Training a model in a network

Abstract

Example embodiments relate to methods for training one or more models in a network, and apparatus, systems, non-transitory computer-readable media, chips, and computer program products associated with said methods. A method implemented at a first apparatus is provided. In this method, the first apparatus receives an instruction to perform a model update for one or more models associated with one or more first apparatuses, wherein the first apparatus is one of the first or more apparatuses. The first apparatus then sends a first notification to notify that the model update for the one or more models has been completed. In this way, personalized local models can be supported, and the network entities used to train the models can be dynamically selected; furthermore, the negative impact of data heterogeneity can be mitigated.

Description

Technical Field

[0001] Exemplary embodiments of this disclosure generally relate to the field of communications, and more particularly to methods, apparatus, systems, non-transitory computer-readable media, chips, and computer program products for communications. Background Technology

[0002] In some scenarios, model training for multiple network entities (e.g., artificial intelligence (AI) model training) is required. For example, in an application, a set of network entities, such as user equipment (UE), servers, network functions, radio access network (RAN) nodes, etc., each has a local model and also has (or hosts) a private dataset. The local models may be heterogeneous, meaning they have different structures. In the application, each of these models will be trained using a collection of private datasets without requiring the private datasets to leave their hosting network entity. For example, a scheme for training models for multiple entities is to use federated learning. The goal of federated learning is to train a general model for all network entities. In federated learning, multiple network entities (e.g., UE, servers, network functions, RAN nodes, etc., or combinations thereof) jointly train a general AI model using their private datasets in multiple iterations. During iterations, each of these network entities sends a local version of the model to an aggregator node. The local version of the model sent from the network entity to the aggregator node is obtained by the network entity through local training (i.e., training the model based on, for example, random initialization of the model or a global version of the pattern using the network entity's private dataset). Summary of the Invention

[0003] Typically, the exemplary embodiments of this disclosure provide a scheme for training one or more models in a network.

[0004] It should be understood that the summary portion is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description.

[0005] In a first aspect, a method implemented at a first device is provided. In this method, the first device receives an instruction to perform a model update of one or more models associated with one or more first devices, wherein the first device is one of the first or more first devices. The first device sends a first notification to notify that the model update of the one or more models has been completed. In this manner, personalized local models can be supported, and entities for training the model can be dynamically selected based on the network, thus providing a large selection space and facilitating model training.

[0006] In some embodiments, the indication includes at least one of the following: a list of model identification information used to identify one or more models; version information of one or more models, wherein each of the one or more models has corresponding version information; or one or more status indications, wherein each of the one or more status indications indicates the status of a model among the one or more models, and the status of the model includes frozen or unfrozen. In this way, the entity that obtains the indication can update the model based on the indication in iterations, thereby dynamically selecting the entity used to train the model in each iteration.

[0007] In some embodiments, the indication is carried on a first message from a second device, the first message also including values ​​of model parameters for one or more models. In this way, the indication can be given along with the values ​​of the model parameters to notify the model to update.

[0008] In some embodiments, the indication is carried on a first message from a second device, and the method further includes: obtaining values ​​of model parameters for one or more models at the first device, wherein the values ​​of the model parameters are pre-configured at the first device. Therefore, the values ​​of the model parameters can be obtained in a pre-configured manner.

[0009] In some embodiments, the second device is part of the data plane of the core network, or the second device is connected to the data plane of the core network. This allows for flexible implementation of the second device.

[0010] In some embodiments, the instruction is transmitted from the second device to the first device via a data plane function. In this way, the instruction can be transmitted in a suitable manner, improving transmission efficiency.

[0011] In some embodiments, the indication originates from a third device, and the method further includes receiving values ​​of model parameters for one or more models from a second device at a first device. In this manner, the indication and values ​​of the model parameters can be sent separately as needed, improving transmission efficiency.

[0012] In some embodiments, the third device is part of the control plane of the core network. Therefore, model updates can be instructed via the control plane to improve efficiency.

[0013] In some embodiments, at least one of the following: an instruction is sent from a third device to a first device; or an instruction is sent from a third device to a first device via a control surface function. In this way, instructions are sent in a flexible manner.

[0014] In some embodiments, the method further includes updating at least one of one or more models at a first device. Therefore, a local model can be trained in a personalized manner.

[0015] In some embodiments, the first notification further includes at least one of the following: a list of model identification information used to identify one or more models; or one or more status indicators, wherein each of the one or more status indicators indicates the status of a model in one or more models, the model status including frozen or unfrozen. This is therefore beneficial for model training in the next iteration.

[0016] In some embodiments, the frozen model is a converged model; or the unfrozen model is a non-converged model. This approach facilitates model training based on the model state in the next iteration.

[0017] In some embodiments, at least one of the following occurs: a first notification is sent from a first device to a third device; or a first notification is sent from a first device to a third device via a control plane function. In this manner, the third device can manage, control, or influence control plane decisions, such as decisions regarding traffic routing, to facilitate model training in the next iteration.

[0018] In some embodiments, the method further includes receiving at a first device a second message notifying the first device that it has been selected to perform a model update. In this way, entities for training the model can be dynamically selected based on the network in each iteration, thus providing a large selection space and improving training efficiency.

[0019] In some embodiments, the method further includes sending an acknowledgment (ACK) at the first device to confirm the selection. In this way, the first device can decide whether to agree to the selection based on its own conditions, which helps to improve training efficiency.

[0020] In some embodiments, the second message originates from a fourth device, and a positive ACK is sent to the fourth device, which is used to perform policy control functions; or the second message originates from a fifth device, and a positive ACK is sent to the fifth device, which is a network controller. In this way, the selection of the first device can be notified and confirmed in a flexible manner.

[0021] In some embodiments, the method further includes: receiving a third message at a first device from a third device, the third message being used to notify the first device to send at least one latest version of one or more models obtained through model updates; and sending at least one latest version at the first device. In this way, the first device can know which device the latest version(s) of the models(s) can be sent to, and therefore, the entity for storing at least one latest version can be dynamically selected in each iteration.

[0022] In some embodiments, at least one of the following is true: the third message is received from a third device via a control plane function; or at least one latest version is transmitted via a data plane function, wherein the data plane function is configured to transmit at least one latest version to a sixth device among the second or more devices. In this manner, the third message can be transmitted using a suitable method, improving transmission efficiency.

[0023] In some embodiments, the sixth device is the second device. In this way, the same or different devices can be used to help train the model in each iteration.

[0024] In some embodiments, the control plane functions include at least one of the following: network exposure function (NEF), policy control function (PCF), network storage function (NSF), network controller (NWC), path management function (PMF), or access and mobility management function (AMF). In this way, the control plane functions of the scheme in this embodiment can be flexibly implemented.

[0025] In some embodiments, the third message further includes model identification information for identifying at least one model associated with at least one latest version. In this manner, the first device can be notified to send the latest version of a portion of the at least one model, thereby improving transmission efficiency.

[0026] In some embodiments, the first device is one of a terminal device, a server, or a network function. In this way, the entity used for model training can be diverse.

[0027] In a second aspect, a method implemented at a second device is provided. In this method, the second device receives a fourth message from a third device, the fourth message indicating whether additional training data should be used by the second device to perform model updates of one or more models associated with a first or more devices, wherein the second device is one of a second or more devices, and the additional training data differs from the training data used at the first or more devices to perform the model updates. The second device updates one or more models based on the received fourth message. In this manner, personalized local models can be supported, entities used for training the models can be dynamically selected, and the negative impacts of data heterogeneity can be mitigated.

[0028] In some embodiments, the fourth message includes the following: a list of model identification information used to identify one or more models; version information of one or more models, wherein each of the one or more models has corresponding version information; and one or more status indicators, wherein each of the one or more status indicators indicates the status of a model among the one or more models, the model status including frozen or unfrozen. In this way, the negative impact of data heterogeneity in private datasets can be mitigated.

[0029] In some embodiments, the fourth message also indicates where to obtain the additional training data and / or how to select the additional training data if it should be used. In this way, the second device does not need to store the additional training data locally, thus saving storage space.

[0030] In some embodiments, updating one or more models includes: using the additional training data to update at least one model among the one or more models that has an indication of an unfrozen state, where additional training data should be used by a second device to perform a model update. In this way, the negative impact of data heterogeneity in private datasets can be mitigated.

[0031] In some embodiments, the method further includes: sending an instruction from a second device to a first device among a first or more devices, the instruction instructing the first device to perform model updates for one or more models. In this manner, the instruction can be sent to dynamically selected entities for training the model.

[0032] In some embodiments, the indication is carried on a first message, and the first message also includes values ​​of model parameters for one or more models. In this way, the indication can be given along with the values ​​of the model parameters to notify the model to update.

[0033] In some embodiments, the indication includes the following: a list of model identification information used to identify one or more models; version information of one or more models, wherein each of the one or more models has corresponding version information; and one or more status indicators, wherein each of the one or more status indicators indicates the status of a model among the one or more models, and the model status includes frozen or unfrozen. In this way, personalized local models can be supported, and entities used for training the model can be dynamically selected in each iteration.

[0034] In some embodiments, the frozen model is a converged model; or the unfrozen model is a non-converged model. This approach facilitates model training based on the model state in the next iteration.

[0035] In some embodiments, the method further includes: receiving, at a second device, a second notification from a third device that the second device is selected to receive at least one latest version of one or more models from a first device, wherein the at least one latest version is obtained through a model update performed by the first device. In this manner, the second device selected to receive at least one latest version of one or more models can be dynamically selected for each iteration.

[0036] In some embodiments, the second notification further includes model identification information for identifying at least one model associated with at least one latest version. In this manner, receiving at least one latest version of one or more models is configurable.

[0037] In some embodiments, the method further includes: receiving configuration at a second device from a third device, the configuration for the second device to obtain one or more latest versions from another device among a second or more devices, wherein the one or more latest versions are versions of one or more models that do not include at least one latest version. In this way, the second device can obtain one or more latest versions from another device and is configurable.

[0038] In some embodiments, the first device is one of a terminal device, a server, or a network function. In this way, the entity used for model training can be diverse.

[0039] In some embodiments, the second device is part of the data plane of the core network; or the second device is connected to the data plane of the core network. In this way, the second device is implemented in a flexible manner.

[0040] In some embodiments, the third device is part of the control plane of the core network. Therefore, model updates can be instructed via the control plane to improve efficiency.

[0041] In a third aspect, a method implemented at a third device is provided. In this method, the third device sends a fourth message to a second device, indicating whether additional training data should be used by the second device to perform model updates of one or more models associated with a first or more devices, wherein the second device is one of a second or more devices, and the additional training data differs from the training data used at the first or more devices to perform the model updates. In this manner, personalized local models can be supported, and network entities used for training the models can be dynamically selected. Furthermore, the negative impacts of data heterogeneity can be mitigated.

[0042] In some embodiments, the method further includes: at a third means, determining one of the first or more means as a first means for performing model updates. In this way, network entities for training the model can be dynamically selected.

[0043] In some embodiments, the method further includes: at the third device, determining the second device by selecting an application location associated with the second device at at least one application location. In this way, the second device can be selected based on the application location to improve training efficiency.

[0044] In some embodiments, the method further includes sending a configuration from a third device to a second device, the configuration being used by the second device to obtain version information of one or more models from another device among the second or more devices. In this way, the second device can obtain one or more latest versions from another device, and this is configurable.

[0045] In some embodiments, the method further includes: sending an instruction from a third device to a first device among a first or more devices, the instruction instructing the first device to perform a model update. Entities used for training the model can be dynamically indicated and selected.

[0046] In some embodiments, the indication includes at least one of the following: a list of model identification information used to identify one or more models; version information of one or more models, wherein each of the one or more models has corresponding version information; or one or more status indications, wherein each of the one or more status indications indicates the status of a model among the one or more models, and the status of the model includes frozen or unfrozen. In this way, the entity that obtains the indication can update the model based on the indication in iterations, thereby dynamically selecting the entity used to train the model in each iteration.

[0047] In some embodiments, the instruction is sent from a third device to a first device, or the instruction is sent from a third device to a first device via a control plane function. In this way, the instruction can be sent in a suitable manner, improving transmission efficiency.

[0048] In some embodiments, the fourth message includes at least one of the following: a list of model identification information used to identify one or more models; version information of one or more models, wherein each of the one or more models has corresponding version information; or one or more status indicators, wherein each of the one or more status indicators indicates the status of a model in one or more models, and the status of a model includes frozen or unfrozen.

[0049] In some embodiments, the frozen model is a converged model; or the unfrozen model is a non-converged model. This approach facilitates model training based on the model state in the next iteration.

[0050] In some embodiments, the fourth message also indicates where to obtain the additional training data and / or how to select the additional training data if it should be used. In this way, the second device does not need to store the additional training data locally, thus saving storage space.

[0051] In some embodiments, the method further includes: receiving, at a third device, a first notification from a first device among a first set of devices, the first notification notifying that a model update of one or more models performed by the first device has been completed; and, based on the first notification, determining, at the third device, a sixth device from a second set of devices for receiving at least one latest version of one or more models, wherein the at least one latest version is obtained by performing a model update by the first device, and the latest version corresponds to the updated model. In this manner, entities for training the model can be dynamically selected according to the network, thus providing a large selection space and facilitating model training.

[0052] In some embodiments, the method further includes sending a second notification from a third device to a sixth device that the sixth device is selected to receive at least one latest version from the first device. In this manner, the device selected to receive at least one latest version of one or more models can be dynamically selected and indicated for each iteration.

[0053] In some embodiments, the second notification further includes model identification information for identifying at least one model associated with at least one latest version. In this way, the first device can be notified to send the latest version of a portion of the at least one model, thereby improving transmission efficiency and dynamically indicating the model identification information.

[0054] In some embodiments, the method further includes sending a configuration at a third device to a sixth device for the sixth device to obtain one or more latest versions from another of the second or more devices, wherein the one or more latest versions are versions of one or more models that do not include at least one latest version. In this way, the sixth device can obtain one or more latest versions from another device and is configurable.

[0055] In some embodiments, the method further includes: sending a third message from a third device to a first device among the first one or more devices, the third message notifying the first device to transmit at least one latest version of one or more models obtained through model updates to a sixth device among the second one or more devices. In this way, the first device can know which device the latest version(s) of the models(s) can be sent to, and therefore, the entity for storing at least one latest version can be dynamically selected in each iteration.

[0056] In some embodiments, the sixth device is the second device. In this way, the same or different devices can be used to help train the model in each iteration.

[0057] In some embodiments, the first device is one of a terminal device, a server, or a network function. In this way, the entity used for model training can be diverse.

[0058] In some embodiments, the second device is part of the data plane of the core network; or the second device is connected to the data plane of the core network. In this way, the second device is implemented in a flexible manner.

[0059] In some embodiments, the third device is part of the control plane of the core network. Therefore, model updates can be instructed via the control plane to improve efficiency.

[0060] In a fourth aspect, a method implemented at a third device is provided. In this method, the third device sends a fifth message to a fourth device requesting influence on traffic routing, wherein influencing traffic routing is used to select a second device from a second or more devices and at least one candidate device from the first or more devices; the second device and the at least one candidate device jointly perform model updates of one or more models associated with the first or more devices, the fourth device being configured to perform policy control functions; and a sixth message is received from the fifth device instructing the second device and the at least one candidate device, wherein the fifth device is configured as a network controller. In this manner, personalized local models can be supported, and entities for training the model can be dynamically selected based on the network, thus providing a large selection space and facilitating model training.

[0061] In some embodiments, the method further includes: at a third device, determining the second device by selecting an application location associated with the second device at at least one application location. In this way, a device corresponding to a suitable application location can be selected as the second device, and training efficiency can be improved.

[0062] In some embodiments, the fifth message includes at least one of the following: data traffic information identifying data traffic of at least one application associated with a second or more devices, wherein the data traffic will be routed during traffic routing; information of the first or more devices; information of at least one location of the at least one application; information of one or more requests associated with traffic routing; or traffic filtering information associated with the data traffic. In this way, it is beneficial for determining appropriate traffic routes during the model training process.

[0063] In some embodiments, the method further includes sending feedback from the third device to the fifth device in response to the sixth message, wherein the feedback indicates an affirmative acknowledgment (ACK) or a negative ACK for the selection of an application location associated with the second device. In this manner, the third device can manage / control the second device.

[0064] In some embodiments, the method further includes: at a third device, identifying a candidate device as a first device among at least one candidate device, wherein information identifying the first device is included in the positive ACK. In this manner, the third device can manage / control the first device.

[0065] In some embodiments, the first device is one of a terminal device, a server, or a network function. In this way, the entity used for model training can be diverse.

[0066] In some embodiments, the second device is part of the data plane of the core network; or the second device is connected to the data plane of the core network. In this way, the second device is implemented in a flexible manner.

[0067] In some embodiments, the third device is part of the control plane of the core network. Therefore, model updates can be instructed via the control plane to improve efficiency.

[0068] In a fifth aspect, a method implemented at a fourth device is provided. In this method, the fourth device receives from a third device a fifth message requesting influence on traffic routing, wherein influence on traffic routing is used to select a second device from a second or more devices and to select at least one candidate device from the first or more devices; the second device and the at least one candidate device jointly perform model updates of one or more models associated with the first or more devices; the fourth device is configured to perform policy control functions; and at least one policy regarding influence on traffic routing is generated based on the fifth message. In this manner, personalized local models can be supported, and entities for training the model can be dynamically selected according to the network, thus providing a large selection space and facilitating model training.

[0069] In some embodiments, the method further includes at least one of the following: sending at least one policy from a fourth device to a fifth device, the fifth device being configured as a network controller for determining a second device and at least one candidate device; or determining a second device and at least one candidate device based on at least one policy at the fourth device. In this manner, at least one policy can be generated and executed by the same device, or it can be flexibly generated and executed by different devices.

[0070] In some embodiments, at least one policy includes at least one of the following: data traffic information to be routed in traffic routing; information on at least one location of at least one application, wherein one of the at least one application is associated with a second device; information on one or more requirements associated with traffic routing; or information on the first or more devices. In this way, at least one policy is configurable and may include various information depending on the requirements(s).

[0071] In some embodiments, the method further includes sending a sixth message from a fourth device to a third device, indicating the second device and at least one candidate device. In this manner, the third device can manage / control the second device.

[0072] In some embodiments, the sixth message indicates a data plane management event, which is associated with the selection of an application location associated with the second device in at least one application location. In this way, the second device can be selected based on the application location.

[0073] In some embodiments, the method further includes receiving feedback from the third device at the fourth device in response to the sixth message, wherein the feedback indicates an affirmative acknowledgment (ACK) for selection of confirmation or a negative ACK for selection of rejection. In this manner, the third device can manage / control the second device.

[0074] In some embodiments, the feedback is an affirmative ACK, and the affirmative ACK further includes information identifying the candidate device as a first device among at least one candidate device. In this way, the third device can manage / control the first device.

[0075] In some embodiments, the method further includes sending a second message from the fourth device to the first device to notify the first device that it has been selected to perform a model update. In this way, the first device can determine whether to agree to the selection based on its conditions, which helps improve training efficiency.

[0076] In some embodiments, the method further includes receiving, at the fourth device, a positive ACK acknowledging the execution of the model update or a negative ACK rejecting the execution of the model update from the first device. In this way, the first device can notify whether it agrees to the selection based on its conditions, allowing the fourth device to further determine whether to select a new first device.

[0077] In some embodiments, the method further includes sending traffic routing information indicating traffic routing at a fourth device to a seventh device for selecting a data plane path for traffic routing. In this manner, a suitable data plane path is selected so that information can be transmitted via the data plane.

[0078] In some embodiments, the method further includes at least one of: receiving, at the fourth device, a third notification from the seventh device that a data plane path has been configured for traffic routing; or receiving, at the fourth device, a response from the seventh device to the transmission of traffic routing information, wherein the response includes the third notification that a data plane path has been configured for traffic routing configuration. The implementation of the notification is flexible.

[0079] In some embodiments, the traffic routing information includes at least one of the following: information identifying an application location associated with the second device in at least one application location; information identifying a first device, wherein the first device is determined among at least one candidate device; information associated with at least one policy; and information about the data plane path for traffic routing. In this way, the fourth device can know the configured traffic routing path.

[0080] In some embodiments, the first device is one of a terminal device, a server, or a network function. In this way, the entity used for model training can be diverse.

[0081] In some embodiments, one of the second or more devices is part of the data plane of the core network; or the device of the second or more devices is connected to the data plane of the core network. In this way, the second devices are implemented in a flexible manner.

[0082] In some embodiments, the third device is part of the control plane of the core network. Therefore, model updates can be instructed via the control plane to improve efficiency.

[0083] In a sixth aspect, a method implemented on a fifth device is provided. In this method, the fifth device receives from a fourth device at least one policy influencing traffic routing, wherein influencing traffic routing is used to select a second device from a second or more devices and to select at least one candidate device from the first or more devices; the second device and the at least one candidate device jointly perform model updates of one or more models associated with the first or more devices, the fourth device is configured to perform policy control functions, and the fifth device is configured as a network controller; based on the at least one policy, the second device and the at least one candidate device are selected; a sixth message is sent to a third device to instruct the second device and the at least one candidate device, wherein the third device is configured to control the second or more devices. In this manner, personalized local models can be supported, and entities for training the model can be dynamically selected according to the network, thus providing a large selection space and facilitating model training.

[0084] In some embodiments, at least one policy includes at least one of the following: information about data traffic to be routed in traffic routing; information about at least one location of at least one application, wherein one of the at least one applications is associated with a second device; information about one or more requirements associated with traffic routing; or information about the first or more devices. In this way, at least one policy is configurable and may include various information depending on the requirements(s).

[0085] In some embodiments, the sixth message indicates a data plane management event, which is associated with the selection of an application location associated with the second device in at least one application location. In this way, the second device can be selected based on the application location.

[0086] In some embodiments, the method further includes receiving feedback from the third device at the fifth device in response to the sixth message, wherein the feedback indicates an affirmative acknowledgment (ACK) for the selection or a negative ACK for the rejection of the selection. In this manner, the third device can manage / control the second device.

[0087] In some embodiments, the feedback is an affirmative ACK, and the affirmative ACK further includes information identifying the candidate device as a first device among at least one candidate device. In this way, the third device can manage / control the fifth device.

[0088] In some embodiments, the method further includes sending a second message from the fifth device to the first device to notify the first device that it has been selected to perform a model update. In this way, the first device can determine whether to agree to the selection based on its conditions, which helps improve training efficiency.

[0089] In some embodiments, the method further includes receiving, at the fifth device, a positive ACK acknowledging the execution of the model update or a negative ACK rejecting the execution of the model update from the first device. In this manner, the first device can notify whether it agrees to the selection based on its conditions, allowing the fifth device to further determine whether to select a new first device.

[0090] In some embodiments, the method further includes: sending traffic routing information indicating traffic routing from a fifth device to a seventh device for selecting a data plane path for traffic routing. In this manner, a suitable data plane path is selected so that information can be transmitted via the data plane.

[0091] In some embodiments, the traffic routing information includes at least one of the following: information identifying an application location associated with the second device in at least one application location; information identifying a first device, wherein the first device is determined among at least one candidate device; information associated with at least one policy; and information about the data plane path for traffic routing. In this way, the fifth device can know the configured traffic routing path.

[0092] In some embodiments, the method further includes at least one of: receiving, at the fifth device, a third notification from the seventh device that a data plane path has been configured for traffic routing; or receiving, at the fifth device, a response from the seventh device to the transmission of traffic routing information, wherein the response includes the third notification that a data plane path has been configured for traffic routing. The implementation of the notification is flexible.

[0093] In some embodiments, the fifth and fourth devices are integrated into a single network entity. This simplifies the transmission signaling.

[0094] In some embodiments, the first device is one of a terminal device, a server, or a network function. In this way, the entity used for model training can be diverse.

[0095] In some embodiments, one of the second or more devices is part of the data plane of the core network; or the device of the second or more devices is connected to the data plane of the core network. In this way, the second devices are implemented in a flexible manner.

[0096] In some embodiments, the third device is part of the control plane of the core network. Therefore, model updates can be instructed via the control plane to improve efficiency.

[0097] In a seventh aspect, a first apparatus is provided. The first apparatus includes a transceiver and a processor communicatively coupled to the transceiver. The processor is configured to: receive an instruction instructing the first apparatus to perform a model update of one or more models associated with a first or more apparatus, wherein the first apparatus is one of the first or more apparatuses; and send a first notification notifying that the model update of the one or more models has been completed.

[0098] In an eighth aspect, a second apparatus is provided. The second apparatus includes a transceiver and a processor communicatively coupled to the transceiver. The processor is configured to: receive a fourth message from a third apparatus, the fourth message indicating whether additional training data should be used by the second apparatus to perform a model update of one or more models associated with a first or more apparatus, wherein the second apparatus is one of a second or more apparatuses, and the additional training data differs from the training data used at the first or more apparatuses to perform the model update; and update one or more models based on receiving the fourth message.

[0099] In a ninth aspect, a third device is provided. The third device includes a transceiver and a processor communicatively coupled to the transceiver. The processor is configured to: send a fourth message to a second device, the fourth message indicating whether additional training data should be used by the second device to perform a model update of one or more models associated with a first or more devices, wherein the second device is one of the second or more devices, and the additional training data is different from the training data used at the first or more devices to perform the model update.

[0100] In a tenth aspect, a third apparatus is provided. The third apparatus includes a transceiver and a processor communicatively coupled to the transceiver. The processor is configured to: send a fifth message to a fourth apparatus requesting influence on traffic routing, wherein influencing traffic routing is used to select a second apparatus and at least one candidate apparatus from the first one or more apparatuses; the second apparatus and the at least one candidate apparatus jointly perform model updates of one or more models associated with the first one or more apparatuses, the fourth apparatus being configured to perform policy control functions; and receive a sixth message from a fifth apparatus indicating the second apparatus and the at least one candidate apparatus, wherein the fifth apparatus is configured as a network controller.

[0101] In an eleventh aspect, a fourth apparatus is provided. The fourth apparatus includes a transceiver and a processor communicatively coupled to the transceiver. The processor is configured to: receive from a third apparatus a fifth message requesting influence on traffic routing, wherein influencing traffic routing is used to select a second apparatus in a second or more apparatus and to select at least one candidate apparatus in a first or more apparatus; the second apparatus and the at least one candidate apparatus jointly perform model updates of one or more models associated with the first or more apparatuses, and the fourth apparatus is configured to perform policy control functions; and generate at least one policy regarding influence on traffic routing based on the fifth message.

[0102] In a twelfth aspect, a fifth device is provided. The fifth device includes a transceiver and a processor communicatively coupled to the transceiver. The processor is configured to: receive from a fourth device at least one policy affecting traffic routing, wherein affecting traffic routing is used to select a second device from a second or more devices and to select at least one candidate device from the first or more devices; the second device and the at least one candidate device jointly perform model updates of one or more models associated with the first or more devices, the fourth device being configured to perform policy control functions, and the fifth device being configured as a network controller; select the second device and the at least one candidate device based on the at least one policy; and send a sixth message to a third device for instructing the second device and the at least one candidate device, wherein the third device is configured to control the second or more devices.

[0103] In a thirteenth aspect, a system is provided. The system includes a first device, a second device, a third device, a fourth device, and a fifth device. The third device is configured to send a fourth message indicating whether additional training data should be used by the second device to perform a model update of one or more models associated with the first or more devices, wherein the second device is one of the second or more devices, and the additional training data differs from the training data used at the first or more devices to perform the model update; the second device is configured to receive the fourth message from the third device and update one or more models based on receiving the fourth message; the first device is configured to receive an instruction from the second or third device instructing the first device to perform a model update, and send a first notification notifying that the model update of one or more models has been completed, wherein the first device is one of the first or more devices; the third device is configured to receive an instruction from the second or third device instructing the first device to perform a model update, and send a first notification notifying that the model update of one or more models has been completed, wherein the first device is one of the first or more devices; the fourth device is configured to receive an instruction from the second or third device instructing the first device to perform a model update, and send a first notification notifying that the model update of one or more models has been completed, wherein the first device is one of the first or more devices; the fifth device is configured to receive a fourth message indicating whether additional training data should be used by the second device or the third device to perform a model update, and send a first notification notifying that the model update of one or more models has been completed. The third device is also configured to control a second or more devices and send a fifth message to a fourth device for requesting influence on traffic routing, wherein influencing traffic routing is used to select at least one candidate device among the second device and the first or more devices, and the second device and at least one candidate device jointly perform a model update; the fourth device is configured to receive the fifth message from the third device to generate at least one policy regarding influence on traffic routing based on the fifth message, and send at least one policy to the fifth device; the fifth device is configured as a network controller and receives at least one policy from the fourth device, and the fifth device is also configured to select the second device and at least one candidate device based on the at least one policy, and send a sixth message to the third device for instructing the second device and at least one candidate device.

[0104] In a fourteenth aspect, a method implemented at a system is provided. The system includes a first device, a second device, a third device, a fourth device, and a fifth device. In the method, the third device sends a fourth message indicating whether additional training data should be used by the second device to perform a model update of one or more models associated with the first or more devices, wherein the second device is one of the second or more devices, and the additional training data differs from the training data used at the first or more devices to perform the model update; the second device receives the fourth message from the third device and updates one or more models based on receiving the fourth message; the first device receives an instruction from the second or third device instructing the first device to perform the model update, and sends a first notification notifying that the model update of one or more models has been completed, wherein the first device is one of the first or more devices. The third device controls the second or more devices and sends a fifth message to the fourth device to request influence on traffic routing, wherein influence on traffic routing is used to select at least one candidate device among the second device and the first or more devices, and the second device and the at least one candidate device jointly perform a model update; the fourth device receives the fifth message from the third device to generate at least one policy regarding influence on traffic routing based on the fifth message, and sends the at least one policy to the fifth device; the fifth device receives the at least one policy from the fourth device, and the fifth device is also configured as a network controller to select the second device and the at least one candidate device based on the at least one policy, and sends a sixth message to the third device to instruct the second device and the at least one candidate device.

[0105] In the fifteenth aspect, a system is provided. The system includes at least one of the following: a first device of the seventh aspect, a second device of the eighth aspect, a third device of the ninth or tenth aspect, a fourth device of the eleventh aspect, or a fifth device of the twelfth aspect.

[0106] In a sixteenth aspect, a non-transitory computer-readable medium is provided, including a computer program stored therein, which, when executed on at least one processor, causes the at least one processor to perform the method of any one of the first to sixth aspects.

[0107] In a seventeenth aspect, a chip is provided, including at least one processing circuit configured to perform the methods of any one of the first to sixth aspects.

[0108] In the eighteenth aspect, a computer program product tangibly stored in a computer-readable medium and including computer-executable instructions is provided, which, when executed, cause a device to perform any one of the first to sixth aspects. Attached Figure Description

[0109] Some exemplary embodiments are described below with reference to the accompanying drawings, wherein:

[0110] Figure 1A Exemplary environments in which some embodiments of this disclosure may be implemented are shown;

[0111] Figure 1B Exemplary communication systems in which some embodiments of this disclosure may be implemented are shown;

[0112] Figure 1C Exemplary component structures to which some embodiments of this disclosure may be applied are shown;

[0113] Figure 1D Exemplary module structures to which some embodiments of this disclosure may be applied are shown;

[0114] Figure 2A Exemplary processes according to some embodiments of this disclosure are shown;

[0115] Figure 2B Another exemplary process according to some other embodiments of this disclosure is shown;

[0116] Figure 3 Exemplary training processes according to some embodiments of this disclosure are shown;

[0117] Figure 4A An exemplary process of iterative evolution of a model according to some embodiments of this disclosure is illustrated;

[0118] Figure 4B Exemplary system architectures according to some embodiments of this disclosure are shown;

[0119] Figure 5 An exemplary process for an iteration of model evolution according to some other embodiments of this disclosure is illustrated;

[0120] Figure 6 An exemplary process for a request to influence traffic routing is illustrated according to some embodiments of this disclosure;

[0121] Figure 7 A flowchart illustrating an exemplary method implemented at a first device according to some embodiments of the present disclosure is shown;

[0122] Figure 8 A flowchart illustrating an exemplary method implemented at a second device according to some embodiments of the present disclosure is shown;

[0123] Figure 9 A flowchart illustrating an exemplary method implemented at a third device according to some embodiments of the present disclosure is shown;

[0124] Figure 10A flowchart illustrating an exemplary method implemented at a third device according to some other embodiments of the present disclosure is shown;

[0125] Figure 11 A flowchart illustrating an exemplary method implemented at a fourth device according to some embodiments of the present disclosure is shown;

[0126] Figure 12 A flowchart illustrating an exemplary method implemented at a fifth device according to some embodiments of the present disclosure is shown;

[0127] Figure 13 A flowchart illustrating exemplary methods implemented at a system according to some embodiments of the present disclosure is shown;

[0128] Figure 14 A simplified block diagram of an apparatus suitable for implementing an example embodiment of the present disclosure is shown.

[0129] In the accompanying drawings, the same or similar reference numerals denote the same or similar elements. Detailed Implementation

[0130] The principles of this disclosure will now be described with reference to some exemplary embodiments. It should be understood that these embodiments are described in order to illustrate and assist those skilled in the art in understanding and implementing this disclosure, and not to impose any limitation on the scope of this disclosure. The embodiments of this disclosure described herein can be implemented in various ways other than those described below.

[0131] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0132] References to "an embodiment," "embodiment," "example embodiment," etc., in this disclosure indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes such specific features, structures, or characteristics. The term "another embodiment" should be understood as "at least one other embodiment." Furthermore, these phrases do not necessarily refer to the same embodiment. Additionally, when a specific feature, structure, or characteristic is described in connection with an embodiment, it is understood by those skilled in the art that such a feature, structure, or characteristic will be implemented in conjunction with other embodiments, whether explicitly described or not.

[0133] It should be understood that although the terms “first” and “second”, etc., may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. As used herein, the term “and / or” includes any and all combinations of one or more of the listed items. Other explicit and implicit definitions may be included below.

[0134] The terminology used herein is for the purpose of describing particular embodiments and is not intended to limit the exemplary embodiments. Unless the context clearly indicates otherwise, the singular forms “a” and “an” and “the” as used herein are intended to include the plural forms as well. It should also be understood that the terms “comprising,” “including,” “having,” “having,” “containing,” and / or “containing,” when used herein, indicate the presence of the described features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof.

[0135] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0136] As used herein, the term "terminal device" refers to a terminal equipment or a module / chip within such a terminal equipment. A terminal equipment can refer to any device with wireless or wired communication capabilities. Examples of terminal devices include, but are not limited to: user equipment (UE), personal computers, desktops, mobile phones, cellular phones, smartphones, personal digital assistants (PDAs), portable computers, tablets, wearable devices, Internet of Things (IoT) devices, ultra-reliable and low-latency communications (URLLC) devices, Internet of Everything (IoE) devices, machine-type communication (MTC) devices, vehicular equipment for V2X communication (where X represents pedestrians, vehicles, or infrastructure / network), integrated access and backhaul (IAB) devices, small data transmission (SDT), mobility, multicast and broadcast services (MBS), positioning, dynamic / flexible duplex in commercial networks, reduced capability (RedCap), and spacecraft or aircraft in non-terrestrial networks (NTNs) (including satellites and unmanned aircraft). The following are considered as separate categories of devices: high-altitude platforms (HAPs) of systems (UAS), extended reality (XR) devices (including different types of reality such as augmented reality (AR), mixed reality (MR), and virtual reality (VR)), unmanned aerial vehicles (UAVs) (collectively referred to as unmanned aircraft, which are aircraft without any human pilots), devices on high-speed trains (HSTs), or image capture devices (such as digital cameras), sensors, gaming devices, music storage and playback devices, or internet devices that enable wireless or wired internet access and browsing."Terminal equipment" can also have "multicast / broadcast" features to support public safety and mission-critical applications, V2X applications, transparent IPv4 / IPv6 multicast delivery, IPTV, smart TVs, wireless services, software delivery over the wireless, group communications, and IoT applications. It may also include one or more subscriber identity modules (SIMs), referred to as Multi-SIM. The term "terminal equipment" is used interchangeably with UE, mobile station, subscriber station, mobile terminal, user terminal, wireless device, or reduced-capability terminal equipment.

[0137] As used herein, devices other than terminal devices (e.g., servers, network functions, devices that are part of the data plane / control plane of the core network, or radio access network (RAN) nodes, etc.) can be referred to as network devices. A network device can be a network equipment or a module / chip of the aforementioned network equipment. The term "network equipment" refers to a device capable of providing or hosting a cell or coverage area in which terminal devices can communicate. Examples of network equipment include, but are not limited to, base stations (NodeBs or NBs), evolved NodeBs (eNodeBs or eNBs), next-generation NodeBs (gNBs), transmission reception points (TRPs), remote radio units (RRUs), radio heads (RHs), remote radio heads (RRHs), IAB nodes, low-power nodes (e.g., femtonodes, piconodes), reconfigurable intelligent surfaces (RISs), network control repeaters, etc. In some other embodiments, the term "network device" may refer to a device on the core network side, such as a core network-side entity / element, like a network function on the control plane or a network function on the data plane. In other embodiments, the term "network device" may refer to a device in a data network, such as a data network-side entity / element, like a network server or an application server.

[0138] Terminal or network devices can possess artificial intelligence (AI) or machine learning capabilities. This typically includes a model trained on a specific function from a large amount of collected data and capable of predicting information. As an example, a terminal or network device can operate across several frequency ranges, such as FR1 (410 MHz to 7125 MHz), FR2 (24.25 GHz to 71 GHz), 71 GHz to 114 GHz, bands above 100 GHz, and terahertz (THz). It can also operate on licensed / unlicensed / shared spectrum. In multi-radio dual connectivity (MR-DC) applications, the terminal device and network device can have more than one connection. The terminal or network device can operate in full-duplex, flexible-duplex, and cross-duplex modes.

[0139] Network devices can have network energy saving and self-organizing network (SON) / minimization of drive test (MDT) capabilities. Terminals can have power saving functions.

[0140] The embodiments disclosed herein can be executed in test equipment such as signal generators, signal analyzers, spectrum analyzers, network analyzers, test terminal equipment, test network equipment, and channel simulators.

[0141] Embodiments of this disclosure may be performed according to any known or future generative communication protocol. Examples of communication protocols include, but are not limited to, fourth-generation (4G), 4.5G, fifth-generation (5G) communication protocols, 5.5G, 5G Advanced Networks, Wireless Fidelity (WiFi) Networks, Ultra Wideband (UWB) Networks, or sixth-generation (6G) Networks.

[0142] In some examples, values, processes, or devices are referred to as “best,” “lowest,” “highest,” “smallest,” “maximum,” etc. It should be understood that such descriptions are intended to indicate that there are many functional alternatives that can be chosen, and that such a choice is not necessarily better, smaller, higher, or more preferred than other choices.

[0143] As used herein, the term "circuit system" can refer to hardware circuitry and / or a combination of hardware circuitry and software. For example, a circuit system can be a combination of analog and / or digital hardware circuitry with software / firmware. In another example, a circuit system can be any part of a hardware processor with software, including one or more digital signal processors, software, and one or more memories, which work together to enable devices such as terminal devices or network devices to perform various functions. In yet another example, a circuit system can be hardware circuitry and / or a processor, such as a microprocessor or a part of a microprocessor, whose operation requires software / firmware, but may be absent when operation does not require software. As used herein, the term circuit system also encompasses hardware circuitry alone, or one or more processors, or hardware circuitry or a part of one or more processors, and the implementation of its (or their) accompanying software and / or firmware.

[0144] Some embodiments of this disclosure relate to training one or more models (e.g., AI models, also known as artificial intelligence (AI) or machine learning (ML) (AI / ML) models) in a network. In exemplary applications, each of a set of entities, such as a UE, server, network function, RAN node, etc., has a local model and also has (or hosts) a private dataset. The local models may be heterogeneous, i.e., they have different structures. In the application, each of these models will be trained using the private dataset without requiring the private dataset to leave its hosting entity. The data in the private dataset may be highly heterogeneous (biased), which can negatively impact training performance. It has been shown that using additional training data (e.g., data from a non-privacy-sensitive public dataset) can improve training performance. Therefore, there is a need for systems and methods that use additional training data during the training process (i.e., during the training of the local model).

[0145] The scheme disclosed herein differs from some existing schemes (e.g., schemes using federated learning). In AI model training schemes using federated learning, multiple entities (e.g., UEs, servers, network functions, RAN nodes, etc., or combinations thereof) jointly train a general AI model using their private datasets over multiple iterations. During an iteration, each of these entities sends a local version of the model to an aggregator node. The local version of the model sent from the entity to the aggregator node is obtained by the entity through local training (i.e., training the model based on, for example, a random initialization of the model or a global version of the pattern using the entity's private dataset). The global version of the model was received from the aggregator node in the previous iteration. The aggregator node receives the local versions of the model from the entities and aggregates the local versions of the model (e.g., using FederatedAveraging / FedAverage) to update the global version of the model. FederatedAveraging / FedAverage is a communication-efficient algorithm suitable for distributed training with a large number of clients. The aggregator node then sends the updated global version of the model (e.g., including values ​​of model parameters or gradients) to the entities, each of which performs local training accordingly for the next iteration. When the target version of the model converges, the aggregator node notifies the entity that the learning process has ended.

[0146] The goal of federated learning is to train a general model for all entities, unlike some embodiments of this disclosure which support personalized local models. Federated learning requires all or most / many of the Type 1 processing service functions (PSFs) (i.e., a small selection space) to participate in each iteration of the learning process (corresponding to model evolution in some embodiments of this disclosure), and the space for learning to adapt to network dynamics (e.g., resource availability and performance changes) is limited. In contrast, some embodiments of this disclosure involve a Type 1 PSF in each iteration of model evolution, which is dynamically selected based on the network (a larger selection space). Furthermore, federated learning is affected by biased private datasets, and the solutions it provides cannot mitigate the associated negative impacts, while some embodiments of this disclosure can. In some embodiments, Type 1 PSFs and Type 2 PSFs are different processing service functions. Type 1 PSFs have their own local models to be trained. In some embodiments, during model training of a Type 1 PSF, the network not only acts as a channel but also assists in training the model; for example, some nodes in the network can provide additional training data, and such nodes can be referred to as Type 2 PSFs. Type 2 PSFs do not have (or are not associated with) their own local model to be trained, but can help train the local model of Type 1 PSFs.

[0147] For illustrative purposes, the following will refer to Figures 1A to 13 The principles and exemplary embodiments of this disclosure are described. However, it should be noted that these embodiments are given to enable those skilled in the art to understand the inventive concept of some embodiments of this disclosure and to implement the solutions as presented herein, and are not intended to limit the scope of this disclosure in any way.

[0148] Figure 1A Exemplary environments in which some embodiments of this disclosure may be implemented are shown. References Figure 1A A simplified schematic diagram of a communication system is provided as an illustrative example and not a limitation. Communication system 100 includes a radio access network 120. Radio access network 120 may be a next-generation (e.g., sixth-generation, 6G or later) radio access network or a traditional (e.g., 5G, 4G, 3G, or 2G) radio access network. In radio access network 120, one or more electric devices (EDs) 110a to 120j (collectively referred to as 110) may be interconnected with each other or connected to one or more network nodes (170a, 170b, collectively referred to as 170). Core network 130 may be part of the communication system and may depend on or be independent of the radio access technology used in communication system 100. Communication system 100 also includes a public switched telephone network (PSTN) 140, the Internet 150, and other networks 160. In some embodiments, one or more electric devices (EDs) 110a to 120j or network functions of core network 130 may be a type 1 PSF, such as… Figure 5 The data plane portion of the core network 130 may be a type 2 PSF, such as a first type 2 PSF 540, a second type 2 PSF 550, or a type 2 PSF B 640.

[0149] Figure 1B An exemplary communication system 100-1 in which some embodiments of the present disclosure may be implemented is shown. Figure 1BAn exemplary communication system 100-1 is illustrated. Generally, communication system 100-1 enables multiple wireless or wired components to transmit data and other content. The purpose of communication system 100-1 may be to provide content, such as voice, data, video, and / or text, via broadcast, multicast, and unicast. Communication system 100-1 can operate by sharing resources (e.g., carrier spectrum bandwidth) among its constituent components. Communication system 100-1 may include terrestrial communication systems and / or non-terrestrial communication systems. Communication system 100-1 can provide a wide range of communication services and applications (e.g., earth monitoring, remote sensing, passive sensing and positioning, navigation and tracking, autonomous delivery, and mobility). Communication system 100-1 can provide high availability and robustness through the joint operation of terrestrial and non-terrestrial communication systems. For example, integrating a non-terrestrial communication system (or components thereof) into a terrestrial communication system can enable a heterogeneous network comprising multiple layers. Compared to traditional communication networks, heterogeneous networks can achieve better overall performance through efficient multi-link joint operation, more flexible function sharing, and faster physical layer link switching between terrestrial and non-terrestrial networks.

[0150] Terrestrial and non-terrestrial communication systems can be considered subsystems of a communication system. In the example shown, communication system 100-1 includes electronic devices (EDs) 110a to 110d (collectively referred to as ED 110), radio access networks (RANs) 120a to 120b, a non-terrestrial communication network 120c, a core network 130, a public switched telephone network (PSTN) 140, the Internet 150, and other networks 160. RANs 120a to 120b include corresponding base stations (BSs) 170a to 170b, which can be collectively referred to as terrestrial transmit and receive points (T-TRPs) 170a to 170b. The non-terrestrial communication network 120c includes access nodes 120c, which can be collectively referred to as non-terrestrial transmit and receive points (NT-TRPs) 172.

[0151] Alternatively or additionally, any ED 110 can be used to access, connect to, or communicate with any other T-TRP 170a to 170b, as well as NT-TRP 172, the Internet 150, the core network 130, the PSTN 140, other networks 160, or any combination thereof. In some examples, ED 110a can perform uplink and / or downlink transmissions with T-TRP 170a via interface 190a. In some examples, ED 110a, 110b, and 110d can also communicate directly with each other via one or more sidelink air interfaces 190b. In some examples, ED 110d can perform uplink and / or downlink transmissions with NT-TRP 172 via interface 190c.

[0152] Air interfaces 190a and 190b can use similar communication technologies, such as any suitable wireless access technology. For example, communication system 100-1 can implement one or more channel access methods in air interfaces 190a and 190b, such as code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), or single-carrier FDMA (SC-FDMA). Air interfaces 190a and 190b can use other higher-dimensional signal spaces, which may involve combinations of orthogonal and / or non-orthogonal dimensions.

[0153] The air interface 190c can enable communication between the ED 110d and one or more NT-TRP 172 via a wireless link (or simply link). For some examples, the link is a dedicated connection for unicast transmission, a connection for broadcast transmission, or a connection between a group of EDs and one or more NT-TRPs for multicast transmission.

[0154] RANs 120a and 120b communicate with core network 130 to provide various services, such as voice, data, and other services, to EDs 110a, 110b, and 110c. RANs 120a and 120b and / or core network 130 may communicate directly or indirectly with one or more other RANs (not shown), which may (or may not) be directly served by core network 130 and may (or may not) employ the same radio access technology as RANs 120a, RAN 120b, or both. Core network 130 may also serve as a gateway access between (i) RANs 120a and 120b, or EDs 110a, 110b, and 110c, or both, and (ii) other networks (e.g., PSTN 140, Internet 150, and other networks 160). Furthermore, some or all of EDs 110a, 110b, and 110c may include the ability to communicate with different wireless networks via different radio links using different radio technologies and / or protocols. Instead of wireless communication (or other than wireless communication), ED 110a, 110b, and 110c can also communicate with service providers or exchanges (not shown) via wired communication channels and with the Internet 150. PSTN 140 may include a circuit-switched telephone network for providing plain old telephone service (POTS). The Internet 150 may include computer networks and / or subnets (internal networks) and include protocols such as Internet Protocol (IP), Transmission Control Protocol (TCP), and User Datagram Protocol (UDP). ED 110a, 110b, and 110c may be multimode devices capable of operating according to various wireless access technologies and include multiple transceivers required to support these technologies.

[0155] Figure 1C Exemplary component structures 100-2 to which some embodiments of this disclosure may be applied are shown. Figure 1CAnother example of an ED 110 and base stations 170a, 170b, and / or 170c is shown. The ED 110 is used to connect people, objects, machines, etc. The ED 110 can be widely used in various scenarios, such as cellular communication, device-to-device (D2D), vehicle-to-everything (V2X), peer-to-peer (P2P), machine-to-machine (M2M), machine-type communication (MTC), Internet of Things (IoT), virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, smart cities, drones, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery, and mobility.

[0156] Each ED 110 represents any suitable end-user equipment used for wireless operation and may include (or be referred to as) devices such as: user equipment (UE), wireless transmit / receive unit (WTRU), mobile station, fixed or mobile subscriber unit, cellular phone, station (STA), machine type communication (MTC) device, personal digital assistant (PDA), smartphone, laptop, computer, tablet, wireless sensor, consumer electronics device, smartbook, vehicle, automobile, truck, bus, train, or IoT device, industrial equipment, or devices within the aforementioned equipment (e.g., communication modules, modems, or chips). Next-generation ED 110 may be referred to using other terms. Base stations 170a and 170b are T-TRPs and are referred to hereinafter as T-TRP 170. Also... Figure 1C As shown, NT-TRP is referred to as NT-TRP 172 below. Each ED 110 connected to T-TRP 170 and / or NT-TRP 172 may be dynamically or semi-statically enabled (i.e., established, activated, or enabled), disabled (i.e., released, deactivated, or disabled), and / or configured in response to one or more of the following: connection availability and connection necessity.

[0157] ED 110 includes a transmitter 201 and a receiver 203 coupled to one or more antennas 204. One antenna 204 is shown. Alternatively, one, some, or all of the antennas may be panels. For example, the transmitter 201 and receiver 203 may be integrated as a transceiver. The transceiver is used to modulate data or other content for transmission by at least one antenna 204 or a network interface controller (NIC). The transceiver is also used to demodulate data or other content received by at least one antenna 204. Each transceiver includes any suitable structure for generating signals for wireless or wired transmission and / or for processing signals received wirelessly or wiredly. Each antenna 204 includes any suitable structure for transmitting and / or receiving wireless or wired signals.

[0158] ED 110 includes at least one memory 208. Memory 208 stores instructions and data used, generated, or acquired by ED 110. For example, memory 208 may store software instructions or modules executed by one or more processing units 210, which are used to implement some or all of the functions and / or embodiments described herein. Each memory 208 includes one or more of any suitable volatile and / or non-volatile storage and retrieval devices. Any suitable type of memory can be used, such as random access memory (RAM), read-only memory (ROM), hard disk, optical disk, subscriber identity module (SIM) card, memory stick, secure digital (SD) memory card, processor cache, etc.

[0159] ED 110 may also include one or more input / output devices (not shown) or interfaces (e.g. Figure 1A (Wired interface of Internet 150 in the network). Input / output devices can interact with users or other devices on the network. Each input / output device includes any suitable structure for providing or receiving information from the user, such as a speaker, microphone, keypad, keyboard, display, or touchscreen, including network interface communication.

[0160] ED 110 also includes a processor 210 for performing operations including: operations related to preparing uplink transmissions to NT-TRP 172 and / or T-TRP 170; operations related to processing downlink transmissions received from NT-TRP 172 and / or T-TRP 170; and operations related to processing sidelink transmissions to and from another ED 110. Processing operations related to preparing uplink transmissions may include operations such as encoding, modulation, transmit beamforming, and generating symbols for transmission. Processing operations related to processing downlink transmissions may include operations such as receive beamforming, demodulating and decoding received symbols. Depending on the embodiment, receiver 203 may receive downlink transmissions (possibly using receive beamforming), and processor 210 may extract signaling from the downlink transmissions (e.g., by detecting and / or decoding signaling). Examples of signaling may be reference signals transmitted by NT-TRP 172 and / or T-TRP 170. In some embodiments, processor 276 performs transmit beamforming and / or receive beamforming based on beam direction indications (e.g., beamangle information (BAI)) received from T-TRP 170. In some embodiments, processor 210 may perform operations related to network access (e.g., initial access) and / or downlink synchronization, such as operations related to detecting synchronization sequences, decoding, and obtaining system information. In some embodiments, processor 210 may perform channel estimation using reference signals received from NT-TRP 172 and / or T-TRP 170.

[0161] Although not shown, processor 210 may form part of transmitter 201 and / or receiver 203. Although not shown, memory 208 may form part of processor 210.

[0162] Processor 210, as well as the processing components of transmitter 201 and receiver 203, may each be implemented by the same or different one or more processors, which execute instructions stored in memory (e.g., memory 208). Alternatively, some or all of the processing components of processor 210, as well as transmitter 201 and receiver 203, may be implemented using dedicated circuitry, such as a programmable field-programmable gate array (FPGA), a graphics processing unit (GPU), or an application-specific integrated circuit (ASIC).

[0163] In some implementations, the T-TRP 170 can be called by other names, such as base station, basetransceiver station (BTS), wireless base station, network node, network device, network-side device, transmit / receive node, base station, evolved NodeB (eNodeB or eNB), home eNodeB, next-generation NodeB (gNB), transmission point (TP), site controller, access point (AP), wireless router, relay station, remote radio head, ground node, ground network device, ground base station, base band unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), positioning node, etc. The T-TRP 170 can be a macro BS, pico BS, relay node, host node, or a combination thereof. T-TRP 170 may refer to the aforementioned equipment or a device within the aforementioned equipment (e.g., a communication module, modem, or chip).

[0164] In some embodiments, portions of T-TRP 170 may be distributed. For example, some modules of T-TRP 170 may be located remotely from the device housing the antenna of T-TRP 170 and may be coupled to the device housing the antenna via a communication link (not shown), sometimes referred to as a fronthaul, such as the Common Public Radio Interface (CPRI). Therefore, in some embodiments, the term T-TRP 170 may also refer to modules on the network side that perform processing operations such as determining the location of ED 110, resource allocation (scheduling), message generation, and encoding / decoding, and are not necessarily part of the device housing the antenna of T-TRP 170. These modules may also be coupled to other T-TRPs. In some embodiments, T-TRP 170 may actually be multiple T-TRPs that work together, for example, through coordinated multicast transmissions, to serve ED 110.

[0165] T-TRP 170 includes at least one transmitter 252 and at least one receiver 254 coupled to one or more antennas 256. One antenna 256 is shown. Alternatively, one, some, or all of the antennas may be panels. The transmitter 252 and receiver 254 may be integrated as a transceiver. T-TRP 170 also includes a processor 260 for performing operations, including operations related to: preparing transmissions for downlink transmission to ED 110; processing uplink transmissions received from ED 110; preparing transmissions for backhaul transmission to NT-TRP 172; and processing transmissions received from NT-TRP 172 via backhaul. Processing operations related to preparing transmissions for downlink or backhaul transmission may include operations such as encoding, modulation, precoding (e.g., MIMO precoding), transmit beamforming, and generating symbols for transmission. Processing operations related to processing transmissions received in the uplink or via backhaul may include operations such as receive beamforming, demodulation, and decoding received symbols. Processor 260 can also perform operations related to network access (e.g., initial access) and / or downlink synchronization, such as generating the contents of a synchronization signal block (SSB), generating system information, etc. In some embodiments, processor 260 also generates beam direction indications, such as a BAI, which can be scheduled by scheduler 253 for transmission. Processor 260 performs other network-side processing operations described herein, such as determining the location of ED 110, determining where to deploy NT-TRP 172, etc. In some embodiments, processor 260 can generate signaling, for example, to configure one or more parameters of ED 110 and / or one or more parameters of NT-TRP 172. Any signaling generated by processor 260 is transmitted by transmitter 252. It should be noted that, as used herein, "signaling" can be alternatively referred to as control signaling. Dynamic signaling can be transmitted in the control channel (e.g., the physical downlink control channel, PDCCH), while static or semi-static higher-layer signaling can be included in messages transmitted in the data channel (e.g., the physical downlink shared channel, PDSCH).

[0166] Scheduler 253 may be coupled to processor 260. Scheduler 253 may be included within or operate separately from T-TRP 170, and may schedule uplink, downlink, and / or backhaul transmissions, including issuing scheduling authorizations and / or configuring unscheduled (“configured authorizations”) resources. T-TRP 170 also includes memory 258 for storing information and data. Memory 258 stores instructions and data used, generated, or acquired by T-TRP 170. For example, memory 258 may store software instructions or modules executed by processor 260 for implementing some or all of the functions and / or embodiments described herein.

[0167] Although not shown, processor 260 may form part of transmitter 252 and / or receiver 254. Furthermore, although not shown, processor 260 may implement scheduler 253. Although not shown, memory 258 may form part of processor 260.

[0168] The processor 260, scheduler 253, and processing components of transmitter 252 and receiver 254 may each be implemented by the same or different one or more processors, which execute instructions stored in memory (e.g., memory 258). Alternatively, some or all of the processor 260, scheduler 253, and processing components of transmitter 252 and receiver 254 may be implemented using a dedicated circuit system (e.g., FPGA, GPU, or ASIC).

[0169] Although the NT-TRP 172 is shown as an example of a drone, it can be implemented in any suitable non-terrestrial form. Furthermore, the NT-TRP 172 may be referred to by other names in some implementations, such as a non-terrestrial node, a non-terrestrial network device, or a non-terrestrial base station. The NT-TRP 172 includes a transmitter 272 and a receiver 274 coupled to one or more antennas 280. One antenna 280 is shown. Alternatively, one, some, or all of the antennas may be a panel. The transmitter 272 and receiver 274 may be integrated as a transceiver. The NT-TRP 172 also includes a processor 276 for performing operations, including operations related to: preparing transmissions for downlink transmissions to ED 110; processing uplink transmissions received from ED 110; preparing transmissions for backhaul transmissions to T-TRP 170; and processing transmissions received from T-TRP 170 via backhaul. Processing operations related to preparing a transmission for downlink or backhaul transmission may include operations such as encoding, modulation, precoding (e.g., MIMO precoding), transmit beamforming, and generating symbols for transmission. Processing operations related to processing transmissions received in the uplink or received via backhaul may include operations such as receive beamforming, demodulation, and decoding of received symbols. In some embodiments, processor 276 performs transmit beamforming and / or receive beamforming based on beam direction information (e.g., BAI) received from T-TRP 170. In some embodiments, processor 276 may generate signaling, for example, to configure one or more parameters of ED 110. In some embodiments, NT-TRP 172 implements physical layer processing but does not implement higher-level functions such as medium access control (MAC) or radio link control (RLC) layer functions. Since this is an example, more generally, NT-TRP 172 may implement higher-level functions in addition to physical layer processing.

[0170] The NT-TRP 172 also includes a memory 278 for storing information and data. Although not shown, a processor 276 may form part of the transmitter 272 and / or receiver 274. Although not shown, the memory 278 may form part of the processor 276.

[0171] Processor 276, and the processing components of transmitter 272 and receiver 274, may each be implemented by the same or different one or more processors, which execute instructions stored in memory (e.g., memory 278). Alternatively, some or all of the processing components of processor 276, transmitter 272, and receiver 274 may be implemented using a dedicated circuit system (e.g., a programmable FPGA, GPU, or ASIC). In some embodiments, NT-TRP 172 may actually be multiple NT-TRPs that work together, for example, through coordinated multipoint transmissions, to serve ED 110.

[0172] T-TRP 170, NT-TRP 172 and / or ED 110 may include other components, but for clarity these components are omitted.

[0173] Figure 1D Exemplary module structures to which some embodiments of this disclosure can be applied are shown. According to Figure 1D One or more operations (or steps) of the methods provided in this embodiment can be performed by the corresponding unit or module. Figure 1D The diagram illustrates units or modules within a device (e.g., in ED 110, T-TRP 170, or NT-TRP 172). For example, signals can be transmitted via a transmitting unit or transmitting module. Signals can be received via a receiving unit or receiving module. Signals can be processed via a processing unit or processing module. Other operations (or steps) can be performed by artificial intelligence (AI) or machine learning (ML) modules. The corresponding units or modules can be implemented using hardware, one or more components or devices executing software, or a combination thereof. For example, one or more of the units or modules can be integrated circuits, such as a programmable FPGA, GPU, or ASIC. It should be understood that if the aforementioned modules are implemented using software executed by a processor, etc., these modules can be retrieved by the processor, wholly or partially, individually or collectively, for processing, or in one or more instances as needed, and these modules themselves may include instructions for further deployment and instantiation.

[0174] Additional details regarding ED 110, T-TRP 170, and NT-TRP 172 are known to those skilled in the art. Therefore, for clarity, these details are omitted herein.

[0175] Figure 2A Exemplary process 200-1 according to some embodiments of this disclosure is shown. For example... Figure 2AAs shown, process 200-1 involves a first device 221, a second device 220, and a third device 230. In some examples, the first device 221 may be a terminal device, a server, or a network function. A server is a computer or system that provides resources, data, services, or programs to other computers (called clients) over a network. A network function (NF) is a functional building block within a network infrastructure, having a well-defined external interface and well-defined functional behavior; for example, a network function may be a network node or a physical device. A network function may also be a logical network entity that can be deployed on or located at a network location or physical network node. In some embodiments, a network function is a virtual network function that can be instantiated in a cloud computing environment (e.g., a data center). In some embodiments, examples of the first device 221 may be found... Figure 5 Type 1 PSF 510 or Figure 6 Type 1 PSF A 610. The terminal device may be a terminal device or a module / chip of a terminal device. A PSF is a processing service function. In some examples, the second device 220 is part of the data plane of the core network, or the second device 220 is connected to the data plane of the core network. In some embodiments, the data plane is referred to as the user plane, such as the user plane of a 3GPP 5G system. In some examples, the data plane may be a collection of network functions used to process and / or transmit data traffic. A Type 2 PSF may be an example of the second device 220, such as a Type 2 PSF 403 or a Type 2 PSF 540 or a Type 2 PSF B640 hereinafter. In some embodiments, the third device 230 may be part of the control plane of the core network. The control plane may be a collection of network functions that control how the system operates, including how to process and / or transmit data traffic. In some embodiments, the third device 230 may be Figure 2B The third device 231. In some embodiments, process 200-1 involves one or more other devices ( Figure 2A (not shown in the diagram), for example, a fourth device, a fifth device, and / or a sixth device. The fourth device, the fifth device, and / or the sixth device may be respectively associated with process 200-2 (see reference). Figure 2B The fourth device 240, fifth device 250, and / or sixth device mentioned in the document are the same devices. Some of the optional operations below are not listed in the document. Figure 2A As shown in the diagram, one or more optional operations can also be referenced. Figure 2B .

[0176] First device 221 is one of a first or more devices, and second device 220 is one of a second or more devices. One of the second or more devices (which can be any device) can be part of the data plane of the core network, or connected to the data plane of the core network. First device 221 can be selected from the first or more devices to perform model updates of one or more models associated with the first or more devices. One of the first or more devices can be identified by third device 230 as the first device 221 for performing model updates. The model associated with one of the first or more devices (e.g., the first device) may refer to a local model of a device (i.e., one of the first or more devices), for example, a model stored or maintained by a device that includes that device.

[0177] The second device 220 can be selected from one or more second devices. The third device 230 can send a fifth message to the fourth device (e.g., hereinafter). Figure 2B The fifth message (265) can be received from the third device 230 by the fourth device. The fifth message is used to request influence on traffic routing, wherein influence on traffic routing is used to select the second device 220 from the second or more devices and to select at least one candidate device from the first or more devices. The second device 220 and the at least one candidate device can jointly perform model updates of one or more models associated with the first or more devices, and the fourth device is configured to perform policy control functions. The at least one candidate device includes the first device 221. For each iteration, a device can be selected as the first device 221 from the first or more devices, and the at least one candidate device is a set of devices selected as the first device 221 for each iteration. In some examples, the fifth message may include information such as data traffic information identifying data traffic of at least one application (in some embodiments, at least one application may be associated with the second or more devices), wherein the data traffic will be routed during traffic routing; information of the first or more devices; information of at least one location of at least one application (in some embodiments, at least one location of at least one application corresponds to the second or more devices); information of one or more requests associated with traffic routing; or traffic filtering information associated with the data traffic; or any combination thereof. In some embodiments, an application may be an application program and / or an application server with associated data traffic, and the application may be associated with a device (i.e., one of a second or more devices), indicating that the location of the application corresponds to that device. In some examples, the number of applications may be one or more, i.e., there may be at least one application.

[0178] As described above, at least one application can be associated with a second or more devices (e.g., type 2 PSF), specifically, the location of the application in at least one application corresponds to a type 2 PSF (e.g., type 2 PSF 403 or first type 2 PSF 540 or type 2 PSF B 640 hereinafter). Furthermore, one of the candidate devices, such as a type 1 PSF (e.g., Figure 5 Type 1 PSF 510 or Figure 6 In the Type 1 PSF (A 610), the application can be added to the model training. In other words, the application can be associated with a Type 2 PSF (e.g., the Type 2 PSF corresponds to the location of the application) and can be accessed by a Type 1 PSF (e.g., the Type 1 PSF can send and / or receive application-related data traffic). In the fifth message described above, the data traffic identified by the data traffic information may include data traffic that will be routed to (or targeted to) a Type 1 PSF (referred to as downlink traffic) and data traffic that will be routed to (or targeted to) a Type 2 PSF (referred to as uplink traffic). Based on the fifth message, the fourth device can generate at least one policy regarding the influence of traffic routing. In some embodiments, the at least one policy may include the following information: information about the data traffic to be routed in the traffic routing, i.e., information about data traffic identifying data traffic of at least one application associated with the second or more devices, as included in the fifth message; information about at least one location of at least one application, wherein one of the at least one applications is associated with the second device, for example, the Type 2 PSF corresponds to the location of the application; information about one or more requests associated with the traffic routing; or information about the first or more devices; or any combination thereof. Information included in at least one strategy may be the same as information included in the fifth message. Details of the information included in the fifth message can also be found in [reference needed]. Figure 6 The corresponding description of the request sent from PSC 560 to PCF 630 in operation 603. At least one policy may also be referred to as one or more policies in operation 605 below.

[0179] Furthermore, in some examples, the fourth device can send to the fifth device (e.g., below). Figure 2BThe fifth device 250 sends at least one policy. The fifth device is configured as a network controller to determine the second device 220 and at least one candidate device based on the at least one policy. The at least one candidate device includes the first device 221. In some examples, the at least one candidate device includes more than one candidate device, and one of the at least one candidate device will be selected as the first device 221. In some other examples, the at least one candidate device includes only one candidate device, and that candidate device will be used as the first device 221. Therefore, the fifth device is configured as a network controller to determine the second device 220 and the first device 221 (i.e., the only candidate device). On the fifth device side, after determining the second device 220 and at least one candidate device, the fifth device may send, and the third device 230 may receive, a sixth message (which may be further referenced) indicating the second device 220 and at least one candidate device. Figure 2B (Sixth message 285 in the original text). After receiving the sixth message, the third device 230 can send feedback on the sixth message to the fifth device, and the fifth device can receive feedback from the third device 230 accordingly. The feedback on the sixth message can indicate an acknowledgment (ACK) or a negative ACK for the selection of an application location associated with the second device 220. Specifically, an acknowledgment ACK can be an affirmation of the selection, and a negative ACK can be a rejection of the selection. In some examples, the selection indicates that multiple second devices (including second device 220) are selected, and then the third device 230 can determine the second device 220 based on the selection by selecting an application location associated with the second device 220 in at least one application location. In some examples, information about at least one application can be included in at least one policy, and the fifth device can receive this information. In some examples, the third device 230 can also determine a candidate device as a first device 221 among at least one candidate device, and information identifying the first device 221 (e.g., ID or network address) can be included in the acknowledgment ACK for the sixth message. In some other examples, if only one candidate device exists (i.e., the first device 221), the aforementioned sixth message is used to indicate the second device 220 and the first device 221. In these examples, the first device 221 is indicated in the sixth message, and the third device 230 is unsure of the first device 221.

[0180] In some other examples, the fifth device can be integrated into the fourth device, and the fourth device then has the functionality of the fifth device, i.e., the fourth device can determine the second device 220 and at least one candidate device based on at least one strategy. In such examples, the transmission of at least one strategy is implemented within the fourth device. Similar to the examples above, at least one candidate device includes the first device 221. In some examples, at least one candidate device includes more than one candidate device, and one of the at least one candidate device will be selected as the first device 221. In some other examples, at least one candidate device includes only one candidate device, and that candidate device will be the first device 221; therefore, the fourth device can determine the second device 220 and the first device 221 (i.e., the only candidate device). The sixth message can be sent from the fourth device to the third device 230, i.e., the fourth device sends a sixth message to the third device 230 indicating the second device 221 and at least one candidate device. After receiving the sixth message, the third device 230 can send feedback to the fourth device regarding the sixth message, and the fourth device can receive feedback from the third device 230. The feedback to the sixth message can indicate an affirmative acknowledgement (ACK) for the selection or a negative ACK for the selection. If the feedback to the sixth message is an affirmative ACK, the affirmative ACK to the sixth message also includes information identifying the candidate device among at least one candidate device as the first device 221. In some other examples, if only one candidate device (i.e., the first device 221) exists, the aforementioned sixth message is used to indicate the second device 220 and the first device 221. In these examples, the first device 221 is indicated in the sixth message, and the fourth device is unsure of the first device 221.

[0181] In some examples, the fifth and fourth devices are integrated into a single network entity. In some embodiments, a sixth message from the fourth or fifth device may indicate a data plane management event associated with selecting an application location associated with the second device 220 in at least one application location. The data plane management event is used to indicate the selection of the application location. Figure 2A The fourth and fifth devices are not shown. In some examples, the fourth device may be... Figure 2B The fourth device 240, and the fifth device can be a fifth device 250. In some examples, the fifth device can be an NWC, for example... Figure 5 or Figure 6 The NWC520 in [the example]. In some examples, the fourth device can be a PCF, for example... Figure 6 The PCF 630 in the example. In some examples, the fifth device (e.g., NWC) can be integrated with the fourth device (e.g., PCF).

[0182] The first device 221 may receive a second message notifying it that it has been selected to perform a model update. In some examples, the second message may originate from a fourth or fifth device. In some embodiments, the fifth device is a network controller. The fifth device may send, and the first device 221 may receive, the second message from the fifth device. In some examples, after receiving the second message, the first device 221 may send an acknowledgment (ACK) to the fifth device indicating confirmation of the selection. On the fifth device side, the fifth device may receive the acknowledgment (ACK) confirming the performance of the model update; in other words, the acknowledgment (ACK) confirms the model update. Alternatively, in some embodiments, the fourth device is used to perform policy control functions, and the fourth and fifth devices are integrated into a single entity. The fourth device may send, and the first device 221 may receive, the second message from the fourth device. In some examples, after receiving the second message, the first device 221 may send an acknowledgment (ACK) to the fourth device indicating confirmation of the selection. On the fourth device side, the fourth device may receive the acknowledgment (ACK) confirming the performance of the model update.

[0183] The first device 221 may receive an instruction to instruct the first device 221 to perform a model update of one or more models associated with the first or more devices. In some embodiments, the received instruction may be an instruction 225 from the second device 220; for example, the second device 220 may send (201a) an instruction 225 to the first device 221 of the first or more devices to instruct the first device 221 to perform a model update of one or more models. The second device 220 may obtain the instruction 225 from the third device 230; for example, the third device 230 may send, and the second device 220 may receive, a message including the instruction 225. In some examples, the message including the instruction 225 may be the same message as the fourth message 245 described below. On the first device 221 side, the first device 221 receives (203a) the instruction 225. In some examples, the instruction 225 may be carried in a first message from the second device 220, which may also include values ​​of model parameters for one or more models. In some other examples, indication 225 may be carried in a first message from second device 220, where first device 221 may obtain values ​​of model parameters for one or more models. In some examples, the values ​​of the model parameters may be pre-configured at first device 221. In some other examples, the values ​​of the model parameters may be dynamically provided by second device 220; for example, the values ​​of the model parameters may be provided together with indication 225, or alternatively, the values ​​of the model parameters may be included in indication 225. In some examples, indication 225 may be sent from second device 220 to first device 221 via data plane functionality.

[0184] In some other embodiments, the received instruction may be instruction 235 from a third device 230. For example, the third device 230 may send (205) instruction 235 to the first device 221 of the first or more devices, instructing the first device 221 to perform a model update. In other words, the third device may send instruction 235 to the first device 221 without going through the second device 221. On the first device 221 side, the first device 221 receives (207) instruction 235. Additionally, in some examples, the first device 221 may also receive values ​​of model parameters for one or more models from the second device 220. In some examples, instruction 235 may be sent from the third device 230 to the first device 221, for example, instruction 235 may be sent without going through the control plane function. In some other examples, instruction 235 may be sent from the third device 230 to the first device 221 via the control plane function. In some embodiments, control plane functions may include network exposure function (NEF), policy control function (PCF), network storage function (NSF), network controller (NWC), path management function (PMF), or access and mobility management function (AMF), or a combination of one or more of the above functions.

[0185] In some embodiments, the received indication, such as indication 225 or indication 235, may include a list of model identification information for identifying one or more models; version information of one or more models; one or more status indications; or any combination thereof. Each of the one or more models has corresponding version information. Each of the one or more status indications indicates the status of a model in the one or more models, and the model status includes frozen or unfrozen. In some embodiments, a frozen model is a converged model (in other words, the model has converged), and additionally or alternatively, an unfrozen model is a non-converged model (in other words, the model has not yet converged). During a model update of one or more models, the status of a model in one or more models may change, for example, from a frozen state to an unfrozen state, or from an unfrozen state to a frozen state. For example, the status of the model associated with the first device 221 may change when or after the first device 221 performs a model update.

[0186] On the second device 220 side, in some embodiments, model updates may also be performed at the second device 220. For example, the second device 220 may not have (or be associated with) a local model to be trained, but the second device 220 may assist in training the local model of the first device or more. In some examples, the second device 220 may perform model updates using additional training data. The additional training data may be different from the training data used to perform model updates at the first device or more.

[0187] In some embodiments, the third device 230 may send (209) a fourth message 245 to the second device 220, and correspondingly, the second device 220 may receive (211) a fourth message 245 from the third device 230. The fourth message 245 is used to indicate whether additional training data should be used by the second device 220 to perform model updates of one or more models associated with the first or more devices. In some embodiments, the fourth message 245 may also indicate where the additional training data is obtained, and / or how the additional training data is selected if it should be used. As described above, in some examples, the fourth message 245 may also be used to instruct the first device 221 to perform model updates. In such examples, operations 209 and 211 are performed prior to operations 201a and 203a, as... Figure 2A As shown. In some examples, the fourth message 245 may include: a list of model identification information for identifying one or more models; version information for one or more models, wherein each of the one or more models has corresponding version information; one or more status indicators, wherein each of the one or more status indicators indicates the status of a model in one or more models, the model status including frozen or unfrozen. Based on the receipt of the fourth message 245, the second device 220 updates (213) one or more models. In some examples, during the updating of one or more models, if additional training data should be used by the second device 220 to perform model updates, the second device 220 uses the additional training data to update at least one model in one or more models that has a status indicator indicating unfrozen (i.e., unconverged).

[0188] As described above, the first device 221 can be instructed to perform a model update. The first device 221 will perform the model update accordingly (i.e., according to instruction 225 or 235). During the performance of the model update, the first device 221 can update at least one of one or more models. After the model update is completed, the first device 221 can send (215) a first notification 255 to notify one or more models that the model update is complete. In some examples, the first notification 255 can be sent from the first device 221 to the third device 230. In some other examples, the first notification 255 can be sent from the first device 221 to the third device 230 via a control plane function (e.g., NWC or AMF). On the third device 230 side, the third device 230 can receive (217) the first notification 255. In some embodiments, the first notification 255 may also include a list of model identification information for identifying one or more models, or one or more status indicators, or a combination thereof.

[0189] In some embodiments, based on the first notification 255, the third device 230 can determine from the second or more devices a sixth device for receiving at least one latest version of one or more models. The at least one latest version is obtained through a model update performed by the first device 221, and the latest version corresponds to the updated model.

[0190] Furthermore, the third device 230 can send a third message (e.g., a control command sent in operation 517 below) to the first device 221. The third message is used to notify the first device 221 to send at least one latest version of one or more models obtained through model updates. For example, at least one latest version of one or more models can be sent to a sixth device among the second or more devices. In some examples, the sixth device can be the second device 220, i.e., the sixth device and the second device 220 can be the same device. On the first device 221 side, the first device 221 receives the third message from the third device 230 and then sends at least one latest version. In some examples, the third device 230 sends the message to the first device 221 via a control plane function, and the first device 221 receives the third message from the third device 230 via a control plane function. In some other examples, the third device 230 sends the message via a data plane function, and the first device 221 receives the message via a data plane function. In some embodiments, the first device 221 sends the message, and the data plane function receives the message, the data plane function being configured to transmit at least one latest version to a sixth device among the second or more devices, and the first device 221 may not know which of the second or more devices is the sixth device. In some examples, the third message may also include model identification information for identifying at least one model associated with at least one latest version.

[0191] As described above, in some examples, the sixth device may be the same as the second device 220, in which the third device 230 may send the second notification, and the second device 220 may receive the second notification from the third device 230. In some other examples, the sixth device is different from the second device 220, in which the third device 230 may send the second notification, and the sixth device may receive the second notification from the third device 230. The second notification is used to notify the second device 220 or the sixth device that it has been selected to receive at least one latest version of one or more models from the first device 221, wherein the at least one latest version is obtained by performing a model update by the first device 221. In some examples, the second notification also includes model identification information for identifying at least one model associated with the at least one latest version.

[0192] In some embodiments, the third device 230 may send, and the second device 220 may receive, a configuration for the second device 220 to obtain one or more latest versions from another device among the second or more devices. In some embodiments, the third device 230 may send, and the sixth device may receive, a configuration for the sixth device to obtain one or more latest versions from another device among the second or more devices. In the above embodiments, the one or more latest versions obtained from the other device are among multiple versions of one or more models, excluding at least one latest version to be received by the second device 220 or the sixth device.

[0193] Figure 2B Another exemplary process according to some other embodiments of this disclosure is shown. Figure 2B Exemplary processes 200-2 according to some embodiments of this disclosure are shown. For example... Figure 2B As shown, process 200-2 involves a third device 231, a fourth device 240, and a fifth device 250. In some embodiments, the third device 231, the fourth device 240, and the fifth device 250 may be the same as the third device 230, the fourth device, and the fifth device in process 200-1, respectively. In some embodiments, process 200-2 involves one or more other devices ( Figure 2B (not shown in the diagram), for example, the first device, the second device, and / or the sixth device, which may also be used in process 200-1 (see reference). Figure 2A The first device 221, the second device 220 and / or the sixth device mentioned in the document are the same.

[0194] In process 200-2, the third device 231 may send (202) a fifth message 265 to the fourth device 240. The fifth message 265 is used to request influence on traffic routing. In some examples, the fifth message 265 may include data traffic information identifying data traffic of at least one application associated with a second or more devices (e.g., the second or more devices in process 200-1), wherein the data traffic will be routed during traffic routing; information of the first or more devices (e.g., the first or more devices in process 200-1); information of at least one location of at least one application associated with the second device; information of one or more requests associated with traffic routing; traffic filtering information associated with the data traffic; or a combination thereof. Influencing traffic routing is used to select a second device from the second or more devices and to select at least one candidate device from the first or more devices. The second device and at least one candidate device jointly perform model updates for one or more models associated with the first or more devices, and the fourth device 240 performs policy control functions. On the fourth device 240 side, the fourth device 240 receives (204b) the fifth message 265 from the third device 231. Furthermore, based on the fifth message 265, the fourth device 240 can generate (206) at least one policy 275 affecting traffic routing. In some embodiments, the fourth device 240 can send (208b) at least one policy 275 to the fifth device 250, which is configured as a network controller for determining (selecting) the second device and at least one candidate device. On the fifth device 250 side, the fifth device 250 can receive (212) at least one policy 275 affecting traffic routing from the fourth device 240. Then, the fifth device 250 can select (214) the second device and at least one candidate device based on the at least one policy 275. Furthermore, the fifth device 250 can send (216) a sixth message 285 to the third device 231 for indicating the second device and at least one candidate device, wherein the third device can be used to control or manage the second or more devices. On the third device 231 side, the third device 231 can receive (218) the sixth message 285 from the fifth device. Further details regarding the fifth message 265, at least one strategy 275, and the sixth message 285 can be found in the corresponding descriptions in the above process 200-1.

[0195] Similar to process 200-1, in some embodiments, the fourth device 240 may send a second message to the first device to notify that the first device has been selected to perform the model update. The fourth device 240 can then receive from the first device a positive ACK confirming the performance of the model update or a negative ACK rejecting the performance of the model update. In some other embodiments, the fifth device 250 may send a second message to the first device to notify that the first device has been selected to perform the model update, and then the fifth device can receive from the first device a positive ACK confirming the performance of the model update or a negative ACK rejecting the performance of the model update. If the negative ACK rejecting the performance of the model update is received by either the fourth device 240 or the fifth device 250, a new device, different from the first device that sent the rejection, can be selected as the new first device to perform the model update. The first device performing the model update may also be referred to as the target first device. In embodiments of this disclosure, the operations related to the first device 221 in process 200-1 and the operations related to updating the model of the first device in process 200-2 are performed by the target first device.

[0196] In some embodiments, the fourth device 240 and the fifth device 250 are integrated into a single network entity, meaning that from the perspective of the single network entity, the fourth device 240 has the functionality of the fifth device 250. For example, the fourth device 240 can send traffic routing information indicating traffic routing to a seventh device. The seventh device is used to select a data plane path for traffic routing. In some examples, the fourth device 240 can receive a third notification from the seventh device that a data plane path has been configured for traffic routing. In some other examples, the fourth device 240 can receive a response from the seventh device to the transmission of traffic routing information. This response includes a third notification that a data plane path has been configured for traffic routing.

[0197] In some other embodiments, the fourth device 240 and the fifth device 250 are not integrated. The fifth device 250 may send traffic routing information indicating traffic routing to the seventh device. The fifth device 250 may receive a third notification from the seventh device that a data plane path has been configured for traffic routing, or may receive a response from the seventh device to the transmission of traffic routing information, wherein the response includes the third notification that a data plane path has been configured for traffic routing. In some embodiments, the traffic routing information may include information identifying at least one application location corresponding to the second device 220 (e.g., ID or network address); information identifying the first device (e.g., ID or network address); information associated with (or included therein) at least one policy 275; or information about the data plane path used for traffic routing (e.g., the ID of the DPF in the data plane path, and information describing how the DPF is connected along the data plane path); or any combination thereof.

[0198] In some other embodiments, the fifth device 250 may be integrated within / implemented by the seventh device. Therefore, within this integrated single network entity, the interaction between the fifth device 250 and the seventh device may be an internal process of the seventh device (and thus optional).

[0199] It should be noted that process 200-2 may also include one or more other operations performed by the first device 221, the second device 220, the third device 230, the fourth device, the fifth device and / or the sixth device in process 200-1. In some embodiments, process 200-1 and process 200-2 may be implemented in combination.

[0200] As mentioned above, each network entity, or combination thereof, among a set of network entities such as UE, server, network function, and RAN node has a local AI model (or simply local model) and a private dataset. The local models may be heterogeneous, meaning they have different structures. Within the private datasets, different private datasets can include different data (diversified data). The goal is to train each of these local models (e.g., with) the private datasets (benefiting from data diversity). Figure 2A and Figure 2B The local model described in the relevant embodiments does not require the private dataset to leave its hosting network entity (protecting data privacy). The data in the private dataset may be highly heterogeneous (biased or imbalanced), and data heterogeneity may negatively impact training performance. In some examples of this disclosure, the set of network entities may be examples of a first or more devices, for example, in some embodiments, a processing service function (PSF), such as... Figure 4A Type 1 PSF 1 to 5, or Figure 5 and Figure 6The Type 1 PSF mentioned above. In some examples, one of the first or more devices (e.g., first device 221, and examples of first device 221) is the target Type 1 PSF. The systems and methods of embodiments of this disclosure can achieve the above objectives. In some examples, during the training of the local model, the schemes of embodiments of this disclosure can use additional training data to mitigate the negative impact from biased private datasets. The additional training data is selected from one or more datasets that are different from the private datasets and are transparent to the Type 1 PSF. The additional training data is selected by the processing service controller (PSC) and used at or by one or more Type 2 PSFs. Therefore, using additional training data does not increase the complexity of the Type 1 PSF, which is particularly important for a Type 1 PSF as a wireless device (e.g., UE). A Type 2 PSF is a network entity (e.g., a server, network function, RAN node, etc.) that does not have (or is not associated with) the local model to be trained, but will help train the local model of the Type 1 PSF. A Type 2 PSF can be an example of a second device (e.g., second device 220) in the above embodiments.

[0201] Figure 3 An exemplary training process 300 according to some embodiments of the present disclosure is shown, such as Figure 3 As shown, the model training process 300 may include three phases: a model collection phase 310, a model evolution phase 330, and a model distribution phase 340. The training process 300 begins with the model collection phase 310 and ends with the model distribution phase 340. The model evolution phase 330 is performed iteratively between the model collection phase 310 and the model distribution phase 340. In the model evolution phase 330, model updates (as mentioned in some of the embodiments above) may be performed. Each Type 1 PSF (an example of the first one or more devices described above) can join the training process 300 by connecting to an application corresponding to the training. When the application is connected, the Type 1 PSF may send (e.g., Figure 5 Operation 519 in the middle) and / or receiving (e.g. Figure 6 Operation 609 in the text refers to application-related data traffic. This data traffic is referred to as application traffic. In model evolution phase 330, a Type 1 PSF, as the target Type 1 PSF, can execute... Figure 2A The operation is performed by the first device 221. Type 2 PSF can perform... Figure 2AThe operations are performed by the second device 220. During the model evolution phase 330, there are some interactions between entities, such as Type 1 PSF, Type 2 PSF, and other entities (including PSC (an example of the third device 231), PCF (an example of the fourth device 230), NWC (an example of the fifth device 250), etc.). Details of the interactions can be found in [reference needed]. Figure 2A and Figure 2B .

[0202] In model collection phase 310, each Type 1 PSF trains its local model using its private dataset and then sends that local model to the Type 2 PSF. The trained local model (or updated local model) can be sent to the Type 2 PSF at the end of an iteration during training; the Type 2 PSF used to receive the model can be the same or different for different iterations. The local model and the private dataset are considered associated with or corresponding to the Type 1 PSF. When sending the local model to the Type 2 PSF, the Type 1 PSF can indicate whether the local model is frozen (i.e., whether it is frozen (i.e., converged)) or not frozen (i.e., not converged)). All Type 1 PSFs send their local models to the same Type 2 PSF. These local models are considered in their initial versions and are collected at the Type 2 PSF.

[0203] In model evolution phase 330, model evolution can be performed distributed across different entities (e.g., at least one Type 1 PSF and a Type 2 PSF), and the model evolution process can include multiple iterations. The Type 2 PSF maintains state information for each local model, indicating whether the local model is frozen (i.e., whether the local model is frozen or not), and receives this information from the corresponding Type 1 PSF. After model collection phase 310 or after an iteration of model evolution phase 330, the Type 2 PSF evaluates the termination conditions of the training process based on the state information and determines whether the termination conditions are met (320). The termination conditions are met, for example, when all local models are frozen, or the number of frozen local models is greater than or not less than a certain threshold, or the ratio of frozen local models to unfrozen local models is greater than or not less than a certain threshold. If the termination conditions are met, the Type 2 PSF triggers / executes model distribution phase 340. Otherwise, the Type 2 PSF triggers / executes the (next) iteration of model evolution phase 330. In the iteration of model evolution phase 330, if the local model is frozen, the local model will not (or should not) evolve; otherwise, it will evolve. When the local model does not evolve, it remains unchanged, meaning the values ​​of its model parameters remain constant. When the local model evolves, it progresses from the current version to the next version, where the value of at least one model parameter changes (e.g., from a source value associated with the current version to a target value associated with the next version). The local model can evolve from the current version to the next version through intermediate versions, for example, where the value of at least one model parameter changes from a source value to a target value through an intermediate value associated with the intermediate version.

[0204] In model distribution phase 340, the Type 2 PSF notifies each Type 1 PSF that the model training process has terminated. When the Type 1 PSF is notified of the termination of the model training process 300, the Type 2 PSF may send the latest version of the local model associated with the Type 1 PSF to the Type 1 PSF. Sending may be optional, for example, if the Type 1 PSF already has the latest version of the associated local model.

[0205] exist Figure 3 In the embodiments mentioned above, the model evolution phase 330 is executed iteratively between the model collection phase 310 and the model distribution phase 340. Further details of the iteration of the model evolution phase 330 can be found in the following embodiments. Figure 4AAn exemplary process 400 of model evolution iteration (referred to as the current iteration) according to some embodiments of this disclosure is illustrated. As described above, a Type 2 PSF collects local models from multiple Type 1 PSFs. Five models are used as an example below. These five models, such as M1, M2, M3, M4, and M5, correspond to Type 1 PSF 1, Type 1 PSF 2, Type 1 PSF 3, Type 1 PSF 4, and Type 1 PSF 5, respectively. These local models are initially trained at their respective Type 1 PSFs. In operation 402, as... Figure 4A As shown, model evolution is performed at the Type 2 PSF, and in some examples, additional training data can be used for model evolution. After performing model evolution at the Type 2 PSF, intermediate versions of the local model are obtained. In operation 404, the Type 2 PSF sends the intermediate versions of the local model to the target Type 1 PSF among multiple Type 1 PSFs. For example, if the target Type 1 PSF is Type 1 PSF 1, the intermediate versions of the local model can be M'1, M'2, M'3, M'4, and M'5. The five intermediate versions of the model, namely M'1, M'2, M'3, M'4, and M'5, correspond to Type 1 PSF 1, Type 1 PSF 2, Type 1 PSF 3, Type 1 PSF 4, and Type 1 PSF 5, respectively. In operation 406, model evolution is performed at the target Type 1 PSF, and the updated local model is obtained. As shown in operation 408, M1, M2, M3, M4 and M5 in this operation refer to the updated local model or the evolved local model, which is sent from the target type 1 PSF to the type 2 PSF and can be further evolved in the next iteration of model evolution.

[0206] Specifically, in the current iteration, a target type 1 PSF is selected from the group of type 1 PSFs relative to several factors with the aim of optimizing network and training performance. Factors considered when selecting a type 1 PSF as the target type 1 PSF may include: the network conditions (e.g., throughput, latency, rate) or status (e.g., load, computing power) of the type 1 PSF, or the history of a type 1 PSF being selected as the target type 1 PSF (e.g., number of times), or any combination thereof. Network conditions can be statistical or transient and are related to the type 1 PSF. Network conditions can be notified or obtained from network functions that maintain (estimate or analyze) information about network conditions (e.g., NWDAF in 3GPP 5G systems as described in 3GPP TS 23.501 V18.2.0). The status of a type 1 PSF can be statistical or transient. This status can be notified or obtained from the type 1 PSF. The history of a type 1 PSF being selected as the target type 1 PSF can be maintained locally or notified or obtained from a storage function.

[0207] A type 1 PSF (e.g., Figure 4A Type 1 PSF 1) in the network controller (NWC) can be controlled by the network controller (NWC) (e.g., with...). Figure 4BThe NWC described in the associated system architecture is selected as the target type 1 PSF. Different type 1 PSFs can be selected as the target type 1 PSF in different iterations of model evolution. In the current iteration, if the local model is not frozen (as indicated by the corresponding type 1 PSF), the type 2 PSF updates the local model with additional data (e.g., through knowledge transfer and / or model training) to evolve the local model to an intermediate version. The type 2 PSF sends the intermediate version of the local model to the target type 1 PSF, and can also indicate to the target type 1 PSF that the local model is not frozen. If the local model is frozen (as indicated by the corresponding type 1 PSF), the type 2 PSF will not update the local model, and will send the current version of the local model as an intermediate version to the target type 1 PSF. When sending the current version of the local model to the target type 1 PSF, the type 2 PSF can also indicate to the target type 1 PSF that the local model is frozen. In some embodiments, the additional data can be used to evolve the local model from the current version to the intermediate version, including a set of data samples as described above. Data samples are selected from a dataset that may differ from any private dataset in the private dataset. This dataset can be a public dataset, located in a Type 2 PSF, or not located in a Type 2 PSF. Data samples are selected based on information about the private dataset in the target Type 1 PSF (e.g., dataset size, label distribution) (if this information is available, e.g., provided by the target Type 1 PSF during the training process) such that, for example, the number of data samples meets certain conditions, such as not being greater than or less than a ratio to the size of the private dataset, or the per-label data sample distribution meets specific conditions, such as being related to the per-label data distribution in the private dataset. In some embodiments, data samples may be processed by a processing service controller (PSC) (e.g., in conjunction with...). Figure 4B The PSC selection is described in the associated system architecture.

[0208] In the current iteration, the target type 1 PSF receives local models from the type 2 PSF (which may include its own local model, i.e., the local model associated with the target type 1 PSF), and can use its private dataset to update some or all of the received local models, as well as its own model. The local models received by the target type 1 PSF may be intermediate versions of models determined during previous iterations. When updating local models (whether received or its own), the type 1 PSF may also use other local models (e.g., by transferring knowledge from other local models). If a local model is frozen, the type 1 PSF will not update the local model, but may still use it to update other unfrozen local models. If a local model is not frozen, the target type 1 PSF can update the local module, for example, by continuing knowledge transfer and / or training, using its private dataset, so that the local model (from the intermediate version) evolves to the next version. If the local model is updated, the target type 1 PSF sends the updated local model to the type 2 PSF. If the local model is not updated, the target type 1 PSF may indicate to the type 2 PSF that the local model has not been updated. A Type 1 PSF can indicate to a Type 2 PSF whether its local model is frozen. The Type 1 PSF determines whether to freeze its local model based on its convergence status. For example, if the local model converges during an update, it is determined that the local model is frozen. If the local model does not converge during an update, it is determined that the local model is not frozen.

[0209] Examples of the apparatus mentioned in processes 200-1 and 200-2 can be found in [reference]. Figure 4B , Figure 5 and / or Figure 6 The "model update" mentioned in procedures 200-1 and 200-2 corresponds to... Figure 4AStep 406 in process 200-1 and process 200-2. The first device in process 200-1 and process 200-2 may be a target type 1 PSF. One or more devices in process 200-1 and process 200-2 may be type 1 PSFs. For each iteration of model evolution, a target type 1 PSF is selected among type 1 PSFs. For example, type 1 PSF 401 or type 1 PSF 510 or type 1 PSF A 610 is selected to perform model updates, and may then also be referred to as target type 1 PSF 401, 510 or 610, and is an example of the first device 221. Type 2 PSF 403 or first type 2 PSF 540 or type 2 PSF B 640 may be an example of the second device 220. Processing service controller (PSC) 419 or 560 may be an example of the third device 230. PCF 407 or 630 may be an example of the fourth device 240. NWC 411 or 520 may be an example of the fifth device 250. Type 2 PSF 403 or Type 2 PSF 550 may be examples of a sixth device. PMF 413 or 620 may be examples of a seventh device. Some examples will be further described with reference to one or more figures of this disclosure.

[0210] Figure 4B An exemplary architecture of system 400-1 according to some embodiments of this disclosure is shown. Figure 4B As shown, the system may include a Type 1 PSF 401, a core network (CN), and a Type 2 PSF 403. The CN may include a control plane and a data plane. The control plane includes the following control plane functions: NEF 405, PCF 407, NSF 411a, NWC 411, and PMF413. In some embodiments, the CN's control plane also includes the control plane function AMF (…). Figure 4B(AMF not shown). The data plane includes a data plane function (DPF) 415. In some embodiments, the data plane is referred to as the user plane. Type 1 PSF 401 is connected to both the control plane and the data plane. Type 2 PSF 403 is connected to the data plane. Type 1 PSF 401 has (or is associated with) a model to be trained and has (or is associated with) a private dataset that can be used by Type 1 PSF 401 to train the model. This model is a local model of Type 1 PSF 401. Type 2 PSF 403 does not have (is not associated with) a model to be trained, but can help train the model associated with Type 1 PSF 401. In some examples, Type 1 PSF 401 may be a wireless terminal device, such as a UE, or a server (e.g., AS), or a network function. In some examples where Type 1 PSF 401 is a wireless terminal device, Type 1 PSF 401 is connected to the CN (including the control plane and data plane) via radio access network (RAN) 417. In this case, the system includes RAN 417. In some examples where the Type 1 PSF 401 is not a radio terminal device, the Type 1 PSF 401 can be directly connected to the CN, and in system 400-1, RAN 417 is optional.

[0211] The system may also include a processing service controller (PSC) 419. In some embodiments, PSC 419 is part of the control plane of the CN, and Type 2 PSF 403 is part of the data plane of the CN. PSC 419 can manage / control Type 2 PSF 403. PSC 419 interacts with the control plane of the CN to manage, control, or influence control plane decisions, such as decisions regarding traffic routing. When interacting with the control plane, PSC 419 can interact with associated control plane functions (e.g., NSF 411a, PCF 407, NWC 411, and / or PMF 413). In some embodiments, for example, PSC 419 interacts directly with the associated control plane functions. In some embodiments, PSC 419 interacts indirectly with the associated control plane functions via NEF 405. In some embodiments, NWC 411 and PSC 419 may be integrated into the same network entity.

[0212] Some network elements involved in the architecture of System 400-1 can be further described as follows. The Access and Mobility Management Function (AMF) can implement the same or similar functions as the AMF in the 3GPP 5G system, such as registration management, connection management, reachability management, mobility management, access authentication, and access authorization as described in 3GPP TS 23.501 V18.2.0. In some embodiments, the AMF corresponds to the AMF in the 3GPP 5G system (i.e., 5G PCF for short). The Network Exposure Function (NEF) 405 can implement the following functions, such as, but not limited to: publicizing network information, analytics, capabilities, and events; secure configuration of information from external applications (e.g., AF) to the system; and conversion of internal to external information. In some embodiments, NEF 405 corresponds to the NEF in the 3GPP 5G system described in 3GPP TS 23.501 V18.2.0 (i.e., 5G NEF for short). The policy control function (PCF) 407 can perform functions such as, but not limited to, supporting a unified policy framework to manage network behavior, making policy decisions and providing final policy rules to control plane functions(s)(s) to enforce them, and accessing subscription information and application data related to policy decisions in one or more network storage functions (e.g., NSF 411a). In some embodiments, PCF 407 corresponds to the PCF in the 3GPP 5G system (referred to as 5G PCF) as described in 3GPP TS 23.501 V18.2.0. The network storage function (NSF) 411a can perform functions such as, but not limited to, storing and retrieving subscription data, storing and retrieving policy data, storing and retrieving structured data for public disclosure, storing and retrieving application data (including packet flow descriptions (PFDs) for application detection, information associated with AF requests, and device group information for group management), etc.

[0213] In some embodiments, NSF 411a corresponds to the UDR in the 3GPP 5G system described in 3GPP TS 23.501 V18.2.0. The network controller (NWC) 411 makes traffic routing decisions and coordinates traffic routing as described in embodiments of this disclosure. In some embodiments, NWC 411 is integrated with PMF 413. In some embodiments, NWC 411 is integrated with PCF 407. In some embodiments, NWC 411 is a separate network function, distinct from PMF 413 and PCF 407. The path management function (PMF) 413 can perform functions such as, but not limited to, selecting and controlling DPF 415, maintaining / configuring data plane topology or paths, establishing and releasing data plane tunnels, configuring traffic forwarding at DPF 415 to apply local switching or packet forwarding, configuring traffic guidance at DPF 415 to route traffic to appropriate destinations, etc. PMF 413 configures DPF 415 via a T4 interface. In some embodiments, PMF 413 corresponds to the SMF in a 3GPP 5G system as described in 3GPP TS 23.501 V18.2.0. Data plane function (DPF) 415 may perform functions such as, but not limited to, acting as an interconnection point for DNs, acting as an anchor point for devices in DPs, routing / forwarding messages, enforcing policy rules (e.g., traffic gating, redirection, traffic steering), performing traffic marking in uplink and downlink, performing message buffering and data notification triggering, and performing message detection (e.g., application detection). In some embodiments, the data plane corresponds to the user plane of a 3GPP 5G system, and DPF 415 corresponds to the UPF in a 3GPP 5G system as described in 3GPP TS 23.501 V18.2.0.

[0214] Control plane functions and DPF 415 are logical network functions, and each of them can be instantiated at one or more network locations, resulting in one or more instances. When RAN 417 implements the functionality of a logical network function (either a control plane function or DPF 415), the logical network function is considered to be instantiated at a RAN node. A network location can refer to a data center or a RAN node. There may be more than one instance of a logical network function at the same network location. Type 1 PSF 401 is connected to DPF 415 (actually, an instance of DPF 415) in the data plane of the CN. This DPF 415 is the anchor point of Type 1 PSF 401 in the data plane, which can be called DPA. DPA is in turn connected to Type 2 PSF 403. In some examples, Type 1 PSF 401 can be connected to DPA through one or more intermediate DPFs (i.e., other instances of DPF 415). One or more intermediate DPFs are located between Type 1 PSF 401 and DPA. One or more intermediate DPFs (if any), a DPA, and an intermediate communication tunnel constitute the data plane path connecting Type 1 PSF 401 and Type 2 PSF 403. The data plane path is considered associated with Type 1 PSF 401. When Type 1 PSF 401 is a radio terminal device such as a UE, it is served by the RAN node in RAN 417, i.e., it has a radio connection to the RAN node. In this case, Type 1 PSF 401 is connected to the DPA through the RAN node. The RAN node is considered part of the data plane path. If the data plane path includes one or more intermediate DPFs, these intermediate DPFs are located between the serving RAN node and the DPA.

[0215] Control plane functions (CPFs), such as AMF, PMF 413, or NWC 411, can interact with type 1 PSF 401 via a first interface, such as... Figure 6 The dashed lines in the diagram indicate that, in some embodiments, the interaction is performed via a relay function in the control plane. In this case, when the CPF interacts with the Type 1 PSF 401 via a first interface, the CPF interacts with the relay function via a second interface, which in turn interacts with the Type 1 PSF 401 via a third interface. For example, when the CPF sends a message to the Type 1 PSF 401 via the first interface (e.g., when the NWC (i.e., the CPF) forwards / sends a notification message to the Type 1 PSF 401, as shown in...), Figure 5 In operation 503 or 517 or in Figure 6As described in operation 609(c), the CPF uses the second interface to send a message to the dependent function, which then uses the third interface to forward the message to the Type 1 PSF 401. When the Type 1 PSF 401 sends a message to the CPF via the first interface (e.g., when the Type 1 PSF 401 sends a notification message to the NWC 411, such as...), Figure 5 As described in operation 513, or when type 1 PSF 401 sends an ACK message to NWC 411, such as Figure 6 As described in operation 609(c) of the protocol, type 1 PSF 401 sends a message to the relay function using the third interface, and then the relay function forwards the message to the CPF using the second interface. Before forwarding the message, the relay function may perform message mapping, in which the control plane function changes or updates some of the original information in the message to the mapped information, so that the message includes the mapped information instead of the original information when it is forwarded.

[0216] Therefore, the first interface (between the CPF and the Type 1 PSF) can be considered to include the second interface (between the control plane function and the relay function) and the third interface (between the relay function and the Type 1 PSF 401), or implemented through the second and third interfaces, with the first interface implemented through the relay function. In some embodiments, the first and third interfaces appear to be the same interface to the Type 1 PSF 401. In some embodiments, the first and second interfaces appear to be the same interface to the CPF. In some embodiments, the first, second, or third interface corresponds to a communication tunnel. In some embodiments, when the Type 1 PSF 401 is a wireless terminal device, the first interface is via RAN 417 (e.g., the serving RAN node of the Type 1 PSF 401) and is a non-access stratum (NAS) interface. Messages sent using the first interface are NAS messages, for example, when the NWC (i.e., the CPF) forwards / sends notification messages to the wireless terminal device (i.e., the Type 1 PSF 401) via RAN 417 (e.g., in...). Figure 5 In operation 503 or 517 or in Figure 6As described in operation 609(c), the notification message is a NAS message. If the first interface is implemented through the second and third interfaces, the third interface is through RAN 417 and is also a NAS interface, and the first interface is through RAN 417 because the third interface is through RAN 417. When the relay function forwards a message, the relay function may include the message or the content of the message in another message and send the other message using the third interface (if the message comes from a CPF and is targeted to Type 1 PSF 401) or the second interface (if the message comes from a Type 1 PSF 401 and is targeted to a CPF). If the other message is sent using the third interface, then the other message is a NAS message.

[0217] If the CPF is NWC 411, the relay function can be PMF 413 or AMF. Alternatively, if the CPF is PMF413, the relay function can be AMF. It should be noted that when the Type 1 PSF 401 is a wireless terminal device, the above description of the first interface between the CPF and the Type 1 PSF 401 is recursive and can be applied to the third interface. That is, if the relay function is PMF 413, the third interface can be through another relay function, such as AMF. When the Type 1 PSF 401 is not a wireless terminal device, in some examples, if the CPF is NWC 411, the relay function can be PMF 413 or NEF 405. Alternatively, in some other examples, if the CPF is PMF 413, the relay function can be NEF 405. It should be noted that when the Type 1 PSF 401 is not a wireless terminal device, the above description of the first interface between the CPF and the Type 1 PSF 401 is recursive and can be applied to the third interface. That is, if the relay function is PMF 413, the third interface can be through another relay function, such as NEF 405.

[0218] If Type 1 PSF 401 is a UE-like radio terminal device, PSC 419 notifies Type 1 PSF 401 via a NAS message. The NAS message is transmitted to Type 1 PSF 401 via the RAN node. In some embodiments, PSC 419 sends the NAS message to the RAN node, and the RAN node forwards the NAS message to Type 1 PSF 401. In some embodiments, the NAS message is generated by a network entity, and PSC 419 sends a message to the network entity, which includes the content of that message in the NAS message. In some embodiments, the NAS message is relayed to the RAN node via another network entity. When relaying the NAS message, the other network entity includes the NAS message in another NAS message and sends that other NAS message to Type 1 PSF 401.

[0219] In some embodiments, the above system architecture can be mapped to / correspond to the 3GPP 5G system architecture as described in 3GPP TS 23.501 V18.2.0. For example, DPF 415 is mapped to / corresponds to 5G UPF; AMF is mapped to / corresponds to 5G GAMF; PMF 413 is mapped to / corresponds to 5G SMF; PCF 407 is mapped to / corresponds to 5G PCF; NSF 411a is mapped to / corresponds to 5G UDR; PSC 419 is mapped to / corresponds to AF; Type 2 PSF 403 is mapped to / corresponds to AS; Type 1 PSF 401 is mapped to / corresponds to UE; NWC 411 is integrated inside 5G SMF or 5G PCF, in other words, the functionality of NWC 411 is implemented by 5G GSMF or 5G PCF.

[0220] Figure 5 The following are examples of iterations for model evolution according to some other embodiments of this disclosure (e.g.) Figure 3 An exemplary process (the operation in stage 330 of the process). Figure 5 As shown, process 500 involves at least one Type 1 PSF (taking Type 1 PSF 510 as an example), NWC 520, data plane 530, at least one Type 2 PSF (taking the first Type 2 PSF 540 and the second Type 2 PSF 550 as examples), and PSC 560. In an exemplary iteration process, Type 1 PSF 510 and the first Type 2 PSF 540 may be involved. For example, Type 1 PSF 510 is selected as the target Type 1 PSF. The iteration of model evolution involves at least one model, each of which is a local model associated with / corresponding to a Type 1 PSF. At least one model includes a local model associated with the target Type 1 PSF. For each local model among the at least one local models, PSC 560 maintains state information about the local model received from the corresponding Type 1 PSF (i.e., whether the local model is frozen). Before the model evolution iteration begins, for example, when all Type 1 PSFs send their local schemas to Type 2 PSF 540 during the model distribution phase, Type 2 PSF 540 may have already collected at least one latest version of the model. At the end of the model evolution iteration, Type 2 PSF 550 may collect at least one latest version of the model, for example, as... Figure 5Operation 519 is shown in the diagram. In some embodiments, the first type 2 PSF 540 and the second type 2 PSF 550 are the same entity. In some embodiments, the first type 2 PSF 540 and the second type 2 PSF 550 are different entities. In at least one model, the model can be frozen and evolve during iterations of model evolution (in other words, remain unchanged). In this case, the model parameters have the same values ​​at the end of the iteration of model evolution as before the start of the iteration of model evolution. In some examples, process 500 can be process 200-1 (e.g., ...). Figure 2A (as shown) and / or process 200-2 (as shown) Figure 2B Examples of interactions between entities mentioned in the diagram. In these examples, type 1 PSF 510 may be an example of first device 221, and first type 2 PSF 540 may be an example of second device 220.

[0221] In operation 501, type 1 PSF 510 is selected as the target type 1 PSF, the application location corresponding to the first type 2 PSF 540 is selected from one or more potential application locations corresponding to one or more type 2 PSFs, and the data plane 530 is accordingly configured such that the data traffic in operation 509 will be routed from the first type 2 PSF 540 to the type 1 PSF 510.

[0222] This operation 501 can be combined with Figure 6 The described process is executed where Type 1 PSF A 610 is Type 1 PSF 510 and Type 2 PSF B 640 is Type 2 PSF 540. In this operation, the PSC 560 request affects the traffic routing used by the application. The request from the PSC 560 is transmitted ( Figure 6 Operation 603) to PCF 630. The request includes information about traffic, information about Type 1 PSFs, and information about application locations. In operation 509, the information about traffic identifies data traffic. The information about Type 1 PSFs identifies a set of Type 1 PSFs associated with at least one model. The information about application locations identifies one or more potential application locations corresponding to one or more Type 2 PSFs.

[0223] If the first type 2 PSF 540 does not have the latest version of the model in at least one model, then in operation 501, the PSC 560 can configure the first type 2 PSF 540 to obtain the latest version of the model. The latest version of the model is stored in a third type 2 PSF (which may be the same as or different from the second type 2 PSF 550). For example, when in Figure 6When configuring Type 2 PSF B 640 in operation 609(b) or 619, PSC 560 provides Type 2 PSF 540 with model identification information (e.g., ID, name, or index), version information (e.g., version number), and information about Type 2 PSF (e.g., ID or network address). The model identification information identifies the model. The version information identifies the latest version of the model. Based on the information provided from PSC 560, Type 2 PSF 540 can interact with Type 2 PSF 540 to obtain the latest version of the model.

[0224] In operation 503, PSC 560 instructs Type 1 PSF 510 to perform a model update for at least one model. In operation 503, PSC 560 may notify Type 1 PSF 510 whether one or more models in the at least one model are frozen. The one or more models may include all models in the at least one model. In operation 509, at least one model may be transferred to Type 1 PSF 510.

[0225] In some embodiments, PSC 560 performs operation 503 by sending a control command to Type 1 PSF 510. In some embodiments, PSC 560 performs operation 503 by sending a message (referred to as a notification message) to Type 1 PSF 510. The notification message includes a list of one or more model identification information, one or more corresponding version information, and one or more corresponding status indications. In the list, each model identification information (e.g., ID, name, or index) identifies a model in one or more models, and the corresponding version information (e.g., version number) indicates the freshness of the model, and the corresponding status indication indicates whether the model is frozen. In some embodiments, the model identification information and / or one or more corresponding version information may be generated by PSC 560. In some embodiments, the notification message may be sent to Type 1 PSF 510 via a control plane function (e.g., NWC 520 or AMF). That is, PSC 560 sends the notification message to the control plane function, and then the control plane function forwards / sends the notification message to Type 1 PSF 510. In some embodiments, such as when the first type 2 PSF 540 is configured to send a message to the type 1 PSF including information (i.e., a list of (one or more) model identification information, (one or more) corresponding version information, and (one or more) corresponding status indications) as part of operation 509, operation 503 is optional. In other words, the information sent in operation 503 may alternatively be sent in operations 505 and 509. Control commands or messages (or notification messages) from PSC 560 to the type 1 PSF 510 may be examples of the aforementioned indication 235.

[0226] In operation 505, PSC 560 configures Type 2 PSF 540, specifically instructing Type 2 PSF 540 to perform model updates for at least one model. In this operation 505, PSC 560 may notify Type 2 PSF 540 whether one or more models are frozen. The one or more models may include all models in the at least one model, as described in operation 503. In some embodiments, PSC 560 performs operation 505 by sending a message to Type 2 PSF 540 (an example of the fourth message 245 described above). This message includes a list of model identification information (one or more), corresponding version information (one or more), and corresponding status indications (one or more), as described in operation 503. For the model identified in the message, the message may include an indication of whether additional training data is required. If additional training data is required, the indication may also indicate where the additional training data is obtained and how to select it. The additional training data will be used to perform operation 507.

[0227] In operation 507, the Type 2 PSF 540 performs model evolution; specifically, the Type 2 PSF 540 updates one or more models from at least one model. The models updated in operation 507 can be models from models that are not frozen (as notified by PSC 560 in operation 503). If at least one model is not frozen (e.g., as notified by PSC 560 in operation 505), then in operation 507, the Type 2 PSF 540 can update the model. When updating the model, some or all of the values ​​of the model's model parameters are updated (changed), and the version information associated with the model is also updated / changed (e.g., the version number is incremented). If at least one model is frozen (e.g., as notified by PSC in operation 505), then in operation 507, the Type 2 PSF 540 will not update the model accordingly. Since the model is not updated, the values ​​of the model's model parameters remain unchanged, and the version information associated with the model can also remain unchanged. In some embodiments, even if the model is not updated, the version information associated with the model will be updated / changed (e.g., the version number is incremented).

[0228] In operation 509, Type 2 PSF 540 sends an updated model, i.e., information about at least one model, to Type 1 PSF 510. This information is included in the data traffic routed to Type 1 PSF 510 via data plane 530. The destination of the data traffic is Type 1 PSF. For the model in at least one model, the information may include model identification information (e.g., ID, name, or index), version information (e.g., version number), values ​​of the model's model parameters, and a status indication. The model identification information is used to identify the model. The version information is associated with the model and corresponds to the values ​​of the model's model parameters, and it indicates the freshness of the model. The status indication indicates whether the model is frozen. If the model has been updated in operation 507, the values ​​of the model's model parameters include the updated values, the version information is the updated version information, and the status indication indicates that the model is frozen. If the model has not been updated in operation 507, the values ​​of the model's model parameters and the version information are the same as the values ​​and version information before this procedure. In some embodiments, the values ​​of the model parameters are optional (e.g., when they are pre-configured at Type 1 PSF 510) and can be identified by Type 1 PSF 510 in its local configuration based on model identification information and, possibly, version number information. This status indication is consistent with the status indication described for the model in operation 505. If the model status indication in operation 505 indicates that the model is frozen, then the model status indication in operation 507 also indicates that the model is frozen. If the model status indication in operation 505 indicates that the model is not frozen, then the model status indication in operation 507 also indicates that the model is not frozen. In some examples, the model identification information, version information, and status indication can be in a message from first Type 2 PSF 540 to Type 1 PSF 510 (an example of a first message). In some examples, operation 503 is optional, and if operation 503 is not required, indication 225 instead of indication 235 in the first message can be indicated to Type 1 PSF 510. Instruction 225 includes the same information as instruction 235, but instruction 225 is sent from type 2 PSF 540 to type 1 PSF 510, for example, via the message above (example of the first message).

[0229] In operation 511, type 1 PSF 510 performs model evolution, specifically updating one or more models in at least one model. The one or more models updated in operation 511 may include some or all models that are not frozen (as notified by PSC 560 in operations 505 or 507). In some embodiments, one or more models include models associated with type 1 PSF 510, and in operations 505 or 507, models associated with type 1 PSF 510 may be indicated as frozen. If at least one model is not frozen (i.e., not frozen) (as notified by PSC 560 in operation 505), then in operation 511, type 1 PSF 510 may update the model. When updating a model, the values ​​of some or all model parameters are updated (changed), and the version information associated with the model is also updated / changed (e.g., the version number is incremented). If at least one model is frozen (as notified by PSC 560 in operation 505) (and in some embodiments, if the model does not correspond to Type 1 PSF 510), then in operation 511, Type 1 PSF 510 will not update the model accordingly. Since the model is not updated, the values ​​of the model parameters remain unchanged, and the version information associated with the model can also remain unchanged. In some embodiments, even if the model is not updated, the version information associated with the model will be updated / changed (e.g., the version number is incremented).

[0230] In operation 513, Type 1 PSF 510 notifies PSC 560 of the status (i.e., frozen or not frozen) of a model in at least one model by sending a message (referred to as a notification message, which may be an example of first notification 255) to PSC 560. In some embodiments, the notification message may be sent to PSC 560 via a control plane function (e.g., NWC 520 or AMF). That is, Type 1 PSF 510 sends the notification message to the control plane function, and then the control plane function forwards / sends the message to PSC 560. The notification message may include a status indication to indicate whether the model is frozen. The notification message may also include model identification information (e.g., ID, index, or name) that identifies the model. In various embodiments, the model is associated with Type 1 PSF 510, and the status of the model is determined by Type 1 PSF 510.

[0231] In some embodiments, the state of the model indicated in this operation is inconsistent (or different) from the state of the model indicated in operation 505 or 507. For example, the model is indicated as frozen in operation 505 or 507 but is not frozen in operation 513, or the model is indicated as not frozen in operation 505 or 507 but is frozen in operation 513. In some other embodiments, the state of the model indicated in operation 513 is consistent (or the same) as the state of the model indicated in operation 505 or 507. In some embodiments, the notification message sent to Type 1 PSF 510 also indicates that a model update for at least one model is complete.

[0232] In operation 515, an application location corresponding to type 2 PSF 550 is selected from one or more application locations, and data plane 530 is configured accordingly so that data traffic in operation 519 will be routed to type 2 PSF 550.

[0233] This operation 515 can be combined Figure 6 The described process is executed where Type 1 PSF A 610 is Type 1 PSF 510, and Type 2 PSF B 640 is Type 2 PSF 550. In operation 515, PSC 560 requests to influence the application's traffic routing. The request from PSC 560 is transmitted ( Figure 6 Operation 603 in the process leads to PCF 630. This request includes information about traffic, information about Type 1 PSFs, and information about application locations. In operation 519, the information about traffic identifies data traffic. The information about Type 1 PSFs identifies Type 1 PSF A 610. The information about application locations identifies one or more potential application locations.

[0234] In operation 515, PSC 560 can, for example, notify the second type 2 PSF 550 of at least one model that the second type 2 PSF 550 can expect to receive in the data traffic by providing a list of model IDs, names, or indexes that identify one or more models. For at least one model that the second type 2 PSF 550 will not receive in the data traffic, if the second type 2 PSF 550 does not have the latest version of the model, then in operation 515, PSC 560 can configure the second type 2 PSF 550 to obtain the latest version of the model. The latest version of the model is stored in a fourth type 2 PSF (which may be the same as or different from the second type 2 PSF 550 or the third type 2 PSF). For example, when in Figure 6When configuring Type 2 PSF B 640 in operation 609(b) or 619, PSC 560 provides the second Type 2 PSF 550 with model identification information (e.g., ID, name, or index), version information (e.g., version number), and information about the fourth Type 2 PSF (e.g., ID or network address). The model identification information identifies the model, and the version information identifies the latest version of the model. Based on the information provided from PSC 560, the second Type 2 PSF 550 can interact with the fourth Type 2 PSF to obtain the latest version of the model. In some examples, the first Type 2 PSF 540 and the second Type 2 PSF 550 can be the same Type 2 PSF, and PSC 560 can send a notification to the first Type 2 PSF 540 (an example of the second notification above) to notify that the first Type 2 PSF 540 has been selected to receive at least one latest version of one or more models from Type 1 PSF 510, wherein at least one latest version is obtained through a model update performed by Type 1 PSF 510.

[0235] In operation 517, PSC 560 sends a control command to Type 1 PSF 510. Specifically, PSC 560 notifies Type 1 PSF 510 of the latest version of at least one model by sending a message (referred to as a notification message, an example of the third message mentioned above) to Type 1 PSF 510.

[0236] In some embodiments, the notification message and the response to operation 513 may be integrated. If the notification message from type 1 PSF 510 to PSC 560 in operation 513 is sent via a control plane function (e.g., NWC or AMF), then the notification message in operation 517 is sent to type 1 PSF 510 via the control plane function. That is, PSC 560 sends the notification message to the control plane function, and then the control plane function forwards / sends the notification message to type 1 PSF 510. In some embodiments, PSC 560 notifies type 1 PSF 510 of the latest version of some models among at least one model, and the notification message may include information identifying some of the models among at least one model (e.g., a list of one or more pattern IDs, one or more names, or indexes).

[0237] In operation 519, Type 1 PSF 510 sends an updated model to Type 2 PSF 550. Specifically, Type 1 PSF includes the latest version of at least one model or some of the models identified in operation 519 in the data traffic and sends the traffic to Type 2 PSF 550 via data plane 530. The target of the data traffic is Type 2 PSF.

[0238] For each of these models, the Type 1 PSF 510 includes the following information in the traffic: model identification information (e.g., ID, name, or index), version information (e.g., version number), and the values ​​of the model's parameters. The model identification information is used to identify the model. The version information is associated with the model and corresponds to the values ​​of the model's parameters, and it indicates the model's freshness. The Type 1 PSF 510 may include the latest version of its own local model in the traffic. The Type 1 PSF 510 may also include a status indication of its own local model in the traffic to indicate whether its local model is frozen (i.e., frozen or not frozen). In some embodiments, the Type 1 PSF 510 is configured to include this indication in the traffic only when the instructions in operation 503 are configured to be integrated with operations 505 and 509. Data plane 530 transmits traffic to the second Type 2 PSF 550, such that the second Type 2 PSF 550 receives the latest version of at least one model or some of the models in at least one model.

[0239] In operation 521, type 2 PSF 550 sends a notification to PSC 560. This notification may indicate that PSC 560 has received the model as indicated in operation 515. The notification may also include a status indication of a model, indicating whether the model is frozen. This model is a local model of type 1 PSF 510, and the indication is consistent with (or identical to) the indication described in operation 519.

[0240] Figure 6 Exemplary procedures for requests to influence traffic routing, according to some embodiments of this disclosure, are illustrated. Figure 6 As shown, process 600 involves at least one Type 1 PSF (taking Type 1 PSF A 610 as an example), NWC 520, PMF 620, PCF 630, data plane 530, at least one Type 2 PSF (taking Type 2 PSF B 640 as an example), and PSC 560. PSC 560 can be used... Figure 6 The process request shown affects the application's traffic routing. Traffic routing is affected, causing the application's traffic (e.g., ...) to be disrupted. Figure 5 The traffic in operation 509 or 519) follows the data plane path in type 1 PSF A 610 (e.g. Figure 5 Type 1 PSF 510) and Type 2 PSF B 640 (e.g.) Figure 5Transmission is made between either Type 2 PSF 540 or Type 2 PSF 550. Type 1 PSF A 610, Type 2 PSF B 640, and the data plane path are dynamically selected based on information in the request from PSC 560. Type 1 PSF A 610 is selected from a set of Type 1 PSFs identified in the request. Type 2 PSF B 640 corresponds to the application location and is selected as a result of selecting an application location from one or more potential application locations identified in the request. The data plane path includes one or more DPFs and connects Type 1 PSF A 610 and Type 2 PSF B 640. This dynamic selection may take into account network conditions (e.g., throughput, congestion, and latency) and PSF conditions, i.e., the conditions of Type 1 PSF A 610 and Type 2 PSFs (e.g., load, remaining energy, computing resource availability, or sufficiency), to balance the load and optimize overall performance.

[0241] The request from PSC 560 is transmitted to PCF 630, such as Figure 6 Operation 603 is shown in the diagram. In some embodiments, a request from PSC 560 is sent directly to PCF 630 or via NEF. In some embodiments, a request from PSC 560 is sent directly to NSF or via NEF (NSF does not...). Figure 6 (As shown in the diagram). The NSF stores the request as application data. In this case, the PCF 630 performs a subscription to receive application-related data from the NSF, and the NSF provides the request to the PCF 630 based on the subscription. The PCF 630 can perform a subscription when it receives an application-related policy request (e.g., from the NWC 520).

[0242] When a request is transmitted from PSC 560 to PCF 630 involving NEF as described above (i.e. Figure 6 In operation 603, the NEF receives a request from the PSC 560 and sends it to the next network entity (i.e., the PCF 630 or NSF). The NEF processes the request before sending it to the next network entity. When processing the request, the NEF may perform information mapping, whereby the NEF replaces / updates portions of the information in the request with mapped information so that the next network entity receives the request with the mapped information. For example, the NEF may map a service identifier in the request to a DNN, or a combination of DNN and network slice information, and / or map an external ID to an internal ID.

[0243] PCF 630 translates this request into one or more policies (in other words, generates one or more policies based on the request), such as Figure 6As shown in operation 605, one or more policies are sent to the NWC 520, such as Figure 6 As shown in operation 607, the NWC 520 makes traffic routing decisions based on one or more policies (e.g., ...). Figure 6 Operation 609 in the middle) and interact with PMF 620 (e.g. Figure 6 Operations 611 and 615 in the above steps are used to implement traffic routing decisions. When interacting with the PMF 620, the NWC 520 provides traffic routing information to the PMF 620, which specifies the traffic routing decision. In some embodiments, the NWC 520 and PMF 620 are integrated as a single network entity. In some embodiments, the NWC 520 and PCF 630 are integrated as a single network entity.

[0244] The traffic routing decisions made by the NWC 520 include a Type 1 PSF selection decision (i.e., selecting Type 1 PSF A 610) and / or an application location selection decision (i.e., selecting an application location corresponding to Type 2 PSF B 640). In some embodiments, the traffic routing decisions made by the NWC 520 include a data plane path (re)selection decision. In some embodiments, the traffic routing decisions made by the NWC 520 do not include a data plane path (re)selection decision, and the data plane path (re)selection decision is instead made by the PMF 620 based on the traffic routing decisions made by the NWC 520 (i.e., the Type 1 PSF selection decision and / or the Type 2 PSF selection decision). When implementing traffic routing decisions, the PMF 620 is configured (e.g., ... Figure 6 Operation 613) in the data plane path allows traffic to be transmitted along the data plane path between Type 1 PSF A 610 and Type 2 PSF B 640. Further details of procedure 600 can be found in operations 601 to 621 below.

[0245] In operation 601, PSC 560 selects one or more potential locations for the application (in other words, potential application locations), each potential location corresponding to a Type 2 PSF. The Type 2 PSF(s) corresponding to the one or more potential application locations can be used to support the application. When selecting one or more potential application locations, PSC 560 may consider loading the corresponding Type 2 PSF(s).

[0246] In Operation 603, the PSC 560 request affects traffic routing for the application. The request from PSC 560 is forwarded to PCF 630. In some examples, the request is sent as a message (e.g., the fifth message mentioned in procedures 200-1 or 200-2 above). In some examples, the request may include information about traffic (i.e., examples of data traffic information identifying data traffic of at least one application (in some embodiments, at least one application may be associated with a second or more devices), as included in the fifth message in process 200-1 or 200-2), information about Type 1 PSF (i.e., examples of information about the first or more devices, as included in the fifth message in process 200-1 or 200-2), information about application location (i.e., examples of information about at least one location of at least one application (in some embodiments, at least one application may be associated with a second device), as included in the fifth message in process 200-1 or 200-2), and information about traffic routing requirements (i.e., examples of information about one or more requirements associated with traffic routing, as included in the fifth message in process 200-1 or 200-2).

[0247] In some examples, the traffic information identifies the application's traffic. The traffic identified in this information may include traffic destined for or to a Type 1 PSF (referred to as downlink traffic) and traffic destined for or to a Type 2 PSF (referred to as uplink traffic). In some examples, this information may include a service identifier, or a DNN, or a combination of a DNN and slice information (e.g., S-NSSAI). The service identifier corresponds to a DNN, or a combination of a DNN and slice information. Additionally, this information may also include traffic filtering information, or, when traffic filtering information is pre-configured (e.g., at the DPF), an application identifier corresponding to the traffic filtering information. The application identifier may also identify the application. Traffic filtering information may be represented using any combination of the following: destination address, source address, source port, destination port, transport protocol (e.g., protocol ID or name), and will be used by the DPF to detect traffic. In some examples, the information about a Type 1 PSF may include information identifying the group of Type 1 PSFs, such as a group ID or list of IDs or network addresses, each identifying a Type 1 PSF. Information regarding application locations may describe one or more potential application locations selected in operation 601. In this information, each potential application location may be represented or identified by an ID or network address, and in some embodiments, the ID or network address is associated with a Type 2 PSF corresponding to the potential application location. Information regarding traffic routing requirements describes the traffic routing requirements. This information may indicate that an application location should be selected from one or more potential application locations. This information may indicate that a Type 1 PSF should be selected from this group of Type 1 PSFs. This information may also indicate that traffic routing should be performed between the application location and the Type 1 PSF for traffic identified in the information regarding traffic, wherein uplink traffic (if any) is routed to the application location and downlink traffic (if any) is routed to the Type 1 PSF.

[0248] In operation 605, PCF 630 generates one or more policies based on the request received in operation 601. The one or more policies include information about traffic, information about application location, and information about traffic routing requirements. The one or more policies may also include information about Type 1 PSFs. Details of the information included in the one or more policies can be referenced to the same information included in the request described in operation 603.

[0249] In operation 607, PCF 630 sends one or more policies (at least one example of policy 275) to NWC 520, which makes traffic routing decisions for the application's traffic. In some embodiments, NWC 520 is associated with a service identifier or a DNN, or a combination of DNN and slice information, and is selected using a service identifier or a service DNN, or a combination of DNN and slice information. In some embodiments, NWC 520 is associated with an application and is selected using an application identifier.

[0250] In operation 609, NWC 520 makes a traffic routing decision, wherein type 1 PSF A 610 and type 2 PSF B 640 are selected according to one or more policies. Type 2 PSF B 640 is selected by selecting the application location corresponding to type 2 PSF B 640. Operation 609 may include the following sub-operations 609(a) to 609(c).

[0251] In operation 609(a), NWC 520 selects an application location corresponding to Type 2 PSF B 640 and one or more Type 1 PSFs according to one or more policies. The one or more Type 1 PSFs include Type 1 PSF A 610. In some examples, the application location may be selected from one or more potential application locations based on traffic routing requirements. For each application location, one or more Type 1 PSFs are selected from that group of Type 1 PSFs based on traffic routing requirements. In some embodiments, NWC 520 may consider optimizing the efficiency of data plane 530 when selecting the application location and one or more Type 1 PSFs.

[0252] In operation 609(b), NWC 520 sends a notification (e.g., in the form of a message) to PSC 560. The notification (i.e., the notification message) may be about a data plane management event and is considered an early notification (i.e., a notification sent before the data plane management event occurs). The data plane management event may be about the (re)selection of an application location (or a change in application location). In this case, the notification includes information identifying the application location, such as an ID or network address, and indicates that the application location has been selected. The data plane management event may also be about the (re)selection of one or more Type 1 PSFs. In this case, the notification also includes information identifying one or more Type 1 PSFs, such as a list of one or more IDs or one or more network addresses. In some embodiments, PSC 560 may send an acknowledgement (ACK) to NWC 520, for example, in the form of a message, to acknowledge the recipient of the notification. The ACK may be an affirmative ACK for confirming (agreeing to) the selection of the application location, or a negative ACK for rejecting (disagreeing to) the selection of the application location. The ACK includes information indicating whether the confirmation is an affirmative or negative ACK. When the ACK is a positive ACK, the PSC 560 can be configured with type 2 PSF B 640 corresponding to the application location before sending the ACK, for example, to obtain... Figure 5 The latest version of one or more models described in Operation 501. In some examples, NWC 520 can be integrated into PCF 630, and then a message (example of sixth message 285 above) can be sent from PCF 630 to PSC 560, wherein sixth message 285 indicates type 2 PSF B 240 (example of second device 220) and one or more type 1 PSFs (example of at least one candidate device), specifically, for example, by indicating a data plane management event, sixth message 285 can be about (re)selecting an application location (or application location change), and about (re)selecting (one or more) type 1 PSFs.

[0253] In operation 609(b), when multiple Type 1 PSFs are identified in the notification, PSC 560 can make a further selection (i.e., downward selection), whereby a single Type 1 PSF (i.e., Type 1 PSF A610) is selected from the multiple Type 1 PSFs. In this case, the positive ACK (i.e., ACK message) sent to NWC 520 includes information identifying Type 1 PSF A610, such as its ID or network address. If NWC 520 receives a negative ACK in this operation, NWC 520 can repeat operation 609 so that different Type 2 PSFs can be selected and / or different one or more Type 1 PSFs can be selected.

[0254] In operation 609(c), if the ACK received from PSC 560 is a positive ACK, NWC 520 notifies Type 1 PSF A 610 of its selection by sending a message (referred to as a notification message, an example of the second message above) to Type 1 PSF A 610. The notification message includes information identifying the application. Based on the notification, Type 1 PSF A 610 knows that it will receive data traffic related to the application and will perform a model update. In some examples, Type 1 PSF A 610 may, for example, send an acknowledgement (ACK) to NWC in the form of a message. The ACK can be a positive ACK to acknowledge (agree to) the selection of Type 1 PSF A 610, or a negative ACK to reject (disagree with) the selection of Type 1 PSF A. In some examples, the ACK (i.e., the ACK message) may include information indicating whether the ACK is a positive or negative ACK. If the NWC520 receives a negative ACK in operation 609(c), the NWC520 can repeat operation 609 so that a different type 1PSF can be selected.

[0255] In some embodiments, Type 1 PSF A 610 can be an AS, and NWC 520 can send notifications to the AF (subscribed notification). Figure 6 Notifications (i.e., notification messages) are sent via NEF (not shown in the image). Notifications can be sent via NEF (...). Figure 6 (Not shown in the image) is sent to AF. NEF can be referenced. Figure 4B After receiving the notification, the AF can then notify the AS, enabling the AS to receive the information from the notification sent from the NWC 520.

[0256] In operation 611, NWC 520 provides traffic routing information to PMF 620 for the application. The traffic routing information includes information identifying the application location (e.g., ID or network address) and information identifying the Type 1 PSF A 610 (e.g., ID or network address). The traffic routing information may also include information about the traffic, which is received by NWC 520 from PCF 630 in operation 607. The traffic routing information indicates that traffic routing should be performed between the application location and the Type 1 PSF for the traffic identified in the information about the traffic, wherein uplink traffic (if any) is routed to the application location and downlink traffic (if any) is routed to the Type 1 PSF.

[0257] In some embodiments, NWC 520 performs data plane path (re)selection for Type 1 PSF A 610 (e.g., to improve the efficiency of data plane 530), wherein NWC 520 selects or reselects a data plane path according to one or more policies received from PCF 630. The data plane path may include one or more DPFs and connect Type 1 PSF A 610 and the application location. One or more DPFs may be (re)selected as part of a data plane path relative to the application location. NWC 520 may include information about the data plane path in the traffic routing information sent to PMF 620. The information about the data plane path specifies the data plane path, for example by including information identifying one or more DPFs (e.g., a list of IDs or network addresses). When traffic routing information includes information about the data plane path (i.e., specifies the data plane path), it implies that traffic routing should be performed along the data plane path as described in the information about the data plane path. In some embodiments, the NWC 520 does not perform data plane path (re)selection, and the PMF 620 is used to perform data plane path (re)selection as described in operation 613. In this case, the traffic routing information sent to the PMF 620 does not include information about the data plane path.

[0258] In operation 613, PMF 620 performs a (re)selection of the data plane path for Type 1 PSF A 610 (e.g., to improve the efficiency of data plane 530), wherein PMF 620 selects or reselects the data plane path based on traffic routing information received from NWC 520. The data plane path includes one or more DPFs and connects Type 1 PSF A 610 and the application location. One or more DPFs can be (re)selected as part of the data plane path relative to the application location. The application location is identified in the traffic routing information.

[0259] In some embodiments, operation 613 is optional, for example, when NWC 520 is configured to perform data plane path (re)selection, or when traffic routing information (as described in operation 611) specifies a data plane path. When NWC 520 performs data plane path (re)selection, the traffic routing information includes information about the data plane path that specifies the data plane path.

[0260] In operation 615, the PMF 620 configures the data plane path to support or implement traffic routing. As described above, the data plane path can be selected by the NWC 520 or the PMF 620 and includes one or more DPFs. When configuring the data plane path, the PMF 620 configures one or more DPFs to establish one or more communication tunnels between them. The PMF 620 can also provide traffic processing rules for one or more of the DPFs. In operation 611, the PMF 620 generates traffic processing rules based on the traffic routing requirements received from the NWC 520. One or more DPFs will detect and route traffic according to the traffic processing rules. When routing traffic, one or more DPFs will route the traffic along the data plane path through one or more communication tunnels.

[0261] In operation 617, PMF 620 notifies NWC 520 that traffic routing has been configured in data plane 530; in other words, data plane 530 has been configured to support or implement traffic routing. Operation 617 can be executed when PMF 620 responds to NWC 520 to acknowledge the recipient of the traffic routing information; that is, the notification and the response can be integrated. This notification can be an example of the third notification described above.

[0262] In operation 619, NWC 520 sends a notification to PSC 560. This notification concerns a data plane management event and is considered a post-event notification (i.e., a notification sent after the data plane management event has occurred). In some examples, the data plane management event relates to the (re)selection of an application location (or a change in application location). In this case, the notification includes information identifying the application location, such as an ID or network address, and indicates that the application location has been selected. In some examples, the data plane management event may also relate to the (re)selection of one or more Type 1 PSFs. In this case, the notification may also include information identifying a Type 1 PSF A 610, such as an ID or network address.

[0263] The PSC 560 can, for example, send an acknowledgment (ACK) to the recipient of the acknowledgment notification to the NWC 520 in the form of a message. The ACK can be an affirmative ACK for confirming (agreeing to) the selection of the application location, or a negative ACK for rejecting (disagreeing with) the selection of the application location. The ACK can include information indicating whether the acknowledgment is an affirmative ACK or a negative ACK. When the ACK is an affirmative ACK, the PSC 560 can be configured with the type 2PSF B 640 corresponding to the application location before sending the ACK, for example, obtaining... Figure 5 The latest version of one or more models described in Operation 501.

[0264] Following operation 619, in operation 621, the application's traffic is transmitted between type 1 PSF A 610 and type 2 PSF B 640 via the data plane, more specifically, via the data plane path configured in operation 613.

[0265] Figure 7 A flowchart illustrating an exemplary method 700 implemented at a first device according to some embodiments of the present disclosure is shown. Figure 7 As shown, at block 710, the first device may receive an instruction to perform a model update for one or more models associated with the first or more devices. The first device is one of the first or more devices. At block 720, the first device may send a first notification to notify that a model update for one or more models has been completed. In some embodiments, the first device for performing method 700 may be the first device 221 in process 200-1 and / or the first device in process 200-2. The operations in method 700 performed by the first device may also refer to the embodiments mentioned in processes 200-1 and / or 200-2 above.

[0266] Figure 8 A flowchart illustrating an exemplary method 800 implemented at a second device (e.g., the second device 220 described above) according to some embodiments of the present disclosure is shown. Figure 8 As shown, at block 810, the second device can receive a fourth message from the third device, the fourth message indicating whether additional training data should be used by the second device to perform model updates of one or more models associated with the first or more devices. The second device is one of the second or more devices, and the additional training data differs from the training data used at the first or more devices to perform the model updates. At block 820, the second device can update one or more models based on receiving the fourth message. In some embodiments, the second device for performing method 800 can be the second device 220 in process 200-1 and / or the second device in process 200-2. The operations performed by the second device in method 800 can also refer to the embodiments mentioned in processes 200-1 and / or process 200-2 above.

[0267] Figure 9 A flowchart illustrating an exemplary method 900 implemented at a third device (e.g., the third device 230 or 231 described above) according to some embodiments of the present disclosure is shown. Figure 9As shown, at box 910, the third device can send a fourth message to the second device, the fourth message indicating whether additional training data should be used by the second device to perform model updates of one or more models associated with the first or more devices, wherein the second device is one of the second or more devices, and the additional training data is different from the training data used to perform model updates at the first or more devices.

[0268] Figure 10 A flowchart illustrating an exemplary method 1000 implemented at a third device (e.g., the third device 230 or 231 described above) according to some other embodiments of this disclosure is shown. Figure 10 As shown, at block 1010, the third device can send a fifth message to the fourth device to request influence on traffic routing. Influence on traffic routing is used to select a second device from the second or more devices and to select at least one candidate device from the first or more devices. The second device and the at least one candidate device jointly perform model updates for one or more models associated with the first or more devices. The fourth device is configured to perform policy control functions. At block 1020, the third device can receive a sixth message from the fifth device to instruct the second device and the at least one candidate device, wherein the fifth device is configured as a network controller.

[0269] In some embodiments, the third means for performing method 900 and / or method 1000 may be the third means 230 in process 200-1 and / or the third means 231 in process 200-2. The operation performed by the third means in method 900 and / or method 1000 may also refer to the embodiments mentioned in process 200-1 and / or process 200-2 above.

[0270] Figure 11 A flowchart illustrating an exemplary method 1100 implemented at a fourth device (e.g., the fourth device 240 described above) according to some embodiments of the present disclosure is shown. Figure 11As shown, in block 1110, the fourth device can receive a fifth message from the third device requesting influence on traffic routing. Influence on traffic routing is used to select a second device from the second or more devices and to select at least one candidate device from the first or more devices. The second device and the at least one candidate device jointly perform model updates for one or more models associated with the first or more devices. The fourth device is configured to perform policy control functions. At block 1120, the fourth device can generate at least one policy regarding influence on traffic routing based on the fifth message. In some embodiments, the fourth device for performing method 1100 can be the fourth device in process 200-1 and / or the fourth device 240 in process 200-2. The operations performed by the fourth device in method 1100 can also refer to the embodiments mentioned in processes 200-1 and / or 200-2 above.

[0271] Figure 12 A flowchart illustrating an exemplary method 1200 implemented at a fifth device (e.g., the fifth device described above) according to some embodiments of the present disclosure is shown. Figure 12 As shown, at block 1210, the fifth device can receive from the fourth device at least one policy regarding influencing traffic routing. Influencing traffic routing is used to select a second device from a second or more devices and to select at least one candidate device from the first or more devices. The second device and the at least one candidate device jointly perform model updates for one or more models associated with the first or more devices. The fourth device is configured to perform policy control functions, and the fifth device is configured as a network controller. At block 1220, the fifth device can select the second device and at least one candidate device based on at least one policy. At block 1230, the fifth device can send a sixth message to a third device to indicate the second device and at least one candidate device. The third device is used to control the second or more devices. In some embodiments, the fifth device for performing method 1200 can be the fifth device in process 200-1 and / or the fifth device 250 in process 200-2. The operations performed by the fifth device in method 1200 can also refer to the embodiments mentioned in processes 200-1 and / or 200-2 above.

[0272] Figure 13A flowchart of an exemplary method 1300 implemented at a system according to some embodiments of the present disclosure is shown. Exemplary method 1300 may include one or more operations of a first device, a second device, a third device, a fourth device, and a fifth device as described above. For example, at block 1310, the third device may send a fourth message indicating whether additional training data should be used by a second device to perform a model update of one or more models associated with the first or more devices, wherein the second device is one of the second or more devices, and the additional training data differs from the training data used at the first or more devices to perform the model update. At block 1320, the second device may receive the fourth message from the third device and update one or more models based on receiving the fourth message. At block 1330, the first device may receive an instruction from the second or third device instructing the first device to perform a model update and send a first notification notifying that a model update of one or more models has been completed, wherein the first device is one of the first or more devices. At block 1340, the third device can control the second or more devices and send a fifth message to the fourth device to request influence on traffic routing, wherein influencing traffic routing is used to select at least one candidate device among the second device and the first or more devices, and wherein the second device and the at least one candidate device jointly perform a model update. At block 1350, the fourth device can receive the fifth message from the third device to generate at least one policy regarding influence on traffic routing based on the fifth message, and send the at least one policy to the fifth device. At block 1360, the fifth device can receive at least one policy from the fourth device, and the fifth device is also configured as a network controller and configured to select the second device and at least one candidate device based on the at least one policy, and send a sixth message to the third device to indicate the second device and the at least one candidate device.

[0273] Figure 14 This is a simplified block diagram of a device 1400 suitable for implementing some embodiments of the present disclosure. Device 1400 can be considered as yet another example embodiment of the first, second, third, fourth, or fifth device described above. Therefore, device 1400 can be implemented at or as at least a part of the aforementioned devices.

[0274] As shown in the figure, device 1400 includes a processor 1410, a memory 1420 coupled to the processor 1410, suitable transmitters (TX) and receivers (RX) 1440 coupled to the processor 1410, and a communication interface coupled to the TX / RX 1440. The memory 1420 stores at least a portion of the program 1430. The TX / RX 1440 is used for bidirectional communication. The TX / RX 1440 has at least one antenna to facilitate communication, but in practice, the access node mentioned in this application may have multiple antennas. The communication interface can represent any interface required for communication with other network elements, such as an X2 interface for bidirectional communication between gNBs or eNBs, an S1 interface for communication between a mobility management entity (MME) / serving gateway (S-GW) and a gNB or eNB, an Un interface for communication between a gNB or eNB and a relay node (RN), or a Uu interface for communication between a gNB or eNB and a terminal device.

[0275] Assuming program 1430 includes program instructions that, when executed by the associated processor 1410, enable device 1400 to operate according to embodiments of this disclosure, as incorporated herein by reference. Figures 1A to 13 The embodiments discussed herein can be implemented by computer software executable by the processor 1410 of device 1400, or by hardware, or by a combination of software and hardware. The processor 1410 can be used to implement various embodiments of this disclosure. Furthermore, a combination of the processor 1410 and the memory 1420 can form a processing unit 1450 suitable for implementing various embodiments of this disclosure.

[0276] Memory 1420 can be of any type suitable for a local technology network and can be implemented using any suitable data storage technology, such as, as non-limiting examples, non-transitory computer-readable storage media, semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. While one memory 1420 is shown in device 1400, several physically different memory modules may exist in device 1400. Processor 1410 can be of any type suitable for a local technology network and may include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture, as non-limiting examples. Device 1400 may have multiple processors, such as application-specific integrated circuit chips that are time-dependent on a clock synchronized with the main processor.

[0277] In some embodiments, the first device includes a circuit system configured to perform method 700. In some embodiments, the second device includes a circuit system configured to perform method 800. In some embodiments, the third device includes a circuit system configured to perform method 900 or 1000. In some embodiments, the fourth device includes a circuit system configured to perform method 1100. In some embodiments, the fifth device includes a circuit system configured to perform method 1200.

[0278] In some embodiments, the exemplary system of this disclosure includes the first device, second device, third device, fourth device, and fifth device as described above. For example, in the exemplary system, the third device is configured to send a fourth message indicating whether additional training data should be used by the second device to perform a model update of one or more models associated with the first or more devices, wherein the second device is one of the second or more devices, and the additional training data differs from the training data used at the first or more devices to perform the model update; the second device is configured to receive the fourth message from the third device and update one or more models based on receiving the fourth message; the first device is configured to receive an instruction from the second or third device instructing the first device to perform a model update, and send a first notification notifying that the model update of one or more models has been completed, wherein the first device is one of the first or more devices. The apparatus; the third apparatus is further configured to control the second or more apparatuses and send a fifth message to the fourth apparatus for requesting influence on traffic routing, wherein influencing traffic routing is used to select at least one candidate apparatus among the second apparatus and the first or more apparatuses, and wherein the second apparatus and at least one candidate apparatus jointly perform a model update; the fourth apparatus is configured to receive the fifth message from the third apparatus to generate at least one policy regarding influence on traffic routing based on the fifth message, and send at least one policy to the fifth apparatus; the fifth apparatus is configured as a network controller and receives at least one policy from the fourth apparatus, the fifth apparatus is further configured to select the second apparatus and at least one candidate apparatus based on the at least one policy, and send a sixth message to the third apparatus for instructing the second apparatus and at least one candidate apparatus.

[0279] The components included in the apparatus and / or device of some embodiments of this disclosure can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In one embodiment, one or more units can be implemented using software and / or firmware (e.g., machine-executable instructions stored in a storage medium). In addition to or instead of machine-executable instructions, some or all of the units in the apparatus and / or device can be implemented at least partially by one or more hardware logic components. For example (but not limited to), illustrative types of hardware logic components that can be used include field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip systems (SOCs), complex programmable logic devices (CPLDs), etc.

[0280] Generally, the various embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software, which can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that the blocks, apparatuses, systems, technical end devices, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof, as non-limiting examples.

[0281] Some embodiments of this disclosure also provide at least one computer program product tangibly stored in a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions that execute in a device on a target real or virtual processor, such as those included in a program module, to perform the above-referenced... Figures 3 to 12 Any of the processes or methods described herein. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of a program module can be combined or split among program modules as needed. The machine-executable instructions of a program module can execute on a local or distributed device. In a distributed device, the program module can reside on both local and remote storage media.

[0282] Program code for performing methods of some embodiments of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code implements the functions / operations specified in the flowcharts and / or block diagrams. The program code may be executed entirely on a machine, partially on a machine (as a standalone software package), partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0283] The aforementioned program code can be embodied on a machine-readable medium, which can be any tangible medium that may include or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Further specific examples of machine-readable storage media will include electrical connections having one or more wires, portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0284] Furthermore, although operations are shown in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring the execution of all shown operations to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, while several specific embodiment details are included in the discussion above, these should not be construed as limiting the scope of this disclosure, but rather as descriptions of features specific to particular embodiments. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually in multiple embodiments or in any suitable sub-combination.

[0285] Although some embodiments of this disclosure have been described in specific language of structural features and / or methodological actions, it should be understood that the embodiments of this disclosure as defined in the appended claims are not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as exemplary ways of implementing the claims.

[0286] When these functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, its essential nature, or its contribution to the prior art, or a portion thereof, can be embodied in the form of a software product stored in a storage medium. This software product includes several instructions to instruct a computer device (which may be a personal computer, server, or network device) to execute all or part of the steps (or operations) of the methods described in the various embodiments of this application. The aforementioned storage medium includes any medium capable of storing program code, such as a USB flash drive, a removable hard disk, read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0287] The above description describes some specific implementations of this application and is not intended to limit the scope of protection of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method comprising: Receive an instruction at a first device, the instruction being for instructing the first device to perform a model update of one or more models associated with a first or more devices, wherein the first device is one of the first or more devices; and A first notification is sent at the first device, the first notification being used to notify that the model update of the one or more models has been completed.

2. The method of claim 1, wherein the instruction comprises at least one of the following: A list of model identification information, which is used to identify the one or more models; Version information of the one or more models, wherein each of the one or more models has corresponding version information; or One or more status indicators, wherein each of the one or more status indicators indicates the status of a model in the one or more models, and the status of the model includes frozen or unfrozen.

3. The method according to claim 1 or 2, wherein the indication is carried on a first message, the first message being from a second device, and the first message further comprising values ​​of model parameters of the one or more models.

4. The method according to claim 1 or 2, wherein the indication is carried on a first message, the first message being from a second device, and the method further comprises: The values ​​of model parameters of the one or more models are obtained at the first device, wherein the values ​​of the model parameters are pre-configured at the first device.

5. The method according to claim 3 or 4, wherein: The second device is part of the data plane of the core network, or The second device is connected to the data plane of the core network.

6. The method according to any one of claims 3 to 5, wherein: The instruction is sent from the second device to the first device via a data plane function.

7. The method according to claim 1 or 2, wherein the instruction originates from a third device, and the method further comprises: The first device receives the values ​​of the model parameters of the one or more models from the second device.

8. The method of claim 7, wherein the third device is part of the control plane of the core network.

9. The method according to claim 7 or 8, wherein at least one of the following: The instruction is sent from the third device to the first device; or The instruction is sent from the third device to the first device via a control plane function.

10. The method according to any one of claims 1 to 9, further comprising: Update at least one of the one or more models at the first device.

11. The method according to any one of claims 1 to 10, wherein the first notification further comprises at least one of the following: A list of model identification information, wherein the model identification information is used to identify the one or more models; or One or more status indicators, wherein, Each of the one or more status indicators indicates the status of a model in the one or more models, the status of which includes frozen or unfrozen.

12. The method according to claim 2 or 10, wherein: The frozen model is a convergent model; or The unfrozen model is a non-convergent model.

13. The method according to any one of claims 1 to 12, wherein at least one of the following: The first notification is sent from the first device to the third device; or The first notification is sent from the first device to the third device via a control plane function.

14. The method according to any one of claims 1 to 13, further comprising: The first device receives a second message notifying it that it has been selected to perform the model update.

15. The method of claim 14, further comprising: Send an affirmative ACK at the first device, indicating confirmation of the selection.

16. The method of claim 15, wherein: The second message originates from the fourth device, and the affirmative ACK is sent to the fourth device, wherein the fourth device is used to perform policy control functions; or The second message comes from the fifth device, and the affirmative ACK is sent to the fifth device, wherein the fifth device is a network controller.

17. The method according to any one of claims 1 to 16, further comprising: The first device receives a third message from the third device, the third message being used to notify the first device to send at least one latest version of the one or more models obtained through the model update; as well as Send the at least one latest version at the first device.

18. The method of claim 17, wherein at least one of the following: The third message is received from the third device via a control plane function; or The at least one latest version is sent via a data plane function, wherein the data plane function is configured to transmit the at least one latest version to a sixth device among a second or more devices.

19. The method of claim 18, wherein the sixth device is the second device.

20. The method according to any one of claims 9, 13 and 18, wherein the control plane function includes at least one of the following: Network Open Function (NEF), Policy Control Function (PCF), Network Storage Function (NSF), Network Controller (NWC), Path Management Function (PMF), or Access and Mobility Management Function (AMF).

21. The method according to any one of claims 17 to 19, wherein the third message further comprises: Model identification information used to identify at least one model associated with the at least one latest version.

22. The method according to any one of claims 1 to 21, wherein the first device is one of a terminal device, a server, or a network function.

23. A method comprising: A fourth message is received from a third device at a second device, the fourth message indicating whether additional training data should be used by the second device to perform model updates of one or more models associated with a first or more devices, wherein the second device is one of a second or more devices, and the additional training data is different from the training data used at the first or more devices to perform model updates. as well as At the second device, the one or more models are updated based on receiving the fourth message.

24. The method of claim 23, wherein the fourth message comprises: A list of model identification information, wherein the model identification information is used to identify the one or more models; Version information of the one or more models, wherein each of the one or more models has corresponding version information; One or more status indicators, wherein each of the one or more status indicators indicates the status of a model in the one or more models, and the status of the model includes frozen or unfrozen.

25. The method of claim 24, wherein the fourth message further indicates, where the additional training data is obtained and / or how the additional training data is selected, if the additional training data should be used.

26. The method according to any one of claims 23 to 25, wherein updating the one or more models comprises: When the additional training data should be used by the second device to perform the model update, the additional training data is used to update at least one of the one or more models that has an indication of an unfrozen state.

27. The method according to any one of claims 23 to 26, further comprising: The second device sends an instruction to the first device among the first one or more devices, the instruction being used to instruct the first device to perform the model update of the one or more models.

28. The method of claim 27, wherein the indication is carried on a first message, and the first message further includes values ​​of model parameters of the one or more models.

29. The method of claim 27 or 28, wherein the instruction comprises: A list of model identification information, wherein the model identification information is used to identify the one or more models; Version information of the one or more models, wherein each of the one or more models has corresponding version information; and One or more status indicators, wherein each of the one or more status indicators indicates the status of a model in the one or more models, and the status of the model includes frozen or unfrozen.

30. The method according to claim 24 or 29, wherein: The frozen model is a convergent model; or The unfrozen model is a non-convergent model.

31. The method of claim 27, further comprising: At the second device, a second notification is received from the third device that the second device is selected to receive at least one latest version of the one or more models from the first device, wherein the at least one latest version is obtained through a model update performed by the first device.

32. The method of claim 31, wherein the second notification further comprises: Model identification information used to identify at least one model associated with the at least one latest version.

33. The method according to claim 31 or 32, further comprising: The second device receives configuration from the third device for obtaining one or more latest versions from another device among the second or more devices, wherein the one or more latest versions are versions of the one or more models that do not include the at least one latest version.

34. The method according to claim 27 or 31, wherein: The first device is one of a terminal device, a server, or a network function.

35. The method according to any one of claims 23 to 34, wherein: The second device is part of the data plane of the core network; or The second device is connected to the data plane of the core network.

36. The method according to any one of claims 23 to 35, wherein: The third device is part of the control plane of the core network.

37. A method comprising: A fourth message is sent from a third device to a second device, the fourth message indicating whether additional training data should be used by the second device to perform model updates for one or more models associated with a first or more devices, wherein the second device is one of a second or more devices, and the additional training data is different from the training data used to perform model updates at the first or more devices.

38. The method of claim 37, further comprising: At the third device, one of the first one or more devices is identified as the first device for performing the model update.

39. The method according to claim 37 or 38, further comprising: At the third device, the second device is determined by selecting an application location associated with the second device at at least one application location.

40. The method according to any one of claims 37 to 39, further comprising: The third device sends a configuration to the second device, the configuration being used by the second device to obtain version information of the one or more models from another device among the second or more devices.

41. The method according to any one of claims 37 to 40, further comprising: The third device sends an instruction to the first device among the first one or more devices, the instruction being used to instruct the first device to perform the model update.

42. The method of claim 41, wherein the instruction comprises at least one of the following: A list of model identification information, wherein the model identification information is used to identify the one or more models; Version information of the one or more models, wherein each of the one or more models has corresponding version information; or One or more status indicators, wherein each of the one or more status indicators indicates the status of a model in the one or more models, and the status of the model includes frozen or unfrozen.

43. The method according to claim 41 or 42, wherein: The instruction is sent from the third device to the first device, or The instruction is sent from the third device to the first device via a control plane function.

44. The method according to any one of claims 37 to 43, wherein the fourth message comprises at least one of the following: A list of model identification information, wherein the model identification information is used to identify the one or more models; Version information of the one or more models, wherein each of the one or more models has corresponding version information; or One or more status indicators, wherein each of the one or more status indicators indicates the status of a model in the one or more models, and the status of the model includes frozen or unfrozen.

45. The method according to claim 42 or 44, wherein: The frozen model is a convergent model; or The unfrozen model is a non-convergent model.

46. ​​The method of claim 44, wherein the fourth message further indicates, where the additional training data is obtained and / or how the additional training data is selected, if the additional training data should be used.

47. The method according to any one of claims 37 to 46, further comprising: At the third device, a first notification is received from the first device among the first one or more devices, the first notification being used to notify that the model update of the one or more models performed by the first device has been completed; as well as Based on the first notification, a sixth device is determined at the third device from the second or more devices for receiving at least one latest version of the one or more models, wherein the at least one latest version is obtained by a model update performed by the first device and the latest version corresponds to the updated model.

48. The method of claim 47, further comprising: A second notification is sent from the third device to the sixth device, indicating that the sixth device has been selected to receive the at least one latest version from the first device.

49. The method of claim 48, wherein the second notification further comprises: Model identification information used to identify at least one model associated with the at least one latest version.

50. The method according to claim 48 or 49, further comprising: The third device sends a configuration to the sixth device for the sixth device to obtain one or more latest versions from another device among the second or one or more devices, wherein the one or more latest versions are versions of the one or more models that do not include the at least one latest version.

51. The method of claim 44, further comprising: A third message is sent from the third device to the first device among the first one or more devices, the third message being used to notify the first device to send at least one latest version of the one or more models obtained through the model update to the sixth device among the second one or more devices.

52. The method according to any one of claims 47 to 51, wherein the sixth device is the second device.

53. The method according to any one of claims 38, 41 to 43, 47 to 48 and 51, wherein: The first device is one of a terminal device, a server, or a network function.

54. The method according to any one of claims 37 to 52, wherein: The second device is part of the data plane of the core network; or The second device is connected to the data plane of the core network.

55. The method according to any one of claims 37 to 53, wherein: The third device is part of the control plane of the core network.

56. A method comprising: A fifth message is sent from the third device to the fourth device to request influence on traffic routing, wherein influencing the traffic routing is used to select a second device from the second or more devices and to select at least one candidate device from the first or more devices, the second device and the at least one candidate device jointly performing model updates of one or more models associated with the first or more devices, and the fourth device is configured to perform policy control functions; and At the third device, a sixth message is received from the fifth device to indicate the second device and the at least one candidate device, wherein the fifth device is configured as a network controller.

57. The method of claim 55, further comprising: At the third device, the second device is determined by selecting an application location associated with the second device at at least one application location.

58. The method of claim 56, wherein the fifth message comprises at least one of the following: Data traffic information, which identifies data traffic of at least one application associated with the second or more devices, wherein the data traffic will be routed during the traffic routing; Information about the first one or more devices; Information on at least one location of the at least one application; Information related to one or more requests associated with the traffic route; or Traffic filtering information associated with the data traffic.

59. The method of claim 56, further comprising: Feedback for the sixth message is sent from the third device to the fifth device, wherein the feedback indicates an affirmative ACK or a negative ACK for the selection of the application location associated with the second device.

60. The method of claim 58, further comprising: At the third device, the candidate device is identified as the first device among the at least one candidate device, wherein information identifying the first device is included in the affirmative ACK.

61. The method according to claim 59, wherein: The first device is one of a terminal device, a server, or a network function.

62. The method according to any one of claims 55 to 60, wherein: The second device is part of the data plane of the core network; or The second device is connected to the data plane of the core network.

63. The method according to any one of claims 55 to 61, wherein: The third device is part of the control plane of the core network.

64. A method comprising: At the fourth device, a fifth message is received from the third device to request influence on traffic routing, wherein influencing the traffic routing is used to select a second device in the second or more devices and to select at least one candidate device in the first or more devices, the second device and the at least one candidate device jointly perform model updates of one or more models associated with the first or more devices, and the fourth device is configured to perform policy control functions. as well as At the fourth device, at least one policy affecting the traffic routing is generated based on the fifth message.

65. The method of claim 63, further comprising at least one of the following: The at least one policy is sent from the fourth device to the fifth device, the fifth device being configured as a network controller to determine the second device and the at least one candidate device; or At the fourth device, the second device and the at least one candidate device are determined based on the at least one strategy.

66. The method of claim 64, wherein the at least one strategy comprises at least one of the following: Information about the data traffic that is routed in the traffic routing; Information on at least one location of the at least one application, wherein one of the at least one applications is associated with the second device; Information related to one or more required requests associated with the traffic route; or Information about the first one or more devices.

67. The method of claim 63, further comprising: A sixth message, instructing the second device and the at least one candidate device, is sent from the fourth device to the third device.

68. The method of claim 66, wherein the sixth message indicates a data plane management event, the data plane management event being associated with the selection of an application location associated with the second device in at least one application location.

69. The method of claim 67, further comprising: Feedback for the sixth message is received from the third device at the fourth device, wherein the feedback indicates an affirmative ACK for confirmation of the selection or a negative ACK for rejection of the selection.

70. The method of claim 68, wherein the feedback is the affirmative ACK, and the affirmative ACK further comprises: Information identifying the candidate device as the first device among the at least one candidate device.

71. The method of claim 69, further comprising: The fourth device sends a second message to the first device to notify that the first device has been selected to perform the model update.

72. The method of claim 70, further comprising: The fourth device receives from the first device either an affirmative ACK for confirming the execution of the model update or a negative ACK for rejecting the execution of the model update.

73. The method according to any one of claims 66 to 70, further comprising: The fourth device sends traffic routing information indicating the traffic route to the seventh device, which is used to select a data plane path for the traffic route.

74. The method of claim 70, further comprising at least one of the following: At the fourth device, a third notification is received from the seventh device that the data plane path has been configured for traffic routing; or The fourth device receives a response from the seventh device regarding the transmission of the traffic routing information, wherein the response includes the third notification that the data plane path has been configured for the traffic routing.

75. The method according to claim 72 or 73, wherein the traffic routing information comprises at least one of the following: Information identifying the application location associated with the second device in at least one application location; Information identifying a first device, wherein the first device is determined among the at least one candidate device; Information associated with the at least one strategy; or Information about the data plane path for the traffic routing.

76. The method according to any one of claims 69 to 71 and 74, wherein: The first device is one of a terminal device, a server, or a network function.

77. The method according to any one of claims 63 to 75, wherein: One of the second or more devices is a portion of the data plane of the core network; or The devices in the second or more devices are connected to the data plane of the core network.

78. The method according to any one of claims 63 to 76, wherein: The third device is part of the control plane of the core network.

79. A method comprising: At the fifth device, at the fourth device, at least one policy regarding influencing traffic routing is received, wherein influencing traffic routing is used to select a second device in a second or more devices and to select at least one candidate device in the first or more devices, wherein the second device and the at least one candidate device jointly perform model updates of one or more models associated with the first or more devices, the fourth device is configured to perform policy control functions, and the fifth device is configured as a network controller. At the fifth device, the second device and the at least one candidate device are selected based on the at least one strategy; as well as A sixth message is sent from the fifth device to the third device to instruct the second device and the at least one candidate device, wherein the third device is configured to control the second or more devices.

80. The method of claim 78, wherein the at least one strategy comprises at least one of the following: Information about the data traffic that is routed in the traffic routing; Information on at least one location of the at least one application, wherein one of the at least one applications is associated with the second device; Information related to one or more required requests associated with the traffic route; or Information about the first one or more devices.

81. The method of claim 78 or 79, wherein the sixth message indicates a data plane management event, the data plane management event being associated with the selection of an application location associated with the second device in at least one application location.

82. The method of claim 80, further comprising: Feedback for the sixth message is received from the third device at the fifth device, wherein the feedback indicates an affirmative ACK for confirmation of the selection or a negative ACK for rejection of the selection.

83. The method of claim 79, wherein the feedback is the affirmative ACK, and the affirmative ACK further comprises: Information identifying the candidate device as the first device among the at least one candidate device.

84. The method according to claim 82, further comprising: A second message is sent from the fifth device to the first device to notify the first device that it has been selected to perform the model update.

85. The method of claim 83, further comprising: At the fifth device, a positive ACK is received from the first device to acknowledge the execution of the model update or a negative ACK to reject the execution of the model update.

86. The method according to any one of claims 78 to 84, further comprising: At the fifth device, traffic routing information instructing the traffic route is sent to the seventh device, which is used to select a data plane path for the traffic route.

87. The method of claim 84, wherein the traffic routing information comprises at least one of the following: Information identifying the application location associated with the second device in at least one application location; Information identifying a first device, wherein the first device is determined among the at least one candidate device; Information associated with the at least one strategy; Information about the data plane path for the traffic routing.

88. The method according to claim 85 or 86, further comprising at least one of the following: At the fifth device, a third notification is received from the seventh device that the data plane path has been configured for the traffic routing; or The fifth device receives a response from the seventh device regarding the transmission of the traffic routing information, wherein the response includes the third notification that the data plane path has been configured for the traffic routing.

89. The method according to any one of claims 78 to 87, wherein the fifth device and the fourth device are integrated in a single network entity.

90. The method according to any one of claims 82 to 84 and 86, wherein: The first device is one of a terminal device, a server, or a network function.

91. The method according to any one of claims 78 to 89, wherein: One of the second or more devices is a portion of the data plane of the core network; or The devices in the second or more devices are connected to the data plane of the core network.

92. The method according to any one of claims 78 to 90, wherein: The third device is part of the control plane of the core network.

93. A first device, comprising: transceiver; as well as A processor communicatively coupled to the transceiver, wherein the processor is configured to perform the method according to any one of claims 1 to 22.

94. A second device, comprising: transceiver; as well as A processor communicatively coupled to the transceiver, wherein the processor is configured to perform the method according to any one of claims 23 to 36.

95. A third device, comprising: transceiver; as well as A processor communicatively coupled to the transceiver, wherein the processor is configured to perform the method according to any one of claims 37 to 54.

96. A third device, comprising: transceiver; as well as A processor communicatively coupled to the transceiver, wherein the processor is configured to perform the method according to any one of claims 55 to 62.

97. A fourth device, comprising: transceiver; as well as A processor communicatively coupled to the transceiver, wherein the processor is configured to perform the method according to any one of claims 63 to 77.

98. A fifth device, comprising: transceiver; as well as A processor communicatively coupled to the transceiver, wherein the processor is configured to perform the method according to any one of claims 78 to 91.

99. A system comprising a first device, a second device, a third device, a fourth device, and a fifth device, wherein: The third device is configured to send a fourth message indicating whether additional training data should be used by the second device to perform model updates of one or more models associated with the first or more devices, wherein the second device is one of the second or more devices, and the additional training data is different from the training data used at the first or more devices to perform model updates. The second device is configured to receive the fourth message from the third device and update the one or more models based on receiving the fourth message; The first device is configured to receive from the second device or the third device an instruction to instruct the first device to perform the model update, and to send a first notification to notify that the model update of the one or more models has been completed, wherein the first device is one of the first one or more devices; The third device is further configured to control the second or more devices and send a fifth message to the fourth device for requesting to influence traffic routing, wherein influencing the traffic routing is used to select at least one candidate device among the second device and the first or more devices, and wherein the second device and the at least one candidate device jointly perform the model update; The fourth device is configured to receive the fifth message from the third device, generate at least one policy regarding the traffic routing based on the fifth message, and send the at least one policy to the fifth device; as well as The fifth device is configured as a network controller and receives the at least one policy from the fourth device. The fifth device is also configured to select the second device and the at least one candidate device based on the at least one policy, and send a sixth message to the third device to instruct the second device and the at least one candidate device.

100. A method comprising: A fourth message is sent at a third device, the fourth message indicating whether additional training data should be used by a second device to perform model updates of one or more models associated with a first or more devices, wherein the second device is one of a second or more devices, and the additional training data is different from the training data used at the first or more devices to perform model updates. The fourth message is received from the third device at the second device, and the one or more models are updated based on the receipt of the fourth message; At a first device, an instruction is received from the second device or the third device to instruct the first device to perform the model update, and a first notification is sent to notify that the model update of the one or more models has been completed, wherein the first device is one of the first one or more devices; The third device controls the second or more devices and sends a fifth message to the fourth device to request influence on traffic routing, wherein influencing the traffic routing is used to select at least one candidate device among the second device and the first or more devices, and wherein the second device and the at least one candidate device jointly perform the model update; The fifth message is received from the third device at the fourth device to generate at least one policy regarding the traffic routing based on the fifth message, and the at least one policy is sent to the fifth device. as well as The fifth device receives the at least one policy from the fourth device, the fifth device being configured as a network controller and for selecting the second device and the at least one candidate device based on the at least one policy, and sending a sixth message to the third device to instruct the second device and the at least one candidate device.

101. A system comprising at least one of the following: a first device according to claim 92, a second device according to claim 93, a third device according to claim 94 or 95, a fourth device according to claim 96, or a fifth device according to claim 97.

102. A non-transitory computer-readable medium comprising a computer program stored thereon, which, when executed on at least one processor, causes the at least one processor to perform the method according to any one of claims 1 to 91.

103. A chip comprising at least one processing circuit configured to perform the method according to any one of claims 1 to 91.

104. A computer program product tangibly stored on a computer-readable medium and comprising computer-executable instructions that, when executed, cause a device to perform the method according to any one of claims 1 to 91.

Images