Authorized federated learning

Abstract

A method comprising: checking whether authorization for performing federated learning of a model by a terminal is received from a first network element; monitoring whether a request for performing federated learning of the model through the terminal is received; and if at least one of the following is satisfied, prohibiting execution of federated learning of the model by the terminal: an authorization for federated learning of the model by the terminal is not received, and a request for execution of federated learning of the model by the terminal is not received.

Description

Technical Field

[0001] The present disclosure relates to federated learning, and in particular to delegation thereof.

[0002] Abbreviations

[0003] 3GPP Third Generation Partnership Project

[0004] 5G / 6G / 7G fifth generation / sixth generation / seventh generation 5GC Fifth Generation Core ADRF Analytics Data Repository Capabilities AF Application Function

[0005] AI

[0006] AMF Access and Mobility Management Function CPU Central Processing Unit CRM Customer Relationship Management DB database

[0007] FL Federated Learning

[0008] FLF Federated Learning Network Function gNB Next Generation Node BID Identifier

[0009] IoT

[0010] IVR Interactive Voice Response ML Machine Learning

[0011] MTC Machine Type Communication NAS Non-Access Stratum

[0012] NEF Network Exposure Function SA System Architecture

[0013] SID Research Projects

[0014] SMF session management functions

[0015] SMS text messaging service

[0016] TR Technical Report

[0017] TS Technical Specifications

[0018] UDM Unified Data Management

[0019] UDR Unified Data Repository

[0020] UE User Equipment

[0021] UPU UE parameter update Background Art

[0022] Many applications in mobile networks require training a single common model using large amounts of data from multiple distributed sources, such as UEs or distributed gNBs. In order to minimize the data exchange between the distributed units that generate the data and the (multiple) centralized units that need to create the common model, the concept of federated learning (FL) can be applied. FL is a form of machine learning in which different versions of a model are not trained on a single node, but on different distributed hosts. This is different from distributed machine learning, which trains a single ML model on distributed nodes to use the computing power of different nodes. In other words, FL differs from the case of distributed learning in that: 1) each distributed node in the FL scenario has its own local training data, which may come from a different distribution than the local training data of other nodes; 2) each node computes the parameters for its local ML model; and 3) the central host does not compute versions or parts of the model, but combines the parameters of all distributed models to generate a master model. The goal of this approach is to keep the training dataset where it is generated and perform model training locally at each individual learner in the federation.

[0023] After training the local model, each individual learner transmits its local model parameters, rather than the (original) training dataset, to an aggregation unit, such as an AF or gNB. The aggregation unit utilizes the local model parameters to update the global model, which may eventually be fed back to the local learners for further iterations until the global model converges. Therefore, each local learner benefits from the datasets of other local learners only through the global model shared by the aggregator, without explicitly accessing the large amount of (potentially privacy-sensitive) data available to each of the other local learners. This is particularly useful in Figure 1 It is shown in , where the UE acts as a local learner and the AF (AF2) performs as an aggregation unit.

[0024] In summary, the FL training process can be explained by the following main steps:

[0025] Initialization: Select a machine learning model (e.g. linear regression, neural network) to train and initialize on the local node.

[0026] Client selection: A small number of local nodes are selected to start training on local data. The selected nodes obtain the current statistical model, while other nodes wait for the next round of federation.

[0027] Reporting and aggregation: Each selected node sends its local model to a central function (possibly hosted by a central server) for aggregation. The central function aggregates the received models and sends model updates back to the nodes.

[0028] Termination: Once a predefined termination criterion is met (e.g. reaching a maximum number of iterations),

[0029] The central function aggregates the updates and finalizes the global model.

[0030] In 3GPP SA2 AIML, the following goals are currently discussed:

[0031] The objectives of this study are to focus on identifying key issues, potential threats, requirements and solutions to achieve:

[0032] 1. 5G system assistance for security management (e.g. membership and group management) for distributed / federated learning, splitting, sharing, and model distribution between application AI / ML endpoints (i.e. UE and application AI / ML service / model providers), which requires data transmission support for application layer AI / ML operations on the 5G system;

[0033] 2. Authentication and authorization for third-party applications or application functions to participate in application-layer AI / ML operations involving UE and network data collection and sharing, i.e., UE and network privacy protection, to support application AI / ML services on 5G systems.

[0034] 3.UE and 5G systems to ensure AI / ML-based services and operations.

[0035] 4. Securely configure external parameters required for AI / ML (e.g. expected UE activity behavior, expected UE mobility, etc.). Summary of the invention

[0036] The object of the present invention is to improve the prior art.

[0037] According to a first aspect of the present invention, there is provided an apparatus comprising means for performing the following:

[0038] checking whether authorization for performing federated learning of a model by the terminal is received from the first network element;

[0039] monitoring whether a request for executing federated learning of a model through a terminal is received; and

[0040] Federated learning of a model is prohibited through a terminal if at least one of the following is true:

[0041] Authorization for federated learning of models through the terminal has not been received, and

[0042] Requests to execute federated learning of models through endpoints were not received.

[0043] According to a second aspect of the present invention, there is provided an apparatus comprising means for performing the following:

[0044] monitoring whether a request for authorization to perform federated learning of a first model through a terminal is received from an application function, wherein the request includes: a requirement for a resource of a terminal or data on the terminal for performing federated learning of the first model through the terminal;

[0045] If the request is received, checking whether the requirement is adapted to the relevant restrictions on executing the federated learning of the first model through the terminal; and

[0046] If the requirement does not comply with the relevant restrictions, authorization to perform federated learning of the first model through the terminal is denied.

[0047] According to a third aspect of the present invention, there is provided an apparatus comprising means for performing the following:

[0048] Monitors whether the database receives an overall restriction on the execution of federated learning for any model through the endpoint;

[0049] If the overall limit is received, storing the overall limit in a database;

[0050] supervising whether the database receives a request to provide a first restriction on federated learning, the first restriction being for executing federated learning of a first model through a terminal; and

[0051] In response to receiving the request, a first restriction is provided, wherein the first restriction includes at least one of: an overall restriction and a related restriction for performing federated learning of the first model by the terminal, and the related restriction is based on the overall restriction.

[0052] According to a fourth aspect of the present invention, there is provided a method, the method comprising:

[0053] checking whether authorization for performing federated learning of a model by the terminal is received from the first network element;

[0054] monitoring whether a request for executing federated learning of a model through an endpoint is received; and

[0055] Federated learning of a model is prohibited through a terminal if at least one of the following is true:

[0056] Authorization for federated learning of models through the terminal has not been received, and

[0057] Requests to execute federated learning of models through endpoints were not received.

[0058] According to a fifth aspect of the present invention, there is provided a method, the method comprising:

[0059] monitoring whether a request for authorization to perform federated learning of a first model through a terminal is received from an application function, wherein the request includes: a requirement for a resource of a terminal or data on the terminal for performing federated learning of the first model through the terminal;

[0060] If the request is received, checking whether the requirement is adapted to the relevant restrictions on executing the federated learning of the first model through the terminal; and

[0061] If the requirement does not comply with the relevant restrictions, authorization to perform federated learning of the first model through the terminal is denied.

[0062] According to a sixth aspect of the present invention, there is provided a method, the method comprising:

[0063] Monitors whether the database receives an overall restriction on the execution of federated learning for any model through the endpoint;

[0064] If the overall limit is received, storing the overall limit in a database;

[0065] supervising whether the database receives a request to provide a first restriction on federated learning, the first restriction being for executing federated learning of a first model through a terminal; and

[0066] In response to receiving the request, a first restriction is provided, wherein the first restriction includes at least one of: an overall restriction and a related restriction for performing federated learning of the first model by the terminal, and the related restriction is based on the overall restriction.

[0067] Each of the methods in the fourth to sixth aspects can be a federated learning method.

[0068] According to a seventh aspect of the present invention, there is provided a computer-readable medium comprising program instructions for causing an apparatus to execute a method according to any one of the fourth to sixth aspects.

[0069] According to some embodiments of the present invention, at least one of the following advantages can be achieved:

[0070] The network maintains overall control over authorized learning;

[0071] The UE maintains control over its involvement in authorized learning.

[0072] It should be understood that any of the above modifications may be applied to the various aspects to which they refer, alone or in combination, unless an alternative is explicitly stated to be excluded. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Further details, features, objects and advantages will become apparent from the following detailed description of preferred embodiments of the present invention in conjunction with the accompanying drawings, in which:

[0074] Figure 1 shows message flows according to some example embodiments of the present invention;

[0075] Figure 2 An apparatus according to an example embodiment of the present invention is shown;

[0076] Figure 3 A method according to an example embodiment of the present invention is shown;

[0077] Figure 4 An apparatus according to an example embodiment of the present invention is shown;

[0078] Figure 5 A method according to an example embodiment of the present invention is shown;

[0079] Figure 6 An apparatus according to an example embodiment of the present invention is shown;

[0080] Figure 7 A method according to an example embodiment of the present invention is shown; and

[0081] Figure 8 An apparatus according to an example embodiment of the invention is shown. DETAILED DESCRIPTION

[0082] Hereinafter, certain embodiments of the present invention will be described in detail with reference to the accompanying drawings, wherein the features of the embodiments may be freely combined with each other unless otherwise described. However, it should be clearly understood that the description of certain embodiments is given by way of example only and is by no means intended to be understood as limiting the present invention to the disclosed details.

[0083] Furthermore, it should be understood that the apparatus is configured to perform the corresponding method, although in some cases only the apparatus or the method is described.

[0084] Some example embodiments of the present invention relate to authorization of the FL of a model if the FL process is initiated by the AF, which may be located inside or outside the network to which the UE is connected. For example, if the Amazon AF wants to start the FL process on the UE, which requires 10,000 cycles of model transfer between the UE and the AF, how will the UE authorize the request from the AF?

[0085] According to AIML studies per SA1, AIML traffic will increase in the near future. Many AFs will continuously train models and push them to UE for retraining (FL use case).

[0086] Example: Model X is expected to consume:

[0087] CPU: 0.01% to 0.02%

[0088] Memory: <5MB

[0089] Space: 30MB

[0090] And the data:

[0091] Image = Yes

[0092] Sensor Information: No

[0093] Audio: No.

[0094] If multiple AFs are pushing their models to UE, even if each model will consume less than 1% CPU, then how will the AF be restricted from having any new model on UE? How will coordination work be done at multiple AFs? If AF1 pushes a model to UE1 for 5 hours, and only 1 model is allowed on UE at that time, then how will AF2 get that information? How will the AF be authorized in this regard?

[0095] The user who owns the device (UE) may have his own preferences and standards (e.g. entertainment models should not consume images stored in the UE, or correspondingly should not be used for e.g. sensor information, audio, keyboard input, etc.). Some example embodiments provide a method how the user (UE) can authorize models from different AFs that may consume images (data).

[0096] In addition, FL learning may be at risk of abuse. For example, Model X expects to consume certain UE resources (e.g., CPU, memory, space) and use certain specific data (images, sensor information, audio, input keyboard, etc.) for ML model training, but on the contrary, Model X is malicious and is collecting additional resources and using additional training data from unauthorized UEs.

[0097] According to some example embodiments, the UE may provide UE resource level preference information to the 5GC. The UE resource level preference information includes restrictions on the use of certain resources by the FL. The UE may provide the UE resource level preference information via any operator portal. Alternatively, the UE resource level preference information may be stored in the 5GC by the AF. In some example embodiments, the UE resource level preference information may be predefined in the 5GC, for example, for a certain type of UE and / or a certain type of user.

[0098] Some examples of UE resource level preferences are shown in Table 1:

[0099]

[0100] Table 1: Example of UE resource level preference

[0101] In some example embodiments, the UE provides restrictions on 5GC data access. Such restrictions may relate to, for example, voice, video, camera, SMS, keyboard input, etc. The UE may provide data access restrictions via any operator portal. Alternatively, the data access restrictions may be stored in the 5GC by the AF. In some example embodiments, the data access restrictions may be predefined in the 5GC, for example, for a certain type of UE and / or a certain type of user.

[0102] In some example embodiments, the restriction (for resources and / or for data access) may depend on the category of the model used for the FL (model category preference information). The model category preference information is provided to the 5GC via any operator portal. Alternatively, the UE model category preference information may be stored in the 5GC by the AF. In some example embodiments, the model category preference information may be predefined in the 5GC, for example for a certain type of UE and / or a certain type of user.

[0103] Table 2 shows an example of model category preference information for data access:

[0104]

[0105] Table 2: Example of model category preference information for data access

[0106] Tables 1 and 2 provide only non-limiting examples. More categories or custom categories are possible based on the use case. For example, if the UE is an IoT device, a new category can be defined.

[0107] The UE resource level preferences, data access restrictions, and / or model class preference information (summarized below as "UE restrictions") may be stored in the UDR / UDM or any other suitable database, such as a dedicated database for the FL.

[0108] When the AF wants to send a model for FL to the UE, it provides the model characteristics to the 5G Core / FL server (e.g. model size, expected number of cycles, FL processing duration, local size of the model returned by the UE, UE identity(ies) involved in the FL process, model category, UE data required for model training, etc.) If the AF is outside the network, the request is usually sent to the NEF, but it can also be sent to any NF handling the FL aspects, such as the FL server.

[0109] The 5GC (represented by, for example, a FL server or NEF) may authorize the request based on UE restrictions stored, for example, in UDM / UDR or any other DB (e.g., ADRF). For example, if the current request is the only request to perform FL on the UE, and if the model characteristics fit the UE restrictions, the 5GC accepts the request, otherwise it rejects the request. If the 5GC accepts the request, the 5GC stores the model characteristics and the time when the FL process is started, which will involve UE resources.

[0110] Examples of model features stored in 5GC:

[0111] Model X, AF-ID

[0112] FL processing time (e.g., 10am to 3pm on day X).

[0113] CPU / memory requirements of the model

[0114] If the second AF wishes to perform FL (or the first AF wishes to perform FL for another model), and the number of models expected to be executed in parallel is 1, the 5G core must reject the request because the authorization has failed. That is, if the maximum number of models used for FL at one time is set to 1 under UE restrictions, and one FL process is in progress, and the second AF requests to perform a FL process for another model, the 5G core should reject the request.

[0115] In some example embodiments, the 5GC keeps track of authorized FL learning for the UE. That is, it deducts the resources assigned for authorized ML learning from the corresponding UE limit. Only the remainder of the UE limit is relevant to the next request for FL at the same time. This new limit may be referred to as a relevant limit. For example, if the UE's limit on CPU usage for ML is 1%, and the first request for ML has been authorized and requires 0.3% CPU usage, the relevant limit for subsequent requests for FL for another model is 0.7% CPU usage. In the 5GC, keeping track of the learning of models authorized by the UE and calculating relevant limits may be performed, for example, by the NEF / FLF and / or by the UDM / ADRF. In the latter case, the UDM / ADRF is notified of the authorization granted by the UE for FL or each model and the resources assigned to the FL of these models. In response to a request from the NEF / FLF, the UDM / ADRF may provide relevant limits, whether or not the overall UE limit is included.

[0116] If any changes occur at the AF (e.g. the AF wants to change the FL time), the AF should ask for the updated authorization at the 5GC.

[0117] If the 5GC accepts the request, it will notify the UE accordingly. In addition, depending on the implementation, it may notify the requesting AF accordingly. The 5GC may notify the UE about the authorization information via NAS (NAS container) or UPU procedure (or another procedure, preferably a secure procedure). The information notified to the UE may include at least the model ID. Typically, it may include:

[0118] AF ID

[0119] Model ID

[0120] Time (e.g. start time and end time, or start time and duration)

[0121] Model characteristics (e.g. which UE training data the model is allowed to use)

[0122] The UE may save this information and use it to approve or reject a request for federated learning of a model received from the AF. That is, the request may include relevant information (at least the model ID). The UE compares this information in the request with the stored information. If the corresponding information is not stored in the UE, the UE rejects the request.

[0123] Figure 1 Message sequence charts according to some example embodiments of the invention are shown. Figure 1 The actions in are as follows:

[0124] Action 1, 2: UE sends a message via portal, IVR or SMS (in Figure 1 AF1 / CRM) provides its UE restrictions (resources, data access and / or model category) to the 5G Core. 5GC stores the information in a DB such as UDM / UDR or ADRF.

[0125] Action 3: AF2 wants to transmit the model for FL to the UE. Therefore, AF2 asks for authorization via 5GC. The request for authorization includes the relevant model features. Figure 1 In the example, the 5GC receives a request for authorization at a network exposure function (NEF) or a new federated learning network function (FLF). In some example embodiments, the FLF may be hosted by another network function, such as the NEF. Figure 1 In the related description, the authorized network function is represented as NEF / FLF.

[0126] Action 4: NEF / FLF obtains UE restrictions and information about the already authorized FL learned for the UE from UDM / ADRF (to be updated in action 6). Therefore, it can calculate the relevant restrictions for the authorized FL request.

[0127] Action 5: NEF / FLF checks whether the requirement for FL learning requested by AF2 fits the relevant constraints. If yes, NEF / FLF authorizes the request, e.g. Figure 1 as shown in the example.

[0128] Action 6: Once the request is authorized, the NEF / FLF stores the authorization information (particularly: requirements for FL) into the UDM / ADRF. This information will help to further authorize new requests for executing FL by the UE. For example, if the UE only allows 1 FL at a time, the NEF / FLF should reject a request from another AF asking for authorization to execute another FL at the same time.

[0129] Action 7: The NEF / FLF pushes information about the authorized FL to the UE. The information includes at least the model ID and may include further information about the requesting AF (AF2) and requirements. For example, the NEF / FLF may provide the information to the UE via a NAS container, i.e., the NEF / FLF asks the SMF, and the SMF provides the information to the UE via NAS. As another option, the information about the authorized FL may be integrity protected and delivered to the UE via the UPU. In addition, Figure 1 Not shown in the figure, the NEF / FLF may directly inform the AF of the authorization (instead of or in addition to action 10).

[0130] Actions 8, 9: The UE stores information about the authorized FL (e.g., in the "Authorized FL List") and sends a response ("Agree") back to the 5GC represented by the NEF / FLF.

[0131] Action 10: The NEF / FLF sends a response back to AF2, informing AF2 that the request to execute FL on the UE has been authorized.

[0132] Action 11: AF2 requests UE to start FL. For this purpose, AF2 provides authorization information (model ID, time window, training data to be used, etc.) to the UE.

[0133] Actions 12, 13: The UE checks whether the information received from AF2 matches the information stored in the authorized FL list updated according to action 8. If the received information matches the information stored in the authorized FL list, then Figure 1 As shown, the UE allows FL processing and notifies AF2 accordingly. Otherwise, the UE rejects the request. That is, if the information involved in the model ID is not available in the UE's authorized FL list, the UE rejects the request ( Figure 1 not shown).

[0134] In some example embodiments, if the information includes the requirement, the UE monitors resource usage of the federated learning of the model with respect to the information from action 7 (stored in the UE in action 8). In case of any abuse (i.e., if the federated learning of the model uses more resources than authorized, or uses some other resource than one of the resources it is authorized to use according to the requirement), the UE may discard the federated learning of the model during operation.

[0135] Figure 2 An apparatus according to an exemplary embodiment of the present invention is shown. The apparatus may be a terminal, such as a UE, an MTC device or an IoT device, or an element thereof. Figure 3 A method according to an exemplary embodiment of the present invention is shown. Figure 2 The device can perform Figure 3 method, but not limited to this method. Figure 3 The method can be Figure 2 The present invention is performed by a device, but is not limited to being performed by the device.

[0136] The device includes a component 110 for checking, a component 120 for monitoring, and a component 130 for prohibiting. The component 110 for checking, the component 120 for monitoring, and the component 130 for prohibiting can be an inspection component, a monitoring component, and a prohibiting component, respectively. The component 110 for checking, the component 120 for monitoring, and the component 130 for prohibiting can be an inspector, a monitor, and a prohibitor, respectively. The component 110 for checking, the component 120 for monitoring, and the component 130 for prohibiting can be an inspection processor, a monitoring processor, and a prohibiting processor, respectively.

[0137] The component 110 for checking checks whether authorization for federated learning of a model through a terminal is received from the core network (S110). The component 120 for monitoring monitors whether a request for performing federated learning of a model through a terminal is received (S120). S110 and S120 may be performed in any order. They may be performed in whole or in part in parallel. Figure 3 An example is shown in which checking S110 is performed before monitoring S120, and the result of checking S110 is negative, while the result of monitoring S120 is positive. That is, for example, although no separate authorization is received (S110=No), the UE receives a request for performing federated learning (S120=Yes).

[0138] If at least one of the following conditions is met:

[0139] Authorization for federated learning of the model by the terminal is not received (S110=No), and

[0140] A request for executing federated learning of a model by a terminal is not received (S120=No),

[0141] Then the component for prohibiting 130 prohibits the execution of the federated learning of the model through the terminal (S130). Otherwise, the component for instructing can instruct the execution of the federated learning of the model through the terminal.

[0142] Figure 4 An apparatus according to an exemplary embodiment of the present invention is shown. The apparatus may be a core network, or a function representing a core network, such as a NEF or FL server, or an element thereof. Figure 5 A method according to an exemplary embodiment of the present invention is shown. Figure 4 The device can perform Figure 5 method, but not limited to this method. Figure 5 The method can be Figure 4 The present invention is performed by a device, but is not limited to being performed by the device.

[0143] The device includes a component 220 for inspection, a component 210 for monitoring, and a component 230 for rejection. The component 220 for inspection, the component 210 for monitoring, and the component 230 for rejection can be an inspection component, a monitoring component, and a rejection component, respectively. The component 220 for inspection, the component 210 for monitoring, and the component 230 for rejection can be an inspector, a monitor, and a rejector, respectively. The component 220 for inspection, the component 210 for monitoring, and the component 230 for rejection can be an inspection processor, a monitoring processor, and a rejection processor, respectively.

[0144] The monitoring component 210 monitors whether a request for authorizing federated learning of a model by a terminal is received (S210). The request includes a request for performing federated learning of a first model by a terminal, about resources of the terminal or about data on the terminal. The request may be received from an application function.

[0145] If a request is received (S210=Yes), the component 220 for checking checks whether the request is adapted to the relevant restrictions for executing federated learning of the model through the terminal (S220).

[0146] If the requirement does not meet the relevant restrictions (S220=No), the component 230 for rejecting rejects the authorization to execute the federated learning of the model through the terminal (S230). Otherwise, the federated learning of the model can be authorized to execute through the terminal.

[0147] Figure 6 An apparatus according to an exemplary embodiment of the present invention is shown. The apparatus may be a database, such as UDM or ADRF, or an element thereof. Figure 7A method according to an exemplary embodiment of the present invention is shown. Figure 6 The device can perform Figure 7 method, but not limited to this method. Figure 7 The method can be Figure 6 The present invention is performed by a device, but is not limited to being performed by the device.

[0148] The device includes a component 310 for monitoring, a component 320 for storing, a component 330 for supervising, and a component 340 for providing. The component 310 for monitoring, the component 320 for storing, the component 330 for supervising, and the component 340 for providing may be a monitoring component, a storage component, a supervision component, and a providing component, respectively. The component 310 for monitoring, the component 320 for storing, the component 330 for supervising, and the component 340 for providing may be a monitor, a storage device, a supervisor, and a provider, respectively. The component 310 for monitoring, the component 320 for storing, the component 330 for supervising, and the component 340 for providing may be a monitoring processor, a storage processor, a supervisory processor, and a providing processor, respectively.

[0149] The monitoring component 310 monitors whether the database (eg, UDM or ADRF) receives the overall restriction for executing federated learning of any model through the terminal (S310). If the overall restriction is received (S310=Yes), the storage component 320 stores the overall restriction in the database (S320).

[0150] The component 330 for monitoring monitors whether the database receives a request to provide a first restriction (S330). The first restriction is for performing federated learning of the first model through the terminal. If the request is received (S330=Yes), in response to receiving the request, the component 340 for providing provides a first restriction (S340). The first restriction includes at least one of the following: an overall restriction and a related restriction for performing federated learning of the first model through the terminal. The related restriction is based on the overall restriction. Specifically, the related restriction can be calculated based on the overall restriction minus the resources assigned to the federated learning of other models other than the first model.

[0151] Figure 8 An apparatus according to an example embodiment of the present invention is shown, the apparatus comprising at least one processor 810, at least one memory 820, including computer program code, and the at least one processor 810 is arranged with the at least one memory 820 and the computer program code so that the apparatus at least performs according to Figure 3 , 5 and at least one of 7 and the described methods involved.

[0152] Some example embodiments are explained for 5G networks. However, the present invention is not limited to 5G. It can also be used in other communication networks, such as in previous generations or upcoming 3GPP networks, such as 4G, 6G or 7G, etc. It can also be used in non-3GPP communication networks.

[0153] The functions of 5GC (e.g., NEF, UDM, etc.) mentioned above are only examples. The functional division may be different from that described. Specifically, some or all actions of 5GC may be performed by a dedicated function for the corresponding purpose, or another existing function may take over some or all of these actions.

[0154] A piece of information can be sent from one entity to another in one or more messages. Each of these messages can include further (different) information.

[0155] The names of network elements, network functions, protocols and methods are based on current standards. In other versions or other technologies, the names of these network elements and / or network functions and / or protocols and / or methods may be different, as long as they provide the corresponding functions. The same applies to terminals.

[0156] If not otherwise stated or clear from the context, different references to two entities mean that they perform different functions. This does not necessarily mean that they are based on different hardware. That is, each of the entities described in this specification can be based on different hardware, or some or all of the entities can be based on the same hardware. This does not necessarily mean that they are based on different software. That is, each of the entities described in this specification can be based on different software, or some or all of the entities can be based on the same software. Each of the entities described in this specification can be deployed in the cloud.

[0157] From the above description, it should be apparent that the exemplary embodiments of the present invention provide, for example, a terminal, such as a UE or an MTC device, or a component thereof, an apparatus embodying the terminal, a method for controlling and / or operating the terminal, (multiple) computer programs for controlling and / or operating the terminal, and a medium carrying such (multiple) computer programs and forming (multiple) computer program products. From the above description, it should be apparent that the exemplary embodiments of the present invention provide, for example, a core network function, such as an AF, a CRM, an UDM, an ADRF, or a NEF, or a component thereof, or a combination of these core network functions, an apparatus embodying the core network function, a method for controlling and / or operating the core network function, (multiple) computer programs for controlling and / or operating the core network function, and a medium carrying such (multiple) computer programs and forming (multiple) computer program products.

[0158] Implementations of any of the above blocks, devices, systems, techniques, or methods include, as non-limiting examples, implementations as hardware, software, firmware, dedicated circuits or logic, general purpose hardware or controllers or other computing devices, or some combination thereof. Each of the entities described in this specification can be embodied in the cloud.

[0159] It should be understood that what has been described above is what is currently considered to be the preferred exemplary embodiment of the present invention. However, it should be noted that the description of the preferred exemplary embodiment is given by way of example only, and various modifications may be made without departing from the scope of the present invention defined by the appended claims.

[0160] The phrase “at least one of A and B” includes options of only A, only B, and both A and B. The terms “first X” and “second X” include options of “first X” and “second X” being the same and “first X” and “second X” being different, unless otherwise specified.

Claims

1. An apparatus comprising means for performing the following: checking whether authorization for performing federated learning of a model by the terminal is received from the first network element; monitoring whether a request for executing the federated learning of the model through the terminal is received; and If at least one of the following is satisfied, executing the federated learning of the model through the terminal is prohibited: the authorization for the federated learning of the model through the terminal is not received, and The request to perform the federated learning of the model by the terminal is not received.

2. The apparatus of claim 1 , wherein the component is further configured to perform: If the authorization for the federated learning of the model is received and the request for the federated learning of the model is received, instructions perform the federated learning of the model.

3. The apparatus according to any one of claims 1 and 2, wherein the component is further configured to perform: A first application function is provided with constraints for performing the federated learning of the model.

4. An apparatus according to claim 3, wherein the restriction comprises at least one of: a restriction on a proportion of the first resource to be used for the federated learning of the model, and a restriction on data access on the terminal to be used for the federated learning of the model.

5. The apparatus according to any one of claims 3 and 4, wherein the restriction is related to the category of the model.

6. The apparatus according to any one of claims 3 to 5, if directly or indirectly dependent on claim 2, wherein the component is further configured to perform: monitoring whether the federated learning executing the model violates the restriction; and If executing the federated learning of the model violates the restriction, abandoning executing the federated learning of the model.

7. The apparatus according to any one of claims 1 to 6, wherein the authorization indicates that the second application function is authorized to request to perform the federated learning of the model, and wherein the component is further configured to perform: The second application function is notified that the authorization is received.

8. The apparatus according to any one of claims 1 to 7, wherein the component is further configured to perform: If the request to execute the federated learning of the model by the terminal is received from a third application function and the authorization to execute the federated learning of the model by the terminal is not received from the first network element, the third application function is notified that execution of the federated learning is prohibited.

9. The apparatus according to any one of claims 1 to 8, wherein the checking comprises: It is checked whether the authorization is received from the first network element via a non-access stratum container or as parameter update data for the terminal.

10. The apparatus according to any one of claims 1 to 9, wherein the first network element comprises: Access and mobility management function AMF, or session management function SMF.

11. The apparatus according to any one of claims 1 to 10, wherein the apparatus is included in the terminal, or the apparatus serves as the terminal.

12. An apparatus comprising means for performing: Monitoring whether a request for executing federated learning of a first model through terminal authorization is received from an application function, wherein the request includes: A requirement for resources of the terminal or data on the terminal for performing the federated learning of the first model through the terminal; If the request is received, checking whether the requirement is adapted to the relevant restrictions for executing the federated learning of the first model through the terminal; as well as If the requirement does not match the relevant restriction, denying the authorization to perform the federated learning of the first model through the terminal.

13. The apparatus of claim 12, wherein the component is further configured to perform: If the requirement is adapted to the relevant restriction, authorizing the execution of the federated learning of the first model by the terminal; and If the requirement is adapted to the relevant restriction, the terminal is notified that execution of the federated learning of the first model is authorized.

14. The apparatus of claim 13, wherein the component is further configured to perform: receiving the associated constraints from a database; and If the requirement fits the relevant constraints, the database is notified that execution of the federated learning of the first model is authorized.

15. The apparatus according to any one of claims 13 and 14, wherein the notifying the terminal and the database respectively comprises at least: The identifier of the first model is notified.

16. The apparatus according to claim 15, wherein notifying the terminal and the database separately further comprises: At least one of: an identifier of the application function, a duration of performing the federated learning of the first model by the user device, and a characteristic of the first model is notified.

17. The apparatus according to any one of claims 13 to 16, wherein the component is further configured to perform: If the requirement fits the relevant constraints, the application function is informed that execution of the federated learning of the first model is authorized.

18. The apparatus according to any one of claims 12 to 17, wherein the relevant restriction is based on at least one of the following: a proportion of the total resources to be used for performing the federated learning for any model through the terminal, and a limit on the proportion of the resources authorized for performing the federated learning for each of one or more second models different from the first model; and Restrictions on access to the data to be accessed for performing the federated learning of the first model.

19. The apparatus according to any one of claims 12 to 18, wherein the associated restrictions are associated with a category; and wherein the component is further configured to perform: monitoring whether the first model belongs to the category; and If the first model does not belong to the category, checking whether the requirement fits the relevant constraints is inhibited.

20. The apparatus according to any one of claims 12 to 19, wherein the apparatus is included in a federated learning network function FLF or a network open function NEF, or wherein the apparatus serves as the federated learning network function FLF or the network open function NEF.

21. An apparatus comprising means for performing: Monitors whether the database receives an overall restriction on the execution of federated learning for any model through the endpoint; if the overall limit is received, storing the overall limit in the database; supervising whether the database receives a request to provide a first restriction for performing federated learning of a first model through the terminal; as well as In response to receiving the request, the first restriction is provided, wherein the first restriction includes at least one of the following: the overall restriction, and a related restriction for performing federated learning of the first model through the terminal, and the related restriction is based on the overall restriction.

22. The apparatus of claim 21, wherein: The overall restriction includes at least one of: an overall proportion of total resources to be used for executing the federated learning of any model through the terminal; and a restriction on access to the data to be accessed for executing the federated learning of any model through the terminal.

23. The apparatus of claim 22, wherein: The first restriction includes at least one of: a first proportion of the resources to be used for performing the federated learning of the first model through the terminal; and restrictions on access to the data to be accessed for performing the federated learning of the first model.

24. The apparatus according to any one of claims 21 to 23, wherein the component is further configured to perform: checking whether the database has received information on authorization for performing federated learning of a second model through the terminal; and If the database receives the information about the authorization for performing the federated learning of the second model through the terminal, the database stores the information about the authorization for performing the federated learning of the second model through the terminal.

25. The apparatus of claim 24, wherein the component is further configured to perform: If the information about the authorization for executing the federated learning of the second model through the terminal is received, calculating the relevant restriction for executing the federated learning of the first model based on the overall restriction and the information about the authorization for executing the federated learning of the second model through the terminal; and In response to receiving the request, providing the associated restrictions, wherein The second model is different from the first model.

26. The apparatus according to any one of claims 24 and 25, wherein the component is further configured to perform: If the relevant limit is provided, then providing the overall limit is prohibited.

27. Apparatus according to any one of claims 24 to 26, if directly or indirectly dependent on claim 22, wherein said information about said authorization comprises: A second proportion of the resources is authorized for performing the federated learning of the second model.

28. An apparatus according to any one of claims 21 to 27, wherein the apparatus is included in a user data management function UDM, or an analysis data repository function ADRF, or wherein the apparatus serves as the user data management function UDM, or the analysis data repository function ADRF.

29. The apparatus according to claim 28, wherein the database is composed of the user data management function UDM and the analytical data repository function ADRF, respectively.

30. The apparatus according to any one of claims 1 to 29, wherein the terminal is a user equipment (UE).

31. The device according to any one of claims 1 to 30, wherein the component comprises: at least one processor; as well as At least one memory including computer program code, the at least one memory and the computer program code being configured to, together with the at least one processor, cause the performance of the apparatus.

32. A method comprising: checking whether authorization for performing federated learning of a model by the terminal is received from the first network element; monitoring whether a request for executing the federated learning of the model through the terminal is received; and If at least one of the following is satisfied, executing the federated learning of the model through the terminal is prohibited: the authorization for the federated learning of the model through the terminal is not received, and The request to perform the federated learning of the model by the terminal is not received.

33. The method of claim 32, further comprising: If the authorization for the federated learning of the model is received and the request for the federated learning of the model is received, instructions perform the federated learning of the model.

34. The method according to any one of claims 32 and 33, further comprising: A first application function is provided with constraints for performing the federated learning of the model.

35. A method according to claim 34, wherein the restriction comprises at least one of: a restriction on a proportion of the first resource to be used for the federated learning of the model, and a restriction on data access on the terminal to be used for the federated learning of the model.

36. A method according to any one of claims 34 and 35, wherein the restriction is related to the class of the model.

37. A method according to any one of claims 34 to 36, if directly or indirectly dependent on claim 2, further comprising: monitoring whether the federated learning executing the model violates the restriction; as well as If executing the federated learning of the model violates the restriction, abandoning executing the federated learning of the model.

38. The method according to any one of claims 32 to 37, wherein the authorization indicates that the second application function is authorized to request to perform the federated learning of the model, and the method further comprises: The second application function is notified that the authorization is received.

39. The method according to any one of claims 32 to 38, further comprising: If the request to execute the federated learning of the model by the terminal is received from a third application function and the authorization to execute the federated learning of the model by the terminal is not received from the first network element, the third application function is notified that execution of the federated learning is prohibited.

40. The method of any one of claims 32 to 39, wherein the checking comprises: It is checked whether the authorization is received from the first network element via a non-access stratum container or as parameter update data for the terminal.

41. The method according to any one of claims 32 to 40, wherein the first network element comprises: Access and mobility management function AMF, or session management function SMF.

42. A method comprising: monitoring whether a request for authorization to perform federated learning of a first model through a terminal is received from an application function, wherein the request includes: a requirement for resources of the terminal or data on the terminal for performing the federated learning of the first model through the terminal; If the request is received, checking whether the requirement is adapted to the relevant restrictions for executing the federated learning of the first model through the terminal; and If the requirement does not match the relevant restriction, denying the authorization to perform the federated learning of the first model through the terminal.

43. The method of claim 42, further comprising: If the requirement is adapted to the relevant restriction, authorizing the execution of the federated learning of the first model by the terminal; as well as If the requirement is adapted to the relevant restriction, the terminal is notified that execution of the federated learning of the first model is authorized.

44. The method of claim 43, further comprising: receiving the relevant restriction from a database; as well as If the requirement fits the relevant constraints, the database is notified that execution of the federated learning of the first model is authorized.

45. The method according to any one of claims 43 and 44, wherein the notifying the terminal and the database respectively comprises at least: The identifier of the first model is notified.

46. ​​The method according to claim 44, wherein notifying the terminal and the database separately further comprises: At least one of: an identifier of the application function, a duration of performing the federated learning of the first model by the user device, and a characteristic of the first model is notified.

47. The method according to any one of claims 43 to 46, further comprising: If the requirement fits the relevant constraints, the application function is informed that execution of the federated learning of the first model is authorized.

48. The method of any one of claims 42 to 47, wherein the associated restriction is based on at least one of: a proportion of the total resources to be used for performing the federated learning of any model through the terminal, and a limit on the proportion of the resources authorized for performing the federated learning of each of one or more second models different from the first model; and Restrictions on access to the data to be accessed for performing the federated learning of the first model.

49. A method according to any one of claims 42 to 48, wherein the relevant restrictions are related to categories; and the method further comprises: monitoring whether the first model belongs to the category; as well as If the first model does not belong to the category, checking whether the requirement fits the relevant constraints is inhibited.

50. A method comprising: Monitors whether the database receives an overall restriction on the execution of federated learning for any model through the endpoint; if the overall limit is received, storing the overall limit in the database; supervising whether the database receives a request to provide a first restriction for performing federated learning of a first model through the terminal; as well as In response to receiving the request, the first restriction is provided, wherein the first restriction includes at least one of the following: the overall restriction, and a related restriction for performing federated learning of the first model through the terminal, and the related restriction is based on the overall restriction.

51. The method of claim 50, wherein: The overall restriction includes at least one of: an overall proportion of total resources to be used for performing the federated learning of any model through the terminal; and a restriction on access to the data to be accessed for performing the federated learning of any model through the terminal.

52. The method of claim 51, wherein: The first restriction includes at least one of: a first proportion of the resources to be used for performing the federated learning of the first model through the terminal; and restrictions on access to the data to be accessed for performing the federated learning of the first model.

53. The method according to any one of claims 50 to 52, further comprising: checking whether the database has received information about authorization for performing federated learning of a second model through the terminal; as well as If the database receives the information about the authorization for performing the federated learning of the second model through the terminal, the database stores the information about the authorization for performing the federated learning of the second model through the terminal.

54. The method of claim 53, further comprising: If the information about the authorization for executing the federated learning of the second model through the terminal is received, calculating the relevant restriction for executing the federated learning of the first model based on the overall restriction and the information about the authorization for executing the federated learning of the second model through the terminal; as well as In response to receiving the request, providing the associated restrictions, wherein The second model is different from the first model.

55. The method according to any one of claims 53 and 54, further comprising: If the relevant limit is provided, then providing the overall limit is prohibited.

56. A method according to any one of claims 53 to 55, if directly or indirectly dependent on claim 51, wherein said information about said authorization comprises: A second proportion of the resources is authorized for performing the federated learning of the second model.

57. The method according to any one of claims 50 to 56, wherein the database consists of a User Data Management Function (UDM), or an Analysis Data Repository Function (ADRF).

58. The method according to any one of claims 32 to 57, wherein the terminal is a user equipment (UE).

59. A computer readable medium comprising program instructions for causing an apparatus to perform the method according to any one of claims 32 to 58.

60. A computer-readable medium comprising instructions for causing an apparatus to at least perform the following: checking whether authorization for performing federated learning of a model by the terminal is received from the first network element; monitoring whether a request for executing the federated learning of the model through the terminal is received; and If at least one of the following is satisfied, executing the federated learning of the model through the terminal is prohibited: the authorization for the federated learning of the model through the terminal is not received, and The request to perform the federated learning of the model by the terminal is not received.

61. A computer-readable medium comprising program instructions for causing an apparatus to at least perform the following operations: Monitoring whether a request for executing federated learning of a first model through terminal authorization is received from an application function, wherein the request includes: A requirement for resources of the terminal or data on the terminal for performing the federated learning of the first model through the terminal; If the request is received, checking whether the requirement is adapted to the relevant restrictions for executing the federated learning of the first model through the terminal; as well as If the requirement does not match the relevant restriction, denying the authorization to perform the federated learning of the first model through the terminal.

62. A computer-readable medium comprising program instructions for causing an apparatus to perform at least the following operations: Monitors whether the database receives an overall restriction on the execution of federated learning for any model through the endpoint; if the overall limit is received, storing the overall limit in the database; supervising whether the database receives a request for a first restriction, the first restriction being for performing federated learning of a first model through the terminal; as well as In response to receiving the request, the first restriction is provided, wherein the first restriction includes at least one of the following: the overall restriction, and a related restriction for performing federated learning of the first model through the terminal, and the related restriction is based on the overall restriction.

Images