Interrupt avoidance during model training using joint learning

Abstract

An apparatus configured to train a model in a communication network using joint learning, the apparatus comprising means for: selecting at least two further apparatuses to train a local model; further selecting an alternative device for at least one of the selected at least two further devices; and configuring each of the at least two further devices for training the local model, and configuring an alternative device of at least one of the selected two further devices for training the local model; receiving a local training result from at least one of the at least two further devices, and receiving a local training result from an alternative device of the at least one of the selected two further devices; and combining the local training results to generate an aggregated training result of the model.

Description

Technical Field

[0001] The present application relates to a method, apparatus, system, and computer program for performing model training using federated learning, and in particular, but not limited to, to a method, apparatus, system, and computer program for avoiding interruptions during model training using federated learning. Background Art

[0002] A communication system can be considered as a facility that enables communication between two or more entities (such as terminals and / or other nodes) or provides connection services for entities. A communication system may include a communication network and one or more compatible terminals (also referred to as communication devices). Communications may carry, for example, voice, video, electronic mail (email), text messages, multimedia data, and / or content data. Non-limiting examples of connection services provided by a communication system may include enhanced mobile broadband, ultra-reliable low-latency communications, mission-critical communications, large-scale Internet of Things (IoT), and multimedia services.

[0003] In a communication system, at least part of the communication between at least two entities takes place over a wireless link. Examples of networks in a communication system are public land mobile networks (PLMNs), radio access networks such as terrestrial radio access networks or non-terrestrial radio access networks (e.g. satellite networks), and different wireless local area networks such as wireless local area networks (WLANs). Radio access networks may include cells and are therefore often referred to as cellular networks.

[0004] The terminal may be referred to as a user equipment (UE) or user device. The terminal is equipped with appropriate signal receiving and sending means to enable communication, such as enabling access to a communication network or communicating directly with other terminals. The terminal may access a carrier provided by a base station (e.g., a base station of a radio access network) and send and / or receive communications on the carrier.

[0005] A communication system and associated compatible terminals typically operate according to a given standard or specification that specifies what the various network entities of the communication system are allowed to do and how it should be implemented. Communication protocols and / or parameters for communication are also typically defined. An example of a communication system is a Universal Mobile Telecommunications System (UMTS) system (e.g., a communication system using 3G radio access technology). Other examples of communication systems are so-called 4G systems (e.g., a communication system operating using 4G radio access technology) and 5G or New Radio (NR) systems (e.g., a communication system operating using 5G or NR radio access technology). The radio access technology used by the communication system is standardized by the Third Generation Partnership Project (3GPP). Summary of the invention

[0006] According to one aspect, a device configured to train a model in a communication network using federated learning is provided, the device comprising components for: selecting at least two additional devices to train a local model; selecting an alternative device for at least one of the selected at least two additional devices; and configuring each of the at least two additional devices for training the local model, and configuring the alternative device of at least one of the selected two additional devices for training the local model; receiving local training results from at least one of the at least two additional devices, and receiving local training results from the alternative device of at least one of the selected two additional devices; and combining the local training results to generate an aggregated training result for the model.

[0007] The component for further selecting an alternative device for at least one of the selected at least two further devices can be used to select the alternative device based on information indicating at least one of the following items: the similarity of the data distribution of data of the local data set of at least one further device and the data distribution of data of the local data set of the alternative device; the location of the further device; the location of the alternative device; the proximity between the further device and the alternative device; the mobility pattern of the alternative device relative to the further device; the quality of communication on the sidelink between the further device and the alternative device; at least one characteristic of the wireless link between the further device and the base station of the radio access network.

[0008] The component may be configured to receive information indicating one or more candidate alternative devices from at least one further device, wherein the component for selecting an alternative device for at least one further device of the selected at least two further devices may be configured to select the alternative device from the one or more candidate alternative devices identified by the further device.

[0009] The component may also be configured to generate a FL report configuration and send the FL report configuration to each of at least two other devices, wherein the FL report configuration includes an indicator that is enabled to enable at least two other devices to generate a FL report, the FL report including: information identifying one or more potential alternative devices.

[0010] A component for configuring each of at least two additional devices to train a local model at the at least two additional devices and configuring each alternative device to train a local model at the alternative device can be used to generate an alternative training UE configuration for at least one of the at least two additional devices and the alternative device, the alternative training UE configuration comprising at least one of the following items: an additional device identifier configured to uniquely identify at least one of the at least two additional devices; an alternative device identifier configured to uniquely identify the additional alternative device; a condition identifier configured to identify a condition in which at least one of the at least two additional devices cannot train a local model and causes the alternative device to perform a local training model.

[0011] The condition may include at least one of: a minimum quality of a Uu link between the further device and a base station of the radio access network; a minimum availability of computing resources at the further device; a minimum availability of power resources at the further device; and a minimum security / integrity level associated with a local data set of the further device.

[0012] The component is used to obtain local training results from at least two additional devices, and when at least one of the at least two additional devices is unable to train a local model, the device can also be used to receive an indicator from an alternative device, the indicator being configured to identify that the local training results of the trained local model will be used as a substitute for the local training results of the trained local model.

[0013] The component for configuring each of the at least two additional devices for training a local model and configuring an alternative device of at least one of the selected two additional devices for training the local model can also be used to generate a global model and a training configuration for the selected at least two additional devices and the alternative device, wherein the training of the local model is based on the global model and the training configuration.

[0014] The component may also be configured to: receive an indication from at least one of the at least two additional devices that at least one of the at least two additional devices is unable to train a local model; and generate a request to perform local model training on an alternative device.

[0015] The request may include at least one of: an indicator of a reason why at least one of the at least two additional apparatuses was unable to train the local model; and a time indicator indicating a time when the replacement apparatus will perform local model training.

[0016] The component for configuring each of the at least two additional devices for training a local model and configuring an alternative device for at least one of the selected two additional devices for training a local model can also be used to: receive acceptance or rejection of an alternative training UE configuration from at least one of the at least two additional devices; receive acceptance or rejection of an alternative training UE configuration from the alternative device; and reselect and reconfigure an additional alternative device for at least one of the at least two additional devices based on receiving at least one rejection of the alternative training configuration from at least one of the at least two additional devices or the alternative device.

[0017] The device can be one of the following items: a base station of a radio access network, wherein at least two further devices and the alternative devices are user equipment; a network data analysis entity, wherein at least two further devices and the alternative devices are distributed network data analysis entities; and an operation, administration and maintenance entity, wherein at least two further devices and the alternative devices are base stations.

[0018] According to a second aspect, a device configured to train a local model during federated learning is provided, the device comprising components for: receiving an alternative training UE configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device of the device, and a condition identifier configured to identify a condition under which the device cannot train the local model and causes the alternative device to train the local model at the alternative device; and training the local model and sending a local training result to the other device, or determining that the device cannot train the local model based on the condition under which the device cannot train the local model, and sending a local model training request to one of the other device or the alternative device, causing the alternative device to perform local training at the alternative device using a local data set.

[0019] The component may also be used to generate information indicating a candidate replacement device based on information indicating at least one of the following items: data distribution of data in a local data set at the device, and data distribution of data in a local data set at a candidate replacement device; spread / distribution of local data of the device and the candidate replacement device; range of data in local data sets at the device and the candidate replacement device; interquartile range of data in local data sets at the device and the candidate replacement device; standard deviation of data in local data sets at the device and the candidate replacement device; variance of data in local data sets at the device and the candidate replacement device; proximity between the device and the candidate replacement device; and mobility pattern between the device and the candidate replacement device.

[0020] The component may also be configured to receive a request from another device to generate information indicative of a candidate replacement device.

[0021] At least one condition may include at least one of: a minimum quality of a Uu link between the device and a base station of the radio access network; a minimum availability of computing resources at the device; a minimum availability of power resources at the device; and a minimum security / integrity level associated with a local data set of another device.

[0022] The local model training request may include: an indicator identifying a condition that prevents the device from training a local model; and a time indicator indicating a time when an alternative device will train a local model.

[0023] The component may also be used to generate an acceptance or rejection of an alternative training UE configuration to the further apparatus, wherein the further apparatus may be caused to reselect and reconfigure the further alternative apparatus for the apparatus.

[0024] The apparatus may be a user equipment, wherein the alternative apparatus may be a user equipment and the further apparatus may be a base station of a radio access network.

[0025] The apparatus may be a wireless communication device, wherein the alternative apparatus may be a wireless communication device and the further apparatus may be a base station of a radio access network.

[0026] The apparatus may be a distributed network data analysis entity, wherein the further apparatus may be a centralized network data analysis entity, and the alternative apparatus may be a distributed network data analysis entity.

[0027] The apparatus may be a base station of the radio access network, wherein the further apparatus may be an operation, administration and maintenance entity and the replacement apparatus may be a base station of the radio access network.

[0028] The apparatus may be an open radio access network function, wherein the further apparatus may be an open radio access network function, and the replacement apparatus may be an open radio access network function.

[0029] According to a third aspect, a device configured to train a local model for federated learning is provided, the device comprising components for: receiving an alternative training UE configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the other device as a device for training a local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition in which the other device cannot train a local model and causes the device to train a local model at the device; and when the other device cannot train the local model, receiving a local model training request from the other device or the other device to train the local model; training the local model using the local data set and sending the training result to the other device after receiving the request.

[0030] The local model training request may include at least one of: an indicator identifying a condition that prevents another device from training a local model; and a time indicator indicating a time when the device will train the local model.

[0031] The component may also be used to generate an acceptance or rejection message to the further device, wherein the further device may be caused to reselect and reconfigure a further alternative device for the other device.

[0032] The means for training a local model and, upon receiving the request, sending the trained local model to the further device may also be used to send an indicator to identify that an update to the parameters of the trained local model is to be used as an alternative update to the parameters of the trained local model.

[0033] The apparatus may be a user equipment, wherein the other apparatus may be a user equipment and the further apparatus may be a base station of a radio access network.

[0034] The apparatus may be a wireless communication device, wherein the other apparatus may be a wireless communication device, and the further apparatus may be a base station of a radio access network.

[0035] The apparatus may be a distributed network data analysis entity, wherein a further apparatus may be a centralized network data analysis entity, and another apparatus may be a distributed network data analysis entity.

[0036] The apparatus may be a base station of a radio access network, wherein the further apparatus may be an operation, administration and maintenance entity and the further apparatus may be a base station of the radio access network.

[0037] The apparatus may be an open radio access network entity, wherein the further apparatus may be an open radio access network entity, and the further apparatus may be an open radio access network entity.

[0038] According to a fourth aspect, a method for a device configured to train a model in a communication network using federated learning is provided, the method comprising: selecting at least two additional devices to train a local model; selecting an alternative device for at least one of the selected at least two additional devices; and configuring each of the at least two additional devices to train the local model, and configuring the alternative device of at least one of the selected two additional devices to train the local model; receiving local training results from at least one of the at least two additional devices, and receiving local training results from the alternative device of at least one of the selected two additional devices; and combining the local training results to generate an aggregated training result of the model.

[0039] Selecting an alternative device for at least one of the selected at least two additional devices may include selecting the alternative device based on information indicating at least one of the following items: similarity of data distribution of data of a local data set of at least one additional device and data distribution of data of a local data set of the alternative device; location of the additional device; location of the alternative device; proximity between the additional device and the alternative device; mobility pattern of the alternative device relative to the additional device; quality of communication on a sidelink between the additional device and the alternative device; at least one characteristic of a Uu link between the additional device and a base station of the radio access network.

[0040] The method may include receiving information indicating one or more potential replacement devices from at least one further device, wherein selecting a replacement device for at least one further device of the selected at least two further devices may include selecting the replacement device from the potential replacement devices identified by the further device.

[0041] The method may further include generating a FL report configuration and sending the FL report configuration to each of at least two other devices, wherein the FL report configuration includes an indicator that is enabled to enable at least two other devices to generate a FL report, the FL report including information identifying one or more potential alternative devices.

[0042] Configuring each of the at least two additional devices to train a local model at the at least two additional devices and configuring each alternative device to train a local model at the alternative device may include generating an alternative training UE configuration for at least one of the at least two additional devices and the alternative device, the alternative training UE configuration comprising at least one of the following items: an additional device identifier configured to uniquely identify at least one of the at least two additional devices; an alternative device identifier configured to uniquely identify the additional alternative device; and a condition identifier configured to identify a condition in which at least one of the at least two additional devices cannot train a local model and causes the alternative device to train a local model at the alternative device.

[0043] The condition may include at least one of: a minimum quality of a Uu link between the further device and a base station of the radio access network; a minimum availability of computing resources at the further device; a minimum availability of power resources at the further device; and a minimum security / integrity level associated with a local data set of the further device.

[0044] Obtaining local training results from at least two additional devices, and when at least one of the at least two additional devices is unable to train a local model, may include receiving an indicator from an alternative device that is configured to identify that the local training results of the trained local model are to be used as a substitute for the local training results of the trained local model.

[0045] Configuring each of the at least two additional devices for training a local model and configuring an alternative device of at least one of the selected two additional devices for training the local model may also include generating a global model and a training configuration for the selected at least two additional devices and the alternative device, wherein the training of the local model is based on the global model and the training configuration.

[0046] The method may also include: receiving an indication from at least one of the at least two additional devices that at least one of the at least two additional devices is unable to train the local model; and generating a request to the alternative device to train the local model at the alternative device.

[0047] The request may include at least one of: an indicator of a reason why at least one of the at least two additional devices was unable to train the local model; and a time indicator indicating a time when the alternative device will train the local model at the alternative device.

[0048] Configuring each of the at least two additional devices for training a local model and configuring an alternative device for at least one of the selected two additional devices for training a local model may also include: receiving an acceptance or rejection of an alternative training UE configuration from at least one of the at least two additional devices; receiving an acceptance or rejection of an alternative training UE configuration from the alternative device; and reselecting and reconfiguring an additional alternative device for at least one of the at least two additional devices based on receiving at least one rejection of an alternative training UE configuration from at least one of the at least two additional devices or the alternative device.

[0049] The device can be one of the following items: a base station of a radio access network, wherein at least two further devices and the alternative devices are user equipment; a network data analysis entity, wherein at least two further devices and the alternative devices are distributed network data analysis entities; and an operation, administration and maintenance entity, wherein at least two further devices and the alternative devices are base stations.

[0050] The device may be an open radio access network application function, wherein the at least two further devices and the alternative device are open radio access network applications.

[0051] According to a fifth aspect, a method for a device configured to train a local model during joint learning is provided, the method comprising: receiving an alternative training UE configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device of the device, and a condition identifier configured to identify a condition under which the device cannot train the local model and causes the alternative device to train the local model at the alternative device; and training the local model and sending a local training result to the other device, or determining that the device cannot train the local model based on the condition under which the device cannot train the local model, and sending a local model training request to one of the other device or the alternative device, causing the alternative device to perform local training at the alternative device using a local data set.

[0052] The method may also include generating information indicating a candidate replacement device based on information indicating at least one of the following items: data distribution of data in a local data set at the device and data distribution of data in a local data set at a candidate replacement device; spread / distribution of local data of the device and the candidate replacement device; range of data in local data sets at the device and the candidate replacement device; interquartile range of data in local data sets at the device and the candidate replacement device; standard deviation of data in local data sets at the device and the candidate replacement device; variance of data in local data sets at the device and the candidate replacement device; proximity between the device and the candidate replacement device; and mobility pattern between the device and the candidate replacement device.

[0053] The method may further include receiving a request from another device to generate information indicative of a candidate replacement device.

[0054] At least one condition may include at least one of: a minimum quality of a Uu link between the device and a base station of the radio access network; a minimum availability of computing resources at the device; a minimum availability of power resources at the device; and a minimum security / integrity level associated with a local data set of another device.

[0055] The local model training request may include: an indicator identifying a condition that prevents the device from training a local model; and a time indicator indicating a time when an alternative device will train a local model.

[0056] The method may further comprise generating an acceptance or rejection of the alternative training UE configuration to the further apparatus, wherein the further apparatus may be caused to reselect and reconfigure the further alternative apparatus for the apparatus.

[0057] The apparatus may be a user equipment, wherein the alternative apparatus may be a user equipment and the further apparatus may be a base station of a radio access network.

[0058] The apparatus may be a wireless communication device, wherein the alternative apparatus may be a wireless communication device and the further apparatus may be a base station of a radio access network.

[0059] The apparatus may be a distributed network data analysis entity, wherein the further apparatus may be a centralized network data analysis entity, and the alternative apparatus may be a distributed network data analysis entity.

[0060] The apparatus may be a base station of the radio access network, wherein the further apparatus may be an operation, administration and maintenance entity and the replacement apparatus may be a base station of the radio access network.

[0061] The apparatus may be an open radio access network function, wherein the further apparatus may be an open radio access network function, and the replacement apparatus may be an open radio access network function.

[0062] According to a sixth aspect, a method for a device configured to train a local model for federated learning is provided, the method comprising: receiving an alternative training UE configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the other device as a device for training a local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition in which the other device cannot train a local model and causes the device to train a local model at the device; and when the other device cannot train the local model, receiving a local model training request from the other device or the other device to train the local model; training the local model using the local data set and sending the training result to the other device after receiving the request.

[0063] The local model training request may include at least one of: an indicator identifying a condition that prevents another device from training a local model; and a time indicator indicating a time when the device will train the local model.

[0064] The method may further include generating an acceptance message or a rejection message to the further apparatus, wherein the further apparatus may be caused to reselect and reconfigure a further replacement apparatus for the other apparatus.

[0065] Training the local model and sending the trained local model to the further apparatus after receiving the request may include sending an indicator to identify that an update to the parameters of the trained local model is to be used as an alternative update to the parameters of the trained local model.

[0066] The apparatus may be a user equipment, wherein the other apparatus may be a user equipment and the further apparatus may be a base station of a radio access network.

[0067] The apparatus may be a wireless communication device, wherein the other apparatus may be a wireless communication device, and the further apparatus may be a base station of a radio access network.

[0068] The apparatus may be a distributed network data analysis entity, wherein a further apparatus may be a centralized network data analysis entity, and another apparatus may be a distributed network data analysis entity.

[0069] The apparatus may be a base station of a radio access network, wherein the further apparatus may be an operation, administration and maintenance entity and the further apparatus may be a base station of the radio access network.

[0070] The apparatus may be an open radio access network entity, wherein the further apparatus may be an open radio access network entity, and the further apparatus may be an open radio access network entity.

[0071] According to a seventh aspect, there is provided a device configured to train a model in a communication network using federated learning, the device comprising: at least one processor and at least one memory storing instructions, which instructions, when executed by the at least one processor, cause the device to at least: select at least two additional devices to train a local model; select an alternative device for at least one of the selected at least two additional devices; and configure each of the at least two additional devices for training the local model, and configure the alternative device of at least one of the selected two additional devices for training the local model; receive local training results from at least one of the at least two additional devices, and receive local training results from the alternative device of at least one of the selected two additional devices; and combine the local training results to generate an aggregated training result of the model.

[0072] The device that is caused to select an alternative device for at least one of the selected at least two further devices may also be caused to select the alternative device based on information indicating at least one of the following items: similarity of data distribution of data of a local data set of at least one further device and data distribution of data of a local data set of the alternative device; location of the further device; location of the alternative device; proximity between the further device and the alternative device; mobility pattern of the alternative device relative to the further device; quality of communication on a sidelink between the further device and the alternative device; at least one characteristic of a Uu link between the further device and a base station of the radio access network.

[0073] The apparatus may also be caused to receive information indicating one or more potential alternative devices from at least one further apparatus, wherein the apparatus caused to select an alternative device for at least one further apparatus of the selected at least two further apparatuses may be caused to select the alternative device from the potential alternative devices identified by the further apparatuses.

[0074] The apparatus may also be caused to generate and send a FL report configuration to each of at least two other apparatuses, wherein the FL report configuration includes an indicator that is caused to enable the at least two other apparatuses to generate a FL report including information identifying one or more potential replacement apparatuses.

[0075] An apparatus that is caused to configure each of at least two additional apparatuses to train a local model at the at least two additional apparatuses and to configure each alternative apparatus to train a local model at the alternative apparatus may be caused to generate an alternative training configuration for at least one of the at least two additional apparatuses and the alternative apparatus, the alternative training configuration comprising at least one of the following items: an additional apparatus identifier configured to uniquely identify at least one of the at least two additional apparatuses; an alternative apparatus identifier configured to uniquely identify the additional alternative apparatus; a condition identifier configured to identify a condition under which at least one of the at least two additional apparatuses cannot train a local model and causes the alternative apparatus to train a local model at the alternative apparatus.

[0076] The condition may include at least one of: a minimum quality of a Uu link between the further device and a base station of the radio access network; a minimum availability of computing resources at the further device; a minimum availability of power resources at the further device; and a minimum security / integrity level associated with a local data set of the further device.

[0077] The device is configured to obtain local training results from at least two additional devices, and when at least one of the at least two additional devices is unable to train a local model, the device may also be configured to receive an indicator from an alternative device, the indicator being configured to identify that the local training results of the trained local model are to be used as a substitute for the local training results of the trained local model.

[0078] The device that is caused to configure each of the at least two additional devices for training a local model and to configure an alternative device of at least one of the selected two additional devices for training the local model can also be caused to generate a global model and training configuration for the selected at least two additional devices and the alternative device, wherein the training of the local model is based on the global model and the training configuration.

[0079] The apparatus may also be caused to: receive an indication from at least one of the at least two additional apparatuses that at least one of the at least two additional apparatuses is unable to train a local model; and generate a request to an alternative apparatus to train a local model at the alternative apparatus.

[0080] The request may include at least one of: an indicator of a reason why at least one of the at least two additional devices was unable to train the local model; and a time indicator indicating a time when the alternative device will train the local model at the alternative device.

[0081] The device that is caused to configure each of the at least two additional devices for training a local model and to configure an alternative device for at least one of the selected two additional devices for training the local model can also be caused to receive acceptance or rejection of the alternative training configuration from at least one of the at least two additional devices; receive acceptance or rejection of the alternative training configuration from the alternative device; and reselect and reconfigure an additional alternative device for at least one of the at least two additional devices based on receiving at least one rejection of the alternative training configuration from at least one of the at least two additional devices or the alternative device.

[0082] The device can be one of the following items: a base station of a radio access network, wherein at least two further devices and the alternative devices are user equipment; a network data analysis entity, wherein at least two further devices and the alternative devices are distributed network data analysis entities; and an operation, administration and maintenance entity, wherein at least two further devices and the alternative devices are base stations.

[0083] The device may be an open radio access network application function, wherein the at least two further devices and the alternative device are open radio access network applications.

[0084] According to an eighth aspect, a device configured to train a local model during federated learning is provided, the device comprising: at least one processor and at least one memory storing instructions, the instructions, when executed by the at least one processor, causing the device to at least: receive an alternative training configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device of the device, and a condition identifier configured to identify a condition in which the device cannot train the local model and causes the alternative device to train the local model at the alternative device; and train the local model and send a local training result to the other device, or determine that the device cannot train the local model based on the condition in which the device cannot train the local model, and send a local model training request to one of the other device or the alternative device, causing the alternative device to perform local training at the alternative device using a local data set.

[0085] The device may also be caused to generate information indicating a candidate replacement device based on information indicating at least one of the following items: data distribution of data in a local data set at the device and data distribution of data in a local data set at a candidate replacement device; spread / distribution of local data of the device and the candidate replacement device; range of data in local data sets at the device and the candidate replacement device; interquartile range of data in local data sets at the device and the candidate replacement device; standard deviation of data in local data sets at the device and the candidate replacement device; variance of data in local data sets at the device and the candidate replacement device; proximity between the device and the candidate replacement device; and mobility pattern between the device and the candidate replacement device.

[0086] The apparatus may also be caused to receive a request from another apparatus to generate information indicative of a candidate replacement apparatus.

[0087] At least one condition may include at least one of: a minimum quality of a Uu link between the device and a base station of the radio access network; a minimum availability of computing resources at the device; a minimum availability of power resources at the device; and a minimum security / integrity level associated with a local data set of another device.

[0088] The local model training request may include: an indicator identifying a condition that prevents the device from training a local model; and a time indicator indicating a time when an alternative device will train a local model.

[0089] The apparatus may also be caused to generate an acceptance or rejection of the alternate training configuration to the further apparatus, wherein the further apparatus may be caused to reselect and reconfigure the further alternate apparatus for the apparatus.

[0090] The apparatus may be a user equipment, wherein the alternative apparatus may be a user equipment and the further apparatus may be a base station of a radio access network.

[0091] The apparatus may be a wireless communication device, wherein the alternative apparatus may be a wireless communication device and the further apparatus may be a base station of a radio access network.

[0092] The apparatus may be a distributed network data analysis entity, wherein the further apparatus may be a centralized network data analysis entity, and the alternative apparatus may be a distributed network data analysis entity.

[0093] The apparatus may be a base station of the radio access network, wherein the further apparatus may be an operation, administration and maintenance entity and the replacement apparatus may be a base station of the radio access network.

[0094] The apparatus may be an open radio access network function, wherein the further apparatus may be an open radio access network function, and the replacement apparatus may be an open radio access network function.

[0095] According to a ninth aspect, a device configured to train a local model for federated learning is provided, the device comprising: at least one processor and at least one memory storing instructions, the instructions, when executed by the at least one processor, causing the device to at least: receive an alternative training configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify another device as a device for training a local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition in which the other device cannot train a local model and causes the device to train a local model at the device; and when the other device cannot train the local model, receive a local model training request from the other device or the other device to train the local model; train the local model using the local data set and send the training result to the other device after receiving the request.

[0096] The local model training request may include at least one of: an indicator identifying a condition that prevents another device from training a local model; and a time indicator indicating a time when the device will train the local model.

[0097] The apparatus may also be caused to generate an accept or reject alternate training configuration message to the further apparatus, wherein the further apparatus may be caused to reselect and reconfigure the further alternate apparatus for the other apparatus.

[0098] The device that is caused to train a local model and, upon receiving the request, send the trained local model to another device may also be caused to send an indicator to identify that an update to the parameters of the trained local model is to be used as a replacement update to the parameters of the trained local model.

[0099] The apparatus may be a user equipment, wherein the other apparatus may be a user equipment and the further apparatus may be a base station of a radio access network.

[0100] The apparatus may be a wireless communication device, wherein the other apparatus may be a wireless communication device, and the further apparatus may be a base station of a radio access network.

[0101] The apparatus may be a distributed network data analysis entity, wherein a further apparatus may be a centralized network data analysis entity, and another apparatus may be a distributed network data analysis entity.

[0102] The apparatus may be a base station of a radio access network, wherein the further apparatus may be an operation, administration and maintenance entity and the further apparatus may be a base station of the radio access network.

[0103] The apparatus may be an open radio access network entity, wherein the further apparatus may be an open radio access network entity, and the further apparatus may be an open radio access network entity.

[0104] According to a tenth aspect, there is provided an apparatus configured to train a model in a communication network using federated learning, the apparatus comprising: a component for selecting at least two additional devices to train a local model; a component for selecting an alternative device for at least one of the selected at least two additional devices; and a component for configuring each of the at least two additional devices for training the local model and configuring an alternative device of at least one of the selected two additional devices for training the local model; a component for receiving local training results from at least one of the at least two additional devices and receiving local training results from an alternative device of at least one of the selected two additional devices; and a component for combining the local training results to generate an aggregated training result of the model.

[0105] According to an eleventh aspect, a device configured to train a local model during joint learning is provided, the device comprising: a component for receiving an alternative training configuration from another device, the other device being configured to train the local model in a communication network including the device, the alternative configuration comprising a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device to the device, and a condition identifier configured to identify a trigger condition in which the device cannot train the local model and causes the alternative device to perform local model training; and a component for training the local model and sending the local training result to the other device, or determining that the device cannot train the local model based on the condition in which the device cannot train the local model, and sending the result to one of the other device or the alternative device so that the alternative device performs local model training.

[0106] According to the twelfth aspect, a device configured to train a local model for federated learning is provided, the device comprising: a component for receiving an alternative training configuration from another device, the other device being configured to train the local model in a communication network including the device, the alternative configuration comprising a device identifier configured to uniquely identify the other device as a device for training the local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition under which the other device cannot train the local model and causes the device to train the local model at the device; and a component for receiving a local model training request for training the local model from another device or the other device when the other device cannot train the local model; and a component for training the local model using the local data set and sending the local training result to the other device after receiving the request.

[0107] According to a thirteenth aspect, there is provided a computer program [or a computer-readable medium comprising program instructions], the instructions being used to cause a device configured to train a model in a communication network using federated learning to perform at least the following items: selecting at least two additional devices to train a local model; selecting an alternative device for at least one of the selected at least two additional devices; and configuring each of the at least two additional devices to train the local model, and configuring an alternative device for at least one of the selected two additional devices to train the local model; receiving local training results from at least one of the at least two additional devices, and receiving local training results from an alternative device for at least one of the selected two additional devices; and combining the local training results to generate an aggregated training result for the model.

[0108] According to a fourteenth aspect, there is provided a computer program [or a computer-readable medium comprising program instructions] comprising instructions for causing a device configured to train a local model during federated learning to perform at least the following items: receiving an alternative training configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device of the device, and a condition identifier configured to identify a condition under which the device cannot train the local model and causes the alternative device to train the local model at the alternative device; and training the local model and sending a local training result to the other device, or determining that the device cannot train the local model based on the condition under which the device cannot train the local model, and sending a local model training request to one of the other device or the alternative device causing the alternative device to perform local training at the alternative device using a local data set.

[0109] According to a fifteenth aspect, there is provided a computer program [or a computer-readable medium comprising program instructions] comprising instructions for causing a device configured to train a local model for federated learning to perform at least the following items: receiving an alternative training UE configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising a device identifier configured to uniquely identify the other device as a device for training a local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition under which the other device cannot train a local model and causes the device to train a local model at the device; and when the other device cannot train the local model, receiving a local model training request from the other device or the other device to train the local model; training the local model using the local data set and sending a local training result to the other device after receiving the request.

[0110] According to the sixteenth aspect, a non-transitory computer-readable medium comprising program instructions is provided, the program instructions being used to cause a device configured to train a model in a communication network using federated learning to perform at least the following items: selecting at least two additional devices to train a local model; selecting an alternative device for at least one of the selected at least two additional devices; and configuring each of the at least two additional devices to train the local model, and configuring an alternative device for at least one of the selected two additional devices to train the local model; receiving local training results from at least one of the at least two additional devices, and receiving local training results from an alternative device for at least one of the selected two additional devices; and combining the local training results to generate an aggregated training result for the model.

[0111] According to the seventeenth aspect, a non-transitory computer-readable medium comprising program instructions is provided, the program instructions being used to cause a device configured to train a local model during federated learning to perform at least the following items: receiving an alternative training configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device of the device, and a condition identifier configured to identify a condition under which the device cannot train the local model and causes the alternative device to train the local model at the alternative device; and training the local model and sending a local training result to the other device, or determining that the device cannot train the local model based on the condition under which the device cannot train the local model, and sending a local model training request to one of the other device or the alternative device, causing the alternative device to perform local training at the alternative device using a local data set.

[0112] According to an eighteenth aspect, a non-transitory computer-readable medium comprising program instructions is provided, the program instructions being used to cause a device configured to train a local model for federated learning to perform at least the following items: receiving an alternative training configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify another device as a device for training a local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition in which the other device cannot train a local model and causes the device to train a local model at the device; and when the other device cannot train the local model, receiving a local model training request from the other device or the other device to train the local model; training the local model using the local data set and sending the training result to the other device after receiving the request.

[0113] According to the nineteenth aspect, there is provided a device configured to train a model in a communication network using federated learning, the device comprising: a selection circuit system configured to select at least two additional devices to train a local model; a selection circuit system configured to select an alternative device for at least one of the selected at least two additional devices; and a configuration circuit system configured to configure each of the at least two additional devices for training the local model and to configure an alternative device of at least one of the selected two additional devices for training the local model; a receiving circuit system configured to receive local training results from at least one of the at least two additional devices and to receive local training results from an alternative device of at least one of the selected two additional devices; and a combining circuit system configured to combine the local training results to generate an aggregated training result of the model.

[0114] According to the twentieth aspect, there is provided a device configured to train a local model during federated learning, the device comprising: a receiving circuit system configured to receive an alternative training configuration from another device, the other device being configured to train the local model in a communication network including the device, the alternative configuration comprising a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device of the device, and a condition identifier configured to identify a condition under which the device cannot train the local model and causes the alternative device to train the local model at the alternative device; a training circuit system configured to train the local model; a control circuit system configured to control sending a local training result to another device; a determination circuit system configured to determine that the device cannot train the local model based on the condition under which the device cannot train the local model; and a control circuit system configured to control sending a local model training request to one of the other device or the alternative device so that the alternative device performs local model training at the alternative device using a local data set.

[0115] According to the twenty-first aspect, a device configured to train a local model for federated learning is provided, the device comprising: a receiving circuit system configured to receive an alternative training configuration from another device, the other device being configured to train the local model in a communication network including the device, the alternative configuration comprising a device identifier configured to uniquely identify the other device as a device for training the local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition under which the other device cannot train the local model and causes the device to train the local model at the device; and a receiving circuit system configured to receive a local model training request for training the local model from the other device or the other device when the other device cannot train the local model; a training circuit system configured to train the local model using a local data set; and a control circuit system configured to control the sending of a local training result to the other device after receiving the local model training request.

[0116] According to the twenty-second aspect, a computer-readable medium comprising program instructions is provided, the program instructions being used to cause a device configured to train a model in a communication network using federated learning to perform at least the following items: selecting at least two additional devices to train a local model; selecting an alternative device for at least one of the selected at least two additional devices; and configuring each of the at least two additional devices to train the local model, and configuring an alternative device for at least one of the selected two additional devices to train the local model; receiving local training results from at least one of the at least two additional devices, and receiving local training results from an alternative device for at least one of the selected two additional devices; and combining the local training results to generate an aggregated training result for the model.

[0117] According to a twenty-third aspect, a computer-readable medium comprising program instructions is provided, the program instructions being used to cause a device configured to train a local model during joint learning to perform at least the following items: receiving an alternative training configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising: a device identifier configured to uniquely identify the device for training the local model, an alternative device identifier configured to uniquely identify an alternative device of the device, and a condition identifier configured to identify a condition under which the device cannot train the local model and causes the alternative device to train the local model at the alternative device; and training the local model and sending a local training result to the other device, or determining that the device cannot train the local model based on the condition under which the device cannot train the local model, and sending a local training model request to one of the other device or the alternative device, causing the alternative device to perform local training at the alternative device using a local data set.

[0118] According to the twenty-fourth aspect, a computer-readable medium comprising program instructions is provided, the program instructions being used to cause a device configured to train a local model for federated learning to perform the following items: receiving an alternative training configuration from another device, the other device being configured to train a local model in a communication network including the device, the alternative configuration comprising a device identifier configured to uniquely identify the other device as a device for training a local model using a local data set, an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device, and a condition identifier configured to identify a condition under which the other device cannot train a local model and causes the device to train a local model at the device; and when the other device cannot train the local model, receiving a local model training request from the other device or the other device to train the local model; training the local model using the local data set and sending the local training result to the other device after receiving the request.

[0119] An apparatus comprises means for performing the actions of the above method.

[0120] An apparatus is configured to perform the actions of the above method.

[0121] A computer program includes program instructions for causing a computer to execute the above method.

[0122] A computer program product stored on a medium may cause an apparatus to perform the methods described herein.

[0123] According to one aspect, a non-transitory computer-readable medium including program instructions for causing an apparatus to perform a method according to at least any one of the preceding aspects is provided.

[0124] In the above, many different embodiments have been described. It should be understood that other embodiments can be provided by combining any two or more of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0125] Embodiments will now be described, by way of example only, with reference to the accompanying drawings, in which:

[0126] Figure 1 shows a representation of a network system according to some example embodiments;

[0127] Figure 2 shows a representation of a control device according to some example embodiments;

[0128] Figure 3 shows representations of apparatus according to some example embodiments;

[0129] Figure 4 A flowchart showing an example of training a model using federated learning (FL); and

[0130] Figures 5 to 7 A flow diagram illustrating an example of interruption avoidance during training of a model using federated learning (FL) according to some example embodiments. DETAILED DESCRIPTION

[0131] In the following, certain embodiments are explained with reference to devices capable of communicating with a communication system serving these devices. Figure 1 , Figure 2 ,as well as Figure 3 Certain general principles of a communication system (e.g., a 5G communication system) and a device (e.g., a terminal served by the communication system) including an access network (AN) and a core network are briefly explained to help understand the technology on which the examples are based.

[0132] Figure 1 A schematic diagram of a 5G wireless communication system (5GS) is shown. The 5GS may be composed of a radio access network (RAN) (e.g., a 5G radio access network or a next generation radio access network), a 5G core network (5GC), one or more application functions (AFs), and multiple data networks (DNs). In some embodiments, the AF is a client of the 5GC and is connected to a user plane function (UPF) of the 5GC via the DN, and is connected to a network function (NF) of the 5GC via a network exposure function (NEF) of the 5GC. In some embodiments, the AF is a trusted application function, and thus the trusted AF is implemented in the 5GC and is directly connected to other NFs of the 5GC. The AF may include the functionality of performing training using federated learning and selecting terminals that are connected to participating FLs of the 5GC, as described in further detail below. It will be understood that although Figure 1 Only one UPF is shown in the figure, but 5GS can be composed of a series of UFPs, including a UPF anchor connected to the DN. The connection between AF and NEF and UPF (or AF and NF of 5GC) is realized via the interface defined in the 3GPP standard.

[0133] 5GC may include, for example, the following network functions (NFs) (also referred to as network entities): network slice selection function (NSSF); network exposure function (NEF); network repository function (NRF); network data analysis function (NWDAF), policy control function (PCF); unified data management (UDM); authentication server function (AUSF); access and mobility management function (AMF); and session management function (SMF). The NFs of 5GC may have a service-based architecture as described in TR 23.501 of the 3GPP standard. The NF services that can be provided by the NFs of 5GC and the service-based interfaces of the NFs of 5GC are described in the 3GPP standard, in particular, in TR 23.501 and 23.502 of the 3GPP standard.

[0134] Access of a terminal to a 5GC may typically be via an access network, such as a 5G radio access network (5G-RAN). The 5G-RAN may include one or more base stations (e.g., gNodeBs (gNBs)). The gNBs of the 5G-RAN may include a gNB distributed unit connected to a gNB central unit and a remote radio head connected to the gNB distributed unit. In some embodiments, one or more base stations of the 5G-RAN may be an evolved NodeB eNodeB (eNB). In some embodiments, the 5G-RAN may be a 3GPP radio access network (e.g., a RAN operating using NR or LTE radio access technology defined in the 3GPP standard). Although Figure 15G-RAN is illustrated, but those skilled in the art will understand that access to 5GC can be made via any wireless or wired access network, such as a non-3GPP access network (e.g., an untrusted wireless local area network (WLAN) accessing 5GC via a non-3GPP interworking function (N3IWF), a trusted WLAN accessing 5GC via a trusted non-3GPP gateway function (TNGF), or a wired network accessing 5G via a wired access gateway function (W-AGF). A non-3GPP access network is an access network that is not configured to communicate directly with a core network having an architecture and operation defined in the 3GPP standard.

[0135] Figure 2 The diagram shows that it can be achieved Figure 1 An example of an apparatus 200 for one or more NFs of 5GC shown in FIG. The apparatus 200 may include: at least one random access memory (RAM) 211a, at least one read-only memory (ROM) 211b, at least one processor 212, 213, and a network interface 214. At least one processor 212, 213 may be coupled to the RAM 211a and the ROM 211b. At least one processor 212, 213 may be configured to execute software code 215. The software code 215 may, for example, include instructions for performing actions or operations of one or more NFs of 5GC. In some embodiments, the software code 215 may include instructions for performing one or more actions or operations of a federated learning (FL) aggregator according to aspects of the present disclosure. The software code 215 may be stored in the ROM 211b. The apparatus 200 may implement one or more NFs of 5GC and may be interconnected with another apparatus 200 that implements one or more other NFs in 5GC. In such an embodiment, the apparatus 200 may be part of a distributed computing system. In some embodiments, each NF of 5GC may be implemented on a single apparatus 200. In such an embodiment, apparatus 200 may be a cloud computing system.

[0136] Figure 3 Pictured Figure 1. The apparatus 300 may be any wireless communication device capable of sending and receiving radio signals. Non-limiting examples of the apparatus 300 include a terminal, a wireless communication device, a user equipment, a mobile station (MS), or a mobile device (such as a mobile phone or a so-called "smartphone"), a computer equipped with a wireless interface card or other wireless interface facilities (e.g., a USB dongle), a personal data assistant (PDA) or tablet computer equipped with wireless communication functions, a machine type communication (MTC) device, an Internet of Things (IoT) communication device, or any combination of these devices, etc. The apparatus 300 may be configured to communicate with a base station (e.g., NG-eNB or gNB) of an access network such as 5G-RAN and 5GC via a base station of a 5G-RAN using non-access layer (NAS) signaling, for example, for data communication. The communication may include or carry one or more of the following: voice, email, text message, multimedia, data, machine data, etc.

[0137] The apparatus 300 may receive wireless signals (e.g., radio or cellular signals) via an appropriate means 306 for receiving wireless signals over an air or radio interface 307 (commonly referred to as a Uu interface), and may transmit wireless signals (e.g., radio or cellular signals) via an appropriate means for transmitting wireless signals. Figure 3 In the embodiment, the apparatus 306 comprises one or more antennas (or an antenna array comprising a plurality of antennas) and a transceiver and is schematically designated by the block 306. The apparatus 306 may be provided, for example, by means of a radio component and an associated antenna arrangement comprising one or more antennas. The antenna arrangement may be arranged internally or externally to the mobile device.

[0138] The apparatus 300 may include: at least one processor 301, at least one memory ROM 302a, at least one RAM 302b, and other possible components 303 for software and hardware-assisted execution of the tasks it is designed to perform, including controlling access to and communication with access networks such as 5G-RAN and other apparatuses 300. The at least one processor 301 is coupled to RAM 311a and ROM 311b. The at least one processor 301 may be configured to execute appropriate software code 308. The software code 308 may include, for example, instructions that, when executed by at least one processor 301, perform one or more actions or operations of the present aspect. The software code 308 may be stored in ROM 311b.

[0139] At least one processor 301, storage means, and other related control means may be provided on an appropriate circuit board and / or in a chipset. This feature is indicated by reference numeral 304. The terminal 300 may optionally have a user interface, such as a keypad 305, a touch-sensitive display screen or touch-sensitive pad, a combination thereof, etc. Depending on the type of device, one or more of the following may be optionally provided: a display, a speaker, and a microphone.

[0140] Control or configuration of such communication systems is traditionally performed by control mechanisms that operate based on defined rules. In order to improve network performance (i.e., the performance of a network in a communication system (such as 5GS)), control mechanisms for implementing machine learning (ML) models have been proposed, wherein network and / or management data from many network entities of the communication system can be processed by the ML model to generate control outputs suitable for such a communication system. The training of the machine learning (ML) model is centralized. Network and / or management data are collected by network entities (usually referred to as distributed nodes) and provided to a single network entity (usually referred to as a central node) to train the ML model using the received data.

[0141] In order to minimize the amount of data exchanged between distributed nodes and a central node (model training is typically performed in the central node (hereinafter generally referred to as model training)), and to prevent data privacy loss at each node, a model can be trained using federated learning (FL). In FL, instead of training a model at a central node using centralized data, the central node provides a global model including parameters or data to the distributed nodes, and each distributed node performs local training on a local model using a dataset including the data of the distributed nodes during an FL iteration (hereinafter referred to as local model training). In other words, each distributed node has a dataset (hereinafter referred to as a local dataset) and trains a local model using its own local dataset (ie, performs local model training). In the following disclosure, the terms local training and local model training are interchangeable. For each iteration of FL, each distributed node then sends its local training results (e.g., the 'learned' parameters of its local model) to the central node.

[0142] During each iteration of FL, the central node receives local training results of local models from distributed nodes and combines or aggregates the local training results to obtain new global model parameters or data (global training results of the global model).

[0143] The local training result may include values ​​of parameters of the local model, and the global training result may include aggregated values ​​of parameters of the global model.

[0144] Then, in further iterations of FL, new global model parameters or data (e.g., aggregated received local training parameter values) can be sent to the distributed nodes, and the distributed nodes use these 'new' global model parameters as parameters for their local models, and further local training is performed on the local models to learn the new local model parameters. This process can be repeated until the values ​​of the parameters of the global model are optimized. The process of training local models at different distributed nodes is known to those skilled in the art and is therefore not described in further detail. For example, federated learning and local and global models are further described in "Federated Optimization: Distributed Optimization Beyond the Datacenter" by Konecny, Jakub, HBMcMahan, and Daniel Ramage (ArXiv abs / 1511.03575 (2015)).

[0145] Example of communication between a central node and distributed nodes for training a model (e.g., a global model) using FL Figure 4 is shown and summarized as follows, where the central node and the distributed nodes are communication systems (such as Figure 1 part of 5GS) or connected by a communication system.

[0146] Figure 4 The example shown in FIG. 4 shows a central node (referred to as FL aggregator 400) and three distributed nodes (i.e., Figure 4 The communication between the three devices 300 (represented as UE1 300a, UE2 300b, and UE3 300c) is used for two iteration sets of FL, iteration set N 451 and iteration set N+1 491. However, the FL can have more than two iteration sets. For each iteration set of FL, there can be an initial communication sequence. The communication between the FL aggregator 400 and the three distributed nodes includes signaling and / or data communication. The signaling can be sent by the FL aggregator 400 to the distributed nodes via the NEF and AMF and 5G-RAN, or can be sent by the FL aggregator via the AMF and 5G-RAN. Data communication can be sent by the FL aggregator 400 to the distributed nodes via the 5G-RAN, UPF, and DN, or sent by the wireless communication device. In Figure 4In the foregoing, for ease of explanation, 5G-RAN, AMF, NEF, UPF, and DN and the communication between these network entities are omitted. For example, the FL aggregator 400 may be configured to: generate a FL report configuration, and send the FL report configuration to each of UE1 300a, UE2 300b, and UE3 300c. The FL report configuration includes information defining the content included in the FL report, the format of the FL report, and how to generate the FL report. For example, as shown in the first set of iterations 451, the FL aggregator 400 may be configured to: at 401, generate a FL report configuration, and send the FL report configuration to UE1 300a, UE2 300b, and UE3 300c.

[0147] Then, each of UE1 300a, UE2 300b, and UE3 300c may be configured to generate a FL report based on the FL report configuration and send the FL report to the FL aggregator 400. For example, UE1 300a is shown generating a FL report and sending the FL report to the FL aggregator at 403, UE2 300b is shown generating a FL report and sending the FL report to the FL aggregator at 405, and UE3 300c is shown generating a FL report and sending the FL report to the FL aggregator at 407. The FL report may include information useful to the FL aggregator 400 to help the FL aggregator 400 determine which UEs are selected for FL (i.e., which UEs are selected to train the local model in the current iteration of FL, which may be abbreviated as local training in the following description). For example, such information may include information indicating the availability of resources at the UE, such as the availability of processing resources (e.g., the number of CPUs available at the UE), the availability of memory resources (the amount of memory available at the UE), the available energy at the UE, measurements indicating the quality of a radio link (e.g., a Uu link) between the UE and a base station of the 5G-RAN (which may be part of a link between the UE and the FL aggregator 400) (e.g., signal-to-noise ratio (SNR), reference signal received power (RSRP)), and the availability of data at the UE for training a local model.

[0148] The FL aggregator 400 is then configured to select K UEs (in Figure 4 In the example shown in , two UEs are used to train a local model (also referred to as local training in this article), such as Figure 4 409 in FIG. 409. The selection of the UE by the FL aggregator 400 may be random in some cases, or based on any suitable UE selection scheme taking into account the acquired FL report (provided by the UE).

[0149] In this example, the FL aggregator 400 (for iteration N 451) selects UE1 300a and UE3 300c for FL.

[0150] The FL aggregator 400 is configured to send the global model and training configuration, such as Figure 4 413 in . In other words, the FL aggregator 400 is configured to broadcast a global model (e.g., aggregated training results or aggregated model parameter values ​​from a previous iteration) and a training configuration to the K selected UEs. In addition, the FL aggregator 400 may send a training configuration to the K selected UEs so that the selected UEs can perform local training. The FL aggregator 400 is configured to then send a signal to each of the selected UEs (e.g., UE1 and UE3) to perform local training of its local model. For example, for iteration N451, FL with UE1 and UE3 is performed, as shown in the dashed box 411. In other words, the FL aggregator 400 sends a configuration to the selected UEs.

[0151] Then, each of the selected UEs is configured to perform local training. For example, UE1 is configured to: train a local model based on a local data set stored at UE1, such as Figure 4 As shown in 415, UE3 is configured to: train a local model based on a local data set stored at UE3. After receiving the global model and the training configuration, each of the K selected UEs configures its local model based on the global model and the training configuration, and updates the parameters of its local model to the received aggregated parameters. Then, the K selected UEs are configured to perform local training.

[0152] After performing local training, the K selected UEs are then configured to report the results of their local training (also referred to herein as local training results). For example, each UE is configured to report its local training results to the FL aggregator 400. For example, for UE1 300a, UE1 300a reports (i.e., sends) the local training results to the FL aggregator 400, such as Figure 4 As shown in 419 in FIG. 4, UE3 300c reports (ie, sends) its local training result to FL aggregator 400, as shown in FIG. Figure 4 As shown in 423 in . The local training result reported by the specific UE may include an update of the parameters of the local model at the specific UE. In some embodiments, the update of the parameters of the local model is the value of the parameters of the local model when the training of the local model is completed at the UE. In some embodiments, the update of the parameters of the local model is the gradient of the parameters of the local model when the training of the local model is completed at the UE.

[0153] Furthermore, the FL aggregator 400 is then configured to perform aggregation of the local training results received from each of the K UEs (eg, UE1 and UE3) (ie, aggregation or combination of updates of the parameters of the local models), such as Figure 4 As shown in 423.

[0154] After performing aggregation of the local training results, FL is repeated with UE1 and UE3 for a predetermined number of iterations. In other words, FL with UE1 and UE3 at 411 is repeated a plurality of times.

[0155] also, Figure 4 Another example iteration set N+1 491 is shown in which UEs are reselected for FL. In iteration set N+1 491, UE2 300b and UE3 300c are selected for FL by FL aggregator 400. For example, at 461, FL aggregator 400 is shown configured to generate FL report configurations and send the FL report configurations to UE1 300a, UE2 300b, and UE3 300c.

[0156] Then, each of the UEs may be configured to generate a FL report based on the FL report configuration and send the FL report to the FL aggregator 400. For example, at 463, UE1 300a is shown to generate a FL report and send the FL report to the FL aggregator, at 465, UE2 300b is shown to generate a FL report and send the FL report to the FL aggregator, and at 467, UE3 300c is shown to generate a FL report and send the FL report to the FL aggregator.

[0157] Then, the FL aggregator 400 is configured to select K UEs (2 UEs in this example) for local training, such as Figure 4 As shown in 469.

[0158] In this example, the FL aggregator 400 selects UE2 300 b and UE3 300 c for FL, and implements FL for iteration set N+1 491 with UE2 and UE3, as indicated by the dashed box 471 .

[0159] Then, the FL aggregator 400 is configured to perform broadcasting of the global FL model and training configuration, such as Figure 4 473 in FIG. The broadcasting of the global FL model and the training configuration includes sending the global model and the training configuration to each of UE2 300b and UE3 300c.

[0160] Then, each of the selected UEs is configured to perform local training. For example, UE2 300b is configured to train a local model based on UE2 data, such as Figure 4As shown in step 475 in, UE3 is configured to: train the local module based on UE3 data, such as Figure 4 As shown in 477. Perform local training of each of the selected UEs.

[0161] After performing local training, the K selected UEs then report their local training results to the FL aggregator 400. For example, for UE2 300b, the local training results are provided (eg, sent) to the FL aggregator 400, such as Figure 4 As shown in 479 in FIG. 4 , for UE3 300c, the local training result is provided (for example, sent) to the FL aggregator 400, such as Figure 4 As shown in 481. The local training result can be the value of the parameter of the local model after the training is performed.

[0162] Furthermore, the FL aggregator 400 is then configured to perform a combination or aggregation of the local training results received from the K UEs, such as Figure 4 As shown in 483 in , the aggregated training results of the global model are obtained.

[0163] After performing the FL aggregation operation to combine or aggregate the local training results to obtain model parameters of the local model to generate the aggregated training results of the global model, the 'local' training process of UE2 and UE3 is repeated (another) predetermined number of times. In other words, the training of the local UE2 and UE3 models as shown in block 461 is repeated multiple times.

[0164] The set of FL iterations may then be repeated or continued until the FL model converges (ie, until the parameters of the global model are optimized).

[0165] When FL is implemented in a communication system (such as a 5G system), during each FL iteration, the UE receives an instruction to perform training of a local model using a local data set from a FL aggregator 400 located within a network entity (also referred to as a network function) of a core network of the 5G system, for example. The UE then reports the local training results (e.g., the values ​​of the parameters of the local model after the local training is completed, the updates of the parameters of the local model after the local training is completed, or the gradients of the parameters of the local model after the local training is completed) to the FL aggregator 400. The FL aggregator 400 then combines or aggregates the local training results of the local models received from the UE to obtain the aggregated training results of the global model (e.g., obtain the values ​​of the parameters of the global model). Next, in subsequent iterations of FL, the aggregated training results (e.g., the values ​​of the parameters of the global model) are included in the global model sent to the UE. The UE can then start training for the next FL iteration.

[0166] An example use case of the federated learning embodiments described below may be training a machine learning model (i.e., a global model) configured to predict the quality of service (QoS) of a device-to-device (D2D) communication between two devices (such as two apparatuses 300) over a sidelink (SL). In some embodiments, the global model may require data such as the geographic location (or region) of the two devices, a SL received signal strength indicator (SL RSSI) of the SL between the two devices, a SL reference signal received power (SL RSRP) between the two devices, SL channel state information (SLCSI) of the SL between the two devices, a SL channel busy rate (SL CBR) of the SL between the two devices, SL transmission parameters of the SL between the two devices (e.g., modulation and coding scheme (MCS), transmission power of communication on the SL between the two devices, number of retransmissions of the SL between the two devices, priority, etc.), and / or SL hybrid automatic repeat request (SL HARQ) feedback to train the global model.

[0167] Traditionally, for models that are centrally trained (i.e., trained at a central node), this data must be sent from the two devices to the central node via a communication system (e.g., a 5G system). For the use case described previously, an ML model can be implemented in a network entity of the core network of the 5G system, which is configured to predict the QoS of D2D communication between two devices over a SL. The ML model requires SL measured data (e.g., to predict the QoS of D2D communication over a SL between two devices). However, the SL measured data is typically only available at the device, and the SL measured data is typically not sent to the core network of the 5G system. Any attempt to send the SL measured data to the core network of the 5G system will significantly increase the signaling sent between the device and the base station of the 5G-AN over the air interface (e.g., Uu interface) 3, which may not be ideal. Therefore, in this case, it can be beneficial to train the ML model using federated learning (such as described in the embodiments herein) to eliminate the need for the device (e.g., UE) to send its local training data to the core network, which reduces signaling between the device (e.g., UE) and the core network and eliminates privacy issues because only local training results (e.g., values ​​of model parameters of the locally trained model or gradients of parameters of the locally trained model) are sent from the device (e.g., UE) to the core network.

[0168] The embodiments described herein are also intended to improve the performance of FL in 5GS. Current FL involves reselecting UEs for FL (hereinafter referred to as training UEs), which increases the signaling between the UEs selected by the FL aggregator 400 for FL and sent over the air interface, and increases the latency of model training using FL. Note that selecting and configuring training UEs for FL requires more signaling to be exchanged between the FL aggregator 400 and the UEs.

[0169] Furthermore, when an already selected training UE cannot effectively participate in the training of the global model (also referred to as model training) using FL, for example due to poor quality of the air interface (e.g., Uu link) between the UE and the base station of the RAN, or due to (temporary) unavailability of local resources of the UE (e.g., computing resources, such as the amount of memory or processing resources available at the UE, power, etc.), the FL aggregator 400 may need to reselect a training UE to participate in the model training using FL. Such reselection of the training UE may further cause additional signaling to be sent between the FL aggregator 400 and the selected UE, and may further cause delays in the model training using FL, as the latency increases.

[0170] These problems may become particularly significant when UE unavailability (UE on and off availability) frequently causes frequent reselection of training UEs to participate in model training using FL at the FL aggregator 400. In addition, among UEs with local data sets of similar data distribution, the FL aggregator 400 may select only one (fewer) UEs to participate in model training using FL to avoid biasing the trained global model towards a specific data distribution. This naturally puts one or more UEs selected as training UEs at a disadvantage, because the selected UEs must participate in model training of the global model using FL and pay the associated costs to apply the resources (such as power consumption and / or processor and / or memory) required when locally training the local model using the local data set.

[0171] Therefore, the following embodiments are intended to provide a FL aggregator that minimizes the reselection of UEs participating in model training using FL, minimizes the signaling sent to the selected UEs for model training using FL (i.e., the signaling overhead between the FL aggregator and the UE), avoids interruptions during model training using FL, and avoids unfair utilization of UEs during model training using FL.

[0172] For example, in some embodiments, the FL aggregator configures the training UE (which is referred to as the first UE or primary UE in this application) to have an alternative training UE (which is also referred to as the second UE or secondary UE in this application). In these embodiments, the second UE is activated as a training UE (i.e., performs local training) only when the first UE is not available for local training.

[0173] Figure 5 , the communication (ie, signaling) between the FL aggregator 500 and the first UE and the second UE according to an embodiment of the present disclosure is shown. Figure 5 In the embodiment shown in , each of the UEs (UE 300a, UE 300b, UE 300c) communicates with the FL aggregator 500 via a link, which includes: a wireless link (e.g., a Uu link) between the UE and a base station of the RAN of the communication system and a wired link between the base station and the FL aggregator 500.

[0174] The initial operation performed by the FL aggregator 500 is similar to that performed by Figure 4 The initial operations performed by the FL aggregator 400 shown in and described above.

[0175] The FL aggregator 500 may be configured to generate a FL report configuration and send it to each UE. The FL report configuration may include information defining the content to be included in the FL report, the format of the FL report, and how to generate the FL report. For example, the FL aggregator 500 may be configured to generate a FL report configuration at 401 and send the FL report configuration to UE1 300a, UE2 300b, and UE3 300c.

[0176] Then, each of the UEs (UE 300a, UE 300b, UE 300c) may be configured to generate a FL report based on the FL report configuration and send the FL report to the FL aggregator 500. For example, at 403, UE1 300a is shown to generate a FL report and send the FL report to the FL aggregator, at 405, UE2 300b is shown to generate a FL report and send the FL report to the FL aggregator, and at 407, UE3 300c is shown to generate a FL report and send the FL report to the FL aggregator. The FL report may include information useful to the FL aggregator 500 to help the FL aggregator 500 determine which UEs to select for FL. For example, such information may include training resource availability at the UE, such as CPU, energy, measurements related to the quality of the radio link (which may be part of the link between the UE and the FL aggregator 500) (e.g., RSRP), availability of local data at the UE for training a local model at the UE (also referred to as training data availability).

[0177] Then, the FL aggregator 500 is configured to select K UEs for local training, such as Figure 5As shown in 409 in . In some cases, the selection of K UEs for local training can be random or based on any suitable UE selection scheme that takes into account the information included in the received FL report (provided by the UE). In the following example, one of the selected UEs is UE1 300a. In the following description, the selected UE is the first UE. In the following example, FL signaling for the selected UE UE1 300a for local training is shown. It should be understood that similar signaling flows will also be implemented for other selected UEs, such as, for example, UE3 300c, which can also be selected as the first UE together with UE1 300a.

[0178] Furthermore, the FL aggregator 500 is configured to select the second UE as a replacement training UE for the first UE, such as Figure 5 As shown in 501 in . Selecting the second UE as a replacement training UE for the first UE (UE2 300b in the example shown herein) may be selecting one or more candidate UEs based on at least one of the following selection criteria:

[0179] Training UE selection assistance information received from a first UE (e.g., UE1 300a), wherein the first UE includes in the training UE selection assistance information: information identifying a candidate replacement training UE;

[0180] a data distribution of data in a local data set at the first UE, and a data distribution of data in a local data set at each candidate replacement training UE;

[0181] a mobility pattern of the candidate alternative training UE relative to the first UE;

[0182] at least one characteristic of a radio link (i.e., a Uu link) between each candidate alternative training UE and a base station of the RAN;

[0183] The location of the first UE and each candidate alternative training UE.

[0184] For example, with respect to the data distribution of the local data set of the candidate alternative training UE, it is noted that the FL aggregator 500 generally does not select two training UEs (in the same iteration) with very similar data distributions to avoid bias in the trained global model. Therefore, at 409, a UE that is not selected as a training UE due to having a data distribution similar to the data distribution of the local data set of the already selected training UE may be selected as a replacement training UE for the already selected training UE. In this way, even when the training UE is not available for local training, the replacement training UE may perform local training using a local data set similar to the local data set of the already selected training UE. In addition, selecting the replacement training UE or the second UE based on the location of each of the first UE and the candidate replacement training UEs may be based on whether the candidate replacement training UE located near the first UE (e.g., within a predetermined distance of the first UE) is likely to have similar data, particularly when the training data in the local data set includes a measurement of the quality of the Uu link between the candidate replacement training UE and a base station of the RAN or a measurement of the quality of the sidelink between the first UE and the candidate replacement training UE. Therefore, a nearby candidate replacement training UE may be selected as the replacement training UE or the second UE.

[0185] Selecting an alternative training UE or a second UE based on a mobility pattern of the candidate alternative training UE relative to the first UE The selection may be based on whether the nearby candidate alternative training UE or the candidate alternative training UE moves with the first UE (e.g., the first UE and the candidate alternative UE are queuing). The mobility pattern may, for example, take into account the trajectory and / or speed of the candidate alternative training UE relative to the first UE. Selecting a second UE based on proximity and relative mobility makes it possible for the second UE to have similar data, particularly when the training data set includes training data indicating radio link (i.e., Uu link) measurements or sidelink measurements. Therefore, a nearby UE may be selected as an alternative training UE.

[0186] In some embodiments, selecting the alternative training UE or the second UE from the candidate alternative training UEs may be based on a sidelink reachability parameter value. The sidelink reachability parameter value may be related to the ability of the first UE to communicate directly with the candidate alternative training UE. In some embodiments, the sidelink reachability parameter value may be determined based on information available at a base station (e.g., gNB) of the RAN from SL-related measurement reports in SL communication and SL relay-related scenarios. For example, selecting the alternative training UE or the second UE from the candidate alternative training UEs may be configured to: select an alternative training UE having a high-quality communication link between the first UE and the candidate alternative training UE so that when selected, the handover between the first (training) UE and the second (alternative training) UE is less likely to fail.

[0187] The selection criteria of at least one characteristic of the radio link (e.g., Uu link) between the candidate replacement training UE and the base station or RAN may be based on ensuring that the candidate replacement training UE can participate in FL when the training UE is unable to participate in FL due to a failure of the radio link (e.g., Uu link).

[0188] The FL aggregator 500 is then caused to configure the selected second UE (UE2 300b in this example) as a replacement training UE for the first UE UE1 300a, such as Figure 5 The operation of configuring the second UE as a replacement training UE for the first UE may include: generating a replacement training UE configuration, and sending the replacement training UE configuration to the first UE and the second UE, wherein the replacement training UE configuration sent to the first UE and the second UE respectively includes: the identifiers of the second UE and the first UE, such as Figure 5 In some embodiments, the alternative training UE configuration sent to the first UE and the second UE may include:

[0189] Unique ID of the training UE;

[0190] Replace the unique ID of the training UE;

[0191] A condition identifier is configured to identify a trigger condition that triggers the transfer of local model training from the training UE to the substitute training UE.

[0192] In some example embodiments, the triggering conditions that trigger the transfer of local model training from the training UE to the replacement training UE may include:

[0193] a minimum quality of the radio link (e.g., Uu link) between the training UE and a base station (e.g., gNB) of the RAN, which the training UE must have in order to continue as a training UE;

[0194] the minimum computational and power resources that a training UE must have in order to continue to be a training UE; and

[0195] The minimum security / integrity level associated with training the UE's local data set that must be met.

[0196] After sending the alternative training UE configuration to the first UE UE1 300a and the second UE UE2 300b, so that the second UE is the alternative training UE of the first UE UE1 300a, the FL aggregator 500 can perform FL using the training of the local model at the first UE, such as Figure 5 As shown in 507.

[0197] In some embodiments, during FL with the first UE UE1 300a (or FL with UE1), model training at the first UE UE1 300a may include: the FL aggregator 500 broadcasts or otherwise sends a global model and training configuration to the selected first UE and the selected second UE, such as Figure 5 As shown in 509 in FIG. In other words, the FL aggregator 500 is configured to broadcast the global model and the training configuration to the K selected first UEs (and the selected second UEs associated with the selected first UEs).

[0198] Then, each of the selected first UEs is configured to perform local training. For example, UE1 performs local training, such as Figure 5 As shown in 511 in . Local training is performed at all selected first UEs among the K selected first UEs that have received the global model and the training configuration, and the first UE is configured to update its corresponding local model based on the global model and the training configuration.

[0199] After performing local training, each of the K selected first UEs is then configured to report the result of its local training (i.e., local training result). For example, the selected first UE is configured to send the local training result to the FL aggregator 500. For UE1 300a, the local training result is sent to the FL aggregator 500, such as Figure 5 As shown in 513 in . The local training result sent by each of the selected first UEs may be the value of the parameter of the local model when the local training is completed, the update of the parameter of the local model, or the gradient of the parameter of the local model. The update of the parameter of the local model may be the difference between the original value of the parameter and the final value of the parameter after the local training is completed.

[0200] Furthermore, the FL aggregator 500 is then configured to aggregate the training results received from the distributed nodes (the selected first UE), such as Figure 5 In other words, the FL aggregator 500 aggregates or combines local training results to obtain an aggregated training result of the global model, and updates the parameters of the global model based on the aggregated training result.

[0201] In some embodiments, the second UE is configured to perform local model training when it is determined that the first UE is not available for local model training. In some embodiments, the second UE (also referred to as an alternative UE or auxiliary UE) is configured to perform local model training based on receiving a local training activation request (including an indication that the first UE cannot perform local model training) from the selected first UE or FL aggregator 500.

[0202] So, for example, Figure 5 As shown in FIG. 3 , the first UE UE1 300a is configured to: determine that the first UE is experiencing a local training unavailable event, such as Figure 5 517 in . For example, in some embodiments, the training UE (first UE) is configured to: evaluate at least one trigger condition that triggers the transmission of local model training to the alternative training UE (included in the alternative training UE configuration). An example of a trigger condition that triggers the transmission of local model training to the monitored alternative training UE may be when the quality of the wireless link (e.g., Uu link) between the base station and the first UE drops below a threshold, which causes the first UE UE1300a to initiate the transmission of local model training to the second UE UE2 300b.

[0203] Therefore, the first UE UE1 300a may be configured to generate a request and send a request to the second UE UE2 300b to perform local model training (or model training transmission request), such as Figure 5 519 in. In some embodiments, the request to perform local model training (also referred to herein as a local model training request) may also include: an indication of the reason for transmitting the local model training to the second UE UE2 300b, and / or the duration for performing the local model training at the second UE UE2 300b.

[0204] Then, the second UE UE2 300b may be configured to: in response to accepting to perform local model training, generate a local training acceptance message (which may be a confirmation or response message in some embodiments), and send the local training acceptance message to the first UE, such as Figure 5 As shown in 521 in . The local training acceptance message generated by the second UE UE2 300b indicates to the FL aggregator 500 that the second UE UE2 300b will use the local data set to perform local model training.

[0205] In some embodiments, the first UE (e.g., UE1) is configured to send an indication of its unavailability to the FL aggregator 500, and the FL aggregator 500 is configured to forward the indication to the second UE or explicitly activate the second UE UE2 300b to perform local model training during the FL iteration (by generating an appropriate trigger message or model training transfer request). Note that the FL aggregator 500 does not perform reselection of the training UE at this stage, but only sends an activation message to the second UE (or alternative UE). In such an embodiment, the second UE UE2 300b is then configured to send a local training acceptance message (or confirmation or response) directly to the FL aggregator 500 instead of the first UE, such as Figure 5 as shown in .

[0206] Then, the first UE UE1 300a is configured to stop monitoring the global model and training configuration sent by the FL aggregator 500, such as Figure 5 As shown in 523 in, and the second UE UE2300b is configured to: start monitoring the global model and training configuration sent by the FL aggregator 500, such as Figure 5 As shown in 525.

[0207] This actually enables the second UE UE2 300b to start local model training. Figure 5 As shown in 527, this is the local training of UE2 300b.

[0208] In some embodiments, local training at UE2 300b during FL iterations may include the FL aggregator 500 performing broadcasting of the global model and training configuration, such as Figure 5 As shown in 529.

[0209] Then, each of the 'active' second UEs in the 'active' second UE is configured to perform local model training. For example, UE2 300b performs local model training (ie, performs local training on the local model using the local data set), such as Figure 5 As shown in 531.

[0210] After performing local training, the 'active' second UE is then configured to report its local training results to the FL aggregator 500. For example, for UE2 300b, the local training results are reported to the FL aggregator 500 by, for example, sending a message including the local training results, such as Figure 5 As shown in 533.

[0211] In an embodiment where the second UE (e.g., UE2 300b) is configured to perform local model training based on receiving a local training activation request (including an indication that the first UE (e.g., UE1 300a) is unable to perform local model training) from the selected first UE or FL aggregator 500, the second UE (e.g., UE2 300b) may also send an indication that the second UE is participating in FL (i.e., an indication that the second UE is a substitute training UE for the first UE) to the FL aggregator 500 in addition to the local training result, so that the FL aggregator 500 does not discard the local training result sent by the second UE. The local training result and the indication may be included in a message sent by the second UE (e.g., UE2 300b) to the FL aggregator 500. The indication that the second UE is participating in the FL of the FL aggregator 500 is also referred to herein as a training switching indication. Figure 5 As shown in 533.

[0212] Furthermore, the FL aggregator 500 is then configured to aggregate the received local training results, such as Figure 5 As shown in 535.

[0213] Therefore, in these embodiments as shown above, the FL aggregator 500 is configured to use the local training results sent by the second UE (or substitute UE) in the FL iteration when the first UE is unable to perform local model training (or unable to report its local training results to the FL aggregator 500). After receiving the local training results, the FL aggregator 500 is configured to perform aggregation or combination of the local training results to obtain the aggregated training results of the global model. In addition, in some embodiments, the FL aggregator 500 is configured to further indicate or register the change of the training UE performing the local model training to maintain a log of the availability of the training UE for local model training or to transmit the local model training to the substitute training UE.

[0214] In some embodiments, the FL aggregator 500 is further configured to: request and obtain candidate replacement training UEs (preferably UEs that serve as replacement training UEs) determined by each selected training UE. In other words, the FL aggregator 500 may request the UE to provide information identifying other UEs, which may be selected as the second UE or replacement training UEs when the UE is selected as the first UE (i.e., the training UE).

[0215] Figure 6 The communication (ie, signaling) between the FL aggregator 600 and the first UE and the second UE according to another embodiment of the present disclosure is shown in more detail in FIG. Figure 6 In the embodiment shown in , the communication (eg, signaling) is modified to include preferred candidate information. In the following examples, a candidate node (eg, a candidate first UE or a candidate second UE) is a UE potentially selected as the first UE or the second UE, respectively.

[0216] exist Figure 6 In the illustrated embodiment, each of the UEs (UE 300a, UE 300b, UE 300c) communicates with the FL aggregator 600 via a link, which includes: a wireless link (e.g., a Uu link) between the UE and a base station of the RAN of the communication system, and a wired link between the base station and the FL aggregator 600.

[0217] The initial operation performed by FL aggregator 600 is similar to that performed by Figure 4 and Figure 5 The initial operations performed by the FL aggregator 400 shown in and described above.

[0218] The FL aggregator 600 may be configured to generate a FL report configuration and send the FL report configuration to each UE. As described above, the FL report configuration may include information defining the content to be included in the FL report, the format of the FL report, and how to generate the FL report. In some embodiments, the FL report configuration (e.g., information defining the content to be included in the FL report) also requests the FL report to include information indicating a candidate (or preferred) alternative training UE. For example, the FL aggregator 600 may be configured to generate a FL report configuration at 601 and send the FL report configuration to UE1 300a, UE2 300b, and UE3 300c, the FL report configuration including a request indicating a candidate alternative training UE.

[0219] In some embodiments, each candidate first UE (when indicated) may be configured to determine one or more candidate alternative training UEs (or candidate second UEs). In some embodiments, the determination of the candidate alternative training UEs may be based at least on the following:

[0220] The data distribution of the data in the local data set at the candidate first UE, and the data distribution of the data in the local data set at the candidate replacement training UE (the reason is similar to the above about Figure 5 The reason described above). In such an embodiment, the candidate first UE (e.g., UE1 300a) may be configured to: request the neighboring UE to report the data distribution of its data in its local data set at the neighboring UE. For example, the data distribution may be any one or any combination of the following items: range, interquartile range, standard deviation, variance;

[0221] The location proximity between the candidate first UE and the candidate alternative training UE and / or the mobility pattern of the candidate alternative training UE relative to the candidate first UE (for the same reasons as described above with respect to Figure 5 For example, the candidate first UE may be configured to request the neighboring candidate replacement training UE to report its position and mobility mode, or the candidate first UE may be configured to monitor the parameters (such as cooperative sensing messages) sent by the neighboring candidate replacement training UE on the sidelink or by monitoring the SL-RSRP.

[0222] Calculate the proximity and relative mobility of neighboring candidate replacement training UEs.

[0223] So, for example, in Figure 6 , UE1 300a determines a candidate alternative training UE list (about UE1 300a) at 603, UE2 300b determines a candidate alternative training UE list (about UE2 300b) at 605, and UE3 300c determines a candidate alternative training UE list (about UE3 300c) at 607.

[0224] Then, each of the UEs may be configured to generate a FL report and send the FL report to the FL aggregator 600 based on the FL report configuration, wherein the FL report configuration also includes: the identified candidate second UE or candidate replacement training UE. For example, at 609, UE1 300a is shown to generate a FL report and send the FL report to the FL aggregator including the candidate replacement training UE, at 611, UE2 300b is shown to generate a FL report and send the FL report to the FL aggregator, and at 613, UE3 300c is shown to generate a FL report and send the FL report to the FL aggregator.

[0225] Then, the FL aggregator 600 is configured to select K UEs for local training, such as Figure 6 615 in FIG. In the following example, one of the selected first UEs is UE1 300a, and another of the selected first UEs is UE3 300c. In the following example figure, FL signaling is shown for the selected UE UE1 300a for local training. It should be understood that similar signaling flows will also be implemented for the other selected UE UE3 300c.

[0226] In addition, the FL aggregator 600 is configured to select the second UE as a replacement training UE for the first UE, such as Figure 6 617 in . The selection of the second UE as a replacement training UE for the first UE (UE2 300b in the example shown herein) may be based on a candidate replacement list included in the FL report received from UE1 300a. In some embodiments, the selection of the second UE as a replacement training UE for the first UE is based on the candidate list and any other selection criteria for the replacement training UE described above.

[0227] The signaling flows between the FL aggregator 600 and the training UE (first UE) and the alternative training UE (second UE) may then be similar to those discussed above.

[0228] For example, the FL aggregator 600 may then be caused to configure the selected second UE (UE2 300b in this example) as a replacement training UE for the first UE UE1 300a, such as Figure 6 The operation of configuring the second UE as a replacement training UE for the first UE may include generating a replacement training UE configuration, and sending the replacement training UE configuration to the first UE and the second UE, wherein the replacement training UE configuration sent to the first UE and the second UE respectively includes the identifiers of the second UE and the first UE, such as Figure 6 As shown in 621.

[0229] After sending the alternative training UE configuration to the first UE UE1 300a and the second UE UE2 300b, so that the second UE is the alternative training UE of the first UE UE1 300a, the FL aggregator 600 can perform FL using the training of the local model at the first UE, such as Figure 6 As shown in 623.

[0230] In some embodiments, during FL with the first UE UE1 300a (or FL with UE1), model training at the first UE UE1 300a may include the FL aggregator 600 broadcasting or otherwise sending a global model and training configuration to the selected first UE and the selected second UE, such as Figure 6 625 in . In other words, the FL aggregator 600 is configured to broadcast the global model and the training configuration to the K selected first UEs (and the selected second UEs associated with the selected first UEs).

[0231] Then, each of the selected first UEs is configured to perform local model training. For example, UE1 300a performs local model training, such as Figure 6 As shown in 626 in the figure, local model training is performed at all K selected UEs that have received the global model and training configuration, and the first UE is configured to update its corresponding local model based on the global model and training configuration.

[0232] After performing local model training, each of the K selected first UEs is then configured to report (its local training results) to the FL aggregator 600. For example, the selected first UE is configured to send the local training results to the FL aggregator 600, such as Figure 6 As shown in 627 in the figure, the local training result is sent by UE1 300a to the FL aggregator 600. (Although not shown, the selected other first UEs (e.g., UE3 300c) are configured to report the results of their local training). The local training result sent by each selected first UE may be the value of the parameter of the local model when the local training is completed, the update of the parameter of the local model, or the gradient of the parameter of the local model. The update of the parameter of the local model may be the difference between the original value of the parameter and the final value of the parameter after the local training is completed.

[0233] Furthermore, the FL aggregator 600 is then configured to perform aggregation on the training results received from the distributed nodes to obtain an aggregated training result of the global model, such as Figure 6 In other words, the FL aggregator 600 aggregates or combines local training results to obtain an aggregated training result of the global model, and updates the parameters of the global model based on the aggregated training result.

[0234] In some embodiments, the second UE is configured to perform local model training when it is determined that the first UE is not available for local training. In some embodiments, the second UE (also referred to as an alternative training UE or auxiliary UE) is configured to: based on receiving a local training activation request (including an indication that the first UE cannot perform local training) from the selected first UE UE1 300a or FL aggregator 600, perform local training of the local model.

[0235] So, for example, Figure 6 As shown in FIG. 3 , the first UE UE1 300a is configured to: determine that the first UE is experiencing a local training unavailable event, such as Figure 6 As shown in 631.

[0236] Then, the first UE UE1 300a is configured to generate and send a request for performing local model training (or a model training transmission request) to the second UE UE2 300b, such as Figure 6 633 in . In some embodiments, the request to perform local model training (also referred to herein as a local model training request) may also include: an indication of the reason for transmitting the local model training to the second UE UE2 300b (or an alternative training UE), and / or the duration for which the local model training is to be performed at the second UE UE2 300b.

[0237] Then, the second UE UE2 300b may be configured to generate a local model training acceptance message (which may be a confirmation or response message in some embodiments) in response to accepting the execution of the local model training, and send the local model training acceptance message to the first UE, such as Figure 6 635. The local model training acceptance message generated by the second UE UE2 300b indicates to the first UE UE1 300a that the second UE UE2 300b will perform local model training (ie, perform local training of the local model using the local data set).

[0238] As described above, in some embodiments, the first UE (e.g., UE1) is configured to send an indication of its unavailability to the FL aggregator 600, and the FL aggregator 600 is configured to: forward the indication to the second UE or explicitly activate the second UE UE2 300b for local model training (by generating a suitable trigger message or model training transfer request). In such an embodiment, the second UE UE2 300b is then configured to: in response to accepting to perform local model training, send a local model training acceptance message (e.g., a suitable confirmation or response message) to the FL aggregator 600.

[0239] Then, the first UE UE1 300a is configured to stop monitoring the global model and training configuration sent by the FL aggregator 600, such as Figure 6 As shown in 637 in the figure, the second UE UE2300b is configured to: start monitoring the global model and training configuration sent by the FL aggregator 600 through step 639.

[0240] This actually enables the second UE UE2 300b to start local model training, such as Figure 6 As shown in 641.

[0241] In some embodiments, local model training performed at UE2 300b during FL iterations may include the FL aggregator 600 performing broadcasting of the global model and training configuration, such as Figure 6 As shown in 643.

[0242] Then, each of the 'active' second UEs is configured to perform local model training (ie, locally train the local model using the local data set). For example, UE2 300b performs local model training (ie, locally trains the local model using the local data set), such as Figure 6 As shown in 645.

[0243] After performing local model training, the 'active' second UE is then configured to report its local training results to the FL aggregator 600. For example, for UE2 300b, the local training results are reported to the FL aggregator 600 by, for example, sending a message including the local training results, such as Figure 6 As shown in 647.

[0244] In an embodiment where the second UE (e.g., UE2 300b) is configured to perform local model training based on receiving a local training activation request (including an indication that the first UE (e.g., UE1 300a) is unable to perform local model training) from the selected first UE or FL aggregator 600, the second UE (e.g., UE2 300b) may also send an indication that the second UE is participating in FL (i.e., an indication that the second UE is a substitute training UE for the first UE) to the FL aggregator 600 in addition to the local training result, so that the FL aggregator 600 does not discard the local training result sent by the second UE. The local training result and the indication may be included in a message sent by the second UE (e.g., UE2 300b) to the FL aggregator 600. The indication that the second UE is participating in FL to the FL aggregator 600 is also referred to herein as a training switching indication. Figure 6 In addition, the FL aggregator 600 is then configured to perform aggregation on the received local training results, such as Figure 6 As shown in 649.

[0245] In some embodiments, the training UE and / or the alternative training UE may be configured to reject the alternative training UE configuration provided by the FL aggregator 600. For example, in the case where the second UE is not reachable from the first UE via a sidelink (SL), the first UE may reject the alternative training UE configuration. Similarly, when the first UE is not reachable from the second UE via a sidelink, the second UE may reject the alternative training UE configuration provided by the FL aggregator 600.

[0246] In other words, the embodiment shows that the first UE or the second UE can accept or reject the alternative training UE configuration received from the FL aggregator 600 based on the determination of whether the second UE will receive (or not receive) a local model training request from the first UE or whether the first UE will receive (or not receive) a local model training acceptance from the second UE. Then, rejecting the alternative training UE configuration can cause or trigger the FL aggregator 600 to provide a new alternative training UE configuration.

[0247] The signaling associated with the example rejection of the alternative training UE configuration is as follows Figure 7 As shown, the figure is based on Figure 5 The example shown is modified to incorporate the ability for the first UE and the second UE to accept or reject an alternative UE training configuration received from the FL aggregator.

[0248] Initial operation and Figure 5 The implementation shown in is similar.

[0249] The FL aggregator 700 may be configured to generate a FL report configuration and send it to each UE. The FL report configuration may include information defining the content to be included in the FL report, the format of the FL report, and how to generate the FL report (in some embodiments, a list of alternative candidates). For example, the FL aggregator 700 may be configured to generate a FL report configuration at 701 and send the FL report configuration to UE1 300a, UE2 300b, and UE3 300c.

[0250] Each of the UEs (UE 300a, UE 300b, UE 300c) may then be configured to generate and send a FL report to the FL aggregator 700 (and in some embodiments also determine and output a candidate list) based on the FL report configuration.

[0251] For example, at 703 UE1 300a is shown generating a FL report and sending the FL report to the FL aggregator, at 705 UE2 300b is shown generating a FL report and sending the FL report to the FL aggregator, and at 707 UE3 300c is shown generating a FL report and sending the FL report to the FL aggregator.

[0252] The FL aggregator 700 is then configured to select K UEs for local model training, such as Figure 7 As shown in 709. The selection of K UEs for local model training may be based on any one of the above selection schemes.

[0253] In addition, the FL aggregator 700 is configured to select the second UE as a replacement training UE for the first UE, such as Figure 7 As shown in 711 in FIG. 8 . The selection of the replacement training UE may be based on any one of the above selection schemes.

[0254] The FL aggregator 700 is then caused to configure the selected second UE (UE2 300b in this example) as a replacement training UE for the first UE UE1 300a, such as Figure 7 As shown in 713.

[0255] Configuring the second UE as a replacement training UE for the first UE UE1 300a may include: generating a replacement training UE configuration, and sending the replacement training UE configuration to the first UE and the second UE, wherein the replacement training UE configuration sent to the first UE and the second UE respectively includes the identifiers of the second UE and the first UE, such as Figure 7 As shown in 715.

[0256] In these embodiments, the first UE and the second UE are configured to evaluate whether the alternative training UE configuration is achievable or acceptable to the UE receiving the alternative training UE configuration.

[0257] Therefore, UE1 300a is configured to evaluate whether the alternative training UE configuration can be implemented at UE1 300a or is acceptable to UE1 300a at 717, and UE2 300b is configured to evaluate whether the alternative training UE configuration can be implemented at UE2 300b or is acceptable to UE2 300b at 719. As described above, the evaluation of whether the alternative training UE configuration can be implemented at the UE or is acceptable to the UE can be based on whether the local model training request or the local model training acceptance can be sent from the first UE to the second UE or from the second UE to the first UE via a sidelink between the first UE and the second UE, respectively.

[0258] The UE (e.g., the first UE UE1 300a and the second UE UE2 300b) may be configured to generate and send an acceptance message when the alternative training UE configuration can be implemented at the UE or is acceptable to the UE, or to generate and send a rejection message when the alternative training UI configuration cannot be implemented at the UE or is unacceptable to the UE. For example, UE1 300a is configured to generate and send an acceptance message or a rejection message at 721 regarding whether the alternative training UE configuration can be implemented at UE1 300a or is acceptable to UE1 300a, and UE2 300b is configured to generate an acceptance message or a rejection message at 723 regarding whether the alternative training UE configuration can be implemented at UE2 300b or is acceptable to UE2 300b.

[0259] The FL aggregator 700 may then optionally (in response to receiving a rejection message from at least one of the first UE or the second UE) generate a new alternative training UE configuration and send the new alternative training UE configuration to the selected first UE and second UE, such as Figure 7 As shown in 725.

[0260] These first UE and second UE selection and configuration steps may be repeated until the configuration is accepted.Then, communication or signaling related to the above-mentioned local model training may be implemented.

[0261] Therefore, after configuring the first UE UE1 300a and the second UE UE2 300b so that the second UE is a replacement training UE for the first UE UE1 300a, the iteration of FL can be performed with the first UE UE1 300a, as shown in FIG. Figure 7 As shown in 731.

[0262] In some embodiments, the iteration of FL training with the first UE UE1 300a at 731 may include the FL aggregator 700 performing a broadcast of a global model and training configuration, such as Figure 7 As shown in 733.

[0263] Then, each of the selected first UEs is configured to perform local model training. For example, UE1 300a performs local model training, such as Figure 7 As shown in 735.

[0264] After implementing the local training, the K selected first UEs are then configured to report the results of the local model training. For example, for UE1 300a, the local training results are sent to the FL aggregator 700, such as Figure 7737 in . (Although not shown, the other selected first UEs (e.g., UE3 300c) are configured to report the results of their local training). The local training results sent by each of the selected first UEs may be the value of the parameter of the local model when the local training is completed, the update of the parameter of the local model, or the gradient of the parameter of the local model. The update of the parameter of the local model may be the difference between the original value of the parameter and the final value of the parameter after the local training is completed.

[0265] Furthermore, the FL aggregator 700 is then configured to perform aggregation on the received local training results to obtain an aggregated training result of the global model, such as Figure 7 As shown in 739.

[0266] In some embodiments, the second UE is configured to perform local model training when it is determined that the first UE is not available for local model training. In some embodiments, the second UE (also referred to as an alternative UE or auxiliary UE) is configured to perform local model training based on receiving a request to perform local model training (or a model training transmission request). The request to perform local model training (also referred to herein as a local model training request) includes an indication that the first UE UE1 300a is unable to perform local model training). The local model training request is received by the second UE from the selected first UE UE1 300a or the FL aggregator 700.

[0267] So, for example, Figure 7 As shown in FIG. 1 , the first UE UE1 300a is configured to: determine that the first UE UE1 300a is experiencing a local training unavailable event, such as Figure 7 741 in . In other words, for example, based on the lack of available computing or power resources at the first UE, the first UE UE1 300a determines that it cannot perform local model training.

[0268] Then, the first UE UE1 300a is configured to generate and send a request for transmitting local model training (also referred to as a model training transmission request) to the second UE UE2 300b, such as Figure 7 As shown in 743. In some embodiments, the local model training request (or model training transmission request) may also include information indicating the reason for transmitting the local model training to the second UE or the alternative training UE and / or the duration of the local model training to be performed at the second UE.

[0269] Then, the second UE UE2 300b may be configured to generate and send a local training acceptance message (eg, a suitable confirmation or response message) to the first UE in response to accepting to perform local model training, and send the local training acceptance message, such as Figure 7As shown in 745.

[0270] In some embodiments, the first UE (e.g., UE1 300a) is configured to send an indication of its unavailability directly to the FL aggregator 700, and the FL aggregator 700 is configured to forward the indication to the second UE or explicitly activate the second UE UE2 300b for local training of the local model (by generating a suitable trigger message, a local model training request, or a model training transfer request, and sending the trigger message, the local model training request, or the model training transfer request to the second UE). In such an embodiment, the second UE UE2 300b is then configured to send a local model training acceptance message (e.g., a suitable confirmation or response message) to the FL aggregator 700 in response to accepting the execution of the local model training.

[0271] Then, the first UE UE1 300a is configured to stop monitoring the global model and training configuration sent by the FL aggregator 700, such as Figure 7 As shown in 747 in , the second UE UE2300b is configured to start monitoring the global model and training configuration sent by the FL aggregator 700.

[0272] This actually enables the second UE UE2 300b to participate in the iteration of FL, such as Figure 7 As shown in 751.

[0273] In some embodiments, during an iteration of FL (also referred to as FL iteration), the FL aggregator 700 may be configured to: perform a broadcast of the global model and training configuration, such as Figure 7 As shown in 753. In other words, the FL aggregator 700 can perform global model and training configuration for UE1 300a and UE2 300b.

[0274] Then, each of the 'active' second UEs in the 'active' second UE is configured to: perform local model training (ie, perform local training on the local model using the local data set). For example, UE2 300b performs local model training, such as Figure 7 As shown in 755.

[0275] After performing local model training, the 'active' second UE is then configured to report its local training results back to the FL aggregator 700. For example, for UE2 300b, the local training results are sent to the FL aggregator 700 by, for example, sending a message including the local training results, such as Figure 7 As shown in 757.

[0276] In an embodiment where the second UE (e.g., UE2 300b) is configured to perform local model training based on receiving a local training activation request (including an indication that the first UE (e.g., UE1 300a) is unable to perform local model training) from the selected first UE or FL aggregator 700, the second UE (e.g., UE2 300b) may send an indication that the second UE is participating in FL (i.e., an indication that the second UE is a substitute training UE for the first UE) to the FL aggregator 700 in addition to the local training result, so that the FL aggregator 700 does not discard the local training result sent by the second UE. The local training result and the indication may be included in a message sent by the second UE (e.g., UE2 300b) to the FL aggregator 700. The indication that the second UE is participating in FL to the FL aggregator 700 is also referred to herein as a training switching indication. Figure 7 As shown in 757.

[0277] Furthermore, the FL aggregator 700 is then configured to aggregate the local training results received from the active second UE UE2 300b and other first and second UEs, such as Figure 7 As shown in 759.

[0278] Thus, the above implementations and embodiments are intended to provide uninterrupted model training using federated learning, even when one or more distributed nodes (e.g., training UEs) fail to perform local training (e.g., when a first UE is temporarily unavailable, a second replacement UE can act as a training UE until the first UE is online again). In addition, when one or more training UEs fail to perform local training, it minimizes the signaling and computation associated with the reselection of training UEs (UEs participating in model training using FL). This is particularly beneficial in situations where the unavailability (on and off availability) of training UEs is frequent (which would otherwise trigger frequent reselection of training UEs to perform model training using FL at the FL aggregator).

[0279] In addition, the above embodiments can provide load balancing in terms of local model training to avoid utilizing a UE or a small group of UEs in local model training using federated learning. Thus, embodiments enable local model training to be temporarily transferred to another UE so that an 'over-adopted' or 'over-used' UE does not always perform local model training.

[0280] The FL aggregators 500, 600, 700 of the present disclosure described herein may be used in Figure 1Alternatively, the FL aggregators 500, 600, 700 of the present disclosure described herein may be implemented in the network data analysis function of the 5GC, the operation, management and maintenance entity of the 5GS, or any network function (NF) of the 5GC (such as AMF, SMF, PCF, etc.).

[0281] It should be understood that these devices may include or be coupled to other units or modules for transmission and / or reception, such as radio components or radio heads. Although these devices are described as one entity, different modules and memories may be implemented in one or more physical or logical entities.

[0282] It should be noted that although some embodiments have been described for 5G systems, similar principles may also be applied to other networks and communication systems. Therefore, although certain embodiments are described above by way of example with reference to certain example architectures of radio access networks and core networks, radio access technologies and standards, the embodiments may be applied to any other suitable form of communication system implementing other radio access technologies than those shown and described herein.

[0283] It is also noted herein that while the above describes exemplifying embodiments, there are numerous variations and modifications which may be made to the disclosed solution without departing from the scope of the present invention.

[0284] Generally, in various embodiments, the FL aggregator may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. Some aspects of the disclosure may be implemented in hardware, while other aspects may be implemented in firmware or software, which may be executed by a controller, microprocessor, or other computing device, but the disclosure is not limited thereto. Although various aspects of the disclosure may be shown and described as block diagrams, flow charts, or using some other graphical representation, it is well understood that, as non-limiting examples, these blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic, general purpose hardware or controllers or other computing devices, or some combination thereof.

[0285] As used in this application, the term "circuitry" may refer to one or more or all of the following:

[0286] (a) hardware circuit implementation only (such as implementation only in analog and / or digital circuitry) and

[0287] (b) a combination of hardware circuitry and software such as (where applicable):

[0288] (i) a combination of analog and / or digital hardware circuits and software / firmware, and

[0289] (ii) hardware processor(s) (including digital signal processor(s)), software, and any portion of memory(s) with software, which work together to enable a device such as a mobile phone or server to perform various functions), and (c) hardware circuit(s) and / or processor(s), such as microprocessor(s) or portion(s) of microprocessor(s), which requires software (e.g. firmware)

[0290] But when it is not needed, the software does not need to exist.

[0291] This definition of circuitry applies to all uses of the term in this application, including in any claims. As another example, as used in this application, the term circuitry also covers an implementation of only a hardware circuit or processor (or multiple processors) or a portion of a hardware circuit or processor and its accompanying software and / or firmware. For example, if applicable to a particular claim element, the term circuitry also includes a baseband integrated circuit or processor integrated circuit for a mobile device, or a similar integrated circuit in a server, cellular network device, or other computing or network device.

[0292] Embodiments of the present disclosure may be implemented by computer software executable by a data processor of a mobile device, such as in a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products, including software routines, applets and / or macros) may be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer executable components that are configured to perform an embodiment when the program is running. One or more computer executable components may be at least one software code or a portion thereof.

[0293] Furthermore, in this regard, it should be noted that any block of the logic flow as shown in the figure may represent program steps, or interconnected logic circuits, blocks and functions, or a combination of program steps and logic circuits, boxes and functions. The software may be stored on physical media implemented within a processor such as memory chips or memory blocks, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants CDs. Physical media are non-transitory media.

[0294] The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. The data processor may be of any type suitable for the local technical environment and may include one or more of a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an FPGA, a gate-level circuit, and a processor based on a multi-core processor architecture, as non-limiting examples.

[0295] Embodiments of the present disclosure may be implemented in various components such as integrated circuit modules. The design of integrated circuits is generally a highly automated process. Complex and powerful software tools are available to convert logic level designs into semiconductor circuit designs for etching and formation on semiconductor substrates.

[0296] The independent claims define the scope of protection sought by various embodiments of the present disclosure. Embodiments and features described in this specification that do not fall within the scope of the independent claims, if any, are to be construed as examples that aid in understanding the various embodiments of the present disclosure.

[0297] The foregoing description provides a complete and informative description of exemplary embodiments of the present disclosure by way of non-limiting examples. However, when read in conjunction with the accompanying drawings and the appended claims, various modifications and adaptations may be apparent to those skilled in the relevant art in view of the foregoing description. However, all of these and similar modifications of the teachings of the present disclosure will still fall within the scope of the present invention as defined in the appended claims. In fact, there are additional embodiments that may include a combination of one or more embodiments with any other embodiments previously discussed.

Claims

1. An apparatus configured to train a model in a communication network using federated learning, the apparatus comprising means for: Select at least two additional devices to train the local model; further selecting an alternative device for at least one of the selected at least two additional devices; as well as configuring each of the at least two further devices to train the local model, and configuring the selected alternative device of the at least one further device of the two further devices to train the local model; receiving local training results from at least one further device of the at least two further devices, and receiving local training results from the selected replacement device for the at least one further device of the two further devices; as well as The local training results are combined to generate an aggregated training result of the model.

2. The apparatus of claim 1 , wherein the means for further selecting the alternative apparatus for at least one of the selected at least two further apparatuses is configured to: select the alternative apparatus based on information indicating at least one of: similarity of the data distribution of the data of the local data set of the at least one further device to the data distribution of the data of the local data set of the replacement device; the location of the further device; the location of the replacement device; the proximity of the additional device to the alternative device; a mobility pattern of the alternative device relative to the further device; the quality of communications on a sidelink between the further device and the alternative device; At least one characteristic of a wireless link between the further apparatus and a base station of the radio access network.

3. An apparatus according to claim 1 or 2, wherein the component is further used to: receive information indicating one or more candidate alternative devices from the at least one other device, wherein the component for selecting the alternative device for the at least one other device among the selected at least two other devices is used to: select the alternative device from the one or more candidate alternative devices identified by the other device.

4. The apparatus according to claim 3, wherein the component is further configured to generate a FL report configuration and send the FL report configuration to each of the at least two further apparatuses, wherein the FL report configuration comprises an indicator, the indicator being enabled to enable the at least two further apparatuses to generate a FL report, the FL report comprising: Information identifying one or more candidate replacement devices.

5. The apparatus according to any one of claims 1 to 4, wherein the means for further configuring each of the at least two further apparatuses to train the local model at the at least two further apparatuses and configuring each alternative apparatus to train the local model at the alternative apparatus is configured to: generate an alternative training UE configuration for the at least one further apparatus of the at least two further apparatuses and the alternative apparatus, the alternative training UE configuration comprising at least one of the following: a further device identifier configured to uniquely identify the at least one further device of the at least two further devices; an alternative device identifier configured to uniquely identify another of said alternative devices; A condition identifier configured to identify a trigger condition, under which at least one of the at least two additional devices is unable to train the local model, and the trigger condition causes the alternative device to perform local model training.

6. The apparatus according to claim 5, wherein the trigger condition comprises at least one of the following: a minimum quality of a radio link between the further device and a base station of the radio access network; a minimum computing resource availability at the further device; a minimum power resource availability at the further device; a minimum security level associated with a local data set of the further device; as well as A minimum integrity level associated with the local data set of the further device.

7. The apparatus of any one of claims 1 to 6, wherein the means is further configured to: receive from the alternative apparatus an indication that the alternative apparatus is an alternative to one of the selected at least two further apparatuses, and wherein combining the local training results comprises: The local training results received from the alternative device and the local training results received from at least one additional device of the at least two additional devices are combined.

8. The device according to any one of claims 1 to 7, wherein the component is further used for: receiving, from the at least one additional device of the at least two additional devices, an indication that the at least one additional device of the at least two additional devices is unable to train the local model; and A request is sent to the substitute device to cause the substitute device to perform local model training.

9. The apparatus of claim 8, wherein the request comprises at least one of: an indicator of a reason why the at least one further device of the at least two further devices was unable to train the local model; and A time indicator indicating the time at which the alternative device will perform local model training.

10. The device according to any one of claims 1 to 9, wherein the device is one of the following: A base station of a radio access network, wherein the at least two further devices and the alternative device are user equipment; a network data analysis entity of the communication system, wherein the at least two further devices and the replacement device are distributed network data analysis entities of the communication system; and An operation, administration and maintenance entity of the communication system, wherein the at least two further devices and the replacement device are base stations.

11. An apparatus configured to train a local model during federated learning, the apparatus comprising means for: Receiving an alternative training UE configuration from a further apparatus, the further apparatus being configured to train a local model in a communication network including the apparatus, the alternative training UE configuration comprising: a device identifier configured to uniquely identify the device used to train the local model; as well as a second device identifier configured to uniquely identify an alternative device to the device; as well as a condition identifier configured to identify a trigger condition under which the at least one other device of the at least two other devices is unable to train the local model, and the trigger condition causes the replacement device to perform local model training; as well as Train the local model and send the local training result to the other device, or determine that the device cannot train the local model based on the trigger condition, and send a local model training request to the other device or one of the alternative devices, so that the other device or one of the alternative devices performs local model training.

12. The apparatus of claim 11, wherein the component is further configured to generate information indicating at least one candidate replacement apparatus based on information indicating at least one of the following: data distribution of the data in a local data set at the device, and data distribution of the data in a local data set at the candidate replacement device; data distribution of local data of the device and the candidate replacement device; the extent of the data in local data sets at the device and the candidate replacement device; an interquartile range of the data in the local data sets at the device and the candidate replacement device; a standard deviation of the data in local data sets at the device and the candidate replacement device; a variance of the data in local data sets at the device and the candidate replacement device; the proximity between the device and the candidate replacement device; as well as A mobility pattern between the device and the candidate replacement device.

13. The apparatus of claim 12, wherein the means is further configured to receive a request from the further apparatus to generate the information indicative of the at least one candidate replacement apparatus.

14. An apparatus according to claim 12, wherein the trigger condition comprises at least one of the following items: a minimum quality of a wireless link between the apparatus and a base station of a radio access network; a minimum computing resource availability at the apparatus; a minimum power resource availability at the apparatus; a minimum security level associated with a local data set of the other apparatus, and a minimum integrity level associated with a local data set of the other apparatus.

15. The apparatus according to any one of claims 11 to 14, wherein the local model training request comprises: an indicator of why the device was unable to train the local model; as well as A time indicator indicating the time at which the alternative device will perform local model training.

16. An apparatus according to any one of claims 11 to 15, wherein the component is further used for one of the following items: generating an acceptance message when the alternative training UE configuration is acceptable to the apparatus, and generating a rejection message when the alternative training UE configuration is unacceptable to the apparatus, and the component is further used for: sending one of the acceptance message and the rejection message to the other apparatus so that the other apparatus reselects or reconfigures the alternative training UE configuration.

17. The apparatus according to any of claims 11 to 16, wherein the apparatus is a first user equipment, wherein the alternative apparatus is a second user equipment, and the further apparatus is a base station of a radio access network.

18. The apparatus according to any one of claims 11 to 16, wherein the apparatus is a distributed network data analysis entity, wherein the further apparatus is a centralized network data analysis entity, and the alternative apparatus is a distributed network data analysis entity.

19. The apparatus according to any of claims 11 to 16, wherein the apparatus is a first base station of a radio access network, wherein the further apparatus is an operations, administration and maintenance entity, and the replacement apparatus is a second base station of the radio access network.

20. An apparatus configured to train a local model during federated learning, the apparatus comprising means for: Receiving an alternative training UE configuration from a further apparatus, the further apparatus being configured to train a local model in a communication network including the apparatus, the alternative configuration comprising: a device identifier configured to uniquely identify another device as a device for training the local model using the local dataset; an alternative device identifier configured to uniquely identify the device as an alternative training device for the another device; as well as a condition identifier configured to identify a trigger condition under which the other device is unable to train the local model, and the trigger condition causes the device to train the local model at the device; as well as When the other device is unable to train the local model, receiving a local model training request to perform local model training from the other device or the additional device; In response to receiving the local model training request, a local model is trained using a local data set to generate a local training result, and the local training result is sent to the other device.

21. The apparatus of claim 20, wherein the local model training request comprises at least one of the following: an indicator identifying the condition that renders the other device unable to train the local model; and A time indicator indicating the time when the device will perform local model training.

22. An apparatus according to claim 20 or 21, wherein the component is further used for one of the following items: generating an acceptance message when the alternative training UE configuration is acceptable to the apparatus, and generating a rejection message when the alternative training UE configuration is unacceptable to the apparatus, and the component is further used for: sending one of the acceptance message and the rejection message to the other apparatus so that the other apparatus reselects or reconfigures the alternative training UE configuration.

23. The device according to any one of claims 20 to 22, wherein the means is further for transmitting an indication that the device is the alternative training device for the other device.

24. The apparatus according to any one of claims 20 to 22, wherein the apparatus is a first user equipment, wherein the other apparatus is a second user equipment, and the further apparatus is a base station of a radio access network.

25. An apparatus according to any one of claims 20 to 22, wherein the apparatus is a distributed network data analysis entity, wherein the further apparatus is a centralized network data analysis entity and the other apparatus is a distributed network data analysis entity.

26. The apparatus according to any of claims 20 to 22, wherein the apparatus may be a first base station of a radio access network, wherein the further apparatus may be an operations, administration and maintenance entity, and the other apparatus is a second base station of the radio access network.

27. A method for an apparatus configured to train a model in a communication network using federated learning, the method comprising: Select at least two additional devices to train the local model; selecting an alternative device for the selected at least one of the at least two further devices; as well as configuring each of the at least two further devices to train the local model, and configuring the selected alternative device of the at least one further device of the two further devices to train the local model; receiving local training results from at least one further device of the at least two further devices, and receiving local training results from the selected replacement device for the at least one further device of the two further devices; as well as The local training results are combined to generate an aggregated training result of the model.

28. A method for an apparatus configured to train a local model during federated learning, the method comprising: receiving an alternative training configuration from another apparatus, the other apparatus being configured to train a local model in a communication network including the apparatus, the alternative configuration comprising: a device identifier configured to uniquely identify the device used to train the local model; an alternative device identifier configured to uniquely identify an alternative device to the device; and a condition identifier configured to identify a condition in which the device is unable to train the local model and cause the substitute device to train the local model at the substitute device; and Train the local model and send the local training result to the other device, or determine that the device cannot train the local model based on the condition in which the device cannot train the local model, and send a local model training request to the other device or one of the alternative devices, so that the alternative device uses the local data set to perform local model training at the alternative device.

29. A method for an apparatus configured to train a local model for federated learning, the method comprising: Receiving an alternative training UE configuration from a further apparatus, the further apparatus being configured to train a local model in a communication network including the apparatus, the alternative configuration comprising: a device identifier configured to uniquely identify another device as a device for training the local model using the local dataset; an alternative device identifier configured to uniquely identify the device as an alternative training device for the other device; and a condition identifier configured to identify a condition in which the other device is unable to train the local model and to cause the device to train the local model at the device; and When the other device is unable to train the local model, a training request for training the local model is received from the other device or the other device to train the local model; the local model is trained using a local data set, and after receiving the request, the training result is sent to the other device.

Images