples and may be any device or medium capable of storing the program.

[0072] The base station 100 and the service management device 200 can realize the functions of the multiple functional units shown in each embodiment by, for example, reading a program stored in a storage medium and executing the read program. The program may also be provided to the base station 100 and the service management device 200 via any transmission medium (such as a communication network or broadcast waves). The base station 100 and the service management device 200 can realize the functions of the multiple functional units shown in each embodiment by executing a program downloaded via the Internet, for example. This program may be executed by the base station 100, the service management device 200, etc.

[0073] The program can be implemented using, for example, a scripting language such as ActionScript or JavaScript (registered trademark), an object-oriented programming language such as Objective-C or Java (registered trademark), or a markup language such as HTML5, but is not limited to these.

[0074] At least a part of the processing in the base station 100 and the service management device 200 may be realized by cloud computing consisting of one or more computers. Furthermore, each functional unit of the base station 100 may be realized by one or more circuits that realize the functions described in the above embodiments, and the functions of multiple functional units may be realized by one circuit.

[0075] (7) The various techniques and processes described in the above embodiments and variations may be combined as appropriate within the scope of achieving the objective of performing federated learning between the base station 100 and the service management device 200 to execute AI processing.

[0076] (8) According to each aspect of the present disclosure described above, it is possible to execute AI processing that enables fast and accurate communication between base stations and terminals, and ultimately between terminals, thereby contributing to the achievement of Goal 9 of the Sustainable Development Goals (SDGs), which is to “build resilient infrastructure, promote inclusive and sustainable industrialization, and promote inclusive and sustainable industrialization.”

[0077] 100, 100a, 100b Base station 110 Communication unit 120 Input unit 130 Control unit 131 Local learning unit 132 Transmission unit 133 AI processing unit 140 Storage unit 150 Output unit 200 Service management device 210 Communication unit 220 Input unit 230 Control unit 231 Global learning unit 232 Feedback unit 240 Storage unit 250 Output unit 300, 300a, 300b, 300c, 300d Terminal

Claims

1. A federated learning system comprising a plurality of base stations and an upper device capable of communicating with each of the plurality of base stations, wherein the base station comprises: a learning unit that learns communication information relating to communication with a terminal and generates a local learning model relating to the communication using real-time RIC (Radio access network Intelligent Controller); and a transmitting unit that transmits information relating to the local learning model to the upper device, and the upper device comprises: a receiving unit that receives information relating to the local learning model; a generating unit that generates an integrated learning model using non-real-time RIC based on the local learning model of each of the plurality of base stations; and a feedback unit that feeds back information relating to the integrated learning model generated by the generating unit to the base station.

2. The federated learning system described in claim 1, characterized in that the information regarding the local learning model is parameter information of a portion of the local learning model that has been specified in advance, the generation of the integrated learning model is performed by averaging each of the parameter information transmitted from each of the multiple base stations, the feedback unit feeds back the averaged parameter information to the base station, and the learning unit replaces parameters corresponding to the parameter information of the local learning model with the averaged parameters.

3. The federated learning system according to claim 2, wherein the generation unit reduces the frequency of generating the integrated learning model using the non-real-time RIC as time passes.

4. A federated learning method in a federated learning system consisting of a plurality of base stations and an upper device capable of communicating with each of the plurality of base stations, the federated learning method executing the following steps: a learning step in which the base station learns communication information regarding communication with a terminal and generates a local learning model related to the communication using real-time RIC (Radio access network Intelligent Controller); a transmission step in which the base station transmits information regarding the local learning model to the upper device; a reception step in which the upper device receives information regarding the local learning model; a generation step in which the upper device generates an integrated learning model using non-real-time RIC based on the local learning models of each of the plurality of base stations; and a feedback step in which information regarding the integrated learning model generated in the generation step is fed back to the base station.

Images