CN118277783A

CN118277783A - Model training processing method and device and related equipment

Info

Publication number: CN118277783A
Application number: CN202211727439.5A
Authority: CN
Inventors: 周通; 孙鹏; 施源; 吴昊; 崇卫微
Original assignee: Vivo Mobile Communication Co Ltd
Current assignee: Vivo Mobile Communication Co Ltd
Priority date: 2022-12-29
Filing date: 2022-12-29
Publication date: 2024-07-02
Also published as: WO2024140510A1

Abstract

The application discloses a model training processing method, a device and related equipment, belonging to the technical field of communication, wherein the model training processing method of the embodiment of the application comprises the following steps: the first device determines first target information according to the first sample data set; the first sample data set includes second target information, tag data, and second output information determined based on a second output of the second model, the second target information includes base station sensitive information or first output information determined based on a first output of the first model, an input of the first model is determined based on the base station sensitive information, the input of the second model is terminal sensitive information, the first target information includes a first bias and a second bias, or the first target information includes a second bias, the first bias is a back propagation bias corresponding to the first output, the first bias is used for parameter update of the first model, the second bias is a back propagation bias corresponding to the second output, and the second bias is used for parameter update of the second model.

Description

Model training processing method and device and related equipment

Technical Field

The application belongs to the technical field of communication, and particularly relates to a model training processing method, device and related equipment.

Background

In a mobile communication system, there are increasing use cases in combination with artificial intelligence (ARTIFICIAL INTELLIGENCE, AI). For example, at the physical layer, there is AI-based Channel State Information (CSI) feedback compression, AI-based beam management, AI-based positioning.

Currently, in AI-based beam management or beam prediction, neither the base station nor the UE want to expose the respective beams and antenna sensitive information from a privacy perspective. Detailed information of the transmission beam and the reception beam cannot be obtained at the time of data collection, resulting in lower accuracy of beam prediction based on the model. Therefore, the problem of low accuracy of the AI model exists in the prior art.

Disclosure of Invention

The embodiment of the application provides a model training processing method, a model training processing device and related equipment, which can solve the problem of low accuracy of an AI model.

In a first aspect, a model training processing method is provided, including:

The first device determines first target information according to the first sample data set;

The first sample data set comprises second target information, label data and second output information determined based on second output of a second model, the second target information comprises base station sensitive information or first output information determined based on first output of a first model, input of the first model is determined based on the base station sensitive information, input of the second model is terminal sensitive information, the first target information comprises first deviation and second deviation, or the first target information comprises second deviation, the first deviation is reverse propagation deviation corresponding to the first output, the first deviation is used for carrying out parameter updating on the first model, the second deviation is reverse propagation deviation corresponding to the second output, and the second deviation is used for carrying out parameter updating on the second model.

In a second aspect, a model training processing method is provided, including:

the terminal performs a first operation, the first operation including any one of:

Determining second output of a second model according to first information received from first equipment, and sending second output information to the first equipment, wherein the second output information is the output information of the second model or is information of the first equipment for post-processing the output information of the second model based on post-processing configuration, the first information comprises first indication information and/or training progress indication, the first indication information is used for indicating a target sample identification set, and the input of the second model is first input information;

Transmitting terminal sensitive information to a second device, wherein the terminal sensitive information is used for determining second output information which is transmitted to a first device and is determined based on second output of a second model, the input of the second model is first input information, and the second output information is output information of the second model or information for the second device to post-process the output information of the second model based on post-processing configuration;

In the case that the second model includes a third sub-model trained on the second device and a fourth sub-model trained on the first device, sending terminal-sensitive information to the second device, where the terminal-sensitive information is used for the second device to send fourth output information to the first device, where the fourth output information is output information of the third sub-model or information obtained by the second device performing post-processing on the output information of the third sub-model based on post-processing configuration, where the fourth output information is used for the first device to input the fourth sub-model, where the output information of the fourth sub-model is second output information, and where the input of the third sub-model is first input information;

In the case that the second model comprises a third sub-model trained on the second device, a fourth sub-model trained on the first device and a fifth sub-model trained on the terminal, taking first input information as input of the fifth sub-model to obtain output information of the fifth sub-model, and sending fifth output information to the second device, wherein the fifth output information is the output information of the fifth sub-model or is obtained by post-processing the output information of the fifth sub-model by the terminal based on post-processing configuration, and the fifth output information is used as input of the fourth sub-model by the first device, and the output information of the fourth sub-model is second output information;

The first input information is terminal sensitive information associated with the target sample identification set or information obtained by preprocessing terminal sensitive information associated with the target sample identification set based on preprocessing configuration, the second output information is second output of the second model or information obtained by post-processing the second output of the second model based on post-processing configuration, the second output information is used for determining first target information by first equipment according to a first sample data set, and the first sample data set comprises second target information, tag data and the second output information; the second target information comprises base station sensitive information or first output information determined based on first output of a first model, the input of the first model is determined based on the base station sensitive information, the first target information comprises first deviation and second deviation, or the first target information comprises second deviation, the first deviation is a back propagation deviation corresponding to the first output, the first deviation is used for updating parameters of the first model, the second deviation is a back propagation deviation corresponding to the second output, and the second deviation is used for updating parameters of the second model; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

In a third aspect, a model training processing method is provided, including:

the base station receives second information from the first equipment, wherein the second information comprises first indication information and/or training progress indication, and the first indication information is used for indicating a target sample identification set;

The base station sends third output information to first equipment, wherein the third output information is used for the first equipment to obtain first output as input of a first submodel, and the first output is used for the first equipment to determine first target information according to a first sample data set;

The third output information comprises output information of a second sub-model or information of the base station for performing post-processing on the output information of the second sub-model based on post-processing configuration, wherein the input of the second sub-model is base station sensitive information associated with the target sample identification set or information of the base station for preprocessing the base station sensitive information associated with the target sample identification set based on preprocessing configuration; the first sample data set includes second target information, label data and second output information determined based on a second output of a second model, the second target information including the first output, an input of the second model being determined based on terminal sensitive information, the first target information including a first bias and a second bias, or the first target information including a second bias, the first bias being a back propagation bias corresponding to the first output and used for parameter updating a first sub-model and determining a third bias for parameter updating a second sub-model, the second bias being a back propagation bias corresponding to the second output and used for parameter updating the second model, a first model including the first sub-model trained on the first device and a second sub-model trained on the base station; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

In a fourth aspect, a model training processing method is provided, including:

The second device receives first information from the first device, wherein the first information comprises first indication information and/or training progress indication, and the first indication information is used for indicating a target sample identification set;

The second device performs a second operation according to the first information, the second operation including at least one of:

Acquiring terminal sensitive information and sending second output information to first equipment, wherein the second output information is output information of a second model or information obtained by the second equipment for carrying out post-processing on the output information of the second model based on post-processing configuration, and the input of the second model is first input information;

Acquiring terminal sensitive information and transmitting fourth output information to first equipment under the condition that a second model comprises a third sub-model trained on the second equipment and a fourth sub-model trained on the first equipment, wherein the fourth output information is output information of the third sub-model or information obtained by post-processing the output information of the third sub-model by the second equipment based on post-processing configuration, the fourth output information is used as input of the fourth sub-model by the first equipment, the output information of the fourth sub-model is second output information, and the input of the third sub-model is first input information;

In the case that the second model comprises a third sub-model trained on the second device, a fourth sub-model trained on the first device and a fifth sub-model trained on the terminal, receiving fifth output information from the terminal and sending the output information of the third sub-model to the first device, wherein the fifth output information is the output information of the fifth sub-model or is obtained by post-processing the output information of the fifth sub-model by the terminal based on post-processing configuration, the fifth output information is used as input of the fourth sub-model by the first device, the output information of the fourth sub-model is second output information, and the input of the third sub-model is first input information;

The first input information is terminal sensitive information associated with the target sample identification set or information obtained by preprocessing terminal sensitive information associated with the target sample identification set based on preprocessing configuration, the second output information is used for determining first target information according to a first sample data set by first equipment, the first sample data set comprises second target information, label data and the second output information, the second target information comprises base station sensitive information or first output information determined based on first output of a first model, the input of the first model is the base station sensitive information, the first target information comprises first deviation and second deviation, or the first target information comprises second deviation, the first deviation is a back propagation deviation corresponding to the first output, the first deviation is used for updating parameters of the first model, the second deviation is a back propagation deviation corresponding to the second output, and the second deviation is used for updating parameters of the second model; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

In a fifth aspect, there is provided a model training processing apparatus comprising:

the first determining module is used for determining first target information according to the first sample data set;

In a sixth aspect, there is provided a model training processing apparatus comprising:

A first execution module for executing a first operation, the first operation comprising any one of:

In a seventh aspect, there is provided a model training processing apparatus comprising:

A third receiving module, configured to receive second information from the first device, where the second information includes first indication information and/or a training progress indication, and the first indication information is used to indicate a target sample identifier set;

A third sending module, configured to send third output information to a first device, where the third output information is used for the first device to obtain a first output as an input of a first sub-model, and the first output is used for the first device to determine first target information according to a first sample data set;

The third output information comprises output information of a second sub-model or information of the base station which carries out post-processing on the output information of the second sub-model based on post-processing configuration, and the input of the second sub-model is base station sensitive information associated with the target sample identification set or information of the base station which carries out pre-processing on the base station sensitive information associated with the target sample identification set based on pre-processing configuration; the first sample data set includes second target information, label data and second output information determined based on a second output of a second model, the second target information including the first output, an input of the second model being determined based on terminal sensitive information, the first target information including a first bias and a second bias, or the first target information including a second bias, the first bias being a back propagation bias corresponding to the first output and used for parameter updating a first sub-model and determining a third bias for parameter updating a second sub-model, the second bias being a back propagation bias corresponding to the second output and used for parameter updating the second model, a first model including the first sub-model trained on the first device and a second sub-model trained on the base station; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

In an eighth aspect, there is provided a model training processing apparatus comprising:

A fourth receiving module, configured to receive first information from a first device, where the first information includes first indication information and/or a training progress indication, and the first indication information is used to indicate a target sample identifier set;

A second execution module, configured to execute a second operation according to the first information, where the second operation includes at least one of:

Acquiring terminal sensitive information and sending second output information to first equipment, wherein the second output information is output information of a second model or information of the second equipment for carrying out post-processing on the output information of the second model based on post-processing configuration, and the input of the second model is first input information;

In a ninth aspect, there is provided a terminal comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the method as described in the second aspect.

In a tenth aspect, a terminal is provided, including a processor and a communication interface, where the processor is configured to perform a first operation, where the first operation includes any one of:

In an eleventh aspect, there is provided a network side device comprising a processor and a memory storing a program or instructions executable on the processor, the program or instructions implementing the steps of the method according to the first aspect or the steps of the method according to the third aspect when executed by the processor.

In a twelfth aspect, a network-side device is provided, including a processor and a communication interface, where,

When the network side equipment is first equipment, the processor is used for determining first target information according to a first sample data set;

When the network side equipment is a base station, the communication interface is used for receiving second information from first equipment, the second information comprises first indication information and/or training progress indication, and the first indication information is used for indicating a target sample identification set; transmitting third output information to a first device, wherein the third output information is used for the first device to obtain a first output as an input of a first sub-model, and the first output is used for the first device to determine first target information according to a first sample data set;

In a thirteenth aspect, a third party server is provided, including a processor and a communication interface, where the communication interface is configured to receive first information from a first device, the first information including first indication information and/or a training progress indication, the first indication information being configured to indicate a target sample identification set;

the processor is configured to perform a second operation according to the first information, where the second operation includes at least one of:

In a fourteenth aspect, there is provided a communication system comprising: the terminal may be configured to perform the steps of the model training processing method according to the second aspect, and the network side device may be configured to perform the steps of the model training processing method according to the first aspect, the steps of the model training processing method according to the third aspect, or the steps of the model training processing method according to the fourth aspect.

A fifteenth aspect provides a readable storage medium having stored thereon a program or instructions which when executed by a processor, performs the steps of the method as described in the first aspect, or performs the steps of the method as described in the second aspect, or performs the steps of the method as described in the third aspect, or performs the steps of the method as described in the fourth aspect.

In a sixteenth aspect, there is provided a chip comprising a processor and a communication interface, the communication interface and the processor being coupled, the processor being adapted to run a program or instructions, to perform the steps of the method as described in the first aspect, or to perform the steps of the method as described in the second aspect, or to perform the steps of the method as described in the third aspect, or to perform the steps of the method as described in the fourth aspect.

In a seventeenth aspect, there is provided a computer program/program product stored in a storage medium, the computer program/program product being executable by at least one processor to perform the steps of the method according to the first aspect, or to perform the steps of the method according to the second aspect, or to perform the steps of the method according to the third aspect, or to perform the steps of the method according to the fourth aspect.

According to the embodiment of the application, the second output information is determined through the second output of the feature extraction of the terminal sensitive information based on the second model, and the first target information is determined by adopting the second output information, so that the parameter of the model can be updated by utilizing the first target information. Therefore, the embodiment of the application can apply the terminal sensitive information to the model training process while protecting the terminal sensitive information, thereby improving the reliability of the model training of the terminal and the network side equipment and improving the accuracy of the trained model.

Drawings

FIG. 1 is a schematic diagram of a network architecture to which embodiments of the present application are applicable;

FIG. 2 is a schematic flow chart of a model training processing method according to an embodiment of the present application;

FIG. 3 is a diagram of a network scenario in which a model training processing method according to an embodiment of the present application may be applied;

FIG. 4 is a second network scenario diagram to which a model training processing method according to an embodiment of the present application may be applied;

FIG. 5 is a third network scenario diagram to which a model training processing method according to an embodiment of the present application may be applied;

FIG. 6 is a second flow chart of a model training method according to the embodiment of the application;

FIG. 7 is a third flow chart of a model training method according to the embodiment of the application;

FIG. 8 is a flowchart of a model training processing method according to an embodiment of the present application;

FIG. 9 is a flowchart of a model training processing method according to an embodiment of the present application;

FIG. 10 is a flowchart of a model training processing method according to an embodiment of the present application;

FIG. 11 is a flowchart of a model training processing method according to an embodiment of the present application;

FIG. 12 is a block diagram of a model training processing device according to an embodiment of the present application;

FIG. 13 is a second block diagram of a model training processing device according to an embodiment of the present application;

FIG. 14 is a third block diagram of a model training processing device according to an embodiment of the present application;

FIG. 15 is a diagram showing a construction of a model training processing device according to an embodiment of the present application;

fig. 16 is a block diagram of a communication device according to an embodiment of the present application;

fig. 17 is a block diagram of a terminal according to an embodiment of the present application;

fig. 18 is a block diagram of a network side device according to an embodiment of the present application.

Detailed Description

The technical solutions of the embodiments of the present application will be clearly described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which are derived by a person skilled in the art based on the embodiments of the application, fall within the scope of protection of the application.

The terms first, second and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the application are capable of operation in sequences other than those illustrated or otherwise described herein, and that the "first" and "second" distinguishing between objects generally are not limited in number to the extent that the first object may, for example, be one or more. Furthermore, in the description and claims, "and/or" means at least one of the connected objects, and the character "/" generally means a relationship in which the associated object is an "or" before and after.

The term "indicated" in the description and claims of the present application may be either an explicit indication or an implicit indication. The explicit indication may be understood as that the sender explicitly informs the receiver of the operation or request result that needs to be performed in the sent indication; the implicit indication is understood as that the receiving side judges according to the indication sent by the sending side, and determines the operation or the request result to be executed according to the judging result.

It should be noted that the techniques described in the embodiments of the present application are not limited to long term evolution (Long Term Evolution, LTE)/LTE evolution (LTE-Advanced, LTE-a) systems, but may also be used in other wireless communication systems, such as code division multiple access (Code Division Multiple Access, CDMA), time division multiple access (Time Division Multiple Access, TDMA), frequency division multiple access (Frequency Division Multiple Access, FDMA), orthogonal frequency division multiple access (Orthogonal Frequency Division Multiple Access, OFDMA), single carrier frequency division multiple access (Single-carrier Frequency Division Multiple Access, SC-FDMA), and other systems. The terms "system" and "network" in embodiments of the application are often used interchangeably, and the techniques described may be used for both the above-mentioned systems and radio technologies, as well as other systems and radio technologies. The following description describes a New Radio (NR) system for exemplary purposes and NR terminology is used in much of the following description, but these techniques may also be applied to applications other than NR system applications, such as 6 th Generation (6G) communication systems.

Fig. 1 shows a block diagram of a wireless communication system to which an embodiment of the present application is applicable. The wireless communication system includes a terminal 11 and a network device 12. The terminal 11 may be a Mobile phone, a tablet Computer (Tablet Personal Computer), a Laptop (Laptop Computer) or a Personal digital assistant (Personal DIGITAL ASSISTANT, PDA), a palm Computer, a netbook, an ultra-Mobile Personal Computer (ultra-Mobile Personal Computer, UMPC), a Mobile internet device (Mobile INTERNET DEVICE, MID), a Mobile terminal, augmented reality (augmented reality, AR)/Virtual Reality (VR) equipment, robots, wearable equipment (Wearable Device), vehicle-mounted equipment (VUE), pedestrian terminals (PUE), smart home (home equipment with wireless communication function, such as refrigerators, televisions, washing machines or furniture), game machines, personal computers (personal computer, PCs), teller machines or self-service machines, and other terminal side equipment, and the wearable equipment includes: intelligent watch, intelligent bracelet, Intelligent headphones, intelligent glasses, intelligent jewelry (intelligent bracelets, intelligent rings, intelligent necklaces, intelligent ankles, intelligent footchains, etc.), intelligent bracelets, intelligent clothing, etc. It should be noted that the specific type of the terminal 11 is not limited in the embodiment of the present application. The network-side device 12 may include an access network device or a core network device, where the access network device may also be referred to as a radio access network device, a radio access network (Radio Access Network, RAN), a radio access network function, or a radio access network element. The access network devices may include base stations, WLAN access points, wiFi nodes, etc., which may be referred to as node bs, evolved node bs (enbs), access points, base transceiver stations (Base Transceiver Station, BTSs), radio base stations, radio transceivers, basic SERVICE SET, BSS, extended SERVICE SET, ESS sets, home node bs, home evolved node bs, transmit and receive points (TRANSMITTING RECEIVING points, TRP) or some other suitable term in the field, the base station is not limited to a specific technical vocabulary as long as the same technical effect is achieved, and it should be noted that in the embodiment of the present application, only the base station in the NR system is described as an example, and the specific type of the base station is not limited. The core network device may include, but is not limited to, at least one of: a core network node, a core network function, a Mobility management entity (Mobility MANAGEMENT ENTITY, MME), an access Mobility management function (ACCESS AND Mobility Management Function, AMF), a session management function (Session Management Function, SMF), a user plane function (User Plane Function, UPF), a policy control function (Policy Control Function, PCF), policy AND CHARGING Rules Function (PCRF), edge application service discovery Function (Edge Application Server Discovery Function, EASDF), unified data management (Unified DATA MANAGEMENT, UDM), unified data repository (Unified Data Repository, UDR), home subscriber server (Home Subscriber Server, HSS), centralized network configuration (Centralized network configuration, CNC), network storage functions (Network Repository Function, NRF), network open functions (Network Exposure Function, NEF), local NEF (or L-NEF), binding support functions (Binding Support Function, BSF), application functions (Application Function, AF), and the like. it should be noted that, in the embodiment of the present application, only the core network device in the NR system is described as an example, and the specific type of the core network device is not limited.

The model training processing method provided by the embodiment of the application is described in detail below through some embodiments and application scenes thereof with reference to the accompanying drawings.

Referring to fig. 2, an embodiment of the present application provides a model training processing method, as shown in fig. 2, including:

Step 201, the first device determines first target information according to the first sample data set;

In the embodiment of the present application, the first device may be referred to as a Network side device, for example, may be a base station or a Network DATA ANALYTICS Function (NWDAF).

Alternatively, the above-described back propagation bias may be referred to as a loss value. The second output information may be the second output of the second model, or information obtained by post-processing the second output of the second model.

Optionally, the content specifically included in the terminal sensitive information and the base station sensitive information may be set according to actual needs, for example, in some embodiments, the terminal sensitive information may include beam information and/or antenna information of the terminal, and the base station sensitive information may include beam information and/or antenna information of the base station.

It should be understood that the first model may be understood as an AI model for feature extraction of base station sensitive information, and the second model may be understood as an AI model for feature extraction of terminal sensitive information.

Optionally, in an embodiment of the present application, the model deployment scenario may include the following scenarios:

Scene 1: as shown in fig. 3, the terminal-side second model, the base station-side first model, and the base station-side third model are included, and the third model is used for an AI model for prediction based on the output of the second model and the output of the first model, and specifically may include a multi-layer neural network.

Scene 2: as shown in fig. 4, the third network model is degraded to be an addition operation, that is, includes the second model on the terminal side, the first model on the base station side, and the addition operation.

Scene 3: as shown in fig. 5, the first model, that is, the second model including the terminal side and the third model including the base station side, is canceled.

It should be noted that, the base station sensitive information may be directly applied to the model training process, or may be applied to the model training process after feature extraction of the first sensitive information by using the first model. The model training process is applied to obtain a final output result, such as a prediction result, by taking the second target information and the second output information as input of a subsequent third model or addition operation.

According to the embodiment of the application, the second output information is determined through the second output of the feature extraction of the terminal sensitive information based on the second model, and the first target information is determined by adopting the second output information, so that the parameter of the model can be updated by utilizing the first target information. Therefore, the embodiment of the application can apply the terminal sensitive information to the model training process while protecting the terminal sensitive information from being acquired by the network equipment, thereby improving the reliability of the model training of the terminal and the network side equipment and improving the accuracy of the trained model.

Optionally, in some embodiments, before the first device determines the first target information from the first sample data set, the method further comprises:

The first device determining a target sample identification set, the target sample identification set being used to determine the first sample data set;

The first device sends first information to the terminal or the second device, the first information comprises first indication information and/or training progress indication, the first indication information is used for indicating the target sample identification set, and the second device is a third-party server.

In the embodiment of the present application, the first sample data set may be understood as a set of sample data included in a batch in performing one iteration training, and the target sample identifier set may be understood as a set including all sample identifiers corresponding to the sample data. The first device may send first information to the terminal or the second device upon determining that the target sample identification set to collect second output information associated with the target sample identification set.

Alternatively, the training progress indication described above may be used to indicate the number of times the training is currently iterated.

Optionally, in some embodiments, the first model satisfies any one of:

the first model is trained on the first device;

the first model includes a first sub-model trained on the first device and a second sub-model trained on a base station.

In the embodiment of the present application, the training of the first model on the first device may be understood as that the first model is not split and trained, and the first model is only trained on the first device. The first model comprising a first sub-model trained on the first device and a second sub-model trained on the base station may be understood as a split training of the first model.

Alternatively, the first sub-model may be understood as a first model global component, and the second sub-model may be referred to as a first model vendor component or a first model personalization component.

It should be appreciated that when the first model is split trained, the first device described above may be understood as a core network function, such as NWDAF.

Optionally, in some embodiments, after the first device determines the first target information from the first sample data set, the method further comprises:

in the case that the first model is trained on the first device, the first device performs parameter updating on the first model according to the first deviation;

In the case that the first model includes the first sub-model and the second sub-model, the first device performs parameter updating on the first sub-model according to a first deviation, determines a third deviation, and sends the third deviation to the base station, wherein the third deviation is used for performing parameter updating on the second sub-model.

In the embodiment of the present application, the above parameter update may be understood as updating a parameter in a model.

Optionally, in some embodiments, where the first model includes the first sub-model and the second sub-model, the first device determines a set of target sample identifications, the method further comprises:

the first equipment sends second information to the base station, wherein the second information comprises the first indication information and/or the training progress indication;

the first device receives third output information from the base station, wherein the third output information is the output information of the second sub-model or is information obtained by the base station through post-processing the output information of the second sub-model based on post-processing configuration, and the post-processing configuration comprises sparse processing configuration and/or privacy processing configuration;

The first device obtains the first output by taking the third output information as the input of the first sub-model;

The first output information is the first output, the input of the second sub-model is base station sensitive information associated with the target sample identification set or information obtained after the base station performs preprocessing on the base station sensitive information associated with the target sample identification set based on preprocessing configuration, and the preprocessing configuration comprises at least one of the following steps: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

It should be noted that, the preprocessing process is specifically determined according to a corresponding preprocessing configuration, for example, when the preprocessing configuration includes a single-hot code dictionary configuration, the preprocessing process includes performing single-hot code processing on the sensitive information; when the data normalization parameter is configured, the preprocessing process comprises the steps of executing normalization processing on sensitive information; when the preprocessing configuration comprises data regularization parameter configuration, the preprocessing process comprises executing regularization processing on the sensitive information; when the preprocessing configuration includes a data normalization parameter configuration, the preprocessing process includes performing normalization processing on the sensitive information.

Likewise, the processing procedure in the post-processing is specifically determined according to the corresponding post-processing configuration, for example, when the post-processing configuration includes a thinning processing configuration, the post-processing procedure includes performing a thinning operation on the output information of the model; when the post-processing configuration comprises a privacy processing configuration, the post-processing procedure comprises performing a privacy operation on the output information of the model.

Optionally, in some embodiments, the method further comprises:

the first device sends a preprocessing instruction and/or a post-processing instruction to the base station, wherein the preprocessing instruction is used for indicating the preprocessing configuration, and the post-processing instruction is used for indicating the post-processing configuration.

In the embodiment of the present application, the preprocessing instruction and/or the post-processing instruction may be carried by the second information carrier, or may be carried by other information carriers, which is not limited herein.

Optionally, after the first device determines the target sample identification set, the method further includes:

the first device determining a second set of sample data, the second set of sample data comprising the base station sensitive information, and the second set of sample data being associated with the set of target sample identifications;

The first device inputs the second sample data set into the first model for forward propagation to obtain the first output, and the first output information is the first output.

In the embodiment of the present application, after the first device determines the target sample identifier set, the first device may determine the second sample data set based on the target sample identifier set, that is, obtain the second sample data associated with each sample identifier in the target sample identifier set, so as to obtain the second sample data set. Wherein the second sample data comprises the base station sensitive information.

Optionally, after the second sample data set is determined, the second sample data set may be directly used as an input of the first model to obtain the first output, or the preprocessing configuration may be adopted to process the second sample data set, and the preprocessed information may be used as an input of the first model to obtain the first output.

Optionally, in some embodiments, the second model satisfies any one of the following:

the second model is trained on a terminal or a second device;

the second model includes a third sub-model trained on a second device and a fourth sub-model trained on the first device;

the second model includes a third sub-model trained on a second device, a fourth sub-model trained on the first device, and a fifth sub-model trained on a terminal;

the second device is a third-party server.

In the embodiment of the present application, the training of the second model on the terminal or the second device may be understood as that the second model is not split-trained. In the case that the second model comprises a third sub-model and a fourth sub-model, or comprises a third sub-model, a fourth sub-model and a fifth sub-model, it is understood that the first model is split-trained.

Alternatively, the third party server may be understood as a service device of the terminal or a service device of a terminal chip manufacturer.

Alternatively, the third sub-model may be understood as a second model global component, the fourth sub-model may be referred to as a second model vendor component or a second model personalization component, and the fifth sub-model may be understood as a second model local component.

Optionally, after the first device determines the first target information according to the first sample data set, the method further comprises at least one of:

In the case of training the second model on the terminal or the second device, the first device sends third information and the second deviation to the terminal or the second device, the third information including a training progress indication;

In the case that the second model comprises the third sub-model and the fourth sub-model, the first device performs parameter updating on the fourth sub-model according to the second deviation, determines a fourth deviation, and sends the fourth deviation to a second device, wherein the fourth deviation is used for the second device to perform parameter updating on the third sub-model;

In the case that the second model includes the third sub-model, the fourth sub-model and the fifth sub-model, the first device performs parameter updating on the fourth sub-model and determines a fourth deviation according to the second deviation, and sends the fourth deviation to a second device, where the fourth deviation is used for the second device to perform parameter updating on the third sub-model and determine a fifth deviation, and the fifth deviation is used for the terminal to perform parameter updating on the fifth sub-model.

Optionally, after the first device sends the first information to the terminal or the second device, the method further includes:

The first device receives fourth information from the terminal or the second device, the fourth information satisfying at least one of:

The fourth information comprises the second output information under the condition that the second model is trained on the terminal, the input of the second model is the first input information, and the second output information is the second output of the second model or information obtained by post-processing the second output of the second model based on post-processing configuration;

In the case of training the second model on the second device, the fourth information includes the second output information, where the input of the second model is the first input information, and the second output information is the second output of the second model or information obtained by post-processing the second output of the second model based on a post-processing configuration;

In the case that the second model includes the third sub-model and the fourth sub-model, the fourth information includes fourth output information, where the fourth output information is used as input of the fourth sub-model to obtain the second output, and the fourth output information is output information of the third sub-model or is information obtained by the second device performing post-processing on the output information of the third sub-model based on post-processing configuration, and the input of the third sub-model is first input information, and the second output information is output information of the fourth sub-model;

In the case that the second model includes the third sub-model, the fourth sub-model, and the fifth sub-model, the fourth information includes output information of the third sub-model, the output information of the third sub-model is used as input of the fourth sub-model to obtain the second output, the input of the third sub-model is fifth output information sent by the terminal, the fifth output information is output information of the fifth sub-model or information obtained by post-processing the output information of the fifth sub-model by the terminal based on post-processing configuration, the input of the fifth sub-model is first input information, and the second output information is output information of the fourth sub-model;

The first input information is terminal sensitive information associated with the target sample identification set or information obtained by preprocessing the terminal sensitive information associated with the target sample identification set based on preprocessing configuration; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

In the embodiment of the application, the first device receives fourth information from the terminal under the condition that the second model is trained on the terminal, and the first device receives fourth information from the second device under the condition that the second model is trained on the second device or the second model is split to be trained.

Optionally, in some embodiments, the fourth information further includes second indication information and/or a training progress indication, the second indication information being used to indicate a sample identification set associated with the fourth information; the first device determining first target information from the first sample data set includes:

in the case that the sample identification set indicated by the second indication information is the same as the target sample identification set, the first device determines first target information according to the first sample data set.

In the embodiment of the application, whether the sample identification set associated with the second output information or the fourth output information which is currently transmitted is the same as the target sample identification set or not can be further checked through the second indication information, and the subsequent operation is only executed under the condition that the sample identification set and the target sample identification set are the same.

Optionally, in some embodiments, the second indication information is used to indicate at least one of:

Identification of all sample data;

collection time stamps for all sample data;

a starting sample identity and a total sample number;

A start sample identity and a stop sample identity;

bitmap information of the sample data;

a start collection time stamp for the sample data and a stop collection time stamp for the sample data.

In the embodiment of the present application, the sample data associated with the second indication information may be understood as terminal sensitive information, and the collection timestamp may be understood as a collection time of the terminal sensitive information or a time of generating the sample data, for example, when the terminal performs measurement, corresponding sensitive information is recorded, so as to obtain the sample data.

Optionally, in some embodiments, the method further comprises:

The first device sends a preprocessing instruction and/or a post-processing instruction to the terminal or the second device, wherein the preprocessing instruction is used for indicating the preprocessing configuration, and the post-processing instruction is used for indicating the post-processing configuration.

Optionally, in an embodiment of the present application, the preprocessing instruction and/or the post-processing instruction may be carried by the first information carrier, or may be carried by other information carriers, which is not limited herein.

Optionally, in some embodiments, the privacy processing configuration includes a privacy method and a parameter configuration associated with the privacy method.

In the embodiment of the present application, the preprocessing is to make different terminals or different second devices use the same coding dictionary. The single hot code dictionary configuration described above may be referred to as a unified code dictionary for unified mapping of inputs having the same physical meaning as inputs to the second model.

Optionally, the privacy method includes any one of the following: differential privacy, homomorphic encryption, and secret sharing.

Optionally, the differential privacy-associated parameter configuration includes at least one of a privacy mechanism and a differential privacy parameter configuration; wherein the differential privacy parameter configuration includes at least one of: privacy budgets; a slack term; clipping values or sensitivities; number of iteration rounds.

Alternatively, after a round of training, if the accuracy is low, it may be indicated to increase the privacy budget and decrease the noise addition.

Optionally, the differential privacy mechanism includes, but is not limited to, a laplace mechanism and a gaussian mechanism.

Optionally, in some embodiments, the third information further includes third indication information, the third indication indicating an optimizer function. The optimizer function shown includes at least one of:

A random gradient descent (Stochastic GRADIENT DESCENT, SGD) optimizer;

momentum (Momentum) optimizer;

A random gradient descent (SDG with Momentum) optimizer with momentum mechanism;

Root mean square transfer (Root Mean Square Prop, RMSprop) optimizer;

An adaptive moment estimation (adaptive moment estimatio, adam) optimizer;

An adaptive gradient (ADAPTIVE GRADIENT, ADAGRAD) optimizer.

Optionally, in some embodiments, the first sample data set further comprises beam quality measured by the terminal.

It should be noted that, for the spatial prediction scheme, the measured beam quality may be a subset of all beam qualities, or the measured beam may be a beam with a 3dB width that is wider, while the beam with a 3dB width that is narrower is to be predicted, where the measured beam quality is not a subset of all beam qualities. For a time domain prediction scheme, the measured beam quality may be a historically measured value and the tag data a future beam quality.

Optionally, the determining, by the first device, the first target information according to the first sample data set includes:

The first device inputting the second target information, the second output, and the measured beam quality to a third model;

The first device determines a first loss function according to the output information of the third model and the tag data;

the first device back-propagates updating parameters of the third model and obtaining the first target information based on the first loss function.

In the embodiment of the present application, the process of obtaining the first target information, including the third model, on the first device is a partial flow in the iterative training process performed for the above scenario 1 and scenario 3.

Optionally, in some embodiments, the first device determining the first target information from the first sample data set includes:

the first device performs addition operation on the second target information and the second output;

the first device determining a second loss function based on the output of the add operation and the tag data;

The first device performs back propagation based on the second loss function to obtain the first target information.

In the embodiment of the present application, the first device does not include the third model, that is, the process of obtaining the first target information is a partial flow in the iterative training process performed for the above scenario 2.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

the beam pair identity with the strongest beam quality among all the transceiving beam pairs.

In the embodiment of the present application, the above-mentioned transmission beam may be understood as a transmission beam of a base station, and the above-mentioned reception beam may be understood as a reception beam of a terminal. The pair of transmit and receive beams includes a transmit beam of the base station and a receive beam of the terminal.

For a better understanding of the present application, the following detailed description is given by way of some examples.

Embodiment 1, for the above scenario 1, the model training process is as shown in fig. 6, and specifically includes the following steps:

In step 601, the base station determines a sample identifier set 1, and sends information 1 to the terminal, where the information 1 includes a first sample identifier set indication and a training progress indication corresponding to the sample identifier set 1, and the first sample identifier set indication is used for aligning terminal sensitive information and base station sensitive information.

In step 602a, the terminal propagates forward through the second model terminal sensitive information associated with the first sample identifier set indication to obtain a second output of the second model, where the terminal sensitive information may include beam information and/or antenna information of the terminal.

In step 602b, the base station performs forward transmission on base station sensitive information associated with the second sample identifier set indication through the first model to obtain a first output of the first model, where the base station sensitive information may include beam information and/or antenna information of the base station.

In step 603, the terminal sends information 2 to the base station, where the information 2 may include a second output, a first sample identification set indication, and a training progress indication.

In step 604, the base station determines a first bias and a second bias based on the first output, the second output, the measured beam quality, and the tag data using a third model. For example, the base station takes the first output, the second output and the measured beam quality as a third model input, and obtains a loss function according to the tag data; the parameters of the third model are then back-propagated based on the obtained loss function to update and obtain a first bias and a second bias.

In step 605, the base station sends information 3 and a second deviation to the terminal, where the information 3 includes a training progress indication and may further include an optimizer function indication.

In step 606a, the terminal updates parameters of the second model based on the second deviation.

In step 606b, the base station updates parameters of the first model based on the first deviation.

In step 607, the terminal indicates to the network side device that parameter updating and training progress indication for the second model are completed.

Repeating the steps until the iteration is completed.

Embodiment 2, for scenario 1 above, the second model is trained by the server of the terminal, and deployed to the terminal after the training is completed. At this time, the UE may upload plaintext data (i.e., plaintext of the sensitive information), and the server of the terminal may pre-process the plaintext data of the terminal, and then pass through the second model, where the second output of the second model is post-processed and then sent to the NEF, and then forwarded to the first device (e.g., the core network function NWDAF). Compared to embodiment 1, in this embodiment, a preprocessing instruction and/or a post-processing instruction may be added to the interaction information.

Embodiment 3, for scenario 1, the second model is split and trained by both the server and NWDAF of the terminal, and deployed to the terminal after training. At this time, the UE may upload plaintext data (i.e., plaintext of the sensitive information), after the server of the terminal pre-processes the plaintext data of the terminal, the plaintext data passes through the second model manufacturer component, after the output information of the second model manufacturer component is post-processed, the output information is sent to the NWDAF, and the NWDAF forwards the received post-processed information to the NWDAF, and obtains the second output of the second model through the second model global component. Compared to embodiment 1, in this embodiment, a preprocessing instruction and/or a post-processing instruction may be added to the interaction information.

Embodiment 4, aiming at the scene 1, the second model is split and trained by the terminal, the server of the terminal and NWDAF, and deployed to the terminal after training. At this time, the terminal may preprocess the plaintext data (i.e., plaintext of the sensitive information) and then pass through the local component of the second model, and after post-processing the output of the model, the output is sent to the server of the terminal. The server of the terminal transmits the received post-processed information to the NEF through a second model manufacturer component, and then forwards the output of the second model manufacturer component to the NWDAF, and the NWDAF obtains the second output of the second model through a second model global component. Compared to embodiment 1, in this embodiment, a preprocessing instruction and/or a post-processing instruction may be added to the interaction information.

Embodiment 5, for scenario 1, the first model is split and trained by both base stations and NWDAF, and deployed to the base station after training. At this time, the base station may preprocess plaintext data (i.e., plaintext of sensitive data of the base station) and then pass through the first model manufacturer component, and after the output information of the first model manufacturer component is post-processed, the output information is sent to the NWDAF, where the NWDAF obtains the second output of the second model by passing through the first model global component. Compared to embodiment 1, in this embodiment, a preprocessing instruction and/or a post-processing instruction may be added to the interaction information.

Embodiment 6, for the above scenario 2, the model training process is as shown in fig. 7, and specifically includes the following steps:

In step 701, the base station determines a sample identifier set 1, and sends information 1 to the terminal, where the information 1 includes a first sample identifier set indication and a training progress indication corresponding to the sample identifier set 1, and the first sample identifier set indication is used for aligning terminal sensitive information and base station sensitive information.

In step 702a, the terminal propagates forward through the second model terminal sensitive information associated with the first sample identification set indication to obtain a second output of the second model, where the terminal sensitive information may include beam information and/or antenna information of the terminal.

In step 702b, the base station performs forward transmission on base station sensitive information associated with the second sample identifier set indication through the first model to obtain a first output of the first model, where the base station sensitive information may include beam information and/or antenna information of the base station.

In step 703, the terminal sends information 2 to the base station, where the information 2 may include a second output, a first sample identification set indication, and a training progress indication.

The base station determines a first bias and a second bias based on the first output, the second output, the measured beam quality, and the tag data using an addition operation, step 704. For example, the base station takes the first output, the second output and the measured beam quality as addition operation input, and then obtains a loss function according to the tag data; the back propagation is then performed based on the obtained loss function to obtain a first deviation and a second deviation.

Step 705, the base station sends information 3 and a second deviation to the terminal, where the information 3 includes a training progress indication, and may further include an optimizer function indication.

In step 706a, the terminal updates parameters of the second model based on the second deviation.

In step 706b, the base station updates parameters of the first model based on the first deviation.

In step 707, the terminal indicates to the network side device that parameter updating and training progress indication are completed for the second model.

Repeating the steps until the iteration is completed.

Embodiment 7, for the above scenario 3, the model training process is as shown in fig. 8, and specifically includes the following steps:

in step 801, the base station determines a sample identifier set 1, and sends information 1 to the terminal, where the information 1 includes a first sample identifier set indication and a training progress indication corresponding to the sample identifier set 1, where the first sample identifier set indication is used to align terminal sensitive information and base station sensitive information.

In step 802, the terminal propagates forward through the second model the terminal sensitive information associated with the first sample identifier set indication to obtain a second output of the second model, where the terminal sensitive information may include beam information and/or antenna information of the terminal.

In step 803, the terminal sends information 2 to the base station, where the information 2 may include a second output, a first sample identification set indication, and a training progress indication.

In step 804, the base station determines a second bias based on the base station sensitive information, the second output, the measured beam quality, and the tag data using a third model. For example, the base station uses the base station sensitive information, the second output and the measured beam quality as the input of a third model, and then obtains a loss function according to the tag data; and then back-propagating based on the obtained loss function to obtain a second deviation.

In step 805, the base station sends information 3 and a second deviation to the terminal, where the information 3 includes a training progress indication and may further include an optimizer function indication.

And step 806, the terminal updates parameters of the second model based on the second deviation.

In step 807, the terminal indicates to the network side device that parameter updating and training progress indication for the second model are completed.

Repeating the steps until the iteration is completed.

Referring to fig. 9, the embodiment of the present application further provides a model training processing method, as shown in fig. 9, where the model training processing method includes:

step 901, the terminal performs a first operation, where the first operation includes any one of the following:

Optionally, before the terminal performs the first operation, the method further includes:

The terminal receives first information from the first device, wherein the first information comprises first indication information and/or training progress indication, the first indication information is used for indicating the target sample identification set, and the second device is a third-party server.

In the case of training the second model on the terminal, the terminal receives third information and the second deviation from the first device, the third information including a training progress indication;

In the case that the second model includes the third sub-model, the fourth sub-model, and the fifth sub-model, the terminal receives a fifth deviation from the second device, and updates parameters of the fifth sub-model based on the fifth deviation, the fifth deviation being determined based on a fourth deviation and the third sub-model, and the fourth deviation being determined based on the second deviation and the fourth sub-model.

Optionally, after the terminal receives the first information from the first device, the method further includes:

Under the condition that the second model is trained on the terminal, the terminal sends fourth information to the first device, wherein the fourth information comprises second output information, the input of the second model is first input information, and the second output information is second output of the second model or information obtained by post-processing the second output of the second model based on post-processing configuration;

Optionally, the fourth information further includes second indication information and/or training progress indication, where the second indication information is used to indicate a sample identification set associated with the fourth information.

Optionally, the second indication information is used to indicate at least one of the following:

Identification of all sample data;

collection time stamps for all sample data;

a starting sample identity and a total sample number;

bitmap information of the sample data;

Optionally, the method further comprises:

The terminal receives a preprocessing instruction and/or a post-processing instruction from the first device, wherein the preprocessing instruction is used for indicating the preprocessing configuration, and the post-processing instruction is used for indicating the post-processing configuration.

Optionally, the privacy processing configuration includes a privacy method and a parameter configuration associated with the privacy method.

Optionally, the third information further includes third indication information, where the third indication is used to indicate an optimizer function.

Optionally, the first sample data set further comprises a beam quality measured by the terminal.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

Referring to fig. 10, the embodiment of the present application further provides a model training processing method, as shown in fig. 10, including:

step 1001, the base station receives second information from the first device, where the second information includes first indication information and/or training progress indication, and the first indication information is used to indicate a target sample identification set;

Step 1002, the base station sends third output information to a first device, where the third output information is used for the first device to obtain a first output as an input of a first sub-model, and the first output is used for the first device to determine first target information according to a first sample data set;

Optionally, after the base station sends the third output information to the first device, the method further includes:

the base station receiving the third deviation from the first device;

And the base station updates parameters of the second sub-model according to the third deviation.

Optionally, the method further comprises:

The base station receives a pre-processing indication and/or a post-processing indication from a first device, the pre-processing indication being used for indicating the pre-processing configuration, the post-processing indication being used for indicating the post-processing configuration.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

According to the embodiment of the application, the first target information is determined through the first output of the feature extraction of the base station sensitive information based on the first model, so that the parameter of the model can be updated by using the first target information. Therefore, the embodiment of the application can apply the base station sensitive information to the model training process while protecting the base station sensitive information, thereby improving the reliability of the model training of the terminal and the network side equipment and improving the accuracy of the trained model.

Referring to fig. 11, the embodiment of the present application further provides a model training processing method, as shown in fig. 11, where the model training processing method includes:

step 1101, the second device receives first information from the first device, where the first information includes first indication information and/or training progress indication, and the first indication information is used to indicate a target sample identification set;

step 1102, the second device performs a second operation according to the first information, where the second operation includes at least one of the following:

Acquiring terminal sensitive information and sending second output information to the first equipment, wherein the second output information is a second model

Or 5 information of the second device for post-processing the output information of the second model based on post-processing configuration, wherein the input of the second model is the first input information;

In the case that the second model includes a third sub-model trained on the second device and a fourth sub-model trained on the first device, acquiring terminal sensitive information and transmitting fourth output information to the first device, where the fourth output information is output information of the third sub-model or information obtained by post-processing the output information of the third sub-model by the second device based on post-processing configuration, the fourth output information is used by the first device as input of the fourth sub-model 0, the output information of the fourth sub-model is second output information, and the input of the third sub-model is first input

Entering information;

In the case where the second model includes a third sub-model trained on the second device, a fourth sub-model trained on the first device, and a fifth sub-model trained on the terminal, receiving fifth output information from the terminal and toward the first device

The output information of the third sub-model is sent, the fifth output information is the output information of the fifth sub-model or 5 is information obtained by post-processing the output information of the fifth sub-model by the terminal based on post-processing configuration, the fifth output information is the output information of the fifth sub-model, and the fifth output information is the output information of the fifth sub-model

The fifth output information is used as the input of the fourth sub-model by the first equipment, the output information of the fourth sub-model is the second output information, and the input of the third sub-model is the first input information;

Wherein the first input information is terminal sensitive information associated with the target sample identification set or is based on

The preprocessing configuration is used for preprocessing the terminal sensitive information associated with the target sample identification set, the 0 th second output information is used for the first equipment to determine first target information according to a first sample data set, and the first sample data set

Together comprising second target information, tag data and said second output information, said second target information comprising base station sensitive information or first output information determined based on a first output of a first model, said first model being input for said base station sensitive information, said first target information comprising a first deviation and a second deviation, or said first target information comprising a second deviation,

The first deviation is the counter-propagation deviation corresponding to the first output and is used for parameter updating of the first model 5, and the second deviation is the counter-propagation deviation corresponding to the second output and is used for parameter updating of the second model 5

The second model carries out parameter updating; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

Optionally, after the second device performs the second operation according to the first information, the method further includes at least one of the following 0:

in the case of training the second model on the second device, the second device receives third information and the second deviation from the first device, the third information including a training progress indication;

in the case where the second model includes the third sub-model and the fourth sub-model, the second device is derived from the first model

The first equipment receives a fourth deviation and updates parameters of the third sub-model according to the fourth deviation, and the fourth deviation is determined based on the second deviation and the fourth sub-model 5;

In the case that the second model includes the third sub-model, the fourth sub-model and the fifth sub-model, the second device performs parameter updating on the third sub-model and determines a fifth deviation according to the fourth deviation received from the first device, and sends the fifth deviation to a terminal, where the fifth deviation is used for the terminal to perform parameter updating on the fifth sub-model.

Optionally, the method further comprises: the second device receives a pre-processing indication and/or a post-processing indication from the first device, the pre-processing indication being used for indicating the pre-processing configuration and the post-processing indication being used for indicating the post-processing configuration.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

According to the model training processing method provided by the embodiment of the application, the execution subject can be a model training processing device. In the embodiment of the application, a model training processing device is taken as an example to execute a model training processing method, and the model training processing device provided by the embodiment of the application is described.

Referring to fig. 12, the embodiment of the present application further provides a model training processing device, as shown in fig. 12, the model training processing device 1200 includes:

a first determining module 1201, configured to determine first target information according to the first sample data set;

Optionally, the model training processing device 1200 further includes: the first transmitting module is configured to transmit the first data,

The first determining module 1201 is further configured to determine a target sample identification set, where the target sample identification set is used to determine the first sample data set;

The first sending module is used for sending first information to the terminal or the second equipment, the first information comprises first indication information and/or training progress indication, the first indication information is used for indicating the target sample identification set, and the second equipment is a third-party server.

Optionally, the first model satisfies any one of:

the first model is trained on the first device;

Optionally, the model training processing device 1200 further includes a first processing module, configured to perform the following operations:

Performing parameter updating on the first model according to the first deviation under the condition that the first model is trained on the first equipment;

And under the condition that the first model comprises the first sub-model and the second sub-model, carrying out parameter updating on the first sub-model according to a first deviation, determining a third deviation, and sending the third deviation to the base station, wherein the third deviation is used for carrying out parameter updating on the second sub-model.

Transmitting second information to the base station, wherein the second information comprises the first indication information and/or the training progress indication;

Receiving third output information from the base station, wherein the third output information is the output information of the second sub-model or is information obtained by the base station through post-processing the output information of the second sub-model based on post-processing configuration, and the post-processing configuration comprises sparse processing configuration and/or privacy processing configuration;

obtaining the first output by taking the third output information as the input of the first sub-model;

Optionally, the model training processing device 1200 further includes:

The first sending module is used for sending a preprocessing instruction and/or a post-processing instruction to the base station, wherein the preprocessing instruction is used for indicating the preprocessing configuration, and the post-processing instruction is used for indicating the post-processing configuration.

Optionally, the model training processing arrangement 1200 further comprises a first processing module,

The first determining module 1201 is further configured to determine a second sample data set, where the second sample data set includes the base station sensitive information, and the second sample data set is associated with the target sample identification set;

the first processing module is configured to input the second sample data set into the first model for forward propagation to obtain the first output.

Optionally, the second model satisfies any one of the following:

the second model is trained on a terminal or a second device;

the second device is a third-party server.

Optionally, the model training processing device 1200 further includes a first processing module configured to perform at least one of:

transmitting third information and the second deviation to the terminal or the second device, wherein the third information comprises a training progress indication, under the condition that the second model is trained on the terminal or the second device;

In the case that the second model comprises the third sub-model and the fourth sub-model, performing parameter updating on the fourth sub-model according to the second deviation, determining a fourth deviation, and sending the fourth deviation to second equipment, wherein the fourth deviation is used for the second equipment to perform parameter updating on the third sub-model;

And under the condition that the second model comprises the third sub-model, the fourth sub-model and the fifth sub-model, carrying out parameter updating on the fourth sub-model and determining a fourth deviation according to the second deviation, and sending the fourth deviation to second equipment, wherein the fourth deviation is used for carrying out parameter updating on the third sub-model and determining a fifth deviation by the second equipment, and the fifth deviation is used for carrying out parameter updating on the fifth sub-model by the terminal.

Optionally, the model training processing device 1200 further includes:

The first receiving module is used for receiving fourth information from the terminal or the second device, and the fourth information meets at least one of the following items:

Optionally, the fourth information further includes second indication information and/or training progress indication, where the second indication information is used to indicate a sample identification set associated with the fourth information; the first determining module 1201 is specifically configured to determine, when the sample identifier set indicated by the second instruction information is the same as the target sample identifier set, first target information according to the first sample data set.

Identification of all sample data;

collection time stamps for all sample data;

A start sample identity and a stop sample identity;

a starting sample identity and a total sample number;

bitmap information of the sample data;

Optionally, the model training processing device 1200 further includes:

the first sending module is used for sending a preprocessing instruction and/or a post-processing instruction to the terminal or the second equipment, wherein the preprocessing instruction is used for indicating the preprocessing configuration, and the post-processing instruction is used for indicating the post-processing configuration.

Optionally, the first determining module 1201 is specifically configured to perform the following operations:

Inputting the second target information, the second output, and the measured beam quality into a third model;

Determining a first loss function according to the output information of the third model and the tag data;

And carrying out back propagation on the basis of the first loss function to update network parameters of the third model and obtain the first target information.

Performing addition operation on the second target information and the second output;

Determining a second loss function based on the output of the add operation and the tag data;

and back-propagating based on the second loss function to obtain the first target information.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

Referring to fig. 13, the embodiment of the present application further provides a model training processing device, as shown in fig. 13, the model training processing device 1300 includes:

A first execution module 1301 configured to execute a first operation, where the first operation includes any one of the following:

Wherein the first input information is terminal sensitive information associated with the target sample identification set or information obtained by preprocessing terminal sensitive information associated with the target sample identification set based on preprocessing configuration, the first input information is terminal sensitive information associated with the target sample identification set

The second output information is the second output of the second model or the information obtained by post-processing the second output 5 of the second model based on post-processing configuration, and the second output information is used for the first device to determine a first target according to the first sample data set

Information, wherein the first sample data set comprises second target information, tag data and the second output information; the second target information includes base station sensitive information or first output information determined based on a first output of a first model, the input of the first model is determined based on the base station sensitive information, the first target information includes a first deviation and a second deviation, or

The first target information includes a second deviation, the first deviation is a counter-propagation deviation corresponding to the first output, 0, and the first deviation is used for updating parameters of the first model, and the second deviation is a counter-propagation deviation corresponding to the second output

Propagating a deviation, wherein the second deviation is used for updating parameters of the second model; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following: the method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

Optionally, the model training processing device 1300 further includes:

5a second receiving module for receiving first information from the first device, the first information including first indication information

And/or training progress indication, wherein the first indication information is used for indicating the target sample identification set, and the second device is a third-party server.

Optionally, the first execution module 1301 is further configured to execute at least one of the following:

Receiving third information and the second 0 bias from the first device with the second model trained on the terminal, the third information including a training progress indication;

In the case where the second model includes the third sub-model, the fourth sub-model, and the fifth sub-model, a fifth bias received from a second device is parameter-updated based on the fifth bias, the fifth bias being determined based on a fourth bias and the third sub-model, the fourth bias being determined based on the second bias and the fourth sub-model.

5 Optionally, the model training processing device 1300 further comprises:

A second sending module, configured to send fourth information to the first device in a case where the second model is trained on the terminal, where the fourth information includes the second output information, an input of the second model is first input information, and the second output information is a second output of the second model or is a second output of the second model based on a post-processing configuration

Information obtained by post-processing is output;

0 wherein the first input information is terminal sensitive information associated with the target sample identification set or is based on

Preprocessing configuration is carried out on information after preprocessing the terminal sensitive information associated with the target sample identification set; the post-processing configuration comprises a sparsification processing configuration and/or a privacy processing configuration, and the preprocessing configuration comprises at least one of the following:

The method comprises the following steps of single-hot code coding dictionary configuration, data normalization parameter configuration, data regularization parameter configuration and data standardization parameter configuration.

Optionally, the fourth information further includes second indication information and/or training progress indication, where the second indication information is used for indicating a sample identification set associated with the fourth information by 5.

Identification of all sample data;

collection time stamps for all sample data;

a starting sample identity and a total sample number;

bitmap information of the sample data;

Optionally, the model training processing device 1300 further includes:

And the second receiving module is used for receiving a preprocessing instruction and/or a post-processing instruction from the first equipment, wherein the preprocessing instruction is used for indicating the preprocessing configuration, and the post-processing instruction is used for indicating the post-processing configuration.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

Referring to fig. 14, the embodiment of the present application further provides a model training processing device, as shown in fig. 14, the model training processing device 1400 includes:

A third receiving module 1401, configured to receive second information from the first device, where the second information includes first indication information and/or a training progress indication, and the first indication information is used to indicate a target sample identifier set;

a third sending module 1402, configured to send third output information to a first device, where the third output information is used for the first device to obtain a first output as an input of a first sub-model, and the first output is used for the first device to determine first target information according to a first sample data set;

Optionally, the training processing device 1400 further comprises an update module,

The third receiving module 1401 is further configured to receive the third deviation from the first device;

and the updating module is used for updating parameters of the second sub-model according to the third deviation.

Optionally, the third receiving module 1401 is further configured to receive a pre-processing indication and/or a post-processing indication from the first device, where the pre-processing indication is used to indicate the pre-processing configuration, and the post-processing indication is used to indicate the post-processing configuration.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

Referring to fig. 15, the embodiment of the present application further provides a model training processing device, as shown in fig. 15, the model training processing device 1500 includes:

a fourth receiving module 1501, configured to receive first information from a first device, where the first information includes first indication information and/or training progress indication, and the first indication information is used to indicate a target sample identification set;

A second execution module 1502, configured to execute a second operation according to the first information, where the second operation includes at least one of:

Optionally, the second execution module 1502 is further configured to execute at least one of:

receiving third information and the second deviation from the first device, with the second model trained on the second device, the third information including a training progress indication;

Receiving a fourth deviation from the first device and performing parameter updating on the third sub-model according to the fourth deviation, the fourth deviation being determined based on the second deviation and the fourth sub-model, in the case that the second model comprises the third sub-model and the fourth sub-model;

And in the case that the second model comprises the third sub-model, the fourth sub-model and the fifth sub-model, performing parameter updating on the third sub-model according to the fourth deviation received from the first equipment, determining a fifth deviation, and sending the fifth deviation to a terminal, wherein the fifth deviation is used for performing parameter updating on the fifth sub-model by the terminal.

Optionally, the fourth receiving module 1502 is further configured to: a pre-processing indication and/or a post-processing indication is received from the first device, the pre-processing indication being used to indicate the pre-processing configuration and the post-processing indication being used to indicate the post-processing configuration.

Optionally, the tag data includes at least one of:

beam quality of all transmit beams;

beam quality of all received beams;

beam quality of all transmit-receive beam pairs;

the beam mark with the strongest beam quality in all the transmitting beams;

The beam mark with the strongest beam quality in all the received beams;

The model training processing device in the embodiment of the application can be an electronic device, such as an electronic device with an operating system, or can be a component in the electronic device, such as an integrated circuit or a chip. The electronic device may be a terminal, or may be other devices than a terminal. By way of example, the terminals may include, but are not limited to, the types of terminals 11 listed above, other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., and embodiments of the present application are not limited in detail.

The model training processing device provided by the embodiment of the application can realize each process realized by the method embodiments of fig. 2 to 11 and achieve the same technical effects, and in order to avoid repetition, the description is omitted here.

Optionally, as shown in fig. 16, the embodiment of the present application further provides a communication device 1600, which includes a processor 1601 and a memory 1602, where the memory 1602 stores a program or an instruction that can be executed on the processor 1601, and the program or the instruction implements each step of the embodiment of the model training processing method when executed by the processor 1601, and the steps achieve the same technical effects, so that repetition is avoided, and no further description is given here.

The embodiment of the application also provides a terminal, which comprises a processor and a communication interface, wherein the processor is used for executing a first operation, and the first operation comprises any one of the following steps:

The terminal embodiment corresponds to the terminal-side method embodiment, and each implementation process and implementation manner of the method embodiment can be applied to the terminal embodiment, and the same technical effects can be achieved. Specifically, fig. 17 is a schematic diagram of a hardware structure of a terminal for implementing an embodiment of the present application.

The terminal 1700 includes, but is not limited to: at least some of the components of the radio frequency unit 1701, the network module 1702, the audio output unit 1703, the input unit 1704, the sensor 1705, the display unit 1706, the user input unit 1707, the interface unit 1708, the memory 1709, the processor 1710, and the like.

Those skilled in the art will appreciate that terminal 1700 may also include a power source (e.g., a battery) for powering the various components, which may be logically connected to processor 1710 via a power management system so as to perform functions such as managing charge, discharge, and power consumption via the power management system. The terminal structure shown in fig. 17 does not constitute a limitation of the terminal, and the terminal may include more or less components than shown, or may combine some components, or may be arranged in different components, which will not be described in detail herein.

It should be appreciated that in embodiments of the present application, the input unit 1704 may include a graphics processing unit (Graphics Processing Unit, GPU) 17041 and a microphone 17042, with the graphics processor 17041 processing image data of still pictures or video obtained by an image capturing device (e.g., a camera) in a video capturing mode or an image capturing mode. The display unit 1706 may include a display panel 17061, and the display panel 17061 may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 1707 includes at least one of a touch panel 17071 and other input devices 17072. Touch panel 17071, also referred to as a touch screen. The touch panel 17071 may include two parts, a touch detection device and a touch controller. Other input devices 17072 may include, but are not limited to, a physical keyboard, function keys (e.g., volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and so forth, which are not described in detail herein.

In the embodiment of the present application, after receiving downlink data from the network side device, the radio frequency unit 1701 may transmit the downlink data to the processor 1710 for processing; in addition, the radio frequency unit 1701 may send uplink data to the network side device. In general, the radio frequency unit 1701 includes, but is not limited to, an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, and the like.

The memory 1709 may be used for storing software programs or instructions and various data. The memory 1709 may mainly include a first storage area storing programs or instructions and a second storage area storing data, wherein the first storage area may store an operating system, application programs or instructions (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like. Further, the memory 1709 may include volatile memory or nonvolatile memory, or the memory 1709 may include both volatile and nonvolatile memory. The nonvolatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable EPROM (EEPROM), or a flash Memory. The volatile memory may be random access memory (Random Access Memory, RAM), static random access memory (STATIC RAM, SRAM), dynamic random access memory (DYNAMIC RAM, DRAM), synchronous Dynamic Random Access Memory (SDRAM), double data rate Synchronous dynamic random access memory (Double DATA RATE SDRAM, DDRSDRAM), enhanced Synchronous dynamic random access memory (ENHANCED SDRAM, ESDRAM), synchronous link dynamic random access memory (SYNCH LINK DRAM, SLDRAM), and Direct random access memory (DRRAM). Memory 1709 in embodiments of the application includes, but is not limited to, these and any other suitable types of memory.

Processor 1710 can include one or more processing units; optionally, the processor 1710 integrates an application processor that primarily handles operations related to the operating system, user interface, and applications, and a modem processor that primarily handles wireless communication signals, such as a baseband processor. It will be appreciated that the modem processor described above may not be integrated into the processor 1710.

Wherein the processor 1710 is configured to perform a first operation, where the first operation includes any one of:

The embodiment of the application also provides network side equipment, which comprises a processor and a communication interface, wherein,

The network side device embodiment corresponds to the network side device method embodiment, and each implementation process and implementation manner of the method embodiment can be applied to the network side device embodiment, and the same technical effects can be achieved.

Specifically, the embodiment of the application also provides network side equipment. As shown in fig. 18, the network side device 1800 includes: an antenna 1801, a radio frequency device 1802, a baseband device 1803, a processor 1804, and a memory 1805. The antenna 1801 is connected to a radio frequency device 1802. In the uplink direction, the radio frequency device 1802 receives information via the antenna 1801, and transmits the received information to the baseband device 1803 for processing. In the downlink direction, the baseband device 1803 processes information to be transmitted, and transmits the processed information to the radio frequency device 1802, and the radio frequency device 1802 processes the received information and transmits the processed information through the antenna 1801.

The method performed by the network side device in the above embodiment may be implemented in the baseband apparatus 1803, where the baseband apparatus 1803 includes a baseband processor.

The baseband apparatus 1803 may, for example, include at least one baseband board, where a plurality of chips are disposed, as shown in fig. 18, where one chip, for example, a baseband processor, is connected to the memory 1805 through a bus interface, so as to call a program in the memory 1805 to perform the network device operation shown in the above method embodiment.

The network-side device may also include a network interface 1806, such as a common public radio interface (common public radio interface, CPRI).

Specifically, the network side device 1800 of the embodiment of the present invention further includes: instructions or programs stored in the memory 1805 and executable on the processor 1804, the processor 1804 invokes the instructions or programs in the memory 1805 to perform the methods performed by the modules shown in fig. 12 or 14, and achieve the same technical effects, so repetition is avoided and will not be described herein.

The embodiment of the application also provides a readable storage medium, wherein the readable storage medium stores a program or an instruction, and the program or the instruction realizes each process of the model training processing method embodiment when being executed by a processor, and can achieve the same technical effect, so that repetition is avoided and redundant description is omitted.

Wherein the processor is a processor in the terminal described in the above embodiment. The readable storage medium includes computer readable storage medium such as computer readable memory ROM, random access memory RAM, magnetic or optical disk, etc.

The embodiment of the application further provides a chip, which comprises a processor and a communication interface, wherein the communication interface is coupled with the processor, and the processor is used for running programs or instructions to realize the processes of the embodiment of the model training processing method, and the same technical effects can be achieved, so that repetition is avoided, and the description is omitted here.

It should be understood that the chips referred to in the embodiments of the present application may also be referred to as system-on-chip chips, or the like.

The embodiment of the present application further provides a computer program/program product, where the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement each process of the above embodiment of the model training processing method, and the same technical effects can be achieved, so that repetition is avoided, and details are not repeated here.

The embodiment of the application also provides a communication system, which comprises: the terminal is used for executing the processes of the embodiments of the model training processing method of the terminal side shown in fig. 9, the network side device is used for executing the processes of the embodiments of the model training processing method of the network side shown in fig. 2 or 10, and the server is used for executing the processes of the embodiments of the model training processing method of the server side shown in fig. 11, and the same technical effects can be achieved, so that repetition is avoided and no further description is given here.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element. Furthermore, it should be noted that the scope of the methods and apparatus in the embodiments of the present application is not limited to performing the functions in the order shown or discussed, but may also include performing the functions in a substantially simultaneous manner or in an opposite order depending on the functions involved, e.g., the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a computer software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present application.

The embodiments of the present application have been described above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present application and the scope of the claims, which are to be protected by the present application.

Claims

1. A model training processing method, comprising:

2. The method of claim 1, wherein prior to the first device determining the first target information from the first set of sample data, the method further comprises:

3. The method of claim 2, wherein the first model satisfies any one of:

the first model is trained on the first device;

4. A method according to claim 3, wherein after the first device determines the first target information from the first set of sample data, the method further comprises:

5. A method according to claim 3, wherein, in case the first model comprises the first sub-model and the second sub-model, the first device determines a set of target sample identities, the method further comprises:

6. The method of claim 5, wherein the method further comprises:

7. The method of claim 3, wherein after the first device determines the set of target sample identifications, the method further comprises:

the first device inputs the second sample data set into the first model for forward propagation to obtain the first output.

8. The method of claim 2, wherein the second model satisfies any one of:

the second model is trained on a terminal or a second device;

the second device is a third-party server.

9. The method of claim 8, wherein after the first device determines the first target information from the first set of sample data, the method further comprises at least one of:

10. The method of claim 8, wherein after the first device sends the first information to the terminal or the second device, the method further comprises:

11. The method according to claim 10, wherein the fourth information further comprises second indication information and/or a training progress indication, the second indication information being used to indicate a set of sample identities associated with the fourth information; the first device determining first target information from the first sample data set includes:

12. The method of claim 11, wherein the second indication information is used to indicate at least one of:

Identification of all sample data;

collection time stamps for all sample data;

A start sample identity and a stop sample identity;

a starting sample identity and a total sample number;

bitmap information of the sample data;

13. The method according to claim 10, wherein the method further comprises:

14. The method according to claim 5 or 13, wherein the privacy processing configuration comprises a privacy method and a privacy method associated parameter configuration.

15. The method of claim 14, wherein the privacy method comprises any one of: differential privacy, homomorphic encryption, and secret sharing.

16. The method of claim 15, wherein the differential privacy-related parameter configuration comprises at least one of: privacy mechanism and differential privacy parameter configuration; wherein the differential privacy parameter configuration includes at least one of: privacy budgets; a slack term; clipping values or sensitivities; number of iteration rounds.

17. The method of claim 9, wherein the third information further comprises third indication information, the third indication indicating an optimizer function.

18. The method according to any of claims 1 to 17, wherein the first set of sample data further comprises a beam quality measured by the terminal.

19. The method of claim 18, wherein the first device determining first target information from the first set of sample data comprises:

The first device back-propagates updating network parameters of the third model and obtaining the first target information based on the first loss function.

20. The method of any of claims 1 to 17, wherein the first device determining first target information from the first sample data set comprises:

21. A model training processing method, comprising:

22. The method of claim 21, wherein prior to the terminal performing the first operation, the method further comprises:

23. The method of claim 21, wherein prior to the terminal performing the first operation, the method further comprises:

24. The method of claim 22, wherein after the terminal receives the first information from the first device, the method further comprises:

25. The method according to claim 24, wherein the fourth information further comprises second indication information and/or a training progress indication, the second indication information being used to indicate a set of sample identities associated with the fourth information.

26. The method of claim 25, wherein the second indication information is used to indicate at least one of:

Identification of all sample data;

collection time stamps for all sample data;

a starting sample identity and a total sample number;

bitmap information of the sample data;

27. The method of claim 24, wherein the method further comprises:

28. The method of claim 27, wherein the privacy processing configuration comprises a privacy method and a privacy method associated parameter configuration.

29. The method of claim 28, wherein the privacy method comprises any one of: differential privacy, homomorphic encryption, and secret sharing.

30. The method of claim 29, wherein the differential privacy-associated parameter configuration comprises at least one of a privacy mechanism and a differential privacy parameter configuration; wherein the differential privacy parameter configuration includes at least one of: privacy budgets; a slack term; clipping values or sensitivities; number of iteration rounds.

31. The method of claim 23, wherein the third information further comprises third indication information, the third indication indicating an optimizer function.

32. The method according to any of claims 21 to 31, wherein the first set of sample data further comprises a beam quality measured by the terminal.

33. A model training processing method, comprising:

34. The method of claim 33, wherein after the base station transmits the third output information to the first device, the method further comprises:

the base station receiving the third deviation from the first device;

35. The method of claim 33, wherein the method further comprises:

36. The method of claim 33, wherein the privacy processing configuration comprises a privacy method and a privacy method associated parameter configuration.

37. The method of claim 36, wherein the privacy method comprises any one of: differential privacy, homomorphic encryption, and secret sharing.

38. The method of claim 37, wherein the differential privacy-associated parameter configuration comprises at least one of a privacy mechanism and a differential privacy parameter configuration; wherein the differential privacy parameter configuration includes at least one of: privacy budgets; a slack term; clipping values or sensitivities; number of iteration rounds.

39. The method according to any of claims 33 to 38, wherein the first set of sample data further comprises a beam quality measured by the terminal.

40. A model training processing method, comprising:

41. The method of claim 40, wherein after the second device performs a second operation according to the first information, the method further comprises at least one of:

In the case that the second model includes the third sub-model and the fourth sub-model, the second device receives a fourth bias from the first device and updates parameters of the third sub-model according to the fourth bias, the fourth bias being determined based on the second bias and the fourth sub-model;

42. The method of claim 40, further comprising:

The second device receives a pre-processing indication and/or a post-processing indication from the first device, the pre-processing indication being used for indicating the pre-processing configuration and the post-processing indication being used for indicating the post-processing configuration.

43. The method of claim 40, wherein the privacy processing configuration includes a privacy method and a privacy method associated parameter configuration.

44. The method of claim 43, wherein the privacy method comprises any one of: differential privacy, homomorphic encryption, and secret sharing.

45. The method of claim 44, wherein the differential privacy-associated parameter configuration includes at least one of a privacy mechanism and a differential privacy parameter configuration; wherein the differential privacy parameter configuration includes at least one of: privacy budgets; a slack term; clipping values or sensitivities; number of iteration rounds.

46. The method of claim 41, wherein the third information further comprises third indication information, the third indication indicating an optimizer function.

47. A model training processing device, comprising:

48. A model training processing device, comprising:

49. A model training processing device, comprising:

50. A model training processing device, comprising:

51. A terminal comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the model training processing method of any of claims 21 to 32.

52. A network side device comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the model training processing method of any of claims 1 to 20 or 33 to 39.

53. A third party server comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the model training processing method of any of claims 40 to 46.

54. A readable storage medium having stored thereon a program or instructions which when executed by a processor perform the steps of the model training processing method of any of claims 1 to 46.