Data processing method and apparatus, communication device, and storage medium

Abstract

The method and device for processing data, the communication device and the storage medium are provided. The method for processing data provided by the embodiment of the present disclosure comprises: determining the local data set distribution characteristics of at least one UE; and based on the local data set distribution characteristics, scheduling a target UE participating in federated learning from the at least one UE.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the field of wireless communication, and more particularly, to a data processing method and apparatus, a communication device, and a storage medium. BACKGROUND

[0002] With the development of artificial intelligence technology, machine learning is applied to more and more fields. However, the training data sources of many machine learning models are distributed among different institutions, and these institutions usually do not share data, and need to consider data privacy and security issues. In addition, the number of wireless network users is growing rapidly, and different users have different specific requirements for models, and using a unified model will greatly reduce the model accuracy. SUMMARY

[0003] The present disclosure provides a data processing method and apparatus, a communication device, and a storage medium.

[0004] According to a first aspect of an embodiment of the present disclosure, a data processing method is provided, applied to a base station, comprising:

[0005] determining a local data set distribution characteristic of at least one user equipment (UE),

[0006] scheduling a target UE participating in federated learning from the at least one UE based on the local data set distribution characteristic.

[0007] In some embodiments, the scheduling of the target UE participating in federated learning from the at least one UE based on the local data set distribution characteristic comprises:

[0008] obtaining distribution difference statistical information of the local data set of each of the at least one UE and a global data set;

[0009] scheduling a target UE participating in federated learning from the at least one UE according to the distribution difference statistical information.

[0010] In some embodiments, the method further comprises:

[0011] obtaining capability information of the at least one UE;

[0012] The scheduling of the target UE participating in federated learning from the at least one UE based on the local data set distribution characteristic comprises:

[0013] scheduling a target UE participating in federated learning from the at least one UE according to the local data set distribution characteristic and the capability information of the at least one UE.

[0014] In some embodiments, the capability information of the at least one UE comprises at least one of:

[0015] computing capability information indicating a computing capability of the UE;

[0016] communication condition information indicating a communication capability and / or a communication channel condition of the UE.

[0017] In some embodiments, the communication condition information comprises channel quality indication (CQI) information detected by the UE.

[0018] In some embodiments, the method further comprises determining a weight coefficient of the target UE in the federated learning according to distribution difference statistical information of a local data set of the target UE and a global data set of a base station.

[0019] In some embodiments, the distribution difference statistical information comprises a probability distribution difference.

[0020] The determining of the weight coefficient of the target UE in the federated learning according to the distribution difference statistical information of the local data set of the target UE and the global data set of the base station comprises:

[0021] determining the weight coefficient of the target UE according to a probability distribution difference corresponding to the single target UE and a sum of probability distribution differences of all target UEs performing the same federated learning.

[0022] In some embodiments, the method further comprises:

[0023] receiving model information of a local model of the target UE for performing the federated learning;

[0024] performing weighted averaging on local models of a plurality of target UEs according to the weight coefficient of the target UE and the model information of the local model, to obtain a global learning model.

[0025] In some embodiments, the method further comprises:

[0026] in response to the global learning model satisfying an OAM subscription requirement, stopping receiving the model information of the local model of the target UE for performing the federated learning.

[0027] In some embodiments, the method further comprises:

[0028] in response to the global learning model not satisfying the OAM subscription requirement, sending model information of the global learning model to the target UE;

[0029] receiving model information of a local model of the target UE updated according to the global learning model;

[0030] update the global learning model according to the model parameters.

[0031] In some embodiments, the method further includes:

[0032] reporting model information of the global learning model and training data for training the global learning model to an OAM;

[0033] receiving model parameters determined by the OAM according to the model information of the global learning model, the training data, and task data of the OAM;

[0034] updating the global learning model according to the model parameters.

[0035] In some embodiments, the method further includes:

[0036] In response to detecting that handover occurs to a base station connected by the target UE, determining that the target UE exits the federated learning.

[0037] According to a second aspect of the embodiments of the present disclosure, a data processing method is provided, which is applied to a UE and includes:

[0038] receiving scheduling information; wherein the scheduling information is sent by a base station based on distribution characteristics of a local data set of the UE for federated learning scheduling.

[0039] In some embodiments, the scheduling information issued by the base station according to the distribution characteristics of the local data set of the UE includes:

[0040] The scheduling information issued by the base station according to the distribution difference statistical information obtained by the base station according to the distribution characteristics of the local data set of the UE and the distribution characteristics of the global data set.

[0041] In some embodiments, the method further includes:

[0042] reporting capability information; wherein the capability information is used for the base station to issue the scheduling information according to the distribution characteristics of the local data set and the capability information.

[0043] In some embodiments, the capability information includes at least one of the following:

[0044] computing capability information indicating computing capability of the UE;

[0045] communication status information indicating communication capability and / or communication channel status of the UE.

[0046] In some embodiments, the communication status information includes CQI information; the method further includes:

[0047] detecting the CQI information of a channel between the UE and the base station.

[0048] In some embodiments, the method further includes:

[0049] reporting model information of the local model of the UE; wherein the local model is used for the base station to perform the federated learning according to the local model and a weight coefficient of the UE; wherein the weight coefficient of the UE is determined by the base station according to distribution difference statistical information of a local data set of the UE and a global data set of the base station.

[0050] In some embodiments, the method further includes:

[0051] generating the local data set according to collected wireless network data;

[0052] extracting data of the local data set to generate a local training data set;

[0053] performing model training using the local training data set to obtain the local model.

[0054] In some embodiments, the method further includes:

[0055] receiving model information of a global learning model issued by the base station;

[0056] performing the federated learning according to the model information of the global learning model to obtain an updated local model;

[0057] in response to the global learning model not meeting the OAM subscription requirement, reporting model information of the updated local model.

[0058] In some embodiments, the method further includes:

[0059] in response to the global learning model meeting the OAM subscription requirement, stopping the federated learning.

[0060] In some embodiments, the method further includes:

[0061] in response to handover of a base station to which the UE is connected, stopping the federated learning.

[0062] According to a third aspect of the embodiments of the present disclosure, a data processing apparatus is provided, which is applied to a base station and includes:

[0063] a first determining module configured to determine local data set distribution characteristics of at least one user equipment (UE),

[0064] The scheduling module is configured to schedule a target UE participating in federated learning from the at least one UE based on the distribution characteristics of the local data set.

[0065] In some embodiments, the scheduling module comprises:

[0066] The first obtaining sub-module is configured to obtain distribution difference statistical information of the local data set of each UE in the at least one UE and a global data set;

[0067] The first scheduling sub-module is configured to schedule a target UE participating in federated learning from the at least one UE according to the distribution difference statistical information.

[0068] In some embodiments, the apparatus further comprises:

[0069] The first obtaining module is configured to obtain capability information of the at least one UE;

[0070] The scheduling module comprises:

[0071] The second scheduling sub-module is configured to schedule a target UE participating in federated learning from the at least one UE according to the distribution characteristics of the local data set and the capability information of the at least one UE.

[0072] In some embodiments, the capability information of the at least one UE comprises at least one of:

[0073] The computing capability information indicates a computing capability of the UE;

[0074] The communication condition information indicates a communication capability and / or a communication channel condition of the UE.

[0075] In some embodiments, the communication condition information comprises channel quality indicator (CQI) information detected by the UE.

[0076] In some embodiments, the apparatus further comprises:

[0077] The second determining module is configured to determine a weight coefficient of the target UE in the federated learning according to distribution difference statistical information of a local data set of the target UE and a global data set of a base station.

[0078] In some embodiments, the distribution difference statistical information comprises a probability distribution difference.

[0079] The second determining module comprises:

[0080] The first determining sub-module is configured to determine the weight coefficient of the target UE according to a probability distribution difference corresponding to a single target UE and a sum of probability distribution differences of all target UEs performing the same federated learning.

[0081] In some embodiments, the apparatus further includes:

[0082] a first receiving module, configured to receive model information of a local model reported by the target UE for the federated learning;

[0083] a processing module, configured to perform weighted averaging on the local models of the target UEs according to the weight coefficients of the target UEs and the model information of the local models, to obtain a global learning model.

[0084] In some embodiments, the apparatus further includes:

[0085] a first stopping module, configured to stop receiving the model information of the local model reported by the target UE for the federated learning, in response to the global learning model satisfying an OAM subscription requirement.

[0086] In some embodiments, the apparatus further includes:

[0087] a first sending module, configured to send model information of the global learning model to the target UE, in response to the global learning model not satisfying the OAM subscription requirement.

[0088] a second receiving module, configured to receive model information of an updated local model of the target UE according to the global learning model;

[0089] a first updating module, configured to update the global learning model according to the updated local model of the target UE and a weight coefficient corresponding to the local model.

[0090] In some embodiments, the apparatus further includes:

[0091] a first reporting module, configured to report model information of the global learning model and training data for training the global learning model to an OAM;

[0092] a third receiving module, configured to receive model parameters determined by the OAM according to the model information of the global learning model, the training data and task data of the OAM;

[0093] a second updating module, configured to update the global learning model according to the model parameters.

[0094] In some embodiments, the apparatus further includes:

[0095] a third determining module, configured to determine that the target UE exits the federated learning, in response to detecting that a handover of a base station connected by the target UE occurs.

[0096] According to a fourth aspect of the embodiments of the present disclosure, a data processing apparatus is provided, which is applied to a UE and comprises:

[0097] a fourth receiving module configured to receive scheduling information, wherein the scheduling information is sent by a base station based on distribution characteristics of a local data set of the UE for federated learning scheduling.

[0098] In some embodiments, the fourth receiving module is specifically configured to:

[0099] receive scheduling information issued by a base station according to distribution difference statistical information obtained by the base station based on distribution characteristics of a local data set of the UE and distribution characteristics of a global data set.

[0100] In some embodiments, the apparatus further comprises:

[0101] a second reporting module configured to report capability information, wherein the capability information is used by the base station to issue the scheduling information according to the distribution characteristics of the local data set and the capability information.

[0102] In some embodiments, the capability information comprises at least one of:

[0103] computing capability information indicating computing capability of the UE;

[0104] communication status information indicating communication capability and / or communication channel status of the UE.

[0105] In some embodiments, the communication status information comprises CQI information, and the apparatus further comprises:

[0106] a detecting module configured to detect the CQI information of a channel between the UE and the base station.

[0107] In some embodiments, the apparatus further comprises:

[0108] a third reporting module configured to report model information of a local model of the UE, wherein the local model is used by the base station to perform the federated learning according to the local model and a weight coefficient of the UE, and wherein the weight coefficient of the UE is determined by the base station according to distribution difference statistical information of the local data set of the UE and the global data set of the base station.

[0109] In some embodiments, the apparatus further comprises:

[0110] a first generating module configured to generate the local data set according to collected wireless network data;

[0111] a second generating module configured to extract data of the local data set to generate a local training data set.

[0112] a training module, configured to perform model training by using the local training data set, to obtain the local model.

[0113] In some embodiments, the apparatus further includes:

[0114] a fifth receiving module, configured to receive model information of a global learning model issued by the base station;

[0115] a third updating module, configured to perform the federated learning according to the model information of the global learning model, to obtain an updated local model;

[0116] a fourth reporting module, configured to report model information of the updated local model in response to the global learning model not meeting the OAM subscription requirement.

[0117] In some embodiments, the apparatus further includes:

[0118] a second stopping module, configured to stop the federated learning in response to the global learning model meeting the OAM subscription requirement.

[0119] In some embodiments, the apparatus further includes:

[0120] a third stopping module, configured to stop the federated learning in response to handover of a base station to which the UE is connected.

[0121] According to a fifth aspect of the embodiments of the present disclosure, a communication device is provided, including at least a processor and a memory for storing executable instructions capable of running on the processor, wherein:

[0122] When the processor runs the executable instructions, the executable instructions perform the steps in any of the methods for determining the processing duration.

[0123] According to a sixth aspect of the embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer executable instructions, and the computer executable instructions are executed by a processor to implement the steps in any of the methods for determining the processing duration.

[0124] In the embodiments of the present disclosure, the base station can utilize the local data set distribution characteristics of the UE to realize scheduling of the UE that can participate in the federated learning from multiple alternative UEs that have a communication connection with the base station, and perform federated learning with the UE that participates in the federated learning. Thus, direct scheduling of the base station can be realized, and the UE does not need to report data to the core network or the data center, which can greatly improve training efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0125] The accompanying drawings, which are incorporated herein and constitute part of this specification, illustrate implementations of the present application and, together with the description, explain the principles of the present application.

[0126] Figure 1 is a structural schematic diagram of a wireless communication system according to an exemplary embodiment;

[0127] Figure 2 is a flowchart of a data processing method according to an exemplary embodiment; Figure 1 ;

[0128] Figure 3 is a flowchart of a data processing method according to an exemplary embodiment; Figure 2 ;

[0129] Figure 4 is a schematic diagram of the principle of federated learning according to an exemplary embodiment;

[0130] Figure 5 is a structural block diagram of a user device of federated learning according to an exemplary embodiment;

[0131] Figure 6 is a structural block diagram of a base station device of federated learning according to an exemplary embodiment;

[0132] Figure 7 is a total flowchart of a data processing method according to an exemplary embodiment;

[0133] Figure 8 is a flowchart of federated learning between a user and a base station according to an exemplary embodiment;

[0134] Figure 9 is a flowchart of a method of model selection in a data processing method according to an exemplary embodiment;

[0135] Figure 10 is a flowchart of a method of user scheduling in a data processing method according to an exemplary embodiment;

[0136] Figure 11 is a flowchart of a method of federated learning in a data processing method according to an exemplary embodiment;

[0137] Figure 12 is a flowchart of a method of data transmission in a data processing method according to an exemplary embodiment;

[0138] Figure 13A is a structural schematic diagram of a data processing device according to an exemplary embodiment; Figure 1 ;

[0139] Figure 13B is a structural diagram of a data processing apparatus according to an example embodiment Figure 2 ;

[0140] Figure 14 is a structural diagram of a communication device according to an example embodiment Figure 1 ;

[0141] Figure 15 is a structural diagram of a communication device according to an example embodiment Figure 2 . DETAILED DESCRIPTION

[0142] The example embodiments will be described in detail herein with reference to the attached drawings. In the following description, unless otherwise indicated, like numbers refer to like elements throughout the description. The following example embodiments are not representative of all embodiments consistent with the present disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0143] The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. As used in the present disclosure and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0144] It will be understood that, although the terms first, second, third, etc. can be used herein to describe various information, these terms are not intended to denote a temporal or chronological order. Rather, these terms are used only to distinguish one from another. For example, a first information can be termed a second information, and similarly, a second information can also be termed a first information, without departing from the scope of the present disclosure. Depending on the context, the word "if' and "when' as used herein can be interpreted to mean "upon determining" or "in response to determining."

[0145] To better describe any of the embodiments of the present disclosure, an embodiment of the present disclosure is exemplarily described in an application scenario of an access control.

[0146] Reference is made to Figure 1 , which shows a structural diagram of a wireless communication system according to an embodiment of the present disclosure. As shown in Figure 1 , the wireless communication system is a communication system based on cellular mobile communication technology, and the wireless communication system can include a plurality of terminals 11 and a plurality of base stations 12.

[0147] The terminal 11 can be a device that provides voice and / or data connectivity to a user. The terminal 11 can communicate with one or more core networks via a Radio Access Network (RAN), and the terminal 11 can be an Internet of Things terminal, such as a sensor device, a mobile phone (or so-called "cellular" phone), and a computer with an Internet of Things terminal, for example, which can be fixed, portable, pocket, hand-held, computer-embedded, or vehicle-mounted. For example, a Station (STA), a subscriber unit, a subscriber station, a mobile station, a mobile, a remote station, an access point, a remote terminal, an access terminal, a user terminal, a user agent, a user device, or a user equipment. Alternatively, the terminal 11 can also be a device of an unmanned aerial vehicle. Alternatively, the terminal 11 can also be a vehicle-mounted device, which can be a vehicle-mounted computer with wireless communication function or a wireless terminal of an external vehicle-mounted computer. Alternatively, the terminal 11 can also be a roadside device, which can be a street lamp, a signal lamp, or other roadside devices with wireless communication function, etc.

[0148] The base station 12 can be a network-side device in a wireless communication system. The wireless communication system can be a 4th generation mobile communication (4G) system, also known as a Long Term Evolution (LTE) system, or the wireless communication system can also be a 5G system, also known as a new radio (NR) system or a 5G NR system. Alternatively, the wireless communication system can also be a next generation of 5G system. In the 5G system, the access network can be referred to as an NG-RAN (New Generation-Radio Access Network).

[0149] The base station 12 can be an evolved NodeB (eNB) in a 4G system. Alternatively, the base station 12 can be a base station (gNB) in a 5G system using a centralized and distributed architecture. When the base station 12 uses a centralized and distributed architecture, it generally includes a central unit (CU) and at least two distributed units (DUs). The central unit is provided with a protocol stack of a packet data convergence protocol (PDCP) layer, a radio link control (RLC) layer, and a media access control (MAC) layer; and the distributed unit is provided with a protocol stack of a physical (PHY) layer. The specific implementation of the base station 12 is not limited in the embodiments of the present disclosure.

[0150] The base station 12 and the terminal 11 can establish a wireless connection through a wireless air interface. In different embodiments, the wireless air interface is a wireless air interface based on a fourth generation mobile communication network technology (4G) standard; or the wireless air interface is a wireless air interface based on a fifth generation mobile communication network technology (5G) standard, such as a new radio interface; or the wireless air interface can also be a wireless air interface based on a more next generation mobile communication network technology standard of 5G.

[0151] In some embodiments, the terminals 11 can also establish an E2E (End to End) connection. For example, V2V (vehicle to vehicle) communication, V2I (vehicle to infrastructure) communication, and V2P (vehicle to pedestrian) communication in vehicle to everything (V2X) communication, and the like.

[0152] In some embodiments, the wireless communication system can also include a network management device 13.

[0153] A plurality of base stations 12 are connected to a network management device 13. The network management device 13 can be a core network device in a wireless communication system, for example, the network management device 13 can be a Mobility Management Entity (MME) in an Evolved Packet Core (EPC). Alternatively, the network management device can also be other core network devices, such as a Serving GateWay (SGW), a Public Data Network GateWay (PGW), a Policy and Charging Rules Function (PCRF), or a Home Subscriber Server (HSS), etc. The implementation form of the network management device 13 is not limited in the embodiments of the present disclosure.

[0154] In the artificial intelligence model training based on the above wireless communication system, federated learning can well solve the data privacy problem. Federated learning is a machine learning framework that can effectively help multiple institutions to use data and machine learning modeling under the requirements of user privacy protection, data security and government regulations. It can directly train the model at the user end and only transmit the model training result, thereby well protecting the data privacy of the user. With the increase in the number of wireless network users, the amount of data transmitted by the network also increases at an amazing speed. How to reduce the loss of model accuracy in wireless transmission and develop more personalized model solutions for each user so that the model is more suitable for the specific requirements of the user is an important topic worthy of study.

[0155] As shown in FIG. 1, the embodiments of the present disclosure provide a data processing method applied to a base station, comprising: Figure 2

[0156] Step S101, determining the local data set distribution characteristics of at least one UE;

[0157] Step S102, based on the local data set distribution characteristics, scheduling a target UE participating in federated learning from the at least one UE.

[0158] In the embodiments of the present disclosure, the base station can establish a communication connection with a plurality of UEs and schedule at least part of the UEs to participate in federated learning. The UE local data needs to meet certain requirements in the process of federated learning, for example, the amount of data and the number of data types of data related to federated learning, etc. Therefore, the base station can determine which UEs to schedule to participate in federated learning according to the local data set distribution characteristics of the plurality of UEs.

[0159] ​Here, the local data set distribution characteristic is a distribution characteristic of a data set generated by the UE locally in use related to federated learning. The distribution of data types or the distribution of data amounts of different data types, etc. can be included.

[0160] In the embodiments of the present disclosure, the base station schedules the UE for federated learning, which can directly determine whether to schedule the UE by receiving the local data set distribution characteristic reported by the UE, or can obtain the distribution characteristic by processing part of the parameters of the local data set reported by the UE.

[0161] In an embodiment, the base station can issue scheduling information to the target UE scheduled to participate in federated learning, for the UE to determine its participation in federated learning. In another embodiment, the base station can also directly issue initial model information to the target UE scheduled to participate in federated learning, so that the UE determines its participation in federated learning and starts federated learning based on the initial model information.

[0162] In this way, the base station can use the local data set distribution characteristic of the UE to schedule the UE that can participate in federated learning from multiple alternative UEs that have a communication connection with the base station, and jointly perform federated learning with the UE participating in federated learning. Thus, direct scheduling of the base station can be realized, without the need for the UE to report data to the core network or the data center, which can greatly improve the training efficiency.

[0163] In some embodiments, the scheduling of the target UE participating in federated learning from the at least one UE based on the local data set distribution characteristic comprises:

[0164] Obtaining distribution difference statistical information of the local data set of each of the at least one UE and the global data set;

[0165] Scheduling the target UE participating in federated learning from the at least one UE according to the distribution difference statistical information.

[0166] In the embodiments of the present disclosure, the base station can determine the above distribution difference statistical information by the above local data set distribution characteristic of each of the at least one UE and the global data set distribution characteristic of the base station, and schedule according to the distribution difference statistical information.

[0167] Here, since the base station can interact with multiple UEs, multiple UEs can participate in the federated learning corresponding to the base station. There is a difference in the probability distribution between the local data set of each of the at least one UE and the data set of the multiple UEs associated with the base station or the global data set obtained by operation or processing. Therefore, the above distribution difference statistical information can be used to determine which UEs can participate in federated learning, and then schedule the UEs.

[0168] Here, the distribution difference statistical information refers to a difference between a distribution of various data types or data values in the local data set and the global data set. Since the global data set is at least composed of data of a plurality of at least one UE associated with the base station, the distribution of the global data set reflects the overall distribution of the data. Therefore, there is a difference between the distribution of the local data set of each at least one UE and the distribution of the global data set. In the embodiment of the present disclosure, the above-mentioned distribution difference statistical information is used to reflect the above-mentioned difference.

[0169] Here, the above-mentioned distribution can include a type distribution of data in the data set, a proportion of each type of data, or a data amount of different types of data. For the distribution difference statistical information, it is a difference between the above-mentioned data types of the local data set, a difference between the proportions of each data type, or a difference between the data amounts of each data type.

[0170] Exemplarily, the above-mentioned distribution can be a probability distribution of each data type in the data set. The probability distribution of the local data set obtained by the UE segment statistics is denoted as P(X m )=[P(x1),P(x2),…,P(x n )], where P(x i ) represents a probability of X m taking event x i . The base station performs statistics on the global data set distribution based on the statistical results of the probability distribution of the local data set reported by each UE, and the probability distribution is denoted as P(X g )=ΣP(X m ). The base station can obtain the above-mentioned distribution difference statistical information according to the probability distribution of the above-mentioned UE and the global probability distribution, which is denoted as ΔP m =||P(X g )-P(X m )||. Its meaning can be a difference in numerical value of the probability distribution of each data type, or a difference in data types contained in the probability distribution. The base station can schedule the corresponding UE according to the numerical value of the distribution difference statistical information.

[0171] It should be noted that the base station can receive the local data set reported by each at least one UE, and then obtain the above-mentioned global data set by statistics. Then, the probability distribution is calculated respectively, and then the above-mentioned distribution difference statistical information is obtained. The base station can also directly receive the probability distribution obtained by the local data set reported by each at least one UE, calculate the probability distribution of the global data set, and then obtain the distribution difference statistical information.

[0172] In an embodiment, the federated learning generates a local data set by sensing and collecting data of the UE, and processes the local data set to generate a local training set; the UE randomly initializes local model parameters and performs local learning model training using the local training set, and uploads the training result to the core network or the data center; the base station requests the local training result of the UE from the core network or the data center, and performs federated average learning using the local learning result of each UE to obtain an update result of the global learning model; the base station feeds back the update result to the UE through the network, and the UE fine-tunes the local model according to the feedback result; the above process is repeated until the model accuracy meets the requirements. When the model training is completed, each base station reports the model training result and the training data statistical characteristics to the network, and the network selects a suitable model according to the task data characteristics.

[0173] The data interaction between the base station and the UE needs to be performed through the core network or the data center, the UE needs to upload the training result data to the core network or the data center, and the base station requests the data. This way does not support direct federated learning between the base station and the UE, reduces the efficiency of federated learning and the utilization rate of wireless network resources. Moreover, the model training is not performed after data adaptation according to different UE requirements, which causes insufficient model accuracy.

[0174] To this end, in the embodiment of the present disclosure, the above steps are used to directly schedule the UE to participate in federated learning by the base station, without the need for the UE to report data to the core network or the data center, which can greatly improve the training efficiency. Moreover, the base station schedules the corresponding UE according to the probability distribution of each at least one UE data set, which takes into account the adaptability of different UEs to participate in federated learning, thereby facilitating the improvement of the accuracy of model training.

[0175] In some embodiments, the method further comprises:

[0176] obtaining the capability information of the at least one UE;

[0177] The scheduling of the target UE participating in federated learning from the at least one UE based on the distribution characteristics of the local data set comprises:

[0178] Scheduling the target UE participating in federated learning from the at least one UE according to the distribution characteristics of the local data set and the capability information of the at least one UE.

[0179] In the embodiment of the present disclosure, the capability of the at least one UE participating in federated learning can also be considered, and the target UE participating in federated learning is determined based on the above distribution characteristics of the local data set and the capability information.

[0180] The capability information of the at least one UE can be a capability of the at least one UE for federated learning, and can include whether the at least one UE can meet a processing capability required by federated learning if participating in federated learning. If the capability of the at least one UE is insufficient for federated learning, the at least one UE cannot be scheduled to participate in federated learning even if a distribution characteristic of a local data set of the at least one UE meets a requirement of the base station.

[0181] In an embodiment, a performance requirement of a model required by federated learning can also be considered, and if the at least one UE itself cannot meet the performance requirement of the model, the at least one UE cannot be scheduled to participate in federated learning. If the at least one UE meets the performance requirement of the model and the capability information of the at least one UE meets the processing capability required by federated learning, the base station can determine whether to schedule the at least one UE according to distribution difference statistical information corresponding to the at least one UE.

[0182] In this way, the base station judges whether the UE can be scheduled to participate in federated learning from multiple aspects such as data of the UE, processing capability, and performance requirement of the model, thereby improving efficiency and accuracy of federated learning and improving adaptability of the model to user data.

[0183] In some embodiments, the capability information of the at least one UE includes at least one of the following:

[0184] The capability information of the at least one UE includes at least one of the following:

[0185] The capability information of the at least one UE includes at least one of the following:

[0186] The capability required by the at least one UE in federated learning can include computing capability of the UE. Since the UE needs to collect a large amount of data and perform model training according to the data when performing federated learning, the UE that does not have sufficient computing capability cannot perform the federated learning.

[0187] The capability information can also include communication capability of the UE. In the process of federated learning, the UE needs to report training results to the base station and receive an updated model issued by the base station, and the like. Therefore, in order to ensure efficiency and accuracy of federated learning, communication capability of the UE participating in federated learning and communication channel condition between the UE and the base station also need to be considered.

[0188] The communication capability of the UE can be device hardware capability of the UE itself, for example, network type supported by the UE, bandwidth, and the like. The communication channel condition of the UE is a condition of a channel established between the UE and the base station, including bandwidth, transmission rate, congestion condition, and interference condition of the channel, and the like.

[0189] In some embodiments, the communication condition information comprises channel quality indicator (CQI) information detected by the UE.

[0190] In the embodiments of the present disclosure, the base station can learn the communication channel condition of the UE by obtaining the CQI information. The CQI is measured by the UE and can include downlink channel quality or uplink channel quality.

[0191] Since the CQI is carried by only a few bits, the base station can quickly and easily learn the basic condition of the communication channel corresponding to the UE by obtaining the CQI detected by the UE, and schedule the UE according to the value of the CQI.

[0192] In some embodiments, the method further comprises determining a weight coefficient of the target UE in the federated learning according to distribution difference statistical information of a local data set of the target UE and a global data set of the base station.

[0193] Considering that different UEs have different data characteristics, the importance of UEs to the global is different in the process of federated learning, therefore, the base station can determine the weight coefficient corresponding to each target UE according to the above distribution difference statistical information of each UE.

[0194] In this way, in the process of federated learning, the related model parameters of each UE are processed according to the weight coefficient of each target UE, and a final training result of federated learning is trained. In this way, the adaptability of the model obtained by federated learning to each UE can be improved, and a more accurate model can be obtained.

[0195] In some embodiments, the distribution difference statistical information comprises a probability distribution difference.

[0196] The determination of the weight coefficient of the target UE according to the distribution difference statistical information of the local data set stored locally by the target UE and the global data set comprises:

[0197] The weight coefficient of the target UE is determined according to the probability distribution difference corresponding to the single target UE and the sum of the probability distribution differences of all target UEs performing the same federated learning.

[0198] In the embodiments of the present disclosure, the base station can obtain the probability distribution of the local data set of each target UE, and obtain the probability distribution of the global data set according to the probability distribution of each target UE. There is a difference between the probability distribution of each target UE and the probability distribution of the global data set. The probability distribution difference, i.e., the difference between the probability distribution of the data in the local data set of the single target UE and the probability distribution of the data in the global data set of the base station, can be the difference in numerical value of each data type probability distribution, or the difference in data types contained in the probability distribution, etc.

[0199] Each target UE of the base station has a respective local data set, and thus, there is a corresponding probability distribution difference for each target UE. Here, the sum of the probability distribution differences can be obtained by summing the probability distribution differences corresponding to each target UE of the base station.

[0200] Therefore, the base station can determine the distribution difference statistics information by counting the above-mentioned probability distribution differences corresponding to each target UE and the sum of the above-mentioned probability distribution differences.

[0201] For example, the weight coefficient of a user in federated average learning can be calculated according to the distribution difference statistics information of the local data set and the global data set of each target UE, which can be represented by the following formula (1):

[0202]

[0203] wherein M represents the total number of target UEs participating in federated learning, a m represents the weight of the local learning model of user m in federated average processing, ΔP m represents the probability distribution difference between the local data set and the global data set of each user.

[0204] In some embodiments, the method further comprises:

[0205] receiving the model information of the local model reported by the target UE for performing the federated learning;

[0206] determining the weight coefficient of the target UE according to the distribution difference statistics information of the local data set stored locally by the target UE and the global data set;

[0207] performing weighted average on the local models of multiple target UEs according to the weight coefficient of the target UE and the model information of the local model, to obtain a global learning model.

[0208] In the process of federated learning, each target UE locally trains a model using a local data set, and reports the training result, including model information such as model parameters, to the base station. The base station then trains a global model according to the model information reported by each target UE to obtain the above-mentioned global learning model.

[0209] In addition, since in the process of the above-mentioned federated learning, each target UE only needs to report its own training result, i.e., model information, to the base station, the local data of the target UE itself will not be reported to the base station, thereby reducing the risk of leakage of personal information and other private data.

[0210] In some embodiments, the method further comprises:

[0211] In response to the global learning model satisfying the OAM subscription requirement, stop receiving the model information of the local model of the target UE for reporting the federated learning.

[0212] Since multiple target UEs and base stations jointly participate in the process of federated learning, multiple target UEs constantly acquire local data, perform federated learning, update local models, and report model information to the base station; the base station performs global learning according to the acquired model information of the local models of the target UEs, thereby training a global model, and then the base station can distribute the global model to each target UE for updating the local model.

[0213] Therefore, the above process of federated learning can be regarded as a cyclic interaction process between the base station and each target UE. In the embodiments of the present disclosure, whether the federated learning process can be stopped can be determined by considering the OAM subscription requirement of the current federated learning corresponding service.

[0214] The OAM subscription requirement contains specific requirements for the model accuracy required by the subscribed service. Therefore, when the global learning model satisfies the OAM subscription requirement, it means that the current global learning model has reached sufficient accuracy, so the federated learning can be stopped, and a usable global learning model can be obtained.

[0215] In some embodiments, the method further comprises:

[0216] In response to the global learning model not satisfying the OAM subscription requirement, sending the model information of the global learning model to the target UE;

[0217] Receiving the model information of the local model updated by the target UE according to the global learning model;

[0218] Updating the global learning model according to the local model updated by the target UE and the weight coefficient corresponding to the local model.

[0219] Here, after the base station acquires the model information of the local model of each target UE, it performs global learning to obtain a global learning model. If the global learning model does not satisfy the OAM subscription requirement, the model information of the global learning model can be sent to each target UE to facilitate the target UE to update the local model.

[0220] It should be noted that during the process of updating the local model, the local data set and the global data set of the base station can also change, and therefore the weight coefficient can also be updated. That is, during the process of federated learning, the weight coefficient is also constantly updated along with the update of the model.

[0221] Then the target UE continues federated learning according to the updated local model, obtains updated model information, and reports the model information to the base station. In this way, the base station and the UE form a cycle of federated learning interaction until the final global learning model meets the OAM subscription requirements.

[0222] In some embodiments, the method further comprises:

[0223] reporting model information of the global learning model and training data for training the global learning model to an OAM;

[0224] receiving model parameters determined by the OAM based on the model information of the global learning model, the training data, and task data of the OAM;

[0225] updating the global learning model according to the model parameters.

[0226] In the embodiments of the present disclosure, the base station can obtain the subscription requirements for terminating federated learning through the OAM, and in addition, the OAM can also update the target global learning model based on the global learning model obtained through federated learning.

[0227] The OAM can be an operation and maintenance management module applied to a core network, and based on each cell corresponding to a service, collect training data of global learning models corresponding to federated learning of different base stations. The OAM can obtain task data reported by task cells through each base station, and based on the task data and the training data of each base station, determine corresponding probability distribution difference information, and issue model information of training models obtained by fusing multiple base stations to the base station, so that the base station updates the global learning model according to the received model parameters.

[0228] In some embodiments, the method further comprises:

[0229] In response to detecting that the base station connected by the target UE is switched, determining that the target UE exits the federated learning.

[0230] In the embodiments of the present disclosure, the target UE needs to maintain communication connection with the base station when performing federated learning, so as to maintain data interaction. Therefore, if the base station detects that the base station connected by the target UE participating in federated learning is switched, for example, the UE performs cell reselection, etc., the federated learning result of the target UE cannot be continued to be used. Therefore, at this time, the base station can determine that the target UE exits the above federated learning.

[0231] For example, if the base station detects that the base station connected by the target UE is switched, the weight coefficient corresponding to the target UE is adjusted to 0, so that the base station will not continue to issue model information to the UE, and will not continue to receive the model information reported by the UE, etc.

[0232] The embodiments of the present disclosure further provide a data processing method, which is applied to a base station and includes the following steps:

[0233] obtaining distribution difference statistical information of a local data set of each of at least one UE and a global data set;

[0234] scheduling a target UE participating in federated learning from the at least one UE according to the distribution difference statistical information.

[0235] Here, since the base station can interact with multiple UEs, multiple UEs can participate in the federated learning corresponding to the base station. The local data set of each candidate UE is different from the probability distribution of the data in the data set of the multiple UEs associated with the base station or the global data set obtained through operation and the like, so the distribution difference statistical information can be used to determine which UEs can participate in the federated learning, and then schedule the UEs.

[0236] Here, the distribution difference statistical information refers to the difference between the distribution of various data types or data values in the local data set and the global data set. Since the global data set is at least composed of the data of multiple candidate UEs associated with the base station, it reflects the overall distribution of the data. Therefore, there is a difference between the distribution of the local data set of each candidate UE and the distribution of the global data set. In the embodiments of the present disclosure, the distribution difference statistical information is used to reflect the above difference.

[0237] Here, the above distribution can include the type distribution of the data in the data set, the proportion of each type of data, or the data amount of different types of data. For the distribution difference statistical information, it is the difference of the above data types of the local data set, the difference of the proportion of each data type, the difference of the data amount of each data type, and the like.

[0238] The base station can directly receive the distribution difference statistical information reported by the UE. For example, the base station issues the distribution information of the global data set to the UE, and the UE determines the above distribution difference statistical information according to the distribution characteristics of the local data set and the received distribution information of the global data set, and reports it to the base station. The base station can also receive the distribution information of the local data set reported by the UE, and determine the above distribution difference statistical information according to the distribution information of the UE and the distribution information of the global data set, and then determine whether to schedule the UE to participate in the federated learning.

[0239] In an embodiment, the base station scheduling the UE to participate in the federated learning can issue scheduling information to the UE, indicating that the UE participates in the federated learning.

[0240] The embodiment can be independently executed, or can be combined with any at least one of the above-mentioned embodiments. Any at least two of the above-mentioned embodiments of the present disclosure can also be split and combined, and the order between steps can be adjusted according to actual application scenarios, which is not limited here.

[0241] As shown in Figure 3 The embodiment of the present disclosure provides a data processing method, applied to a UE, comprising:

[0242] Step S201, receiving scheduling information issued by a base station according to a local data set distribution characteristic of the UE; wherein the scheduling information is used to determine whether the UE is a target UE scheduled to participate in federated learning.

[0243] In the embodiment of the present disclosure, the UE can report its own local data set distribution characteristic to the base station, or can report part of the data in the local data set to the base station to determine the above-mentioned distribution characteristic and issue scheduling information.

[0244] After the UE receives the scheduling information, it can know that it has been scheduled by the base station to participate in federated learning as a target UE, and can perform the above-mentioned federated learning according to the local data set.

[0245] In the embodiment of the present disclosure, the base station can establish a communication connection with multiple UEs, and schedule at least part of the UEs to participate in federated learning. In the process of federated learning, the UE local data needs to meet certain requirements, for example, the amount of data and the number of data types of data related to federated learning, etc. Therefore, the base station can determine which UEs to schedule to participate in federated learning according to the local data set distribution characteristics of multiple UEs.

[0246] Here, the local data set distribution characteristic is the distribution characteristic of the data set related to federated learning generated by the UE locally in the use process. It can include the distribution of data types or the distribution of data amounts of different data types, etc.

[0247] In some embodiments, the receiving scheduling information issued by the base station according to the local data set distribution characteristic of the UE comprises:

[0248] Receiving scheduling information issued by the base station according to the distribution difference statistical information obtained by the local data set distribution characteristic of the UE and the global data set distribution characteristic.

[0249] In the embodiments of the present disclosure, the UE can report the probability distribution information of its local data set to the base station, so that the base station determines the distribution difference statistical information. The UE can also receive the probability distribution information of the global data set issued by the base station, and determine the distribution difference statistical information and report it to the base station. Since the base station can determine whether to schedule the candidate UE as the target UE according to the distribution difference statistical information, after the UE receives the scheduling information, it can know that it has been scheduled by the base station as the target UE participating in federated learning, and can perform the federated learning according to the local data set.

[0250] Of course, if the UE does not receive the scheduling information, it does not participate in federated learning.

[0251] In some embodiments, the method further comprises:

[0252] reporting capability information; wherein the capability information is used for the base station to issue the scheduling information according to the local data set distribution characteristics and the capability information.

[0253] Since the capability of the candidate UE participating in federated learning can also be considered, and the target UE participating in federated learning is determined based on the local data set distribution characteristics and the capability information. Therefore, the UE can report its capability information to the base station, so that the base station determines whether the UE meets the requirements of federated learning.

[0254] It should be noted that the UE can report its capability information after establishing a communication connection with the base station, or report the capability information based on the request of the base station. After reporting the capability information, if the scheduling information of the base station is received, it can participate in federated learning. If no scheduling information is received, it does not participate in federated learning.

[0255] In some embodiments, the capability information comprises at least one of:

[0256] The computing capability information indicates the computing capability of the UE;

[0257] The communication condition information indicates the communication capability and / or communication channel condition of the candidate UE.

[0258] The capability required by the candidate UE in federated learning can include the computing capability of the UE. Since the UE needs to collect a large amount of data when performing federated learning, and train the model according to the data, the UE without sufficient computing capability cannot perform the federated learning.

[0259] The capability information can further include a communication capability of the UE. During the federated learning, the UE needs to report training results to the base station and receive an updated model issued by the base station, and the like. Therefore, in order to ensure the efficiency and accuracy of the federated learning, the communication capability of the UE participating in the federated learning and the communication channel status between the UE and the base station also need to be considered.

[0260] Here, the communication capability of the UE can be a device hardware capability of the UE itself, for example, a network type supported by the UE, a bandwidth, and the like. The communication channel status of the UE is a status of a channel established between the UE and the base station, including a bandwidth, a transmission rate, a congestion status, an interference status, and the like of the channel.

[0261] In some embodiments, the communication status information includes CQI information; and the method further includes:

[0262] detecting the CQI information of the channel between the UE and the base station.

[0263] Whether the UE can participate in the federated learning needs to consider the communication status between the UE and the base station. Therefore, the UE can detect the CQI information of the channel in real time and report the CQI information to the base station, so that the base station determines whether the communication channel between the UE and the base station meets the requirements of the federated learning.

[0264] In some embodiments, the method further includes:

[0265] reporting model information of a local model of the UE; wherein the local model is used for the base station to perform the federated learning according to the local model and according to a weight coefficient of the UE; and wherein the weight coefficient of the UE is determined by the base station according to distribution difference statistical information of a local data set of the UE and a global data set of the base station.

[0266] If the UE is scheduled to participate in the federated learning, the UE can perform training of a local model according to a local data set and report model information obtained by the training to the base station. In this way, the base station can perform training of a global learning model according to the model information reported by the UE and other UEs.

[0267] It should be noted that the federated learning is a model training process jointly performed by the base station and the UEs. The UE performs training of a local model locally, and after reporting the training result to the base station, the base station performs weighted average processing on the results reported by the UEs and the weight coefficients of the UEs, thereby obtaining a global learning model.

[0268] In some embodiments, the method further includes:

[0269] generating the local data set according to collected wireless network data;

[0270] extract data of the local data set to generate a local training data set;

[0271] perform model training by using the local training data set to obtain the local model.

[0272] In the process of federated learning, the UE needs to use locally collected data. The locally collected data can be wireless network data, i.e., data generated by the user in the process of service use. The UE generates a local data set according to the collected data, and if the data quantity of the local data set is large, data extraction can also be performed, for example, part of the data is extracted by using a sampling method as a local training data set. In some embodiments, if the data quantity of the local data set is small, the local data set can also be directly used as the training data set.

[0273] By using the above local training data set and the initial model obtained from the base station, the above federated learning, i.e., model training, can be performed to obtain the above local model.

[0274] When reporting, the UE can only report the model information of the local model, without reporting the local data, so as to reduce the possibility of privacy leakage and improve the model training efficiency.

[0275] In some embodiments, the method further includes:

[0276] receiving model information of a global learning model issued by the base station;

[0277] performing the federated learning according to the model information of the global learning model to obtain an updated local model;

[0278] in response to the global learning model not meeting the OAM subscription requirement, reporting model information of the updated local model.

[0279] In the process of federated learning, the base station also issues model information of a global learning model obtained according to model information of each UE to each UE. Therefore, after receiving the model information of the global learning model, the UE can continue to perform federated learning to update the model information of the local model. Then, the UE continues to report to the base station until the base station obtains a global learning model meeting the OAM subscription requirement.

[0280] In some embodiments, the method further includes:

[0281] in response to the global learning model meeting the OAM subscription requirement, stopping the federated learning.

[0282] The process of the federated learning can be regarded as a process of cyclic interaction between the base station and the target UEs. In the embodiments of the present disclosure, whether the federated learning process can be stopped can be determined by considering the OAM subscription requirement of the current federated learning corresponding service.

[0283] The OAM subscription requirement contains an analysis ID list, i.e., the IDs of different analysis types. The analysis ID list specifies the required model accuracy of the service. Therefore, when the global learning model meets the OAM subscription requirement, it means that the current global learning model has reached sufficient accuracy, and thus the federated learning can be stopped, and the global learning model available for use can be obtained.

[0284] Here, the UE can also obtain the final global learning model for local use.

[0285] In some embodiments, the method further includes:

[0286] In response to handover of the base station to which the UE is connected, stopping the federated learning.

[0287] In the embodiments of the present disclosure, the target UE needs to maintain the communication connection with the base station when performing the federated learning, so as to maintain the data interaction. Therefore, if the base station to which the UE is connected is handed over, for example, the UE performs cell reselection, etc., the federated learning with the base station scheduling the UE cannot continue. Therefore, the UE can exit the federated learning at this time.

[0288] The embodiments of the present disclosure also provide a data processing method applied to a UE, and the method includes:

[0289] Reporting a local data set distribution characteristic, wherein the local data set distribution characteristic is used for the base station to determine whether to schedule the UE to perform the federated learning.

[0290] In the embodiments of the present disclosure, the base station can establish a communication connection with multiple UEs, and schedule at least part of the UEs to perform the federated learning. The UE local data needs to meet certain requirements in the process of the federated learning, for example, the data volume and the number of data types of the data related to the federated learning, etc. Therefore, the base station can determine which UEs to schedule to participate in the federated learning according to the local data set distribution characteristics of the multiple UEs.

[0291] Here, the local data set distribution characteristic is the distribution characteristic of the data set related to the federated learning generated by the UE locally in the use process. It can include the distribution of data types or the distribution of data volumes of different data types, etc.

[0292] It should be noted that if the UE reports the local data set distribution characteristics, the base station determines to schedule the UE for federated learning, and the UE can determine to participate in federated learning according to the received indication of the base station, such as scheduling information, and perform model training based on the data in the local data set.

[0293] The embodiment can be independently executed or combined with any at least one of the above-mentioned embodiments. Any at least two of the above-mentioned embodiments of the present disclosure can also be split and combined, and the order of the steps can be adjusted according to the actual application scene, which is not limited here.

[0294] The embodiment of the present disclosure also provides the following examples:

[0295] In the embodiment of the present disclosure, a data processing method is provided:

[0296] The UE perceives and collects wireless network data, generates a local data set, and statistically analyzes the distribution characteristics of the local data set and transmits the distribution characteristic information to the base station through a wireless channel, while reporting the computing capacity and real-time communication condition of the UE to the base station.

[0297] The base station receives the information transmitted by the UE, and statistically analyzes the distribution difference between the local data set of the UE and the global data set. The base station schedules the UE based on the statistical result of the distribution difference, the computing capacity and real-time communication condition reported by the UE, and the performance requirement of the learning model, and determines whether the UE participates in this federated learning. After each UE receives the UE scheduling result, if participating in learning, the local data set is randomly sampled to generate a local training data set, and the parameters of the local learning model are randomly initialized.

[0298] The UE and the base station start federated learning. In each round of iterative learning, the UE determines the number of local learning model training according to the current available computing power and real-time communication condition, and transmits the result to the base station through a wireless channel after completing the local learning model training. The base station updates the weight coefficient of the UE in the federated average according to whether the UE switches and the statistical result of the distribution difference between the local data set of each UE and the global data set, and when the UE switches in the training, the UE can directly exit the federated learning process, and the base station updates the weight coefficient of the exiting UE to zero. The base station performs federated average learning to obtain the update result of the global learning model. The base station feeds back the update result of the global learning model to the UE through a wireless channel, and the UE updates the local learning model according to the result;

[0299] The base station monitors the training effect of the global learning model, and when the model of a certain round of update meets the subscription demand of the OAM, a signal is sent to each UE to terminate the training, and the federated learning is ended. After the model training is completed, each base station reports the model training result and training data statistical characteristics to the OAM, and the OAM selects a suitable model according to the task data characteristics.

[0300] The embodiments of the present disclosure also provide a federated learning system, comprising one base station device and M user terminal devices, the base station device and the user terminal devices communicate through a wireless channel. The base station device belongs to a functional unit of the base station, and the user terminal is a terminal accessing the base station, and the base station coordinates the terminals to participate in wireless federated learning.

[0301] The base station device for federated learning provided by the embodiments of the present disclosure is responsible for: according to the UE reported data, the distribution difference of the UE local data set and the global data set is counted; according to the data distribution difference, the computing ability of UE, the communication condition, the learning model performance requirement, the UE is scheduled; according to the distribution characteristics of the UE local data set, the weight coefficient of the UE in the federated learning is calculated; the federated average learning is carried out, and the global learning model is updated.

[0302] The base station device for federated learning provided by the embodiments of the present disclosure specifically comprises:

[0303] The base station communication module is used for data transmission and control signaling interaction with the UE through the wireless channel, and the UE is controlled;

[0304] The user scheduling module is used for scheduling the UE according to the data, the computing ability, the communication condition, the learning model performance requirement and other information sent by the UE;

[0305] The model calculation and processing module is used for federated averaging of the local learning model training and updating results fed back by the UE, and generating the averaged global learning model;

[0306] The transmission control module is used for specifying the data transmission scheme according to the characteristics of the data to be transmitted and the wireless communication condition;

[0307] The model updating module is used for updating the generated averaged global learning model, and transmitting the updating result of the global learning model to the UE through the wireless channel.

[0308] The user device for federated learning provided by the embodiments of the present disclosure is responsible for: sensing and collecting local wireless network data, and counting the distribution characteristics of the collected data; randomly and uniformly sampling the collected data to generate a local training set, and initializing a local learning model; determining the local training times according to the real-time communication condition and the computing ability and performing local training; updating the local learning model according to the global learning model updating result.

[0309] The user device for federated learning provided by the embodiments of the present disclosure specifically comprises:

[0310] The user communication module is used for data transmission and control signaling interaction with the base station through the wireless channel;

[0311] Data sensing and storage module, for sensing and collecting UE generated data, generating UE local training dataset, and storing the dataset;

[0312] Model training and calculation module, for training and updating local learning model using UE locally sensed and stored data;

[0313] Transmission control module, for specifying data transmission scheme according to to-be-transmitted data characteristics and wireless communication conditions.

[0314] As Figure 4 For the embodiments of the present disclosure, the principle diagram of federated learning is as shown in Figure 4 The system of wireless federated learning includes one base station device 10 and M user devices 20 (i.e. the above-mentioned UEs), and the user devices 20 and the base station device 10 communicate through a wireless channel. Each UE stores its own local data, and the base station device stores a large amount of data. The base station device in the present disclosure belongs to a functional unit of a base station, the user device is a terminal accessing the base station, and the base station coordinates each terminal to perform wireless federated learning using local data.

[0315] As Figure 5 The user device 20 of the federated learning in the present disclosure specifically includes a user communication module 510, a data sensing and storage module 520, a model training and calculation module 530, and a transmission control module 540. The specific functions and architectures of each module are described as follows:

[0316] User communication module 510: the main function of this module is to interact with the base station through a wireless channel for data transmission and control signaling, mainly including a radio frequency functional unit and a baseband signal processing functional unit.

[0317] Data sensing and storage module 520: the main function of this module is to sense and collect UE generated data, generate UE local training dataset, and store the dataset, mainly including a data sensing functional unit and a data storage functional unit.

[0318] Model training and calculation module 530: the main function of this module is to train and update the local learning model using the UE locally sensed and stored data, mainly including a data cache functional unit and a data calculation processing functional unit.

[0319] Transmission control module 540: the main function of this module is to specify the data transmission scheme according to the to-be-transmitted data characteristics and wireless communication conditions.

[0320] As Figure 6As shown, the base station device 10 for federated learning in this embodiment of the present disclosure specifically includes a base station communication module 610, a user scheduling module 620, a model calculation and processing module 630, a transmission control module 640, and a model update module 650. The specific functions and architecture of each module are detailed below:

[0321] Base station communication module 610: The main function of this module is to transmit data and interact with the UE via a wireless channel, and to manage and control the UE. It mainly includes a radio frequency function unit, a baseband signal processing function unit, and a user management function unit.

[0322] User scheduling module 620: The main function of this module is to schedule the UE based on information such as data sent by the UE, computing power, communication conditions, and learning model performance requirements.

[0323] Model Calculation and Processing Module 630: The main function of this module is to perform joint averaging on the training and update results of the local learning model fed back by the UE, and generate an averaged global learning model. It mainly includes a model cache function unit and a model calculation and processing function unit.

[0324] Transmission control module 640: The main function of this module is to specify the data transmission scheme according to the characteristics of the data to be transmitted and the wireless communication conditions.

[0325] Model update module 650: The main function of this module is to update the generated average global learning model and transmit the updated result of the global learning model to the UE through the wireless channel. It mainly includes a discrimination function unit and an update function unit.

[0326] like Figure 7 The following is a flowchart of the data processing method in an embodiment of this disclosure:

[0327] In step S701, the UE senses and collects wireless network data through the data sensing and storage module, generates a local dataset, and performs statistical analysis on the probability distribution information of the local dataset.

[0328] In step S702, the UE reports the statistical results of the local dataset to the base station via the wireless channel, and also reports the computing power and communication conditions.

[0329] Step S703: The base station performs statistical analysis on the probability distribution of the global network dataset based on the statistical results reported by each UE, and records the difference in probability distribution between each UE's local dataset and the global network dataset. The specific steps are as follows:

[0330] Step S31: The UE calculates the probability distribution of its local dataset. Using the local dataset D of UEm... m For example, the probability distribution of its statistical local dataset is denoted as P(X).m = [P(x1),P(x2),…,P(x)] n )], where P(x i ) represents X m Get event x i The probability of.

[0331] Step S32: The base station performs statistical analysis on the distribution of the global network dataset based on the statistical results of the probability distribution of the local datasets reported by each UE. The probability distribution is denoted as P(X). g )=∑P(X m ).

[0332] Step S33: The base station calculates and records the probability distribution difference between the local dataset of each UE and the global dataset of the network, and denotes it as ΔP. m =||P(X) g )-P(X m )||.

[0333] In step S704, the base station schedules the UE based on the probability distribution difference between the UE's local dataset and the network global dataset, the UE's computing power and communication conditions, and the specific learning model performance requirements, determines whether the UE should participate in federated learning, and sends the UE scheduling result to the UE.

[0334] In step S705, the UE and the base station perform federated learning iteratively multiple times until the model meets the OAM subscription requirements.

[0335] In one embodiment, the UE generates a local training dataset and uses it to train and update the local learning model. The UE then transmits the training and update results of the local learning model to the base station via a wireless channel. The base station updates the weight coefficients of each UE in the federated averaging process based on whether a UE has switched over and the probability distribution differences between each UE's local dataset and the network's global dataset. When a UE switches over during training, it can directly exit the federated learning process, and the base station updates the weight coefficients of the exiting UE to zero. The base station performs federated averaging learning to obtain the updated results of the global learning model. The base station transmits the updated results of the global learning model to the UE via a wireless channel. The UE updates its local learning model based on these results, and this process is repeated iteratively until the global learning model meets the OAM subscription requirements.

[0336] Furthermore, such as Figure 8 As shown, the process of federated learning between the UE and the base station includes the following steps:

[0337] In step S801, the UE generates a local training dataset and initializes the parameters of the local learning model.

[0338] In an embodiment, the UE obtains a local training data set by uniformly randomly sampling the local data set. Taking the UE m as an example, the generated local training data set is denoted as D m The UE randomly initializes a set of model parameters as the initial parameters of the local learning model, and the initialized local learning model result is denoted as

[0339] In step S802, the UE performs local learning model training and transmits the training result to the base station through the wireless channel.

[0340] In an embodiment, the UE first determines the number of local training rounds K according to the currently available computing power and real-time communication conditions. Then, the UE performs K rounds of training updates on the local learning model using the local training data set. In each training process, the training of the local learning model is realized based on the stochastic gradient descent algorithm. Taking the UE m as an example, the model update result can be represented by the following formula:

[0341]

[0342] wherein, denotes the local learning model obtained in the kth round of training before the tth federal average, η denotes the learning rate, denotes the training data set with a data amount of N randomly sampled from the local data set D m in the kth round of updating the local learning model, x denotes the data in the training set , and G(·) denotes the empirical risk term.

[0343] The UE transmits the training update result of the local learning model after K rounds of training to the base station through the wireless channel. Taking the tth federal average process as an example, the training update result of the local learning model transmitted by each UE can be represented as

[0344] In step S803, the base station updates the weight coefficients of each UE in the federal average according to whether the UE has switched and the statistical distribution characteristics of the local data set of each UE. When the UE switches during training, the UE can directly exit the federal learning process, and the base station updates the weight coefficient of the exiting UE to zero.

[0345] In an embodiment, the weight coefficient of the UE in the federal average learning is calculated according to the probability distribution difference between the local data set of the UE and the global data set, which can be represented by the following formula:

[0346]

[0347] wherein, M denotes the total number of UEs participating in the federal learning, a m denotes the weight of the local learning model of the UE m in the federal average process, and ΔPm This represents the difference in probability distribution between each user's local dataset and the global network dataset.

[0348] In step S804, the base station performs federated averaging to obtain the updated result of the global learning model, and feeds back the updated result to the user through the wireless channel.

[0349] In one embodiment, the base station receives the training update results of the local learning models of all users, and performs a federated average based on the probability distribution differences between each user's local training dataset and the network's global dataset to obtain the update result of the global learning model. Taking the t-th federated average process as an example, the update result of the global learning model is:

[0350]

[0351] The base station transmits the updated results of the global learning model to all users via a wireless channel. Taking the t-th federated averaging process as an example, the updated result of the global learning model W sent by the base station is w. t , where t is the number of federated averaging processes, and K is the total number of rounds of local training performed when the UE reports model information of the local model.

[0352] Step S805: The user updates the local learning model based on the feedback results from the base station.

[0353] In one embodiment, the user receives the update result of the global learning model and updates the local learning model based on this result. Taking the t-th federated averaging process as an example, the update result of user m's local learning model is:

[0354] Step S806: Repeat steps S802 to S805 until the updated global training model meets the OAM subscription requirements. The federated learning process ends, and the final training result w of the global learning model is obtained. T .

[0355] The list of analytics IDs in the OAM subscription requirements sets specific requirements for model accuracy. After the global model is updated in each iteration, the base station compares the global model training results with the specific requirements in the OAM analytics ID list. If the requirements are met, the training is terminated.

[0356] like Figure 9 As shown, the network model selection in the data processing method provided in this embodiment includes the following steps:

[0357] Step S901: After completing the training, each base station reports the model accuracy and the distribution characteristics of the training data to OAM.

[0358] Step S902, the OAM perceives the task data and counts the probability distribution characteristics of the task data.

[0359] Step S903, the OAM counts the distribution difference information of the training data of each base station and the task data, and selects one or more base station models according to the statistical information and the model accuracy of each base station.

[0360] In an embodiment, the OAM can select the training models of multiple base stations, and use the multiple base station models after fusion.

[0361] Step S904, the OAM issues the model selection result to each base station, and each base station reports specific model parameter information according to the result.

[0362] As shown in Figure 10 The protocol and interface principle of the user scheduling part in the data processing method provided by the embodiment of the present disclosure mainly relates to the user end data perception and storage module, the user end model training and calculation module, the user end communication module, the base station end communication module, the base station end model training and calculation module, and the base station end user management module in the wireless federated learning device provided by the embodiment of the present disclosure, and specifically as follows:

[0363] S1. The user end data perception and storage module perceives the wireless network data, obtains a local data set, and counts the distribution characteristics of the local data set.

[0364] S2. The user end data perception and storage module sends the local data set distribution characteristic information signaling to the user end communication module, and this process and the corresponding signaling are newly added by the present application. The signaling indication content is: sending the user local data set distribution characteristic information to the receiver.

[0365] S3. The user end model training and calculation module sends the user's calculation capability and learning model performance requirement signaling to the user end communication module, and this process and the corresponding signaling are newly added by the present application. The signaling indication content is: sending the user calculation capability and learning model performance requirement to the receiver.

[0366] S4. The user end communication module encapsulates and packages the user data distribution characteristics, calculation capability, and learning model performance requirement as user scheduling information.

[0367] S5a. The user end communication module sends the user scheduling information data packet signaling to the base station end communication module, and this process and the corresponding signaling are newly added by the present application. The signaling indication content is: sending the user scheduling information data packet to the receiver.

[0368] S5b. The user end communication module sends the CQI[4] measurement and reporting signaling to the base station end communication module. The signaling indication content is: the user performs CQI measurement and reports the CQI information to the receiver.

[0369] S6. The base station end communication module sends a send user scheduling information signaling to the base station end model training and calculation module, this process and the corresponding signaling are newly added by the application, and the signaling indicates that the aggregated user scheduling information is sent to the receiving party, and at this time, the transmitted is the unpackaged data.

[0370] S7. The base station end model training and calculation module statistics the distribution difference between the local data set and the global data set.

[0371] S8. The base station end model training and calculation module reports the distribution difference statistics result to the base station end user management module through a signaling, this process and the corresponding signaling are newly added by the application, and the signaling indicates that the distribution difference statistics result is reported to the receiving party.

[0372] S9a. The base station end user management module sends a request user scheduling information and CQI information signaling to the base station end communication module, this process and the corresponding signaling are newly added by the application, and the signaling indicates that the aggregated user scheduling information and CQI information are requested.

[0373] S9b. The base station end communication module sends a send user scheduling information and CQI information signaling to the base station end user management module, this process and the corresponding signaling are newly added by the application, and the signaling indicates that the aggregated user scheduling information and CQI information are sent to the receiving party.

[0374] S10. The base station end user management module performs user scheduling according to the distribution difference statistics result, the learning model performance requirement, the calculation ability and the communication condition.

[0375] S11a. The base station end user management module sends a send user scheduling result signaling to the base station end communication module, this process and the corresponding signaling are newly added by the application, and the signaling indicates that the scheduling scheme result of each user is sent to the receiving party.

[0376] S11b. The base station end communication module sends a send user scheduling result signaling to the user end communication module.

[0377] S12a. The user end communication module sends a request to establish an RRC connection signaling to the target base station communication module, and the signaling indicates that the RRC connection with the target base station is requested to be established.

[0378] S12b. The base station communication module sends an establish an RRC connection signaling to the user end communication module, and the signaling indicates that the receiving party is notified that the RRC connection is agreed to be established.

[0379] S12c. The user end communication module sends an RRC connection establishment completion signaling to the target base station communication module, and the signaling indicates that the receiving party is notified that the RRC connection establishment is completed.

[0380] As Figure 11 shown, the protocol and interface principle of the part of federated learning between the user and the base station in the data processing method provided by the embodiments of the present disclosure mainly involve the user-side data sensing and storage module, the user-side model training and calculation module, the base station-side model training and calculation module, and the base station-side model updating module in the wireless federated learning device provided by the embodiments of the present disclosure, and are as follows:

[0381] S1. The user-side data sensing and storage module senses and collects the wireless network data to generate a local data set.

[0382] S2. The user-side data sensing and storage module sends the local data set signaling to the user-side model training and calculation module. This process and the corresponding signaling are newly added by the present disclosure, and the signaling indication content is: sending the user local data set to the receiver.

[0383] S3. The user-side model training and calculation module uniformly and randomly extracts the local data set to generate a local training data set.

[0384] S4. The user-side model training and calculation module randomly initializes the local learning model parameters and trains and updates the local learning model by using the local training data set.

[0385] S5. The user-side model training and calculation module sends the local learning model training result signaling to the base station-side model training and calculation module. This process and the corresponding signaling are newly added by the present disclosure, and the signaling indication content is: sending the local learning model training result to the receiver.

[0386] S6. The user-side data sensing and storage module sends the local data set distribution characteristic information signaling to the base station-side model training and calculation module.

[0387] S7. The base station-side model training and calculation module updates the weight coefficient of the federated average according to whether the user has switched and the data distribution statistical characteristics of each user.

[0388] S8. The base station-side model training and calculation module performs federated average learning to obtain the update result of the global learning model.

[0389] S9a. The base station-side model training and calculation module sends the global learning model update result signaling to the base station-side model updating module. This process and the corresponding signaling are newly added by the present disclosure, and the signaling indication content is: sending the global learning model update result to the receiver.

[0390] S9b. The base station-side model updating module updates the global learning model according to the model update result.

[0391] 9c. The base station model update module will send the global learning model update result signaling to the user-end model training and calculation module.

[0392] 9d. The user-side model training and computation module updates the local learning model based on the global learning model update results.

[0393] like Figure 12 As shown in this disclosure, the data transmission protocol and interface principle of a data processing method provided in this embodiment mainly involve a transmitter model training and calculation / model update module, a transmitter transmission control module, a transmitter communication module, a receiver communication module, and a receiver transmission control module. The data transmission involved in this disclosure embodiment is divided into two cases: one is that the user terminal transmits local learning model update parameters to the base station, in which case the transmitter is the user terminal and the receiver is the base station; the other is that the base station transmits global learning model update parameters to the user terminal, in which case the transmitter is the base station and the receiver is the user terminal. In the following description, "model parameters" refers to the local learning model update parameters and the global learning model update parameters. Specifically:

[0394] S1. The transmitter model training and calculation / model update module will send model parameter signaling to the transmission control module. The signaling content is: send updated model parameters to the receiver.

[0395] S2. The transmitter communication module will measure the CQI and send the signaling to the transmission control module.

[0396] S3. The transmitter transmission control module formulates a data transmission scheme based on communication conditions and model parameter characteristics.

[0397] S4. The transmitting end transmission control module sends the data transmission scheme information signaling to the transmitting end communication module. This process and the corresponding signaling are new additions to this invention. The signaling instruction content is: send the data transmission scheme information to the receiver, including information such as modulation method and code rate.

[0398] S5. The transmitter model training and calculation / model update module will send model parameter signaling to the transmitter communication module.

[0399] S6. The transmitter communication module encapsulates and packages the model parameters according to the data transmission scheme.

[0400] S7a. The transmitting end communication module sends the transmission model parameter data packet signaling to the receiving end communication module. The signaling indicates: Transmit the encapsulated and packaged model parameter data packet.

[0401] S7b. The receiving end communication module sends the sending model parameter signaling to the receiving end transmission control module. At this time, the data transmitted is the decapsulated data.

[0402] S7c. The receiving end transmission control module sends a notification of correct data reception signaling to the receiving end communication module, and the signaling indicates that the receiving end has received correct data.

[0403] S7d. The receiving end communication module sends a notification of correct data reception signaling to the sending end communication module.

[0404] As shown in Figure 13A The embodiments of the present disclosure further provide a data processing apparatus 1300 applied to a base station, comprising:

[0405] A first determination module 1301 configured to determine local data set distribution characteristics of at least one user equipment (UE),

[0406] A scheduling module 1302 configured to schedule target UEs participating in federated learning from the at least one UE based on the local data set distribution characteristics.

[0407] In some embodiments, the scheduling module 1302 comprises:

[0408] A first acquisition sub-module configured to acquire distribution difference statistical information of the local data set and the global data set of each UE in the at least one UE;

[0409] A first scheduling sub-module configured to schedule target UEs participating in federated learning from the at least one UE according to the distribution difference statistical information.

[0410] In some embodiments, the apparatus further comprises:

[0411] A first acquisition module configured to acquire capability information of the at least one UE;

[0412] The scheduling module 1302 comprises:

[0413] A second scheduling sub-module configured to schedule target UEs participating in federated learning from the at least one UE according to the local data set distribution characteristics and the capability information of the at least one UE.

[0414] In some embodiments, the capability information of the at least one UE comprises at least one of the following:

[0415] Computing capability information indicating computing capability of the UE;

[0416] Communication condition information indicating communication capability and / or communication channel condition of the UE.

[0417] In some embodiments, the communication condition information comprises channel quality indicator (CQI) information detected by the UE.

[0418] In some embodiments, the apparatus further includes:

[0419] a second determining module, configured to determine a weight coefficient of the target UE in the federated learning according to distribution difference statistical information of a local data set of the target UE and a global data set of a base station.

[0420] In some embodiments, the distribution difference statistical information includes: a probability distribution difference.

[0421] The second determining module includes:

[0422] a first determining submodule, configured to determine the weight coefficient of the target UE according to a probability distribution difference corresponding to a single target UE and a sum of probability distribution differences of all target UEs performing the same federated learning.

[0423] In some embodiments, the apparatus further includes:

[0424] a first receiving module, configured to receive model information of a local model reported by the target UE for performing the federated learning;

[0425] a processing module, configured to perform weighted averaging on local models of multiple target UEs according to the weight coefficient of the target UE and the model information of the local model, to obtain a global learning model.

[0426] In some embodiments, the apparatus further includes:

[0427] a first stopping module, configured to stop receiving the model information of the local model reported by the target UE for performing the federated learning, in response to the global learning model satisfying an OAM subscription requirement.

[0428] In some embodiments, the apparatus further includes:

[0429] a first sending module, configured to send model information of the global learning model to the target UE, in response to the global learning model not satisfying the OAM subscription requirement.

[0430] a second receiving module, configured to receive model information of a local model updated by the target UE according to the global learning model;

[0431] a first updating module, configured to update the global learning model according to the local model updated by the target UE and a weight coefficient corresponding to the local model.

[0432] In some embodiments, the apparatus further includes:

[0433] The first reporting module is configured to report model information of the global learning model and training data for training the global learning model to the OAM.

[0434] The third receiving module is configured to receive model parameters determined by the OAM according to the model information of the global learning model, the training data, and task data of the OAM.

[0435] The second updating module is configured to update the global learning model according to the model parameters.

[0436] In some embodiments, the apparatus further includes:

[0437] The third determining module is configured to determine that the target UE exits the federated learning in response to detecting that handover of a base station to which the target UE is connected occurs.

[0438] The present disclosure also provides a data processing apparatus applied to a base station, which includes:

[0439] The second obtaining module is configured to obtain distribution difference statistical information of a local data set of each UE in the at least one UE and a global data set.

[0440] The second scheduling module is configured to schedule a target UE participating in federated learning from the at least one UE according to the distribution difference statistical information.

[0441] As Figure 13B The present disclosure also provides a data processing apparatus 1310 applied to a UE, which includes:

[0442] The fourth receiving module 1311 is configured to receive scheduling information issued by a base station according to distribution characteristics of a local data set of the UE; wherein the scheduling information is used to determine whether the UE is a target UE scheduled to participate in federated learning.

[0443] In some embodiments, the fourth receiving module 1311 is specifically configured to:

[0444] Receive scheduling information issued by a base station according to distribution difference statistical information obtained by the base station according to the distribution characteristics of the local data set of the UE and distribution characteristics of a global data set.

[0445] In some embodiments, the apparatus further includes:

[0446] The second reporting module is configured to report capability information; wherein the capability information is used for the base station to issue the scheduling information according to the distribution characteristics of the local data set and the capability information.

[0447] In some embodiments, the capability information includes at least one of the following:

[0448] computing capability information indicating a computing capability of the UE;

[0449] communication condition information indicating a communication capability and / or a communication channel condition of the UE.

[0450] In some embodiments, the communication condition information comprises CQI information; and the apparatus further comprises:

[0451] a detecting module configured to detect the CQI information of a channel between the UE and the base station.

[0452] In some embodiments, the apparatus further comprises:

[0453] a third reporting module configured to report model information of a local model of the UE; wherein the local model is used for the base station to perform the federated learning according to the local model and a weight coefficient of the UE; wherein the weight coefficient of the UE is determined by the base station according to distribution difference statistical information of a local data set of the UE and a global data set of the base station.

[0454] In some embodiments, the apparatus further comprises:

[0455] a first generating module configured to generate the local data set according to collected wireless network data;

[0456] a second generating module configured to extract data of the local data set to generate a local training data set;

[0457] a training module configured to perform model training using the local training data set to obtain the local model.

[0458] In some embodiments, the apparatus further comprises:

[0459] a fifth receiving module configured to receive model information of a global learning model issued by the base station;

[0460] a third updating module configured to perform the federated learning according to the model information of the global learning model to obtain an updated local model;

[0461] a fourth reporting module configured to report model information of the updated local model in response to the global learning model not meeting OAM subscription requirements.

[0462] In some embodiments, the apparatus further comprises:

[0463] a second stopping module configured to stop the federated learning in response to the global learning model meeting OAM subscription requirements.

[0464] In some embodiments, the apparatus further includes:

[0465] a third stopping module, configured to stop the federated learning in response to a handover of a base station to which the UE is connected.

[0466] Embodiments of the present disclosure further provide a data processing apparatus applied to a UE, the apparatus comprising:

[0467] a fifth reporting module, configured to report a local dataset distribution characteristic, wherein the local dataset distribution characteristic is used for a base station to determine whether to schedule the UE for federated learning.

[0468] As to the apparatus in the above embodiments, specific manners in which various modules perform operations have been described in details in embodiments of the method, and will not be described in details here.

[0469] Figure 14 is a structural block diagram of a communication device provided by embodiments of the present disclosure. The communication device can be a terminal. For example, the communication device 1400 can be a mobile phone, a computer, a digital broadcast user device, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0470] Referring to Figure 14 , the communication device 1400 can include at least one of the following components: a processing component 1402, a memory 1404, a power supply component 1406, a multimedia component 1408, an audio component 1410, an input / output (I / O) interface 1412, a sensor component 1414, and a communication component 1416.

[0471] The processing component 1402 usually controls overall operations of the communication device 1400, such as operations associated with displaying, making phone calls, data communications, camera operations and recording operations. The processing component 1402 can include at least one processor 1420 to execute instructions to complete all or part of steps of the methods described above. Further, the processing component 1402 can include at least one module to facilitate interaction between the processing component 1402 and other components. For example, the processing component 1402 can include a multimedia module to facilitate the interaction between the multimedia component 1408 and the processing component 1402.

[0472] The memory 1404 is configured to store various types of data to support the operation of the communication device 1400. Examples of such data include instructions for any application or methods operating on the communication device 1400, contact data, phonebook data, messages, pictures, videos, and the like. The memory 1404 can be implemented by any type of volatile or nonvolatile storage devices or a combination thereof such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read only memory (PROM), read only memory (ROM), magnetic memory, flash memory, magnetic disks, or optical disks.

[0473] The power component 1406 provides power to the various components of the communication device 1400. The power component 1406 can include a power management system, at least one power supply, and other components associated with generating, managing, and distributing power for the communication device 1400.

[0474] The multimedia component 1408 includes a screen providing an output interface between the communication device 1400 and a user. In some embodiments, the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes at least one touch sensor to sense touch, swiping, and gestures on the touch panel. The touch sensor can not only sense a boundary of a touching or swiping action, but also detect a pressure associated with the touching or swiping action. In some embodiments, the multimedia component 1408 includes a front camera and / or a rear camera. The front and / or rear camera can receive external multimedia data when the communication device 1400 is in an operation mode, such as a shooting mode or a video mode. Each of the front and rear camera can be a fixed optical lens system or have a focal length and optical zoom capability.

[0475] The audio component 1410 is configured to output and / or input audio signals. For example, the audio component 1410 includes a microphone (MIC) configured to receive external audio signals when the communication device 1400 is in an operation mode, such as a calling mode, a recording mode, and a voice recognition mode. The received audio signals can be further stored in the memory 1404 or transmitted via the communication component 1416. In some embodiments, the audio component 1410 also includes a speaker for outputting audio signals.

[0476] The I / O interface 1412 provides an interface between the processing component 1402 and peripheral interface modules, which can be a keyboard, a click wheel, buttons, and the like. The buttons can include, but are not limited to, a home button, a volume button, a start button, and a lock button.

[0477] The sensor component 1414 includes at least one sensor to provide various aspects of status assessment for the communication device 1400. For example, the sensor component 1414 can detect an open / closed status of the communication device 1400, relative positioning of components of the communication device 1400, such as a display and keypad of the communication device 1400, a change in position of the communication device 1400 or a component of the communication device 1400, presence or absence of user contact with the communication device 1400, orientation or acceleration / deceleration of the communication device 1400, and temperature changes of the communication device 1400. The sensor component 1414 can include a proximity sensor configured to detect presence of a nearby object without any physical contact. The sensor component 1414 can also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor component 1414 can also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0478] The communication component 1416 is configured to facilitate wired or wireless communication between the communication device 1400 and other devices. The communication device 1400 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof. In an example embodiment, the communication component 1416 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an example embodiment, the communication component 1416 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra-WideBand (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0479] In an example embodiment, the communication device 1400 can be implemented by at least one application-specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), controller, microcontroller, microprocessor, or other electronic component, for performing the above-described methods.

[0480] In an example embodiment, a non-transitory computer-readable storage medium including instructions, such as the memory 1404 including instructions, is also provided, which can be executed by the processor 1420 of the communication device 1400 to complete the above-described methods. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disc, and an optical data storage device, etc.

[0481] As Figure 15As shown, an embodiment of the present disclosure shows the structure of another communication device. The communication device can be a base station involved in the embodiments of the present disclosure. For example, the communication device 1500 can be provided as a network device. Referring to Figure 15 The communication device 1500 includes a processing component 1522, which is further composed of at least one processor, and a memory resource represented by a memory 1532 for storing instructions, such as an application program, executable by the processing component 1522. The application program stored in the memory 1532 can include one or more than one module each corresponding to a set of instructions. In addition, the processing component 1522 is configured to execute the instructions to perform any of the methods described above or the methods described above in the communication device.

[0482] The communication device 1500 can also include a power supply component 1526 configured to perform power management of the communication device 1500, a wired or wireless network interface 1550 configured to connect the communication device 1500 to a network, and an input / output (I / O) interface 1558. The communication device 1500 can operate based on an operating system stored in the memory 1532, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, FreeBSD TM or the like.

[0483] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the present disclosure cover any and all variations of the present application which come within the scope of the claims and a concept of the application. It is intended that the specification and examples be considered exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0484] It is to be understood that the application is not limited to the precise construction described and as shown in the attached figures, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the present application. The scope of the application is to be determined by the claims appended hereto.

Claims

1. A data processing method, wherein, The method is applied to a base station, and comprises: determining local data set distribution characteristics of at least one user equipment (UE); scheduling a target UE participating in federated learning from the at least one UE based on the local data set distribution characteristics; The method further comprises: determining a weight coefficient of the target UE in the federated learning according to distribution difference statistical information of a local data set of the target UE and a global data set of the base station.

2. The method of claim 1, wherein, The method further comprises: obtaining distribution difference statistical information of the local data set and the global data set of each of the at least one UE; scheduling a target UE participating in federated learning from the at least one UE according to the distribution difference statistical information.

3. The method of claim 1, wherein, The method further comprises: obtaining capability information of the at least one UE; The method further comprises: scheduling a target UE participating in federated learning from the at least one UE according to the local data set distribution characteristics and the capability information of the at least one UE.

4. The method of claim 3, wherein, The capability information of the at least one UE comprises at least one of: computing capability information indicating computing capability of the UE; communication status information indicating communication capability and / or communication channel status of the UE.

5. The method of claim 4, wherein, The communication status information comprises channel quality indicator (CQI) information detected by the UE.

6. The method of claim 1, wherein, The distribution difference statistical information comprises a probability distribution difference. The method further comprises: determining the weight coefficient of the target UE according to a probability distribution difference corresponding to a single target UE and a sum of probability distribution differences of all target UEs performing the same federated learning.

7. The method of claim 1, wherein, The method further comprises: receiving model information of a local model of the target UE for performing the federated learning; performing weighted averaging on local models of multiple target UEs according to the weight coefficient of the target UE and the model information of the local model to obtain a global learning model.

8. The method of claim 7, wherein, The method further comprises: stopping receiving model information of the local model of the target UE for performing the federated learning in response to the global learning model satisfying an OAM subscription requirement.

9. The method of claim 8, wherein, The method further comprises: sending model information of the global learning model to the target UE in response to the global learning model not satisfying the OAM subscription requirement; receiving model information of a local model of the target UE updated according to the global learning model; updating the global learning model according to the local model of the target UE updated and a weight coefficient corresponding to the local model.

10. The method of any one of claims 7 to 9, wherein, The method further comprises: reporting model information of the global learning model and training data for training the global learning model to an OAM; receiving model parameters determined by the OAM according to the model information of the global learning model, the training data, and task data of the OAM; update the global learning model according to the model parameters.

11. The method of claim 1, wherein, The method further includes: in response to detecting that handover of a base station to which the target UE is connected occurs, determining that the target UE exits the federated learning.

12. A data processing method, wherein, The method is applied to a UE, and includes: receiving scheduling information; wherein the scheduling information is sent by a base station based on local data set distribution characteristics of the UE for federated learning scheduling; the local data set distribution characteristics of the UE are also used for a weight coefficient determined by the base station according to distribution difference statistical information of the local data set of the UE and a global data set of the base station, and the weight coefficient is a weight coefficient of the UE in the federated learning.

13. The method of claim 12, wherein, receiving scheduling information issued by a base station according to the local data set distribution characteristics of the UE, including: receiving scheduling information issued by a base station according to distribution difference statistical information obtained by the base station according to the local data set distribution characteristics of the UE and global data set distribution characteristics.

14. The method of claim 12, wherein, The method further includes: reporting capability information; wherein the capability information is used for the base station to issue the scheduling information according to the local data set distribution characteristics and the capability information.

15. The method of claim 14, wherein, The capability information includes at least one of: computing capability information indicating the computing capability of the UE; communication status information indicating the communication capability and / or communication channel status of the UE.

16. The method of claim 15, wherein, The communication status information includes CQI information; the method further includes: detecting the CQI information of the channel between the UE and the base station.

17. The method of any one of claims 12 to 16, wherein, The method further includes: reporting model information of a local model of the UE; wherein the local model is used for the base station to perform the federated learning according to the local model and the weight coefficient of the UE.

18. The method of claim 17, wherein, The method further includes: generating the local data set according to collected wireless network data; extracting data of the local data set to generate a local training data set; performing model training using the local training data set to obtain the local model.

19. The method of claim 17, wherein, The method further includes: receiving model information of a global learning model issued by the base station; performing the federated learning according to the model information of the global learning model to obtain an updated local model; in response to the global learning model not meeting OAM subscription requirements, reporting model information of the updated local model.

20. The method of claim 19, wherein, The method further includes: in response to the global learning model meeting OAM subscription requirements, stopping the federated learning.

21. The method of claim 12, wherein, The method further includes: in response to handover of a base station to which the UE is connected occurring, stopping the federated learning.

22. A data processing apparatus, wherein, The apparatus is applied to a base station, and includes: a first determination module configured to determine local data set distribution characteristics of at least one user equipment (UE); a scheduling module configured to schedule a target UE participating in federated learning from the at least one UE based on the local data set distribution characteristics; a second determination module configured to determine a weight coefficient of the target UE in the federated learning according to distribution difference statistical information of the local data set of the target UE and a global data set of the base station.

23. The apparatus of claim 22, wherein, The scheduling module includes: The first obtaining submodule is configured to obtain distribution difference statistical information of the local data set and the global data set of each of the at least one UE; The first scheduling submodule is configured to schedule a target UE participating in federated learning from the at least one UE according to the distribution difference statistical information.

24. The apparatus of claim 22, wherein, The apparatus further comprises: The first obtaining module is configured to obtain capability information of the at least one UE; The scheduling module comprises a second scheduling submodule configured to schedule a target UE participating in federated learning from the at least one UE according to the local data set distribution characteristics and the capability information of the at least one UE.

25. The apparatus of claim 24, wherein, The capability information of the at least one UE comprises at least one of: The computing capability information indicates the computing capability of the UE; The communication condition information indicates the communication capability and / or communication channel condition of the UE.

26. The apparatus of claim 25, wherein, The communication condition information comprises channel quality indication (CQI) information detected by the UE.

27. The apparatus of claim 22, wherein, The distribution difference statistical information comprises a probability distribution difference. The second determining module comprises: The first determining submodule is configured to determine a weight coefficient of a single target UE according to a probability distribution difference corresponding to the target UE and a sum of probability distribution differences of all target UEs performing the same federated learning.

28. The apparatus of claim 22, wherein, The apparatus further comprises: The first receiving module is configured to receive model information of a local model reported by the target UE for performing the federated learning; The processing module is configured to perform weighted averaging on local models of multiple target UEs to obtain a global learning model according to the weight coefficient of the target UE and the model information of the local model.

29. The apparatus of claim 28, wherein, The apparatus further comprises: The first stopping module is configured to stop receiving model information of the local model reported by the target UE for performing the federated learning in response to the global learning model meeting an OAM subscription requirement.

30. The apparatus of claim 29, wherein, The apparatus further comprises: The first sending module is configured to send model information of the global learning model to the target UE in response to the global learning model not meeting the OAM subscription requirement; The second receiving module is configured to receive model information of a local model updated by the target UE according to the global learning model; The first updating module is configured to update the global learning model according to the local model updated by the target UE and a weight coefficient corresponding to the local model.

31. The apparatus of any one of claims 28 to 30, wherein, The apparatus further comprises: The first reporting module is configured to report model information of the global learning model and training data for training the global learning model to an OAM; The third receiving module is configured to receive model parameters determined by the OAM according to the model information of the global learning model, the training data, and task data of the OAM; The second updating module is configured to update the global learning model according to the model parameters.

32. The apparatus of claim 22, wherein, The apparatus further comprises: The third determining module is configured to determine that the target UE exits the federated learning in response to detecting that a base station connected by the target UE has switched.

33. A data processing apparatus, wherein, The apparatus is applied to a UE and comprises: The fourth receiving module is configured to receive scheduling information; wherein the scheduling information is sent by a base station based on local data set distribution characteristics of the UE for scheduling of federated learning; and the local data set distribution characteristics of the UE are further used for a weight coefficient determined by the base station according to distribution difference statistical information of the local data set of the UE and a global data set of the base station, and the weight coefficient is a weight coefficient of the UE in the federated learning.

34. The apparatus of claim 33, wherein, The fourth receiving module is specifically configured to: receive scheduling information issued by the base station according to distribution difference statistical information obtained by the base station based on the local data set distribution characteristics of the UE and global data set distribution characteristics.

35. The apparatus of claim 33, wherein, The device further comprises: A second reporting module configured to report capability information; wherein the capability information is used by the base station to issue the scheduling information according to the local data set distribution characteristics and the capability information.

36. The apparatus of claim 35, wherein, The capability information comprises at least one of: computing capability information indicating the computing capability of the UE; communication status information indicating the communication capability and / or communication channel status of the UE.

37. The apparatus of claim 36, wherein, The communication status information comprises CQI information; and the device further comprises: A detection module configured to detect the CQI information of the channel between the UE and the base station.

38. The apparatus of any one of claims 33 to 37, wherein, The device further comprises: A third reporting module configured to report model information of a local model of the UE; wherein the local model is used by the base station to perform the federated learning according to the local model and the weight coefficient of the UE.

39. The device of claim 38, wherein, The device further comprises: A first generation module configured to generate the local data set according to collected wireless network data; A second generation module configured to extract data of the local data set to generate a local training data set; A training module configured to perform model training using the local training data set to obtain the local model.

40. The apparatus of claim 38, wherein, The device further comprises: A fifth receiving module configured to receive model information of a global learning model issued by the base station; A third updating module configured to perform the federated learning according to the model information of the global learning model to obtain an updated local model; A fourth reporting module configured to report the model information of the updated local model in response to the global learning model not meeting the subscription requirements of an OAM.

41. The apparatus of claim 40, wherein, The device further comprises: A second stopping module configured to stop the federated learning in response to the global learning model meeting the subscription requirements of the OAM.

42. The apparatus of claim 33, wherein, The device further comprises: A third stopping module configured to stop the federated learning in response to handover of a base station connected by the UE.

43. A communications device, comprising: The communication device at least comprises a processor and a memory for storing executable instructions capable of running on the processor, wherein: When the processor runs the executable instructions, the executable instructions perform the steps in the data processing method provided by any one of the preceding claims 1 to 11 or 12 to 21.

44. A non-transitory computer-readable storage medium, wherein, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the steps in the data processing method provided by any one of the preceding claims 1 to 11 or 12 to 21.

Images