Information processing device and information processing system

Abstract

This information processing system includes a first information processing device and a plurality of second information processing devices, wherein the plurality of second information processing devices each transmit an associative learning profile including the degree to which the second information processing devices contribute to associative learning, the first information processing device receives the associative learning profiles of the plurality of second information processing devices, and the first information processing device determines an information processing device participating in associative learning on the basis of the associative learning profiles.

Description

Information processing device and information processing system

[0001] The present disclosure relates to an information processing device and an information processing system

[0002] As machine learning technology is gaining attention, research and development is being conducted on federated learning, which trains models without consolidating distributed learning data in one place, from the viewpoint of protecting data privacy (Non-Patent Document 1). Furthermore, Non-Patent Document 2 discloses federated learning using multiple NWDAFs (Network Data Analysis Functions) in a 5G core system, and in it, a procedure for discovering NWDAFs that will participate in the federated learning process.

[0003] Japanese Patent Application Laid-Open No. 2023-141510

[0004] Feng, Siwei, and Han Yu. "Multi-participant multi-class vertical federated learning." 2020.<URL: https: / / doi.org / 10.48550 / arXiv.2001.11154> 3GPP TS 23.288 V18.3.0 6.2C

[0005] One aspect of the present disclosure aims to provide a technology that enables more appropriate selection of information processing devices that will participate in federated learning.

[0006] One aspect of the present disclosure is an information processing device having a processor, the processor performing the steps of: acquiring a federated learning profile regarding another information processing device, the federated learning profile including at least a degree of contribution to the federated learning; and determining an information processing device to participate in the federated learning based on the federated learning profile.

[0007] Another aspect of the present disclosure is an information processing system including a first information processing device and a plurality of second information processing devices, wherein the plurality of second information processing devices transmit federated learning profiles including contribution levels to federated learning, the first information processing device receives the federated learning profiles of the plurality of second information processing devices, and the first information processing device determines the information processing devices to participate in the federated learning based on the federated learning profiles.

[0008] According to aspects of the present disclosure, it is possible to more preferably select information processing devices that will participate in federated learning.

[0009] FIG. 1 is a diagram illustrating components of an information processing system according to an embodiment. FIG. 2 is a diagram showing an example of the configuration of an information processing device constituting an information processing system according to an embodiment. FIG. 3 is a diagram showing the functional configuration of an information processing device constituting an information processing system according to an embodiment. FIG. 4 is a diagram explaining a registration and discovery procedure according to an embodiment. FIG. 5 is a diagram explaining a registration and discovery procedure according to an embodiment. FIG. 6 is a diagram explaining a basic procedure and a contribution update procedure of federated learning according to an embodiment. FIG. 7 is a diagram explaining a basic procedure and a contribution update procedure of federated learning according to an embodiment.

[0010] (Summary) In federated learning processing, the coordinator or active participant (hereinafter collectively referred to as active participant) controls the entire learning process. The active participant selects passive participants to participate in federated learning, provides global model information to passive participants, obtains local learning results from passive participants, and performs processes such as updating the global model by integrating the local learning results. The passive participant performs local learning processing using the global model obtained from active participants and its own learning data, and transmits the learning results to the active participant.

[0011] Non-Patent Document 2 discloses that each NF registers profile information indicating the type of federated learning it can participate in in a repository, and active participants select passive participants based on this profile information. However, the method in Non-Patent Document 2 can determine whether an NF can handle federated learning for a specific analysis, but cannot determine the extent to which the NF can contribute to the federated learning. For efficient learning, it is desirable to select an NF that is estimated to be able to make a significant contribution to the federated learning.

[0012] In consideration of the above-described problems, one aspect of the present disclosure includes a contribution level in a federated learning profile, and an active participant selects passive participants to participate in the federated learning by referring to the contribution level. More specifically, one aspect of the present disclosure is an information processing system including a first information processing device (active participant) and multiple second information processing devices (candidate passive participants), wherein the multiple second information processing devices transmit federated learning profiles including the contribution levels to the federated learning, the first information processing device receives the federated learning profiles of the multiple second information processing devices, and the first information processing device determines the information processing devices to participate in the federated learning based on the federated learning profiles.

[0013] In this aspect, the first information processing apparatus selects a predetermined number of second information processing apparatuses with the highest contribution levels, or second information processing apparatuses with the contribution levels equal to or greater than a threshold level.

[0014] According to this aspect, active participants can select NFs with high contribution levels by referring to the contribution levels included in the profile information, thereby enabling more efficient federated learning.

[0015] In one embodiment, the federated learning profile may be transmitted and received via a repository. More specifically, the plurality of second information processing devices may transmit the federated learning profile to a repository, and the first information processing device may receive the federated learning profile from the repository. In another embodiment, the federated learning profile may be transmitted and received directly between the first information processing device and the second information processing device without going through a repository.

[0016] In one embodiment, the contribution level is set for each learning type (analysis type) in which the second information processing device can participate. When the second information processing device can participate in federated learning for multiple different analyses, it is expected that the contribution level to each federated learning will be different, so it is preferable to set the contribution level for each learning type. More specifically, in one embodiment using a repository, the multiple second information processing devices transmit the federated learning profiles related to the multiple learning types to the repository, and the first information processing device obtains from the repository the federated learning profile corresponding to the learning type to be performed.

[0017] In one embodiment, the contribution level may be calculated when the first information processing device performs a model integration process based on local model information acquired from the second information processing device participating in the federated learning. The contribution level in the federated learning may be calculated using any existing method, such as a method based on the Shapley value (in the case of vertical federated learning) or a method using deletion diagnostics or an influence function (in the case of horizontal federated learning). The first information processing device may directly register (update) the calculated contribution level in a repository, or may transmit the calculated contribution level to the second information processing device, which then registers (updates) the calculated contribution level in the repository.

[0018] In one embodiment, the information processing system is a mobile communication network system, and the first information processing device and the second information processing device are network functions therein. For example, the first information processing device may be a Network Data Analytics Function (NWDAF) or an Application Function (AF). The second information processing device may also be an NWDAF or an AF. Furthermore, the mobile communication core network may be a core network of any generation, such as 4G, 5G, or 6G.

[0019] (Embodiment 1) Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. The following embodiments are merely examples for the purpose of explanation, and the present disclosure is not limited to the configurations of the embodiments. For example, although an example in which the present disclosure is applied to a fifth-generation mobile communication system will be described below, the present disclosure may also be applied to a fourth-generation mobile communication system or a generation after the fifth generation. The present disclosure may also be applied to a mobile communication system defined by a party other than 3GPP, or to any wireless communication system or wired communication system other than a mobile communication system.

[0020] In addition, although the following description is based on an example in which the NFs participating in the federated learning are NWDAFs, the NFs participating in the federated learning may be any NFs. Furthermore, the federated learning may be vertical federated learning (VFL) or horizontal federated learning (HFL). Vertical federated learning is a technique in which each NF learns using different features of the same sample, while horizontal federated learning is a technique in which each NF learns using the same features of different samples.

[0021] <Configuration of Information Processing System> Figure 1 shows components that make up a fifth-generation mobile communication system (5G network). In Figure 1, UE (User Equipment) 2 is a user (subscriber) terminal. RAN (Radio Access Network) 3 is an access network to a 5G core network (5GC). RAN 3 is composed of base stations (gNBs). The 5G network has a 5G core network (5GC) and an access network ((R)AN), and UE 2, DN 5, and AF 12 are connected to the 5G network. Each of NFs 11a to 11k is a function realized by one or more computers (information processing devices) executing a program. However, a single computer may realize two or more of NFs 11a to 11k. Each of NFs 11a to 11k can also be referred to as a network node or network component.

[0022] 5GC is composed of a set of components with specific functions called NFs (Network Functions). Figure 1 shows the following NFs 11 that make up 5GC. In Figure 1, they are indicated by bold rectangles.

[0023] UPF (User Plane Function) 11a AMF (Access and Mobility Management Function) 11b SMF (Session Management Function) 11c PCF (Policy Control Function) 11d NEF (Network Exposure Function) 11e NRF (Network Repository Function) 11g NSSF (Network Slice Selection Function) 11h AUSF (Authentication Server) Function) 11i UDM (Unified Data Management) 11j NWDAF (Network Data Analytics Function) 11k

[0024] The UPF 11a performs routing and forwarding of user packets (user plane packets transmitted and received by the UE 2), packet inspection, and QoS processing.

[0025] The AMF 11b terminates the N1 and N2 interfaces and performs registration management, connection management, mobility management, etc. of the UE 2. The AMF 11b also performs SMF selection for session management.

[0026] The SMF 11c manages PDU (Protocol Data Unit) sessions and controls the UPF 11a to implement QoS (Quality of Service) control and policy control. The PDU session is a virtual communication path for exchanging data between the UE 2 and a DN (Data Network) 5. The DN 5 is a data network (such as the Internet) outside the 5GC.

[0027] The PCF 11d performs QoS control, policy control, billing control, etc. under the control of the SMF 11c. QoS control involves controlling the quality of communication, such as prioritized packet forwarding. Policy control involves controlling communication, such as QoS based on network or subscriber information, whether packet forwarding is permitted, and billing.

[0028] The NEF11e mediates communication between external nodes (AFs) and nodes within the control plane.

[0029] The NRF 11g stores and manages information on NFs (e.g., AMF, SMF, UPF, etc.) within the 5GC. In response to an inquiry about an NF desired to be used, the NRF 11g can return multiple NF candidates to the inquiry source. The information stored and managed by the NRF 11g includes the federated learning profile of the NF.

[0030] The NSSF 11h has a function of selecting a network slice to be used by a subscriber from among the network slices generated by network slicing. A network slice is a virtual network having specifications according to the application.

[0031] The AUSF 11i is a subscriber authentication server that performs subscriber authentication under the control of the AMF 11b.

[0032] The UDM 11j holds subscriber-related information, provides subscriber information, or acquires, registers, deletes, and changes the status of the UE 2.

[0033] The NWDAF11k has the function of collecting data from each NF11, OAM terminal, external server, etc., analyzing it using AI / machine learning algorithms, and providing analytical information. Multiple instances exist within a single 5G core network, and model learning and analytical information are provided using federated learning processing between these multiple NWDAF instances.

[0034] The AF 12 is an NF that provides application services via the NRF 11g as part of the 5GC, or an NF that is external to the 5GC and provides application services via the NRF 11e. Alternatively, the UE 2 or an application program executed on the UE 2 may operate as the AF 12.

[0035] In 5GC, multiple NFs of the same type may be prepared. For example, an NF 11 may be prepared for each data center (station). Also, one NF 11 may be shared between data centers. Also, multiple NFs 11 of the same type may be configured in one data center. The number of data centers, the number of NFs 11, and the correspondence between the NFs 11 and the data centers may be set appropriately.

[0036] <Configuration of Information Processing Device> Figure 2 is a diagram showing an example configuration of an information processing device that can operate as each of NFs 11a to 11k, an OAM terminal, and an external server. In Figure 2, the information processing device 20 can be configured using a dedicated or general-purpose information processing device (computer) such as a personal computer (PC), a workstation (WS), or a server machine. However, the information processing device 20 may also be a collection (cloud) of one or more computers.

[0037] The information processing device 20 includes a processor 21 as a processing unit or control unit (controller), a storage device 22, a communication interface 23 (communication IF23), an input device 24, and a display 25, all of which are interconnected via a bus 26.

[0038] The storage device 22 includes a main storage device and an auxiliary storage device. The main storage device is used as at least one of a storage area for programs and data, a program development area, a program work area, and a buffer area for communication data. The main storage device is configured with RAM (Random Access Memory) or a combination of RAM and ROM (Read Only Memory). The auxiliary storage device is used as a storage area for data and programs. A non-volatile storage medium is used as the auxiliary storage device. Examples of non-volatile storage media include a hard disk, a solid state drive (SSD), a flash memory, and an EEPROM (Electrically Erasable Programmable Read-Only Memory). The storage device 22 may also include a drive device for a disk recording medium.

[0039] The communication IF 23 is a circuit that performs communication processing. For example, the communication IF 23 is a network interface card (NIC). The communication IF 23 may also be a wireless communication circuit that performs wireless communication (5G, wireless LAN (Wi-Fi (registered trademark)), BLE, etc.). The communication IF 23 may also be a combination of a circuit that performs wired communication processing and a wireless communication circuit.

[0040] The input device 24 includes keys, buttons, a pointing device, a touch panel, etc., and is used to input information. The display 25 is, for example, a liquid crystal display, etc., and displays information and data.

[0041] The processor 21 performs various processes by executing various programs stored in the storage device 22. By the processor 21 executing the programs stored in the storage device 22, the information processing device 20 can operate as each of the NFs 11a to 11k, an OAM terminal, and an external server.

[0042] <Detailed Functions of NWDAF> Figure 3 is a diagram illustrating the NWDAF 30 (11g) in more detail. The NWDAF 30 analyzes 5G network data. Examples of analysis information include user data congestion information, network performance information, UE mobility information, UE communication information, slice load level information, etc.

[0043] The NWDAF 30 includes an analysis logic unit 32, a model learning logic unit 31, and a data collection unit 33. Note that one NWDAF 30 does not need to have both the model learning logic unit 31 and the analysis logic unit 32, and may perform only the learning process (federated learning as an active participant or a passive participant), or may acquire a trained model and perform only analysis.

[0044] The model learning logic unit 31 is a functional unit that performs federated learning, and includes the function of local learning performed as a passive participant, and the function of selecting passive participants, transmitting model parameters, and performing model integration processing as an active participant. Note that the active participant functions can be omitted in NWDAF 30 that functions only as a passive participant. The model learning logic unit 31 performs model learning processing using learning data acquired from the data collection unit 33. The model learning logic unit 31 is also referred to as MTLF (Model Training Logical Function).

[0045] The analysis logic unit 32 acquires the integrated model (global model) learned by federated learning from the model learning logic unit 31, inputs data obtained from the data collection unit 33 into the model, and acquires the analysis results. The analysis results are provided to each NF 11 or the OAM terminal 34. The analysis logic unit 32 is also referred to as AnLF (Analytics Logical Function).

[0046] The data collection unit 33 exchanges data with each NF 11 in the 5GC and the OAM terminal 34 .

[0047] It should be noted that the network functions participating in federated learning do not have to be NWDAFs: any network function that has model learning logic can be a participant in federated learning.

[0048] <Registration and Discovery Processing> The registration and discovery processing in the federated learning processing will be described below with reference to Fig. 4 and Fig. 5. In Fig. 4, the active NF 41 and the passive NFs 42a to 42c may be NWDAFs or any other NFs.

[0049] FIG. 4 is a schematic diagram illustrating the registration and discovery process. As shown in the figure, (1) the passive NFs 42a to 42c (hereinafter, collectively referred to as passive NFs 42 when there is no need to distinguish between them) register their federated learning profiles in the NRF 43. As described below, the federated learning profile includes information indicating the type of analysis learning that the passive NFs 42 can perform and their contribution to that learning type. Next, the active NF 41 (2) acquires the passive NF 42 federated learning profile from the NRF 43 and (3) selects passive NFs 42 to participate in the learning based on the federated profile. The selection criteria are assumed to be a predetermined number of passive NFs 42 with the highest contribution levels, or those with contribution levels above a threshold, among the passive NFs 42 that can perform the learning to be performed. (4) The active NF 41 requests the selected passive NFs 42 to participate in the federated learning, and (5) the passive NFs 42 reply to the active NF 41 regarding whether or not they agree to participate in the federated learning. For the sake of simplicity, FIG. 4 shows the join request being sent only to the passive NF 42c, but in reality, a plurality of passive NFs are selected and the join request is sent to all of them.

[0050] 5 is a sequence diagram showing the flow of the registration and discovery process. In steps S1a and S1b, the active NF 41 and the passive NF 42 send a registration request (Nnrf_NFManagement_NFRequest) including a federated learning profile to the NRF 43. The federated learning profile may be included as part of the NF profile (including an NF instance identifier, an NF type, an IP address, information on the services provided, etc.). The federated learning profile includes capability information indicating whether the NF can participate in federated learning as an active participant, a passive participant, or both. The federated learning profile further includes information on the types of learning that the NF can perform, the contribution level for each learning type, and the data it holds.

[0051] Examples of learning types include learning of a model that estimates user data congestion (User Data Congestion), learning of a model that estimates network performance (Network Performance), learning of a model that estimates UE mobility information (UE Mobility), learning of a model that estimates UE communication information (UE Communication), and learning of a model that estimates a slice load level (Load Level Information). Since learning and analysis are related, learning types are also referred to as analysis types. Therefore, in the present disclosure, learning types are represented by analysis IDs (Analytics IDs).

[0052] The contribution level is an index that indicates the degree of contribution that can be made in the target associative learning. Since the contribution level is expected to differ depending on the learning type, the contribution level is set for each learning type. Details of the contribution level will be described later.

[0053] The information about the retained data is, for example, information indicating what data the NF retains (including when it can be acquired), and includes information about what characteristics (attributes) of the sample data it retains, what sample data it retains, etc. The information about the retained data may include, for example, the acquisition date and time (timestamp) of the data, the validity period of the data, related areas, application ID, group ID, etc.

[0054] In steps S2a and S2b, the NRF 43 notifies the active NF 41 and the passive NF 42 that the federated learning profile has been successfully registered (Nnrf_NFManagement_NFRegisterResponse).

[0055] In step S3, the active NF 41 starts an associative learning process for performing a specific analysis. The type of associative learning that the active NF 41 is about to perform is specified by the analysis ID.

[0056] In step S4, the active NF 41 transmits a discovery request (Nnrf_NFDiscovery_RequestRequest) to the NRF 43. The discovery request includes the learning type (analysis ID) of the federated learning to be executed by the NRF 43.

[0057] In step S5, the NRF 43 extracts federated learning profiles for NFs that can participate in learning of the learning type included in the discovery request, and notifies the active NF 41 of a discovery response (Nnrf_NFDiscovery_RequestResponse).

[0058] In step S6, the active NF 41 selects passive NFs to participate in the federated learning based on the federated learning profile included in the discovery response notified from the NRF 43. For example, the active NF 41 selects a predetermined number of passive NFs with the highest contribution levels or passive NFs with contribution levels equal to or greater than a threshold level.

[0059] In step S7, the active NF 41 transmits a request to the selected passive NF to participate in the federated learning, and in step S8, the passive NF 42 transmits a response indicating whether or not it accepts the participation to the active NF 41. Note that the active NF 41 may first inquire of all passive NFs 42 whether or not they wish to participate in the federated learning, and then select passive NFs to participate in the federated learning from among the passive NFs that can participate, based on their contribution levels.

[0060] <Federated Learning and Contribution Level Update Processing> The basic processing of federated learning and the contribution level update processing will be described below with reference to Fig. 6 to Fig. 8. The passive NF 42 in Fig. 6 to Fig. 8 is a passive NF that has agreed to participate in federated learning.

[0061] FIG. 6 is a schematic diagram illustrating the basic process of federated learning and the process of updating contributions. As shown in the figure, (1) the active NF 41 provides initial model parameters to the passive NF 42. The passive NF 42 (2) performs local learning using the received initial model parameters and learning data it holds or can acquire, and (3) reports local model information representing the learning results to the active NF 41. (4) The active NF 41 integrates the local model information acquired from each passive NF 42 to create (update) an integrated model and calculates the contribution of each passive NF 42 to the learning. (5) The active NF 41 transmits the integrated model parameters to the passive NF 42. A two-step procedure can be used to register contributions to the NRF 43. In one procedure, (6a) the active NF 41 registers the contribution in the NRF 43. In the other procedure, (6b-1) the active NF 41 transmits the contribution to the passive NF 42, and (6b-2) the passive NF 42 registers the contribution to the NRF 43.

[0062] Local learning and model integration in federated learning are well known, so a brief explanation will be given here. Model parameters such as weights w are transmitted from the active NF 41 to the passive NF 42. The passive NF 42 applies local learning data to the model to which the weights w are set to calculate a loss and calculates updated weights w' according to the loss. The active NF 41 integrates the updated weights w' obtained from each passive NF 42 to update the integrated model. The average is, for example, a weighted average according to the number of data samples held by the passive NF 42.

[0063] The contribution may be calculated by any method, such as a Shapley value-based method (in the case of vertical associative learning) or a method using deletion diagnostics or an influence function (in the case of horizontal associative learning). The NRF 43 may replace the contribution stored for the passive NF with the contribution calculated by the active NF 41, or may replace it with a value (e.g., a weighted average) that combines the stored contribution and the contribution calculated by the active NF 41.

[0064] FIG. 7 is a sequence diagram showing the basic process of federated learning and the process of updating contributions. In step S11, the active NF 41 provides the passive NF 42 with initial model parameters. In step S12, the passive NF 42 collects learning data, and in step S13, it performs local learning. Local learning is, for example, a process of updating weight parameters based on the loss (error) when learning data is input to a model. In step S14, the passive NF 42 reports local model information to the active NF 41. In step S15, the active NF 41 integrates the local model information obtained from each passive NF 42 to update the integrated model. For example, the weight parameters of the integrated model can be obtained by weighted averaging the weight parameters obtained from each passive NF 42 according to the number of data items for each passive NF 42, etc. In step S18, the active NF 41 transmits the integrated model parameters to the passive NF 42, and in step S19, the passive NF 42 updates the local model. The processes of steps S12 to S19 are repeated until a predetermined condition is met (for example, a predetermined number of times, or until the error becomes equal to or smaller than a threshold value).

[0065] In step S20, the active NF 41 calculates the contribution of each passive NF 42 in the federated learning process. In step S21, an NF profile update request is sent to the NRF 43. The profile update request includes the identifier of the passive NF and the calculated contribution. Note that, in order for the active NF 41 to update the profile of the passive NF 42 to the NRF 43, it is assumed that the active NF 41 has the authority to perform this update. Note that the calculation and update of the contribution may be performed every time or at predetermined intervals during the repeated processing of steps S12 to S19.

[0066] 8 is a sequence diagram showing another example of the flow of the basic process of federated learning and the contribution update process. In this example, this can be realized even when the active NF 41 does not have the authority to update the profile of the passive NF 42. The processes from step S11 to step S20 are the same as those in FIG. 7, so their explanations are omitted. In step S22, the active NF 41 transmits an NF profile update notification to the passive NF. The profile update notification includes the calculated contribution. Then, in step S23, the passive NF 42 transmits an NF profile update request to the NRF 43.

[0067] <Advantageous Effects of the Present Embodiment> According to the present embodiment, the active NF can select a passive NF that makes a high contribution to the associative learning, and therefore the associative learning can be performed more efficiently.

[0068] <Other Modifications> The above-described embodiment is merely an example, and the present disclosure can be implemented with appropriate modifications within the scope that does not deviate from the gist thereof.

[0069] The method for calculating the contribution degree is not limited to the one described above and may be calculated based on any criteria. Furthermore, the contribution degree is not limited to being calculated by the active NF; the passive NF may calculate the contribution degree itself and register it in the NRF. When the passive NF calculates the contribution degree itself, for example, it may calculate an evaluation value based on the number of samples of data held or available, or the compatibility with the learning type, as the contribution degree. Furthermore, the contribution degree calculated by the active NF and the contribution degree calculated by the passive NF may be registered in the NRF. In this case, the active NF obtains two types of contribution degrees from the NRF and selects passive NFs to participate in the federated learning based on both or either of these two contribution degrees.

[0070] In the above example, each NF registers an NF profile with the NRF, and the active NF discovers the passive NFs using the NRF. However, in an embodiment in which the active NF has passive NFs pre-configured and no discovery procedure is required, the active NF may directly request the federated learning profile from the passive NF. That is, the federated learning profile may be exchanged between the active NF and the passive NF without going through the NRF.

[0071] In the above description, an example in which both the active NF and the passive NF are NWDAFs has been mainly described, but an AF or any other NF may be an active NF or a passive NF. Also, in the above description, an information processing system constituting a 5G core network has been described as an example, but the technique of the present disclosure may be applied to an information processing system constituting a core network of a mobile wireless communication service other than 5G, or to any information processing system unrelated to a mobile wireless communication service.

[0072] The present disclosure can also be realized by supplying a computer program that implements the functions described in the above embodiments to a computer, and having one or more processors in the computer read and execute the program. Such a computer program may be provided to the computer on a non-transitory computer-readable storage medium connectable to the computer's system bus or via a network. Non-transitory computer-readable storage media include, for example, any type of disk, such as a magnetic disk (e.g., a floppy disk, a hard disk drive (HDD), etc.), an optical disk (e.g., a CD-ROM, a DVD disk, a Blu-ray disk), a read-only memory (ROM), a random access memory (RAM), an EPROM, an EEPROM, a magnetic card, a flash memory, an optical card, or any type of medium suitable for storing electronic instructions.

[0073] 11g: NRF 11k, 30: NWDAF 31: Model learning logic unit 41: Active NF 42: Passive NF 43: NRF

Claims

1. An information processing device having a processor, the processor performing the steps of: acquiring federated learning profiles related to other information processing devices, the federated learning profiles including at least a degree of contribution to the federated learning; and determining information processing devices to participate in the federated learning based on the federated learning profiles.

2. The information processing apparatus according to claim 1, wherein the step of acquiring the federated learning profile includes acquiring the federated learning profile from a repository.

3. The information processing device according to claim 1, characterized in that the contribution level is set for each learning type, and in the step of acquiring the federated learning profile, an federated learning profile corresponding to the learning type to be performed is acquired.

4. The information processing device described in claim 1, characterized in that the processor further executes the steps of: calculating the contribution of each information processing device to the federated learning in the model integration process based on local model information obtained from the information processing devices participating in the federated learning; and transmitting the calculated contribution to the repository or each information processing device.

5. The information processing device according to claim 1, characterized in that in the step of determining the information processing devices to participate in the federated learning, a predetermined number of information processing devices with the highest contribution levels or information processing devices with the contribution levels equal to or greater than a threshold level are selected.

6. The information processing device according to claim 1, wherein the associative learning is horizontal associative learning or vertical associative learning.

7. The information processing device according to claim 1, wherein the information processing device is one of network functions in a mobile communication core network.

8. The information processing device according to claim 7, wherein the information processing device is an NWDAF.

9. The information processing device according to claim 1, wherein the information processing device is an application function that is located outside a mobile communication core network and provides a service via the mobile communication core network.

10. An information processing system including a first information processing device and a plurality of second information processing devices, wherein the plurality of second information processing devices transmit federated learning profiles including a degree of contribution to federated learning, the first information processing device receives the federated learning profiles of the plurality of second information processing devices, and the first information processing device determines the information processing devices to participate in the federated learning based on the federated learning profiles.

11. The information processing system according to claim 10, wherein the plurality of second information processing devices transmit the federated learning profile to a repository, and the first information processing device receives the federated learning profile from the repository.

12. The information processing system described in claim 11, characterized in that the contribution level is set for each learning type, the plurality of second information processing devices transmit the federated learning profiles related to the plurality of learning types to the repository, and the first information processing device obtains from the repository the federated learning profile corresponding to the learning type to be performed.

13. The information processing system described in claim 11, characterized in that the first information processing device calculates the contribution of each second information processing device to the federated learning in the model integration process based on local model information obtained from the second information processing devices participating in the federated learning, and transmits the calculated contribution to the repository.

14. The information processing system described in claim 11, characterized in that the first information processing device calculates the contribution of each second information processing device to the federated learning in the model integration process based on local model information obtained from the second information processing devices participating in the federated learning, transmits the calculated contribution to each second information processing device, and the second information processing device transmits the contribution to the repository.

15. The information processing system described in claim 10, characterized in that in the step of determining the information processing devices to participate in the federated learning, a predetermined number of second information processing devices with the highest level of contribution or a second information processing device with the level of contribution equal to or greater than a threshold level is selected.

16. The information processing system according to claim 10, wherein the associative learning is horizontal associative learning or vertical associative learning.

Images