Data processing method in a communication network, network-side equipment, and readable storage medium

Abstract

The present application discloses a data processing method in a communication network and a network side device, which belongs to the field of communication technology. The data processing method in a communication network of an embodiment of the present application includes: a first network element determines a first accuracy of a first model, the first accuracy being for indicating the accuracy degree of the first model used for actual inference; and when the first accuracy satisfies a predetermined condition, the first network element retrains the first model or reselects a second model.

Description

Technical Field

[0001] This application belongs to the field of communication technologies, and specifically relates to a data processing method and a network-side device in a communication network.

Background Art

[0002] In a communication network, usually, several network elements are introduced to perform intelligent model training, an inference task is executed based on the trained model, and an inference result can be obtained. This inference result can assist policy decisions by devices inside and outside the network to improve the intelligent level of the devices' policy decisions.

[0003] However, in actual applications, the accuracy achieved by the model after the training stage does not represent the inference accuracy that can be achieved when this model is actually used for inference. That is, when an inference task is executed based on the model, the accuracy of the inference result may be lower than the accuracy in the model training stage. In this case, if the inference result is provided to devices inside and outside the network, it will lead to incorrect policy decisions or the execution of inappropriate operations.

Summary of the Invention

Problems to be Solved by the Invention

[0004] Embodiments of this application provide a data processing method and a network-side device in a communication network, which can solve the problem that the accuracy during actual inference of the model affects policy decisions when the accuracy is low.

Means for Solving the Problems

[0005] In a first aspect, a step in which a first network element determines a first accuracy of a first model, where the first accuracy is for indicating the accuracy degree of the first model used for actual inference The present invention provides a data processing method in a communication network, which includes the step of, if the first accuracy satisfies a predetermined condition, the first network element retrains the first model or re-selects a second model.

[0006] In the second aspect, A decision module used to determine the first accuracy of a first model, wherein the first accuracy indicates the degree of accuracy of the first model used in actual inference; The present invention provides a data processing device in a communication network, comprising: a model training module used to retrain the first model or re-select a second model if the first accuracy satisfies predetermined conditions.

[0007] In the third aspect, A step in which a second network element performs an inference task based on a first model, wherein the first model is trained by the first network element, and the first network element includes a model training function network element. A data processing method in a communication network is provided, comprising the steps of: transmitting usage information of the first model and at least one of the first data to the first network element; and / or transmitting first instruction information to a seventh network element, wherein the first instruction information is for instructing the seventh network element to store the first data of the inference task, and the seventh network element includes a data storage function network element.

[0008] In the fourth aspect, A task execution module used to perform an inference task based on a first model, wherein the first model is trained by a first network element, and the first network element includes a task execution module that includes a model training function network element. A data processing device in a communication network is provided, comprising: a transmission module used to transmit at least one of the usage information of the first model and the first data to the first network element, and / or to transmit the first instruction information to the seventh network element, wherein the first instruction information is for instructing the seventh network element to store the first data of the inference task, and the seventh network element is a transmission module including a data storage function network element.

[0009] In the fifth aspect, a network-side device is provided, comprising a processor and a memory storing a program or command executable on the processor, wherein when the program or command is executed by the processor, the steps of the method described in the first aspect are realized.

[0010] In the sixth aspect, the present invention provides a processor and a communication interface, wherein the processor determines a first accuracy of a first model, the first accuracy being used to indicate the degree of accuracy of the first model to be used for actual inference, and provides network-side equipment used to retrain the first model or reselect a second model if the first accuracy satisfies predetermined conditions.

[0011] In the seventh aspect, a network-side device is provided, comprising a processor and a memory storing a program or command executable on the processor, wherein when the program or command is executed by the processor, the steps of the method described in the third aspect are realized.

[0012] In the eighth aspect, the present invention provides a network-side device comprising a processor and a communication interface, wherein the processor is used to perform an inference task based on a first model, the first model being trained by a first network element, the first network element including a model training function network element, and the communication interface is used to transmit at least one of the usage information and first data of the first model to the first network element and / or to transmit first instruction information to a seventh network element, the first instruction information being for instructing the seventh network element to store first data of the inference task, and the seventh network element including a data storage function network element.

[0013] In the ninth aspect, the present invention provides a data processing system in a communication network comprising a first network-side device and a second network-side device, wherein the first network-side device is capable of performing the steps of the method described in the first aspect, and the second network-side device is capable of performing the steps of the method described in the third aspect.

[0014] The present invention provides a readable storage medium in which a program or command is stored on the tenth side, and when the program or command is executed by a processor, a step of the method described on the first side or a step of the method described on the third side is realized.

[0015] The present invention provides a chip comprising a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor executes a program or command to implement the method described in the first aspect or the method described in the third aspect.

[0016] On the 12th aspect, there is provided a computer program / program product stored in a memory medium, which, when executed by at least one processor, realizes the steps of the method described in the 1st aspect or realizes the steps of the method described in the 3rd aspect.

Advantages of the Invention

[0017] In the embodiments of the present application, since the first accuracy of the first model when used in actual inference can be determined and the first model can be retrained when the first accuracy does not meet the predetermined conditions, when the accuracy of the first model when used in actual inference decreases, the accuracy of the first model can be adjusted by retraining the first model or reselecting the second model, and the accuracy of the first model when used in actual inference can be increased. As a result, it can better assist correct policy decisions or appropriate action operations by devices inside and outside the network.

Brief Description of the Drawings

[0018] [Figure 1] It is a schematic diagram of a wireless communication system according to an embodiment of the present application. [Figure 2] It is a schematic flowchart of a data processing method in a communication network according to an embodiment of the present application. [Figure 3] It is a schematic flowchart of a data processing method in a communication network according to an embodiment of the present application. [Figure 4] It is a schematic flowchart of a data processing method in a communication network according to an embodiment of the present application. [Figure 5] It is a schematic structural diagram of a data processing device in a communication network according to an embodiment of the present application. [Figure 6] It is a schematic structural diagram of a data processing device in a communication network according to an embodiment of the present application. [Figure 7] It is a schematic structural diagram of a communication device according to an embodiment of the present application. [Figure 8]It is a structural schematic diagram of a network-side device according to an embodiment of the present application.

Embodiments for Carrying out the Invention

[0019] In the following, while referring to the drawings in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly described. Naturally, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application.

[0020] Terms such as "first", "second", etc. in the specification and claims of the present application are not for describing a specific order or sequence, but for distinguishing similar objects. When used in this way, these terms may be interchangeable in some cases so that the embodiments of the present application can be implemented in an order other than that illustrated or described here. Objects distinguished by "first" and "second" generally belong to the same category, and the number of objects is not limited. For example, it should be understood that the first object may be one or more. Also, in the specification and claims, "and / or" represents at least one of the connected objects, and the symbol " / " generally represents that the related objects before and after are in an "or" relationship.

[0021] It should be noted that the technologies described in the embodiments of this application are not limited to Long Term Evolution (LTE) / LTE-Advanced (LTE-A) systems, but are also applicable to other wireless communication systems such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), and Single-carrier Frequency Division Multiple Access (SC-FDMA), as well as to other systems. In the embodiments of this application, the terms "system" and "network" are often used interchangeably, and the technologies described herein may be used with the above-mentioned systems and wireless communication technologies, or with other systems and wireless communication technologies. However, for illustrative purposes, the following description uses a 5th Generation (5G) system, and most of the following descriptions use 5G terminology. These technologies are applicable to systems other than 5G, such as 6th Generation (6G) communication systems.

[0022] Figure 1 shows a block diagram of a wireless communication system to which an embodiment of the present invention can be applied. The wireless communication system comprises a terminal 11 and a network-side device 12. The terminal 11 includes mobile phones, tablet personal computers, laptop computers (also called notebook computers), personal digital assistants (PDAs), palmtop computers, netbooks, ultra-mobile personal computers (UMPCs), mobile internet devices (MIDs), augmented reality (AR) / virtual reality (VR) devices, robots, wearable devices, vehicle user equipment (VUEs), pedestrian user equipment (PUEs), smart home devices (household appliances with wireless communication capabilities such as refrigerators, televisions, washing machines, or furniture), game consoles, and personal computers. Terminal-side equipment may include computers (PCs), ATMs, or kiosks. Wearable devices include smartwatches, smart bracelets, smart headphones, smart glasses, smart jewelry (smart bangles, smart bracelets, smart rings, smart necklaces, smart anklets, etc.), smart wristbands, smart wear, etc. It should be noted that the specific type of terminal 11 is not limited in the embodiments of this application. Network-side equipment 12 may include access network equipment or core network equipment, and access network equipment 12 may be called wireless access network equipment, radio access network (RAN), radio access network function, or radio access network unit.The access network equipment 12 may include a base station, a Wireless Local Area Network (WLAN) access point, or a WiFi node. The base station may also be called a Node B, an evolved Node B (eNB), an access point, a Base Transceiver Station (BTS), a radio base station, a radio transceiver, a Basic Service Set (BSS), an Extended Service Set (ESS), a Home B node, a Home evolved B node, a Transmitting Receiving Point (TRP), or any other appropriate term in the above-mentioned field, and the base station is not limited to any particular technical term as long as similar technical effects can be achieved. It should be noted that, although the embodiments of this application describe a base station in an NR system as an example, the specific type of base station is not limited.The core network equipment includes core network nodes, core network functions, Mobility Management Entity (MME), Access and Mobility Management Function (AMF), Session Management Function (SMF), User Plane Function (UPF), Policy Control Function (PCF), Policy and Charging Rules Function (PCRF), Edge Application Server Discovery Function (EASDF), Unified Data Management (UDM), Unified Data Repository (UDR), Home Subscriber Server (HSS), Centralized network configuration (CNC), Network Repository Function (NRF), Network Exposure Function (NEF), Local NEF (or L-NEF), Binding Support Function (BSF), and Application Function (Application It may include, but is not limited to, at least one of the following: Function (AF), etc. It should be noted that in the embodiments of this application, the core network equipment in a New Radio (NR) system is used as an example, but the specific type of core network equipment is not limited.

[0023] In communication networks, several network elements can be introduced to perform intelligent data analysis and generate data analysis results for several tasks. These data analysis results can assist in policy decisions by devices inside and outside the network, with the aim of improving the intelligence level of policy decisions by devices using artificial intelligence (AI) techniques.

[0024] For example, a Network Data Analytics Function (NWDAF) can train an artificial intelligence / machine learning (AI / ML) model based on training data to obtain a model suitable for a particular AI task. The NWDAF performs model inference based on the AI / ML model and inference input data to obtain data analysis results (analytics) corresponding to a specific AI inference task; these data analysis results (analytics) are also called inference results. A PCF within the network executes intelligent policy control and charging (PCC) policies based on certain inference results. For example, it might formulate intelligent user stay policies based on the analysis results of user business behavior to improve the user's business experience. Alternatively, an AMF might perform intelligent mobility management operations based on certain inference results or analytics. For example, it might intelligently page users based on the analysis results of user movement trajectories to improve paging reachability.

[0025] For devices inside and outside the network to make correctly optimized policy decisions based on AI data analysis results, it is essential that these decisions are based on accurate data analysis results. If the accuracy of the data analysis results is low and they are provided to devices inside and outside the network as incorrect information for reference, it will ultimately lead to incorrect policy decisions or inappropriate operations; therefore, it is necessary to guarantee the accuracy of the data analysis results. However, in actual applications, due to differences in data distribution, insufficient generalization ability of the model, etc., the model accuracy achieved after the training phase does not represent the inference accuracy that this model can achieve when used for actual inference (generally, the accuracy of the model's actual inference is lower than the accuracy during the model training phase). Thus, the above-mentioned problems of incorrect policy decisions or inappropriate operations are likely to occur.

[0026] To solve the above technical problems, the embodiment of the present application provides a data processing method and network-side equipment in a communication network, which can improve the accuracy of the first model when it is used for actual inference by adjusting the accuracy of the first model by retraining the first model or re-selecting the second model when the accuracy of the first model decreases when it is used for actual inference, thereby better assisting correct policy decisions or appropriate actions by devices inside and outside the network.

[0027] The data processing method and network-side equipment in the communication network provided in the embodiment of this application will be described in detail below with reference to the drawings, using several embodiments and their application scenarios.

[0028] As shown in Figure 2, an embodiment of the present invention provides a data processing method 200 in a communication network, which can be performed by a first network element, which may be a network-side device in the embodiment shown in Figure 1. In other words, this method can be performed by software or hardware installed on the first network element or network-side device. This method includes the following steps.

[0029] S202: The first network element determines the first accuracy of the first model, and the first accuracy indicates the degree of accuracy of the first model used in actual inference.

[0030] The first network element may be a model training function network element in a communication network, and the model training function network element has AI / ML model training capabilities and can be used to train an AI / ML model based on training data. Optionally, the first network element may be a Model Training Logical Function (MTLF).

[0031] The first model can be obtained by training the first network element. Specifically, the first network element can acquire training data from other network elements (e.g., network elements capable of providing training data), and after performing AI / ML training based on this training data, the first model can be obtained. The training data includes input data and label data, and the label data corresponds to the input data and may be the actual values ​​of the data (ground truth), i.e., facts and data that actually occurred.

[0032] The first network element, when obtained by training the first model, can determine the first accuracy of the first model, which may be the accuracy in use of the first model's inference. In this embodiment, the first accuracy may indicate the degree of accuracy of the first model used in actual inference, and specifically, the first accuracy may indicate the degree of correctness and / or error of the inference result during actual inference.

[0033] It should be explained that the first accuracy can be expressed in various forms, such as a specific percentage value like 90%, a category like high, medium, or low, or normalized data like 0.9, where the form of expression for the first accuracy is not specifically limited. The first accuracy can positively or negatively indicate the degree of accuracy or error of the inference results of the first model on the task. For example, the accuracy of the inference of the first model can be negatively indicated by calculating the model's inference error or inference error rate. There are various methods for calculating the inference error or inference error rate, such as the Mean Absolute Error (MAE) or the Mean Squared Error (MSE).

[0034] Optionally, as one embodiment, the step of a first network element determining the first accuracy of the first model is: The first network element acquires the first data, The process may include the step of a first network element determining a first accuracy based on first data.

[0035] The first data is, Inference input data and, Inference result data corresponding to the inference input data, It includes at least one of the inference input data and the corresponding label data.

[0036] The inference input data may be the model input data when performing an inference task based on the first model, the inference result data may be the model output result obtained after inferring the inference input data based on the first model, and the label data may be the actual result data corresponding to the inference input data.

[0037] In this embodiment, the step of the first network element acquiring the first data can be implemented in multiple ways. Optionally, in the first implementation method, the step of the first network element acquiring the first data is: The steps include receiving usage information of the first model transmitted by the second network element, A step of determining the source information of the first data according to the usage information, The procedure may also include the step of obtaining first data according to the source information.

[0038] In the second implementation method, the step in which the first network element acquires the first data is: The step may also include receiving the first data transmitted by the second network element.

[0039] The second network element can perform an inference task based on the first model. The second network element may be a model inference function network element in a communication network, and the model inference function network element has a model inference function and can infer inference input data corresponding to the inference task based on the first model to obtain inference result data. Optionally, the second network element may be an Analytics Logical Function (AnLF). Specifically, the second network element may be a network element that previously requests the first network element to obtain the first model, or a network element that previously requests the first network element to obtain model information of the first model. When the first network element receives the first data transmitted by the second network element, it can receive the first data from the network element that previously requested the first network element to obtain the first model or model information of the first model.

[0040] In the third implementation method, the step in which the first network element acquires the first data is: The step of receiving the first data transmitted by the seventh network element may include the step of the seventh network element including a data storage function network element, which may specifically be an Analytics Data Repository Function (ADRF).

[0041] In the first implementation described above, the first network element can train and acquire the first model before receiving usage information of the first model transmitted by the second network element, and then transmit the model information of the first model to the second network element. Specifically, when the second network element performs an inference task using the first model, it can send a model request message to the first network element, which requests that the first model be acquired. After receiving the model request message, the first network element can transmit the model information of the first model to the second network element if the first network element has trained and acquired the first model; if the first network element has not trained the first model, it can train the first model, and after training and acquiring the first model, transmit the model information of the first model to the second network element. Alternatively, the first network element may actively transmit the model information of the first model to the second network element after training and acquiring the first model, and the second network element can perform an inference task using the first model without sending a model request message to the first network element if it needs to perform an inference task. The first network element, when training the first model, obtains training data from other network elements (network elements capable of providing training data), and then trains the first model based on that training data. When the first network element sends model information of the first model to the second network element, it can do so via Nnwdaf_MLModelProvision_Notify or Nnwdaf_MLModelInfo_Response.

[0042] The model information for the first model mentioned above is: The first model and, This may include at least one of the following: a second accuracy of the first model, which indicates the degree of accuracy of the model output results presented by the first model during the training or testing phase.

[0043] The second accuracy may be the accuracy in training (AiT) of the first model, and can be used to indicate the degree of accuracy during the model training phase. Specifically, the second accuracy can be used to indicate the degree of correctness and / or erroneousness of the training results during the model training phase. Optionally, the second accuracy may be equal to the number of times the model's decision results are correct divided by the total number of decisions, i.e., second accuracy = number of correct decisions / total number of decisions. Here, a correct decision result may mean that the decision result matches the label data, and / or that the difference between the decision result and the label data is within an acceptable range. Optionally, the first network element may be configured with a validation dataset for evaluating the model's second accuracy, and this validation set may include the model input data and the actual label data. The first network element can input validation input data into the trained model to obtain output data, compare the output data with the actual label data to determine whether the decision result is correct or not, and finally obtain the second accuracy of the first model using the second accuracy calculation formula.

[0044] It should be explained that the second accuracy can be expressed in various forms, such as a specific percentage value like 90%, a category like high, medium, or low, or normalized data like 0.9, and the form of expression for the second accuracy is not specifically limited. The second accuracy can positively or negatively indicate the degree of accuracy or error during the model training phase of the first model. For example, the degree of accuracy during the training phase of the first model can be negatively indicated by calculating the model training error or training error rate. There are various methods for calculating the training error or training error rate, such as MAE or MSE.

[0045] After the first network element transmits model information of the first model to the second network element, the second network element can perform an inference task based on the first model. For specific implementation methods of the second network element's execution of the inference task, refer to the specific implementation methods of the corresponding steps in the embodiment shown in Figure 3, and will not be described in detail here. After performing the inference task, the second network element can transmit its usage information of the first model to the first network element, and the first network element can receive the usage information of the first model transmitted by the second network element. Optionally, the first network element may send a first request message to the second network element before receiving the usage information of the first model transmitted by the second network element, the first request message is intended to request that the second network element obtain the usage information of the first model. Upon receiving the first request message, the second network element can transmit its usage information of the first model to the first network element. In other words, information on the use of the first model by the second network element may be actively transmitted to the first network element by the second network element, or it may be transmitted to the first network element by the second network element when it receives a first request message from the first network element, but is not specifically limited. Information on the use of the first model by the second network element is Model identifier information for the first model, Task identifier information for inference tasks performed based on the first model, Conditional information for the inference task, It may include at least one of the target information for the inference task.

[0046] Model identifier information (Model ID) can be used to indicate which model a second network element uses, and the model identifier information of the first model can be used to indicate the first model. Task identifier information (analytic ID) can identify the type of inference task and can be used to determine the corresponding model, and the task identifier information of an inference task performed based on the first model can be used to determine the first model. Analytics filter information can be used to limit the scope of execution of an inference task, such as time range or geographical range. Analytics target information can be used to indicate the target of an inference task, which may be, for example, a Target terminal (User Equipment, UE) (if there is a target UE, i.e., a task objective), or a Network Function (NF) instance.

[0047] Optionally, the information on the use of the first model by the second network element may include at least one of three things: inference input data when the second network element performs an inference task based on the first model, inference result data corresponding to the inference input data, and label data corresponding to the inference input data. The inference input data can be collected by the second network element; for example, the second network element can collect and perform inference on the inference input data when it receives a task request for an inference task from another network element (e.g., a consumer network element). Alternatively, the inference input data can be actively collected by the second network element. The inference result data can be obtained by the second network element based on the first model and the inference input data when performing an inference task. The label data can be obtained by the second network element from another network element (e.g., a source device for label data); for example, the second network element can send a data acquisition request to another network element and request that it acquire label data.

[0048] After receiving usage information of the first model from the second network element, the first network element can determine the source information of the first data according to that usage information. The source information of the first data is: A third network element for providing inference input data corresponding to the inference task, It includes at least one of a fourth network element for providing label data corresponding to an inference task.

[0049] The third network element may include source devices for inference input data. The third network element can be determined by the first network element, which can determine the target and scope of an inference task based on information such as analytics filter information and analytics target information within the usage information, and can determine from which specific network elements to obtain inference input data corresponding to the inference task based on said target and scope, as well as metadata information, and these determined network elements are the third network element.

[0050] The fourth network element may include a source device for label data. The fourth network element can be determined by the first network element, which specifically determines a corresponding network element device type (NF type) capable of providing the output data type (data type) of the first model, determines a network element instance corresponding to that network element device type according to the target of the inference task, limited information, etc., and can make that network element instance the fourth network element. For example, the first model is a UE mobility model, the output data type of the first model is UE location, and based on that output data type, it can be determined that the AMF type is UE location data information. If, according to limited information such as the target UE1 and area of ​​interest (AOI) of the inference task, the UDM or NRF queries that the corresponding AMF instance is AMF1, then AMF1 is the fourth network element, and the first network element can use AMF1 as the source of label data and obtain the actual UE location label data value from AMF1.

[0051] Optionally, the above source information is The first model is designed to handle inference tasks, The input data type for the first model, It may further include at least one of the output data types of the first model.

[0052] The first model can be determined by the first network element based on the model identifier information and / or task identifier information in the usage information. For example, if the usage information includes task identifier information (analytics ID) and there is a mapping relationship between the task identifier information and the model identifier information (model ID), then for a certain analytics ID (e.g., analytics ID=UE mobility for predicting a user's movement trajectory), it can be determined based on the mapping relationship that the corresponding model identifier information is model1, and the model corresponding to model1 is the first model corresponding to the inference task.

[0053] The input data types (which can be called metadata) and output data types of the first model relate to the resulting data for the inference task or prediction to which the first model is specifically applied. For example, if the first model is used to predict a user's movement trajectory, the input data types of the first model may include UE ID, time, the current business state of the UE, etc., and the output data types may include UE location (e.g., Tracking Area (TA) / cell) etc.

[0054] The first network element can determine the source information of the first data and then acquire the first data according to the source information. The acquisition of the first data according to the source information by the first network element is: Sending an input data acquisition request message to the third network element, wherein the input data acquisition request message is intended to request the acquisition of inference input data, This may include at least one of the following: sending a label data acquisition request message to a fourth network element, wherein the label data acquisition request message is for requesting the acquisition of label data.

[0055] The input data acquisition request message can be used by the third network element to determine which inference input data to feed back to the first network element, and the input data acquisition request message contains: Type information of the inference input data, The target information corresponding to the inference input data, The inference input data may include at least one of the following: time information corresponding to the data.

[0056] The type information of the inference input data, the target information corresponding to the inference input data, and the time information corresponding to the inference input data are determined by the first network element according to the input data type of the inference process, the inference target, and the time period to be inferred. In other words, the first network element determines the type of inference input data that needs to be acquired according to the input data type of the inference process, determines the target of the inference input data that needs to be acquired according to the inference target, and determines the time information (timestamp, time zone, etc.) of the inference input data according to the time period to be inferred. If the inference process is a statistical calculation performed on a past time or a prediction performed on a future time, the time information may be past time or future time.

[0057] A label data acquisition request message can be used by the fourth network element to determine which label data to feed back to the first network element, and the label data acquisition request message includes: Label data type information, The target information corresponding to the label data, The label data may include at least one of the following: time information corresponding to the label data.

[0058] The type information of the label data, the target information corresponding to the label data, and the time information corresponding to the label data are determined by the first network element according to the output data type of the inference process, the inference target, and the time period to which the inference is performed. In other words, the first network element determines the type of label data that needs to be acquired according to the output data type of the inference process, the target of the label data that needs to be acquired according to the inference target, and the time information (timestamp, time zone, etc.) of the label data according to the time period to which the inference is performed. If the inference process is a statistical calculation performed on a past time or a prediction performed on a future time, the time information may be past time or future time.

[0059] For example, MTLF (first network element) sends a label data acquisition request message to AMF or LMF (third network element), and the request message If the label data type being transported is UE location, the target information is UE ID1, and the time information is a specific time period, then the request message This is used to request the AMF / LMF to provide feedback on the UE location value of UE ID1 for a specific time period.

[0060] After receiving an input data acquisition request message, the third network element can send the corresponding inference input data to the first network element. In this way, the first network element can acquire the inference input data from the third network element. Similarly, after receiving a label data acquisition request message, the fourth network element can send the corresponding label data to the first network element, and the first network element can acquire the label data from the fourth network element. It should be noted that when the second network element performs one or more inference processes during the execution of an inference task and obtains multiple inference output results, the first network element needs to acquire corresponding label data values ​​from the fourth network element that correspond to the multiple inference output results.

[0061] Furthermore, it should be explained that the first network element acquires inference input data and / or label data as described above, and regarding the inference result data, the first network element can acquire it from the second network element when acquiring the inference result data, and the inference result data acquired from the second network element can be obtained after the second network element has performed an inference task based on the first model. Also, since the inference result data can be obtained by inference based on the inference input data and the first model, when the first network element acquires the inference result data, it may be possible for the first network element to acquire the inference input data and then perform inference based on the inference input data and the first model to obtain the inference result data. In this way, the first network element does not need to acquire the inference result data from other network elements, thus simplifying the data acquisition step.

[0062] In the second implementation described above, that is, in a situation where the first network element acquires the first data by receiving the first data transmitted by the second network element, the first data acquired by the first network element may include at least one of three: inference input data, inference result data, and label data. The inference input data can be collected by the second network element; for example, the second network element can collect the inference input data and perform inference when it receives a task request for an inference task from another network element (e.g., a consumer network element). Alternatively, the inference input data can be actively collected by the second network element. The inference result data can be obtained by the second network element based on the first model and the inference input data when performing an inference task. The label data can be acquired by the second network element from another network element (e.g., a source device for label data); for example, the second network element can acquire label data by sending a data acquisition request to another network element.

[0063] Optionally, before the first network element receives the first data transmitted by the second network element, The first network element sends a second request message to a second network element, which may further include the step of requesting the second network element to retrieve the first data collected by the second network element.

[0064] In other words, when the first network element acquires the first data, it can send a second request message to the second network element, which requests the second network element to acquire the first data that it has collected. When the second network element receives the second request message, it can send the first data to the first network element.

[0065] Optionally, the second request message may be a subscribe message. message teeth, Identifier information for the inference task, Limiting conditions information for the inference task, The target information for the inference task, Identifier information for the first model, The input data type information of the first model, It includes at least one of the output data type information of the first model.

[0066] Optionally, the second request message may further include a reason for the request, such as the need to retrain the first model, or the accuracy of the first model not meeting or degrading accuracy requirements.

[0067] In the third implementation described above, that is, in a situation where the first network element acquires the first data by receiving the first data transmitted by the seventh network element, the first data acquired by the first network element may include at least one of three: inference input data, inference result data, and label data. The first data can be stored in the seventh network element by the second network element, specifically, the second network element can transmit first instruction information to the seventh network element after performing an inference task, and the first instruction information is for instructing the seventh network element to store the first data of the inference task. Optionally, the first instruction information is Identifier information for the inference task, Limiting conditions information for the inference task, The target information for the inference task, Inference input data corresponding to the inference task, Inference result data corresponding to the inference task, It includes at least one of the label data corresponding to the inference task.

[0068] The information regarding the reason for memory retention may include, for example, that the second network element has completed an inference task, that the seventh network element needs to periodically store the first data, or that the accuracy of the first model used when the second network element performs the inference task does not meet the accuracy requirement or is reduced.

[0069] The first data stored in the seventh network element can be transmitted to the seventh network element by the second network element. The method by which the second network element obtains the first data may be that the inference input data is collected by the second network element. For example, when the second network element receives a task request for an inference task from another network element (e.g., a consumer network element), it can collect the inference input data and perform inference. Alternatively, the inference input data can be actively collected by the second network element. The inference result data can be obtained by the second network element based on the first model and the inference input data when the inference task is executed. Label data can be obtained by the second network element from another network element (e.g., a source device for label data), for example, by sending a data acquisition request to another network element.

[0070] Optionally, before the first network element receives the first data transmitted by the seventh network element, The step of a first network element sending a third request message to a seventh network element may further include the step of the third request message requesting the acquisition of first data.

[0071] In other words, when the first network element retrieves the first data, it can send a third request message to the seventh network element, which requests the seventh network element to retrieve the first data stored in it. When the seventh network element receives the third request message, it can send the first data to the first network element.

[0072] Optionally, the third request message may be a subscribe message. message teeth, Identifier information for the inference task, Limiting conditions information for the inference task, The target information for the inference task, Identifier information for the first model, The input data type information of the first model, It includes at least one of the output data type information of the first model.

[0073] Optionally, the third requirement message The requirements may further include reasons for the requirement, such as the need to retrain the first model, or the accuracy of the first model not meeting or decreasing accuracy requirements.

[0074] What needs to be explained is that in actual applications, the first network element can acquire the first data by one or more of the three methods described above; that is, the first network element can acquire the first data from the second network element, and / or determine the source information of the first data according to the usage information of the first model by the second network element and acquire the first data according to that source information, and / or acquire the first data from the seventh network element.

[0075] After acquiring the first data, the first network element can determine the first accuracy of the first model based on the first data.

[0076] When determining the first accuracy based on the first data, optionally, if the first data includes inference input data and label data but not inference result data, the first network element can input the inference input data into the first model, determine the inference result data corresponding to the inference input data, and then determine the first accuracy based on the inference result data and label data. If the first data includes inference result data and label data, the first network element can directly determine the first accuracy based on the inference result data and label data.

[0077] When determining the first accuracy based on the inference result data and label data, the inference result data and label data are specifically compared, and the ratio of the number of correct inferences to the total number of inferences is determined. This ratio is the first accuracy of the first model. Here, a correct inference result may mean that the inference result data and label data match, or that the difference between the inference result data and label data is within an acceptable range. The expression of the first accuracy may be a percentage value (e.g., 90%), a category (e.g., high, medium, low), or a normalized number (e.g., 0.9), but is not specifically limited to these.

[0078] S204: If the first accuracy satisfies the predetermined conditions, the first network element retrains the first model or Model 2 Select again.

[0079] The first network element, after determining the first accuracy of the first model, can determine whether the first accuracy satisfies predetermined conditions and decide whether it is necessary to retrain the first model or re-select the second model.

[0080] Optionally, as one embodiment, the first accuracy satisfying the predetermined conditions is The first accuracy is less than the second accuracy, which indicates the degree of accuracy of the model output results presented by the first model during the training or testing phase. The first accuracy is less than the second accuracy, and the difference between the first accuracy and the second accuracy is greater than a predetermined value. This may include at least one of the following: a first accuracy is less than a predetermined accuracy, where the predetermined accuracy can be set according to actual needs and is not specifically limited thereto.

[0081] If the first accuracy satisfies a predetermined condition, it can be indicated that the accuracy of the first model used for actual inference does not meet the actual needs, in which case the first network element can retrain the first model or reselect the second model. Retraining the first model may involve correcting the first model without changing its model structure, or it may involve changing the model structure of the first model and training the first model with the new model structure. Optionally, there may be two ways of implementing model retraining: one is to restart model training from scratch based on the training data, and the other is to fine-tune the first model based on the training data, which can converge faster and save resources. Reselecting the second model may involve reselecting another new existing model, the reselected second model may be one whose accuracy is higher than the first threshold or which meets the performance requirements of the model.

[0082] Optionally, as one embodiment, when the first network element retrains the first model, A step of acquiring target training data, wherein the target training data includes target input data and target label data corresponding to the target input data, The procedure may include a step of retraining the first model based on target training data.

[0083] The target training data is different from the training data used when training the first model previously. First network element When retraining the first model, the model can be trained using new training data to adjust the accuracy of the first model, thereby improving the accuracy of the retrained first model.

[0084] Optionally, as one embodiment, the first network element acquires the target training data. To obtain the first training data used during the training of the first model, The fifth network element is determined, and the second training data is obtained from the fifth network element, provided that the fifth network element is for providing the training data. This may include determining a sixth network element and obtaining inference data from the sixth network element, wherein the sixth network element is for providing the inference data.

[0085] Each of the first training dataset, the second training dataset, and the inference dataset includes input data and label data.

[0086] The fifth network element may include the source device for training data. Optionally, in one embodiment, the first network element may determine the fifth network element. This may include determining a fifth network element based on the second piece of information.

[0087] The second set of information includes task identifier information and / or conditional information for inference tasks performed based on the first model. The task identifier information (analytic ID) can identify the type of inference task and can be used to determine the corresponding model. The conditional information for inference tasks (analytics filter information) can be used to limit the scope of execution of the inference task, such as the time range and geographical range.

[0088] The sixth network element may include a source device for inference data. Optionally, in one embodiment, the first network element may determine the sixth network element. This may include determining the sixth network element based on the second piece of information described above.

[0089] The first network element can obtain the second model after retraining the first model or re-selecting the second model. After the second model is obtained, optionally, as in one embodiment, The model information of the retrained second model or the re-selected second model is transmitted to the second network element, and the second network element performs an inference task based on the second model. The process may further include transmitting model information of a retrained or re-selected second model to a seventh network element, and storing the model information of the second model by the seventh network element.

[0090] The second network element here may be the second network element in the current task, i.e., the second network element that previously performed an inference task based on the first model and then sent usage information of the first model to the first network element. After the first network element sends the model information of the second model to the second network element, the second network element can re-execute the previous inference task based on the second model or perform a new inference task. Because the second model is a retrained model, the accuracy of the inference result data is high. Optionally, the second network element may be another network element that performs an inference task based on the second model, and after the first network element sends the model information of the second model to the other second network element, the other second network element can perform an inference task based on the second model. Because the second model is a retrained model, the accuracy of the inference result data is high.

[0091] The seventh network element includes a data storage function network element, i.e., a network element that stores model information for the second model. Optionally, the seventh network element may be an Analytics Data Repository Function (ADRF). Storing model information for the second model in the seventh network element makes it easier for other network elements to find models or data.

[0092] The model information for the second model is as follows: Model identifier information for the second model, Task identifier information for inference tasks performed based on the second model, Information on the scope of application of the second model, The third accuracy of the second model indicates the degree of accuracy of the model output results presented by the second model during the training or testing phase, Training data for the second model, It may include at least one of the second model.

[0093] The Model Identifier (Model ID) of the second model is used to identify the second model. The Analytic Identifier (analytic ID) of the inference task performed based on the second model can be used to determine the corresponding second model. The scope information of the second model can be used to limit the scope of execution of the inference task, such as the time range and domain range. The third accuracy of the second model can be called the accuracy in training (AiT) of the second model and describes the degree of accuracy of identification or decision that the model can achieve after it has been trained. Specifically, the third accuracy can be used to indicate the degree of correctness and / or incorrectness of the model output results presented by the second model during the training or testing phase. The method for determining the third accuracy is the same as the method for determining the second accuracy of the first model described above, and the expression format of the third accuracy may be the same as the expression format of the second accuracy, but this will not be explained again here. The training data of the second model is the training data used when training the second model, and may include input data and label data, and may specifically be the target training data obtained when retraining the first model described above. The second model includes, but is not limited to, descriptive information and / or a model file for the second model, and the model file may include elements such as the complete network structure and parameter information that generate the second model.

[0094] It should be explained that the model information transmitted by the first network element to the second and seventh network elements may be the same or different. For example, the first network element can transmit the second model to the second network element, and can transmit the model identifier information of the second model, the task identifier information of the inference task performed based on the second model, the scope information of the second model, the third accuracy of the second model, the training data of the second model, and the second model to the seventh network element.

[0095] In the embodiment of the present invention, the first accuracy of the first model when used for actual inference is determined, and the first model can be retrained if the first accuracy does not meet predetermined conditions. Therefore, if the accuracy of the first model when used for actual inference decreases, the accuracy of the first model can be adjusted by retraining the first model or re-selecting the second model, thereby improving the accuracy of the first model when used for actual inference. This better assists devices inside and outside the network in making correct policy decisions or appropriate actions.

[0096] As shown in Figure 3, an embodiment of the present invention provides a data processing method 300 in a communication network, which can be performed by a second network element, which may be a network-side device in the embodiment shown in Figure 1. In other words, this method can be performed by software or hardware installed on the second network element or network-side device. It should be noted that the second network element may be deployed independently as a separate network element device from the first network element in the embodiment shown in Figure 2, or it may be deployed together with the same network element device, for example, with an NWDAF, in which case the NWDAF can provide model inference functionality while also providing model training functionality. The method shown in Figure 3 includes the following steps.

[0097] S302: The second network element performs an inference task based on the first model, the first model is trained by the first network element, and the first network element includes a model training function network element.

[0098] The first model can be obtained by training a first network element on training data, which can be acquired by the first network element from other network elements (network elements capable of providing training data). In this step, the second network element can perform an inference task based on the first model trained by the first network element.

[0099] Optionally, as one embodiment, the second network element may obtain the first model from the first network element before performing an inference task based on the first model, specifically, A step of sending a model request message to a first network element, wherein the model request message is for requesting to obtain a first model, The step may include receiving model information of a first model transmitted by a first network element.

[0100] After the second network element sends a model request message to the first network element, if the first network element has trained and obtained the first model, it can send the model information of the first model to the second network element, and the second network element can receive the model information of the first model sent by the first network element. If the first network element has not trained the first model, it can obtain training data from another network element (a network element that can provide training data), train the first model based on the training data, and send the model information of the first model to the second network element, and the second network element can receive the model information of the first model sent by the first network element. When the second network element receives the model information of the first model, it can receive it via Nnwdaf_MLModelProvision_Notify or Nnwdaf_MLModelInfo_Response.

[0101] The model information for the first model is: The first model and, It includes at least one of the following: a second accuracy of the first model, which indicates the degree of accuracy of the model output results presented by the first model during the training or testing phase.

[0102] The second accuracy of the first model may also be the training accuracy (AiT) of the first model, and can be used to indicate the degree of accuracy during the model training phase. Specifically, the second accuracy can be used to indicate the degree of correctness and / or error of the training results during the model training phase. Optionally, the second accuracy may be equal to the number of times the model's decision results are correct divided by the total number of decisions, i.e., second accuracy = number of correct decisions / total number of decisions. Here, a correct decision result may mean that the decision result matches the label data, and / or that the difference between the decision result and the label data is within an acceptable range. The expression of the second accuracy may be a percentage value (e.g., 90%), a categorical expression (e.g., high, medium, low), or a normalized number (e.g., 0.9), but is not specifically limited. The second accuracy can indicate, positively or negatively, the degree of accuracy or error during the model training phase of the first model. For example, the accuracy during the training phase of the first model can be negatively indicated by calculating the model training error or training error rate. There are various methods for calculating the training error or training error rate, such as MAE or MSE.

[0103] The second network element, after receiving model information from the first model, can perform an inference task based on the first model.

[0104] Optionally, as one embodiment, the second network element may perform an inference task based on the first model. The eighth network element receives a task request for an inference task and executes the inference task based on the first model. This includes at least one of the following: obtaining a task request for an inference task that has been simulated and triggered by a second network element, and performing the inference task based on a first model.

[0105] In other words, the execution of an inference task by the second network element based on the first model may be triggered when it receives an inference task transmitted by the eighth network element, and / or the second network element may set up a validation test phase in which the second network element itself simulates triggering an inference task to estimate the accuracy of the model inference. The eighth network element includes a consumer network element, which may specifically be a consumer network function entity (consumer NF), and the consumer NF may be a 5G network element or an AF terminal. When the eighth network element transmits an inference task to the second network element, it can do so via Nnwdaf_AnalyticsSubscription_Subscribe or Nnwdaf_AnalyticsInfo_Request.

[0106] When an inference task is transmitted to the second network element by the eighth network element, if the inference task has inference input data, the inference input data can be input to the first model, thereby obtaining inference result data. If the inference task does not have inference input data, when the second network element performs the inference task based on the first model, Sending an input data acquisition request message to a third network element, wherein the input data acquisition request message is intended to request the acquisition of inference input data corresponding to the inference task, Receiving inference input data transmitted by the third network element, This may also include inputting inference input data into a first model to obtain inference result data.

[0107] The third network element here may be the third network element in the embodiment shown in Figure 2, and the third network element can provide inference input data corresponding to the inference task. An input data acquisition request message can be used by the third network element to determine which inference input data to feed back to the second network element, and the input data acquisition request message may include at least one of the following: type information of the inference input data, target information corresponding to the inference input data, and time information corresponding to the inference input data. Specifically, refer to the correspondence in the embodiment shown in Figure 2, and will not be described again here. When the second network element sends an input data acquisition request message to the third network element, it can send it using Nnwdaf_AnalyticsSubscription_Subscribe or Nnwdaf_AnalyticsInfo_Request.

[0108] After receiving an input data acquisition request message, the third network element can send the corresponding inference input data to the second network element. After receiving the inference input data sent by the third network element, the second network element can input the inference input data into the first model, thereby obtaining the corresponding inference result data.

[0109] For example, the second network element can perform an inference calculation using model1, which corresponds to analytics ID=UE mobility (identifying the inference task type, for example, analytics ID=UE mobility to predict the user's movement trajectory), and the values ​​of the input data corresponding to model1 (e.g., UE ID, time, current business state of the UE), and obtain the output value of the inference result UE location.

[0110] When an inference task is triggered by a second network element, the second network element can generate or determine inference input data when it performs the inference task based on the first model, and the inference result data can be obtained by inputting the inference input data into the first model.

[0111] What needs to be explained is that when a second network element performs an inference task, it can obtain multiple output values ​​by executing the inference computation process once. Alternatively, it can obtain multiple inference output values ​​by performing the inference multiple times.

[0112] Optionally, in one embodiment, when an inference task performed by a second network element is transmitted by an eighth network element, the second network element may, after performing the inference task using the first model, transmit the obtained inference result data to the eighth network element to assist in policy decisions made by the eighth network element.

[0113] S304: Transmits usage information of the first model and at least one of the first data to a first network element, and / or transmits first instruction information to a seventh network element, which instructs the seventh network element to store the first data of the inference task, wherein the seventh network element includes a data storage function network element.

[0114] After performing an inference task, the second network element can transmit usage information of the first model and at least one of the first data to the first network element. After receiving the usage information and at least one of the first data, the first network element can decide whether to retrain the first model or re-select the second model. A specific implementation can be found in the corresponding steps in the embodiment shown in Figure 2, which will not be described again here.

[0115] The usage information of the first model by the second network element is: Model identifier information for the first model, Task identifier information for inference tasks performed based on the first model, Conditional information for the inference task, It includes at least one of the target information for the inference task.

[0116] The Model Identifier (Model ID) of the first model can be used to indicate that the model used by the second network element is the first model. The Analytic Identifier (analytic ID) can identify the type of inference task, and the Task Identifier of an inference task performed based on the first model can be used to determine the first model. The Analytics Filter Information can be used to limit the scope of execution of the inference task by the second network element, such as time range or geographical range. The Analytics Target can be used to indicate the target of the inference task performed by the second network element, which may be, for example, a Target UE (i.e., a UE where the task objective is located) or a certain NF instance. Optionally, the information on the second network element's use of the first model may include inference input data and / or inference result data when the second network element performs an inference task based on the first model.

[0117] Optionally, as one embodiment, before the second network element transmits the usage information of the first model to the first network element, The second network element receives the first request message sent by the first network element. and Furthermore, the first request message is intended to request that the second network element obtain information on the use of the first model.

[0118] In other words, when the second network element receives the first request message from the first network element, it can send usage information for the first model to the first network element.

[0119] The first data is, Inference input data and, Inference result data corresponding to the inference input data, It may include at least one of the following: the inference input data and the corresponding label data.

[0120] The inference input data may be the model input data when performing an inference task based on the first model, the inference result data may be the model output result obtained after inferring the inference input data based on the first model, and the label data may be the actual result data corresponding to the inference input data.

[0121] Optionally, as one embodiment, before the second network element transmits the first data to the first network element, The second network element collects inference input data, The second network element further includes at least one of collecting label data corresponding to the inference input data.

[0122] In other words, if the first data includes inference input data, this inference input data can be collected by the second network element. If the first data includes label data corresponding to the inference input data, this label data can also be collected by the second network element. Optionally, if the first data includes inference result data, this inference result data can be obtained by the second network element after inference based on the first model and the inference input data.

[0123] Optionally, the second network element can collect inference input data. When the second network element receives a task request for an inference task from the eighth network element, it collects inference input data, provided that the eighth network element includes a consumer network element. The second network element may include at least one of the following: actively collecting inference input data.

[0124] In other words, the second network element may collect inference input data to perform inference when it receives a task request for an inference task, or it may actively collect inference input data. A task request for an inference task can be transmitted to the second network element by the eighth network element, which includes a consumer network element, which may specifically be a consumer NF, and the consumer NF may be a 5G network element or an AF terminal, etc.

[0125] Optionally, the second network element may collect label data corresponding to the inference input data. The second network element may also send a data retrieval request to the fourth network element, the data retrieval request being for the retrieval of label data corresponding to the inference input data.

[0126] The fourth network element may be the source device for the label data. When the second network element collects label data, it sends the data to the fourth network element. acquisition A request can be sent, and if the fourth network element receives a data acquisition request, it can send the corresponding label data to the second network element.

[0127] Optionally, before the second network element sends the first data to the first network element, The step of receiving a second request message transmitted by a first network element may further include the step of requesting that the second request message retrieve first data collected by the second network element.

[0128] In other words, if the second network element receives the second request message sent by the first network element, it can send the first data to the first network element.

[0129] Optionally, the second request message may be a subscribe message. message teeth, Identifier information for the inference task, Limiting conditions information for the inference task, The target information for the inference task, Identifier information for the first model, The input data type information of the first model, It includes at least one of the output data type information of the first model.

[0130] Optionally, the second requirement message The requirements may further include reasons for the requirement, such as the need to retrain the first model, or the accuracy of the first model not meeting or decreasing accuracy requirements.

[0131] In this embodiment, the second network element performs the inference task, and then the seventh Network elements A first instruction information may be transmitted to the seventh network element, which is intended to instruct the seventh network element to store first data for an inference task, and the seventh network element includes a data storage function network element, which may specifically be an ADRF.

[0132] Optionally, the first instruction information is: Identifier information for the inference task, Limiting conditions information for the inference task, The target information for the inference task, Inference input data corresponding to the inference task, Inference result data corresponding to the inference task, It includes at least one of the label data corresponding to the inference task.

[0133] Regarding memory reason information, for example, the second network element needs to complete an inference task, the seventh network element needs to periodically store the first data, and the accuracy of the first model used when the second network element performs task inference does not meet or decreases the accuracy requirement. It is fine to do so. .

[0134] After the second network element transmits the first instruction information to the seventh network element, the seventh network element can store the first data according to the first instruction information. The first data can be determined by the second network element in the manner described above, and will not be explained again here. After the seventh network element has stored the first data, the first network element can obtain the first data from the seventh network element when it needs to. The specific implementation method is described in the embodiment shown in Figure 2, and will not be explained again here.

[0135] Optionally, as one embodiment, the second network element is: Receives model information of the second model transmitted by the first network element. do Often, the second model is obtained by retraining the first model with the elements of the first network, or the second model is a model re-selected by the elements of the first network.

[0136] Specifically, after the first network element receives usage information and / or first data of the first model from the second network element, if it determines that the first accuracy of the first model satisfies a predetermined condition, it can retrain the first model or re-select the second model. After retraining to obtain the second model or re-selecting the second model, the first network element can transmit the model information of the second model to the second network element, at which point the second network element can receive the model information of the second model transmitted by the first network element. Subsequently, the second network element can re-execute a previously performed inference task using the second model or execute a new inference task. Since the second model is a retrained model, the accuracy of the inference result data is high.

[0137] The model information for the second model is as follows: Model identifier information for the second model, Task identifier information for the inference task executed based on the second model described above, Information on the scope of application of the second model, The third accuracy of the second model indicates the degree of accuracy of the model output results presented by the second model during the training or testing phase, Training data for the second model, It may include at least one of the second model.

[0138] The Model Identifier (Model ID) of the second model is used to identify the second model. The Analytic Identifier (analytic ID) of the inference task performed based on the second model can be used to determine the corresponding second model. The scope information of the second model can be used to limit the scope of execution of the inference task, such as the time range and domain range. The third accuracy of the second model can be called the training accuracy (AiT) of the second model and describes the degree of accuracy of identification or decision that the model can achieve after it has been trained. Specifically, the third accuracy can be used to indicate the degree of correctness and / or error of the model output results presented by the second model during the training or testing phase. The method for determining the third accuracy is the same as the method for determining the second accuracy of the first model in the embodiment shown in Figure 2 above, and the expression format of the third accuracy may be the same as the expression format of the second accuracy, but this will not be explained redundantly here. The training data for the second model is the training data used when training the second model, and may include input data and label data, and may specifically be the target training data obtained when retraining the first model as described above. The second model includes, but is not limited to, descriptive information and / or a model file for the second model, and the model file may include elements such as the complete network structure and parameter information that generate the second model.

[0139] In the embodiment of the present invention, the first accuracy of the first model when used for actual inference is determined, and if the first accuracy does not meet predetermined conditions, the first model can be retrained. Therefore, if the accuracy of the first model when used for actual inference decreases, the accuracy of the first model can be adjusted by retraining the first model to improve its accuracy when used for actual inference, thereby better assisting correct policy decisions or appropriate actions by devices inside and outside the network.

[0140] In possible application scenarios, the data processing method in a communication network provided in the embodiment of the present invention may be shown in Figure 4. The data processing method in a communication network shown in Figure 4 may include the following steps:

[0141] Step 1: The first network element obtains training data from the fifth network element.

[0142] Step 2: The first network element trains the first model based on the training data.

[0143] Step 3: The second network element sends a request to the first network element to retrieve the model of the first model.

[0144] Step 4: The first network element transmits the model information of the first model to the second network element.

[0145] It should be explained that the execution order of steps 1 to 4 above may be step 1, step 2, step 3, step 4, or step 3, step 1, step 2, step 4.

[0146] Step 5: The second network element receives the inference task from the eighth network element.

[0147] It should be explained that the execution order of steps 1-4 and step 5 above may be steps 1-4 followed by step 5, or step 5 followed by steps 1-4.

[0148] Step 6: The second network element obtains inference input data from the third network element and performs the inference task based on the inference input data and the first model.

[0149] Step 7: The second network element retrieves label data from the fourth network element.

[0150] Step 8: The second network element sends the inference result data to the eighth network element.

[0151] What needs to be explained is that the second network element can set up a validation test phase, in which the second network element itself can simulate the trigger of an inference task to estimate the accuracy of the model inference. Specifically, after performing step 4, the second network element can simulate the trigger of an inference task and execute the inference task based on the first model. In this case, steps 5-7 above are replaced by steps in which the second network element simulates the trigger of an inference task and executes the inference task based on the first model. Figure 3 illustrates this by showing only the second network element performing steps 5-7.

[0152] Step 9: The first network element sends a first request message to the second network element, which requests the second network element to retrieve information on the use of the first model and / or the first data collected by the second network element.

[0153] Step 10: The second network element transmits usage information and / or first data of the first model to the first network element.

[0154] Information on using the first model is: Model identifier information for the first model, Task identifier information for inference tasks performed based on the first model, Conditional information for the inference task, It includes at least one of the target information for the inference task.

[0155] The first data is, Inference input data and, Inference result data corresponding to the inference input data, It includes at least one of the inference input data and the corresponding label data.

[0156] Step 11: The first network element determines the source information of the first data based on the usage information.

[0157] The first data is, Inference input data and, Inference result data corresponding to the inference input data, It includes at least one of the inference input data and the corresponding label data.

[0158] The source information for the first data is: A third network element for providing inference input data corresponding to the inference task, It includes at least one of a fourth network element for providing label data corresponding to an inference task.

[0159] Step 12a: The first network element sends an input data acquisition request message to the third network element.

[0160] The input data acquisition request message is used to request the acquisition of inference input data. The input data acquisition request message includes: Type information of the inference input data, The target information corresponding to the inference input data, It includes at least one of the following: the inference input data and the corresponding time information.

[0161] Step 12b: The first network element sends a label data acquisition request message to the fourth network element.

[0162] A label data retrieval request message is used to request the retrieval of label data. A label data retrieval request message includes: Label data type information, The target information corresponding to the label data, It includes at least one of the following: label data or time information corresponding to the label data.

[0163] What needs to be explained is that the first network element can perform at least one of steps 12a and 12b described above.

[0164] Furthermore, it should be explained that the first network element can selectively perform steps 11, 12a, and 12b. For example, if the second network element does not transmit usage information for the first model, but only transmits the first data, and the first data includes inference input data, inference result data, and label data, the first network element does not need to perform steps 11, 12a, and 12b. If the second network element transmits both usage information for the first model and the first data, the first network element can perform steps 11 and 12a if the first data includes only label data, and can perform steps 11 and 12b if the first data includes only inference input data. The specific steps performed by the first network element can be determined according to the actual situation and are not specifically limited here, as long as it is guaranteed that the first network element can obtain the inference input data, inference result data, and label data.

[0165] Furthermore, when the first network element acquires the first data, it can acquire the first data from the seventh network element, and Figure 4 illustrates only the acquisition of usage information and / or the first data of the first model from the second network element as an example.

[0166] Step 13: The first network element determines the first accuracy of the first model.

[0167] The first accuracy is intended to indicate the degree of accuracy of the first model used in the actual reasoning, and may be a degree of correctness or incorrectness.

[0168] Step 14: The first network element determines whether the first accuracy satisfies a predetermined condition.

[0169] The first accuracy satisfies the predetermined conditions. The first accuracy is less than the second accuracy, which indicates the degree of accuracy of the model output results presented by the first model during the training or testing phase. The first accuracy is less than the second accuracy, and the difference between the first accuracy and the second accuracy is greater than a predetermined value. This includes at least one of the following: the first accuracy is less than a predetermined accuracy.

[0170] If the first accuracy satisfies the predetermined conditions, at least one of steps 15a to 15c can be performed. If the first accuracy does not satisfy the predetermined conditions, it is not necessary to perform the subsequent steps, and here, as an example, it will be explained that at least one of steps 15a to 15c is performed.

[0171] Step 15a: The first network element obtains the first training data, which will be used when training the first model.

[0172] Step 15b: The first network element obtains the second training data from the fifth network element.

[0173] Step 15c: The first network element obtains inference data from the sixth network element.

[0174] Step 16: The first network element re-selects the first model, or the first model is retrained based on the target training data to obtain the second model.

[0175] The target training data here includes data obtained in at least one of steps 15a to 15c above. That is, the target training data is The first training data used during training of the first model, The second training data obtained from the fifth network element, It includes at least one of the inference data obtained from the sixth network element.

[0176] Step 17: The first network element transmits model information of the second model to the second network element.

[0177] The model information for the second model is as follows: Model identifier information for the second model, Task identifier information for inference tasks performed based on the second model, Information on the scope of application of the second model, The third accuracy of the second model indicates the degree of accuracy of the model output results presented by the second model during the training or testing phase, Training data for the second model, Includes at least one of the second model.

[0178] Step 18: The first network element transmits model information of the second model to other second network elements.

[0179] Step 19: The first network element transmits model information of the second model to the seventh network element.

[0180] The first network element may or may not perform steps 17-19 above if it is retraining the first model to obtain the second model or re-selecting the second model, and if it is performed, it may perform at least one of steps 17-19.

[0181] The specific implementation methods for each step shown in Figure 4 can be found by referring to the specific implementation methods for the corresponding steps in Figures 2 and 3, and will not be explained again here.

[0182] In the embodiment of the present invention, the first accuracy of the first model when used for actual inference is determined, and the first model can be retrained if the first accuracy does not meet predetermined conditions. Therefore, if the accuracy of the first model when used for actual inference decreases, the accuracy of the first model can be adjusted by retraining the first model or re-selecting the second model, thereby improving the accuracy of the first model when used for actual inference. This better assists devices inside and outside the network in making correct policy decisions or appropriate actions.

[0183] The data processing method in a communication network provided in the embodiment of the present application can be executed by a data processing device in the communication network. The data processing device in a communication network provided in the embodiment of the present application will be described using the example of the data processing method in the communication network being executed by the data processing device in the communication network.

[0184] Figure 5 is a schematic diagram of the structure of a data processing device in a communication network according to an embodiment of the present invention, and this device can correspond to the first network element in other embodiments. As shown in Figure 5, the device 500 is A decision module 501 used to determine the first accuracy of the first model, wherein the first accuracy indicates the degree of accuracy of the first model used in actual inference, The system includes a model training module 502 used to retrain the first model or re-select the second model if the first accuracy satisfies predetermined conditions.

[0185] Optionally, as one embodiment, the first accuracy is: The degree of accuracy of the inference results when the first model is used in actual inference, This is to indicate at least one of the following: the degree of error in the inference result when the first model is used in actual inference.

[0186] Optionally, as one embodiment, the decision module 501 is: It is used to acquire first data and to determine the first accuracy based on the first data.

[0187] The first data mentioned above is: Inference input data and, The inference result data corresponding to the inference input data, This includes at least one of the following: the inference input data and the corresponding label data.

[0188] Optionally, as one embodiment, the decision module 501 is: Receiving usage information of the first model transmitted by the second network element, The source information of the first data is determined according to the usage information, and This is used to obtain the first data according to the source information.

[0189] Optionally, as one embodiment, the decision module 501 is: Used to receive the first data transmitted by a second network element, the second network element includes a model inference function network element.

[0190] Optionally, as one embodiment, the decision module 501 is: Used to receive the first data transmitted by the seventh network element, the seventh network element includes a data storage function network element.

[0191] Optionally, as one embodiment, the decision module 501 further: It is used to send a first request message to the second network element, and the first request message is intended to request the second network element to obtain usage information of the first model.

[0192] Optionally, as one embodiment, the decision module 501 further: The aforementioned It is used to send a second request message to a second network element, which requests that the second request message retrieve the first data collected by the second network element.

[0193] Optionally, as one embodiment, the decision module 501 further: This is used to send a third request message to the seventh network element, and the third request message acquires the first data. to request It is for that purpose.

[0194] Optionally, as one embodiment, the model training module 502 further: This is used to train and acquire the first model, and to transmit the model information of the first model to the second network element. The model information for the 1 model of the 1st model is: The first model and, The system includes at least one of the following: a second accuracy of the first model, which indicates the degree of accuracy of the model output results presented by the first model during the training or testing phase.

[0195] Optionally, as one embodiment, the usage information is: The model identifier information of the previous model, Task identifier information for an inference task performed based on the first model described above, The conditional information for the aforementioned inference task, It includes at least one of the target information for the aforementioned inference task.

[0196] Optionally, as one embodiment, the source information is: A third network element for providing inference input data corresponding to the aforementioned inference task, It includes at least one of the following: a fourth network element for providing label data corresponding to the aforementioned inference task.

[0197] Optionally, as one embodiment, the decision module 501 is: Sending an input data acquisition request message to the third network element, wherein the input data acquisition request message is for requesting the acquisition of the inference input data, It is used to send a label data acquisition request message to the fourth network element, wherein the label data acquisition request message is for requesting the acquisition of the label data, The aforementioned input data acquisition request message includes: The type information of the aforementioned inference input data, The target information corresponding to the aforementioned inference input data, This includes at least one of the following: the inference input data and time information corresponding to the inference input data, The aforementioned label data acquisition request message includes: The type information of the aforementioned label data, The target information corresponding to the aforementioned label data, This includes at least one of the following: the label data and time information corresponding to it.

[0198] Optionally, as one embodiment, the decision module 501 is: The inference input data is input to the first model to determine the inference result data, and, This is used to determine the first accuracy based on the inference result data and the label data.

[0199] Optionally, as one embodiment, the first accuracy satisfying a predetermined condition is The first accuracy is less than the second accuracy, which indicates the degree of accuracy of the model output results presented by the first model during the training or testing phase. The first accuracy is smaller than the second accuracy, and the difference between the first accuracy and the second accuracy is greater than a predetermined value. This includes at least one of the following: the first accuracy is less than a predetermined accuracy.

[0200] Optionally, as one embodiment, the second accuracy is: The degree of accuracy of the model output results presented by the first model during the training or testing phase, This is for indicating at least one of the following: the degree of error of the model output results presented by the first model during the training or testing phase.

[0201] Optionally, as one embodiment, the model training module 502 is: To acquire target training data, wherein the target training data includes target input data and target label data corresponding to the target input data, and This is used to retrain the first model based on the aforementioned target training data.

[0202] Optionally, as one embodiment, the model training module 502 is: To acquire the first training data used during training of the first model, The process involves determining a fifth network element and obtaining second training data from the fifth network element, wherein the fifth network element is for providing training data. This is used to determine a sixth network element and to obtain inference data from the sixth network element, wherein the sixth network element is for providing inference data.

[0203] Optionally, as one embodiment, the model training module 502 is: Used to determine the fifth network element according to the second information, The second information includes task identifier information and / or conditional information for an inference task performed based on the first model.

[0204] Optionally, as one embodiment, the model training module 502 is: This is used to determine the sixth network element according to the second information.

[0205] Optionally, as one embodiment, the model training module 502 further: The model information of the retrained second model or the re-selected second model is transmitted to the second network element, and the second network element performs an inference task based on the second model. It is used for at least one of the following: transmitting model information of a second model obtained by retraining or a re-selected second model to a seventh network element, and storing the model information of the second model by the seventh network element. The model information for the two additional models is as follows: The model identifier information of the previous 2nd model, Task identifier information for the inference task executed based on the second model described above, The scope information of the second model mentioned above, A third accuracy of the second model, which indicates the degree of accuracy of the model output results presented by the second model during the training or testing phase, The training data for the second model described above, It includes at least one of the above second models.

[0206] Optionally, as one embodiment, the third accuracy is: The degree of accuracy of the model output results presented by the second model during the training or testing phase, This is for indicating at least one of the following: the degree of error of the model output results presented by the second model during the training or testing phase.

[0207] Optionally, as one embodiment, the first network element includes a model training function network element. The second network element includes a model inference function network element. The third network element includes the source device for the inference input data. The fourth network element includes the source device for the label data. The fifth network element includes the source equipment for the training data. The sixth network element includes the source device for inference data, The seventh network element includes a data storage function network element.

[0208] Apparatus 500 according to the embodiment of the present application can refer to the flow of Method 200 of the corresponding embodiment of the present application, and each unit / module and other operation and / or function in apparatus 500 is for realizing the corresponding flow in Method 200, and can achieve similar or equivalent technical effects, which are not described again here for the sake of brevity.

[0209] Figure 6 is a schematic diagram of the structure of a data processing device in a communication network according to an embodiment of the present invention, and this device can correspond to a second network element in other embodiments. As shown in Figure 6, the device 600 is A task execution module 601 used to perform an inference task based on a first model, wherein the first model is trained by a first network element, and the task execution module 601 includes a model training function network element. A transmission module 602 used for transmitting at least one of the usage information of the first model and the first data to the first network element and / or for transmitting first instruction information to the seventh network element, where the first instruction information is for instructing the seventh network element to store the first data of the inference task, and the seventh network element includes a data storage function network element, and the transmission module 602 is provided.

[0210] Optionally, as an example, the first data inference input data, inference result data corresponding to the inference input data, and at least one of label data corresponding to the inference input data.

[0211] Optionally, as an example, the apparatus 600 further includes a first receiving module 603 used for receiving a first request message transmitted by the first network element, where the first request message is for requesting to obtain the usage information of the first model by the second network element.

[0212] Optionally, as an example, the apparatus 600 further includes a second receiving module 604 used for receiving a second request message transmitted by the first network element, where the second request message is for requesting to obtain the first data collected by the second network element.

[0213] Optionally, as an example, the apparatus 600 The aforementioned collecting inference input data, The aforementioned and further includes a collecting module 605 used for at least one of collecting label data corresponding to the inference input data.

[0214] Optionally, as one embodiment, the collection module 605 may further be: When a task request for the inference task is received by the eighth network element, the inference input data is collected, wherein the eighth network element includes a consumer network element. The aforementioned It is used for at least one of the following: actively collecting inference input data.

[0215] Optionally, as one embodiment, the collection module 605 may further be: 4th It is used to send a data acquisition request to a network element, which requests that the network element acquire label data corresponding to the inference input data.

[0216] Optionally, as one embodiment, the task execution module 601 further: Used to receive model information of a second model transmitted by the first network element, wherein the second model is obtained by retraining the first model by the first network element or the second model is a model re-selected by the first network element. The model information for the two additional models is as follows: The model identifier information of the previous 2nd model, Task identifier information for the inference task executed based on the second model described above, The scope information of the second model mentioned above, A third accuracy of the second model, which indicates the degree of accuracy of the model output results presented by the second model during the training or testing phase, The training data for the second model described above, It includes at least one of the above second models.

[0217] Optionally, as one embodiment, the third accuracy is: The degree of accuracy of the model output results presented by the second model during the training or testing phase, This is for indicating at least one of the following: the degree of error of the model output results presented by the second model during the training or testing phase.

[0218] Optionally, as one embodiment, the task execution module 601 further: Sending a model request message to the first network element, wherein the model request message is for requesting the acquisition of the first model, and Used to receive model information of the first model transmitted by the first network element, The model information for the 1 model of the 1st model is: The first model and, The system includes at least one of the following: a second accuracy of the first model, which indicates the degree of accuracy of the model output results presented by the first model during the training or testing phase.

[0219] Optionally, as one embodiment, the second accuracy is: The degree of accuracy of the model output results presented by the first model during the training or testing phase, This is for indicating at least one of the following: the degree of error of the model output results presented by the first model during the training or testing phase.

[0220] Optionally, as one embodiment, the task execution module 601 is: Receiving a task request for the inference task transmitted by the eighth network element, and executing the inference task based on the first model, wherein the eighth network element includes a consumer network element, It is used for at least one of obtaining the task requirements of the inference task simulated and triggered by the second network element and executing the inference task based on the first model.

[0221] Optionally, as an example, the task execution module 601 is to send an input data acquisition request message to a third network element, where the input data acquisition request message is for requesting to acquire inference input data corresponding to the inference task, receiving the inference input data sent by the third network element, and is used for inputting the inference input data into the first model to obtain inference result data.

[0222] Optionally, as an example, the task execution module 601 further is used for sending the inference result data to the eighth network element.

[0223] Optionally, as an example, the usage information includes at least one of the model identifier information of the first model, the task identifier information of the inference task executed based on the first model, the conditional limitation information of the inference task, and the target information of the inference task.

[0224] The device 600 according to the embodiments of the present application can refer to the flow of the method 300 of the corresponding embodiments of the present application, and each unit / module in the device 600 and the above other operations and / or functions are respectively for realizing the corresponding flow in the method 300, and can achieve the same or equivalent technical effects. For the sake of brevity, they are not described repeatedly here.

[0225] The data processing device in the communication network in the embodiments of this application may be, for example, an electronic device having an operating system, or it may be a component in an electronic device such as an integrated circuit or a chip. The electronic device may be a terminal, or it may be other devices other than terminals. Exemplaryly, a terminal may include, but is not limited to, the types of terminals 11 listed above. Other devices may be servers, network attached storage (NAS), etc., but are not specifically limited in the embodiments of this application.

[0226] The data processing device in the communication network provided in the embodiment of the present application can realize each step realized in the embodiment of the method shown in Figures 2 to 4 and achieve similar technical effects, and to avoid repetition, it is omitted here.

[0227] Optionally, as shown in Figure 7, the embodiment of the present application further provides a communication device 700 comprising a processor 701 and a memory 702, wherein the memory 702 stores a program or command executable on the processor 701, and when the communication device 700 is a network-side device, the execution of the program or command by the processor 701 realizes each step of the embodiment of the data processing method in the communication network described above and achieves similar technical effects, which are omitted here to avoid repetition.

[0228] Embodiments of the present application further provide a network-side device comprising a processor and a communication interface. The processor is used to determine a first accuracy of a first model, the first accuracy being used to indicate the degree of accuracy of the first model to be used for actual inference, and, if the first accuracy satisfies predetermined conditions, to retrain the first model or reselect a second model. Alternatively, the processor is used to perform an inference task based on the first model, the first model being trained by a first network element, the first network element including a model training function network element. The communication interface is used to transmit at least one of the usage information of the first model and first data to the first network element, and / or to transmit first instruction information to a seventh network element, the first instruction information being used to instruct the seventh network element to store first data for the inference task, the seventh network element including a data storage function network element. Embodiments of the network-side device correspond to embodiments of the method for network-side devices described above, and each implementation step and implementation of embodiments of the method described above is applicable to embodiments of the network-side device and can achieve similar technical effects.

[0229] Specifically, the embodiment of the present invention further provides network-side equipment. As shown in Figure 8, the network-side equipment 800 comprises a processor 801, a network interface 802, and memory 803. The network interface 802 is, for example, a common public radio interface (CPRI).

[0230] Specifically, the network-side device 800 of the embodiment of the present invention further comprises commands or programs stored in memory 803 and executable on processor 801, the processor 801 calling the commands or programs in memory 803 and performing the methods executed by each module shown in Figure 5 or Figure 6, achieving similar technical effects and being omitted here to avoid repetition.

[0231] The embodiment of the present application further provides a readable storage medium in which a program or command is stored, and when the program or command is executed by a processor, each step of the embodiment of the data processing method in the communication network described above is realized and similar technical effects are achieved, and to avoid repetition, it is omitted here.

[0232] The processor is the processor in the terminal described in the above embodiment. The readable storage medium includes computer-readable storage media such as computer read-only memory ROM, random access memory RAM, magnetic disk, or optical disk.

[0233] Embodiments of the present invention further provide a chip comprising a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor executes a program or command to realize each step of the embodiment of the data processing method in the communication network described above, and can achieve similar technical effects, which are omitted here to avoid repetition.

[0234] The chip described in the embodiments of this application may also be called a system-on-a-chip, system chip, chip system, or SoC, etc.

[0235] Embodiments of the present application further provide a computer program / program product which is stored in a storage medium and executed by at least one processor to realize each step of the embodiment of the data processing method in the communication network described above and achieve similar technical effects, and are omitted here to avoid repetition.

[0236] Embodiments of the present invention further provide a data processing system in a communication network comprising a first network-side device and a second network-side device. The first network-side device is usable to perform the steps of the data processing method in the communication network described in Figure 2 above. 2nd Network-side equipment can be used to perform the steps of the data processing method in the communication network described in Figure 3 above.

[0237] It should be noted that in this specification, the terms “including,” “consisting of,” or any other variation thereof are intended to include non-exclusive inclusion, so that a process, method, article, or apparatus containing a set of elements includes not only those elements but also other elements not explicitly stated, or elements specific to such process, method, article, or apparatus. Unless otherwise specified, an element limited by the phrase “including one…” does not preclude the existence of other identical elements in a process, method, article, or apparatus containing that element. Furthermore, the scope of the methods and apparatus in embodiments of this application is not limited to performing functions in the order shown or discussed herein, but may further include performing functions almost simultaneously or in the opposite order, depending on the relevant function. For example, the above methods may be performed in an order different from the order described, and further, each step may be added, omitted, or combined. Also, features described with reference to some examples may be combined with other examples.

[0238] From the above description of the embodiments, it will be clear to those skilled in the art that the methods of the above embodiments can be implemented in the form of a combination of software and a necessary common hardware platform, although they may, of course, be implemented in hardware, but in many cases the former is a more preferred embodiment. Based on this view, the technical solutions of the present application can be implemented substantially or in part in the form of a computer software product, which is stored in a storage medium (e.g., ROM / RAM, magnetic disk, optical disk) and includes a plurality of commands that cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of each embodiment of the present application.

[0239] Although embodiments of the present application have been described above with reference to the drawings, the present application is not limited to the above-described specific embodiments. The above-described specific embodiments are merely illustrative and not limiting. Many forms that a person skilled in the art could make based on the suggestions of the present application without departing from the spirit of the present application and the scope of protection of the claims are all within the scope of protection of the present application.

[0240] [Cross-reference of related applications] This invention claims priority to the Chinese patent applications filed with the Chinese National Intellectual Property Office on March 7, 2022, with application number 202210224517.3, titled "Data Processing Method and Network-Side Equipment in a Communication Network," and filed with the Chinese National Intellectual Property Office on August 9, 2022, with application number 202210950629.7, titled "Data Processing Method and Network-Side Equipment in a Communication Network," and all contents of these applications are incorporated into this invention by reference.

Claims

1. A step in which a first network element determines a first accuracy of a first model, wherein the first accuracy indicates the degree of accuracy of the first model used for actual inference, The step includes, if the first accuracy satisfies a predetermined condition, the first network element retrains the first model, The step in which the first network element determines the first accuracy of the first model is: The steps include: the first network element acquiring first data; and the first network element determining the first accuracy based on the first data. The first data is, Inference input data and, The inference result data corresponding to the inference input data, Includes label data corresponding to the aforementioned inference input data, The step in which the first network element acquires the first data is: A data processing method in a communication network, comprising the step of receiving the first data transmitted by a second network element, wherein the second network element includes a model inference function network element.

2. The data processing method according to claim 1, wherein the label data is the actual value of the data corresponding to the inference input data.

3. Before the step of receiving the first data transmitted by the second network element, The data processing method according to claim 1, further comprising the step of the first network element sending a second request message to the second network element, wherein the second request message is for requesting the second network element to retrieve the first data collected by the second network element.

4. The step in which the first network element determines the first accuracy of the first model is: If the first data includes the inference input data, the steps include inputting the inference input data into the first model and determining the inference result data, The data processing method according to claim 1, comprising the step of determining the first accuracy according to the inference result data and the label data.

5. The first accuracy satisfies the predetermined conditions. The data processing method according to claim 1, wherein the first accuracy is less than a predetermined accuracy.

6. After the first network element retrains the first model, The process further includes the steps of transmitting model information of the retrained second model to a second network element, and having the second network element perform an inference task based on the second model, The model information for the two models mentioned above is: The third accuracy of the second model is used to indicate the degree of accuracy of the model output results of the second model during the training phase, The data processing method according to claim 1, including the second model.

7. A step in which a second network element performs an inference task based on a first model, wherein the first model is trained by the first network element, the first network element includes a model training function network element, and the second network element includes a model inference function network element. The process includes the steps of transmitting usage information of the first model and first data to the first network element, or transmitting first data to the first network element, The first data is, Inference input data and, The inference result data corresponding to the inference input data, Includes label data corresponding to the aforementioned inference input data, The first data is used by the first network element to determine the first accuracy of the first model. A data processing method in a communication network.

8. Before the step of transmitting the first data to the first network element, The data processing method according to claim 7, further comprising the step of the second network element receiving a second request message transmitted by the first network element, wherein the second request message is for requesting the second network element to obtain the first data collected by the second network element.

9. The step of receiving model information of a second model transmitted by the first network element, further comprising the step that the second model is a model obtained by retraining the first model by the first network element, The model information for the two models mentioned above is: The third accuracy of the second model is used to indicate the degree of accuracy of the model output results of the second model during the training phase, The data processing method according to claim 7, including the second model.

10. A network-side device comprising a processor and a memory storing a program or command executable on the processor, wherein when the program or command is executed by the processor, the steps of the data processing method described in any one of claims 1 to 6 are realized.

11. A network-side device comprising a processor and a memory storing a program or command executable on the processor, wherein when the program or command is executed by the processor, the steps of the data processing method described in any one of claims 7 to 9 are realized.

12. A readable storage medium in which a program or command is stored, and when the program or command is executed by a processor, the steps of the data processing method according to any one of claims 1 to 6 are realized.

13. A readable storage medium on which a program or command is stored, and when the program or command is executed by a processor, the steps of the data processing method according to any one of claims 7 to 9 are realized.

Images