Network node and program for mobile communication network
The mechanism allows for inference-based analysis results to be shared between network functions in a mobile communication network without transmitting sensitive information, addressing the lack of defined procedures in existing vertical federated learning configurations.
Patent Information
- Application Number
- JP2024031586
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-03-01
- Publication Date
- 2025-09-11
AI Technical Summary
Existing vertical federated learning configurations in mobile communication networks lack a defined procedure for a first network function (NF) to provide analysis results to a second NF without transmitting application-specific information (APP information) to the second NF when they are different entities.
A mechanism is introduced where a first server generates and transmits first intermediate information using a first learning model, which is processed by a second server using a second learning model, with error information being exchanged to update the models without direct transmission of sensitive information.
Enables inference-based analysis results to be made at the second server using information from the first server without sharing sensitive data, ensuring privacy and compliance with data protection regulations.
Smart Images
Figure 2025133562000001_ABST
Abstract
Description
[Technical Field]
[0001] The present disclosure relates to a technology for utilizing learning models in a mobile communication system. [Background technology]
[0002] Non-Patent Document 1 discloses a configuration for data analytics services in a mobile communication network. A network function (NF) that provides a service is also called a producer NF, and an NF that uses (subscribes to) a service provided by a producer NF is also called a consumer NF. Note that a producer NF can also operate as a consumer NF at the same time.
[0003] According to Non-Patent Document 1, a network data analysis function (NWDAF) that is a producer NF that provides data analysis services is provided in the core network of a mobile communication network. A consumer NF obtains analysis results by subscribing to the data analysis service provided by the NWDAF.
[0004] A consumer NF is implemented, for example, in an external data network (DN) connected to the mobile communication network, such as a server located on the Internet. The server is operated, for example, by a service provider that provides a specific service, such as a video distribution service, to wireless devices (WDs) in the mobile communication network. The service provider can obtain analysis results, such as the quality of experience (QoE), of users of WDs that use the service provider's services from the mobile communication network by subscribing the server (NF) operated by the service provider to a data analysis service.
[0005] To obtain the analysis results of the quality of experience, etc., information held by the mobile communication network (hereinafter referred to as NW information) and information held by the service provider, such as information on the application for using the service (hereinafter referred to as APP information), can be used. When using both NW information and APP information for analysis, the NWDAF performs inference using the NW information collected within the mobile communication network and the APP information notified from the NF as inputs to a learning model, and notifies the NF of the result (analysis result).
[0006] However, since the APP information may contain privacy information that the service provider cannot submit to the outside, there may be cases where the APP information cannot be transmitted to the NWDAF. Therefore, it is possible to consider a configuration in which the NWDAF distributes a learning model to the NF, and the NF performs inference using the learning model. However, since the NW information may also contain privacy information that the mobile communication network operator cannot submit to the outside, there may be cases where the NW information cannot be transmitted to the NF.
[0007] In such a case, the configuration of vertical federated learning (VFL) disclosed in Non-Patent Document 2 can be applied. Specifically, as shown in FIG. 1, the NWDAF generates intermediate information using network information as input to learning model A and transmits the intermediate information to the NF. The NF performs inference using the intermediate information acquired from the NWDAF and the APP information collected by the NF as input to learning model B. By applying the VFL configuration, the information transmitted from the NWDAF to the NF can be treated as intermediate information different from network information, and the NF can perform inference to obtain analysis results. [Prior art documents] [Non-patent literature]
[0008] [Non-Patent Document 1] 3GPP TS23.288,V18.3.0,September 2023 [Non-patent document 2] 3GPP S2-2306424 Summary of the Invention [Problem to be solved by the invention]
[0009] When applying the VFL configuration to the analysis service provided by NWDAF, the NF determines the error in the output of learning model B and updates learning model B, while also calculating the error in the intermediate information and feeding this error back to NWDAF. Then, NWDAF updates learning model A based on the error in the intermediate information.
[0010] However, when an NF (hereinafter referred to as the first NF) that has the APP information for obtaining the analysis results is different from an NF (hereinafter referred to as the second NF) that obtains the analysis results, no procedure is defined for the first NF to provide the analysis results to the second NF without sending the APP information to the second NF.
[0011] The present disclosure provides a mechanism that enables a second server to make inferences based on information held by a first server without transmitting the information to the second server. [Means for solving the problem]
[0012] According to one embodiment of the present disclosure, a network node of a mobile communication network includes a first receiving means for receiving first intermediate information generated by a first server using a first learning model from the first server, a first transmitting means for transmitting second intermediate information based on the first intermediate information to a second server using a second learning model, a second receiving means for receiving second error information from the second server indicating an error in the second intermediate information, and a second transmitting means for transmitting first error information indicating an error in the first intermediate information derived from the second intermediate information to the first server. [Effects of the Invention]
[0013] According to the present disclosure, it is possible to make inferences based on information held by a first server at a second server without transmitting the information to the second server. [Brief explanation of the drawings]
[0014] [Figure 1] An explanatory diagram of VFL. [Figure 2] system configuration diagram. [Figure 3] FIG. [Figure 4] FIG. 10 is a diagram showing an example of a learning configuration. [Figure 5] FIG. 10 is a diagram showing an example of a learning configuration. [Figure 6] FIG. 10 is a diagram showing an example of a learning configuration. [Figure 7] FIG. 1 is a diagram showing an example of the configuration of a network node. DETAILED DESCRIPTION OF THE INVENTION
[0015] Hereinafter, the embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention as claimed, and not all combinations of features described in the embodiments are necessarily essential to the invention. Two or more of the features described in the embodiments may be arbitrarily combined. Furthermore, the same reference numerals are used for the same or similar components, and redundant explanations will be omitted.
[0016] Fig. 2 is a configuration diagram of a communication system according to this embodiment. According to Fig. 2, a core network 100 of a mobile communication network includes a network node 1 that implements an NWDAF and a network node 2 that implements a network repository function (NRF). Note that the network node 1 and the network node 2 may be the same network node. In the following description, the network node 1 will also be referred to as an NWDAF, and the network node 2 will also be referred to as an NRF.
[0017] The NWDAF can operate as a consumer NF within the mobile communication network or as a producer NF that provides analysis services to consumer NFs outside the mobile communication network. The NRF has a function of intermediating between the producer NF in the mobile communication network and the consumer NF that wants to use the service provided by the producer NF. Note that the core network 100 also includes various network functions of types different from the NWDAF and NRF.
[0018] Server 4 in FIG. 2 is located in a DN connected to core network 100, for example, the Internet, and is operated by a service provider that provides services such as video distribution services to WDs in a mobile communication network. Server 4 implements a consumer NF that uses an analysis service provided by NWDAF. Server 3 is located in a DN connected to core network 100 and has the function of collecting APP information necessary for the service provider operating server 4 to obtain analysis results, such as information on the location of WDs that use the services of the service provider operating server 4. In the following description, the NF implemented in server 3 will also be referred to as the first NF, and the NF implemented in server 4 will also be referred to as the second NF.
[0019] The first NF can also operate as a second NF, and the second NF can also operate as the first NF. That is, an NF can operate to obtain necessary analysis results using APP information held by other NFs, and can also operate to provide the APP information held by the NF so that other NFs can obtain analysis results.
[0020] 3 is a sequence diagram of the process of generating a learning model for providing the 2NF with the analysis results requested by the 2NF. Note that multiple servers, including servers 3 and 4, which can operate as consumer NFs of the NWDAF, are assumed to have registered in advance with the NRF at least one of the collectable APP information, the desired analysis results, and the information required to obtain the analysis results.
[0021] In S10, the NWDAF determines the learning configuration by searching the information registered in the NRF to determine the NF that is requesting the analysis results and the NF that can collect the APP information necessary to obtain the analysis results. Here, it is assumed that the first NF has determined to provide the analysis results to the second NF based on the APP information it can collect.
[0022] In S11, the NWDAF performs the necessary processing to start training with the first NF and the second NF. For example, the NWDAF may notify the first NF of the APP information to be used and notify the second NF of the analysis results to be provided. In S12, the NWDAF distributes the first training model to be used by the first NF to the first NF, and in S13, the NWDAF distributes the second training model to be used by the second NF to the second NF and starts training.
[0023] FIG. 4 shows a learning configuration in one embodiment. The first NF generates first intermediate information by using specified APP information (hereinafter referred to as first APP information) as input to a first learning model. In S14, the first NF transmits the first intermediate information to the NWDAF. The NWDAF generates second intermediate information by using the first intermediate information and NW information collected by the NWDAF from within the network as input to a third learning model. In S15, the NWDAF transmits the second intermediate information to the second NF. The second NF obtains analysis results by using the second intermediate information and the APP information collected by the second NF (hereinafter referred to as second APP information) as input to a second learning model.
[0024] Next, the second NF updates the second learning model based on the error of the analysis result, and calculates the error of the second intermediate information (hereinafter referred to as the second error), and in S16 transmits second error information indicating the second error to the NWDAF. The NWDAF updates the third learning model based on the second error, and calculates the error of the first intermediate information (hereinafter referred to as the first error), and in S17 transmits first error information indicating the first error to the first NF. The first NF updates the first learning model based on the first error. Thereafter, S14 to S17 are repeated.
[0025] With the above configuration, the first NF can provide the analysis results to the second NF without transmitting the information collected by the first NF to the second NF.
[0026] If the analysis result requested by the second NF does not require the second APP information, the second NF obtains the analysis result by inputting only the second intermediate information to the second learning model. If the analysis result requested by the second NF does not require NW information, the NWDAF does not use the third learning model and transmits the first intermediate information received from the first NF in S14 as the second intermediate information in S15. In this case, the NWDAF transmits the second error information received from the second NF in S16 as the first error information to the first NF in S17.
[0027] The learning configuration is not limited to that shown in FIG. 4 and can be configured as shown in FIG. 5, for example. In FIG. 5, the NWDAF uses a third learning model and a fourth learning model. The NWDAF generates third intermediate information by inputting NW information to the third learning model, and generates second intermediate information by inputting the first intermediate information and the third intermediate information to the fourth learning model. The NWDAF also updates the fourth learning model based on the second error, calculates the error in the third intermediate information and the first error in the first intermediate information, and updates the third learning model based on the error in the third intermediate information.
[0028] Furthermore, the learning configuration can also be as shown in FIG. 6. In FIG. 6, the NWDAF generates third interim information by inputting the NW information to the third learning model, and transmits the first interim information received from the first NF and the third interim information generated by the NWDAF to the second NF. In this case, the second interim information in S15 of FIG. 3 includes the first interim information and the third interim information. The second NF outputs the analysis result by inputting the first interim information and the third interim information included in the second interim information and the second APP information to the second learning model. Furthermore, the second NF updates the second learning model based on the error of the analysis result, calculates the first error of the first interim information and the third error of the third interim information, and transmits second error information including the first error information and the third error information indicating the third error to the NWDAF in S16. The NWDAF updates the third learning model based on the third error and transmits the first error information to the first NF.
[0029] <Device configuration> FIG. 7 is a diagram illustrating a configuration example of a network node 1 that implements NWDAF. The network node 1 includes, for example, one or more processors and one or more memory devices. The one or more memory devices may include volatile memory devices and non-volatile memory devices. Each functional block illustrated in FIG. 7 may be implemented by one or more processors executing a computer program stored in the one or more memory devices. The network node 1 may also be implemented by a single device. Alternatively, the network node 1 may be implemented by multiple devices that can communicate with each other. Note that FIG. 7 illustrates only parts necessary for understanding the embodiment, and the network node 1 may include other functional blocks not illustrated.
[0030] The first transmitter 15 and the second receiver 18 perform transmission and reception processing with the server 3 that implements the first NF, and the second transmitter 16 and the first receiver 17 perform transmission and reception processing with the server 4 that implements the second NF. Note that the first transmitter 15 and the first transmitter 16 can transmit signals via the same interface. Also, the first receiver 17 and the second receiver 18 can receive signals via the same interface.
[0031] The first receiving unit 17 may receive first intermediate information from the server 3. The first intermediate information is generated by the server 3 using the first APP information as input to the first learning model. The first transmitting unit 15 may transmit second intermediate information based on the first intermediate information to the server. The second receiving unit 18 may receive second error information indicating an error in the second interim information from the server 4. The server 4 may obtain an analysis result by using the second interim information as input to the second learning server, and may determine an error in the second interim information based on the error in the analysis result. The second transmitting unit 16 may transmit first error information indicating an error in the first interim information derived from the second interim information to the server 3.
[0032] In one embodiment, the second intermediate information may be the same as the first intermediate information, and the first error information may be the same as the second error information.
[0033] In one embodiment, the generation unit 11 may generate the second intermediate information by inputting the NW information and the first intermediate information collected by the network node 1 to the third learning model. In this case, the update unit 12 updates the third learning model based on the second error information and derives the first error information.
[0034] In one embodiment, the generation unit 11 may generate third intermediate information by inputting the NW information collected by the network node 1 to a third learning model, and may generate second intermediate information by inputting the third intermediate information and the first intermediate information to a fourth learning model. In this case, the update unit 12 updates the fourth learning model based on the second error information, derives third error information indicating the error between the first error information and the third intermediate information, and updates the third learning model based on the third error information.
[0035] In one embodiment, the generation unit 11 generates third intermediate information by using the NW information collected by the network node 1 as input to the third learning model, and the first transmission unit 15 may transmit information including the first intermediate information and the third intermediate information as second intermediate information to the server 4. In this case, the second reception unit 18 receives information including first error information indicating an error in the first intermediate information and third error information indicating an error in the third intermediate information as second error information.
[0036] In one embodiment, the acquisition unit 14 may acquire the first learning model or the second learning model from a learning model stored in the network node 1 or a learning model stored in a storage device of the mobile communication network. Then, the first transmission unit 15 may transmit the second learning model to the server, and the second transmission unit 16 may transmit the first learning model to the server 3.
[0037] In one embodiment, the selection unit 13 can select server 3 and server 4 from multiple servers based on information that can be collected at the servers and that is registered in the network node 2 by multiple servers including server 3 and server 4, and information indicating the analysis results requested by the servers.
[0038] The present disclosure also provides a computer program that, when executed by one or more processors in an apparatus having one or more processors, causes the apparatus to function as network nodes 1 and 2 or servers 3 and 4, and a computer-readable storage medium storing the computer program. The present disclosure also provides methods executed by network nodes 1 and 2 or servers 3 and 4 for the processing shown in Figure 3, computer programs that cause network nodes 1 and 2 or servers 3 and 4 to execute the methods shown in Figure 3, and computer-readable storage media storing the computer programs.
[0039] As described above, a mechanism is provided that enables inference based on information held by a first server on a second server without transmitting that information to the second server. This makes it possible to contribute to Goal 9 of the United Nations' Sustainable Development Goals (SDGs), which is to "Build resilient infrastructure, promote sustainable industrialization and foster innovation."
[0040] The invention is not limited to the above-described embodiment, and various modifications and variations are possible within the scope of the gist of the invention. [Explanation of symbols]
[0041] 15: First transmitting unit, 16: Second transmitting unit, 17: First receiving unit, 18: Second receiving unit
Claims
1. A network node of a mobile communication network, comprising: a first receiving means for receiving, from the first server, first intermediate information generated by the first server using a first learning model; a first transmitting means for transmitting second intermediate information based on the first intermediate information to a second server that uses a second learning model; second receiving means for receiving second error information indicating an error in the second intermediate information from the second server; a second transmitting means for transmitting first error information indicating an error in the first intermediate information derived from the second intermediate information to the first server; A network node comprising:
2. the second intermediate information is the same as the first intermediate information, The network node of claim 1 , wherein the first error information is the same as the second error information.
3. a generation means for generating the second intermediate information by using the first information collected by the network node and the first intermediate information as inputs of a third learning model; an update means for updating the third learning model based on the second error information and deriving the first error information; The network node of claim 1 further comprising:
4. a generating means for generating third intermediate information by using the first information collected by the network node as an input of a third learning model, and generating the second intermediate information by using the third intermediate information and the first intermediate information as input of a fourth learning model; an updating means for updating the fourth learning model based on the second error information, deriving third error information indicating an error between the first error information and the third intermediate information, and updating the third learning model based on the third error information; The network node of claim 1 further comprising:
5. a generating means for generating third intermediate information by using the first information collected by the network node as an input of a third learning model; the second interim information includes the first interim information and the third interim information, The network node according to claim 1 , wherein the second error information includes third error information indicating an error between the first error information and the third intermediate information.
6. the second learning model is transmitted by the first transmission means to the second server, The network node according to claim 1 , wherein the first learning model is transmitted to the first server by the second transmission means.
7. the first server generates the first intermediate information by using the second information as an input of the first learning model; the second server acquires an analysis result by using the second intermediate information as an input of the second learning model; 2. The network node according to claim 1, further comprising: a selection means for selecting the first server and the second server from the plurality of servers based on information indicating information that can be collected at the servers, the information being registered by the plurality of servers with other network nodes of the mobile communication network, and information indicating analysis results requested by the servers.
8. The network node of claim 7 , wherein the other network node is a node implementing a Network Repository Function (NRF).
9. The network node of claim 1 , wherein the network node is a node implementing a Network Data Analysis Function (NWDAF).
10. A program that, when executed by one or more processors of a device having one or more processors, causes the device to function as a network node according to any one of claims 1 to 9.