Federal learning data processing method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a federal learning data processing method and device, electronic equipment and a storage medium, and the method comprises the steps that a first participant sends a first sample alignment request to an NEF network element, and the NEF network element sends a second sample alignment request to a second participant according to the first sample alignment request; the second participant sends a sample identification set of the second participant to the NEF network element according to the second sample alignment request; the NEF network element performs sample alignment on an expected sample identifier set included in the first sample alignment request and a sample identifier set of the second participant to obtain an aligned sample identifier set, and sends the aligned sample identifier set to the first participant; and if the first participant detects that the alignment sample identifier set satisfies the training demand information, the first participant sends confirmation information to the NEF network element, so that the NEF network element sends the alignment sample identifier set to the second participant according to the confirmation information. According to the technical scheme provided by the embodiment of the invention, the data security in the sample alignment process of federal learning can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of federated learning, in particular to a data processing method and device for federated learning, an electronic device, a storage medium and a program product. BACKGROUND

[0002] Federated learning is a new artificial intelligence basic technology, which can train a model among multiple participants without sharing original data. Before model training, sample alignment needs to be performed among the participants to ensure that the same sample space is used for vertical federated learning training. However, the sample alignment process may cause data leakage and other problems, thereby reducing data security. SUMMARY

[0003] Embodiments of the present application provide a data processing method and device for federated learning, an electronic device, a storage medium and a program product, which can improve data security in the sample alignment process of federated learning.

[0004] In a first aspect, the embodiments of the present application provide a data processing method for federated learning, applied to a vertical federated learning system comprising a first participant and a second participant. The first participant is one of an application function (AF) network element and a network data analysis function (NWDAF) network element. The second participant is the other of the AF network element and the NWDAF network element. The method is performed by the first participant, and the method comprises:

[0005] sending a first sample alignment request to a network exposure function (NEF) network element, so that the NEF network element obtains a set of expected sample identifiers of the first participant from the first sample alignment request, and sends a second sample alignment request to the second participant according to the first sample alignment request;

[0006] receiving a set of aligned sample identifiers sent by the NEF network element, wherein the set of aligned sample identifiers is obtained by the NEF network element performing sample alignment on the set of expected sample identifiers and a set of sample identifiers of the second participant, and the set of sample identifiers of the second participant is sent by the second participant to the NEF network element according to the second sample alignment request;

[0007] if it is detected that the set of aligned sample identifiers meets training requirement information, sending confirmation information to the NEF network element, so that the NEF network element sends the set of aligned sample identifiers to the second participant according to the confirmation information.

[0008] In a second aspect, the embodiments of the present application provide a data processing method for federated learning, applied to a vertical federated learning system comprising a first participant and a second participant, the first participant being one of an AF network element and an NWDAF network element, and the second participant being the other of the AF network element and the NWDAF network element, the method comprising:

[0009] obtaining a set of expected sample identifiers of the first participant from a first sample alignment request sent by the first participant, and sending a second sample alignment request to the second participant according to the first sample alignment request;

[0010] receiving a set of sample identifiers of the second participant sent by the second participant according to the second sample alignment request, and performing sample alignment on the set of expected sample identifiers and the set of sample identifiers of the second participant to obtain an aligned set of sample identifiers;

[0011] sending the aligned set of sample identifiers to the first participant, so that the first participant generates confirmation information after detecting that the aligned set of sample identifiers meets training requirement information;

[0012] receiving the confirmation information sent by the first participant, and sending the aligned set of sample identifiers to the second participant according to the confirmation information.

[0013] In a third aspect, the embodiments of the present application provide a data processing method for federated learning, applied to a vertical federated learning system comprising a first participant and a second participant, the first participant being one of an AF network element and an NWDAF network element, and the second participant being the other of the AF network element and the NWDAF network element, the method being performed by the second participant, and the method comprising:

[0014] receiving a second sample alignment request sent by an NEF network element; wherein the second sample alignment request is generated by the NEF network element according to a first sample alignment request sent by the first participant;

[0015] finding sample identifiers according to the second sample alignment request to obtain a set of sample identifiers of the second participant;

[0016] sending the set of sample identifiers of the second participant to the NEF network element, so that the NEF network element performs sample alignment on the set of sample identifiers of the second participant and a set of expected sample identifiers contained in the first sample alignment request to obtain an aligned set of sample identifiers, and sends the aligned set of sample identifiers to the first participant;

[0017] receive the alignment sample identifier set sent by the NEF network element according to the confirmation information, wherein the confirmation information is generated by the first participant after detecting that the alignment sample identifier set meets the training requirement information.

[0018] In a fourth aspect, the embodiments of the present application provide a data processing apparatus for federated learning, applied to a vertical federated learning system comprising a first participant and a second participant, the first participant being one of an AF network element and an NWDAF network element, the second participant being the other of the AF network element and the NWDAF network element, the apparatus being configured in the first participant, and the apparatus comprising:

[0019] a request module configured to send a first sample alignment request to a network exposure function (NEF) network element, so that the NEF network element obtains an expected sample identifier set of the first participant from the first sample alignment request, and sends a second sample alignment request to the second participant according to the first sample alignment request;

[0020] a receiving module configured to receive an alignment sample identifier set sent by the NEF network element, wherein the alignment sample identifier set is obtained by the NEF network element performing sample alignment on the expected sample identifier set and a sample identifier set of the second participant, and the sample identifier set of the second participant is sent by the second participant to the NEF network element according to the second sample alignment request;

[0021] a sending module configured to, if it is detected that the alignment sample identifier set meets training requirement information, send confirmation information to the NEF network element, so that the NEF network element sends the alignment sample identifier set to the second participant according to the confirmation information.

[0022] In a fifth aspect, the embodiments of the present application provide a data processing apparatus for federated learning, applied to a vertical federated learning system comprising a first participant and a second participant, the first participant being one of an AF network element and an NWDAF network element, the second participant being the other of the AF network element and the NWDAF network element, and the apparatus comprising:

[0023] a receiving module configured to obtain an expected sample identifier set of the first participant from a first sample alignment request sent by the first participant, and send a second sample alignment request to the second participant according to the first sample alignment request;

[0024] an alignment module configured to receive a sample identifier set of the second participant sent by the second participant according to the second sample alignment request, and perform sample alignment on the expected sample identifier set and the sample identifier set of the second participant to obtain an alignment sample identifier set;

[0025] a sending module, configured to send the set of aligned sample identifiers to the first participant, so that the first participant generates confirmation information after detecting that the set of aligned sample identifiers meets the training requirement information;

[0026] a notification module, configured to receive the confirmation information sent by the first participant, and send the set of aligned sample identifiers to the second participant according to the confirmation information.

[0027] In a sixth aspect, an embodiment of the present application provides a data processing apparatus for federated learning, applied to a vertical federated learning system including a first participant and a second participant, the first participant being one of an AF network element and an NWDAF network element, the second participant being the other of the AF network element and the NWDAF network element, the apparatus being configured in the second participant, and the apparatus comprising:

[0028] a receiving module, configured to receive a second sample alignment request sent by an NEF network element; wherein the second sample alignment request is generated by the NEF network element according to a first sample alignment request sent by the first participant;

[0029] a searching module, configured to search for sample identifiers according to the second sample alignment request, to obtain a set of sample identifiers of the second participant;

[0030] a sending module, configured to send the set of sample identifiers of the second participant to the NEF network element, so that the NEF network element performs sample alignment on the set of sample identifiers of the second participant and a set of expected sample identifiers contained in the first sample alignment request, to obtain a set of aligned sample identifiers, and send the set of aligned sample identifiers to the first participant;

[0031] an obtaining module, configured to receive the set of aligned sample identifiers sent by the NEF network element according to confirmation information, wherein the confirmation information is generated by the first participant after detecting that the set of aligned sample identifiers meets the training requirement information.

[0032] In a seventh aspect, an embodiment of the present application provides an electronic device, comprising:

[0033] one or more processors;

[0034] a storage apparatus, configured to store one or more computer programs, when the one or more computer programs are executed by the one or more processors, the electronic device implements the method as described above.

[0035] In an eighth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, when the computer program is executed by a processor of an electronic device, the electronic device implements the method as described above.

[0036] In a ninth aspect, the embodiments of the present application provide a computer program product comprising a computer program which, when executed by a processor, implements the method as described above.

[0037] In the technical solutions provided by the embodiments of the present application, the vertical federated learning system comprises a first participant and a second participant, the first participant is one of an AF network element and an NWDAF network element, the second participant is the other of the AF network element and the NWDAF network element, the first participant sends a first sample alignment request to a NEF network element, so that the NEF network element obtains a set of expected sample identifiers of the first participant from the first sample alignment request, and sends a second sample alignment request to the second participant according to the first sample alignment request; the second participant sends a set of sample identifiers of itself to the NEF network element according to the second sample alignment request; the NEF network element performs sample alignment on the set of expected sample identifiers and the set of sample identifiers of the second participant, obtains a set of aligned sample identifiers, and sends the set of aligned sample identifiers to the first participant; if the first participant detects that the set of aligned sample identifiers meets training requirement information, the first participant sends confirmation information to the NEF network element, so that the NEF network element sends the set of aligned sample identifiers to the second participant according to the confirmation information. Compared with related technologies, not only the sample alignment between the AF network element and the NWDAF network element can be implemented, but also the AF network element and the NWDAF network element as participants do not need to directly interact data in the sample alignment process, and the NEF network element performs sample alignment, thereby improving the data security in the sample alignment process of federated learning.

[0038] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1A is a schematic diagram of an implementation environment shown by an exemplary embodiment of the present application;

[0040] Figure 1A is a schematic diagram of an implementation environment shown by an exemplary embodiment of the present application;

[0041] Figure 1A is a schematic diagram of an implementation environment shown by an exemplary embodiment of the present application;

[0042] Figure 2 is a flowchart of a data processing method of federated learning shown by an exemplary embodiment of the present application;

[0043] Figure 3 is a flowchart of a data processing method of federated learning shown by another exemplary embodiment of the present application;

[0044] Figure 4is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0045] Figure 5 is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0046] Figure 6 is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0047] Figure 7 is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0048] Figure 8 is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0049] Figure 9 is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0050] Figure 10 is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0051] Figure 11 is a data processing method flow chart of federated learning shown by another example embodiment of the present application;

[0052] Figure 12 is a schematic diagram of a data processing device of federated learning shown by an example embodiment of the present application;

[0053] Figure 13 is a schematic diagram of a data processing device of federated learning shown by another example embodiment of the present application;

[0054] Figure 14 is a schematic diagram of a data processing device of federated learning shown by another example embodiment of the present application;

[0055] Figure 15 A structural schematic diagram of a computer system of an electronic device suitable for implementing embodiments of the present application is shown. DETAILED DESCRIPTION

[0056] The example embodiments are now described in a more detailed manner with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms, and should not be understood as being limited to only these examples; rather, the purpose of providing these embodiments is to make the present application more comprehensive and complete, and to fully convey the ideas of the example embodiments to those skilled in the art.

[0057] Furthermore, the features, structures, or characteristics described in this application can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to provide a full understanding of the embodiments of this application. However, those skilled in the art will recognize that when implementing the technical solutions of this application, not all the detailed features in the embodiments may be used, one or more specific details may be omitted, or other methods, elements, devices, steps, etc., may be employed.

[0058] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0059] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0060] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0061] It should also be noted that "multiple" as mentioned in this application refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0062] The technical solutions of the embodiments of this application are described in detail below:

[0063] In related technologies, data leakage and other problems can easily occur during sample alignment, reducing data security. Therefore, embodiments of this application provide a data processing method and apparatus, electronic device, storage medium, and program product for federated learning, which can improve data security during the sample alignment process in federated learning.

[0064] Please see Figure 1A ,Figure 1A This is a schematic diagram of an implementation environment related to this application. The implementation environment includes a first participant 110, a Network Exposure Function (NEF) network element 120, and at least one second participant 130. The first participant 110 and the NEF network element 120 communicate via wired or wireless network, and the NEF network element 120 and the second participant 130 communicate via wired or wireless network. The first participant 110 is an active participant in the vertical federated learning system, i.e., a federated learning server, and the second participant 130 is a passive participant in the vertical federated learning system, i.e., a federated learning client. Specifically, the first participant 110 is one of an Application Function (AF) network element and a Network Data Analytics Function (NWDAF) network element, and the second participant 130 is the other of an AF network element and an NWDAF network element; that is, for example, see [link to relevant documentation]. Figure 1A As shown, if the first participant 110 is an AF network element, then the second participant 130 is an NWDAF network element. See [link / reference]. Figure 1A As shown, the first participant 110 is an NWDAF network element, and the second participant 130 is an AF network element.

[0065] The NEF (Network Element) is responsible for managing publicly accessible network data. All external applications wishing to access data within the core network must go through the NEF, which can reside between the core network and external third-party application functionalities. The NEF can provide corresponding security guarantees to ensure the security of external applications accessing the core network.

[0066] AF network elements can implement the control plane functions of third-party application servers, interacting through AF network element-NEF network element-Policy Control function (PCF) network element or AF network element-PCF network element. AF network elements can also implement the user plane functions of third-party application servers, interacting through the application server (AS)-Internet Protocol (IP) transport network-User plane function (UPF) interface.

[0067] The NWDAF network element is used to provide specific network data analysis services to the network. The NWDAF network element can contain at least one of the Analysis Logical Function (AnLF) module and the Model Training Logical Function (MTLF) module. The AnLF module is responsible for model inference and provides general NWDAF service interfaces such as Nnwdaf_AnalyticsSubscription and Nnwdaf_AnalyticsInfo. It can generate analysis results (including static statistical data and dynamic inference results) based on the requests of consumer network elements. The MTLF module is responsible for model training and can provide the trained model to the AnLF module.

[0068] It should be noted that, Figure 2 The number of the first participant 110, NEF network element 120 and second participant 130 is merely illustrative. Depending on actual needs, there can be any number of the first participant 110, NEF network element 120 and second participant 130.

[0069] In one exemplary embodiment, a first participant 110 sends a first sample alignment request to a NEF network element 120, enabling the NEF network element 120 to obtain the desired sample identifier set of the first participant 110 from the first sample alignment request, and to send a second sample alignment request to a second participant 130 according to the first sample alignment request; the second participant 130 sends its own sample identifier set to the NEF network element 120 according to the second sample alignment request; the NEF network element 120 performs sample alignment on the desired sample identifier set and the sample identifier set of the second participant 130 to obtain an aligned sample identifier set, and sends it to the first participant 110; if the first participant 110 detects that the aligned sample identifier set meets the training requirements, it sends a confirmation message to the NEF network element 120. This allows NEF network element 120 to send the aligned sample identifier set to the second participant 130 based on the confirmation information. Compared with related technologies, this approach enables sample alignment between AF network elements and NWDAF network elements. Furthermore, during the sample alignment process, the AF network elements and NWDAF network elements, as participants, do not need to directly interact with data; the NEF network element 120 performs the sample alignment, thus improving data security during the sample alignment process in federated learning. On the other hand, the aligned sample identifier set is only sent to the second participant 130 when it meets the training requirements, which can improve the performance of the federated learning model. Moreover, the aligned sample identifier set is obtained based on the expected sample identifier set of the first participant 110, enabling customization of the training samples participating in the federated learning model.

[0070] The data processing method for federated learning in this application can be applied to federated learning scenarios between AF network elements and NWDAF network elements. The embodiments of this application involve user-related data such as samples. When the method of this application is applied to specific products or technologies, user permission or consent is obtained, and the extraction, use, and processing of related data comply with local security standards and local laws and regulations.

[0071] See Figure 2 , Figure 2 This is a flowchart illustrating a data processing method for federated learning, as shown in an exemplary embodiment of this application. This method can be applied to... Figure 2 The implementation environment shown can be composed of Figure 3 The first participant 110 in the implementation environment shown performs the action.

[0072] like Figure 3 As shown, in an exemplary embodiment, the data processing method for federated learning may include S210-S230, which are described in detail below:

[0073] S210, a first sample alignment request is sent to the Network Open Function (NEF) network element so that the NEF network element can obtain the expected sample identifier set of the first participant from the first sample alignment request, and send a second sample alignment request to the second participant according to the first sample alignment request.

[0074] It should be noted that the first participant in federated learning is either an AF network element or an NWDAF network element, while the second participant in a vertical federated learning system is either an AF network element or an NWDAF network element. In other words, if the first participant is an AF network element, then the second participant is an NWDAF network element, and vice versa.

[0075] The first sample alignment request is generated by the first participant and is used to request sample alignment. The first sample alignment request contains a desired sample identifier set, which is the set of sample identifiers included by the first participant that corresponds to the samples they wish to use for federated learning training. The desired sample identifier set contains at least one sample identifier. The sample identifier is used to distinguish other samples. The sample identifier can be the identifier of the sample object to which the sample belongs. The sample object includes, but is not limited to, user equipment (UE). The identifier of the UE includes, but is not limited to, at least one of the following: Subscription Concealed Identifier (SUCI), IP address, service account, etc. Optionally, the sample identifier set can be a list.

[0076] When sample alignment is required for federated learning training, the first participant can generate a first sample alignment request and send it to the NEF network element. After receiving the first sample alignment request, the NEF network element parses out the desired sample identifier set from it, generates a second sample alignment request, and sends the second sample alignment request to the second participant.

[0077] Optionally, the first sample alignment request may include the identification information of the second participant, so that the NEF network element can determine the second participant based on the identification information. The identification information of the second participant is used to uniquely identify the second participant, and includes, but is not limited to, the second participant's IP address.

[0078] S220, receive the alignment sample identifier set sent by the NEF network element; wherein, the alignment sample identifier set is obtained by the NEF network element by aligning the expected sample identifier set and the sample identifier set of the second participant, and the sample identifier set of the second participant is sent to the NEF network element by the second participant according to the second sample alignment request.

[0079] The first participant receives the aligned sample identifier set from the NEF network element. The process of obtaining the aligned sample identifier set is as follows: After the NEF network element sends the second sample alignment request to the second participant, the second participant searches for its own sample identifiers to construct a sample identifier set, and then sends its own sample identifier set to the NEF network element. The NEF network element performs sample alignment on the desired sample identifier set and the second participant's sample identifier set to obtain the aligned sample identifier set, and then sends the aligned sample identifier set to the first participant.

[0080] S230, if the alignment sample identifier set is detected to meet the training requirements, a confirmation message is sent to the NEF network element so that the NEF network element sends the alignment sample identifier set to the second participant according to the confirmation message.

[0081] To ensure the performance of the trained federated learning model, training requirement information is set to describe the requirements for federated learning training. This includes requirements for the samples participating in federated learning training, such as the number of samples, sample type, etc. In an optional example, the requirement for the number of samples can be that the number of samples participating in training is greater than a sample number threshold. The sample number threshold can be 1000, 2000, etc., and the specific value can be flexibly set according to actual needs.

[0082] The first participant can check whether the aligned sample identifier set meets the training requirements. For example, it can check whether the sample type corresponding to the aligned sample identifier set meets the requirements for that sample type, and whether the number of samples corresponding to the aligned sample identifier set meets the requirements for that sample type. If the aligned sample identifier set meets the training requirements, the representation can be trained using the samples corresponding to the aligned sample identifier set through federated learning. Therefore, it generates confirmation information and sends it to the NEF network element. After receiving the confirmation information, the NEF network element sends the aligned sample identifier set to the second participant, thereby completing the sample alignment between the first and second participants.

[0083] Optionally, the training requirement information includes a sample number threshold. The first participant can calculate whether the number of sample identifiers contained in the aligned sample identifier set is greater than the sample number threshold. If so, it is determined that the training requirement information is met, and confirmation information is sent to the NEF network element.

[0084] In an optional implementation, during the sample alignment process, sample features can also be aligned. Correspondingly, the first participant can receive multiple sample feature type sets corresponding to the second participants sent by the NEF network element, and merge the received sample feature type sets to obtain a merged sample feature type set. If the aligned sample identifier set and the merged sample feature type set are detected to meet the training requirements, confirmation information is sent to the NEF network element.

[0085] Among them, the sample feature type is the type to which the sample feature belongs, and the sample feature is the feature data of the sample. For example, if the sample object is a UE, the latency of UE "****" is 1 second, and the packet loss rate is 1%, then the sample identifier can be "****", the sample feature includes latency of 1 second and packet loss rate of 1%, and the sample feature type includes latency and packet loss rate.

[0086] The merged sample feature type set contains at least one sample feature type. The merged sample feature type set is obtained by merging the sample feature type sets corresponding to multiple second participants. In other words, the merged sample feature type set is the union of the sample feature type sets of multiple second participants. The sample feature type set of the second participants contains the sample feature types that the second participants can provide.

[0087] The training requirements also include requirements for the features of the samples participating in the training, such as requirements for the types of sample features and the number of sample feature types required for each participant. The requirements for sample feature types indicate which types of sample features can participate in the training. Confirmation information is generated only after both the merged sample feature type set and the aligned sample identifier set satisfy the training requirements.

[0088] Given that the training requirements include a threshold for the number of samples and a threshold for the number of feature types, if the number of sample identifiers in the aligned sample identifier set is greater than the sample number threshold, and the number of sample feature types in the merged sample feature type set is greater than the feature type threshold, then the training requirements are deemed met, and a confirmation message is sent to the NEF network element.

[0089] Optionally, if the aligned sample identifier set is detected to not meet the training requirements, or if the merged sample feature type set is detected to not meet the training requirements, the sample identifiers contained in the expected sample identifier set are updated to obtain the updated expected sample identifier set; the first sample alignment request is regenerated based on the updated expected sample identifier set, and the regenerated first sample alignment request is sent to the NEF network element to re-align the samples.

[0090] exist Figure 3 In the illustrated embodiment, not only can sample alignment between AF network elements and NWDAF network elements be achieved, but also, during the sample alignment process, the AF network elements and NWDAF network elements, as participants, do not need to directly interact with data. The NEF network element performs the sample alignment, which improves the data security in the sample alignment process of federated learning.

[0091] In one exemplary embodiment, see Figure 3 , Figure 2 Is Figure 3 The flowchart illustrates a data processing method for federated learning proposed based on the given example. This method can be applied to... Figure 3 The implementation environment shown can be composed of Figure 3 The first participant 110 in the implementation environment shown performs the action.

[0092] like Figure 4 As shown, before S210, the method may also include S310-S330, which are described in detail below:

[0093] S310, send a participant lookup request to the NEF network element so that the NEF network element can find candidate participants that support the vertical federated learning service according to the participant lookup request and obtain the sample identifier set of the candidate participants; the network element type of the candidate participants matches the network element type of the second participant.

[0094] The participant lookup request is used to request the discovery of a second participant. To discover a second participant, the first participant can send a participant lookup request to the NEF network element. The NEF network element, based on this request, searches for network elements that support vertical federated learning services and whose network element type matches that of the second participant. The found network elements are then designated as candidate participants. Specifically, if the first participant is an AF network element, then NWDAF network elements supporting vertical federated learning services are searched as candidate participants; if the first participant is an NWDAF network element, then AF network elements supporting vertical federated learning services are searched as candidate participants. After identifying candidate participants, the NEF network element can obtain the sample identifier set of the candidate participants.

[0095] Optionally, the NEF network element can obtain a sample identifier set of candidate participants from the User Data Management (UDM) network element. Specific methods for obtaining this information will be described later and will not be repeated here. The UDM network element is a system or component responsible for managing and storing user-related data.

[0096] S320, receive the participant lookup response sent by the NEF network element; wherein, the participant lookup response contains the anonymous identification information of the candidate participants and a sample identification set, and the anonymous identification information of the candidate participants is obtained by the NEF network element through anonymization processing of the identification information of the candidate participants.

[0097] To enhance data security, NEF network elements can anonymize the identification information of candidate participants, obtaining anonymous identification information for each candidate. Then, based on the anonymous identification information and sample information of the candidate participants, a participant lookup response is generated and sent to the first participant. In other words, the participant lookup response contains a mapping relationship between anonymous identification information and the sample identification set.

[0098] Anonymization methods include, but are not limited to, encryption.

[0099] S330, Select a second participant from the candidate participants based on the anonymous identification information of the candidate participants and the sample identification set.

[0100] After receiving the participant lookup response, the first participant parses it to obtain the anonymous identifier information and sample identifier set of each candidate participant. Then, based on the parsed information, it selects at least one second participant from the candidate participants.

[0101] The specific selection method can be flexibly set according to actual needs. For example, in one optional implementation, a reference sample identifier set of the first participant can be obtained. With the goal of maximizing the intersection between the sample identifier set of the second participant and the reference sample identifier set, at least a predetermined number of candidate participants can be selected as the second participants, thereby ensuring the number of second participants and increasing the number of training samples. Alternatively, a second participant can be selected from the candidate participants whose corresponding sample identifier set is a superset of the reference sample identifier set. That is, the reference sample identifier set is a subset of the sample identifier set of the second participant, thus allowing more samples from the first participant to be used for federated learning.

[0102] After selecting a candidate reference party, the anonymous identification information of the second reference party can be sent to the NEF network element so that the NEF network element can determine the candidate reference party based on the anonymous identification information of the candidate reference party.

[0103] When generating the first sample alignment request, the anonymous identification information of the second reference party can be added to the first sample alignment request so that the NEF network element can deanonymize the anonymous identification information. Based on the identification information obtained from the deanonymization process, the second sample alignment request is sent to the second participant.

[0104] It should be noted that, Figure 4 For specific implementation details of S210-S230 shown, please refer to Figure 4 S210-S230 shown will not be described again here.

[0105] exist Figure 4 In the illustrated embodiment, during the discovery of the second participant, the NEF network element searches for candidate participants supporting federated learning based on the participant lookup request from the first participant. After obtaining the sample identifier set of the candidate participants, it anonymizes the identifier information of the candidate participants to obtain anonymous identifier information. The anonymous identifier information and the sample identifier set of the candidate participants are then sent to the first participant. This allows the first participant to select the second participant from the candidate participants based on the anonymous identifier information and the sample identifier set. This not only allows the selection of a second participant whose sample identifier set meets the requirements, but also, because the identifier information of the candidate participants has been anonymized, the first participant cannot identify the participant to which the sample identifier set belongs, thus improving data security.

[0106] In one exemplary embodiment, see Figure 5 , Figure 5 Is Figure 5 The flowchart illustrates a data processing method for federated learning proposed based on the given example. This method can be applied to... Figure 5 The implementation environment shown can be composed ofFigure 1A The first participant 110 in the implementation environment shown performs the action.

[0107] like Figure 1A As shown, when there are multiple candidate participants, method S330 may include S410-S420, which are described in detail below:

[0108] S410, obtain the sample feature type set of each candidate participant from the participant lookup response.

[0109] In an optional implementation, if the participant lookup response also includes a set of sample feature types for each candidate participant, then the set of sample identifiers and the set of sample feature types for each candidate participant can be obtained from it.

[0110] S420: Select a second participant from multiple candidate participants based on the sample identifier set and sample feature type set corresponding to each candidate participant.

[0111] In the process of selecting a second reference party, not only the sample identifier set of each candidate reference party can be considered, but also the sample feature type set of each candidate reference party needs to be considered.

[0112] In an alternative implementation, vertical federated learning is applicable to participants with the same sample space but different feature spaces. Therefore, when selecting a second participant, the goal is to maximize the intersection between the sample identifier set of the second participant and the reference sample identifier set, and minimize the number of overlapping sample feature types between the second participant and the first participant. At least one second reference participant is selected from multiple candidate reference participants.

[0113] In another optional implementation, a set of reference sample identifiers can be obtained for selecting a second reference party, and multiple target reference parties whose corresponding sample identifier sets match the set of reference sample identifiers can be selected from multiple candidate reference parties. Overlapping sample feature types between the sample feature type set of each target reference party and the sample feature type set of other target reference parties are found, and a second reference party is selected from multiple target reference parties based on the found overlapping sample feature types. The number of overlapping sample feature types for each target reference party is negatively correlated with the probability that each target reference party is selected as the second reference party. That is, the more overlapping sample feature types between the sample feature type set of each target reference party and the sample feature type set of other target reference parties, the lower the probability that the target reference party is selected as the second reference party, thus ensuring that among the selected second reference parties, each second reference party has fewer overlapping sample feature types with other second reference parties, resulting in different reference parties having different feature spaces.

[0114] The matching of the candidate reference sample identifier set with the reference sample identifier set can mean that the candidate reference sample identifier set is a subset of the reference sample identifier set, or that the number of overlapping sample identifiers between the candidate reference sample identifier set and the reference sample identifier set reaches a set threshold.

[0115] During the selection process, multiple second participants can be selected from the target participants, ensuring that there are no overlapping sample feature types among the multiple second participants.

[0116] It should be noted that, Figure 5 For specific implementation details of S210-S230 shown, please refer to Figure 5 S210-S230 shown, Figure 6 For specific implementation details of S310-S320 shown, please refer to Figure 6 S310-S320 shown will not be described again here.

[0117] exist Figure 6 In the illustrated embodiment, a second participant can be selected from multiple candidate participants based on their respective sample identifier sets and sample feature type sets, thereby ensuring that the samples and sample features provided by the second participant meet the training requirements and improving the performance of the federated learning model.

[0118] See Figure 6 , Figure 7 This is a flowchart illustrating a data processing method for federated learning, as shown in an exemplary embodiment of this application. This method can be applied to... Figure 7 The implementation environment shown can be composed of Figure 7 The NEF120 is executed in the implementation environment shown.

[0119] like Figure 7 As shown, in an exemplary embodiment, the data processing method for federated learning may include S510-S540, which are described in detail below:

[0120] S510: Obtain the expected sample identifier set of the first participant from the first sample alignment request sent by the first participant, and send the second sample alignment request to the second participant according to the first sample alignment request.

[0121] After receiving the first sample alignment request, the NEF network element generates a second sample alignment request and sends it to the second participants. If there are multiple second participants, the second sample alignment request is sent to each of them.

[0122] For a detailed description of the first participant, the second participant, and the first sample alignment request, please refer to the aforementioned records, which will not be repeated here.

[0123] S520, receive the sample identifier set of the second participant sent by the second participant according to the second sample alignment request, and perform sample alignment between the expected sample identifier set and the sample identifier set of the second participant to obtain the aligned sample identifier set.

[0124] In response to the second sample alignment request from the first participant, the second participant obtains the sample identifier, gets the sample identifier set, and then sends the sample identifier set to the NEF network element, so that the NEF network element receives the sample identifier set from the second participant.

[0125] The NEF network element obtains the desired sample identifier set from the first sample alignment request and performs sample alignment between the desired sample identifier set and the sample identifier set of the second participant to obtain the aligned sample identifier set. Optionally, the aligned sample identifier set can be the intersection between the desired sample identifier set and the sample identifier set of the second participant.

[0126] Optionally, if there are multiple second participants, the desired sample identifier set is aligned with the sample identifier sets corresponding to the multiple second participants to obtain an aligned sample identifier set.

[0127] S530, the alignment sample identifier set is sent to the first participant, so that the first participant can generate confirmation information after detecting that the alignment sample identifier set meets the training requirements.

[0128] The sample alignment identifier set is sent to the first participant. The first participant checks whether the aligned sample identifier set meets the training requirements and generates confirmation information based on the detection result. The specific generation process can be found in the description of the foregoing embodiments, and will not be repeated here.

[0129] S540: Receive confirmation information from the first participant and send the aligned sample identifier set to the second participant based on the confirmation information.

[0130] After receiving confirmation from the first participant, the NEF network element sends the aligned sample identifier set to the second participant to complete the sample alignment.

[0131] exist Figure 5 In the illustrated embodiment, not only can sample alignment between AF network elements and NWDAF network elements be achieved, but also, during the sample alignment process, the AF network elements and NWDAF network elements, as participants, do not need to directly interact with data. The NEF network element performs the sample alignment, which improves the data security in the sample alignment process of federated learning.

[0132] In one exemplary embodiment, see Figure 7 , Figure 7 Is Figure 7The flowchart illustrates a data processing method for federated learning proposed based on the given example. This method can be applied to... Figure 8 The implementation environment shown can be composed of Figure 8 The NEF network element 120 is executed in the implementation environment shown.

[0133] like Figure 8 As shown, when there are multiple second participants, the method may further include S610-S620, which are detailed below:

[0134] S610, obtain the sample feature type set for each second participant.

[0135] NEF network elements can obtain the sample feature type set of each second participant.

[0136] S620, send the sample feature type sets corresponding to multiple second participants to the first participant, so that the first participant can merge the sample feature type sets corresponding to multiple second participants to obtain a merged sample feature type set, and generate confirmation information after detecting that the aligned sample identifier set and the merged sample feature type set meet the training requirements.

[0137] The NEF network element sends the sample feature type sets corresponding to multiple second participants to the first participant. The first participant merges the sample feature type sets corresponding to the multiple second participants to obtain a merged sample feature type set, and detects whether the aligned sample identifier set and the merged sample feature type set meet the training requirements. For details, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0138] In an optional implementation, after receiving a first feature alignment request from a first participant, the NEF network element may, in response to the first feature alignment request, send a second feature alignment request to each second participant, so that each second participant obtains its own sample feature type set according to the second feature alignment request and sends it to the NEF network element.

[0139] Optionally, the first feature alignment request may include feature requirement information. NEF network elements can add the feature requirement information to the second feature alignment request so that each second participant can find the sample feature type that matches the feature requirement information and obtain its own sample feature type set.

[0140] Feature requirement information describes the requirements for sample feature types participating in federated learning training. This includes, but is not limited to, at least one of feature configuration information and function configuration information. The feature configuration information contains a desired set of sample feature types, indicating the sample feature types required for federated learning training. The function configuration information indicates from which Network Function (NF) elements the data corresponding to each sample feature type is collected. For example, the desired set of sample feature types may include sample feature type 1 and sample feature type 2. If the second participant can obtain sample feature type 1 and sample feature type 3, it will only provide sample feature type 1 to NEF elements.

[0141] In another optional implementation, after receiving the first sample alignment request, the NEF network element may regard it as having received the first feature alignment request, obtain the sample feature type set of each second participant, and after receiving the second sample alignment request, each second participant may regard it as having received the second feature alignment request, obtain not only its own sample identifier set, but also its own sample feature type set, and send the sample identifier set and sample feature type set to the NEF network element.

[0142] Under these conditions, the first sample alignment request contains feature requirement information, and the second sample alignment request also contains feature requirement information. The NEF network element can generate a first sample alignment response containing the alignment sample identifier set and the sample feature type set corresponding to each of the multiple second participants, and send the first sample alignment response to the first participant. Each second participant can generate a second alignment response containing its own sample identifier set and sample feature type set, and send the second alignment response to the NEF network element.

[0143] It should be noted that, Figure 8 For specific implementation details of S510-S540 shown, please refer to Figure 9 The S510-S540 shown will not be described in detail here.

[0144] exist Figure 9 In the embodiment shown, the NEF network element sends the sample feature type sets corresponding to multiple second participants to the first participant, so that the first participant merges the sample feature type sets corresponding to multiple second participants to obtain a merged sample feature type set. After detecting that the aligned sample identifier set and the merged sample feature type set meet the training requirements, confirmation information is generated, thereby ensuring that the samples and sample features using federated learning meet the requirements and improving the performance of the federated learning model.

[0145] In one exemplary embodiment, see Figure 9 , Figure 9 Is Figure 1AThe flowchart illustrates a data processing method for federated learning proposed based on the given example. This method can be applied to... Figure 1A The implementation environment shown can be composed of Figure 9 The NEF network element 120 is executed in the implementation environment shown.

[0146] like Figure 9 As shown, when there are multiple second participants, S510 can include S710-S720, which are described in detail below:

[0147] S710, obtain the expected sample identifier set of the first participant from the first sample alignment request sent by the first participant.

[0148] The first sample alignment request contains the expected set of sample identifiers.

[0149] S720, generate a second sample alignment request based on the expected sample identifier set, and send the second sample alignment request to each second participant, so that each second participant can find a sample that matches the sample identifier contained in the expected sample identifier set, and construct a sample identifier set for each second participant based on the sample identifier of the found sample.

[0150] The second sample alignment request may include a desired set of sample identifiers. Each second participant can find samples that match the sample identifiers contained in the desired set of sample identifiers, and thus construct its own set of sample identifiers based on the sample identifiers corresponding to the found samples. In other words, each second participant's set of sample identifiers is a subset of the desired set of sample identifiers.

[0151] Under these conditions, S520 includes S730, detailed as follows:

[0152] S730, receive the sample identifier set of the second participant sent by the second participant according to the second sample alignment request, find the intersection between the sample identifier sets corresponding to multiple second participants respectively, and generate an aligned sample identifier set according to the found intersection.

[0153] Since the sample identifier set of each second participant is already a subset of the expected sample identifier set, the intersection between the sample identifier sets corresponding to multiple second participants can be obtained, and the intersection can be used as the aligned sample identifier set.

[0154] In an optional implementation, the type of sample identifier used by the first participant may differ from the type of sample identifier used by the second participant. Therefore, during the process of generating a second sample alignment request based on the desired sample identifier set, each sample identifier contained in the desired sample identifier set can be mapped from a first identifier space to a second identifier space to obtain a mapped desired sample identifier set, and a second sample alignment request containing the mapped desired sample identifier set can be generated. During the process of sending the aligned sample identifier set to the first participant, each sample identifier contained in the aligned sample identifier set can be mapped from the second identifier space to the first identifier space to obtain a mapped aligned sample identifier set, and the mapped aligned sample identifier set can be sent to the first participant so that the first participant can identify the received aligned sample identifier set. Here, the first identifier space is used to characterize the type of sample identifier used by the first participant, and the second identifier space is used to characterize the type of sample identifier used by the second participant. For example, suppose the first participant is an AF network element, in which IP addresses are used as sample identifiers. The second participant is an NWDAF network element, in which SUCIs are used as sample identifiers. The expected sample identifier set contains sample identifiers as IP addresses. The IP addresses are mapped to SUCIs to obtain the mapped expected sample identifier set, so that the second participant can identify the mapped expected sample identifier set. The sample identifier set of the second participant contains SUCIs. The aligned sample identifier set generated by the intersection of the sample identifier sets corresponding to multiple second participants also contains SUCIs. In order for the first participant to identify the aligned sample identifier set, the SUCIs contained in the aligned sample identifier set are mapped to IP addresses, thereby obtaining the mapped aligned sample identifier set. The mapped aligned sample identifier set is then sent to the first participant.

[0155] It should be noted that, Figure 10 For specific implementation details of S530-S540 shown, please refer to Figure 10 The S530-S540 shown will not be described in detail here.

[0156] exist Figure 10 In the illustrated embodiment, the second sample alignment request includes a desired sample identifier set, so that the second participant only needs to find the sample identifiers contained in itself and existing in the desired sample identifier set to obtain the sample identifier set. The NEF network element only needs to generate the aligned sample identifier set based on the intersection between the sample identifier sets corresponding to the multiple second participants, thereby improving the sample alignment efficiency.

[0157] In one exemplary embodiment, see Figure 10 , Figure 11 Is Figure 11 The flowchart illustrates a data processing method for federated learning proposed based on the given example. This method can be applied to...Figure 12 The implementation environment shown can be composed of Figure 12 The NEF network element 120 is executed in the implementation environment shown.

[0158] like Figure 12 As shown, prior to S510, the method may also include S810-S840, which are detailed below:

[0159] S810 receives a participant lookup request sent by the first participant and, based on the participant lookup request, looks up candidate participants that support the vertical federated learning service; the network element type of the candidate participant matches the network element type of the second participant.

[0160] The second participant needs to participate in vertical federated learning. Therefore, the candidate participants need to support the vertical federated learning service. Optionally, the second participant can be a client of the vertical federated learning service. Therefore, candidate participants that support the vertical federated learning client service can be searched so that the candidate participants can act as clients of the vertical federated learning service.

[0161] Optionally, the participant search request may include participant requirement information to find candidate participants that meet the requirements. The participant requirement information includes the vertical federated learning client service, and may also include service requirement information corresponding to the data analysis service, location requirement information, business type requirement information, and at least one of the following from the request sample identifier set: Service requirement information may be the analysis ID corresponding to the data analysis service, which can be the ID corresponding to the data analysis service provided by the vertical federated learning system; the candidate participant being searched needs to support the data analysis service corresponding to this analysis ID. Location requirement information refers to the requirement for the location of the candidate participant, and may include location information, which can be represented by geographic location, TA, Cell ID, etc. Business type requirement information refers to the requirement for the business type associated with the candidate participant, and may include a business type ID (e.g., application ID); the business type associated with the candidate participant needs to match this business type ID. The request sample identifier set contains multiple sample identifiers, and the sample identifier set of the candidate participant can be obtained based on the request sample identifier set. Here, TA stands for Tracking Area, which is used to represent the location, and ID stands for Identity document, which is the identifier.

[0162] Optionally, considering that the parameters representing the same content are of different types in AF network elements and NWDAF network elements, before searching for candidate participants based on participant demand information, parameter mapping can be performed on the data contained in the participant demand information so that the mapped data meets the parameter types in the candidate participants. For example, AF network elements use geographic location as location information, while NWDAF network elements use TA as location information. If the first participant is an AF network element, then the geographic location contained in the participant demand information needs to be mapped to TA, and then candidate participants are searched based on TA.

[0163] In an optional implementation, during the process of searching for candidate participants supporting the vertical federated learning service based on the participant search request, the NEF network element can request the Network Repository Function (NRF) network element to search for candidate participants supporting the vertical federated learning service, so that the NRF network element can search for candidate participants supporting the vertical federated learning service and send the identification information of the candidate participants to the NEF network element.

[0164] Optionally, NEF network elements can send participant requirement information to NRF network elements, enabling NRF network elements to find candidate participants that meet these requirements. For example, NEF network elements can send the vertical federated learning client service and the analysis ID to NRF network elements, allowing NRF network elements to find candidate participants that support the vertical federated learning client service, support the data analysis service corresponding to the analysis ID, and whose network element type matches the network element type of the second participant. For instance, if the second participant is an NWDAF network element, the NRF network element will find NWDAF network elements that support the federated learning client service and support the data analysis service corresponding to the analysis ID, select the found NWDAF network elements as candidate participants, and send the candidate participant's identification information to the NEF network element. The NRF network element, as one of the core network elements, plays a crucial role in providing efficient, scalable, and secure network function discovery and management.

[0165] S820, obtain the sample identifier set of candidate participants.

[0166] In order for the first participant to select a suitable second participant, the NEF network element can obtain sample information of the candidate participants.

[0167] Optionally, the NEF network element can request the UDM network element to find the sample identifier set of the candidate participants based on the identification information of the candidate participants. The identification information of the candidate participants can be sent from the NRF network element to the NEF network element. The UDM network element finds the sample identifier set of the candidate participants based on its own data and sends it to the NEF network element. Network elements that cannot provide services to the sample object have a low probability of obtaining the sample object's data as a sample. The UDM network element can use the identification information of the candidate participants to filter out those that can provide services to the sample object, then obtain the identification information of the sample object served by the filtered candidate participants, using the obtained sample object identification information as the sample identifier to obtain the sample identifier set. Based on this sample identifier set, it generates the sample information of the candidate participants and returns it to the NEF network element. For candidate participants that cannot provide services to the sample object, their sample information is not returned.

[0168] S830, anonymize the identification information of the candidate participants to obtain the anonymous identification information of the candidate participants, and generate the participant search response based on the anonymous identification information of the candidate participants and the sample identification set.

[0169] To prevent the leakage of sample information of candidate participants, NEF network elements can anonymize the identification information of candidate participants to obtain anonymous identification information of the candidate participants. The specific methods of anonymization can be found in the description of the foregoing embodiments, and will not be repeated here.

[0170] The participant lookup response includes anonymous identifiers and sample information for candidate participants.

[0171] Optionally, the NEF network element can obtain the sample feature type set of candidate participants and add it to the participant lookup response. Specifically, it can request the UDM network element to look up the sample feature type set of candidate participants.

[0172] In an optional implementation, if the participant's search request includes a requested sample identifier set, only the sample identifiers of each candidate participant that exist in the requested sample identifier set can be obtained, thus obtaining the sample identifier set of each candidate participant. That is, the sample identifier set of each candidate participant is a subset of the requested sample identifier set. For example, if the requested sample identifier set contains identifiers 1, 2, and 3, and candidate participant 1 can obtain samples with identifiers 1, 3, and 4, then only identifiers 1 and 3 are added to candidate participant 1's sample identifier set.

[0173] Optionally, if the first identifier space corresponding to the first participant is different from the second identifier space corresponding to the second participant, the request sample identifier set can be mapped from the first identifier space to the second identifier space to obtain the mapped request sample identifier set. Then, the sample identifier of each candidate participant that exists in the mapped request sample identifier set is found to obtain the sample identifier set of each candidate participant. The sample identifier set of each candidate participant is then mapped from the second identifier space to the first identifier space to obtain the mapped sample identifier set of each candidate participant. Finally, the mapped sample identifier set is added to the participant lookup response.

[0174] S840 sends the participant lookup response to the first participant, so that the first participant can select the second participant from the candidate participants based on the anonymous identification information of the candidate participants and the sample identification set.

[0175] The participant lookup response is sent to the first participant, enabling the first participant to select the second participant. The specific selection method can be found in the description of the foregoing embodiments, and will not be repeated here.

[0176] It should be noted that, Figure 12 For specific implementation details of S510-S540 shown, please refer to Figure 13 The S510-S540 shown will not be described in detail here.

[0177] exist Figure 13 In the illustrated embodiment, during the discovery of the second participant, the NEF network element searches for candidate participants supporting federated learning based on the participant lookup request from the first participant. After obtaining the sample identifier set of the candidate participants, it anonymizes the identifier information of the candidate participants to obtain anonymous identifier information. The anonymous identifier information and the sample identifier set of the candidate participants are then sent to the first participant. This allows the first participant to select the second participant from the candidate participants based on the anonymous identifier information and the sample identifier set. This not only allows the selection of the second participant whose sample identifier set meets the criteria, but also, due to the anonymization of the candidate participants' identifier information, prevents the first participant from identifying the participant to which the sample identifier set belongs, thus improving data security.

[0178] See Figure 13 , Figure 13 This is a flowchart illustrating a data processing method for federated learning, as shown in an exemplary embodiment of this application. This method can be applied to... Figure 14 The implementation environment shown can be composed of Figure 14 The second participant 130 performs in the implementation environment shown.

[0179] like Figure 14As shown, in an exemplary embodiment, the data processing method for federated learning may include S910-S930, which are described in detail below:

[0180] S910, receive the second sample alignment request sent by the NEF network element; wherein, the second sample alignment request is generated by the NEF network element based on the first sample alignment request sent by the first participant.

[0181] The first sample alignment request is generated by the first participant, and the second sample alignment request is generated by the NEF network element. For a detailed description of the first participant, the second participant, the NEF network element, and the sample alignment request, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0182] S920, based on the second sample alignment request, find the sample identifier and obtain the sample identifier set of the second participant.

[0183] In response to the second sample alignment request, the second participating party searches for the sample identifiers it contains, thereby obtaining its own sample identifier set. Specifically, if the second sample alignment request contains a desired sample identifier set (or a mapped desired sample identifier set), then it searches for the sample identifiers it contains that exist in the desired sample identifier set (or the mapped desired sample identifier set), thereby obtaining its own sample identifier set.

[0184] In an optional implementation, the sample identifier set of the second participant may be generated by the second participant based on the sample identifiers contained in the historically collected sample data.

[0185] In another optional implementation, after receiving the second sample alignment request, the second participant may collect data related to the sample object to which the sample belongs from the data providing network element, and then obtain a sample identifier set based on the sample identifiers contained in the collected data. The data providing network element includes network function (NF) network elements, etc.

[0186] Optionally, after receiving the second feature alignment request and the second sample alignment request, the second participant may also obtain its own sample feature type set and send it to the NEF network element. If the received request contains feature requirement information, it collects data related to the sample object to which the sample belongs from the data providing network element; then, it searches for sample feature types that match the feature requirement information from the sample feature types contained in the collected data to obtain a sample feature type set, and constructs the sample feature type set based on the sample feature types contained in the collected data. The feature requirement information includes, but is not limited to, at least one of feature configuration information and function configuration information. The feature configuration information describes the sample feature types required for federated learning training, and the function configuration information indicates which NF network elements to collect data from.

[0187] Optionally, the second sample alignment request may also include the analysis ID, business type ID, etc.

[0188] S930, the sample identifier set of the second participant is sent to the NEF network element so that the NEF network element performs sample alignment between the sample identifier set of the second participant and the expected sample identifier set contained in the first sample alignment request, obtains the aligned sample identifier set, and sends the aligned sample identifier set to the first participant.

[0189] The second participant sends its own sample identifier set to the NEF network element. The NEF network element obtains the aligned sample identifier set based on the second participant's sample identifier set. For details, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0190] S940 receives the alignment sample identifier set sent by the NEF network element according to the confirmation information, wherein the confirmation information is generated by the first participant after detecting that the alignment sample identifier set meets the training requirements.

[0191] The second participant receives the aligned sample identifier set, thereby completing the sample alignment. Subsequently, it can perform vertical federated learning with the first participant based on the samples corresponding to the aligned sample identifier set.

[0192] exist Figure 14 In the illustrated embodiment, not only can sample alignment between AF network elements and NWDAF network elements be achieved, but also, during the sample alignment process, the AF network elements and NWDAF network elements, as participants, do not need to directly interact with data. The NEF network element performs the sample alignment, which improves the data security in the sample alignment process of federated learning.

[0193] To better understand this invention, this explanation uses a vertical federated learning system as an example, with the server (i.e., the first participant) as the AF network element, the client (i.e., the second participant) as the NWDAF network element, and the sample object as the UE. See also... Figure 15 As shown, the data processing methods of federated learning include S1001-S1014, which are described in detail below:

[0194] S1001, the AF network element sends a client lookup request to the NEF network element.

[0195] When an AF network element needs to train a VFL with an NWDAF network element to build an NFL model, if the AF network element is not trusted and cannot directly obtain data from the NWDAF network element, the AF network element can send a client lookup request (i.e., a participant lookup request) to the NEF network element to find participants in Vertical Federated Learning (VFL), that is, to find clients of the VFL system.

[0196] Optionally, the client lookup request may include at least one of the following: VFL client service, analysis ID, application ID, request external UE ID list of the AF network element (i.e., request sample identifier set), VFL client service, and external location information. The external UE ID is the UE's ID in the AF network element.

[0197] S1002, NEF network element requests NRF network element discovery client.

[0198] The NEF network element performs parameter mapping on the data contained in the client's lookup request. During the mapping process, each external UE ID in the request external UEID list is mapped to an internal UE ID, resulting in a request internal UE ID list. The internal UE ID is the UE's ID within the NWDAF network element, which is also the UE's ID within the core network. The external location information (the location information used in the AF network element) is mapped to TA or Cell-ID.

[0199] After parameter mapping, NEF network elements can request NRF network elements to discover clients. During the request process, NEF network elements can send VFL client services and analytics IDs to NRF network elements. NRF network elements then search for NWDAF network elements that support VFL client services and analytics IDs to obtain candidate NWDAF network elements.

[0200] S1003, the NRF network element sends the ID of the candidate NWDAF network element to the NEF network element.

[0201] The NRF network element will return the ID of the candidate NWDAF network element to the NEF network element. The candidate NWDAF network element is the candidate client.

[0202] S1004, the NEF network element searches for the internal UE ID list of candidate NWDAF network elements from the UDM network element.

[0203] After receiving the ID of the candidate NWDAF network element from the NRF network element, the NEF network element sends the ID of the candidate NWDAF network element to the UDM network element. The UDM network element determines whether the candidate NWDAF network element is available, and searches for the internal UE ID of the available candidate NWDAF network element that exists in the request internal UE ID list, obtains the internal UE ID list of the candidate NWDAF network element, and sends it to the NEF network element.

[0204] S1005, the NEF network element anonymizes the ID of the candidate NWDAF network element to obtain the anonymous ID of the candidate NWDAF network element, and sends the client lookup response to the AF network element.

[0205] The NEF network element maps the internal UE IDs in the internal UE ID list of each candidate NWDAF network element to external UE IDs, obtains the external UE ID list of each candidate NWDAF network element, adds the external UE ID list and anonymous ID of each candidate NWDAF network element to the client lookup response, and sends it to the AF network element.

[0206] S1006, the AF network element generates the first sample alignment request and sends the first sample alignment request to the NEF network element.

[0207] If the client search response contains a set of sample feature types for candidate NWDAF network elements, then a set of NWDAF network elements with no overlap or minimal overlap between sample feature types is selected from the candidate NWDAF network elements. Furthermore, the intersection of the external UE ID lists of this set of NWDAF network elements needs to cover the reference external UE ID list (i.e., the reference sample identifier set) of the AF network elements.

[0208] If the client's search response does not contain a sample feature type set of candidate NWDAF network elements, then at least one NWDAF network element can be selected based on the external UE ID list of candidate NWDAF network elements and the reference external UE ID list.

[0209] The selected NWDAF network element is the client of the vertical federated learning system and is denoted as the target NWDAF network element. The first sample alignment request includes the expected external UE ID list (i.e., the expected sample identifier set), the anonymous ID, analysis ID, application ID, feature configuration information, and function configuration information of the target NWDAF network element. The function configuration information is used to indicate from which NF network elements to collect data, and it can indicate function description information rather than the specific value of the function.

[0210] S1007, the NEF network element sends the second sample alignment request to the target NWDAF network element.

[0211] The data in the first sample alignment request is parameter-mapped to obtain the second sample alignment request. Each external UE ID in the expected external UE ID list is mapped to an internal UE ID to obtain the expected internal UE ID list. The second sample alignment request includes the expected internal UE ID list, application ID, analysis ID, feature profile, function profile, etc.

[0212] S1008, the target NWDAF network element collects UE-related data from the NF network element.

[0213] Based on the functional configuration file and analysis ID, the target NWDAF network element collects UE-related data from the NF network element. Optionally, it can collect data related to UEs corresponding to the expected internal UE ID list.

[0214] S1009, the target NWDAF network element generates a second sample aligned response based on the collected data.

[0215] The target NWDAF network element determines which UE-related data is available and which sample feature types are available based on the collected data, to ensure that the available sample feature types meet the feature profile, i.e., whether the available sample feature types are the sample feature types required in the feature profile. It generates an internal UE ID list for the target NWDAF based on the internal UE IDs of the available UEs that exist in the expected internal UE ID list, and generates a sample feature type set for the target NWDAF based on the available sample feature types that meet the feature profile. The internal UE ID list and sample feature type set of the target NWDAF are then added to the second sample alignment response.

[0216] S1010, the target NWDAF network element sends the second sample alignment response to the NEF network element.

[0217] S1011, NEF network element generates the first sample alignment response.

[0218] If multiple target NWDAF network elements are involved, the NEF network element will collect the second sample alignment responses corresponding to the multiple target NWDAF network elements respectively, and then generate the first sample alignment response.

[0219] Specifically, the intersection of the internal UE ID lists of multiple target NWDAF network elements is calculated to obtain an aligned internal UE ID list. The internal UE IDs in the aligned internal UE ID list are converted into external UE IDs to obtain an aligned external UE ID list. The aligned external UE ID list and the sample feature type sets corresponding to the multiple target NWDAF network elements are added to the first sample alignment response.

[0220] S1012, the NEF network element sends the first sample alignment response to the AF network element.

[0221] S1013, the AF network element verifies the alignment response of the first sample.

[0222] The AF network element merges the sample feature type sets corresponding to multiple target NWDAF network elements to obtain a merged sample feature type set. It verifies whether the number of external UE IDs contained in the aligned external UE ID list and the number of sample feature types contained in the merged sample feature type set meet the requirements. If they meet the requirements, it sends a confirmation message to the NEF network element, and the NEF network element sends the aligned internal UE ID list to the target NWDAF network element. If they do not meet the requirements, it evaluates whether to stop the VFL process. Then, it selects a different expected external UE ID list or a different target NWDAF network element to start a new sample alignment process.

[0223] S1014 When a new target NWDAF network element is introduced or the current target NWDAF network element needs to be updated, the AF network element evaluates whether the ongoing VFL process needs to be re-aligned or whether a new VFL process needs to be started.

[0224] pass Figure 15 The data processing method of federated learning shown enables the secure and efficient discovery of NWDAF network elements as clients of vertical federated learning, and secure sample alignment between them, under the condition that the AF network element is the server of vertical federated learning.

[0225] This section uses an example from a vertical federated learning system where the server (i.e., the first participant) is the NWDAF network element, the client (i.e., the second participant) is the AF network element, and the sample object is the UE. See also... Figure 15 As shown, the data processing methods of federated learning include S1101-S1114, which are described in detail below:

[0226] S1101, the NWDAF network element sends a client lookup request to the NEF network element.

[0227] When an NWDAF network element needs to perform VFL training with an AF network element, if the AF network element is not trusted, the NWDAF network element can send a client lookup request to the NEF network element. Optionally, the client lookup request may include at least one of the following: VFL client service, analytics ID, application ID, request for an internal UE ID list, VFL client service, and internal location information.

[0228] S1102, NEF network element requests NRF network element discovery client.

[0229] The NEF element maps the request for an internal UE ID list to a request for an external UE ID list; and maps TA or Cell-ID to external location information.

[0230] After parameter mapping, NEF network elements can request NRF network elements to discover clients. The process is similar to S902 and will not be described in detail here.

[0231] S1103, the NRF network element sends the ID of the candidate AF network element to the NEF network element.

[0232] The NRF network element will return the ID of the candidate AF network element as a candidate client to the NEF network element.

[0233] S1104, the NEF network element searches for the external UE ID list of candidate AF network elements from the UDM network element.

[0234] The search process is similar to that in S1103, and will not be described in detail here.

[0235] S1105, the NEF network element anonymizes the ID of the candidate AF network element to obtain the anonymous ID of the candidate AF network element, and sends the client lookup response to the NWDAF network element.

[0236] For each candidate AF network element, the external UE ID list is mapped to the internal UE ID list of that candidate AF network element. The client lookup response contains the internal UE ID list and anonymous ID of each candidate AF network element.

[0237] S1106, the NWDAF network element generates the first sample alignment request and sends the first sample alignment request to the NEF network element.

[0238] The selection process for the target AF network element is similar to that in S1006, and will not be repeated here. The first sample alignment request includes the expected internal UE ID list, the anonymous ID, analysis ID, application ID, feature configuration information, and function configuration information of the target AF network element.

[0239] S1107, the NEF network element sends the second sample alignment request to the target AF network element.

[0240] The expected internal UE ID list in the first sample alignment request is mapped to the expected external UE ID list. The second sample alignment request contains the expected external UE ID list, analysis ID, application ID, feature configuration information, and function configuration information.

[0241] S1108, the target AF network element collects UE-related data from the NF network element.

[0242] The process is similar to S1008, and will not be described in detail here.

[0243] S1109, the target AF element generates a second sample aligned response based on the collected data.

[0244] This process is similar to S1009 and will not be described in detail here. The second sample alignment response contains a list of external UE IDs of the target AF network element and a set of sample feature types.

[0245] S1110, the target AF element sends the second sample alignment response to the NEF element.

[0246] S1111, NEF network element generates the first sample alignment response.

[0247] If multiple target AF network elements are involved, the NEF network element calculates the intersection of the external UE ID lists of the multiple target AF network elements to obtain the aligned external UE ID list, and maps the aligned external UE ID list to the aligned internal UE ID list. The first sample alignment response contains the aligned internal UE ID list and the sample feature type set corresponding to the multiple target AF network elements respectively.

[0248] S1112, the NEF network element sends the first sample alignment response to the NWDAF network element.

[0249] S1113, the NWDAF network element verifies the alignment response of the first sample.

[0250] The verification process is similar to that of S1114, and will not be described in detail here.

[0251] S1114 When a new target AF network element is introduced or the current target AF network element needs to be updated, the NWDAF network element evaluates whether the ongoing VFL process needs to be re-aligned or whether a new VFL process needs to be started.

[0252] pass ​ The data processing method of federated learning shown enables the secure and efficient discovery of AF network elements as clients of vertical federated learning, and secure sample alignment between AF network elements, under the condition that NWDAF network elements are servers for vertical federated learning.

[0253] See ​ , ​ This is a block diagram illustrating a data processing apparatus for federated learning, as shown in an exemplary embodiment of this application. It is applied to a federated learning system comprising a first participant and a second participant. The first participant is one of an Application Function (AF) network element and a Network Data Analysis Function (NWDAF) network element, and the second participant is the other of an AF network element and an NWDAF network element. The apparatus is configured within the first participant, as shown below. ​ As shown, the device includes:

[0254] The request module 1201 is configured to send a first sample alignment request to the Network Open Function (NEF) network element, so that the NEF network element can obtain the expected sample identifier set of the first participant from the first sample alignment request, and send a second sample alignment request to the second participant according to the first sample alignment request.

[0255] The receiving module 1202 is configured to receive the alignment sample identifier set sent by the NEF network element; wherein, the alignment sample identifier set is obtained by the NEF network element by aligning the expected sample identifier set and the sample identifier set of the second participant, and the sample identifier set of the second participant is sent to the NEF network element by the second participant according to the second sample alignment request.

[0256] The sending module 1203 is configured to send confirmation information to the NEF network element if it detects that the alignment sample identifier set meets the training requirements, so that the NEF network element sends the alignment sample identifier set to the second participant according to the confirmation information.

[0257] In an exemplary embodiment, based on the foregoing scheme, the device further includes a lookup module configured to:

[0258] Before sending the first sample alignment request to the NEF network element, a participant lookup request is sent to the NEF network element so that the NEF network element can find candidate participants that support the vertical federated learning service according to the participant lookup request and obtain the sample identifier set of the candidate participants; the network element type of the candidate participants matches the network element type of the second participant.

[0259] Receive the participant lookup response sent by the NEF network element; wherein, the participant lookup response contains the anonymous identification information of the candidate participants and a sample identification set, and the anonymous identification information of the candidate participants is obtained by the NEF network element through anonymization processing of the identification information of the candidate participants;

[0260] A second participant is selected from the candidate participants based on the anonymous identification information of the candidate participants and the sample identification set.

[0261] In an exemplary embodiment, based on the aforementioned scheme, when there are multiple candidate participants, the search module is specifically configured as follows:

[0262] Obtain the sample feature type set for each candidate participant from the participant lookup response;

[0263] Based on the sample identifier set and sample feature type set corresponding to multiple candidate participants, a second participant is selected from multiple candidate participants.

[0264] In an exemplary embodiment, based on the foregoing scheme, the search module is specifically configured as follows:

[0265] Obtain a set of reference sample identifiers for selecting a second reference party, and select multiple target reference parties from multiple candidate reference parties whose corresponding sample identifier sets match the set of reference sample identifiers;

[0266] Find the overlapping sample feature types between the sample feature type set of each target reference party and the sample feature type set of other target reference parties, and select a second reference party from multiple target reference parties based on the found overlapping sample feature types; wherein, the number of overlapping sample feature types corresponding to each target reference party is negatively correlated with the probability of each target reference party being selected as the second reference party.

[0267] In an exemplary embodiment, based on the foregoing scheme, the sending module 1203 is specifically configured as follows:

[0268] The system receives sample feature type sets corresponding to multiple second participants sent by NEF network elements, and merges the received sample feature type sets to obtain a merged sample feature type set.

[0269] If the number of sample identifiers in the aligned sample identifier set is greater than the sample number threshold, and the number of sample feature types in the merged sample feature type set is greater than the feature type number threshold, then the training requirement information is satisfied, and the confirmation information is sent to the NEF network element.

[0270] In one exemplary embodiment, based on the foregoing solution, the device further includes a realignment module configured as follows:

[0271] If the aligned sample identifier set is found to not meet the training requirements, the sample identifiers contained in the expected sample identifier set are updated to obtain the updated expected sample identifier set.

[0272] The first sample alignment request is regenerated based on the updated expected sample identifier set, and the regenerated first sample alignment request is sent to the NEF network element.

[0273] It should be noted that, ​ The data processing apparatus for federated learning provided is based on the same concept as the data processing method for federated learning on the first participant side provided in the above embodiments. The specific ways in which each module and unit performs operations have been described in detail in the method embodiments and will not be repeated here.

[0274] See ​ , ​ This is a block diagram illustrating a data processing apparatus for federated learning, as shown in an exemplary embodiment of this application. It is applied to a federated learning system including a first participant and a second participant. The first participant is one of an AF network element and an NWDAF network element, and the second participant is the other of an AF network element and an NWDAF network element, as shown below. ​As shown, the device includes:

[0275] The receiving module 1301 is configured to obtain the expected sample identifier set of the first participant from the first sample alignment request sent by the first participant, and send a second sample alignment request to the second participant according to the first sample alignment request.

[0276] Alignment module 1302 is configured to receive the sample identifier set of the second participant sent by the second participant according to the second sample alignment request, and to perform sample alignment between the expected sample identifier set and the sample identifier set of the second participant to obtain an aligned sample identifier set;

[0277] The sending module 1303 is configured to send the alignment sample identifier set to the first participant, so that the first participant generates confirmation information after detecting that the alignment sample identifier set meets the training requirements.

[0278] The notification module 1304 is configured to receive confirmation information sent by the first participant and send the alignment sample identifier set to the second participant based on the confirmation information.

[0279] In an exemplary embodiment, based on the foregoing scheme, when there are multiple second participants, the device further includes an acquisition module configured to:

[0280] Obtain the sample feature type set for each second participant;

[0281] The sample feature type sets corresponding to multiple second participants are sent to the first participant, so that the first participant can merge the sample feature type sets corresponding to multiple second participants to obtain a merged sample feature type set. After detecting that the aligned sample identifier set and the merged sample feature type set meet the training requirements, confirmation information is generated.

[0282] In an exemplary embodiment, based on the aforementioned scheme, when there are multiple second participants, the device further includes an acquisition module. The receiving module 1301 is specifically configured to: generate a second sample alignment request based on the desired sample identifier set, and send the second sample alignment request to each second participant, so that each second participant can find a sample that matches the sample identifier contained in the desired sample identifier set, and construct a sample identifier set for each second participant based on the sample identifier of the found sample.

[0283] The alignment module 1302 is specifically configured to: find the intersection between the sample identifier sets corresponding to multiple second participants, and generate an aligned sample identifier set based on the found intersection.

[0284] In an exemplary embodiment, based on the foregoing scheme, the receiving module 1301 is specifically configured to: map each sample identifier contained in the desired sample identifier set from a first identifier space to a second identifier space to obtain a mapped desired sample identifier set; wherein, the first identifier space is the identifier space of the sample in the first participant, and the second identifier space is the identifier space of the sample in the second participant; and generate a second sample alignment request containing the mapped desired sample identifier set.

[0285] The notification module 1304 is specifically configured to: map each sample identifier contained in the alignment sample identifier set from the second identifier space to the first identifier space to obtain the mapped alignment sample identifier set, and send the mapped alignment sample identifier set to the first participant.

[0286] In an exemplary embodiment, based on the foregoing scheme, the device further includes a lookup module configured to:

[0287] Before obtaining the expected sample identifier set of the first participant from the first sample alignment request sent by the first participant, a participant lookup request sent by the first participant is received, and according to the participant lookup request, the network storage function (NRF) network element is requested to look up candidate participants that support the vertical federated learning service; the network element type of the candidate participant matches the network element type of the second participant.

[0288] Based on the identification information of the candidate participants sent by the NRF network element, request the User Data Management (UDM) network element to find the sample identification set of the candidate participants;

[0289] The identification information of the candidate participants is anonymized to obtain the anonymous identification information of the candidate participants. Based on the anonymous identification information of the candidate participants and the sample identification set, a participant search response is generated.

[0290] The participant lookup response is sent to the first participant, so that the first participant can select the second participant from the candidate participants based on the anonymous identification information of the candidate participants and the sample identification set.

[0291] It should be noted that, ​ The data processing apparatus for federated learning provided is based on the same concept as the data processing method for federated learning on the NEF network element side provided in the above embodiments. The specific ways in which each module and unit performs operations have been described in detail in the method embodiments and will not be repeated here.

[0292] See ​ , ​This is a block diagram illustrating a data processing apparatus for federated learning, as shown in an exemplary embodiment of this application. It is applied to a federated learning system comprising a first participant and a second participant. The first participant is one of an AF network element and an NWDAF network element, and the second participant is the other of an AF network element and an NWDAF network element. The apparatus is configured within the second participant, as shown below. ​ As shown, the device includes:

[0293] The receiving module 1401 is configured to receive a second sample alignment request sent by the NEF network element; wherein the second sample alignment request is generated by the NEF network element based on the first sample alignment request sent by the first participant.

[0294] The lookup module 1402 is configured to look up sample identifiers based on the second sample alignment request to obtain the sample identifier set of the second participant;

[0295] The sending module 1403 is configured to send the sample identifier set of the second participant to the NEF network element, so that the NEF network element performs sample alignment on the sample identifier set of the second participant and the expected sample identifier set contained in the first sample alignment request, obtains the aligned sample identifier set, and sends the aligned sample identifier set to the first participant.

[0296] The acquisition module 1404 is configured to receive the alignment sample identifier set sent by the NEF network element according to the confirmation information, wherein the confirmation information is generated by the first participant after detecting that the alignment sample identifier set meets the training requirements.

[0297] It should be noted that, ​ The data processing apparatus for federated learning provided is based on the same concept as the data processing method for federated learning on the second participant side provided in the above embodiments. The specific ways in which each module and unit performs operations have been described in detail in the method embodiments and will not be repeated here.

[0298] Embodiments of this application also provide an electronic device, including: one or more processors; and a storage device for storing one or more computer programs, which, when executed by one or more processors, cause the electronic device to implement the federated learning data processing method provided in the above embodiments.

[0299] ​ A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0300] It should be noted that, ​ The computer system 1500 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0301] like​ As shown, the computer system 1500 includes a Central Processing Unit (CPU) 1501, which can perform various appropriate actions and processes based on a computer program stored in Read-Only Memory (ROM) 1502 or a computer program loaded from storage portion 1508 into Random Access Memory (RAM) 1503, such as executing the federated learning data processing method in the above embodiments. Various computer programs and data required for system operation are also stored in RAM 1503. The CPU 1501, ROM 1502, and RAM 1503 are interconnected via bus 1504. An input / output (I / O) interface 1505 is also connected to bus 1504.

[0302] In some embodiments, the following components are connected to the I / O interface 1505: an input section 1506 including a keyboard, mouse, etc.; an output section 1507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1508 including a hard disk, etc.; and a communication section 1509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 1509 performs communication processing via a network such as the Internet. A drive 1510 is also connected to the I / O interface 1505 as needed. A removable medium 1511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1510 as needed so that computer programs read from it can be installed into the storage section 1508 as needed.

[0303] In particular, according to embodiments of this application, a computer program implementing the data processing method of federated learning can be carried on a computer-readable medium, which can be downloaded and installed from a network via the communication section 1509, and / or installed from a removable medium 1511.

[0304] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a computer program that can be used by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer program contained in the computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0305] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and a computer program.

[0306] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0307] Another aspect of this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor of an electronic device, enables the electronic device to implement the data processing method of federated learning as described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0308] Another aspect of this application provides a computer program product, which includes a computer program that, when executed by a processor, implements the federated learning data processing method provided in the various embodiments described above. The computer program can be stored in a computer-readable storage medium. The computer program product can be a computer program as a product, such as an APP (Application), webpage, mini-program, etc.; or, the computer program product can also be a storage medium, device, terminal, virtual machine, etc., containing the computer program.

[0309] The above description is merely a preferred exemplary embodiment of this application and is not intended to limit the implementation of this application. Those skilled in the art can easily make corresponding modifications or alterations based on the main concept and spirit of this application. Therefore, the scope of protection of this application should be determined by the scope of protection claimed in the claims.

Claims

1. A data processing method for federated learning, characterized in that, A method is applied to a vertical federated learning system comprising a first participant and a second participant, wherein the first participant is one of an application function (AF) network element and a network data analysis function (NWDAF) network element, and the second participant is the other of an AF network element and an NWDAF network element. The method is executed by the first participant and includes: A first sample alignment request is sent to the Network Open Function (NEF) network element so that the NEF network element can obtain the expected sample identifier set of the first participant from the first sample alignment request, and send a second sample alignment request to the second participant according to the first sample alignment request. The system receives an aligned sample identifier set sent by the NEF network element; wherein the aligned sample identifier set is obtained by the NEF network element by aligning the expected sample identifier set and the sample identifier set of the second participant, and the sample identifier set of the second participant is sent to the NEF network element by the second participant according to the second sample alignment request; If the alignment sample identifier set is detected to meet the training requirements, a confirmation message is sent to the NEF network element so that the NEF network element sends the alignment sample identifier set to the second participant based on the confirmation message.

2. The method as described in claim 1, characterized in that, Before sending the first sample alignment request to the Network Open Function (NEF) network element, the method further includes: A participant lookup request is sent to the NEF network element so that the NEF network element can find candidate participants that support the vertical federated learning service according to the participant lookup request and obtain the sample identifier set of the candidate participants; the network element type of the candidate participants matches the network element type of the second participant. The system receives a participant lookup response sent by the NEF network element; wherein the participant lookup response includes the anonymous identification information of the candidate participant and a sample identification set, and the anonymous identification information of the candidate participant is obtained by the NEF network element through anonymization processing of the identification information of the candidate participant; The second participant is selected from the candidate participants based on the anonymous identification information of the candidate participants and the sample identification set.

3. The method as described in claim 2, characterized in that, The number of candidate participants is multiple, and the step of selecting the second participant from the candidate participants based on the anonymous identification information of the candidate participants and the sample identification set includes: Obtain the sample feature type set for each candidate participant from the participant search response; The second participant is selected from the multiple candidate participants based on their respective sample identifier sets and sample feature type sets.

4. The method as described in claim 3, characterized in that, The step of selecting the second participant from the multiple candidate participants based on their respective sample identifier sets and sample feature type sets includes: Obtain a set of reference sample identifiers for selecting the second reference party, and select multiple target reference parties from the multiple candidate reference parties whose corresponding sample identifier sets match the set of reference sample identifiers; Find the overlapping sample feature types between the sample feature type set of each target reference party and the sample feature type set of other target reference parties, and select the second reference party from the plurality of target reference parties based on the found overlapping sample feature types; wherein, the number of overlapping sample feature types corresponding to each target reference party is negatively correlated with the probability of each target reference party being selected as the second reference party.

5. The method as described in claim 1, characterized in that, If the alignment sample identifier set is detected to meet the training requirements, then a confirmation message is sent to the NEF network element, including: The system receives sample feature type sets corresponding to multiple second participants sent by the NEF network element, and merges the received sample feature type sets to obtain a merged sample feature type set. If the number of sample identifiers in the aligned sample identifier set is greater than the sample number threshold, and the number of sample feature types in the merged sample feature type set is greater than the feature type number threshold, then the training requirement information is satisfied, and the confirmation information is sent to the NEF network element.

6. The method as described in claim 1, characterized in that, The method further includes: If it is detected that the aligned sample identifier set does not meet the training requirement information, the sample identifiers contained in the expected sample identifier set are updated to obtain the updated expected sample identifier set. The first sample alignment request is regenerated based on the updated expected sample identifier set, and the regenerated first sample alignment request is sent to the NEF network element.

7. A data processing method for federated learning, characterized in that, A method applied to a vertical federated learning system comprising a first participant and a second participant, wherein the first participant is one of an AF network element and an NWDAF network element, and the second participant is the other of an AF network element and an NWDAF network element, the method comprising: Obtain the desired sample identifier set of the first participant from the first sample alignment request sent by the first participant, and send a second sample alignment request to the second participant according to the first sample alignment request; Receive the sample identifier set of the second participant sent by the second participant according to the second sample alignment request, and perform sample alignment between the expected sample identifier set and the sample identifier set of the second participant to obtain the aligned sample identifier set; The alignment sample identifier set is sent to the first participant, so that the first participant generates confirmation information after detecting that the alignment sample identifier set meets the training requirements. The system receives the confirmation information sent by the first participant and sends the alignment sample identifier set to the second participant based on the confirmation information.

8. The method as described in claim 7, characterized in that, The number of the second participants is multiple, and the method further includes: Obtain the sample feature type set for each second participant; The sample feature type sets corresponding to multiple second participants are sent to the first participant, so that the first participant merges the sample feature type sets corresponding to the multiple second participants to obtain a merged sample feature type set, and generates the confirmation information after detecting that the aligned sample identifier set and the merged sample feature type set meet the training requirement information.

9. The method as described in claim 7, characterized in that, The number of second participants is multiple; the step of sending a second sample alignment request to the second participants according to the first sample alignment request includes: The second sample alignment request is generated based on the expected sample identifier set, and the second sample alignment request is sent to each second participant, so that each second participant finds a sample that matches the sample identifier contained in the expected sample identifier set, and constructs a sample identifier set for each second participant based on the sample identifier of the found sample; The step of aligning the desired sample identifier set with the sample identifier set of the second participant to obtain an aligned sample identifier set includes: Find the intersection between the sample identifier sets corresponding to multiple second participants, and generate the aligned sample identifier set based on the found intersection.

10. The method as described in claim 9, characterized in that, The step of generating the second sample alignment request based on the desired sample identifier set includes: Each sample identifier contained in the desired sample identifier set is mapped from a first identifier space to a second identifier space to obtain the mapped desired sample identifier set; wherein, the first identifier space is the identifier space of the sample in the first participant, and the second identifier space is the identifier space of the sample in the second participant; Generate a second sample alignment request containing the mapped desired sample identifier set; Sending the aligned sample identifier set to the first participant includes: Each sample identifier contained in the alignment sample identifier set is mapped from the second identifier space to the first identifier space to obtain the mapped alignment sample identifier set, and the mapped alignment sample identifier set is sent to the first participant.

11. The method as described in claim 7, characterized in that, Before obtaining the desired sample identifier set of the first participant from the first sample alignment request sent by the first participant, the method further includes: The system receives a participant lookup request sent by the first participant and, based on the participant lookup request, requests the Network Storage Function (NRF) network element to look up candidate participants that support the vertical federated learning service; the network element type of the candidate participant matches the network element type of the second participant. Based on the identification information of the candidate participants sent by the NRF network element, the user data management UDM network element is requested to search for the sample identification set of the candidate participants; The identification information of the candidate participants is anonymized to obtain the anonymous identification information of the candidate participants, and a participant search response is generated based on the anonymous identification information of the candidate participants and the sample identification set. The participant lookup response is sent to the first participant, so that the first participant selects the second participant from the candidate participants based on the anonymous identifier information of the candidate participants and the sample identifier set.

12. A data processing method for federated learning, characterized in that, An application is made to a vertical federated learning system comprising a first participant and a second participant, wherein the first participant is one of an AF network element and an NWDAF network element, and the second participant is the other of an AF network element and an NWDAF network element, wherein the method is executed by the second participant, and the method includes: Receive a second sample alignment request sent by a NEF network element; wherein the second sample alignment request is generated by the NEF network element based on the first sample alignment request sent by the first participant; Based on the second sample alignment request, the sample identifier is found to obtain the sample identifier set of the second participant; The sample identifier set of the second participant is sent to the NEF network element, so that the NEF network element performs sample alignment on the sample identifier set of the second participant with the expected sample identifier set contained in the first sample alignment request to obtain an aligned sample identifier set, and sends the aligned sample identifier set to the first participant; The first participant receives the alignment sample identifier set sent by the NEF network element based on the confirmation information, wherein the confirmation information is generated after the first participant detects that the alignment sample identifier set meets the training requirements.

13. A data processing apparatus for federated learning, characterized in that, An apparatus is applied to a vertical federated learning system comprising a first participant and a second participant, wherein the first participant is one of an AF network element and an NWDAF network element, and the second participant is the other of an AF network element and an NWDAF network element. The apparatus is configured within the first participant and comprises: The request module is configured to send a first sample alignment request to the Network Open Function (NEF) network element, so that the NEF network element can obtain the expected sample identifier set of the first participant from the first sample alignment request, and send a second sample alignment request to the second participant according to the first sample alignment request. The receiving module is configured to receive the alignment sample identifier set sent by the NEF network element; wherein the alignment sample identifier set is obtained by the NEF network element by performing sample alignment on the expected sample identifier set and the sample identifier set of the second participant, and the sample identifier set of the second participant is sent to the NEF network element by the second participant according to the second sample alignment request; The sending module is configured to send confirmation information to the NEF network element if it detects that the alignment sample identifier set meets the training requirements, so that the NEF network element sends the alignment sample identifier set to the second participant according to the confirmation information.

14. A data processing apparatus for federated learning, characterized in that, An apparatus for use in a vertical federated learning system comprising a first participant and a second participant, wherein the first participant is one of an AF network element and an NWDAF network element, and the second participant is the other of an AF network element and an NWDAF network element, the apparatus comprising: The receiving module is configured to obtain the expected sample identifier set of the first participant from the first sample alignment request sent by the first participant, and send a second sample alignment request to the second participant according to the first sample alignment request. The alignment module is configured to receive the sample identifier set of the second participant sent by the second participant according to the second sample alignment request, and to perform sample alignment between the expected sample identifier set and the sample identifier set of the second participant to obtain an aligned sample identifier set; The sending module is configured to send the alignment sample identifier set to the first participant, so that the first participant generates confirmation information after detecting that the alignment sample identifier set meets the training requirements. The notification module is configured to receive the confirmation information sent by the first participant and send the alignment sample identifier set to the second participant according to the confirmation information.

15. A data processing apparatus for federated learning, characterized in that, An apparatus is applied to a vertical federated learning system comprising a first participant and a second participant, wherein the first participant is one of an AF network element and an NWDAF network element, and the second participant is the other of an AF network element and an NWDAF network element. The apparatus is configured within the second participant and comprises: The receiving module is configured to receive a second sample alignment request sent by a NEF network element; wherein the second sample alignment request is generated by the NEF network element based on a first sample alignment request sent by the first participant. The search module is configured to search for sample identifiers based on the second sample alignment request to obtain the sample identifier set of the second participant; The sending module is configured to send the sample identifier set of the second participant to the NEF network element, so that the NEF network element performs sample alignment on the sample identifier set of the second participant with the expected sample identifier set contained in the first sample alignment request to obtain an aligned sample identifier set, and sends the aligned sample identifier set to the first participant; The acquisition module is configured to receive the alignment sample identifier set sent by the NEF network element according to the confirmation information, wherein the confirmation information is generated by the first participant after detecting that the alignment sample identifier set meets the training requirements.

16. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more computer programs that, when executed by one or more processors, cause the electronic device to perform the method of any one of claims 1-12.

17. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the processor of the electronic device, causes the electronic device to perform the method described in any one of claims 1-12.

18. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-12.

Images