Communication method, and apparatus

Abstract

Disclosed in the present application are a communication method and an apparatus, which are applied to the technical field of communications. The method comprises: a management service producer acquires first data of a first model, the first data being data acquired from an actual network; on the basis of the first data, the management service producer determines a data distribution difference of the first model; and, when the data distribution difference satisfies a first condition, the management service producer sends to a management service consumer a first message, the first message comprising first instruction information, and the first instruction information being used for instructing the management service consumer to update the first model. Using the present application can proactively sense model degradation and improve sensing capability, so that the first model can be updated without frequently transmitting a large amount of data, thereby reducing model update costs and improving model performance.

Description

A communication method and apparatus

[0001] This application claims priority to Chinese Patent Application No. 202510486847.3, filed on April 17, 2025, with the Chinese National Intellectual Property Administration, entitled “A Communication Method and Apparatus”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of communication technology, and in particular to a communication method and apparatus. Background Technology

[0003] With the continuous development of artificial intelligence (AI) and machine learning (ML) technologies, these technologies can be applied to the telecommunications industry. For example, they can be used for the management and orchestration of 5G systems and 5G networks. However, in practical applications, because AI / ML model training is typically based on laboratory-constructed data, this data cannot fully reflect the complexity of real-world network environments. Therefore, changes in network conditions can lead to model degradation or render the model unsuitable for the changed conditions. Furthermore, model degradation is usually detected passively, only becoming apparent after network performance declines or customer feedback, resulting in poor perceptual ability. Summary of the Invention

[0004] This application provides a communication method and apparatus that can proactively detect model degradation, enhance perception capabilities, and improve model performance.

[0005] In a first aspect, embodiments of this application provide a communication method applied to a management service producer, the method comprising:

[0006] Obtain first data for the first model, wherein the first data is data collected from the actual network; determine the data distribution difference of the first model based on the first data; when the data distribution difference meets a first condition, send a first message to the management service consumer, wherein the first message includes first indication information, and the first indication information is used to instruct the management service consumer to update the first model.

[0007] By acquiring the first data, the management service producer can perform a difference analysis between the data in the actual network and the original data used to train the first model, or the management service producer can perform a difference analysis between the data in the actual network and the inference results of the first model, thereby determining the data distribution difference of the first model. By judging whether the data distribution difference meets the first condition, the management service producer can proactively perceive whether the first model has deteriorated, enhancing the perception capability and overcoming the limitations of passively perceiving model deterioration. By sending the first message, the management service consumer can update the first model without frequently transmitting large amounts of data, which can reduce the cost of model updates and improve model performance.

[0008] In one possible design, the data distribution difference includes the distribution difference between the first data and the second data of the first model, where the second data is part or all of the original data used to train the first model; the value of a distribution metric between the first data and the second data is determined; when the value of the distribution metric is greater than a first threshold, the data distribution difference is determined to satisfy the first condition. By analyzing the distribution difference between the data in the actual network and the original data used to train the first model, management service producers can proactively perceive whether the first model has deteriorated, enhancing their perception capabilities and overcoming the limitations of passively perceiving model deterioration.

[0009] In another possible design, the first message further includes third data of the first model. This third data is either a first number of data points in the first data that have the largest distribution metric value between themselves and the second data, or data points in the first data that have a distribution metric value greater than a second threshold between themselves and the second data. This allows management service consumers to subsequently update the first model based on the third data point, which has the largest distribution difference between themselves and the second data, without requiring frequent large-scale data transfers. This reduces the cost of model updates and improves model performance.

[0010] In another possible design, the data distribution difference includes the distribution difference between degraded data and sampled data. The sampled data is part or all of the first data, and the degraded data is the sampled data whose accuracy after inference by the first model is lower than a third threshold. The ratio of the number of degraded data to the number of sampled data is determined. When the ratio is greater than a fourth threshold, the data distribution difference is determined to satisfy the first condition. By analyzing the distribution difference between degraded data and sampled data, management service producers can proactively perceive whether the first model has degraded, enhancing their perception capabilities and overcoming the limitations of passively perceiving model degrade.

[0011] In another possible design, the first message further includes third data of the first model. This third data is either the second number of data points in the first data that have the smallest distribution metric value compared to the degraded data, or the third data is data points in the first data whose distribution metric value compared to the degraded data is less than a fifth threshold. This allows management service consumers to subsequently update the first model based on the third data point in the first data that has the smallest distribution difference from the degraded data, eliminating the need for frequent large-scale data transmissions. This reduces the cost of model updates and improves model performance.

[0012] In another possible design, a first request is received from the management service consumer. This first request requests the management service producer to update the first model. The first request includes first information, which is determined by the management service consumer based on the third data. The first model is then updated based on this first information. By receiving the first request, the management service producer can update the first model based on the first information in the first request, eliminating the need for frequent large-scale data transmission. This helps reduce the cost of model updates and improves model performance.

[0013] In another possible design, the first information includes the update method of the first model, which includes remote update or local update. The remote update is used to instruct the management service consumer to perform the model update, and the local update is used to instruct the management service producer to perform the model update. This allows the management service producer to update the first model using either remote or local update methods, thereby improving model performance.

[0014] In another possible design, when the update method includes remote update, the first information also includes the update weight of the first model. This allows management service producers to update the first model based on the update weight in the remote update method, without frequent large-scale data transmission, thus reducing the cost of model updates and improving model performance.

[0015] In another possible design, when the update method includes local updates, the first information further includes at least one of the following: fourth data of the first model, a first time range, or a first algorithm; the fourth data is the data in the third data used to update the first model; the first time range is the time range for updating the first model; and the first algorithm is used to determine the degree of degradation of the first model. This allows management service producers to update the first model based on the fourth data, the first time range, and / or the first algorithm in a local update manner, without frequent large-scale data transfers, thus reducing the cost of model updates and improving model performance.

[0016] In another possible design, a second request is received from the management service consumer, requesting the management service producer to deploy the first model. Receiving the second request allows the management service producer to deploy the first model, which facilitates resource management within the communication network, thereby improving network reliability and stability.

[0017] In another possible design, the second request includes information about the first model, which includes the collection strategy for the third data. This allows management service producers to collect the third data according to the collection strategy, enabling management service consumers to subsequently update the first model based on the third data without frequent large-scale data transfers. This reduces the cost of model updates and improves model performance.

[0018] In another possible design, the information of the first model further includes at least one of the following: the second data, the second algorithm, the first threshold, the second threshold, the third threshold, the fourth threshold, or the fifth threshold; the second algorithm is used to determine the value of the distribution metric among different data. This allows management service producers to proactively detect whether the first model has deteriorated and collect third data, enabling management service consumers to subsequently update the first model based on the third data without frequent large-scale data transmission, thus reducing the cost of model updates and improving model performance.

[0019] In another possible design, the first message also includes identification information for the first model. This allows management service consumers to distinguish each model that needs updating and the corresponding information and / or data for each model.

[0020] Secondly, embodiments of this application provide a communication method applied to managing service consumers, the method comprising:

[0021] When the data distribution difference of the first model meets the first condition, a first message is received from the management service producer. The first message includes first indication information, which is used to instruct the management service consumer to update the first model. The data distribution difference is determined by the management service producer based on the first data of the first model. The first data is the data collected in the actual network.

[0022] By determining whether the data distribution difference of the first model meets the first condition, the management service producer can proactively perceive whether the first model has deteriorated, enhance its perception capability, and overcome the limitations of passively perceiving model deterioration. By receiving the first message, the management service consumer can update the first model without frequently transmitting large amounts of data, which can reduce the cost of model updates and improve model performance.

[0023] In one possible design, the data distribution difference includes the distribution difference between the first data and the second data of the first model, where the second data is part or all of the original data used to train the first model. This allows management service producers to proactively detect whether the first model has deteriorated, enhancing their perception capabilities and overcoming the limitations of passively detecting model deterioration.

[0024] In another possible design, the first message further includes third data of the first model. This third data is either a first number of data points in the first data that have the largest distribution metric value between themselves and the second data, or data points in the first data that have a distribution metric value greater than a second threshold between themselves and the second data. This allows management service consumers to subsequently update the first model based on the third data point, which has the largest distribution difference between themselves and the second data, without requiring frequent large-scale data transfers. This reduces the cost of model updates and improves model performance.

[0025] In another possible design, the data distribution difference includes the distribution difference between degraded data and sampled data, where the sampled data is part or all of the first data, and the degraded data is the sampled data whose accuracy after inference by the first model is lower than a third threshold. This allows management service producers to proactively perceive whether the first model has degraded, enhancing their perception capabilities and overcoming the limitations of passively perceiving model degradation.

[0026] In another possible design, the first message further includes third data of the first model. This third data is either the second number of data points in the first data that have the smallest distribution metric value compared to the degraded data, or the third data is data points in the first data whose distribution metric value compared to the degraded data is less than a fifth threshold. This allows management service consumers to subsequently update the first model based on the third data point in the first data that has the smallest distribution difference from the degraded data, eliminating the need for frequent large-scale data transmissions. This reduces the cost of model updates and improves model performance.

[0027] In another possible design, based on the third data, first information is determined; a first request is sent to the management service producer, the first request requesting the management service producer to update the first model, the first request including the first information, the first information being used by the management service producer to update the first model. By sending the first request, the management service producer can update the first model based on the first information in the first request, without frequently transmitting large amounts of data, thereby reducing the cost of model updates and improving model performance.

[0028] In another possible design, the first information includes the update method of the first model, which includes remote update or local update. The remote update is used to instruct the management service consumer to perform the model update, and the local update is used to instruct the management service producer to perform the model update. This allows the management service producer to update the first model using either remote or local update methods, thereby improving model performance.

[0029] In another possible design, when the update method includes remote update, the first information also includes the update weight of the first model. This allows management service producers to update the first model based on the update weight in the remote update method, without frequent large-scale data transmission, thus reducing the cost of model updates and improving model performance.

[0030] In another possible design, when the update method includes local updates, the first information further includes at least one of the following: fourth data of the first model, a first time range, or a first algorithm, wherein the fourth data is data from the third data used to update the first model; the first time range is the time range for updating the first model; and the first algorithm is used to determine the degree of degradation of the first model. This allows management service producers to update the first model based on the fourth data, the first time range, and / or the first algorithm in a local update manner, without frequent large-scale data transfers, thus reducing the cost of model updates and improving model performance.

[0031] In another possible design, a second request is sent to the management service producer, requesting the producer to deploy the first model. Sending the second request allows the management service producer to deploy the first model, thereby enabling it to manage resources in the communication network and improve network reliability and stability.

[0032] In another possible design, the second request includes information about the first model, which includes the collection strategy for the third data. This allows management service producers to collect the third data according to the collection strategy, enabling management service consumers to subsequently update the first model based on the third data without frequent large-scale data transfers. This reduces the cost of model updates and improves model performance.

[0033] In another possible design, the information of the first model further includes at least one of the following: the second data, the second algorithm, the first threshold, the second threshold, the third threshold, the fourth threshold, or the fifth threshold; the second algorithm is used to determine the value of the distribution metric among different data. This allows management service producers to proactively detect whether the first model has deteriorated and collect third data, enabling management service consumers to subsequently update the first model based on the third data without frequent large-scale data transmission, thus reducing the cost of model updates and improving model performance.

[0034] In another possible design, the first message also includes identification information for the first model. This allows management service consumers to distinguish each model that needs updating and the corresponding information and / or data for each model.

[0035] Thirdly, embodiments of this application provide a communication device configured to implement the methods and functions described in the first aspect. This communication device is implemented in hardware / software. It includes modules corresponding to the aforementioned functions. This communication device can be the aforementioned management service producer.

[0036] Fourthly, embodiments of this application provide a communication device configured to implement the methods and functions described in the second aspect. The communication device is implemented in hardware / software. It includes modules corresponding to the aforementioned functions. This communication device can serve as a management service consumer.

[0037] Fifthly, embodiments of this application provide a communication device, which includes one or more processors. The one or more processors enable the communication device to implement the methods in any possible design or implementation of the first aspect described above. The communication device can be a management service producer, a chip, chip system, or processor that supports the management service producer in implementing the above methods, or a logical node, logical module, or software capable of implementing all or part of the management service producer's functions.

[0038] In one possible design, the communication device may further include an interface circuit, wherein the processor is used to communicate with other devices or components through the interface circuit.

[0039] In one possible design, the communication device may further include a memory. The memory stores part or all of the computer program or instructions necessary to implement the functions described in the first aspect above. The one or more processors can execute the computer program or instructions, which, when executed, cause the communication device to implement the methods in any possible design or implementation of the first aspect above.

[0040] Sixthly, embodiments of this application provide a communication device, which includes one or more processors. The one or more processors enable the communication device to implement the methods in any possible design or implementation of the second aspect described above. The communication device can be a management service consumer, a chip, chip system, or processor that supports the management service consumer in implementing the above methods, or a logical node, logical module, or software capable of implementing all or part of the management service consumer functions.

[0041] In one possible design, the communication device may further include an interface circuit, wherein the processor is used to communicate with other devices or components through the interface circuit.

[0042] In one possible design, the communication device may further include a memory. The memory stores part or all of the computer program or instructions necessary to implement the functions described in the second aspect above. The one or more processors can execute the computer program or instructions, which, when executed, cause the communication device to implement the methods in any possible design or implementation of the second aspect above.

[0043] In a seventh aspect, embodiments of this application provide a communication system including at least one management service producer and at least one management service consumer. The management service producer is used to perform the method in the first aspect described above, and the management service consumer is used to perform the method in the second aspect described above.

[0044] Eighthly, embodiments of this application provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the methods described above.

[0045] Ninthly, embodiments of this application provide a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods described above.

[0046] In a tenth aspect, embodiments of this application provide a chip including a processor and a communication interface for communicating with external or internal devices, the processor enabling the chip to implement the methods described in the above aspects.

[0047] In one possible design, the chip may further include a memory storing computer programs or instructions, which the processor executes, either from the stored computer programs or instructions or derived from other programs or instructions. When the computer program or instructions are executed, the processor causes the chip to implement the methods described above.

[0048] In another possible design, the chip can be integrated into either the management service producer or the management service consumer. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the background art, the accompanying drawings used in the embodiments of this application or the background art will be described below.

[0050] Figure 1 is a schematic diagram of a network architecture applicable to the communication method of this application embodiment;

[0051] Figure 2 is a schematic diagram of the model's operation process;

[0052] Figure 3 is a flowchart illustrating a model update method;

[0053] Figure 4 is a schematic diagram of the structure of a model update system;

[0054] Figure 5 is a schematic diagram of data distribution;

[0055] Figure 6 is a flowchart illustrating a communication method provided in an embodiment of this application;

[0056] Figure 7 is a flowchart illustrating another communication method provided in an embodiment of this application;

[0057] Figure 8 is a flowchart illustrating a collection strategy provided in an embodiment of this application;

[0058] Figure 9 is a flowchart illustrating another collection strategy provided in an embodiment of this application;

[0059] Figure 10 is a schematic diagram of an application scenario for model updating provided in an embodiment of this application;

[0060] Figure 11 is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0061] Figure 12 is a schematic diagram of another communication device provided in an embodiment of this application;

[0062] Figure 13 is a schematic diagram of the structure of a communication device provided in an embodiment of this application. Detailed Implementation

[0063] The following explanations of some of the terms used in this application are provided to facilitate understanding by those skilled in the art.

[0064] (1) Model: In the field of artificial intelligence (AI), a model refers to a mathematical structure trained by an algorithm that can be used for prediction, classification, or decision-making. Examples include neural network models, long short-term memory (LSTM) models, and Transformer models.

[0065] (2) Data distribution: refers to the arrangement of data within a certain range, reflecting the frequency or probability of data values. Common distribution types include normal distribution, uniform distribution, and exponential distribution. Data distribution can be described using statistical indicators such as mean, median, and variance.

[0066] The embodiments of this application are described below with reference to the accompanying drawings.

[0067] It should be understood that in the description of this application, "at least one" means one or more, and "multiple" means two or more. In addition, the words "first," "second," etc., unless otherwise stated, are used only for the purpose of distinguishing descriptions and should not be construed as indicating or implying relative importance or order.

[0068] It should be understood that in the description of this application, the indication includes direct indication (also known as explicit indication) and implicit indication. Direct indication information A refers to information A being included; implicit indication information A refers to information A being indicated through the correspondence between information A and information B, and the direct indication information B. The correspondence between information A and information B can be predefined, pre-stored, pre-burned, or pre-configured.

[0069] It should be understood that, in the description of this application, information C is used to determine information D, including both situations where information D is determined solely based on information C and situations where it is determined based on information C and other information. Furthermore, information C can also be used to determine information D indirectly, for example, where information D is determined based on information E, and information E is determined based on information C.

[0070] Furthermore, in this application, "device A sends information A to device B" can be understood as device B being the destination of information A or an intermediate device in the transmission path between the destination and device B, which may include sending information directly or indirectly to device B. Similarly, "device B receives information A from device A" can be understood as device A being the source of information A or an intermediate device in the transmission path between the source and device A, which may include receiving information directly or indirectly from device A. Information may undergo necessary processing, such as format changes, between the source and destination, but the destination can understand the valid information from the source. Similar expressions in this application can be interpreted in a similar way and will not be elaborated further here.

[0071] The technical solutions provided in this application can be applied to various communication systems, such as 5G or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area network (WLAN) systems, satellite communication systems, future communication systems, or integrated systems of multiple systems. The technical solutions provided in this application can also be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems. Furthermore, they can be applied to future communication systems or integrated systems of multiple systems.

[0072] In this communication system, one network element can send signals to or receive signals from another network element. Signals may include information, signaling, or data; a network element can also be replaced by an entity, network entity, device, communication device, communication module, node, or communication node, etc. This application embodiment uses a network element as an example for description.

[0073] For example, as shown in FIG1, FIG1 is a schematic diagram of a network architecture applicable to the communication method of the present application embodiment. Specifically, the network architecture may include at least one management service consumer (MnS Consumer), such as the management service consumer 1 shown in FIG1; the network architecture may also include at least one management service producer (MnS Producer), such as the management service producer 1 shown in FIG1. ​​The management service consumer 1 and the management service producer 1 are communicatively connected, and this communication connection is used for signal transmission.

[0074] Among them, management service producer 1 can provide large model management services, including but not limited to large model instance registration service, large model instance discovery service, large model instance update service, large model instance deregistration service, and large model instance subscription service. Through these services, large model instances can be managed in a unified manner.

[0075] Management service consumer 1 can invoke the large model instance discovery service and large model instance subscription service provided by management service producer 1. For example, management service consumer 1 can be a user of large model instances that invokes the large model instance discovery service and large model instance subscription service, that is, a consumer of large model instances or a large model service consumer (large model MnS consumer).

[0076] It should be noted that the five names—Large Model Instance Registration Service, Large Model Instance Discovery Service, Large Model Instance Update Service, Large Model Instance Deregistration Service, and Large Model Instance Subscription Service—are for illustrative purposes only and do not constitute a limitation on the embodiments of this application. With the development of communication technology, AI technology, and ML technology, these five services may adopt other names. For example, Large Model Instance Registration Service may also be called Large Model Registration Service, Model Registration Service, etc.; Large Model Instance Discovery Service may also be called Large Model Discovery Service, Model Discovery Service, etc.; Large Model Instance Update Service may also be called Large Model Update Service, Model Update Service, etc.; Large Model Instance Deregistration Service may also be called Large Model Deregistration Service, Model Deregistration Service, etc.; and Large Model Instance Subscription Service may also be called Large Model Subscription Service, Model Subscription Service, etc.

[0077] Optionally, the management service producer 1 can be a management function, entity, management entity, network equipment (NE), network element, etc., and this application does not limit it.

[0078] Optionally, the management service producer 1 can be configured in a network management system (NMS), an element management system (EMS), or an NE, and this application does not limit this.

[0079] For example, the management service provider 1 can be an operation and maintenance center (OMC), or it can be a network management unit provided by the device.

[0080] Optionally, the management service consumer 1 (such as the user of a large model instance) can be a network function (NF), entity, management service consumer entity, network device, network element, etc., and this application does not limit it.

[0081] For example, in the core network, network elements such as access and mobility management function (AMF), session management function (SMF), and policy control function (PCF) can all use large model instances to assist in completing related services. In other words, network elements such as AMF, SMF, and PCF can be users of large model instances.

[0082] For example, the managed service consumer 1 could be an operation-support system (OSS), or a network management unit provided by an operator, or a third-party cloud server or equipment vendor.

[0083] It should be noted that Figure 1 is a simplified schematic diagram for ease of understanding. The network architecture may also include other devices, such as model providers, which are not shown in Figure 1. In practical applications, this network architecture may include multiple management service consumers or multiple management service producers. This application embodiment does not limit the number of management service consumers and management service producers included in the network architecture.

[0084] The AI ​​and ML technologies in the embodiments of this application will be further illustrated below.

[0085] With the continuous development of AI and ML technologies, these technologies and their related applications are being widely adopted by an increasing number of industries, such as mobile networks in the telecommunications industry. AI and ML technologies can generally be categorized from different perspectives, as follows:

[0086] (1) Classification according to training methods: including supervised learning, semi-supervised learning, unsupervised learning, and reinforcement learning. Each of these learning methods is applicable to one or more specific categories of reasoning (such as prediction) and requires specific types of training data.

[0087] For example, the reasoning methods and training data used for different learning approaches are shown in Table 1. Specifically, supervised learning includes regression (numerical) and classification as its reasoning methods, and its training data includes labeled data; semi-supervised learning includes regression (numerical) and classification as its reasoning methods, and its training data includes labeled or unlabeled data; unsupervised learning includes clustering and association as its reasoning methods, and its training data includes unlabeled data; reinforcement learning includes reward-based behavior as its reasoning method, and its training data does not need to be predefined. Here, labeled data refers to a set of training and test data, which are assigned one or more labels.

[0088] Table 1

[0089] (2) Classification by complexity: Based on learning complexity, it can be divided into machine learning (i.e. basic learning) and deep learning.

[0090] (3) Classification by architecture: Based on the topology and location where the learning task occurs, it can be divided into centralized learning, distributed learning and federated learning.

[0091] (4) Classified according to the continuity of learning: including offline learning and continuous learning.

[0092] AI and ML technologies can also be applied to multiple areas of 5G systems, such as management and orchestration (e.g., model-driven architecture (MDA)) and 5G networks (e.g., network data analytics function (NWDAF)). AI / ML inference functions in 5G systems can use ML models for inference. For each AI / ML technology, depending on its specific characteristics, it may be suitable for specific application scenarios in 5G systems. To implement and promote AI / ML capabilities in 5G systems, and to enable the adoption of appropriate AI / ML technologies, ML models and AI / ML inference functions need to be managed. AI / ML management capabilities can include ML model training, ML model testing, AI / ML inference emulation, ML model deployment, and AI / ML inference. Specifically:

[0093] (1) ML model training: This includes training one or more ML models, which includes initial training and retraining. ML model training may also include ML model validation to evaluate the model's performance on training and validation data. If the validation results do not meet expectations (e.g., unacceptable variance), the ML model needs to be retrained.

[0094] (2) ML Model Testing: Test the validated ML model to evaluate its performance on the test data. If the test results meet expectations, the ML model can proceed to the next step; if the test results do not meet expectations, the ML model needs to be retrained.

[0095] (3) AI / ML Inference Simulation: Run the ML model for inference in a simulation environment. Evaluate the inference performance of the ML model in the simulation environment before applying the model to the target network or system. If the simulation results do not meet expectations (e.g., the inference performance does not meet the target, or the ML model has a negative impact on other existing functions), the ML model needs to be retrained.

[0096] (4) ML Model Deployment: This includes the ML model loading process (i.e., a series of atomic operations) to enable the trained ML model to be used in the target AI / ML inference function. In some cases (e.g., when training and inference functions coexist), ML model deployment may not be necessary.

[0097] (5) AI / ML Inference: Use a trained ML model to perform inference in the AI / ML inference function. AI / ML inference may also trigger model retraining or updates based on performance monitoring and evaluation.

[0098] Optionally, depending on the system implementation and AI / ML functional arrangements, AI / ML inference simulation and ML model deployment can be skipped. That is to say, the ML model lifecycle can include ML model training, ML model testing, and AI / ML inference.

[0099] Figure 2 illustrates a model execution flow. Specifically, the model lifecycle shown in Figure 2 can include steps such as ML model training, ML model testing, AI / ML inference simulation, ML model deployment, and / or AI / ML inference. After ML model training, ML model testing can be performed. After ML model testing, AI / ML inference simulation can be performed; alternatively, ML model deployment or AI / ML inference can also be performed. After AI / ML inference simulation, ML model deployment, AI / ML inference, or ML model training can also be performed. After ML model deployment, AI / ML inference can be performed. After AI / ML inference, ML model training can be performed. The connections between these different steps can refer to a sequence of flows.

[0100] In practical applications, changes in network conditions can lead to a decline in the model's capabilities or render the model unsuitable for the changed conditions. In such cases, the management service consumer can still trigger updates. For example, it can trigger an update when it realizes that the insights or decisions generated by the function are no longer applicable to the observed network state, or when it observes a decline in the inference performance of the ML model.

[0101] Specifically, managed service consumers can request managed service producers to use updated ML models for inference and specify certain performance requirements. This approach provides managed service producers with flexibility to take action based on requests, such as updating the ML model. Updating the ML model might involve loading an already trained ML model, or it might require requesting the management service (MnS) to train or retrain the ML model. Furthermore, depending on the model's configuration, the AI / ML inference function may learn new features during use. For example, if the model is configured to learn through reinforcement learning, or if the model is configured to download new versions of its component ML models, in this case, the AI / ML authorized consumer can be notified of the new capabilities of the AI / ML inference function by the managed service producer (e.g., operator, management function, or network function).

[0102] Currently, in order to reduce the degradation of the model, the deployed model can be updated through periodic updates or passively triggered updates, thereby improving the accuracy of model inference.

[0103] Figure 3 shows a flowchart of a model update method. Specifically, periodic updates or passively triggered updates mainly include the following steps: S301, OSS sends a deployment request to OMC; S302, OMC sends data collection requests to other network elements; S303, OMC receives live network data from other network elements; S304, OMC performs boundary positioning based on the deployed model and live network data; S305, OMC sends configuration information to other network elements; S306, OMC sends processing results to OSS.

[0104] The deployment request here is used to request the OMC to deploy services / models; this deployment request may include model update cycles (e.g., two updates per year) or updates following user complaints. The data collection request is used to request other network elements to collect relevant messages, key performance indicators (KPIs), log data, etc., within their scope. Other network elements can refer to network elements managed by the OMC. Configuration information can be used by other network elements to adjust communication channels and transmit power of various access points (APs).

[0105] However, the method shown in Figure 3 relies on a fixed update cycle or user feedback, which often fails to respond promptly to dynamic changes in the network environment. While periodic updates can ensure continuous model optimization to some extent, their drawbacks are also significant. First, model degradation is usually passively perceived, only becoming apparent after network performance declines or customer feedback, making timely adjustments impossible. Second, long update cycles may result in significant performance degradation before the update, impacting the overall network performance. Furthermore, model training is typically based on laboratory-constructed data, which cannot fully reflect the complexity of real-world network environments, leading to poor adaptability in practical applications, especially prone to severe performance degradation in live network environments. Finally, due to differences in network environments, the same version of the model may exhibit varying degrees of degradation in different application scenarios, further exacerbating the lag and inconsistency of model updates.

[0106] To address the aforementioned shortcomings, model updates can be performed in the cloud. Specifically, the device can periodically collect data and upload it to the cloud for processing and analysis. When the cloud detects model aging, it can update the model, then send the updated model back to the device manufacturer for redeployment on the device. This update method can promptly detect model degradation through inference results and perform corresponding updates.

[0107] For example, as shown in Figure 4, Figure 4 is a schematic diagram of a model update system. Specifically, the model update system may include an application (APP) 401, an OSS 402, a fiber-to-the-room (FTTR) module 403, and a cloud 404. The detailed descriptions of each module are as follows.

[0108] APP 401: Used for on-site troubleshooting of issues causing system lag. For example, the APP here could refer to an installation and maintenance APP.

[0109] Specifically, APP 401 can receive the root cause of the lag from OSS, which may include the cause of the lag problem and / or the location result; then, APP 401 can resolve the lag problem in a targeted manner based on the root cause of the lag.

[0110] OSS 402: This may include a data acquisition module and a stuttering problem localization module. The data acquisition module receives data from the FTTR module. The stuttering problem localization module is used to locate and analyze network stuttering issues. For example, the network here can refer to a home network, a passive optical network (PON), a bearer network, and content sources, etc.

[0111] The process of the stuttering problem delineation module to locate and analyze network stuttering problems can include the following steps: First, the data acquisition module can send FTTR data to the stuttering problem delineation module; second, after receiving the FTTR data, the stuttering problem delineation module can identify stuttering in the FTTR data; then, the stuttering problem delineation module can delineate the faults in the data after stuttering identification; finally, the stuttering problem delineation module can locate the faults in the data after fault delineation to obtain the location result.

[0112] Optionally, the OSS here can be replaced by a home decoration business quality management system.

[0113] FTTR module 403: Used to collect data in the network.

[0114] Cloud-based 404: Used for training and knowledge accumulation of AI models.

[0115] Among them, the cloud 404 can receive sample data from OSS 402, and then the cloud 404 can train the AI ​​model and / or accumulate knowledge based on the sample data, thereby sending the updated model to OSS 402.

[0116] It should be noted that Figure 4 is only an exemplary structural diagram of a model update system. In practical applications, the model update system in Figure 4 can be transformed according to specific circumstances.

[0117] Although the above method can effectively maintain the performance of the model, it requires frequent and large-scale data transfer, resulting in high data transfer costs.

[0118] Therefore, how to proactively detect whether a model has deteriorated is a technical problem that urgently needs to be solved.

[0119] To address the aforementioned technical challenges, model degradation can be detected by analyzing differences in data distribution. In laboratory settings, models are typically trained on specific, carefully designed datasets, often collected under controlled conditions and exhibiting a degree of regularity and consistency. However, data in real-world network environments is often more complex and diverse, encompassing data from different geographical locations, time periods, and network conditions, resulting in significant discrepancies. Consequently, when the model is applied to a real-world network, the distribution of real-world data deviates from that of laboratory data. This distributional difference leads to a decline in the model's predictive performance, meaning the model performs poorly in practical applications.

[0120] Furthermore, laboratory data typically exhibits ideal conditions with minimal noise and relatively uniform patterns and characteristics. In contrast, live network data is affected by various factors, such as network congestion, equipment failure, and changes in user behavior, all of which alter the data distribution. Because models are not trained to account for these complex and dynamically changing factors, they struggle to effectively infer and predict from live network data, resulting in performance degradation. Therefore, changes in data distribution are the root cause of model deterioration.

[0121] Figure 5 illustrates a data distribution diagram, showing the differences between laboratory data and live data. The horizontal axis represents dimension 1, and the vertical axis represents dimension 2. As shown in Figure 5, both laboratory and live data exhibit data distribution drift in both dimensions 1 and 2. Live data can include data with the same distribution as the pre-defined model, as well as data with a different distribution. To improve model performance, data from the live data that differs from the pre-defined model distribution can be used as high-value data for subsequent model updates.

[0122] It should be understood that the data distributed in the same way as the pre-set model can refer to laboratory data.

[0123] In view of this, this application provides a communication method and apparatus that detects whether the first model has deteriorated by analyzing the differences in the data distribution of the first model, which is beneficial to enhancing the sensing capability.

[0124] The technical solution provided in this application will be described in detail below with reference to more accompanying drawings.

[0125] This application uses an OMC (Operational Management Center) as the managed service producer and an OSS (Operational Service Provider) as an example to illustrate the technical solution of this application. Here, OMC can refer to one of multiple OMCs, and OSS can refer to one of multiple OSSs. This application only uses OMCs and OSSs as examples and does not limit the number or type of OMCs and OSSes.

[0126] As shown in Figure 6, Figure 6 is a flowchart illustrating a communication method provided in an embodiment of this application. This communication method includes, but is not limited to, the following steps:

[0127] S601: OMC obtains the first data of the first model.

[0128] The first data refers to the data collected from the actual network. For example, the first data could be live network operation and maintenance data.

[0129] Optionally, the first model can be a pre-trained model. For example, the first model can be a large language model (LLM).

[0130] In one possible approach, other network elements can proactively report the first data they have collected to the OMC.

[0131] Among them, the other network elements are network elements managed by the OMC.

[0132] In another possible approach, the OMC can first send a data acquisition request to other network elements, and after receiving the data acquisition request, the other network elements can then report the first data they have acquired to the OMC.

[0133] It should be understood that the above implementation is merely an illustrative example and does not constitute a limitation on the method of obtaining the first data in the embodiments of this application.

[0134] S602: OMC determines the data distribution differences of the first model based on the first data.

[0135] The data distribution difference of the first model may include the distribution difference between the first data and the second data, or the distribution difference between the degraded data and the sampled data. The second data is part or all of the original data used to train the first model; the sampled data is part or all of the data in the first data; the degraded data is the data in the sampled data whose accuracy after inference by the first model is lower than a third threshold.

[0136] It should be understood that the third threshold can be an empirical parameter, or it can be adjusted according to different application requirements; this application does not limit this. Furthermore, the first, second, fourth, and fifth thresholds mentioned later are similar to the third threshold and will not be repeated hereafter.

[0137] In one possible implementation, the data distribution difference of the first model may include the distribution difference between the first data and the second data.

[0138] Specifically, OMC can first determine the value of the distribution metric between the first data and the second data, and then determine whether the value of the distribution metric is greater than a first threshold. If the value of the distribution metric is greater than the first threshold, OMC can determine that the data distribution difference meets the first condition. If the value of the distribution metric is less than or equal to the first threshold, OMC can determine that the data distribution difference does not meet the first condition.

[0139] It should be understood that the data distribution difference here satisfies the first condition, meaning that compared to the distribution of the second data, the distribution of the first data has changed significantly, and the first model may have deteriorated or become unsuitable. The data distribution difference here does not satisfy the first condition, meaning that the distribution difference between the first and second data is small, the inference performance of the first model should be relatively stable, there is no obvious deterioration, and no further processing is required.

[0140] Optionally, a distance metric can be used to determine the value of the distribution metric between the first and second data points. For example, distance metrics can include Kullback-Leibler divergence, Wasserstein distance, Jensen-Shannon divergence, cosine similarity, etc.

[0141] KL divergence measures the difference between the probability distributions of different data points; a larger KL divergence value indicates a greater difference in distribution between the first and second data points. Wasserstein distance and JS divergence measure the similarity between the distributions of different data points; a smaller distance value indicates a more similar distribution between the first and second data points. Cosine similarity measures the similarity between the distributions of two data points by calculating the angle between vectors; a cosine similarity value closer to 1 indicates a more similar distribution between the first and second data points.

[0142] For example, OMC can determine the value of the distribution metric between the first and second data points by calculating the KL divergence. When the KL divergence value between the first and second data points is greater than a first threshold, it can be determined that the data distribution difference meets the first condition.

[0143] For example, OMC can also determine the distribution metric between the first and second data points by calculating cosine similarity. When 1-cos(x,y) is greater than a first threshold, it can be determined that the data distribution difference meets the first condition. Here, cos(x,y) represents the cosine similarity value between the first and second data points, where x represents the first data point and y represents the second data point.

[0144] In another possible implementation, the data distribution difference in the first model may include the distribution difference between degraded data and sampled data.

[0145] Specifically, OMC can first determine the number of degraded data and the number of sampled data, and then determine whether the ratio of the number of degraded data to the number of sampled data is greater than the fourth threshold. If the ratio is greater than the fourth threshold, OMC can determine that the data distribution difference meets the first condition, that is, the proportion of degraded data is high, and the first model has degraded significantly. If the ratio is less than or equal to the fourth threshold, OMC can determine that the data distribution difference does not meet the first condition, that is, the proportion of degraded data is low, the inference effect of the first model should be relatively stable, there is no significant degradation, and no further processing is required.

[0146] It should be understood that the distribution difference here satisfies the first condition, meaning that the proportion of degraded data is relatively high, and the first model has significantly degraded. The data distribution difference here does not satisfy the first condition, meaning that the proportion of degraded data is relatively low, and the first model has not significantly degraded, requiring no further processing.

[0147] Optionally, OMC can determine the amount of degraded data and the amount of sampled data by constructing a decision tree. Specifically, firstly, OMC can select part or all of the data from the first dataset as sampled data, then input the sampled data into the first model for inference, obtain the inference result, and then, based on the inference result, assign pseudo-labels to each sampled data (e.g., label 0 if the first model infers correctly, label 1 if the first model infers incorrectly), thus constructing a pseudo-labeled dataset; then, using the pseudo-labeled dataset, a decision tree is constructed, where each node of the decision tree represents a part of the data, and each node represents a sample set under a specific condition. These nodes are not only used for classification and decision-making, but also for calculating the accuracy of the first model's inference at each node; after calculating the accuracy of the first model's inference at each node of the decision tree, the performance of samples of different categories can be evaluated. If the accuracy of some nodes is significantly lower than expected and the proportion of inference errors is high, it indicates that the first model may have degraded under certain data distributions or scenarios, and data in the sampled data whose accuracy of the first model's inference is lower than a third threshold can be considered degraded data; finally, the amount of degraded data and the amount of sampled data are counted.

[0148] It should be understood that the selection of sampling data can be random or based on certain screening criteria; this application does not limit this selection. Furthermore, this pseudo-label dataset, constructed through the correctness of inference, can effectively identify the performance of the first model on specific data. The annotations in this dataset reflect the accuracy and bias of the first model in the current live network environment.

[0149] S603: When the data distribution difference meets the first condition, the OMC sends the first message to the OSS.

[0150] The first message may include a first instruction message, which can be used to instruct OSS to update the first model.

[0151] It should be understood that the first instruction information here can also be a notification information, which can be used to notify OSS to update the first model.

[0152] Specifically, when the data distribution difference meets the first condition, it can be understood that the distribution of the first data has changed significantly compared to the distribution of the second data, and the first model may have deteriorated or become unsuitable; or, it can be understood that the proportion of deteriorated data is high, and the first model has deteriorated significantly. In order to reduce the degree of deterioration of the first model, it is necessary to update the first model. By sending the first message, the OMC can instruct the OSS to update the first model, or the OMC can notify the OSS to update the first model.

[0153] Optionally, the first message may also include third data from the first model.

[0154] In one possible implementation, when the data distribution difference of the first model includes the distribution difference between the first data and the second data, the third data can be the first number of data in the first data that has the largest value of the distribution metric between the first data and the second data, or the third data can be data in the first data that has a value of the distribution metric between the first data and the second data that is greater than a second threshold.

[0155] Specifically, based on the distribution difference between the first data and the second data, a subset of samples with the largest distribution difference from the second data can be selected from the first data (for example, the first number of data with the largest distribution metric between the first data and the second data, or data with a distribution metric between the first data and the second data that is greater than a second threshold). These samples can be used as a fine-tuning set for updating or retraining the first model.

[0156] It should be understood that the fine-tuning set here mainly includes the key distributional differences between the first and second data. These samples can help the first model better adapt to the live network environment.

[0157] In another possible implementation, when the data distribution difference of the first model includes the distribution difference between degraded data and sampled data, the third data can be the second number of data in the first data that has the smallest value of the distribution metric between the first data and the degraded data, or the third data can be data in the first data whose value of the distribution metric between the first data and the degraded data is less than a fifth threshold.

[0158] Specifically, based on the identified degraded data, a subset of data with the smallest distribution difference from the degraded data can be selected from the first data (for example, the second number of data in the first data with the smallest distribution metric value between the degraded data and the first data, or data in the first data with a distribution metric value between the degraded data and the first data that is less than a fifth threshold). These samples can be used as a fine-tuning set for updating or retraining the first model.

[0159] It should be understood that the fine-tuning set here represents typical cases in the first dataset that the first model cannot currently handle accurately. It mainly includes key samples of the degraded parts of the first model. Fine-tuning these data helps the first model quickly adapt to changes in the live network environment. Furthermore, this process usually requires less training time and has lower computational resource requirements, which helps reduce the cost of model updates.

[0160] Optionally, the method for calculating the distribution metric between the first data and the degraded data includes, but is not limited to, distance measurement methods, feature matching, etc., and this application does not limit it in this regard.

[0161] Optionally, the values ​​of the first quantity and the second quantity can be the same or different, and this application does not limit this.

[0162] Optionally, the first message may also include identification information for the first model. Specifically, OSS can distinguish each model that needs to be updated, as well as the information and / or data corresponding to each model, based on the identification information in the first message.

[0163] For example, the information and / or data contained in the first message may be as shown in Table 2. Specifically, the first message may include two attributes: the identification information of the first model and the third data of the first model. The support qualifier corresponding to the identification information of the first model is conditional mandatory (CM); the support qualifier for the third data of the first model is also CM.

[0164] Table 2

[0165] In this embodiment, by acquiring the first data, the OMC can perform a difference analysis between the data in the actual network and the original data used to train the first model. Alternatively, the OMC can first input the first data into the first model for inference, and then perform a difference analysis between the data in the actual network and the inference results of the first model, thereby determining the data distribution difference of the first model. By judging that the data distribution difference meets the first condition, the OMC can actively perceive the degradation of the first model, enhance its perception capability, and overcome the limitations that may exist in passively perceiving model degradation. By sending the first message, the OSS can update the first model without frequently transmitting large amounts of data, which can reduce the cost of model updates and improve model performance.

[0166] The technical solution provided in this application can also be described in detail through the method embodiment shown in FIG7. It should be understood that the method embodiment shown in FIG7 describes the method in a three-party interactive manner, and the steps performed by each executing entity can be treated as a separate method.

[0167] As shown in Figure 7, which is a flowchart of another communication method provided in an embodiment of this application, the details are as follows.

[0168] S701: OSS sends a second request to OMC.

[0169] The second request is used to request OMC to deploy the first model. The second request includes information about the first model, which may include at least one of the following: a third data collection strategy, second data, a second algorithm, a first threshold, a second threshold, a third threshold, a fourth threshold, or a fifth threshold. The second algorithm is used to determine the numerical values ​​of distribution metrics between different data points. For example, the second algorithm could be KL divergence, Wasserstein distance, JS divergence, or cosine similarity, etc.

[0170] It should be understood that the information in the first model can be carried and transmitted by signaling, or the information in the first model can be predefined by the protocol. The signaling here can be physical layer signaling or higher layer signaling, and this application does not limit it.

[0171] In one possible approach, OSS sends a second request to OMC, which may include information about the first model. The information about the first model may include at least one of the following: a third data collection strategy, second data, a second algorithm, a first threshold, a second threshold, a third threshold, a fourth threshold, or a fifth threshold.

[0172] In another possible approach, OSS and OMC obtain the mapping relationship between collection strategies and indexes. This mapping relationship includes multiple collection strategies, each corresponding to a different index. OSS sends a second request to OMC, which may include information about the first model. The information about the first model may include at least one of the following: the index of the third data collection strategy, the second data, the second algorithm, a first threshold, a second threshold, a third threshold, a fourth threshold, or a fifth threshold. After receiving the index of the third data collection strategy, OMC determines the third data collection strategy by querying the mapping relationship between collection strategies and indexes.

[0173] The mapping relationship between the collection strategy and the index can be predefined, pre-stored, pre-burned, or pre-configured. Predefinition can include predefined elements, such as protocol definitions. Alternatively, the mapping relationship can be configured or pre-configured. Pre-configuration can be achieved by pre-saving the corresponding code, table, or other means that can be used to indicate relevant information in the device. This application does not limit the specific implementation method.

[0174] For example, as shown in Figure 8, which is a flowchart illustrating a collection strategy provided in an embodiment of this application, the third data can be selected by calculating the value of a distribution metric between the first data and the second data. This mainly includes the following steps: S801, calculating the value of a distribution metric between the first data and the second data; S802, determining whether the value of the distribution metric is greater than a first threshold; S803, when the value of the distribution metric is greater than the first threshold, determining that the first model has deteriorated; S804, selecting the first number of data points with the largest distribution metric value between the first data and the second data as the third data, or selecting the data points with a distribution metric value greater than the second threshold as the third data; S805, when the value of the distribution metric is less than or equal to the first threshold, determining that the first model has not deteriorated.

[0175] Specifically, after executing step S802, if the value of the distribution metric is greater than the first threshold, then execute steps S803 to S804; if the value of the distribution metric is less than or equal to the first threshold, then execute step S805.

[0176] For example, as shown in Figure 9, which is a flowchart illustrating another collection strategy provided in this application embodiment, third data can be collected by calculating the value of a distribution metric between degraded data and first data. This mainly includes the following steps: S901, selecting sampled data from the first data; S902, inputting the sampled data into a first model for inference to obtain inference results; S903, labeling the sampled data based on the inference results to obtain a pseudo-label dataset; S904, constructing a decision tree based on the pseudo-label dataset; S905, calculating the accuracy of the first model inference at each node of the decision tree; S906, removing third data from the sampled data whose accuracy is lower than that of the first model inference. The threshold data is used as degraded data; S907, the ratio of the number of degraded data to the number of sampled data is calculated; S908, it is determined whether the ratio is greater than the fourth threshold; S909, when the ratio is greater than the fourth threshold, the first model is determined to be degraded; S910, the second number of data in the first data with the smallest distribution metric value between it and the degraded data is used as the third data, or the data in the first data with the distribution metric value between it and the degraded data is less than the fifth threshold is used as the third data; S911, when the ratio is less than or equal to the fourth threshold, the first model is determined not to be degraded.

[0177] Specifically, after executing step S908, if the ratio is greater than the fourth threshold, then execute steps S909 to S910; if the ratio is less than or equal to the fourth threshold, then execute step S911.

[0178] It should be understood that the methods shown in Figures 8 and 9 are merely illustrative examples and do not constitute a limitation of this application.

[0179] Optionally, the second request can also be used to request the OMC to deploy other models. This application does not limit the number of other models or the types of other models.

[0180] S702: OSS sends a fault handling request to OMC.

[0181] S703: OMC sends data subscriptions required for services to other network elements.

[0182] Here, "other network elements" can refer to network elements managed by the OMC.

[0183] For example, the data subscriptions required by the business can be user call logs or network element measurement reports.

[0184] S704: Other network elements send the first data to the OMC.

[0185] The first data refers to the data collected from the actual network. For example, the first data could be live network operation and maintenance data.

[0186] S705: OMC determines the data distribution differences of the first model based on the first data.

[0187] S706: When the data distribution difference meets the first condition, the OMC sends the first message to the OSS.

[0188] The specific implementation methods of steps S705 to S706 are the same as those of steps S602 to S603 in the above embodiment. You can refer to steps S602 to S603, and they will not be repeated here.

[0189] S707: OSS sends the first request to OMC.

[0190] The first request can be used to request the OMC to update the first model; the first request can include first information, which can include the update method of the first model, including remote update or local update. Remote update is used to indicate that the model update is performed by OSS, and local update is used to indicate that the model update is performed by OMC.

[0191] In one possible approach, the update method may include remote updates, and the first information may also include the update weights of the first model.

[0192] Here, updating the weights can refer to modifying the model weight increment matrix.

[0193] Specifically, if the model update is performed by OSS, OSS can retrain the first model based on the third data to obtain the updated weights of the first model.

[0194] In another possible approach, the update method may include local updates, and the first information may further include at least one of the following: fourth data, a first time range, or a first algorithm. The first time range is the time range for updating the first model; the first algorithm is used to determine the degree of degradation of the first model. For example, the first algorithm may be KL divergence.

[0195] Specifically, if the model update is performed by OMC, OSS can filter the third data in the first message to obtain the fourth data, which is the data in the third data used to update the first model.

[0196] Optionally, the update mode of the first model can be set to "remote" to indicate a remote update; or, the update mode of the first model can be set to "local" to indicate a local update.

[0197] Optionally, the first information may also include the identification information of the first model.

[0198] For example, the information and / or data included in the first information may be as shown in Table 3. Specifically, the first information may include at least one of the following attributes: the update method of the first model, the identification information of the first model, the update weight of the first model, the fourth data of the first model, the first time range, or the first algorithm. The support qualifier corresponding to the update weight of the first model is CM; the support qualifiers for the update method of the first model, the identification information of the first model, the fourth data of the first model, the first time range, and the first algorithm are mandatory (mandatory, M).

[0199] Table 3

[0200] Specifically, for each attribute in Table 3, if the update method includes remote update, the first information may include the update method of the first model, the identification information of the first model, and / or the update weight of the first model; if the update method includes local update, the first information may include the update method of the first model, the identification information of the first model, the fourth data of the first model, the first time range, and / or the first algorithm.

[0201] Optionally, the first request may also include a first response, which may be used to instruct OSS to determine whether to update the first model.

[0202] S708: OMC updates the first model based on the first request.

[0203] In one possible approach, when the update method includes remote updates, the OMC can update the first model based on the update weight in the first request.

[0204] In another possible implementation, when the update method includes local update, the OMC can update the first model based on the fourth data in the first request, the first time range, and / or the first algorithm.

[0205] As shown in Figure 10, Figure 10 is a schematic diagram of an application scenario for model updating provided by an embodiment of this application. Specifically, model updating can include an offline pre-setting stage and an online evolution stage. In the offline pre-setting stage, the first model can be updated based on data from multiple application scenarios to obtain the first model to be adjusted. For example, the data from multiple application scenarios can include data A, data B, ... In the online evolution stage, the first model to be adjusted can be retrained based on different application scenarios and different data. For example, different application scenarios can include new application scenario 1, new application scenario 2, ... new application scenario N.

[0206] In the new application scenario 1, the process of retraining the first model to be adjusted can include the following steps: when t=0, the first model to be adjusted can be retrained based on data C0; when t=1, the first model to be adjusted can be retrained based on data C1; when t=2, the first model to be adjusted can be retrained based on data C2; ... where t refers to a statistic.

[0207] In the new application scenario 2, the process of retraining the first model to be adjusted can include the following: when t=0, the first model to be adjusted can be retrained based on data D0; when t=1, the first model to be adjusted can be retrained based on data D1; when t=2, the first model to be adjusted can be retrained based on data D2; ...

[0208] S709: OMC sends updated configuration information to other network elements.

[0209] For example, the updated configuration information may include radio frequency (RF) parameters.

[0210] S710: OMC sends the processing result to OSS.

[0211] The processing result may include at least one of the following: the inference result of the updated first model or the second response. The second response may be used to instruct the OMC to determine to update the first model.

[0212] Furthermore, the technical solution provided in this application can also be applied to updating multiple models. Moreover, to ensure network reliability and stability and improve update efficiency, the OMC can also request the OSS to update multiple models.

[0213] Specifically, in this scenario, the OMC can obtain the first data of each of the M models. Then, the OMC can determine the data distribution differences of each of the M models based on the first data of each of the M models. When the data distribution differences of N models among the M models meet the first condition, the OMC can send a first message to the OSS. The first message may include first indication information, which is used to instruct the OSS to update the N models. Alternatively, the first message may also include N first indication messages, one of which is used to instruct the OSS to update any one of the N models.

[0214] Optionally, the first message here may also include the third data for each of the N models.

[0215] Where M is an integer greater than 0, and N is an integer greater than 0 and less than or equal to M. The specific process for calculating the data distribution difference of each of the M models can be referred to step S602 in the above embodiment, and will not be repeated here. The specific process for requesting updates to the N models can be referred to step S603 in the above embodiment, and will also not be repeated here.

[0216] By adopting the embodiments of this application, receiving a second request enables the OMC to deploy a first model, which facilitates the management of resources in the communication network using the first model, thereby improving the reliability and stability of the network; by acquiring first data, the OMC can perform difference analysis between the data in the actual network and the original data used to train the first model, or the OMC can perform difference analysis between the data in the actual network and the inference results of the first model, thereby determining the data distribution differences of the first model; by judging whether the data distribution differences meet the first condition, the OMC can proactively perceive whether the first model has deteriorated, enhancing its perception capability and overcoming the limitations of passively perceiving model deterioration; by sending a first message, the OSS can update the first model without frequent large-scale data transmission, which reduces the cost of model updates and improves model performance; by receiving a first request, the OMC can adjust the first model according to the first information in the first request, which helps the updated first model to better adapt to the environment in the actual network and obtain better model inference accuracy.

[0217] The methods of the embodiments of this application have been described in detail above, and the apparatus of the embodiments of this application is provided below.

[0218] As shown in Figure 11, Figure 11 is a schematic diagram of a communication device provided in an embodiment of this application. This communication device can be a management service provider, such as the aforementioned OMC, or a chip or processing system within the OMC. This device can be used to implement any method and function related to the OMC in any of the foregoing embodiments. The device may include a receiving module 1101, a processing module 1102, and a transmitting module 1103. Optionally, the transmitting module 1103 may correspond to the radio frequency circuit and baseband circuit included in the OMC. The detailed descriptions of each module are as follows.

[0219] The receiving module 1101 is used to acquire the first data of the first model, which is the data collected from the actual network.

[0220] The processing module 1102 is used to determine the data distribution differences of the first model based on the first data.

[0221] The sending module 1103 is used to send a first message to the management service consumer when the data distribution difference meets a first condition. The first message includes first instruction information, which is used to instruct the management service consumer to update the first model.

[0222] Optionally, the data distribution difference includes the distribution difference between the first data and the second data of the first model, wherein the second data is part or all of the original data used to train the first model; the processing module 1102 is further used to determine the value of the distribution metric between the first data and the second data; when the value of the distribution metric is greater than a first threshold, it is determined that the data distribution difference meets the first condition.

[0223] Optionally, the first message may also include third data of the first model, wherein the third data is the first number of data in the first data that has the largest value of the distribution metric between the first data and the second data, or the third data is data in the first data that has a value of the distribution metric between the first data and the second data that is greater than a second threshold.

[0224] Optionally, the data distribution difference includes the distribution difference between degraded data and sampled data, where the sampled data is part or all of the data in the first data, and the degraded data is the data in the sampled data whose accuracy after inference by the first model is lower than the third threshold; the processing module 1102 is also used to determine the ratio of the number of degraded data to the number of sampled data; when the ratio is greater than the fourth threshold, it is determined that the data distribution difference satisfies the first condition.

[0225] Optionally, the first message may also include third data of the first model, wherein the third data is the second number of data in the first data that has the smallest value of the distribution metric between the first data and the degraded data, or the third data is data in the first data whose value of the distribution metric between the first data and the degraded data is less than a fifth threshold.

[0226] Optionally, the receiving module 1101 is further configured to receive a first request from the management service consumer, the first request being used to request the management service producer to update the first model, the first request including first information, the first information being determined by the management service consumer based on third data.

[0227] Optionally, the processing module 1102 is also used to update the first model based on the first information.

[0228] Optionally, the first information includes the update method of the first model, which includes remote update or local update. Remote update is used to instruct the management service consumer to perform the model update, and local update is used to instruct the management service producer to perform the model update.

[0229] Optionally, when the update method includes remote update, the first information may also include the update weights of the first model.

[0230] Optionally, when the update method includes local update, the first information further includes at least one of the following: fourth data of the first model, a first time range, or a first algorithm; the fourth data is the data in the third data used to update the first model; the first time range is the time range for updating the first model; and the first algorithm is used to determine the degree of degradation of the first model.

[0231] Optionally, the receiving module 1101 is also configured to receive a second request from the management service consumer, the second request being used to request the management service producer to deploy the first model.

[0232] Optionally, the second request includes information about the first model, which includes a third data collection strategy.

[0233] Optionally, the information of the first model may also include at least one of the following: second data, a second algorithm, a first threshold, a second threshold, a third threshold, a fourth threshold, or a fifth threshold; the second algorithm is used to determine the value of the distribution metric among the different data.

[0234] Optionally, the first message may also include identification information of the first model.

[0235] It should be noted that the implementation of each module can also correspond to the descriptions of the method embodiments shown in Figures 6-10, and execute the methods and functions performed by the management service producer in the above embodiments.

[0236] As shown in Figure 12, Figure 12 is a schematic diagram of a communication device provided in an embodiment of this application. This communication device can be a management service consumer, such as the aforementioned OSS, or a chip or processing system within the OSS. This device can be used to implement any method and function related to the OSS in any of the foregoing embodiments. The device may include a receiving module 1201, a processing module 1202, and a transmitting module 1203. Optionally, the transmitting module 1203 may correspond to the radio frequency circuit and baseband circuit included in the OSS. The detailed description of each module is as follows.

[0237] The receiving module 1201 is used to receive a first message from the management service producer when the data distribution difference of the first model meets a first condition. The first message includes first instruction information, which is used to instruct the management service consumer to update the first model. The data distribution difference is determined by the management service producer based on the first data of the first model. The first data is the data collected in the actual network.

[0238] Optionally, the data distribution difference includes the distribution difference between the first data and the second data of the first model, where the second data is part or all of the original data used to train the first model.

[0239] Optionally, the first message may also include third data of the first model, wherein the third data is the first number of data in the first data that has the largest value of the distribution metric between the first data and the second data, or the third data is data in the first data that has a value of the distribution metric between the first data and the second data that is greater than a second threshold.

[0240] Optionally, the data distribution difference includes the distribution difference between degraded data and sampled data, where the sampled data is part or all of the data in the first data, and the degraded data is the data in the sampled data whose accuracy after inference by the first model is lower than the third threshold.

[0241] Optionally, the first message may also include third data of the first model, wherein the third data is the second number of data in the first data that has the smallest value of the distribution metric between the first data and the degraded data, or the third data is data in the first data whose value of the distribution metric between the first data and the degraded data is less than a fifth threshold.

[0242] Optionally, the processing module 1202 is used to determine the first information based on the third data.

[0243] Optionally, the sending module 1203 is used to send a first request to the management service producer. The first request is used to request the management service producer to update the first model. The first request includes first information, which is used by the management service producer to update the first model.

[0244] Optionally, the first information includes the update method of the first model, which includes remote update or local update. Remote update is used to instruct the management service consumer to perform the model update, and local update is used to instruct the management service producer to perform the model update.

[0245] Optionally, when the update method includes remote update, the first information may also include the update weights of the first model.

[0246] Optionally, when the update method includes local update, the first information further includes at least one of the following: fourth data of the first model, a first time range, or a first algorithm, wherein the fourth data is the data in the third data used to update the first model; the first time range is the time range for updating the first model; and the first algorithm is used to determine the degree of degradation of the first model.

[0247] Optionally, the sending module 1203 is also used to send a second request to the management service producer, the second request being used to request the management service producer to deploy the first model.

[0248] Optionally, the second request includes information about the first model, which includes a third data collection strategy.

[0249] Optionally, the information of the first model may also include at least one of the following: second data, a second algorithm, a first threshold, a second threshold, a third threshold, a fourth threshold, or a fifth threshold; the second algorithm is used to determine the value of the distribution metric among the different data.

[0250] Optionally, the first message may also include identification information of the first model.

[0251] It should be noted that the implementation of each module can also correspond to the descriptions of the method embodiments shown in Figures 6-10, and execute the methods and functions performed by the management service consumers in the above embodiments.

[0252] Figure 13 is a schematic diagram of a communication device provided in an embodiment of this application. This communication device is used to perform the functions of OMC and OSS in the above method embodiments, or to implement the steps or processes executed by OMC and OSS in the above method embodiments.

[0253] As shown in Figure 13, the communication device includes a processor 1301 and a transceiver 1302. Optionally, the communication device also includes a memory 1303. The processor 1301, transceiver 1302, and memory 1303 can communicate with each other via internal connections to transmit control and / or data signals. The memory 1303 stores computer programs. The processor 1301 retrieves and runs the computer program from the memory 1303 to control the transceiver 1302 to transmit and receive signals. Optionally, the communication device may also include an antenna for transmitting uplink data or uplink control signaling output by the transceiver 1302 via wireless signals.

[0254] The processor 1301 and the memory 1303 can be combined into a single processing device. The processor 1301 executes the program code stored in the memory 1303 to achieve the above-mentioned functions. In specific implementations, the memory 1303 can be integrated into the processor 1301 or be independent of the processor 1301.

[0255] The transceiver 1302 described above can also be referred to as a transceiver unit or transceiver module. The transceiver 1302 may include a receiver (or receiver circuit) and a transmitter (or transmitter circuit). The receiver is used to receive signals, and the transmitter is used to transmit signals.

[0256] It should be understood that the communication device shown in Figure 13 can implement the various processes involving OMC and OSS in the method embodiments shown in Figures 6-10. The operation and / or function of each module in the communication device are respectively for implementing the corresponding processes in the above method embodiments. For details, please refer to the description in the above method embodiments; to avoid repetition, detailed descriptions are appropriately omitted here.

[0257] The processor 1301 described above can be used to execute the actions implemented internally by the OMC and OSS as described in the preceding method embodiments, while the transceiver 1302 can be used to execute the receiving or transmitting actions of the OMC and OSS as described in the preceding method embodiments. Please refer to the descriptions in the preceding method embodiments for details, which will not be repeated here.

[0258] The processor 1301 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor 1301 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, etc. The communication bus 1304 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in Figure 13, but this does not indicate that there is only one bus or one type of bus. The communication bus 1304 is used to realize the connection and communication between these components. In this embodiment, the transceiver 1302 is used for signaling or data communication with other node devices. Memory 1303 may include volatile memory, such as nonvolatile random access memory (NVRAM), phase change RAM (PRAM), magnetoresistive RAM (MRAM), etc., and may also include non-volatile memory, such as at least one disk storage device, electrically erasable programmable read-only memory (EEPROM), flash memory devices, such as NOR flash memory or NAND flash memory, semiconductor devices, such as solid-state disk (SSD), etc. Memory 1303 may also be at least one storage device located remotely from the aforementioned processor 1301. Memory 1303 may also store a set of computer program code or configuration information. Processor 1301 may also execute the program stored in memory 1303. The processor can cooperate with the memory and transceiver to execute any of the methods and functions related to OMC and OSS in the above-described embodiments.

[0259] This application also provides a chip, including a processor and a communication interface, the communication interface being used to communicate with external or internal devices, and the processor being used to implement the methods described above.

[0260] In one possible design, the chip may further include a memory storing computer programs or instructions, which the processor executes, either from the stored computer programs or instructions or derived from other programs or instructions. When the computer program or instructions are executed, the processor implements the methods described above.

[0261] In another possible design, the chip can be integrated into either the management service producer or the management service consumer.

[0262] This application also provides a processor coupled to a memory for executing any method and function involving a management service producer or a management service consumer in any of the above embodiments.

[0263] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform any method and function involving a management service producer or a management service consumer in any of the above embodiments.

[0264] This application also provides an apparatus for performing any of the methods and functions related to management service producers or management service consumers in any of the above embodiments.

[0265] This application also provides a communication system, which includes at least one management service producer and at least one management service consumer involved in any of the above embodiments.

[0266] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the communication device, the unit or module within the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0267] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital versatile discs (DVDs)), or semiconductor media (e.g., SSDs), etc.

[0268] It should be understood that the "and / or" appearing in the embodiments of this application is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0269] It should be understood that in the embodiments of this application, "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0270] It should be understood that the symbol " / " appearing in the embodiments of this application can indicate that the preceding and following objects are in an "or" relationship. Additionally, the symbol " / " can also represent a division sign, i.e., performing a division operation. For example, A / B can mean A divided by B.

[0271] It is understood that in the embodiments of this application, the management service producer and / or management service consumer may perform some or all of the steps in the embodiments of this application. These steps or operations are merely examples, and other operations or variations thereof may also be performed in the embodiments of this application. Furthermore, the various steps may be performed in different orders as presented in the embodiments of this application, and it is not necessary to perform all the operations in the embodiments of this application.

[0272] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. Any modifications, equivalent substitutions, or improvements made within the principles of this application should be included within the scope of protection of this application.

Claims

1. A communication method, characterized in that, Applied to managed service producers, the method includes: Obtain the first data of the first model, where the first data is the data collected from the actual network. Based on the first data, determine the data distribution differences of the first model; When the data distribution difference meets the first condition, a first message is sent to the management service consumer. The first message includes first indication information, which is used to instruct the management service consumer to update the first model.

2. The method as described in claim 1, characterized in that, The data distribution difference includes the distribution difference between the first data and the second data of the first model, wherein the second data is part or all of the original data used to train the first model; The step of determining the data distribution difference of the first model based on the first data includes: Determine the value of the distribution metric between the first data and the second data; When the value of the distribution metric is greater than the first threshold, it is determined that the data distribution difference satisfies the first condition.

3. The method as described in claim 2, characterized in that, The first message also includes third data of the first model, wherein the third data is a first number of data in the first data whose distribution metric value is the largest among the first data and the second data, or the third data is data in the first data whose distribution metric value is greater than a second threshold among the second data.

4. The method as described in claim 1, characterized in that, The data distribution difference includes the distribution difference between degraded data and sampled data. The sampled data is part or all of the data in the first data. The degraded data is the data in the sampled data whose accuracy after inference by the first model is lower than the third threshold. The step of determining the data distribution difference of the first model based on the first data includes: Determine the ratio of the quantity of the degraded data to the quantity of the sampled data; When the ratio is greater than the fourth threshold, it is determined that the data distribution difference satisfies the first condition.

5. The method as described in claim 4, characterized in that, The first message also includes third data of the first model, wherein the third data is the second number of data in the first data whose distribution metric value between the first data and the degraded data is the smallest, or the third data is data in the first data whose distribution metric value between the first data and the degraded data is less than a fifth threshold.

6. The method as described in claim 3 or 5, characterized in that, The method further includes: Receive a first request from the management service consumer, the first request being used to request the management service producer to update the first model, the first request including first information, the first information being determined by the management service consumer based on the third data; Based on the first information, the first model is updated.

7. The method as described in claim 6, characterized in that, The first information includes the update method of the first model, which includes remote update or local update. The remote update is used to instruct the management service consumer to perform the model update, and the local update is used to instruct the management service producer to perform the model update.

8. The method as described in claim 7, characterized in that, When the update method includes remote update, the first information also includes the update weights of the first model.

9. The method as described in claim 7, characterized in that, When the update method includes the local update, the first information further includes at least one of the following: fourth data of the first model, a first time range, or a first algorithm; the fourth data is the data used to update the first model from the third data; the first time range is the time range for updating the first model; the first algorithm is used to determine the degree of degradation of the first model.

10. The method according to any one of claims 6-9, characterized in that, The method further includes: A second request is received from the management service consumer, the second request being used to request the management service producer to deploy the first model.

11. The method as described in claim 10, characterized in that, The second request includes information about the first model, which includes the collection strategy for the third data.

12. The method as described in claim 11, characterized in that, The information of the first model also includes at least one of the following: the second data, the second algorithm, the first threshold, the second threshold, the third threshold, the fourth threshold, or the fifth threshold; the second algorithm is used to determine the value of the distribution metric among different data.

13. The method according to any one of claims 1-12, characterized in that, The first message also includes the identification information of the first model.

14. A communication method, characterized in that, Applied to managing service consumers, the method includes: When the data distribution difference of the first model meets the first condition, a first message is received from the management service producer. The first message includes first indication information, which is used to instruct the management service consumer to update the first model. The data distribution difference is determined by the management service producer based on the first data of the first model. The first data is the data collected in the actual network.

15. The method as described in claim 14, characterized in that, The data distribution difference includes the distribution difference between the first data and the second data of the first model, wherein the second data is part or all of the original data used to train the first model.

16. The method as described in claim 15, characterized in that, The first message also includes third data of the first model, wherein the third data is a first number of data in the first data whose distribution metric value is the largest among the first data and the second data, or the third data is data in the first data whose distribution metric value is greater than a second threshold among the second data.

17. The method as described in claim 14, characterized in that, The data distribution difference includes the distribution difference between degraded data and sampled data. The sampled data is part or all of the data in the first data. The degraded data is the data in the sampled data whose accuracy after inference by the first model is lower than a third threshold.

18. The method as described in claim 17, characterized in that, The first message also includes third data of the first model, wherein the third data is the second number of data in the first data whose distribution metric value between the first data and the degraded data is the smallest, or the third data is data in the first data whose distribution metric value between the first data and the degraded data is less than a fifth threshold.

19. The method as described in claim 16 or 18, characterized in that, The method further includes: Based on the third data, the first information is determined; A first request is sent to the management service producer. The first request is used to request the management service producer to update the first model. The first request includes the first information, which is used by the management service producer to update the first model.

20. The method as described in claim 19, characterized in that, The first information includes the update method of the first model, which includes remote update or local update. The remote update is used to instruct the management service consumer to perform the model update, and the local update is used to instruct the management service producer to perform the model update.

21. The method as described in claim 20, characterized in that, When the update method includes remote update, the first information also includes the update weights of the first model.

22. The method as described in claim 20, characterized in that, When the update method includes the local update, the first information further includes at least one of the following: fourth data of the first model, a first time range, or a first algorithm, wherein the fourth data is the data used to update the first model from the third data; the first time range is the time range for updating the first model; and the first algorithm is used to determine the degree of degradation of the first model.

23. The method according to any one of claims 19-22, characterized in that, The method further includes: A second request is sent to the management service producer, the second request being used to request the management service producer to deploy the first model.

24. The method as described in claim 23, characterized in that, The second request includes information about the first model, which includes the collection strategy for the third data.

25. The method as described in claim 24, characterized in that, The information of the first model further includes at least one of the following: the second data, the second algorithm, the first threshold, the second threshold, the third threshold, the fourth threshold, or the fifth threshold; the second algorithm is used to determine the value of the distribution metric among different data.

26. The method according to any one of claims 14-25, characterized in that, The first message also includes the identification information of the first model.

27. A communication device, characterized in that, It includes modules for performing the method of any one of claims 1-13, or modules for performing the method of any one of claims 14-26.

28. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a computer program that, when executed by a processor, causes the method of any one of claims 1-13 or any one of claims 14-26 to be implemented.

29. A communication system, characterized in that, It includes management service producers and management service consumers, wherein the management service producers are used to perform the method of any one of claims 1-13, and the management service consumers are used to perform the method of any one of claims 14-26.

30. A computer program product, characterized in that, The computer program product includes a computer program that, when run on a computer, causes the computer to perform the method of any one of claims 1-13 or any one of claims 14-26.

Images