Model delivery methods, computer-readable storage medium, electronic apparatus and computer program product

Abstract

Provided in the embodiments of the present disclosure are model delivery methods, a computer-readable storage medium, an electronic apparatus and a computer program product. A method comprises: sending a first model delivery request to a second network element, the first model delivery request comprising at least one of the following: an artificial intelligence / machine learning (AI / ML) inference requirement, a model performance requirement, model delivery information, a first model identifier, and model performance satisfaction information; and receiving a target model delivery response sent by the second network element on the basis of a first matching result, the first matching result being obtained by the second network element performing model matching on the basis of the first model delivery request and model delivery restriction information. The present disclosure can at least solve the problem in the prior art of how to deliver ML models in networks, so as to implement model delivery in the networks.

Description

Model transmission methods, computer-readable storage media, electronic devices and computer program products

[0001] Cross-references to related applications

[0002] This disclosure is based on and claims priority to Chinese Patent Application No. 202510053003X, filed on January 10, 2025, entitled “Model Transfer Method, Computer-Readable Storage Medium, Electronic Device and Computer Program Product”, and incorporates the entire contents of that patent application by reference. Technical Field

[0003] This disclosure relates to the field of text retrieval technology, and more specifically, to a model transmission method, a computer-readable storage medium, an electronic device, and a computer program product. Background Technology

[0004] The lifecycle management of a machine learning (ML) model can be divided into several phases: training, simulation, deployment, and inference. In the deployment phase, model deployment can be further divided into two categories: ML Model Loading, which supports loading the ML model onto the inference function for inference use; and ML Model Delivery, which supports transferring the ML model from its host entity to its destination entity. The destination entity may not necessarily use the received model and may further transfer it to the next entity. Currently, 3GPP ML model lifecycle management only supports ML Model Loading but not ML Model Delivery. A solution for how to transfer ML models over the network has not yet been proposed. Summary of the Invention

[0005] This disclosure provides a model transmission method, a computer-readable storage medium, an electronic device, and a computer program product to at least address the problem of how to transmit ML models over a network in the related art.

[0006] According to an embodiment of this disclosure, a model transmission method is provided, applied to a first network element, comprising:

[0007] Send a first model transmission request to the second network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning AI / ML inference requirements, model performance requirements, model transmission information, first model identifier, and model performance satisfaction information;

[0008] The system receives a target model transmission response sent by the second network element based on a first matching result, wherein the first matching result is obtained by the second network element through model matching based on the first model transmission request and model transmission restriction information.

[0009] According to another embodiment of this disclosure, a model transmission method is provided, applied to a second network element, comprising:

[0010] The system receives a first model transmission request sent by a first network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning (AI / ML) inference requirements, model performance requirements, model transmission information, a first model identifier, and model performance satisfaction information;

[0011] Model matching is performed based on the first model transmission request and model transmission restriction information to obtain the first matching result;

[0012] Based on the first matching result, the target model transmission response is sent to the first network element.

[0013] According to yet another embodiment of this disclosure, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to perform the steps in any of the above method embodiments when it is run.

[0014] According to yet another embodiment of this disclosure, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0015] According to yet another embodiment of this disclosure, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in any of the above method embodiments. Attached Figure Description

[0016] Figure 1 is a schematic diagram of the service-oriented management architecture based on related technologies;

[0017] Figure 2 is a flowchart of a model transmission method according to an embodiment of the present disclosure;

[0018] Figure 3 is a flowchart of a model transmission method according to an embodiment of the present disclosure;

[0019] Figure 4 is a flowchart of model transmission according to an embodiment of the present disclosure;

[0020] Figure 5 is a flowchart of a UE request model transmission according to an embodiment of the present disclosure;

[0021] Figure 6 is a flowchart of the model transmission of a UE request according to an embodiment of the present disclosure;

[0022] Figure 7 is a flowchart of a network-initiated model transmission according to an embodiment of the present disclosure;

[0023] Figure 8 is a flowchart of a network-initiated model transmission according to an embodiment of the present disclosure;

[0024] Figure 9 is a flowchart of model training and training data collection triggered by model transmission according to an embodiment of the present disclosure. Detailed Implementation

[0025] The embodiments of this disclosure will be described in detail below with reference to the accompanying drawings and examples.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0027] Figure 1 is a schematic diagram of a service-based management architecture based on related technologies. As shown in Figure 1, the service-based management architecture includes a Business Support System (BSS), a Cross Domain Management Function (CDMF), a Domain Management Function (DMF), and a Network Element (NE). The CDMF manages one or more DMFs. The DMF can manage one or more NEs.

[0028] A business support system (BSS) is oriented towards communication services and provides functions and management services such as billing, settlement, accounting, customer service, sales, network monitoring, communication service lifecycle management, and service intent translation. The BSS can be an operator's operating system or a vertical OT system.

[0029] The cross-domain management function unit, also known as the network management function unit (NMF), can be a network management entity such as a network management system (NMS), a network management service producer (MnS Produoer), a management and network service consumer (MnS Consumer), or a network function management service consumer (NFMS_C). The cross-domain management function unit provides one or more of the following management functions or services: network lifecycle management, network deployment, network fault management, network performance management, network configuration management, network assurance, network optimization, and translation of network intents from communication service providers (Intent-CSPs). The network referred to in the above management functions or services can include one or more network elements or subnetworks, or it can be a network slice. In other words, a network management function unit can be a Network Slice Management Function (NSMF), a Management Data Analytical Function (MDAF), a Self-Organization Network Function (SON), an Intent-Driven Management Service (Intent Driven MnS), or a Close Control Loop Management Service.

[0030] The domain management function unit, also known as the network subnet management function (NSMF) or network element management function unit, can be a wireless automation engine (MBB automation engine, MAE), an element management system (EMS), a network function management service provider (NFMS_P), a network slice subnet management function (NSSMF), a domain management data analysis function (Domain MDAF), a self-organization network function (SON Function), a domain intent management function unit, or a close control loop management service, an MnS producer, an MnS consumer, or other network element management entities. Domain management function units can be classified in the following ways: By network type, they can be divided into: Radio access network (RAN) domain management function units (RAN domain MnF), Core network domain management function units (CN domain MnF), and Transport network domain management function units (TN domain MnF), etc. It should be noted that a domain management function unit can also be a domain network management system, managing one or more of the access network, core network, or transport network. By administrative region, they can be divided into: domain management function units for a specific region, such as the domain management function unit for city A, the domain management function unit for city B, etc.The domain management functional unit provides one or more of the following functions or management services: lifecycle management of subnetworks or network elements, deployment of subnetworks or network elements, fault management of subnetworks or network elements, performance management of subnetworks or network elements, assurance of subnetworks or network elements, optimization functions of subnetworks or network elements, and translation of subnetwork or network element intents (Intent from Network Operator, abbreviated as Intent-NOP). Here, a subnetwork includes one or more network elements. A subnetwork can also include other subnetworks, that is, one or more subnetworks forming a larger subnetwork. A subnetwork can also be a network slice subnetwork.

[0031] A network element is an entity that provides network services. Network elements include core network elements, radio access network elements, and transport network elements. Specifically, core network elements may include, but are not limited to, Access and Mobility Management Function (AMF) entities, Session Management Function (SMF) entities, Policy Control Function (PCF) entities, Network Data Analysis Function (NWDAF) entities, Network Repository Function (NRF) entities, and gateways. Wireless access network elements may include, but are not limited to: various base stations (e.g., Generation Node B (gNB), Evolved Node B (eNB), Central Unit Control Panel (CUCP), Central Unit (CU), Distributed Unit (DU), Central Unit User Panel (CUUP), etc.). In this disclosure, network function (NF) is also referred to as network element (NE). A network element can provide one or more of the following management functions or services: network element lifecycle management, network element deployment, network element fault management, network element performance management, network element assurance, network element optimization functions, and translation of network element intent, etc.

[0032] The fundamental component of Service Based Management Architecture (SBMA) is the Management Service (MnS). MnS is a set of capabilities for managing and orchestrating networks and services. Entities that provide MnS are called MnS producers, and entities that consume MnS are called MnS consumers. Any entity with proper authorization and authentication can consume MnS provided by MnS producers (as shown in the units listed in Figure 1). MnS producers provide their services through standardized service interfaces composed of individually designated MnS components.

[0033] MnS is specified using different independent components. A specific MnS consists of at least two of these components. Currently, three different component types are defined, referred to as MnS component type A, MnS component type B, and MnS component type C.

[0034] MnS component type A is a set of management operations and / or notifications that are independent of the managed entity. These operations and notifications themselves do not involve any information related to the managed network. These operations and notifications are referred to as generic or network-independent. For example, operations that create, read, update, and delete instances of managed objects, where the managed object instance to be operated on is specified only in the signature of the operation, are generic.

[0035] MnS Component Type B refers to management information represented by an information model, representing the managed entity. MnS Component Type B is also known as Network Resource Model (NRM). Examples of MnS Component Type B include network resource models defined in TS 28.622 and TS 28.541.

[0036] MnS component type C contains performance and fault information for the managed entity. Examples of management service component type C include: alarm information as defined in TS 28.532 and TS 28.545; and performance data as defined in TS 28.552, TS 28.554, and TS 32.425.

[0037] 3GPP defines four types of management services (MnSType): ProvMnS, FaultSupervisionMnS, StreamingDataReportingMnS, and FileDataReportingMnS. When MnSType is ProvMnS, MnsCapability includes MANAGEMENT_DATA_CONTROL, FAULT_MANAGEMENT, FILE_MANAGEMENT, NR_PROVISIONING, 5GC_PROVISIONING, NETWORK_SLICING_PROVISIONING, EDGE_COMPUTING_PROVISIONING, AI / ML_MANAGEMENT, MDA, SON_POLICY, RANSC_MANAGEMENT, INTENT_DRIVEN_MANAGEMENT, MNS_REGISTRY_AND_DISCOVERY, COMMUNICATION_SERVICE_ASSURANCE, NSOEU, MSAC_MANAGEMENT, and CCL.

[0038] This embodiment provides a model transmission method operating on the above-described network architecture. Figure 2 is a flowchart of the model transmission method according to an embodiment of this disclosure. As shown in Figure 2, the method is applied to a first network element, and the process includes the following steps:

[0039] Step S202: Send a first model transmission request to the second network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning AI / ML inference requirements, model performance requirements, model transmission information, first model identifier, and model performance satisfaction information;

[0040] In this embodiment of the disclosure, the first network element refers to the ML Model Delivery Consumer, and the second network element refers to the ML Model Delivery Producer. Specifically, it can be any entity in the network, such as cross-domain operation administration and maintenance (OAM), single-domain OAM, gNB, NWDAF, etc.

[0041] The first model identifier mentioned above can be a single model identifier or a collection of multiple model identifiers. That is, the first model step can be: a single ML model ID, multiple ML Model IDs, a combined ML Model ID; an Inference Index ID indicating one or more ML Model IDs, etc. This description also applies to all model identifiers and target models in this disclosure.

[0042] Step S204: Receive the target model transmission response sent by the second network element based on the first matching result, wherein the first matching result is obtained by the second network element through model matching based on the first model transmission request and model transmission restriction information.

[0043] In step S204 above, model matching refers to matching and selecting a model that meets the transmission requirements based on model requirements (such as inference requirements, performance requirements, etc.). Specifically, the selected model's transmission restriction information satisfies the requirements in the transmission information. The model transmission restriction information may include at least one of the following: a model transmittable identifier, a model loadable identifier, or model transmission restrictions.

[0044] Through the above steps S202 to S204, the problem of how to transmit ML models over a network can be solved at least in the related technologies, and model transmission over a network can be realized.

[0045] The embodiments disclosed herein can achieve accurate model matching and efficient transmission, and are particularly suitable for AI / ML (Artificial Intelligence / Machine Learning) applications in 5G network environments, such as autonomous driving, smart healthcare, virtual reality and other services or channel state information CSL compression scenarios. It can quickly obtain the best model according to real-time needs and improve application performance.

[0046] In this embodiment of the disclosure, step S204 may include: if the first matching result is a match with a first target model, receiving a first model transmission response sent by a second network element, wherein the first model transmission response includes at least one of the following: model transmission status, model address, model file, specifically, the model address may be an OAM address; if the first matching result is a match with multiple first target models, receiving a second model transmission response sent by a second network element, wherein the second model transmission response includes a transmittable model identifier, and the first network element may select the model identifier to be transmitted from the transmittable model identifier; if the first matching result is no match with a first target model, the second network element determines to train a new model, interacts with a third network element to train a second target model, and transmits the trained second target model to the first network element, and the first network element receives the second target model transmitted by the second network element; the second network element may also directly inform the first network element that no model that meets the requirements is available, and the first network element receives a third model transmission response sent by the second network element, wherein the third model transmission response includes: no transmittable model indication and / or no transmittable model reason. This mechanism ensures the flexibility and intelligence of model transfer, and can meet the needs by training new models even when model resources are insufficient. It is suitable for large-scale AI / ML service deployments, such as smart cities and industrial automation.

[0047] In one embodiment, after receiving the second model transmission response sent by the second network element, a second model transmission request can be sent to the second network element, wherein the second model transmission request includes a second model identifier; the second network element then obtains the third target model corresponding to the second model identifier and sends it to the first network element for transmission, and the first network element receives the third target model transmitted by the second network element. The second model identifier is similar to the first model identifier described above, and will not be repeated here. This interaction method enables precise selection and transmission of models, and is suitable for application scenarios that require highly customized AI / ML models, such as personalized recommendation systems and domain-specific intelligent analysis.

[0048] In another embodiment, after receiving the second model transmission response from the second network element, the first network element can adjust the first model transmission request to obtain a third model transmission request based on the reason why no model can be transmitted; send the third model transmission request to the second network element; the second network element performs model matching based on the second model transmission request and model transmission restriction information to obtain a fourth target model, and transmits it to the first network element, which then receives the fourth target model sent by the second network element. This dynamic adjustment mechanism ensures that even if the initial model request cannot be satisfied, a suitable model can be re-matched by adjusting parameters, improving the reliability and efficiency of model transmission, and is suitable for real-time updating and optimization scenarios of AI / ML models.

[0049] The AI / ML inference requirements in this disclosure include at least one of the following: AI / ML inference requirements in natural language form, structured AI / ML inference requirements, and parameterized AI / ML inference requirements; and / or model performance requirements include at least one of the following: AI / ML inference name, model performance information, model energy consumption information, model complexity information, model size, and model version; and / or model transmission information includes at least one of the following: transmission type, inference latency, model destination address, model transmission conditions, model transmission method, and model transmission range. The transmission type includes at least one of the following: direct model transmission (direct transmission to the consumer), indirect model transmission (transmission to a third party), direct model loading, and indirect model loading; the model destination address can be either a transmission address or a loading address. These detailed requirements and information parameters ensure the accuracy and efficiency of model transmission, are applicable to various AI / ML applications such as speech recognition, image processing, and data analysis, and allow for customized model transmission strategies based on specific needs, thereby improving the application experience.

[0050] This embodiment also provides a model transmission method operating on the above-described network architecture. Figure 3 is a flowchart of the model transmission method according to an embodiment of this disclosure. As shown in Figure 3, the method is applied to a second network element and includes the following steps:

[0051] Step S302: Receive a first model transmission request sent by the first network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning AI / ML inference requirements, model performance requirements, model transmission information, first model identifier, and model performance satisfaction information;

[0052] Step S304: Perform model matching based on the first model transmission request and model transmission restriction information to obtain the first matching result;

[0053] The aforementioned model transmission restriction information includes at least one of the following: model transmittable identifier, model loadable identifier, and model transmission restriction conditions.

[0054] The second network element checks model transmission restriction information, specifically including the following parameters: a model transmittable identifier, indicating whether the model can be transmitted; a model loadable identifier, indicating whether the model can be loaded; and model transmission restrictions, indicating the restrictions on model transmission, such as the types of entities that can be transmitted (if the model can only be used in the core network, it cannot be transmitted to the base station), the identifiers of entities that can be transmitted (such as base station identifier, core network identifier, slice identifier, subnet identifier), the area where transmission is restricted / allowed, the time where transmission is restricted / allowed, the number of transmissions allowed, and whether multiple transmissions are allowed. Then, based on the first model transmission request, the second network element analyzes the parameters in the first model transmission request, including AI / ML inference requirements, model performance requirements, and model transmission information, to match and select the most suitable ML model. The second network element's model matching may or may not find a match for the target model.

[0055] Step S306: Send the target model transmission response to the first network element according to the first matching result.

[0056] Through the above steps S302 to S304, the problem of how to transmit ML models over a network can be solved at least in the related technologies, and model transmission over a network can be realized.

[0057] As a provider of model resources, the second network element can efficiently respond to the model needs of the first network element through an intelligent matching mechanism. It is suitable for scenarios such as cloud service providers and AI platforms, and can provide users with fast and accurate model services.

[0058] In this embodiment of the disclosure, step S306 may include:

[0059] If the first matching result is a match with a first target model, the first target model is transmitted to the first network element, and a first model transmission response is sent to the first network element. The first model transmission response includes at least one of the following: model transmission status, model address, and model file. That is, the second network element transmits the first target model to the first network element and sends a first model transmission response to the first network element, which includes at least one of the following information: model transmission status (e.g., "completed" or "transmitting"); model address (if the transmission method is a report address); and model file (if the model file is transmitted directly).

[0060] If the first matching result is that multiple first target models are matched, a second model transmission response is sent to the first network element, wherein the second model transmission response includes a list of models that can be transmitted; if the first matching result is that multiple first target models are matched, the second network element will not transmit the model immediately, but will send a second model transmission response to the first network element, wherein the second model transmission response contains a list of all models that can be transmitted, for the first network element to further select.

[0061] If the first matching result is that no first target model is matched, the second target model is transmitted to the first network element. This second target model is trained by the third network element after initial training. Upon receiving the request, according to the model matching strategy, no matching model is found. The second target model is then trained: the second network element decides to train the second target model and notifies the third network element (e.g., an ML Model Training Producer) to begin training. Alternatively, a third model transmission response is sent to the first network element, explaining that no model can be transmitted and the reason why. In another scenario, a third model transmission response is sent to the first network element, including an indication that no model can be transmitted and / or a reason why no model can be transmitted. This response mechanism ensures the transparency and controllability of model transmission and is suitable for scenarios requiring model resource management and scheduling, such as AI / ML model library management and edge computing, effectively improving the utilization efficiency of model resources.

[0062] In one embodiment, after sending a second model transmission response to the first network element, a second model transmission request is received from the first network element. The second model transmission request includes a second model identifier, which is the identifier of one or more models in a list of transmittable models. A third target model corresponding to the second model identifier is then transmitted to the first network element. This interaction method enables precise model selection and transmission, and is suitable for application scenarios requiring highly customized AI / ML models, such as personalized recommendation systems and domain-specific intelligent analysis, ensuring the accuracy and applicability of the models.

[0063] In another embodiment, after sending a third model transmission response to the first network element, a third model transmission request is received from the first network element. This third model transmission request is obtained by adjusting the first model transmission request based on the reason why no model can be transmitted. Model matching is performed based on the second model transmission request and model transmission restriction information to obtain a second matching result. If the second matching result matches a fourth target model, the fourth target model is transmitted to the first network element. This dynamic adjustment mechanism ensures that even if the initial model request cannot be met, a suitable model can be re-matched by adjusting parameters, improving the reliability and efficiency of model transmission. It is suitable for real-time updates and optimization scenarios of AI / ML models, such as online learning and dynamic adjustment of model parameters, and can quickly respond to changing needs.

[0064] In an optional embodiment, the method further includes: creating a model transfer request instance, a model transfer report instance, a model transfer process instance, and a model transfer strategy instance for model loading and / or model transfer; configuring the model transfer request instance, model transfer report instance, model transfer process instance, and model transfer strategy instance according to parameters in the target request, wherein the target request includes at least one of the following: a first model transfer request, a first target model, a second target model, a third target model, a fourth target model, a first model transfer response, a second model transfer request, and a third model transfer request; executing the transfer target request based on the configured model transfer request instance and model transfer strategy instance, and configuring the model transfer report instance and model transfer process instance according to the transfer status. This instantiation and configuration mechanism enables the automation and intelligence of model transfer, is suitable for large-scale AI / ML service deployments, such as smart cities and industrial automation scenarios, and can significantly improve the efficiency and accuracy of model transfer.

[0065] Specifically, this could involve extending the existing ML Model Loading functionality. The three IOCs (Information Item Classes) – MLModelLoadingRequest, MLModelLoadingProcess, and MLModelPolicy – ​​would all be name-contained under the AIMLInferenceFunction. These three IOCs would then be name-contained under the ManagedEntity. Furthermore, the MLModelLoadingRequest, MLModelLoadingProcess, and MLModelPolicy IOCs would be enhanced, and the concept of Loading would be expanded to support ML Model Delivery. For example, the MLModelLoadingRequest IOC could include the model capability requirements, management requirements, and control requirements of the model being delivered.

[0066] In another optional embodiment, the method further includes: creating a model task instance to support model loading and / or model transmission; configuring the model task instance according to parameters in the target request, wherein the target request includes at least one of the following: a first model transmission request, a first target model, a second target model, a third target model, a fourth target model, a first model transmission response, a second model transmission request, and a third model transmission request; transmitting the target request based on the configured model task instance and updating the model task instance according to the transmission status. This task instantiation mechanism enables efficient management and execution of model transmission, and is suitable for batch transmission and update scenarios of AI / ML models, such as batch model updates for a large number of UE accesses, effectively improving the management efficiency of model transmission.

[0067] For example, a second network element receives a first model transmission request from a first network element, requesting the first network element to perform traffic flow prediction. The second network element responds to this request using the previously created MLModelDeliveryRequest instance, MLModelDeliveryReport instance, MLModelDeliveryProcess instance, and MLModelDeliveryPolicy instance. The second network element maps the first model transmission request to the MLModelDeliveryRequest instance, setting its parameters to the gNB's request content. The second network element also checks the MLModelDeliverypolicy instance to ensure that the transmission conforms to the preset policy conditions.

[0068] Specifically, this can involve defining a dedicated Model Transfer IOC (MLModel Job IOC). When a Consumer requests a model transfer, it needs to request the Producer to create an ML Model Job IOC. The model transfer request includes the Consumer's requirements for the model's capabilities, management, and control, among other things.

[0069] Furthermore, the above method also includes: after determining the training of the second target model, sending a model training request to the third network element, wherein the model training request instructs the third network element to collect data, and the model training request includes at least one of the following: data collection source information, data collection control information, and data collection range; receiving a training report sent by the third network element after training the second target model based on the data collected based on the data collection source information, wherein the training report carries the data collection source information involved in training the second target model. This data-driven model training mechanism can ensure the accuracy and timeliness of the model, and is suitable for AI / ML applications that require real-time data updates, such as intelligent transportation and environmental monitoring, and can train models based on the latest data to improve application effectiveness.

[0070] In this embodiment of the disclosure, AI / ML inference requirements include at least one of the following: AI / ML inference requirements in natural language form, structured AI / ML inference requirements, and parameterized AI / ML inference requirements; and / or model performance requirements include at least one of the following: AI / ML inference name, model performance information, model energy consumption information, model complexity information, model size, and model version; and / or model transmission information includes at least one of the following: transmission type, inference latency, model destination address, model transmission conditions, model transmission method, and model transmission range. These detailed requirements and information parameters ensure the accuracy and efficiency of model transmission, are applicable to various AI / ML applications such as CSL compression, speech recognition, image processing, and data analysis, and allow for customized model transmission strategies based on specific needs, thereby improving the application experience.

[0071] The model transmission mechanism disclosed herein not only enables intelligent matching and efficient transmission of AI / ML models, but also dynamically adjusts the model transmission strategy according to actual needs, ensuring the accuracy and applicability of the models. This method has significant advantages in 5G and future network environments, providing fast, accurate, and efficient model services for various AI / ML applications, improving application performance and user experience. Particularly in scenarios such as large-scale data processing, real-time demand response, and resource optimization, this mechanism plays a crucial role in promoting the widespread application and development of AI / ML technology across various fields. Furthermore, by creating and configuring model transmission request instances, model transmission report instances, model transmission process instances, model transmission strategy instances, and model task instances, the automation and intelligence of model transmission can be achieved, reducing the need for manual intervention, improving system scalability and response speed, and providing strong technical support for building intelligent networks and intelligent applications.

[0072] The following example illustrates the embodiments of this disclosure, using the first network element as an ML Model Delivery Consumer and the second network element as an ML Model Delivery Producer.

[0073] The Consumer and Producer in this disclosure, such as ML Model Delivery Consumer and ML Model Delivery Producer, can be any functional entity in Figure 1 (base station, core network element, base station network management, core network management, cross-domain network management, service support system, etc.).

[0074] Figure 4 is a flowchart of model transmission according to an embodiment of the present disclosure. As shown in Figure 4, it includes:

[0075] S401, the ML Model Delivery Consumer sends an ML model transfer request (corresponding to the first model transfer request mentioned above) to the ML Model Delivery Producer. The ML model transfer request contains at least one of the following parameters:

[0076] AI / ML inference requests refer to AI / ML inference requests originating from the Consumer itself or relayed from it. Specifically, these can include: AI / ML inference requests in natural language; and structured AI / ML inference requests, such as structured intents issued in the form of intents, which indicate AI / ML requests.

[0077] Parameterized AI / ML inference requirements specifically include one of the following parameters: a list of model inference capabilities (e.g., AI / MLInferenceNameList, which includes multiple AIML inference names), indicating the inference capabilities that the model should support; and an inference problem (InferenceType), indicating the broad category of problem types, such as energy saving, traffic prediction, fault management, etc.

[0078] Model performance requirements indicate the required performance parameters of the model, which may include one of the following parameters: AIMI Inference Name (AI / MLInferenceName); model performance information, such as accuracy; model energy consumption information, indicating the upper limit of energy consumption for model inference; model complexity information, such as FLOPs (Floating Point Operations per Second), MACs (Multiplication and Accumulation Operations), and MAdds (Million Accumulations and Multiplications per Second). FLOPs refers to the number of floating-point operations performed per second, MACs refers to a combination of one multiplication and one addition operation (usually followed by a multiplication operation), and MAdds refers to millions of multiplication and accumulation operations performed per second; model size; and model version.

[0079] Model transmits information;

[0080] The transmission categories include "direct model transmission" (directly transmitted to the consumer), "indirect model transmission" (transmitted to a third party), "direct model loading" (directly loaded to the consumer), and "indirect model loading" (loaded to a third party).

[0081] Inference latency, such as the upper limit of the expected latency from model transmission to inference execution, or the upper limit of the expected latency from model transmission to generating inference results;

[0082] Model transmission address, such as Consumer address or third-party address, etc.;

[0083] Model transfer conditions, such as time conditions (time point, time window, etc.), model transfer frequency, number of model transfers, model version updates, or a combination of multiple conditions. When the above conditions are met, the Producer should automatically execute the model transfer;

[0084] Transmission models include encrypted / unencrypted transmission; file-based / stream-based transmission; and file address-based reporting.

[0085] The model transmission range can be the type of the object being transmitted (such as base station, network element, slice, subnet, etc.), the list of objects to be transmitted, the area range where the object being transmitted is located, the object being transmitted under specific conditions, or other related objects under specific conditions (such as a PM of the object being transmitted exceeding a certain threshold, such as energy consumption exceeding a certain threshold, requesting transmission or loading of the model for inference).

[0086] Model identifier;

[0087] Model performance satisfaction information can be an indication of whether the model is satisfactory or not, or a satisfaction score, such as a specific score from 0 to 100.

[0088] S402, ML Model Delivery, after receiving an ML model delivery request, the Producer performs requirement matching / model selection. The ML Model Delivery Producer first checks the delivery-related attributes of the ML Model, specifically including the following parameters:

[0089] Model transferability identifier. Indicates whether the model can be transferred;

[0090] The model-loadable flag indicates whether the model can be loaded.

[0091] Model transmission restriction information indicates the restrictions on model transmission, such as the types of entities that are restricted or allowed to be transmitted (if the model can only be used in the core network, it cannot be transmitted to the base station), the identifiers of entities that are restricted or allowed to be transmitted (such as base station identifier, core network identifier), the restricted / allowed transmission area, the restricted / allowed transmission time, the restricted number of transmissions, and whether multiple transmissions are allowed, etc.

[0092] When the Consumer determines the model to be transmitted (i.e., S401 does not contain a model identifier), execute S403-S4024.

[0093] S403, the ML Model Delivery MnS Producer sends an ML model transmission response (corresponding to the first and third model transmission responses mentioned above) to the ML Model Delivery MnS Consumer. The model transmission response contains at least one of the following parameters:

[0094] A list of models that can be transmitted or a list of model identifiers, including the model identifiers that can be transmitted;

[0095] No Transmittable Model Indicator: Indicates that the Producer has no transmittable model.

[0096] There are no reasons why the model cannot be transmitted, such as mismatched requirements or limitations on model transmission (including the reasons for the limitations).

[0097] S404, The ML Model Delivery MnS Consumer sends an ML model transfer request (corresponding to the second model transfer request mentioned above) to the ML Model Delivery MnS Producer. The request includes the following parameter: Model Identifier (ML Model ID).

[0098] When the Producer indicates in S403 that there are no models to transmit, the Consumer may modify the model transmission request based on the given reason and repeat S401.

[0099] When the Producer determines the model to be transmitted or the Consumer indicates the model identifier to be transmitted, execute S405-S406.

[0100] S405, ML Model Delivery MnS Producer executes model delivery. The Producer's model delivery execution includes the following specific schemes:

[0101] Option 1: ML Model Delivery. The MnS Producer creates four Invocation Controls (IOCs): MLModelDeliveryRequest, MLModelDeliveryProcess, MLModelDeliverPolicy, and MLModelDeliveryReport. Based on the received delivery request, each of the three IOCs contains the following parameters:

[0102] The MLModelDeliveryRequest IOC contains AI / ML inference requirements, model performance requirements, model delivery information, and ML task type.

[0103] The MLModelDeliveryProcess IOC contains the model delivery status and the ML task number;

[0104] The MLModelDelivery IOC contains model transfer information;

[0105] The MLModelDeliveryReport IOC contains a list of transferable models, an indication that no transferable models are available, and a reason why no transferable models are available.

[0106] Option 2: ML Model Delivery Producer creates ML Model Job IOCs (MLModel Job IOCs), which include one of the following parameters:

[0107] ML task types (MLJobType): MLModelDelivery, MLModelLoading;

[0108] ML Job ID (MLJobID);

[0109] The model index points to the model identifier;

[0110] Model performance requirements;

[0111] Model transmits information;

[0112] Model transmission status.

[0113] S406, the ML Model Delivery Producer sends an ML model delivery response (corresponding to the first model delivery response mentioned above) to the ML Model Delivery Consumer, which includes the following parameters:

[0114] Model transmission status, including "paused", "completed", and "transmitting";

[0115] The address where the model is located (applicable when the transmission method specified in the Consumer transmission information is to report the model transmission address).

[0116] Figure 5 is a flowchart of a UE request model transmission according to an embodiment of the present disclosure, as shown in Figure 5, including:

[0117] In S501, the UE sends an ML model transmission request (corresponding to the first model transmission request mentioned above) to the gNB, requesting the base station to transmit the model to the UE. The parameters included in the ML model transmission request are the same as those in S401 above;

[0118] In S502, the gNB sends an ML model transmission request to the RAN-OAM, requesting the OAM to transmit the model to the gNB. The parameters in the ML model transmission request are the same as those in S401 above;

[0119] S503, RAN-OAM performs requirement matching / ML model selection, the same as S402 above;

[0120] S504, RAN-OAM sends an ML model transmission response to gNB, as in S403 above;

[0121] S505, gNB sends an ML model transmission request to RAN-OAM, as in S404 above;

[0122] S506, RAN-OAM performs ML model transmission, the same as S405 above;

[0123] S507, RAN-OAM sends an ML model transmission response to gNB, as in S406 above;

[0124] S508, gNB sends an ML model transmission response to UE, which contains the model file or model address.

[0125] Figure 6 is a flowchart of a UE request model transmission according to an embodiment of the present disclosure, as shown in Figure 6, including:

[0126] S601, same as S501 above;

[0127] In S602, the gNB sends an ML model transfer request to the AMF, requesting the transfer of the model. The parameters in the request message are the same as those in S401 above;

[0128] S603, AMF sends an ML model transfer request to NWDAF, requesting the transfer of the model. Same as S401 above;

[0129] S604, NWDAF sends an ML model transfer request to CN-OAM, requesting model transfer. Same as S401 above;

[0130] S605, CN-OAM performs ML model transmission, the same as S405 above;

[0131] S606, CN-OAM sends the ML model transmission response to NWDAF, as in S406 above;

[0132] S607, NWDAF sends the ML model transmission response to gNB, as in S406 above;

[0133] S608, gNB sends ML model transmission response to UE, as in S406 above;

[0134] S609 is the same as S506 above.

[0135] Figure 7 is a flowchart of a network-initiated model transmission according to an embodiment of the present disclosure. As shown in Figure 7, it includes:

[0136] S701, the operator / base station pre-configures the model transmission / automatic model update strategy. When a UE that meets the conditions enters the area, if the pre-configured strategy is satisfied, the base station will send the qualified model to it.

[0137] Model transfer / automatic update strategies, such as model transfer information.

[0138] S702, the UE sends an access request to the base station. The access request contains the UE's AI capability information, including: model performance information, parameters that are the same as the model performance requirements; and UE capability information, such as computing power and supported model categories.

[0139] S703, the base station determines whether the model transmission / model update strategy is satisfied based on the UE model information in the access request and the model information on the base station side;

[0140] S704-S708, The base station sends an ML model transmission request to the OAM / / The OAM performs ML model transmission / / The OAM sends an ML model transmission response to the base station;

[0141] S709, the base station sends an access response to the UE. The access response contains a model address or a model file, and there can be multiple model files.

[0142] Figure 8 is a flowchart of a network-initiated model transmission according to an embodiment of the present disclosure, as shown in Figure 8, including:

[0143] S801, the operator / AMF pre-configures the model transmission / automatic model update policy. When a UE that meets the conditions enters the area, if the policy is satisfied, the core network sends the qualified model to it.

[0144] S802, when registering or making any changes, AMF obtains the contract information from UDM (Unified Data Management), which includes the UEML (User Equipment Machine Learning) model contract information;

[0145] S803, When the PLMN (Public Land Mobile Network) changes or the AMF changes, the PCF (Policy Control Function) updates the model transmission / automatic model update policy.

[0146] S804-S813 are similar to S702-S709 above, and will not be described again here. The model transmission path is CN-OAM->NWDAF->AMF->gNB-UE.

[0147] Figure 9 is a flowchart of model training and training data collection triggered by model transmission according to an embodiment of the present disclosure. As shown in Figure 9, it includes:

[0148] S901, similar to S401 above, the ML model transmission request also includes model performance satisfaction information, which includes at least one of the following parameters:

[0149] Model performance satisfaction index, 1-100, the higher the value, the more satisfied the surface.

[0150] Model performance degradation parameters: A list of model performance parameters that the consumer is dissatisfied with, along with their values. These parameters fall into two categories: first, performance parameters inherent to the model itself, such as accuracy; second, performance parameters of devices or networks affected by the model, such as energy consumption, power consumption, and throughput.

[0151] The reasons for unsatisfactory model performance indicate to the Consumer why they are dissatisfied with the performance.

[0152] S902, no ML model is available at the Producer, so it is decided to train / retrain a new model.

[0153] S903, the ML Model Delivery Producer sends a model training request to the ML Model Training Producer (third network element). The model training request indicates data collection source information, which can be a data source address or data type. In this embodiment, the data collection source includes one of the following:

[0154] UE training data collection, the data source address is the address or identifier of UE Data Collection Producer, and the data category is UE model training data;

[0155] Digital twin simulation data collection, the data source address is the address or identifier of the NDT (Network Digital Twin) Producer, and the data category is NDT simulation data;

[0156] Generative AI data collection, with the data source address being the address or identifier of GenAI Producer, and the data category being GenAI (Generative AI) data.

[0157] S904, Collect data from the UE Data Collection Producer. The data collection request includes the following information:

[0158] Data collection control information includes the time and frequency of data collection, the amount of data collected from each UE, the data collection mode (such as collection from UE one by one, collection from batch UEs, continuous collection or non-continuous collection, etc.), and the data collection method (Trace, MDT (Measurement Data Transmission), QoE (Quality of Experience), etc.).

[0159] The scope of data collection includes: data source scope information, such as the number of UEs for which data is collected, the region of the UEs, the status of the UEs (Idle, active, etc.), the UE category, etc.; data scope information, such as the name of the data (PM / KPI), the data category (standardized / non-standardized), etc.

[0160] S905, collect data from the NDT Producer. The data collection request includes the following information:

[0161] Data collection control information includes the time and frequency of data collection, the amount of data collected from each NDT, the data collection mode (such as continuous or non-continuous collection), and the data collection method (Trace, MDT, QoE, etc.).

[0162] The scope of data collection includes: data source scope information, such as the number of NDT components for data collection, the status of the NDT (running, idle, etc.), and the type of NDT; and data scope information, such as the name of the data (PM / KPI) and the data type (standardized / non-standardized / signaling / business data / measurement data).

[0163] S906, Collect data from GenAI Producer. The data collection request includes the following information:

[0164] Data collection control information includes the time, frequency, amount of data collected, and data collection mode.

[0165] The scope of data collection includes:

[0166] Data source range: one or more GenAI Producers;

[0167] Data scope information, such as the data name (Performance Measurement / Key Performance Indicator, PM / KPI), data category (standardized / non-standardized / signaling / business data / measurement data), etc.

[0168] S907, the ML Training Producer sends a training report to the ML Model Delivery Producer, which includes the following parameters:

[0169] The data source information involved in this model training is the same as that in S903 above;

[0170] The model data source identifier indicates that the training data source contains simulation data or data generated by GenAI.

[0171] The model transmission process from S908 to S909, as shown in Figure 4, will not be described in detail here.

[0172] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this disclosure.

[0173] This embodiment also provides a model transmission device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated. Applied to a first network element, the device includes:

[0174] The first sending module is configured to send a first model transmission request to the second network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning (AI / ML) inference requirements, model performance requirements, model transmission information, first model identifier, and model performance satisfaction information;

[0175] The first receiving module is configured to receive the target model transmission response sent by the second network element based on the first matching result, wherein the first matching result is obtained by the second network element through model matching based on the first model transmission request and model transmission restriction information.

[0176] This embodiment also provides a model transmission device, which should be configured as a second network element. The device includes:

[0177] The second receiving module is configured to receive a first model transmission request sent by the first network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning (AI / ML) inference requirements, model performance requirements, model transmission information, first model identifier, and model performance satisfaction information;

[0178] The matching module is configured to perform model matching based on the first model transmission request and model transmission restriction information to obtain a first matching result.

[0179] The second sending module is configured to send a target model transmission response to the first network element based on the first matching result.

[0180] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0181] Embodiments of this disclosure also provide a computer-readable storage medium storing a computer program configured to perform the steps in any of the above method embodiments when executed.

[0182] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0183] Embodiments of this disclosure also provide an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0184] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0185] Specific examples in this embodiment can be found in the examples described in the above embodiments and exemplary implementations, and will not be repeated here.

[0186] It is obvious to those skilled in the art that the modules or steps of this disclosure described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this disclosure is not limited to any particular combination of hardware and software.

[0187] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A model transmission method, applied to a first network element, comprising: Send a first model transmission request to the second network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning AI / ML inference requirements, model performance requirements, model transmission information, first model identifier, and model performance satisfaction information; The system receives a target model transmission response sent by the second network element based on a first matching result, wherein the first matching result is obtained by the second network element through model matching based on the first model transmission request and model transmission restriction information.

2. The method according to claim 1, wherein, Receiving the target model transmission response sent by the second network element based on the first matching result includes: If the first matching result is a match to a first target model, the first model transmission response sent by the second network element is received, wherein the first model transmission response includes at least one of the following: model transmission status, model address, and model file; If the first matching result is a match for multiple first target models, the second model transmission response sent by the second network element is received, wherein the second model transmission response includes a model identifier that can be transmitted; If the first matching result is that no first target model is matched, the second target model transmitted by the second network element is received, wherein the second target model is obtained by the third network element after the second network element determines the training; or, the third model transmission response sent by the second network element is received, wherein the third model transmission response includes at least one of the following: no transmission model indication, no transmission model reason.

3. The method according to claim 2, wherein, After receiving the second model transmission response sent by the second network element, the method further includes: Send a second model transmission request to the second network element, wherein the second model transmission request includes a second model identifier; The third target model transmitted by the second network element is received, wherein the third target model is the model corresponding to the second model identifier.

4. The method according to claim 2, wherein, After receiving the third model transmission response sent by the second network element, the method further includes: Based on the reason why no model can be transmitted, the first model transmission request is adjusted to obtain the third model transmission request; Send the third model transmission request to the second network element; The second network element receives a fourth target model sent by the second network element, wherein the fourth target model is obtained by the second network element through model matching based on the third model transmission request and model transmission restriction information.

5. The method according to any one of claims 1 to 4, wherein, The AI / ML inference requirements include at least one of the following: AI / ML inference requirements in natural language form, structured AI / ML inference requirements, parameterized AI / ML inference requirements; and / or The model performance requirements include at least one of the following: AI / MI inference name, model performance information, model energy consumption information, model complexity information, model size, model version; and / or The model transmission information includes at least one of the following: transmission type, inference latency, model destination address, model transmission conditions, model transmission method, and model transmission range.

6. A model transmission method, applied to a second network element, comprising: The system receives a first model transmission request sent by a first network element, wherein the first model transmission request includes at least one of the following: artificial intelligence / machine learning (AI / ML) inference requirements, model performance requirements, model transmission information, a first model identifier, and model performance satisfaction information; Model matching is performed based on the first model transmission request and model transmission restriction information to obtain the first matching result; Based on the first matching result, the target model transmission response is sent to the first network element.

7. The method according to claim 6, wherein, The model transmission restriction information includes at least one of the following: model transmittable identifier, model loadable identifier, and model transmission restriction conditions.

8. The method according to claim 6, wherein, Sending the target model transmission response to the first network element based on the first matching result includes: If the first matching result is a match with a first target model, the first target model is transmitted to the first network element, and a first model transmission response is sent to the first network element, wherein the first model transmission response includes at least one of the following: model transmission status, model address, and model file; If the first matching result is that multiple first target models are matched, a second model transmission response is sent to the first network element, wherein the second model transmission response includes: a list of models that can be transmitted; If the first matching result is that no first target model is matched, the second target model is transmitted to the first network element, wherein the second target model is obtained by training by the third network element after the determination training; or, a third model transmission response is sent to the first network element, wherein the third model transmission response includes at least one of the following: no model to transmit, reason for no model to transmit.

9. The method according to claim 6, wherein, After sending the second model transmission response to the first network element, the method further includes: The system receives a second model transmission request sent by the first network element, wherein the second model transmission request includes a second model identifier, which is an identifier of one or more models in a list of models that can be transmitted. The third target model corresponding to the second model identifier is transmitted to the first network element.

10. The method according to claim 6, wherein, After sending the third model transmission response to the first network element, the method further includes: The third model transmission request is received from the first network element, wherein the third model transmission request is obtained by the first network element by adjusting the first model transmission request based on the reason that there is no transmission model; Model matching is performed based on the third model transmission request and model transmission restriction information to obtain a second matching result; If the second matching result matches the fourth target model, the fourth target model is transmitted to the first network element.

11. The method according to claim 8, wherein, The method further includes: Create a model transfer request instance, a model transfer report instance, a model transfer process instance, and a model transfer strategy instance for model loading and / or model transfer; The model transmission request instance, the model transmission report instance, the model transmission process instance, and the model transmission strategy instance are configured according to the parameters in the target request, wherein the target request includes at least one of the following: the first model transmission request, the first target model, the second target model, the third target model, the fourth target model, the first model transmission response, the second model transmission request, and the third model transmission request.

12. The method according to claim 8, wherein, The method further includes: Create model task instances to support model loading and / or model transfer; The model task instance is configured according to the parameters in the target request, wherein the target request includes at least one of the following: the first model transmission request, the first target model, the second target model, the third target model, the fourth target model, the first model transmission response, the second model transmission request, and the third model transmission request.

13. The method according to claim 8, wherein, The method further includes: After determining the training of the second target model, a model training request is sent to the third network element, wherein the model training request is used to instruct the third network element to collect data, and the model training request includes at least one of the following: data collection source information, data collection control information, and data collection range; The system receives a training report sent by the third network element, wherein the training report carries information about the data collection sources involved in training the second target model.

14. The method according to any one of claims 6 to 13, wherein, The AI / ML inference requirements include at least one of the following: AI / ML inference requirements in natural language form, structured AI / ML inference requirements, parameterized AI / ML inference requirements; and / or The model performance requirements include at least one of the following: AI / MI inference name, model performance information, model energy consumption information, model complexity information, model size, model version; and / or The model transmission information includes at least one of the following: transmission type, inference latency, model destination address, model transmission conditions, model transmission method, and model transmission range.

15. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the steps of the method according to any one of claims 1 to 5, 6 to 14.

16. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the method described in any one of claims 1 to 5, 6 to 14.

Images