A model learning method, a model learning device, and a storage medium

Abstract

The present disclosure relates to a model learning method, a model learning device and a storage medium. The model learning method is applied to a macro base station and includes: in response to receiving a model training request sent by an operation maintenance management (OAM) entity, sending the model training request to a first number of micro base stations; wherein the communication coverage range of the first number of micro base stations is within the communication coverage range of the macro base station. Through the present disclosure, the interaction between the macro base station and the micro base station is realized, the task allocation of the trained model is realized, the signal quality number is fast, the data transmission rate is improved, and the utilization rate of the wireless access network is improved.

Description

Technical Field

[0001] This disclosure relates to the field of wireless communication technology, and in particular to a model learning method, a model learning device, and a storage medium. Background Technology

[0002] To improve peak data rates and spectrum utilization in communication technology, heterogeneous network technology has been introduced. Heterogeneous network technology refers to the deployment of numerous micro base stations within the coverage area of ​​macro base stations, forming a heterogeneous system with different node types within the same coverage area. Because the geographical distance between the access point and the user equipment being served is reduced, system throughput and overall network efficiency can be effectively improved.

[0003] On the other hand, with the development of artificial intelligence technology, machine learning is being applied to more and more fields, and federated learning is one such learning method. Federated learning refers to a method of machine learning that involves different participants (such as terminals) collaborating to learn. This collaborative learning effectively ensures information security during big data exchange and protects terminal data and personal data privacy. Applying federated learning to multi-source heterogeneous networks enables machine learning modeling of these networks. However, due to the varying performance of network nodes in multi-source heterogeneous networks, the federated learning process suffers from complexity and low efficiency. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this disclosure provides a model learning method, a model learning device, and a storage medium.

[0005] According to a first aspect of the present disclosure, a model learning method is provided, applied to a macro base station, comprising:

[0006] In response to receiving a model training request from the Operation, Maintenance and Management (OAM) entity, the model training request is sent to a first number of micro base stations; wherein the communication coverage of the first number of micro base stations is within the communication coverage of the macro base station.

[0007] In one implementation, the model training request is used to trigger the micro base station to report capability information; after sending the model training request to a first number of micro base stations, the method further includes:

[0008] In response to receiving capability information sent by a micro base station, a model structure and model parameter values ​​are determined based on the capability information, and the model structure and model parameter values ​​are sent to the micro base station; the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure.

[0009] In one embodiment, the capability information includes data type characteristics of the micro base station; the method further includes:

[0010] Receive a first number of first model training results sent by a first number of micro base stations; determine the data type features of different micro base stations in the first number of micro base stations, and determine the first model loss function; after unifying the data type features based on the data type features of different micro base stations in the first number of micro base stations, perform first model alignment on the first number of first model training results with the goal of optimizing the first model loss function; perform global model learning based on the result of the first model alignment, and determine the global model.

[0011] In one implementation, the step of performing global model learning based on the results of the first model alignment to determine the global model includes:

[0012] In response to the global model learning result not satisfying the OAM model training request, the model learning result is sent to the micro base station, and the micro base station receives a first number of first model training results re-determined based on the model learning result; and the first model loss function is re-determined based on the global model learning result, and the first number of first model training results are re-aligned with the goal of optimizing the re-determined first model loss function; based on the result of the re-determined first model alignment, the next global model learning is performed, and the model learning result is re-determined, until the model learning result satisfies the model training request, and the model corresponding to the model learning result that satisfies the model training request is determined as the global model.

[0013] In one implementation, determining the first model loss function includes:

[0014] A first loss function and a first model alignment loss function are determined between the first model training result of a first number of micro base stations and the model learning result obtained from the previous global model learning of the macro base station; based on the first loss function and the first model alignment loss function, a first model loss function is determined.

[0015] In one implementation, the step of performing global model learning based on the first model alignment result to determine the global model includes:

[0016] In response to the model learning result of the global model learning satisfying the model training request of OAM, a stop model training information is sent to the micro base station; the stop training information instructs the micro base station to stop the terminal from executing the model training task; the model corresponding to the model learning result is determined as the global model, and the global model is sent to OAM.

[0017] In one embodiment, the method further includes:

[0018] In response to receiving terminal handover information sent by a micro base station during model training, the macro base station re-determines the terminal to perform the model training task based on the terminal handover information and sends the terminal information to the micro base station; the terminal handover information includes the terminal that exited the model training and the target micro base station to which the terminal reconnects; the terminal handover information is used by the macro base station to re-determine the terminal to perform the model training task.

[0019] According to a second aspect of the present disclosure, a model learning method is provided, applied to a micro base station, comprising:

[0020] The system receives a model training request sent by a macro base station; sends the model training request to a terminal; wherein the number of micro base stations receiving the model training request is a first number; and the communication coverage area of ​​the first number of micro base stations is within the communication coverage area of ​​the macro base station.

[0021] In one implementation, the model training request is used to trigger the terminal to report its communication conditions and data characteristics. After sending the model training request to the terminal, the model learning method further includes:

[0022] The system receives communication conditions and data type characteristics sent by the terminal; processes the communication conditions and data characteristics of the terminal and the micro base station to obtain capability information, and sends the capability information to the macro base station; wherein, the capability information is used by the macro base station to determine the model structure and model parameter values.

[0023] In one embodiment, the method further includes:

[0024] Receive model structure and model parameter values; the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure; determine a second number of terminals to perform model training based on the communication conditions and data type characteristics of the terminals, as well as the model structure and model parameter values; send scheduling information to the second number of terminals; the scheduling information includes the model structure and model parameter values, as well as instruction information instructing the terminals to perform model training.

[0025] In one embodiment, the method further includes:

[0026] Receive the second number of second model training results sent by the second number of terminals; determine the data type features of different terminals in the second number of terminals and determine the second model loss function; after unifying the data type features based on the data type features of different terminals in the second number of terminals, perform second model alignment on the second number of second model training results with the goal of optimizing the second model loss function; perform federated aggregation based on the result of the second model alignment to obtain the first model training result.

[0027] In one implementation, the federated aggregation based on the alignment results of the second model to obtain the training results of the first model includes:

[0028] In response to receiving a request to continue training from a macro base station and receiving model learning results from the macro base station; updating the terminal's model structure and model parameter values ​​based on the model learning results, and sending continued training scheduling information to the terminal; in response to receiving a second number of second model training results again, redetermining the second model loss function based on the first model training results, and performing second model alignment on the second number of second model training results with the goal of optimizing the redetermined second model loss function; and performing the next federated aggregation based on the result of the redetermined second model alignment to redetermine the first model training results.

[0029] In one implementation, determining the second model loss function includes:

[0030] Determine a second loss function and a second model alignment loss function between the second model training result of the second number of terminals and the first model training result obtained from the previous federated aggregation of the micro base station; determine a second model loss function based on the second loss function and the second model alignment loss function.

[0031] In one embodiment, the method further includes:

[0032] The system receives a stop model training message sent by a macro base station; the stop training message instructs the micro base station to stop the terminal from executing the model training task; and based on the stop model training message, instructs the terminal to stop executing the model training task.

[0033] In one embodiment, the method further includes:

[0034] Send terminal switching information; the terminal switching information includes information about the terminal that exited model training and the target micro base station that the terminal reconnected to; the terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task; in response to receiving the terminal information sent by the macro base station, re-determine the terminal to perform the model training task and send the model training task to the terminal.

[0035] In one implementation, sending the model training task to the terminal includes:

[0036] In response to the terminal information including the terminal that last performed the model training task, the target micro base station after the terminal handover is determined, and the target micro base station sends the model training task to the terminal; and / or

[0037] In response to the fact that the terminal information does not include the terminal that previously performed the model training task, it is determined that the terminal will no longer perform the model training task, and a new terminal is determined to perform the model training task, and the model training task is sent to the new terminal.

[0038] According to a third aspect of the present disclosure, a model learning apparatus is provided, applied to a macro base station, comprising:

[0039] The sending module is configured to, in response to receiving a model training request sent by the Operation, Maintenance and Management (OAM) entity, send the model training request to a first number of micro base stations; wherein the communication coverage of the first number of micro base stations is within the communication coverage of the macro base station.

[0040] In one embodiment, the model training request is used to trigger the micro base station to report capability information; the device further includes: a determination module;

[0041] The determining module is configured to, in response to receiving capability information sent by the micro base station, determine the model structure and model parameter values ​​based on the capability information, and send the model structure and model parameter values ​​to the micro base station; the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure.

[0042] In one embodiment, the capability information includes data type characteristics of the micro base station; the device further includes: a receiving module;

[0043] The receiving module is configured to receive a first number of first model training results sent by a first number of micro base stations; determine the data type features of different micro base stations in the first number of micro base stations, and determine a first model loss function; after unifying the data type features based on the data type features of different micro base stations in the first number of micro base stations, perform a first model alignment on the first number of first model training results with the goal of optimizing the first model loss function; and perform global model learning based on the result of the first model alignment to determine a global model.

[0044] In one embodiment, the determining module is configured to:

[0045] In response to the global model learning result not satisfying the OAM model training request, the model learning result is sent to the micro base station, and the micro base station receives a first number of first model training results re-determined based on the model learning result; and the first model loss function is re-determined based on the global model learning result, and the first number of first model training results are re-aligned with the goal of optimizing the re-determined first model loss function; based on the result of the re-determined first model alignment, the next global model learning is performed, and the model learning result is re-determined, until the model learning result satisfies the model training request, and the model corresponding to the model learning result that satisfies the model training request is determined as the global model.

[0046] In one embodiment, the determining module is configured to:

[0047] A first loss function and a first model alignment loss function are determined between the first model training result of a first number of micro base stations and the model learning result obtained from the previous global model learning of the macro base station; based on the first loss function and the first model alignment loss function, a first model loss function is determined.

[0048] In one embodiment, the determining module is configured to:

[0049] In response to the model learning result of the global model learning satisfying the model training request of OAM, a stop model training information is sent to the micro base station; the stop training information instructs the micro base station to stop the terminal from executing the model training task; the model corresponding to the model learning result is determined as the global model, and the global model is sent to OAM.

[0050] In one embodiment, the determining module is further configured to:

[0051] In response to receiving terminal handover information sent by a micro base station during model training, the macro base station re-determines the terminal to perform the model training task based on the terminal handover information and sends the terminal information to the micro base station; the terminal handover information includes the terminal that exited the model training and the target micro base station to which the terminal reconnects; the terminal handover information is used by the macro base station to re-determine the terminal to perform the model training task.

[0052] According to a fourth aspect of the present disclosure, a model learning apparatus is provided, applied to a micro base station, comprising:

[0053] A receiving module is used to receive a model training request sent by a macro base station; a sending module sends the model training request to a terminal; wherein, the number of micro base stations receiving the model training request is a first number; and the communication coverage area of ​​the first number of micro base stations is within the communication coverage area of ​​the macro base station.

[0054] In one implementation, the model training request is used to trigger the terminal to report its communication conditions and data characteristics, and the receiving module is further used to:

[0055] The system receives communication conditions and data type characteristics sent by the terminal; processes the communication conditions and data characteristics of the terminal and the micro base station to obtain capability information, and sends the capability information to the macro base station; wherein, the capability information is used by the macro base station to determine the model structure and model parameter values.

[0056] In one embodiment, the receiving module is further configured to: receive a model structure and model parameter values; the model structure is a model structure instructing a micro base station to train based on the model training request, and the model parameter values ​​are initial parameter values ​​of the model structure; determine a second number of terminals to perform model training based on the communication conditions and data type characteristics of the terminals, as well as the model structure and model parameter values; and send scheduling information to the second number of terminals; the scheduling information includes the model structure, model parameter values, and instruction information instructing the terminals to perform model training.

[0057] In one embodiment, the device further includes: a determining module;

[0058] The receiving module is used to receive the second number of second model training results sent by the second number of terminals; the determining module is used to determine the data type features of different terminals in the second number of terminals and determine the second model loss function; after unifying the data type features based on the data type features of different terminals in the second number of terminals, the second model alignment is performed on the second number of second model training results with the goal of optimizing the second model loss function; federated aggregation is performed based on the result of the second model alignment to obtain the first model training result.

[0059] In one embodiment, the determining module is configured to:

[0060] In response to receiving a request to continue training from a macro base station and receiving model learning results from the macro base station; updating the terminal's model structure and model parameter values ​​based on the model learning results, and sending continued training scheduling information to the terminal; in response to receiving a second number of second model training results again, redetermining the second model loss function based on the first model training results, and performing second model alignment on the second number of second model training results with the goal of optimizing the redetermined second model loss function; and performing the next federated aggregation based on the result of the redetermined second model alignment to redetermine the first model training results.

[0061] In one embodiment, the determining module is configured to:

[0062] Determine a second loss function and a second model alignment loss function between the second model training result of the second number of terminals and the first model training result obtained from the previous federated aggregation of the micro base station; determine a second model loss function based on the second loss function and the second model alignment loss function.

[0063] In one embodiment, the receiving module is further configured to: receive stop model training information sent by a macro base station; the stop training information instructs the micro base station to stop the terminal from executing the model training task; and instruct the terminal to stop executing the model training task based on the stop model training information.

[0064] In one embodiment, the sending module is further configured to: send terminal switching information; the terminal switching information includes information about the terminal that exited model training and the target micro base station that the terminal reconnected to; the terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task; in response to receiving the terminal information sent by the macro base station, re-determine the terminal to perform the model training task and send the model training task to the terminal.

[0065] In one embodiment, the sending module:

[0066] In response to the terminal information including the terminal that last performed the model training task, the target micro base station after the terminal handover is determined, and the target micro base station sends the model training task to the terminal; and / or

[0067] In response to the fact that the terminal information does not include the terminal that previously performed the model training task, it is determined that the terminal will no longer perform the model training task, and a new terminal is determined to perform the model training task, and the model training task is sent to the new terminal.

[0068] According to a fifth aspect of the present disclosure, a model learning apparatus is provided, comprising:

[0069] A processor; a memory for storing processor-executable instructions; wherein the processor is configured to: execute the model learning method described in the first aspect or any embodiment of the first aspect, or execute the model learning method described in the second aspect or any embodiment of the second aspect.

[0070] According to a sixth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein when instructions in the storage medium are executed by a processor of a mobile terminal, the mobile terminal is enabled to execute the model learning method described in the first aspect or any embodiment of the first aspect, or the mobile terminal is enabled to execute the model learning method described in the second aspect or any embodiment of the second aspect.

[0071] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: by sending model training requests to micro base stations through macro base stations, the interaction between macro base stations and micro base stations is realized to allocate model training tasks, thereby improving the utilization efficiency of wireless access network equipment, resulting in higher channel quality and higher model reliability and accuracy.

[0072] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0073] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0074] Figure 1 This is a schematic diagram of a heterogeneous network scenario architecture for a model learning method according to an exemplary embodiment.

[0075] Figure 2 This is a flowchart illustrating a model learning method according to an exemplary embodiment.

[0076] Figure 3 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0077] Figure 4 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0078] Figure 5 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0079] Figure 6 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0080] Figure 7 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0081] Figure 8 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0082] Figure 9 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0083] Figure 10 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0084] Figure 11This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0085] Figure 12 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0086] Figure 13 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0087] Figure 14 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0088] Figure 15 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0089] Figure 16 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0090] Figure 17 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0091] Figure 18 This is a flowchart illustrating yet another model learning method according to an exemplary embodiment.

[0092] Figure 19 This is a main flowchart illustrating a model reasoning method according to an exemplary embodiment.

[0093] Figure 20 This is a flowchart illustrating the federated learning process of model inference in a model learning method according to an exemplary embodiment.

[0094] Figure 21 This is a flowchart illustrating the terminal switching process in a model learning method according to an exemplary embodiment.

[0095] Figure 22 This is a flowchart illustrating a model inference method according to an exemplary embodiment.

[0096] Figure 23 This is a schematic diagram illustrating the protocol and interface principle for signaling and data transmission between micro base stations and macro base stations in a model learning method according to an exemplary embodiment.

[0097] Figure 24 This is a schematic diagram illustrating the protocol and interface principle for signaling and data transmission between a micro base station and a terminal in a model learning method according to an exemplary embodiment.

[0098] Figure 25This is a schematic diagram illustrating the protocol and interface principles for terminal switching in a model learning method according to an exemplary embodiment.

[0099] Figure 26 This is a block diagram of a model learning device according to an exemplary embodiment.

[0100] Figure 27 This is a block diagram of yet another model learning device according to an exemplary embodiment.

[0101] Figure 28 This is a block diagram illustrating an apparatus for model learning according to an exemplary embodiment.

[0102] Figure 29 This is a block diagram illustrating yet another apparatus for model learning according to an exemplary embodiment. Detailed Implementation

[0103] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0104] To improve peak data rates and spectrum utilization in communication technology, heterogeneous network technology has been introduced. Heterogeneous network technology refers to the deployment of numerous micro base stations within the coverage area of ​​macro base stations, forming a heterogeneous system with different node types within the same coverage area. Because the geographical distance between the access point and the served terminal is reduced, system throughput and overall network efficiency can be effectively improved.

[0105] On the other hand, with the development of artificial intelligence technology, machine learning is being applied to more and more fields, and federated learning is one such learning method. Federated learning refers to a method of machine learning that involves different participants (such as terminals) collaborating to learn. This collaborative learning effectively ensures information security during big data exchange and protects terminal data and personal data privacy. Applying federated learning to multi-source heterogeneous networks enables machine learning modeling of these networks, and its implementation can be seen in the following examples.

[0106] The macro base station forwards the specific subscription requests of the Operation Administration and Maintenance (OAM) entity to the terminal. These OAM subscription requests can also be referred to as model training requests. The terminal reports its communication conditions and local data type characteristics to the macro base station. The macro base station allocates tasks based on the terminal's reported information and sends the model structure and hyperparameter information to the terminal. The terminal trains its local model according to the tasks assigned by the macro base station. After training, the terminal sends the locally learned model parameters to the macro base station. The macro base station performs federated averaging based on the terminal's local learning results to obtain the global model. The macro base station checks whether the global learning model meets the OAM subscription requirements. If it does, the macro base station sends the obtained model to the OAM. If not, the terminal updates its local model based on the global learning results and iterates again with the macro base station until the obtained global model meets the OAM subscription requirements.

[0107] As can be seen from the above implementation methods, the related technology has the following shortcomings:

[0108] 1) When terminals are directly connected to macro base stations for data and signaling transmission, the geographical distance between the terminal and the macro base station is large, the channel quality is poor, and the data transmission rate is slow, which affects the overall efficiency of the communication network and results in low efficiency of the federated learning process.

[0109] 2) Macro base stations directly perform federated averaging on the local training results of all terminals. In practical applications, the data structure of the local training sets of different terminals may be different, making direct federated averaging less feasible. This will result in poor model generalization ability and fail to guarantee model reliability and accuracy.

[0110] 3) Data interaction between macro base stations and terminals needs to be conducted through the core network or data center. The terminal needs to upload the training result data to the core network or data center first, and then the macro base station requests the data. Direct data transmission between base stations and terminals for federated learning is not supported, which reduces the efficiency of federated learning and the utilization rate of wireless network resources.

[0111] 4) If a terminal exits the macro base station connection, it will directly exit the federated learning process, and the processing procedure for new terminals joining the connection is not considered. This results in less and less available training data during the federated learning process, which is not conducive to the overall training of the model and the improvement of model accuracy.

[0112] To address the shortcomings of the above-described implementations, related technologies consider combining model learning with heterogeneous networks. In a heterogeneous network, a macro base station's coverage area includes multiple micro base stations, and terminals connect to these micro base stations for data and signaling interaction. Because micro base stations have limited coverage, handover is easily triggered when a terminal moves. However, related technologies do not consider the handover issue, thus making it impossible to determine whether a terminal will continue to support training after a handover. Furthermore, in federated learning, different nodes may use different types of training data, leading to potentially different dimensions in the training results. Related technologies also do not consider methods for model learning based on heterogeneous networks.

[0113] Based on this, this disclosure provides a model learning method that aligns the learned model with the learning results of heterogeneous networks to determine the training model required for OAM. Furthermore, it proposes a handling method after terminal handover, allowing the terminal to continue participating in the training task of the source micro base station or join the training task of the target micro base station, depending on the type of model training task. This effectively solves the problem of continuously decreasing available training data in mobile terminal scenarios. Moreover, by aligning the data used for training the model at different nodes before training, it can support training the same model using different types of data.

[0114] Furthermore, the macro base stations and micro base stations mentioned in this disclosure belong to network equipment, which can also be called wireless access network equipment. This wireless access network equipment can be: a base station, an evolved Node B (eBY), a home base station, an access point (AP) in a Wi-Fi system, a wireless relay node, a wireless backhaul node, a transmission point (TP), or a transmission and reception point (TRP), etc. It can also be a gNB in ​​an NR system, or a component or part of a base station. When it is a vehicle-to-everything (V2X) communication system, the network equipment can also be in-vehicle equipment. It should be understood that the specific technologies and specific equipment forms used in the embodiments of this disclosure are not limited.

[0115] Furthermore, the terminal involved in this disclosure can also be referred to as a terminal device, user equipment (UE), mobile station (MS), mobile terminal (MT), etc., and is a device that provides voice and / or data connectivity to a user. For example, a terminal can be a handheld device with wireless connectivity, an in-vehicle device, etc. Currently, some examples of terminals include: smartphones (Mobile Phones), pocket personal computers (PPCs), handheld computers, personal digital assistants (PDAs), laptops, tablets, wearable devices, or in-vehicle devices, etc. In addition, when it is a vehicle-to-everything (V2X) communication system, the terminal device can also be an in-vehicle device. It should be understood that the embodiments of this disclosure do not limit the specific technology or specific device form adopted by the terminal.

[0116] Figure 1 This is a schematic diagram of a heterogeneous network scenario architecture illustrating a model learning method according to an exemplary embodiment. For example... Figure 1 As shown, the system includes one macro base station, M micro base stations, and N terminals. The terminal devices disclosed herein are primarily responsible for local data acquisition and local model training. The micro base station devices are primarily responsible for terminal scheduling and task allocation, coordinating terminal devices for model training, and managing terminal mobility. The macro base station devices are primarily responsible for coordinating micro base station devices for global model training to obtain a global model that meets OAM subscription requirements.

[0117] The coverage area of ​​the micro base station is within the coverage area of ​​the macro base station. When the macro base station and the micro base station exchange signaling / data, it can be via a wired connection, such as through optical fiber, coaxial cable, or network cable; or it can be via a wireless connection, such as through millimeter waves. The connection between the macro base station and the micro base station can be achieved through the X2 interface, or through other interfaces such as X3. This embodiment of the invention does not limit the specific implementation of the connection.

[0118] A wireless connection can be established between a micro base station and a terminal via a wireless air interface. In different implementations, this wireless air interface is based on the fourth-generation (4G) mobile communication network technology standard; or, it is based on the fifth-generation (5G) mobile communication network technology standard, such as a new air interface; or, it can be based on a next-generation mobile communication network technology standard based on 5G. This disclosure does not specify the specific implementation form of the connection between the terminal and the micro base station within the micro base station's range. Based on this system, a model learning method is proposed in this disclosure.

[0119] Figure 2 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 2 As shown, the model method used in macro base stations includes the following steps.

[0120] In step S11, in response to receiving a model training request sent by the Operation, Maintenance and Management (OAM) entity, a model training request is sent to a first number of micro base stations.

[0121] In this embodiment of the disclosure, the OAM initiates a model training request to the macro base station. The model training request includes the OAM's requirements for the training task type and model accuracy of the subscribed model. Based on the received model training request, the macro base station... Figure 1 The X2 interface shown forwards the model training request to the micro base station. The number of forwarded model training requests is determined based on the number of micro base stations covered by the macro base station. For ease of distinction, this disclosure refers to the number of micro base stations covered by a macro base station as the first number.

[0122] A model training request may include at least: an analysis ID, a notification target address, and analysis report information. The analysis ID identifies the type of analysis requested; the notification target address associates the notifications received by the requested party with this subscription; and the analysis report information includes parameters such as the preferred analysis precision level and analysis time interval. A model training request may also include analysis filter information, which indicates the conditions that the reported analysis information must meet.

[0123] The model learning method provided in this disclosure allows macro base stations to send received model training requests to micro base stations, thereby increasing data rates and further improving the overall efficiency of the communication network.

[0124] In this embodiment of the disclosure, the macro base station sends a model training request to the micro base station to enable the micro base station to report capability information. The capability information reported by the micro base station includes the communication conditions and local data type characteristics of the terminals accessing the micro base station, as well as the communication conditions and local data type characteristics of the micro base station itself.

[0125] Figure 3 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 3 As shown, the model learning method used in macro base stations includes the following steps.

[0126] In step S21, in response to receiving capability information sent by the micro base station, the model structure and model parameter values ​​are determined based on the capability information, and the model structure and model parameter values ​​are sent to the micro base station.

[0127] In this embodiment of the disclosure, the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure.

[0128] Based on the received micro base station transmission capability information, the macro base station allocates model training tasks and determines the model structure and model parameter values ​​for each micro base station among a first number of micro base stations. Specifically, the model training task allocation involves assigning a specific task for federated learning to each micro base station. The corresponding model structure and model parameter values ​​are then sent to each micro base station.

[0129] Figure 4 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 4 As shown, the model learning method used in macro base stations includes the following steps.

[0130] In step S31, a first number of first model training results sent by a first number of micro base stations are received.

[0131] In this embodiment of the disclosure, the macro base station receives the first model training result sent by each of the first number of micro base stations, and obtains the first number of first model training results.

[0132] In step S32, the data type characteristics of different micro base stations in the first number of micro base stations are determined, and the loss function of the first optimization model is determined.

[0133] In the embodiments of this disclosure, different micro base stations have different data type characteristics. For example, one micro base station may have image data as its data type, while another micro base station may have digital data as its data type. Of course, this is merely an example and is not a specific limitation of this disclosure.

[0134] In step S33, after unifying the data type features based on the data type features of different micro base stations in the first number of micro base stations, the first model alignment is performed on the first number of first model training results with the goal of optimizing the first model loss function.

[0135] In this embodiment of the disclosure, the macro base station first performs dimensional unification on the first data and the first model training results after the micro base station federated learning.

[0136] In some embodiments of this disclosure, the macro base station performs one-dimensional convolution on the data type features of all (i.e., a first number) micro base stations within the macro base station's coverage area after federated learning, mapping the data type features of all micro base stations to the same dimension d′, as shown in the following formula:

[0137]

[0138] Where r1, r2...r qThis represents the q micro base stations connected to the macro base station. It is a micro base station {r1, r2...r q The size of the convolution kernel, d′ is the common dimension. After one-dimensional convolution, the features of all terminals are mapped to the same dimension d′.

[0139] Secondly, based on the unified dimensional results of all micro base stations, the macro base station aims to optimize the loss function of the first model and aligns the training results of the first model with the data type features of different micro base stations.

[0140] In step S34, global model learning is performed based on the results of the first model alignment to determine the global model.

[0141] In this embodiment of the disclosure, the macro base station performs global model learning based on the result of the first model alignment to obtain the model learning result. This model learning result is then compared with the model training task type requirements and model accuracy included in the model training request to determine the global model for the OAM request.

[0142] Figure 5 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 5 As shown, the model learning method used in macro base stations includes the following steps.

[0143] In step S41, in response to the fact that the model learning result of global model learning does not meet the model training request of OAM, the model learning result is sent to the micro base station, and the first number of first model training results re-determined by the micro base station based on the model learning result are received.

[0144] In this embodiment of the disclosure, in response to the macro base station determining that the model learning result of the current global model learning does not meet the OAM model training request, the model learning result of the current global model learning is sent to the micro base station for the micro base station to re-determine the first model training result.

[0145] In step S42, the first model loss function is re-determined based on the model learning results of global model learning, and the first model alignment is performed again on the first number of first model training results received with the goal of optimizing the re-determined first model loss function.

[0146] In this embodiment of the disclosure, the first model loss function is re-determined based on the model learning results of the global model learning that does not meet the OAM model training request, and the first model alignment is performed on the first number of first model training results received again with the goal of optimizing the re-determined first model loss function.

[0147] In step S43, based on the result of the redefined first model alignment, the next global model learning is performed, and the model learning result is redefined until the model learning result satisfies the model training request. The model corresponding to the model learning result that satisfies the model training request is determined as the global model.

[0148] In this embodiment, the macro base station performs global model learning again based on the re-determined first model alignment result, i.e., the result of re-optimizing the first model loss function, and obtains the model learning result again. The re-obtained model learning result is compared with the model training request to determine whether it meets the requirements for the model in the model training request. If it does not meet the requirements, the first model loss function is re-determined until the global model learning result meets the model training request. The model corresponding to the model learning result that meets the model training request is then determined as the global model.

[0149] Figure 6 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 6 As shown, the model learning method used in macro base stations includes the following steps.

[0150] In step S51, a first loss function and a first model alignment loss function are determined between the first model training result of a first number of micro base stations and the model learning result obtained from the previous global model learning of macro base stations.

[0151] In this embodiment, the first model loss function comprises two parts: one part is a first loss function between a first number of micro base stations' first model training results and the model learning result obtained from the previous global model learning by the macro base station; the other part is a first model alignment loss function. The macro base station performs first model alignment on a first number of first model training results with the goal of optimizing the first model loss function. In other words, the macro base station performs first model alignment on a first number of first model training results with the goal of optimizing both the first model alignment loss function and the overall first loss function.

[0152] In step S52, the first model loss function is determined based on the first loss function and the first model alignment loss function.

[0153] In this embodiment of the disclosure, the absolute value error function and the squared error loss function used for regression problems, and the cross-entropy loss function used for classification problems are used to determine the first loss function and the first model alignment loss function as the first model loss function.

[0154] In some embodiments of this disclosure, the first model loss function may refer to the following formula.

[0155]

[0156] Where l(·, ·) represents the model's loss function, i.e., the absolute value error function and squared error loss function used for regression problems, the cross-entropy loss function used for classification problems, etc.; l M Let η be the alignment loss function for the first model, Θ be a weighting factor, Θ be all the parameters to be learned, such as weights and bias terms, and q be the total number of micro base stations participating in federated learning. a represents the first model training result of the federated aggregation parameters of micro base station k in the t-th federated learning process. t-1 This represents the model learning result of the macro base station during the (t-1)th global learning process.

[0157] Where, l is the alignment loss function of the first model. M The functional expression of can be expressed as:

[0158]

[0159] in, The norm of the squared Hilbert-Schmidt matrix, C S and C T These represent the covariance matrices before and after model alignment, respectively.

[0160] Figure 7 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 7 As shown, the model learning method used in macro base stations includes the following steps.

[0161] In step S61, in response to the model learning result of global model learning satisfying the OAM model training request, a stop model training message is sent to the micro base station.

[0162] In this embodiment, the stop training message instructs the micro base station to stop the terminal from executing the model training task. The macro base station determines that the current global model learning result satisfies the OAM model training request. In other words, the subscription requirement in the model training request sent by OAM includes specific requirements for the model accuracy needed by the subscribed service. When the global model learning result satisfies the OAM subscription requirement, it indicates that the current global learning model has reached sufficient accuracy, and the training task is terminated to obtain a usable global model. A stop model training message is then sent to the micro base station. This stop training message instructs the micro base station to stop the terminal from executing the model training task.

[0163] In step S62, the model corresponding to the model learning result is determined as the global model, and the global model is sent to OAM.

[0164] In this embodiment of the disclosure, taking the current step of performing the t-th global model learning as an example, the model learning result of the t-th global model learning is represented by a. t If it means that a t Send to OAM.

[0165] Figure 8 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 8 As shown, the model learning method used in macro base stations includes the following steps.

[0166] In step S71, in response to receiving terminal handover information sent by the micro base station during the model training process, the terminal to perform model training is re-determined based on the terminal handover information, and the terminal information is sent to the micro base station.

[0167] In this embodiment, in response to the macro base station receiving terminal handover information from the micro base station, it determines that a terminal performing a model training task has exited, or that a new terminal has joined the micro base station. The macro base station re-determines the terminal performing the model training task based on the received terminal handover information and sends the terminal information of the re-determined terminal to the micro base station. The model handover information includes information about the terminal that exited model training and the target micro base station to which the terminal rejoins; the terminal handover information is used by the macro base station to re-determine the terminal performing the model training task. The macro base station determines whether the terminal that exited or joined the connection should participate in performing the model training task based on the handover situation. The macro base station determines whether the terminal that exited or joined the connection should continue to participate in the training task of the source micro base station based on the training task type in the OAM subscription request.

[0168] In some embodiments of this disclosure, training tasks can be categorized into tasks related to upper-layer applications and tasks related to lower-layer network channels. If the task is related to upper-layer applications, the terminal can continue to participate in the federated learning task of the source micro base station; if the task is related to the lower-layer network channel, the trained model is only applicable to the source micro base station (i.e., the micro base station accessed before the terminal switched), and the terminal cannot continue to participate in the federated learning task of the source micro base station. The macro base station can decide whether the terminal continues to participate in the training of the source micro base station based on the training task type in the OAM subscription request and the specific switching information.

[0169] In one embodiment, if the macro base station decides that the terminal should continue to participate in the model training task of the source micro base station, then the target micro base station (i.e., the micro base station accessed after the terminal handover) will be responsible for forwarding the first model training results between the terminal and the source micro base station. The source micro base station will keep the terminal in the training task list and reassign model training tasks to it. The target micro base station will send the task assignment results of the terminal to the terminal, and the terminal will retain the training information from the source micro base station and continue to participate in the federated learning of the source micro base station.

[0170] In one embodiment, if the macro base station decides that the terminal should continue participating in the training of the source micro base station, the target micro base station will be responsible for forwarding the first model training results between the terminal and the source micro base station. When the terminal completes a round of local model training, the terminal sends the local training results to the target micro base station, which then forwards the results to the source micro base station. When the macro base station completes a round of global model learning, the source micro base station sends the global learning results and signaling indicating whether the terminal should continue training to the target micro base station, which then forwards the data and signaling to the terminal.

[0171] Based on the same / similar concept, embodiments of this disclosure also provide a model learning method.

[0172] Figure 9 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 9 As shown, the model learning method used in micro base stations includes the following steps.

[0173] In step S81, a model training request sent by a macro base station is received.

[0174] In step S82, a model training request is sent to the terminal.

[0175] In this embodiment of the disclosure, the number of micro base stations receiving the model training request is a first number; the communication coverage of the first number of micro base stations is within the communication coverage of the macro base station. After receiving the model training request sent by the macro base station, the micro base station forwards the model training request to the terminal.

[0176] In this embodiment of the disclosure, the micro base station sends a model training request to the terminal. The model training request can be used to trigger the terminal to send its own communication conditions and data characteristics.

[0177] Figure 10 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 10 As shown, the model learning method used in micro base stations includes the following steps.

[0178] In step S91, the receiving terminal sends the communication conditions and data type characteristics.

[0179] In step S92, the communication conditions and data characteristics of the terminal and the micro base station are processed to obtain capability information, and the capability information is sent to the macro base station.

[0180] In this embodiment, after receiving a model training request from a micro base station, the terminal determines its own communication conditions and data characteristics and reports them. The micro base station and the terminal interact with each other via a wireless channel for data and signaling. In one embodiment, the communication conditions reported by the terminal refer to the terminal's communication capabilities or the communication channel status. In another embodiment, the communication conditions reported by the terminal to the micro base station may include Channel Quality Indicator (CQI) information detected by the terminal. The local data characteristics reported by the terminal may include the type of collected data, etc. The micro base station transmits the communication conditions and data characteristics reported by the terminal, as well as the micro base station's communication conditions and data characteristics, to the macro base station via the X2 interface. For ease of description, this disclosure refers to the terminal's communication conditions and data characteristics, as well as the micro base station's communication conditions and data characteristics, as capability information. This capability information is used by the macro base station to determine the model structure and model parameter values.

[0181] Figure 11 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 11 As shown, the model learning method used in micro base stations includes the following steps.

[0182] In step S101, the model structure and model parameter values ​​are received.

[0183] In this embodiment of the disclosure, the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure.

[0184] In step S102, based on the communication conditions and data type characteristics of the terminals, as well as the model structure and model parameter values, a second number of terminals for performing model training is determined.

[0185] In this embodiment of the disclosure, the micro base station determines the second number of terminals to perform the model training task based on the received model structure and model parameter values, as well as the communication conditions and data type characteristics of the access terminals.

[0186] In step S103, scheduling information is sent to the second number of terminals.

[0187] In this embodiment of the disclosure, after determining the second number of terminals, the micro base station sends scheduling information to the second number of terminals. The scheduling information includes the model structure, model parameter values, and instructions to the terminals to perform model training.

[0188] In one approach, if the micro base station determines that the terminal performing the model training task includes one terminal (i.e., the second quantity is one), then the micro base station determines that the terminal's learning mode is a single-terminal training mode. The micro base station directly forwards the training task allocated by the macro base station to the terminal, and the terminal can perform local model training according to the allocated task.

[0189] In another approach, the micro base station determines that the terminals performing the model training task include multiple terminals (i.e., the second number is multiple), and the micro base station determines that the learning mode of the terminals is a multi-terminal collaborative training mode. The micro base station allocates the training tasks assigned by the macro base station according to the communication conditions and local data characteristics of different terminals, assisting multiple terminals in collaboratively completing model training. After receiving the task assigned by the micro base station, each terminal can perform local model training according to the model training task assigned by the micro base station.

[0190] In some embodiments of this disclosure, after receiving scheduling information sent by the micro base station, the terminal initializes the local model parameters, performs local model training according to the model training task requirements allocated by the micro base station, and transmits the training results to the micro base station through a wireless channel.

[0191] Figure 12 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 12 As shown, the model learning method used in micro base stations includes the following steps.

[0192] In step S111, the second number of second model training results sent by the second number of terminals are received.

[0193] In this embodiment of the disclosure, the micro base station receives a second number of model training results sent by a second number of terminals. Taking terminal m in the second number of terminals as an example, terminal m randomly initializes a set of model parameters as the initialization parameters of the local learning model, and the result of the initialized local learning model is denoted as... Terminal m generates a local dataset D by sensing and collecting data. m And randomly select a dataset of size N from the local dataset to generate a local training set T. m After initializing the local model parameters, the terminal trains the local model using the local training set and transmits the training results (i.e., the second model training results) to the micro base station via a wireless channel. Taking the t-th federated learning process as an example, the local learning model training update result transmitted by terminal m can be expressed as:

[0194] In step S112, the data type characteristics of different terminals in the second number of terminals are determined, and the second model loss function is determined.

[0195] In this embodiment of the disclosure, the data type characteristics of each terminal in the second number of terminals are determined, wherein different data type characteristics include image data, digital data, etc.

[0196] In step S113, after unifying the data type features based on the data type features of different terminals in the second number of terminals, the training results of the second number of second models are aligned with the goal of optimizing the loss function of the second model.

[0197] In this embodiment, since the data types of the local datasets of the terminals may differ, the feature dimensions of the trained local models may also differ. Therefore, the feature dimensions of different terminals are unified to facilitate model alignment and federated aggregation. One-dimensional convolutions are performed on the features of all terminals trained under micro base station i, mapping the features of all terminals to the same dimension d. The specific formula is as follows:

[0198]

[0199] Where m1, m2...m n This represents the n terminals connected to micro base station i. It is a terminal {m1, m2...m n The size of the convolution kernel, d, is a common dimension. After one-dimensional convolution, the features of all terminals are mapped to the same dimension d. Based on the unified dimensional result of all terminals, the micro base station aims to optimize the loss function of the first model and aligns the training results of the second model with the data type features of different terminals.

[0200] In step S114, federated aggregation is performed based on the alignment results of the second model to obtain the training results of the first model.

[0201] In this embodiment of the disclosure, the micro base station performs federated learning based on the alignment result of the second model to obtain the training result of the first model. The training result of the first model is then sent to the macro base station.

[0202] Figure 13 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 13 As shown, the model learning method used in micro base stations includes the following steps.

[0203] In step S121, in response to receiving a request to continue training from the macro base station, and receiving the model learning results from the macro base station.

[0204] In this embodiment of the disclosure, if a request to continue training is received from a macro base station, the model learning results sent by the macro base station are further received.

[0205] In step S122, the model structure and model parameter values ​​of the terminal are updated based on the model learning results, and training schedule information is sent to the terminal.

[0206] In this embodiment, the micro base station sends the model learning results sent by the macro base station to the terminal, and the terminal updates the model structure and model parameter values ​​based on the model learning results. The micro base station sends a continue training scheduling message to the terminal, instructing the terminal to continue executing the model training task based on the updated model structure and model parameter values, and resends the newly obtained second model training results to the micro base station.

[0207] In step S123, in response to receiving a second number of second model training results again, the second model loss function is re-determined based on the first model training results, and second model alignment is performed on the second number of second model training results with the goal of optimizing the re-determined second model loss function.

[0208] In this embodiment of the disclosure, after the micro base station receives the second number of second model training results sent by the terminal again, it redetermines the second loss function based on the first model training results sent by the macro base station, and once again performs second model alignment on the second number of second model training results with the goal of optimizing the second loss function.

[0209] In step S124, based on the results of the redefined second model alignment, the next federated aggregation is performed to redefine the training results of the first model.

[0210] In this embodiment, based on the redefined second model alignment result, taking micro base station i as an example, micro base station i performs federated aggregation on the basis of model alignment. After the federated aggregation is completed, the micro base station reports the federated aggregation result to the macro base station through the X2 interface. Taking the t-th federated learning process as an example, the federated aggregation result transmitted by micro base station i can be represented as follows: Federated aggregation is performed to redefine the training results of the first model. A cyclical interaction of federated learning among macro base stations, micro base stations, and terminals is formed until the final macro base station determines a global model that meets the OAM requirements.

[0211] Figure 14 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 14 As shown, the model learning method used in micro base stations includes the following steps.

[0212] In step S131, a second loss function and a second model alignment loss function are determined between the second model training result of the second number of terminals and the first model training result obtained from the previous federated aggregation of the micro base station.

[0213] In step S132, the second model loss function is determined based on the second loss function and the second model alignment loss function.

[0214] In this embodiment, after unifying the feature dimensions of all terminals, model alignment is performed based on the feature alignment results of different terminals. During model alignment, the goal is to optimize the second model loss function. The second model loss function can be determined in two parts: the first part is obtained by calculating the loss function based on the model training results of the t-th federated learning of all terminals and the update results of the (t-1)-th federated learning of the micro base station; the second part is obtained by calculating the loss function before and after model alignment. The goal of model alignment is to optimize the overall loss function of both parts.

[0215] Specifically, the absolute value error function and squared error loss function used for regression problems, and the cross-entropy loss function used for classification problems are adopted to determine the first loss function and the first model alignment loss function as the first model loss function.

[0216] In some embodiments of this disclosure, the first model loss function may refer to the following formula.

[0217] The loss function of microbase station i during the t-th federated training process can be expressed as:

[0218]

[0219] Where l(·,·) represents the model's loss function, i.e., the absolute value error function and squared error loss function used for regression problems, the cross-entropy loss function used for classification problems, etc.; l M The model alignment loss function is defined by η, which represents a weighting factor; Θ represents all parameters to be learned, such as weights and bias terms; and n represents the total number of terminals participating in federated learning under micro base station i. This represents the training and update result of the local learning model in terminal k during the t-th federated learning process; This represents the training update result of the federated aggregation parameters of micro base station i during the (t-1)th federated learning process.

[0220] The model alignment loss function can be expressed as:

[0221]

[0222] in, The norm of the squared Hilbert-Schmidt matrix, C S and C T These represent the covariance matrices before and after model alignment, respectively.

[0223] Figure 15 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 15 As shown, the model learning method used in micro base stations includes the following steps.

[0224] In step S141, the stop model training information sent by the macro base station is received.

[0225] In this embodiment of the disclosure, the stop training information is used to instruct the micro base station to stop the terminal from performing model training tasks.

[0226] In step S142, the terminal is instructed to stop executing the model training task based on the stop model training information.

[0227] In this embodiment of the disclosure, if the micro base station receives a stop model training message, it determines that it will no longer train the model. It then sends the stop model training message to the terminal, instructing the terminal to stop executing the model training task.

[0228] Figure 16 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 16 As shown, the model learning method used in micro base stations includes the following steps.

[0229] In step S151, terminal handover information is sent.

[0230] In this embodiment of the disclosure, the terminal switching information includes information about the terminal that exited model training and the target micro base station that the terminal reconnected to; the terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task.

[0231] In some embodiments of this disclosure, the source micro base station refers to the micro base station to which the terminal was connected before the handover, and the target micro base station refers to the micro base station to which the terminal was connected after the handover. The source micro base station periodically sends measurement control signals to the terminal. The terminal measures the reference signal received power and reference signal received quality, etc., according to the measurement control signals and reports the measurement report to the source micro base station. When the source micro base station detects that other base stations can provide higher service quality for the terminal, the source micro base station makes a terminal handover decision, notifies the terminal to prepare for handover and initiates a handover request to the target micro base station, and simultaneously reports the information of the handover terminal and the target micro base station to the connected macro base station. The source micro base station sends a reconfiguration RRC connection request message to the terminal and sends terminal status information to the target micro base station. The terminal and the target micro base station perform a series of parameter configurations. The terminal successfully accesses the target micro base station, and the target micro base station sends a handover success message to the source micro base station.

[0232] In step S152, in response to receiving terminal information sent by the macro base station, the terminal to perform the model training task is re-determined, and the model training task is sent to the terminal.

[0233] In this embodiment of the disclosure, after receiving the terminal information sent by the macro base station, the micro base station reallocates the model training task of each terminal based on the newly determined terminal to perform the model training task, and sends the corresponding model training task to the terminal.

[0234] Figure 17 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 17 As shown, the model learning method used in micro base stations includes the following steps.

[0235] In step S161, in response to the terminal information including the terminal that previously performed the model training task, the target micro base station after the terminal handover is determined, and the target micro base station sends the model training task to the terminal.

[0236] In this embodiment, after a terminal switches to a different micro base station, the micro base station re-determines the terminal to perform the model training task based on the terminal information sent by the macro base station. If the terminal information includes the terminal that previously performed the model training task, and that terminal has already switched micro base stations, the target micro base station (i.e., the micro base station the terminal accesses after the switch) will be responsible for forwarding the second model training results between the terminal and the source micro base station (i.e., the micro base station the terminal accessed before the switch). The source micro base station will continue to keep the terminal in the training task list and reassign training tasks to it. The target micro base station sends the terminal's model training task to the terminal, and the terminal retains the training information from the source micro base station and continues to participate in the federated learning of the source micro base station.

[0237] Figure 18 This is a flowchart illustrating a model learning method according to an exemplary embodiment. For example... Figure 18 As shown, the model learning method used in micro base stations includes the following steps.

[0238] In step S171, in response to the fact that the terminal information does not include the terminal that previously performed the model training task, it is determined that the terminal will no longer perform the model training task, and a new terminal is determined to perform the model training task, and the model training task is sent to the new terminal.

[0239] In this embodiment, if the training task type of the source micro base station does not support the terminal's continued participation in training, the source micro base station will completely remove the terminal from the training. The new terminal will report its communication conditions and local data characteristics to the target micro base station. The target micro base station will then determine whether the new terminal should participate in the target micro base station's training based on the training task type and the information reported by the terminal. The target micro base station will then send the task assignment results to the terminal.

[0240] Furthermore, after a terminal switches to a new micro base station, the micro base station, based on the terminal information sent by the macro base station, re-determines the terminal to perform the model training task. If the terminal information does not include the terminal that previously performed the model training task, and that terminal has already switched micro base stations, the target micro base station will be responsible for forwarding the second model training results between the terminal and the source micro base station. When the terminal completes a round of local model training, it sends the local training results to the target micro base station, which then forwards the results to the source micro base station. When the macro base station completes a round of global model learning, the source micro base station sends the model learning results and signaling indicating whether the terminal should continue training to the target micro base station, which then forwards the data and signaling to the terminal. The micro base station removes terminals that are no longer performing model training tasks from the training task list. New terminals to participate in the model training task are determined, and these new terminals report their communication conditions and local data characteristics to the target micro base station via an unlimited channel. The target micro base station decides whether a new terminal should participate in the model training task based on the type of training task and the information reported by the new terminal. Regardless of whether the terminal participates in the training of the source micro base station, the target micro base station will send the task assignment results of the terminal to the terminal through the wireless channel.

[0241] In some embodiments of this disclosure, the interaction process between macro base stations, micro base stations, and terminals is described.

[0242] OAM initiates a model training request to the macro base station. Upon receiving the request, the macro base station forwards it to the micro base station, which then forwards it to the terminal. The terminal reports its communication conditions and local data characteristics to the micro base station, which in turn reports the terminal information to the macro base station. The macro base station allocates tasks to the micro base station based on the terminal information reported by the micro base station and sends the model structure and hyperparameter information (i.e., the model parameter values ​​involved in this embodiment) to the micro base station. After receiving the information from the macro base station, the micro base station selects the terminals participating in the training and their learning modes, and allocates tasks to the terminals participating in the model training task. The terminal, micro base station, and macro base station continuously iterate through federated learning until the model meets the OAM subscription requirements (e.g., the model's accuracy requirements). The macro base station then reports the model training results (i.e., the global model learning results) to OAM.

[0243] The OAM subscription requirement includes: an analysis ID, used to identify the type of analysis requested; a notification target address, used to associate the notifications received by the requested party with this subscription; analysis report information, including parameters such as the preferred analysis precision level and analysis time interval; and analysis filter information (optional): indicating the conditions that the report analysis information must meet.

[0244] In some embodiments, the method for iteratively performing federated learning on the specific terminal, micro base station, and macro base station includes:

[0245] During federated learning, the terminal first initializes its local model parameters, then trains its local model according to the task requirements assigned by the micro base station, and transmits the training results (i.e., the second model training results) to the micro base station via a wireless channel. The micro base station aggregates the local training results from all participating terminals, performs model alignment, then federated aggregation, and reports the federated aggregation result (i.e., the first model training result) to the macro base station via the X2 interface. The macro base station, after aggregating the federated aggregation results from all participating micro base stations, performs model alignment, then global model learning, and sends the global learning result to the micro base station via the X2 interface. The micro base station forwards the global model training result to the terminal via a wireless channel, and the terminal updates its locally learned model based on the global model training result. The macro base station determines whether the globally trained model meets the OAM subscription requirements.

[0246] In some embodiments, if the global model performance meets the OAM subscription requirements, the macro base station will report the model training results to OAM and notify the micro base station to stop training.

[0247] In some embodiments, if the global model performance does not meet the OAM subscription requirements, the macro base station needs to arrange the training tasks of the terminal according to the terminal switching information, and the micro base station will then reallocate the tasks according to the terminal switching situation. The terminal will then perform local model learning again and report the results to the micro base station. This process is repeated until the model performance meets the OAM subscription requirements.

[0248] In some embodiments, during terminal handover, the source micro base station represents the micro base station connected to the terminal before the handover, and the target micro base station represents the micro base station connected to the terminal after the handover. The macro base station schedules the terminal to perform model training tasks based on the terminal handover information, including:

[0249] When a source micro base station makes a terminal handover decision during a federated learning cycle, it notifies the terminal to prepare for the handover and reports the terminal exiting the connection and the target micro base station information to the macro base station. Upon receiving the command from the source micro base station, the terminal performs the handover and establishes a connection on the target micro base station. The macro base station determines whether the terminal should continue participating in the source micro base station's training based on the source micro base station's training task type and the terminal's handover information.

[0250] In some embodiments, if the training task type of the source micro base station supports the terminal continuing to participate in training, the target micro base station will be responsible for forwarding the training data between the terminal and the source micro base station, and the terminal will continue to participate in the training task of the source micro base station. The target micro base station will send the task arrangement result of the terminal to the terminal.

[0251] In some embodiments, if the training task type of the source micro base station does not support the terminal's continued participation in training, the source micro base station will completely remove the terminal from the training. The new terminal will report its communication conditions and local data characteristics to the target micro base station. The target micro base station will then determine whether the new terminal should participate in the target micro base station's training based on the training task type and the information reported by the terminal. The target micro base station will then send the task assignment results to the terminal.

[0252] In some embodiments, after macro base stations, micro base stations, and terminals complete the OAM model training task and send the global model to OAM, they can also perform inference on the trained model. The OAM determines the task cell for model inference, and the implementation methods for task cell inference include:

[0253] During task inference, the task cell initiates an inference request to the OAM (Operational Information Management) through its macro base station, reporting the inference task type and specific requirements. The OAM then selects one or more suitable models based on the task type and requirements. Once a suitable model is found, the OAM sends the model selection result to the macro base station, which then reports the specific model parameter information. The OAM forwards the model parameter information reported by the selected macro base station to the macro base station where the task cell is located, and the macro base station performs inference for the task based on the model parameter information.

[0254] The following embodiments will illustrate the interaction process between macro base stations, micro base stations, and terminals with reference to the accompanying drawings. Figure 19 This is a main flowchart illustrating a model reasoning method according to an exemplary embodiment. For example... Figure 19 As shown, it includes the following steps:

[0255] Step 1: OAM sends a model training request to the macro base station, which then forwards the request to the micro base station.

[0256] Step 2: The micro base station forwards the model training request to the terminal, the terminal reports the communication conditions and local data type characteristics to the micro base station, and the micro base station reports the terminal data to the macro base station.

[0257] Step 3: The macro base station allocates tasks based on the information reported by the micro base station and sends the model structure and model parameter values ​​to the micro base station.

[0258] Step 4: The micro base station selects the terminals participating in the model training task and the learning mode of the terminals, and assigns tasks to the terminals participating in the training.

[0259] Step 5: The terminal, micro base station and macro base station continuously iterate in federated learning until the model meets the OAM subscription requirements, and the macro base station reports the model training results to OAM.

[0260] Figure 20This is a flowchart illustrating a federated learning process for a model inference method according to an exemplary embodiment. For example... Figure 20 As shown, the process includes: the terminal initializing local model parameters; the terminal training the local model according to task requirements and transmitting the second model training results to the micro base station via a wireless channel; the micro base station aggregating the second model training results from all terminals, performing model alignment, then federated aggregation, and reporting the results to the macro base station via the X2 interface; the macro base station aggregating the federated aggregation results from all micro base stations, performing model alignment, then global model learning, and sending the global model learning results to the micro base station via the X2 interface; the micro base station sending the model learning results to the terminal via a wireless channel, and the terminal updating its local learning model based on the model learning results; the macro base station determining whether the global model corresponding to the model training results meets the OAM subscription requirements; if the OAM subscription requirements are met, federated learning ends, and the macro base station reports the model learning results to OAM. If the OAM subscription requirements are not met, the macro base station determines whether a terminal that has exited or newly joined the connection should participate in training based on the handover information, and the micro base station reassigns the model training task based on the terminal handover situation.

[0261] Figure 21 This is a flowchart illustrating a terminal switching process for a model inference method according to an exemplary embodiment. For example... Figure 21 As shown, the process includes: the source micro base station notifying the terminal to prepare for handover and reporting the disconnected terminal and target micro base station information to the macro base station; the terminal performing the handover and establishing a connection on the target micro base station; the macro base station deciding whether the terminal should continue participating in the source micro base station's model training task based on the training task type and handover information; if the terminal continues to participate in the source micro base station's model training task, the target micro base station is responsible for forwarding the training data between the terminal and the source micro base station, and the terminal continues to participate in the source micro base station's training task; the target micro base station sending the terminal's task allocation result to the terminal. If the terminal does not continue to participate in the source micro base station's model training task, the source micro base station removes the terminal from the training; newly added terminals reporting communication conditions and local data characteristics to the target micro base station; the target micro base station deciding whether new terminals should participate in training based on the training task type and the information reported by the terminals; the target micro base station sending the terminal's task allocation result to the terminal.

[0262] In some embodiments of this disclosure, after determining the global model, the process further includes inferring the global model. Figure 22 This is a flowchart illustrating a model inference method according to an exemplary embodiment. For example... Figure 22 As shown, it includes the following steps:

[0263] Step 1: The task cell initiates an inference request to OAM through the macro base station and reports the inference task type and specific requirements.

[0264] Step 2: OAM finds one or more suitable models based on the type of inference task and specific requirements.

[0265] In one embodiment, inference task types are categorized into those related to upper-layer applications or those related to lower-layer network channels. When selecting a model, macro base station models with training task types similar to inference task types are given priority.

[0266] In one embodiment, multiple trained models can be selected, and the models can be fused together for inference.

[0267] Step 3: OAM sends the model selection results to the macro base stations, and the selected macro base stations report the specific model parameter information.

[0268] Step 4: OAM forwards the model parameter information to the macro base station where the task cell is located, and the macro base station where the task cell is located performs inference on the task based on the model parameter information.

[0269] In one embodiment, OAM selects multiple trained macro base station models. The macro base station where the task cell is located then fuses the multiple models and then performs inference on the task.

[0270] Figure 23 This is a schematic diagram illustrating the protocol and interface principles for signaling and data transmission between micro base stations and macro base stations in a model learning method according to an exemplary embodiment. Figure 23 As shown, this mainly involves the interaction between micro base stations and macro base stations, as detailed below:

[0271] 1a. The micro base station sends a connection establishment request (X2 Setup Request) signaling to the macro base station, indicating a request to establish a connection with the target base station. 1b. The macro base station allocates resources based on the connection establishment request signaling sent by the micro base station. 1c. The macro base station sends a successful connection establishment (X2 Setup Response) signaling to the micro base station, indicating that the connection has been successfully established. 2a. The micro base station packages the first model training results. 2b. The micro base station sends a training result data packet signaling to the macro base station, indicating that the training data packet is to be sent to the receiver. 3. The macro base station uses the AI ​​service module to perform global model training. 4. The macro base station sends a packaged and sent global model training result data packet signaling to the micro base station, indicating that the global model training results are packaged and sent to the receiver. 5. The macro base station sends a notification signaling whether to continue training to the micro base station, indicating whether to continue training. 6. The macro base station and the micro base station confirm that the transmission is complete. 7. The macro base station sends a resource release signaling message to the micro base station, indicating that the resource should be released.

[0272] Figure 24 This is a schematic diagram illustrating the protocol and interface principles for signaling and data transmission between a micro base station and a terminal in a model learning method according to an exemplary embodiment. Figure 24 As shown, this mainly involves the interaction between the micro base station and the terminal, as detailed below:

[0273] 1a. The terminal sends an RRC Connection Request signaling message to the micro base station, indicating a request to establish an RRC connection with the target base station. 1b. The micro base station sends an RRC Connection Setup confirmation signaling message to the terminal, indicating that the receiver agrees to establish an RRC connection. 1c. The terminal performs radio resource configuration based on the signaling message sent by the micro base station. 1d. The terminal sends an RRC Connection Setup Complete signaling message to the micro base station, indicating that the RRC connection establishment is complete. 2a. The terminal packages its local training results (i.e., the second model training results). 2b. The terminal sends a local training result data packet signaling message to the micro base station, indicating that it is sending the local training result data packet to the receiver. 3. The micro base station and the macro base station collaborate to use the AI ​​service module for model training. 4. The micro base station sends a global model training result signaling message to the terminal, indicating that it is sending the global model training results to the receiver. 5. The micro base station will send a signaling message to the terminal indicating whether to continue training. The signaling message indicates whether to continue training. 6a. The micro base station will send an RRC Connection Release request signaling message to the terminal, indicating a request to release the RRC connection. 6b. The terminal will send an RRC Connection Release Complete signaling message to the micro base station, indicating that the RRC connection has been successfully released.

[0274] Figure 25 This is a schematic diagram illustrating the protocol and interface principles for terminal switching in a model learning method according to an exemplary embodiment. For example... Figure 25 As shown, this mainly involves the interaction between the macro base station, the source micro base station, the target micro base station, and the terminal, as detailed below:

[0275] 1. The source micro base station sends a Measurement Control signal to the terminal, indicating that it is notifying the other party to perform signal strength measurement. 2. The terminal sends a Measurement Report signal to the source micro base station, indicating that it is sending a measurement report to the receiver. 3. The source micro base station makes a handover decision (HO decision). 4a. The source micro base station sends a Handover Request signal to the target micro base station, indicating that it is sending a handover request to the receiver. 4b. The target micro base station sends a Handover Request Acknowledgment signal to the source micro base station, indicating that it is sending a handover request acknowledgment to the receiver. 5. The source micro base station sends an RRC Connection Reconfiguration signal containing Mobility Control Information to the terminal, indicating that it is sending an RRC Connection Reconfiguration request to the receiver. 6. The source micro base station sends an Early Status Transfer (EST) signaling message to the target micro base station, indicating that the terminal status information is being sent to the receiver. 7. The terminal accesses the target micro base station. 8. The terminal sends an RRC Connection Reconfiguration Complete (RRC) signaling message to the target micro base station, indicating that the RRC connection reconfiguration complete message is being sent to the receiver. 9. The target micro base station sends a Handover Success (Handover Success) signaling message to the source micro base station, indicating that the handover success message is being sent to the receiver. 10. The source micro base station sends a Handover Terminal and Target Micro Base Station Information (HTS) signaling message to the macro base station, indicating that the handover terminal and target micro base station information is being sent to the macro base station. 11. The macro base station decides whether the terminal should continue participating in the source micro base station's training task based on the source micro base station's training task type and handover information. 12. The macro base station sends a Decision Result signaling message to the target micro base station, indicating that the decision result is being sent to the receiver. 13. The macro base station sends a decision result signaling message to the source micro base station, indicating that the decision result is to be sent to the receiver. 14. The target micro base station decides whether the handover terminal should participate in its federated learning training task. 15. The target micro base station sends a decision result signaling message to the terminal, indicating that the decision result is to be sent to the receiver.

[0276] Based on the same concept, embodiments of this disclosure also provide a model learning device.

[0277] It is understood that the model learning apparatus provided in this disclosure includes hardware structures and / or software modules corresponding to each function in order to achieve the above-mentioned functions. In conjunction with the units and algorithm steps of the various examples disclosed in this disclosure, this disclosure can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solutions of this disclosure.

[0278] In some embodiments of this disclosure, the model learning apparatus is illustrated by example, comprising one macro base station device, M micro base station devices, and N user devices.

[0279] The user equipment (User Equipment) is the terminal accessing the micro base station, responsible for local data collection and local model training, and can update the local model based on the global model learning results. The micro base station is responsible for selecting terminals and learning modes to participate in model training, allocating training tasks to participating terminals, summarizing the local training results of the terminals, and using the AI ​​service module for model alignment and federated averaging. It is also responsible for terminal handover management and forwarding signaling from the macro base station to the terminals. The macro base station is responsible for interacting with OAM (Operational Information Management), allocating tasks to participating micro base station devices, summarizing the training results of the micro base station devices, and using the AI ​​service module for model alignment and global model learning. It also determines whether a terminal should continue participating in training when a handover occurs.

[0280] Figure 26 This is a block diagram illustrating a model learning device according to an exemplary embodiment. (Refer to...) Figure 26 The model learning device 100 is applied to a macro base station and includes a transmission module 101.

[0281] The sending module is configured to, in response to receiving a model training request from the Operation, Maintenance and Management (OAM) entity, send the model training request to a first number of micro base stations. The communication coverage area of ​​the first number of micro base stations is within the communication coverage area of ​​the macro base station.

[0282] In this embodiment of the disclosure, the model training request is used to trigger the micro base station to report capability information. The apparatus also includes a determination module 102.

[0283] The determining module 102 is used to, in response to receiving capability information sent by the micro base station, determine the model structure and model parameter values ​​based on the capability information, and send the model structure and model parameter values ​​to the micro base station. The model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure.

[0284] In this embodiment of the disclosure, the capability information includes the data type characteristics of the micro base station. The apparatus also includes a receiving module 103.

[0285] The receiving module 103 is used to receive a first number of first model training results sent by a first number of micro base stations. It determines the data type characteristics of different micro base stations within the first number of micro base stations and determines the first model loss function. After unifying the data type characteristics based on these characteristics, it performs first model alignment on the first number of first model training results with the goal of optimizing the first model loss function. Based on the first model alignment result, it performs global model learning to determine the global model.

[0286] In this embodiment, the determining module 102 is configured to, in response to a global model learning result not satisfying the OAM model training request, send the model learning result to the micro base station and receive a first number of first model training results re-determined by the micro base station based on the model learning result. It then re-determines the first model loss function based on the global model learning result and, with the goal of optimizing the re-determined first model loss function, re-aligns the received first number of first model training results. Based on the re-determined first model alignment result, the next global model learning is performed, and the model learning result is re-determined until the model learning result satisfies the model training request. The model corresponding to the model learning result that satisfies the model training request is then determined as the global model.

[0287] In this embodiment of the disclosure, the determining module 102 is used to determine a first loss function and a first model alignment loss function between the first model training result of a first number of micro base stations and the model learning result obtained from the previous global model learning of the macro base station. Based on the first loss function and the first model alignment loss function, the first model loss function is determined.

[0288] In this embodiment of the disclosure, the determining module 102 is configured to send a stop model training message to the micro base station in response to the global model learning result satisfying the OAM model training request. The stop training message instructs the micro base station to stop the terminal from executing the model training task. The model corresponding to the model learning result is determined as the global model, and the global model is sent to the OAM.

[0289] In this embodiment of the disclosure, the determining module 102 is further configured to, in response to receiving terminal handover information sent by a micro base station during model training, re-determine the terminal to perform the model training task based on the terminal handover information, and send the terminal information to the micro base station. The terminal handover information includes information about the terminal that exited model training and the target micro base station to which the terminal reconnects. The terminal handover information is used by the macro base station to re-determine the terminal to perform the model training task.

[0290] Figure 27 This is a block diagram illustrating a model learning device according to an exemplary embodiment. (Refer to...) Figure 27 The model learning device 200, applied to a micro base station, includes a receiving module 201 and a transmitting module 202.

[0291] The receiving module 201 is used to receive model training requests sent by the macro base station. The sending module 202 is used to send model training requests to the terminal. The number of micro base stations receiving the model training requests is a first quantity. The communication coverage area of ​​the first quantity of micro base stations is within the communication coverage area of ​​the macro base station.

[0292] In this embodiment, the model training request is used to trigger the terminal to report its communication conditions and data characteristics. The receiving module 201 is also used to receive the communication conditions and data characteristics sent by the terminal. The communication conditions and data characteristics of the terminal and the micro base station are processed to obtain capability information, which is then sent to the macro base station. The capability information is used by the macro base station to determine the model structure and model parameter values.

[0293] In this embodiment of the disclosure, the receiving module 201 is further configured to: receive a model structure and model parameter values. The model structure is a model structure instructing the micro base station to train based on a model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure. Based on the communication conditions and data type characteristics of the terminals, as well as the model structure and model parameter values, a second number of terminals to perform model training is determined. Scheduling information is sent to the second number of terminals. The scheduling information includes the model structure, model parameter values, and instruction information instructing the terminals to perform model training.

[0294] In this embodiment of the disclosure, the apparatus further includes a determination module 203.

[0295] The receiving module 201 is used to receive the second number of second model training results sent by the second number of terminals. The determining module 203 is used to determine the data type features of different terminals in the second number of terminals and determine the second model loss function. After unifying the data type features based on the data type features of different terminals in the second number of terminals, the second model alignment is performed on the second number of second model training results with the goal of optimizing the second model loss function. Federated aggregation is performed based on the second model alignment results to obtain the first model training results.

[0296] In this embodiment, the determining module 203 is configured to: respond to receiving a continue training request from a macro base station and receiving model learning results from the macro base station; update the terminal's model structure and model parameter values ​​based on the model learning results and send continue training scheduling information to the terminal; respond to receiving a second number of second model training results again; redetermine the second model loss function based on the first model training results; and perform second model alignment on the second number of second model training results with the goal of optimizing the redetermined second model loss function; and perform the next federated aggregation based on the result of the redetermined second model alignment to redetermine the first model training results.

[0297] In this embodiment of the disclosure, the determining module 203 is used to determine a second loss function and a second model alignment loss function between the second model training result of the second number of terminals and the first model training result obtained from the previous federated aggregation of the micro base station. Based on the second loss function and the second model alignment loss function, a second model loss function is determined.

[0298] In this embodiment of the disclosure, the receiving module 201 is further configured to: receive stop model training information sent by the macro base station. The stop training information instructs the micro base station to stop the terminal from executing the model training task. Based on the stop model training information, the terminal is instructed to stop executing the model training task.

[0299] In this embodiment of the disclosure, the sending module 202 is further configured to: send terminal switching information. The terminal switching information includes information about the terminal that exited model training and the target micro base station to which the terminal reconnects. The terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task. In response to receiving the terminal information sent by the macro base station, the macro base station re-determines the terminal to perform the model training task and sends the model training task to the terminal.

[0300] In this embodiment of the disclosure, the sending module 202 is configured to, in response to the terminal information including the terminal that last performed the model training task, determine the target micro base station after the terminal handover, and have the target micro base station send the model training task to the terminal. And / or

[0301] In response to the terminal information not including the terminal that previously performed the model training task, it is determined that the terminal will no longer perform the model training task, and a new terminal is identified to perform the model training task. The model training task is then sent to the new terminal. Regarding the apparatus in the above embodiments, the specific methods by which each module performs its operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0302] Figure 28This is a block diagram illustrating an apparatus 300 for model learning according to an exemplary embodiment. For example, apparatus 300 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0303] Reference Figure 28 The device 300 may include one or more of the following components: processing component 302, memory 304, power component 306, multimedia component 308, audio component 310, input / output (I / O) interface 312, sensor component 314, and communication component 316.

[0304] Processing component 302 typically controls the overall operation of device 300, such as operations associated with display, telephone calls, data communication, camera operation, and recording. Processing component 302 may include one or more processors 320 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 302 may include one or more modules to facilitate interaction between processing component 302 and other components. For example, processing component 302 may include a multimedia module to facilitate interaction between multimedia component 308 and processing component 302.

[0305] Memory 304 is configured to store various types of data to support the operation of device 300. Examples of such data include instructions for any application or method operating on device 300, contact data, phonebook data, messages, pictures, videos, etc. Memory 304 can be implemented by any type of volatile or non-volatile storage device 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 storage, flash memory, magnetic disk, or optical disk.

[0306] The power supply component 306 provides power to the various components of the device 300. The power supply component 306 may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device 300.

[0307] Multimedia component 308 includes a screen that provides an output interface between the device 300 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 308 includes a front-facing camera and / or a rear-facing camera. When the device 300 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0308] Audio component 310 is configured to output and / or input audio signals. For example, audio component 310 includes a microphone (MIC) configured to receive external audio signals when device 300 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 304 or transmitted via communication component 316. In some embodiments, audio component 310 also includes a speaker for outputting audio signals.

[0309] I / O interface 312 provides an interface between processing component 302 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0310] Sensor assembly 314 includes one or more sensors for providing status assessments of various aspects of device 300. For example, sensor assembly 314 may detect the on / off state of device 300, the relative positioning of components such as the display and keypad of device 300, changes in the position of device 300 or a component of device 300, the presence or absence of user contact with device 300, the orientation or acceleration / deceleration of device 300, and temperature changes of device 300. Sensor assembly 314 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 314 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 314 may also include an accelerometer, a gyroscope, a magnetometer, a pressure sensor, or a temperature sensor.

[0311] Communication component 316 is configured to facilitate wired or wireless communication between device 300 and other devices. Device 300 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 316 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 316 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0312] In an exemplary embodiment, the apparatus 300 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0313] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 304 including instructions, which can be executed by a processor 320 of the device 300 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0314] Figure 29 This is a block diagram illustrating an apparatus 400 for model learning according to an exemplary embodiment. For example, apparatus 400 may be provided as a server. (Refer to...) Figure 29 The apparatus 400 includes a processing component 422, which further includes one or more processors, and memory resources represented by memory 432 for storing instructions, such as application programs, that can be executed by the processing component 422. The application programs stored in memory 432 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 422 is configured to execute instructions to perform the methods described above.

[0315] Device 400 may also include a power supply component 426 configured to perform power management of device 400, a wired or wireless network interface 450 configured to connect device 400 to a network, and an input / output (I / O) interface 458. Device 400 may operate on an operating system stored in memory 432, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0316] It can be further understood that in this disclosure, "multiple" refers to two or more, and other quantifiers are similar. "And / or" describes the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. The singular forms "a," "the," and "the" are also intended to include the plural forms unless the context clearly indicates otherwise.

[0317] It is further understood that the terms "first," "second," etc., are used to describe various types of information, but this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another, and do not indicate a specific order or degree of importance. In fact, the expressions "first," "second," etc., are completely interchangeable. For example, without departing from the scope of this disclosure, first information can also be referred to as second information, and similarly, second information can also be referred to as first information.

[0318] It is further understood that although operations are described in a specific order in the accompanying drawings in the embodiments of this disclosure, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all of the shown operations to be performed to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0319] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0320] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A model learning method, characterized in that, Applications in macro base stations include: In response to receiving a model training request from the Operation, Maintenance and Management (OAM) entity, the model training request is sent to a first number of micro base stations, wherein the communication coverage of the first number of micro base stations is within the communication coverage of the macro base station. In response to receiving terminal handover information sent by a micro base station during model training, the terminal to perform the model training task is re-determined based on the terminal handover information, and the terminal information is sent to the micro base station. The terminal switching information includes information about the terminal that exited model training and the target micro base station that the terminal reconnected to; the terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task. The model training request is used to trigger the micro base station to report capability information; after sending the model training request to a first number of micro base stations, the method further includes: in response to receiving the capability information sent by the micro base station, determining the model structure and model parameter values ​​based on the capability information, and sending the model structure and model parameter values ​​to the micro base station; the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure.

2. The model learning method according to claim 1, characterized in that, The capability information includes the data type characteristics of the micro base station; the method further includes: Receive the first number of first model training results sent by the first number of micro base stations; Determine the data type characteristics of different micro base stations among the first number of micro base stations, and determine the first model loss function; After unifying the data type features based on the data type features of different micro base stations in the first number of micro base stations, the first model alignment is performed on the first number of first model training results with the goal of optimizing the first model loss function. Based on the alignment results of the first model, global model learning is performed to determine the global model.

3. The model learning method according to claim 2, characterized in that, The step of performing global model learning based on the alignment results of the first model to determine the global model includes: In response to the global model learning result not satisfying the OAM model training request, the model learning result is sent to the micro base station, and the first number of first model training results re-determined by the micro base station based on the model learning result are received; and Based on the model learning results of the global model learning, the first model loss function is re-determined, and with the goal of optimizing the re-determined first model loss function, the first model alignment is re-performed on the first number of first model training results received. Based on the results of the redefined first model alignment, the next global model learning is performed, and the model learning results are redefined until the model learning results satisfy the model training request. The model corresponding to the model learning results that satisfy the model training request is then determined as the global model.

4. The model learning method according to claim 3, characterized in that, Determine the loss function for the first model, including: A first loss function is defined between the first model training result of a first number of micro base stations and the model learning result obtained from the previous global model learning of the macro base station, and a first model alignment loss function is defined. The first model loss function is determined based on the first loss function and the first model alignment loss function.

5. The model learning method according to claim 2, characterized in that, The step of performing global model learning based on the alignment results of the first model to determine the global model includes: In response to the model training request of OAM being satisfied by the model learning result of the global model learning, a stop model training message is sent to the micro base station; the stop model training message instructs the micro base station to stop the terminal from executing the model training task. The model corresponding to the learning result of the model is determined as the global model, and the global model is sent to the OAM.

6. A model learning method, characterized in that, Applications in micro base stations include: Receive model training requests sent by macro base stations; Send the model training request to the terminal; Send terminal switching information; the terminal switching information includes information about the terminal that exited model training and the target micro base station that the terminal reconnected to; the terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task; In response to receiving terminal information sent by the macro base station, the terminal to perform the model training task is re-determined, and the model training task is sent to the terminal. The number of micro base stations receiving model training requests sent by the macro base station is a first number; the communication coverage area of ​​the first number of micro base stations is within the communication coverage area of ​​the macro base station. The model training request is used to trigger the terminal to report its communication conditions and data characteristics. After sending the model training request to the terminal, the model learning method further includes: receiving the communication conditions and data characteristics sent by the terminal; processing the communication conditions and data characteristics of the terminal and the micro base station to obtain capability information, and sending the capability information to the macro base station; wherein, the capability information is used by the macro base station to determine the model structure and model parameter values. Receive model structure and model parameter values; the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure; Based on the communication conditions and data type characteristics of the terminals, as well as the model structure and model parameter values, a second number of terminals for performing model training is determined. Send scheduling information to the second number of terminals; the scheduling information includes model structure and model parameter values, as well as instructions to the terminals to perform model training.

7. The model learning method according to claim 6, characterized in that, The method further includes: Receive the second number of second model training results sent by the second number of terminals; Determine the data type characteristics of different terminals in the second number of terminals, and determine the loss function of the second model; After unifying the data type features based on the different data type features of the second number of terminals, the second model alignment is performed on the second number of second model training results with the goal of optimizing the second model loss function. Federated aggregation is performed based on the alignment results of the second model to obtain the training results of the first model.

8. The model learning method according to claim 7, characterized in that, The federated aggregation based on the alignment results of the second model yields the training results of the first model, including: In response to receiving a request to continue training from the macro base station, and receiving the model learning results from the macro base station; The model structure and model parameter values ​​of the terminal are updated based on the model learning results, and training schedule information is sent to the terminal. In response to receiving a second number of second model training results again, the second model loss function is re-determined based on the first model training results, and second model alignment is performed on the second number of second model training results with the goal of optimizing the re-determined second model loss function; Based on the results of the redefined second model alignment, the next federated aggregation is performed to redefine the training results of the first model.

9. The model learning method according to claim 8, characterized in that, Determine the loss function for the second model, including: Determine the second loss function between the second model training result of the second number of terminals and the first model training result obtained from the previous federated aggregation of the micro base station, and the second model alignment loss function; The second model loss function is determined based on the second loss function and the second model alignment loss function.

10. The model learning method according to claim 8, characterized in that, The method further includes: Receive a stop model training message sent by a macro base station; the stop training message instructs the micro base station to stop the terminal from executing the model training task; Based on the stopped model training information, the terminal is instructed to stop executing the model training task.

11. The model learning method according to claim 6, characterized in that, Sending the model training task to the terminal includes: In response to the terminal information including the terminal that last performed the model training task, the target micro base station after the terminal handover is determined, and the target micro base station sends the model training task to the terminal; and / or In response to the fact that the terminal information does not include the terminal that previously performed the model training task, it is determined that the terminal will no longer perform the model training task, and a new terminal is determined to perform the model training task, and the model training task is sent to the new terminal.

12. A model learning device, characterized in that, Applications in macro base stations include: The sending module is configured to, in response to receiving a model training request sent by the Operation, Maintenance and Management (OAM) entity, send the model training request to a first number of micro base stations, and, in the event that terminal switching information is received from a micro base station during model training, re-determine the terminal to perform the model training task based on the terminal switching information, and send the terminal information to the micro base station. The terminal switching information includes information about the terminal that exited model training and the target micro base station that the terminal reconnected to. The terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task. The communication coverage area of ​​the first number of micro base stations is within the communication coverage area of ​​the macro base stations; The model training request is used to trigger the micro base station to report capability information; after sending the model training request to a first number of micro base stations, the device further includes: in response to receiving the capability information sent by the micro base station, determining the model structure and model parameter values ​​based on the capability information, and sending the model structure and model parameter values ​​to the micro base station; the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure.

13. A model learning device, characterized in that, Applications in micro base stations include: The receiving module is used to receive model training requests sent by the macro base station; The sending module is used to send the model training request to the terminal and to send terminal switching information, wherein the terminal switching information includes information on the terminal that exited the model training and the target micro base station that the terminal reconnected to. The terminal switching information is used by the macro base station to re-determine the terminal to perform the model training task. In response to receiving the terminal information sent by the macro base station, the macro base station re-determines the terminal to perform the model training task and sends the model training task to the terminal. The number of micro base stations receiving model training requests sent by the macro base station is a first number; the communication coverage area of ​​the first number of micro base stations is within the communication coverage area of ​​the macro base station. The model training request is used to trigger the terminal to report its communication conditions and data characteristics. After sending the model training request to the terminal, the model learning device further includes: receiving the communication conditions and data characteristics sent by the terminal; processing the communication conditions and data characteristics of the terminal and the micro base station to obtain capability information, and sending the capability information to the macro base station; wherein, the capability information is used by the macro base station to determine the model structure and model parameter values. Receive model structure and model parameter values; the model structure is a model structure that instructs the micro base station to train based on the model training request, and the model parameter values ​​are the initial parameter values ​​of the model structure; Based on the communication conditions and data type characteristics of the terminals, as well as the model structure and model parameter values, a second number of terminals for performing model training is determined. Send scheduling information to the second number of terminals; the scheduling information includes model structure and model parameter values, as well as instructions to the terminals to perform model training.

14. A model learning device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to execute the model learning method according to any one of claims 1-5, or to execute the model learning method according to any one of claims 6-11.

15. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the mobile terminal, the mobile terminal is able to perform the model learning method according to any one of claims 1-5, or the mobile terminal is able to perform the model learning method according to any one of claims 6-11.

Images