Federated multi-task learning in network automation

Abstract

A method, apparatus and computer program product are provided for performing federated multi-task learning using a federated learning aggregation environment and federated learning distributed nodes (FLDNs). In the context of a method, the method includes determining respective FLDN sets from a plurality of FLDNs for training corresponding personalized global machine learning (ML) models to perform different personalized tasks, where the different personalized tasks are determined based on one or more ML use cases. The method also includes generating at least a first personalized global ML model for performing a first personalized task and a second personalized global ML model for performing a second personalized task. The method also includes performing a respective plurality of training iterations for training the first and second global personalized ML models using the corresponding FLDN sets and by aggregating local model parameters of local personalized ML models independently generated by the respective FLDNs.

Description

FEDERATED MULTI-TASK LEARNING IN NETWORK AUTOMATIONTECHNICAL FIELD

[0001] An example embodiment relates generally to federated multi-task learning (FMTL) techniques for use in managing network automation, covering both performance and trustworthiness aspects.BACKGROUND

[0002] Federated techniques for machine learning include training different versions of a machine learning model different distributed nodes. However, legacy approaches to federated learning may result in poor model performance trustworthiness due to heterogeneity of the different distributed nodes and local training data involved in training the various model versions. For example, a first model version trained by a first distributed node may demonstrate adequate performance and a second model version trained by a second distributed node may demonstrate poor performance. A typical federated machine learning approach includes averaging the first and second versions of the model, which may result in a third version of the model having less than adequate performance as compared to the first model version. As indicated, current federated learning approaches demonstrate a potential deficiency in that a distributed node may receive a final, aggregated machine learning model that demonstrates worse performance in comparison to a precursor version of the machine learning model that was trained locally and independently by the distributed node.BRIEF SUMMARY

[0003] A method, apparatus and computer program product are provided in accordance with an example embodiment in order to improve federated multi-task learning (FMTL) in network automation by generating personalized global machine learning models targeting different subsets of a plurality of federated learning distributed nodes (FLDNs), where FLDNs in each subset are associated with similar tasks, goals, and / or characteristics.

[0004] Typical machine learning (ML) approaches may require centralizing of all the training data that are collected by distributed nodes on one single central node for model training. To minimize the data exchange between the distributed nodes and the central node where model training is performed, the present techniques perform federated learning (FL). For example, instead of training a model at the central node as in previous approaches, various embodiments of the present techniques may iteratively train different versions of a model at different FLDNs, where each FLDN may include a different set of local training data. In some embodiments, during each training iteration, a central node (referred to herein as the federated learning aggregation (FLA) computing environment) aggregates the different local versions of the model to generate a global version of the model, which may be distributed to the FLDNs for further training iterations until model convergence is achieved.

[0005] In some embodiments, the present techniques perform FMTL including training a shared global machine learning model using groups of FLDNs that are associated with different tasks but cooperatively train different versions of a machine learning model to generate, at the FLA, a respective aggregated global machine learning model for performing each task. For example, a first group of FLDNs may include one or more first training datasets associated with a first source domain and a first target domain, and a second group of FLDNs may include one or more second training datasets associated with a second source domain and a second target domain (e.g., data of the first and second training datasets are heterogenous). In some contexts, a first version of a machine learning model may be provided to the first and second groups of FLDNs and trained using the respective first or second training dataset. In some contexts, global version of the machine learning model may be generated based on a respective trained iteration of the first version of the machine learning model from the first and second groups of FLDNs. However, the global version of the machine learning model may demonstrate low performance (e.g., low statistical accuracy and / or the like) and / or low trustworthiness when applied to the first target domain, the second target domain, or both. Various embodiments of the present techniques may overcome such challenge by providing different group of FLDNs with a personalized version of a machine learning model for training, where each personalized version of the machine learning model may be generated to optimally fit a local data distribution of the corresponding FLDNs.

[0006] In various embodiments, the present apparatus, method, and computer program product generate machine learning models that demonstrate and / or enable various advantages for providingmodel trustworthiness. For example, the machine learning models generated by the present techniques may demonstrate transparency, diversity, non-discrimination and fairness, and technical robustness and safety. As another example, the machine learning models may enable or promote adequate privacy and data governance, accountability (e.g., auditability, minimization and reporting of negative impact, trade-offs, and redress), human agency and oversight, and societal and environmental wellbeing.

[0007] In at least one embodiment, a method is provided that includes (i) determining (503), for at least a first personalized task and a second personalized task of a plurality of personalized tasks, (1) a first set of FLDNs (105) from a plurality of federated learning distributed nodes (FLDNs (105) associated with the first personalized task and (2) a second set of FLDNs (105) from the plurality of FLDNs (105) associated with the second personalized task, (ii) generating (518) a first personalized machine learning (ML) model for performing the first personalized task and a second personalized ML model for performing the second personalized task, and (iii) performing (521) a respective plurality of model training iterations for training the first personalized ML model and the second personalized ML model, the respective plurality of model training iterations comprising (1) providing (521), in a respective initial model training iteration for the first personalized ML model and the second personalized ML model, (A) the first personalized ML model to respective FLDNs (105) of the first set and (B) the second personalized ML model to respective FLDNs (105) of the second set, (2) providing (521), in respective subsequent model training iterations for the first personalized ML model and the second personalized ML model, (A) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the first set and (B) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the second set, where, in a respective model training iteration, respective FLDNs (105) of the first set and respective FLDNs (105) of the second set independently train the corresponding personalized ML model to perform the corresponding personalized task, (3) receiving (524), in a respective model training iteration, local model parameters from the respective FLDNs (105), wherein the local model parameters are associated with the training of the corresponding personalized ML model by the respective FLDNs (105) of the first set or the second set, and (4) generating (530) (A) the subsequent trained iteration of the first personalized ML model based on an aggregation of the local model parameters associated with the respective FLDNs (105) of the first set; and (B) the subsequent trained iteration of the second personalized ML model based on an aggregation ofthe local model parameters associated with respective FLDNs (105) of the second set, where an output of a final iteration of each of the respective plurality of training iterations comprises a first personalized global ML model for performing the first personalized task and a second personalized global ML model for performing the second personalized task.

[0008] In some embodiments, the plurality of FLDNs (105) comprise one or more individual management domains and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a cross-domain management domain (103).

[0009] In some embodiments, the plurality of FLDNs (105) comprise one or more network nodes and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a radio access network (RAN) management domain (103). In some embodiments, the plurality of FLDNs (105) comprise one or more network data analytic functions (NWDAFs) and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a core management domain (103). In some embodiments, the plurality of FLDNs (105) comprise one or more distributed NWDAFs and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a central NWDAF (103).

[0010] In some embodiments, the plurality of FLDNs (105) comprise one or more user equipment (UE) and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a network node (103).

[0011] In some embodiments, the plurality of FLDNs (105) comprise one or more real-time RAN intelligent controllers (RICs) and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a non- real-time RIC (103).

[0012] In some embodiments, the plurality of FLDNs (105) comprise one or more network nodes and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a real-time RIC (103). In some embodiments, the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a first UE (103), and the plurality of FLDNs (105) comprise one or more additional UEs.

[0013] In some embodiments, the method further includes, in a respective model training iteration, (i) receiving (527), from respective FLDNs (105) of the first set and respective FLDNs (105) of the second set, at least one training score associated with training of the corresponding personalized ML model by the FLDNs (105) and (ii) generating (530) a respective aggregated training score for the first personalized ML model and the second personalized ML model based on the training scores received from the corresponding FLDNs (105). In some embodiments, the method further includes providing (536) the at least one aggregated training score associated with the first personalized ML model and the at least one aggregated score associated with the second personalized ML model to a computing device associated with an operator entity.

[0014] In some embodiments, (i) a respective training score of the at least one training score includes a performance score for performing the corresponding personalized task and (ii) a respective training score of the at least one aggregated training score includes an aggregated performance score for performing the corresponding personalized task. In some embodiments, (i) the respective training score further includes a trustworthiness score for performing the corresponding personalized task and (ii) the respective aggregated training score further includes an aggregated trustworthiness score for performing the corresponding personalized task. In some embodiments, the method further includes determining (506, 512) the plurality of personalized tasks based on a user input received from the computing device associated with the operator entity, wherein the user input indicates at least one machine learning model use case.

[0015] In some embodiments, the method further includes receiving (506), from a data store, a plurality of historical personalized tasks based on the at least one machine learning model use case, wherein the plurality of personalized tasks are determined based on the plurality of historical personalized tasks. In some embodiments, the method further includes (i) generating (515) amapping of the at least one machine learning use case to subsets of the plurality of historical personalized tasks and (ii) generating (515), based on the mapping, at least (1) a first category comprising the first subset of the plurality of FLDNs (105) and (2) a second category comprising the second subset of the plurality of FLDNs (105), wherein the plurality of personalized tasks are further determined based on the first category and the second category.

[0016] In some embodiments, the method further includes (i) providing (509) the plurality of historical personalized tasks to the computing device associated with the operator entity, (ii) receiving (512) a second user input from the computing device associated with the operator entity, wherein the second user input indicates at least one of a filtered subset of the plurality of historical personalized tasks or at least one additional personalized task, and (iii) determining (506) the plurality of personalized tasks further based on at least one of the filtered subset of the plurality of historical personalized tasks or the at least one additional personalized task.

[0017] In some embodiments, the method further includes between one or more model training iterations of the plurality of model training iterations, (i) determining (506) a third personalized task, (ii) determining (515) a third set of the plurality of FLDNs (105) associated with the third personalized task, (iii) generating (518) a third personalized machine learning model to perform the third personalized task, and (iv) providing (521) the third personalized machine learning model or a subsequent trained iteration of the third personalized machine learning model to the third set of FLDNs (105), wherein (1) respective FLDNs (105) of the third set train the third personalized machine learning model or the subsequent trained iteration of the third personalized machine learning model in a remaining subset of the plurality of model iterations and (2) in a final model training iteration, a personalized global machine learning model for the third personalized task is generated based on a corresponding plurality of local model parameters received from respective FLDNs (105) of the third set during the remaining subset of the plurality of model training iterations. In some embodiments, one or more of the first set of the plurality of FLDNs (105), the second set of the plurality of FLDNs (105), or the third set of the plurality of FLDNs (105) are determined based on one or more properties or statuses of the plurality of FLDNs (105).

[0018] In some embodiments, the method further includes (i) removing (506) at least one FLDN from the first set of the plurality of FLDNs (105) based on a change to the one or more properties or statuses of the at least one FLDN and (ii) adding (506) the at least one FLDN to the second set of the plurality of FLDNs (105) or the third set of the plurality of FLDNs (105) based on the changeto the one or more properties or statuses of the at least one FLDN. In some embodiments, the one or more properties or statuses of the plurality of FLDNs (105) includes at least one of a computational resource availability, a power availability, a channel quality, a data quality score, or a data trustworthiness score. In some embodiments, the one or more properties or statuses of the plurality of FLDNs (105) includes at least one of location data, velocity data, a trust score, or a quantity of available training data samples.

[0019] In some embodiments, the method further includes (i) providing (515) a notification indicative of the plurality of personalized tasks to the plurality of FLDNs (105), (ii) receiving (512), from at least one FLDN of the plurality of FLDNs (105), a request to associate the at least one FLDN with the first personalized task or the second personalized task, and (iii) determining (515) at least one of the first set of the plurality of FLDNs (105 )or the second set of the plurality of FLDNs (105) based on the request from the at least one FLDN.

[0020] In some embodiments, the method further includes (i) receiving (512), from at least one FLDN, a request for disassociation from the first set of the plurality of FLDNs (105)or the second set of the plurality of FLDNs (105), (ii) determining (506) that at least one training score associated with the at least one FLDN and a personalized machine learning model trained by the at least one FLDN meets a predetermined threshold, and (iii) in response to the at least one training score meeting the predetermined threshold, providing (506), to the at least one FLDN, a request to retain the at least one FLDN in the corresponding set of the plurality of FLDNs (105) for the plurality of training iterations, wherein the request to retain the at least one FLDN indicates a monetary retention incentive. In some embodiments, (i) the request for disassociation includes at least one disassociation factor and (ii) the at least one disassociation factor is at least one of a computational overload, a bandwidth limitation, or a power limitation.

[0021] In various embodiments, as further described below, one or more operations of the abovedescribed methods are performed by an apparatus having means for performing the one or more operations. For example, an apparatus may include (i) means for determining (503), for at least a first personalized task and a second personalized task of a plurality of personalized tasks, (1) a first set of FLDNs (105) from a plurality of federated learning distributed nodes (FLDNs (105)) associated with the first personalized task and (2) a second set of FLDNs (105) from the plurality of FLDNs (105) associated with the second personalized task, (ii) means for generating (518) a first personalized machine learning (ML) model for performing the first personalized task and asecond personalized ML model for performing the second personalized task, and (iii) means for performing (521) a respective plurality of model training iterations for training the first personalized ML model and the second personalized ML model, the respective plurality of model training iterations comprising (1) means for providing (521), in a respective initial model training iteration for the first personalized ML model and the second personalized ML model, (A) the first personalized ML model to respective FLDNs (105) of the first set and (B) the second personalized ML model to respective FLDNs (105) of the second set, (2) means for providing (521), in respective subsequent model training iterations for the first personalized ML model and the second personalized ML model, (A) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the first set and (B) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the second set, where, in a respective model training iteration, respective FLDNs (105) of the first set and respective FLDNs (105) of the second set independently train the corresponding personalized ML model to perform the corresponding personalized task, (3) means for receiving (524), in a respective model training iteration, local model parameters from the respective FLDNs (105), wherein the local model parameters are associated with the training of the corresponding personalized ML model by the respective FLDNs (105) of the first set or the second set, and (4) means for generating (530) (A) the subsequent trained iteration of the first personalized ML model based on an aggregation of the local model parameters associated with the respective FLDNs (105) of the first set; and (B) the subsequent trained iteration of the second personalized ML model based on an aggregation of the local model parameters associated with respective FLDNs (105) of the second set, where an output of a final iteration of each of the respective plurality of training iterations comprises a first personalized global ML model for performing the first personalized task and a second personalized global ML model for performing the second personalized task. In the same example, the apparatus may embody additional aspects and / or include additional means for performing additional operations of the above-described methods.

[0022] As further described below, in some embodiments, one or more operations of the abovedescribed methods are performed by an apparatus including at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform the one or more operations. For example, an apparatus may include at least one processor and at least one memory storing instructions that, when executed by the at least oneprocessor, cause the apparatus to perform functionality including (i) determining (503), for at least a first personalized task and a second personalized task of a plurality of personalized tasks, (1) a first set of FLDNs (105) from a plurality of federated learning distributed nodes (FLDNs (105) associated with the first personalized task and (2) a second set of FLDNs (105) from the plurality of FLDNs (105) associated with the second personalized task, (ii) generating (518) a first personalized machine learning (ML) model for performing the first personalized task and a second personalized ML model for performing the second personalized task, and (iii) performing (521) a respective plurality of model training iterations for training the first personalized ML model and the second personalized ML model, the respective plurality of model training iterations comprising (1) providing (521), in a respective initial model training iteration for the first personalized ML model and the second personalized ML model, (A) the first personalized ML model to respective FLDNs (105) of the first set and (B) the second personalized ML model to respective FLDNs (105) of the second set, (2) providing (521), in respective subsequent model training iterations for the first personalized ML model and the second personalized ML model, (A) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the first set and (B) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the second set, where, in a respective model training iteration, respective FLDNs (105) of the first set and respective FLDNs (105) of the second set independently train the corresponding personalized ML model to perform the corresponding personalized task, (3) receiving (524), in a respective model training iteration, local model parameters from the respective FLDNs (105), wherein the local model parameters are associated with the training of the corresponding personalized ML model by the respective FLDNs (105) of the first set or the second set, and (4) generating (530) (A) the subsequent trained iteration of the first personalized ML model based on an aggregation of the local model parameters associated with the respective FLDNs (105) of the first set; and (B) the subsequent trained iteration of the second personalized ML model based on an aggregation of the local model parameters associated with respective FLDNs (105) of the second set, where an output of a final iteration of each of the respective plurality of training iterations comprises a first personalized global ML model for performing the first personalized task and a second personalized global ML model for performing the second personalized task. In the same example, the apparatus may also perform other operations and / or embody additional aspects of the above-described methods.

[0023] In various embodiments, as further described below, provided herein is a computer program product including at least one non-transitory computer-readable storage medium having computer-executable program code instructions stored therein, the computer-executable program code instructions including program code instructions configured for performing one or more operations and / or embody additional aspects of the above-described methods. For example, a non- transitory computer readable medium including program instructions stored thereon for performing at least the following (i) determining (503), for at least a first personalized task and a second personalized task of a plurality of personalized tasks, (1) a first set of FLDNs (105) from a plurality of federated learning distributed nodes (FLDNs (105) associated with the first personalized task and (2) a second set of FLDNs (105) from the plurality of FLDNs (105) associated with the second personalized task, (ii) generating (518) a first personalized machine learning (ML) model for performing the first personalized task and a second personalized ML model for performing the second personalized task, and (iii) performing (521) a respective plurality of model training iterations for training the first personalized ML model and the second personalized ML model, the respective plurality of model training iterations comprising (1) providing (521), in a respective initial model training iteration for the first personalized ML model and the second personalized ML model, (A) the first personalized ML model to respective FLDNs (105) of the first set and (B) the second personalized ML model to respective FLDNs (105) of the second set, (2) providing (521), in respective subsequent model training iterations for the first personalized ML model and the second personalized ML model, (A) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the first set and (B) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the second set, where, in a respective model training iteration, respective FLDNs (105) of the first set and respective FLDNs (105) of the second set independently train the corresponding personalized ML model to perform the corresponding personalized task, (3) receiving (524), in a respective model training iteration, local model parameters from the respective FLDNs (105), wherein the local model parameters are associated with the training of the corresponding personalized ML model by the respective FLDNs (105) of the first set or the second set, and (4) generating (530) (A) the subsequent trained iteration of the first personalized ML model based on an aggregation of the local model parameters associated with the respective FLDNs (105) of the first set; and (B) the subsequent trained iteration of the second personalized ML model based on an aggregation of thelocal model parameters associated with respective FLDNs (105) of the second set, where an output of a final iteration of each of the respective plurality of training iterations comprises a first personalized global ML model for performing the first personalized task and a second personalized global ML model for performing the second personalized task.

[0024] In at least one embodiment, a method is provided that includes (i) storing (603) training data in memory (204) accessible to an apparatus (105, 200), (ii) receiving (606), from a federated learning aggregation (FLA) computing environment (101), a request for local environment data associated with the apparatus, (iii) providing (609), to the FLA computing environment (101), local environment data including at least one of a computational resource availability, a power availability, a channel quality, a data quality, or a data trustworthiness, wherein, based on the local environment data, the FLA computing environment (101) associates the apparatus with at least one personalized task of a plurality of personalized tasks, (iv) performing a plurality of model training iterations, wherein the performing a respective model training iteration of the plurality of model training iterations include: (1) receiving (621), from the FLA computing environment (101), a personalized machine learning model for performing the at least one personalized task or a subsequent trained iteration of the personalized machine learning model (2) training (624) the personalized machine learning model or a subsequent iteration of the personalized machine learning model using the training data to generate a first plurality of local model parameters, and (3) providing (627) the local model parameters to the FLA computing environment (101), wherein the FLA computing environment (101) generates the subsequent trained iteration of the personalized machine learning model based on an aggregation of the first plurality of local model parameters with a second plurality of local model parameters received by the FLA computing environment (101) from a second apparatus having training the personalized machine learning model or a subsequent trained iteration of the personalized machine learning model, (v) generating (630), following at least a subset of the plurality of model training iterations, at least one training score for the personalized machine learning model or a subsequent iteration of the personalized machine learning model, and (vi) providing (633) the at least one training score to the FLA computing environment (101).

[0025] In some embodiments, the FLA computing environment (101) includes a cross-domain management domain and the apparatus (105) performing the method includes one or more individual management domains. In some embodiments, the FLA computing environment (101)includes a radio access network (RAN) management domain and the apparatus performing the method includes a network node. In some embodiments, the FLA computing environment (101) includes a core management domain and the apparatus (105) performing the method includes a network data analytic function (NWDAF). In some embodiments, the FLA computing environment (101) includes a central NWDAF and the apparatus (105) performing the method includes a distributed NWDAF. In some embodiments, the FLA computing environment (101) includes a network node and the apparatus (105) performing the includes a user equipment (UE). In some embodiments, the FLA computing environment (101) includes a non-real-time RAN intelligent controller (RIC) and the apparatus (105) performing the method includes a real-time RIC. In some embodiments, the FLA computing environment (101) includes a real-time RIC and the apparatus performing the steps of the method includes a network node. In some embodiments, the FLA computing environment (101) includes a first UE; and the apparatus (105) performing the method includes a second UE.

[0026] In some embodiments, the at least one training score includes a performance score for performing the at least one personalized task. In some embodiments, the at least one training score further includes a trustworthiness score for performing the at least one personalized task.

[0027] In some embodiments, the method further includes (i) receiving (606) at least one local environment data threshold, (ii) determining (609) that a change to the local environment data meets the at least one local environment data threshold, and (iii) providing (609), to the FLA computing environment (101), a notification indicative of the change to the local environment data, wherein the notification causes the FLA computing environment (101) to remove the association between the apparatus and the at least one personalized task. In some embodiments, the method further includes (i) providing (609), to the FLA computing environment (101), at least one property or status of the apparatus, wherein (1) the at least one property or status of the apparatus includes at least one of location data indicative of a location of the apparatus, a velocity of the apparatus, a trust score of the apparatus, or a quantity of training data samples of the apparatus, and (2) the FLA computing environment (101) associates the apparatus with the at least one personalized task further based on the at least one property or status of the apparatus, (ii) receiving (606) at least one property or status threshold from the FLA computing environment (101), (iii) determining (609) that a change to the at least one property or status of the apparatus meets the at least one property or status threshold, and (iv) providing (609), to the FLA computingenvironment (101), a notification indicative of the change to the change to the at least one property or status of the apparatus, wherein the notification causes the FLA computing environment (101) to remove the association between the apparatus and the at least one personalized task. In some embodiments, the at least one property or status threshold includes a predetermined proximity to a geozone and the change to the at least one property or status of the apparatus indicates the location of the apparatus is beyond the predetermined proximity of the geozone. In some embodiments, the at least one property or status threshold is a velocity range associated with the at least one task and the change to the at least one property or status of the apparatus indicates the velocity of the apparatus is outside the velocity range.

[0028] In some embodiments, at least one of (i) the notification indicative of the change to the local environment data or (ii) the notification indicative of the change to the at least one property or status of the apparatus further causes the FLA computing environment (101) to associate the apparatus with an additional personalized task of the plurality of tasks. In some embodiments, the method further includes (i) receiving (621), from the FLA computing environment (101), a second personalized machine learning model for performing the additional personalized task or a subsequent trained iteration of the second personalized machine learning model, (ii) training (624) the second personalized machine learning model or a subsequent iteration of the second personalized machine learning model using the training data to generate an additional plurality of local model parameters, and (iii) providing (627) the additional plurality of local model parameters to the FLA computing environment (101).

[0029] In some embodiments, the training data includes a plurality of local training samples. In some embodiments, the method further includes (i) generating (624) synthetic data and (ii) generating (624) an additional plurality of local training samples based on the synthetic data, wherein the training data further includes the additional plurality of local training samples. In some embodiments, the method further includes indicating (609) a quantity of local training samples in the training data to the FLA computing environment (101), wherein the FLA computing environment (101) associates the apparatus (105) with the at least one personalized task of the plurality of personalized tasks further based on the quantity of local training samples in the training data.

[0030] In some embodiments, the method further includes (i) receiving (612) a notification indicative of the plurality of personalized tasks from the FLA computing environment (101) and(ii) providing (615), to the FLA computing environment (101), a request to associate the apparatus with the at least one personalized task, wherein the FLA computing environment (101) associates the apparatus (105) with the at least one personalized task based at least in part on the request. In some embodiments, the method further includes (i) providing (618), to the FLA computing environment (101), a request for disassociation from the at least one personalized task and (ii) receiving (618), from the FLA computing environment (101), a request to remain associated with the at least one personalized task and continue performing the plurality of training iterations, wherein (1) the request to retain the at least one FLDN indicates at least one retention incentive and (2) the FLA computing environment (101) provides the request to the apparatus in response to the at least one training score meeting a predetermined threshold. In some embodiments, the request for disassociation includes at least one disassociation factor. In some embodiments, the at least one disassociation factor is at least one of a computational overload, a bandwidth limitation, or a power limitation. In some embodiments, the request for disassociation further includes a duration of unavailability of the apparatus.

[0031] In various embodiments, as further described below, one or more operations of the abovedescribed methods are performed by an apparatus having means for performing the one or more operations. For example, an apparatus may include (i) means for storing (603) training data in memory (204) accessible to an apparatus (105, 200), (ii) means for receiving (606), from a federated learning aggregation (FLA) computing environment (101), a request for local environment data associated with the apparatus, (iii) means for providing (609), to the FLA computing environment (101), local environment data including at least one of a computational resource availability, a power availability, a channel quality, a data quality, or a data trustworthiness, wherein, based on the local environment data, the FLA computing environment (101) associates the apparatus with at least one personalized task of a plurality of personalized tasks, (iv) means for performing a plurality of model training iterations, wherein the means for performing a respective model training iteration of the plurality of model training iterations include: (1) means for receiving (621), from the FLA computing environment (101), a personalized machine learning model for performing the at least one personalized task or a subsequent trained iteration of the personalized machine learning model (2) means for training (624) the personalized machine learning model or a subsequent iteration of the personalized machine learning model using the training data to generate a first plurality of local model parameters, and (3) means forproviding (627) the local model parameters to the FLA computing environment (101), wherein the FLA computing environment (101) generates the subsequent trained iteration of the personalized machine learning model based on an aggregation of the first plurality of local model parameters with a second plurality of local model parameters received by the FLA computing environment (101) from a second apparatus having training the personalized machine learning model or a subsequent trained iteration of the personalized machine learning model, (v) means for generating (630), following at least a subset of the plurality of model training iterations, at least one training score for the personalized machine learning model or a subsequent iteration of the personalized machine learning model, and (vi) means for providing (633) the at least one training score to the FLA computing environment (101). In the same example, the apparatus may embody additional aspects and / or include additional means for performing additional operations of the abovedescribed methods.

[0032] As further described below, in some embodiments, one or more operations of the abovedescribed methods are performed by an apparatus including at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform the one or more operations. For example, an apparatus may include at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform functionality including (i) storing (603) training data in memory (204) accessible to an apparatus (105, 200), (ii) receiving (606), from a federated learning aggregation (FLA) computing environment (101), a request for local environment data associated with the apparatus, (iii) providing (609), to the FLA computing environment (101), local environment data including at least one of a computational resource availability, a power availability, a channel quality, a data quality, or a data trustworthiness, wherein, based on the local environment data, the FLA computing environment (101) associates the apparatus with at least one personalized task of a plurality of personalized tasks, (iv) performing a plurality of model training iterations, wherein the performing a respective model training iteration of the plurality of model training iterations include: (1) receiving (621), from the FLA computing environment (101), a personalized machine learning model for performing the at least one personalized task or a subsequent trained iteration of the personalized machine learning model (2) training (624) the personalized machine learning model or a subsequent iteration of the personalized machine learning model using the training data to generate a first plurality of local model parameters, and(3) providing (627) the local model parameters to the FLA computing environment (101), wherein the FLA computing environment (101) generates the subsequent trained iteration of the personalized machine learning model based on an aggregation of the first plurality of local model parameters with a second plurality of local model parameters received by the FLA computing environment (101) from a second apparatus having training the personalized machine learning model or a subsequent trained iteration of the personalized machine learning model, (v) generating (630), following at least a subset of the plurality of model training iterations, at least one training score for the personalized machine learning model or a subsequent iteration of the personalized machine learning model, and (vi) providing (633) the at least one training score to the FLA computing environment (101). In the same example, the apparatus may also perform other operations and / or embody additional aspects of the above-described methods.

[0033] In various embodiments, as further described below, provided herein is a computer program product including at least one non-transitory computer-readable storage medium having computer-executable program code instructions stored therein, the computer-executable program code instructions including program code instructions configured for performing one or more operations and / or embody additional aspects of the above-described methods. For example, a non- transitory computer readable medium including program instructions stored thereon for performing at least the following (i) storing (603) training data in memory (204) accessible to an apparatus (105, 200), (ii) receiving (606), from a federated learning aggregation (FLA) computing environment (101), a request for local environment data associated with the apparatus, (iii) providing (609), to the FLA computing environment (101), local environment data including at least one of a computational resource availability, a power availability, a channel quality, a data quality, or a data trustworthiness, wherein, based on the local environment data, the FLA computing environment (101) associates the apparatus with at least one personalized task of a plurality of personalized tasks, (iv) performing a plurality of model training iterations, wherein the performing a respective model training iteration of the plurality of model training iterations include: (1) receiving (621), from the FLA computing environment (101), a personalized machine learning model for performing the at least one personalized task or a subsequent trained iteration of the personalized machine learning model (2) training (624) the personalized machine learning model or a subsequent iteration of the personalized machine learning model using the training data to generate a first plurality of local model parameters, and (3) providing (627) the local modelparameters to the FLA computing environment (101), wherein the FLA computing environment (101) generates the subsequent trained iteration of the personalized machine learning model based on an aggregation of the first plurality of local model parameters with a second plurality of local model parameters received by the FLA computing environment (101) from a second apparatus having training the personalized machine learning model or a subsequent trained iteration of the personalized machine learning model, (v) generating (630), following at least a subset of the plurality of model training iterations, at least one training score for the personalized machine learning model or a subsequent iteration of the personalized machine learning model, and (vi) providing (633) the at least one training score to the FLA computing environment (101). In the same example, the program instructions may also be configured to perform additional operations and / or embody additional aspects of the above-described methods.BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Having thus described certain example embodiments of the present disclosure in general terms, reference will hereinafter be made to the accompanying drawings, which are not necessarily drawn to scale, and wherein:

[0035] FIG. 1 illustrates an example of a communication system in which an example embodiment of the present disclosure may be implemented;

[0036] FIG. 2 illustrates a block diagram of an apparatus that may be configured in accordance with an example embodiment of the present disclosure;

[0037] FIGS. 3A-3B illustrate a cross-functional diagram for federated multi-task learning (FMTL) in accordance with an example embodiment of the present disclosure;

[0038] FIG. 4 is an example schema for management of FMTL in accordance with an example embodiment of the present disclosure;

[0039] FIG. 5 is an example flowchart of an FMTL training process from a federated learning aggregation (FLA) computing environment perspective in accordance with an example embodiment of the present disclosure; and

[0040] FIG. 6 is an example flowchart of an FMTL training process from a federated learning distributed node (FLDN) perspective in accordance with an example embodiment of the present disclosure.DETAILED DESCRIPTION

[0041] Some embodiments will now be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all, embodiments are shown. Indeed, various embodiments may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. Like reference numerals refer to like elements throughout. As used herein, the terms “data,” “content,” “information,” and similar terms may be used interchangeably to refer to data capable of being transmitted, received and / or stored in accordance with the described embodiments. Thus, use of any such terms should not be taken to limit the spirit and scope of the embodiments.

[0042] Additionally, as used herein, the term ‘circuitry’ refers to (a) hardware-only circuit implementations (e.g., implementations in analog circuitry and / or digital circuitry); (b) combinations of circuits and computer program product(s) comprising software and / or firmware instructions stored on one or more computer readable memories that work together to cause an apparatus to perform one or more functions described herein; and (c) circuits, such as, for example, a microprocessor(s) or a portion of a microprocessor(s), that require software or firmware for operation even if the software or firmware is not physically present. This definition of ‘circuitry’ applies to all uses of this term herein, including in any claims. As a further example, as used herein, the term ‘circuitry’ also includes an implementation comprising one or more processors and / or portion(s) thereof and accompanying software and / or firmware. As another example, the term ‘circuitry’ as used herein also includes, for example, a baseband integrated circuit or applications processor integrated circuit for a mobile phone or a similar integrated circuit in a server, a cellular network device, other network device (such as a core network apparatus), field programmable gate array, and / or other computing device.

[0043] The term “comprising” means including but not limited to and should be interpreted in the manner it is typically used in the patent context. Use of broader terms such as comprises, includes, and having should be understood to provide support for narrower terms such as consisting of, consisting essentially of, and comprised substantially of. Furthermore, to the extent that the terms “includes” and “including,” and variants thereof are used in either the detailed description or the claims, these terms are intended to be inclusive in a manner similar to the term “comprising.”

[0044] The phrases “in one embodiment,” “according to one embodiment,” “in some embodiments,” “in various embodiments”, and the like generally refer to the fact that the particular feature, structure, or characteristic following the phrase may be included in at least one embodiment of the present disclosure, but not necessarily all embodiments of the present disclosure. Thus, the particular feature, structure, or characteristic may be included in more than one embodiment of the present disclosure such that these phrases do not necessarily refer to the same embodiment.

[0045] As used herein, the terms “example,” “exemplary,” and the like are used to mean “serving as an example, instance, or illustration.” Any implementation, aspect, or design described herein as “example” or “exemplary” is not necessarily to be construed as preferred or advantageous over other implementations, aspects, or designs. Rather, use of the terms “example,” “exemplary,” and the like are intended to present concepts in a concrete fashion.

[0046] If the specification states a component or feature “may,” “can,” “could,” “should,” “would,” “preferably,” “possibly,” “typically,” “optionally,” “for example,” “often,” or “might” (or other such language) be included or have a characteristic, that particular component or feature is not required to be included or to have the characteristic. Such component or feature may be optionally included in some embodiments, or it may be excluded.

[0047] As used herein, the term “computer-readable medium” refers to signal, non-transitory computer-readable medium and the like. The term ‘non-transitory computer-readable medium’ refers to non-transitory storage hardware, non-transitory storage device or non-transitory computer system memory that may be accessed by a controller, a microcontroller, a computational system or a module of a computational system to encode thereon computer-executable instructions or software programs. A non-transitory “computer-readable medium” may be accessed by a computational system or a module of a computational system to retrieve and / or execute the computer-executable instructions or software programs encoded on the medium. Examples of non- transitory computer-readable media may include, but are not limited to, one or more types of hardware memory, non-transitory tangible media (for example, one or more magnetic storage disks, one or more optical disks, one or more USB flash drives), computer system memory or random-access memory (such as, DRAM, SRAM, EDO RAM), and the like.

[0048] As used herein, the term “personalized task” refers to an action of generating or implementing a machine learning model architecture that satisfies a use case for networkautomation respective to different local data sets or groupings of local data sets. For example, a personalized task may include generating a machine learning model architecture for performing network automation operations at one or more FLDNs such that the machine learning model architecture limits a number of radio link failures (RLFs) experienced by network services and devices to a threshold limit (e.g., where different personalized tasks for that use case may be associated with different values for that threshold limit).

[0049] As another example, personalized tasks may include generating machine learning model architectures for FLDNs associated with varying levels of data heterogeneity. A use case may include a scenario in which different groupings of FLDNs include different data set sizes (e.g., due to differences in data sampling rates). A personalized task may be defined to perform network automation services for each grouping of FLDNs. The present techniques for FMTL training may be performed to generate a respective personalized global machine learning model for performing network automation in each grouping of FLDNs.

[0050] As another example, personalized tasks may include generating machine learning model architectures for FLDNs associated with varying levels of system heterogeneity. A use case may include a scenario in which different groupings of FLDNs are associated with different levels of computational capacity and / or communication capacity. The present techniques for FMTL training may be performed to generate a respective personalized global machine learning model for performing network automation services for each grouping of FLDNs. An operating entity may provide to a gNodeB (e.g., embodying a FLA aggregation environment as described herein) analytics indicative of possible computational and communication capabilities for UEs in a cell coverage area. The gNodeB may perform the present FMTL training techniques to determine different categories of UEs based on computational capacity and / or communication capacity. The gNodeB may generate a respective personalized global machine learning model for performing personalized network automation operations for each category of UEs.

[0051] As another example, personalized tasks may include generating machine learning model architectures for FLDNs having varying levels of objective heterogeneity. A use case may include a scenario in which different groupings of FLDNs utilize machine learning model architectures to support different objectives. For example, a first grouping of FLDNs may utilize federated learning machine learning models for a mobility use case and a second grouping of FLDNs may utilize FL machine learning models for supporting energy efficiency. The present techniques for FMTLtraining may be performed to generate a first personalized global machine learning model for fulfilling the objective of the first grouping of FLDNs and a second personalized global machine learning model for fulfilling the objective of the second grouping of FLDNs.

[0052] As used herein, the term “personalized machine learning model” (including personalized global machine learning model and personalized local machine learning model) refers to a machine learning model configured to perform one or more tasks within a network to support network automation. In some embodiments, satisfaction of the personalized task is based on (i) model performance (e.g., statistical accuracy for performing the personalized task), and / or (ii) model trustworthiness (e.g., the extent to which the machine learning model architecture may be generalized to varying network elements associated with different properties, statuses, explainability requirements, and / or the like).

[0053] As used herein, the term “subsequent trained iteration” refers to a version of a personalized machine learning model that was generated based on local model parameters obtained in a least one iteration of FMTL training by a set of federated learning distributed nodes (FLDNs). For example, a subsequent iteration of a personalized global machine learning model may be a personalized global machine learning model generated based on a plurality of personalized local machine learning models (e.g., defined by local model parameters) generated by respective FLDNs, where the plurality of personalized local machine learning models were generated by the respective FLDNs training a prior iteration of the personalized global machine learning model using respective local training data.Overview

[0054] In federated learning (FL), an objective includes generating a single global machine learning model capable of achieving “average” performance and trustworthiness across all FL distributed nodes (FLDNs) participating in the FL training process. Different FLDNs, or groups of FLDNs, may utilize different sets of local training data that are non-IID (independent and identically distributed). Additionally, the different sets of local training data may demonstrate further heterogeneity due to missing, biased, or noisy features or labels in the local training data. Further, different FLDNs may demonstrate varying computational, storage, bandwidth, and / or power capabilities for performing model training. As a result of the heterogeneity of FLDNs and local training data, the performance and / or trustworthiness of the trained global machine learning model may be adequate or ideal in a first set of participating FLDNs while being inadequate orpoor in a second set of participating FLDNs. For example, scenarios may occur in which, for a subset of participating FLDNs, the trained global machine learning model demonstrates worse performance and trustworthiness as compared to a model trained locally and independently by the subset of participating FLDNs. In various embodiments, the present apparatus, method, and computer program product overcome such challenges using federated multi-task learning (FMTL) techniques that include determining personalized tasks for a machine learning model and intended use case, determining different subsets of FLDNs associated with the different personalized tasks, and generating personalized global machine learning models that each target a corresponding subset of the participating FLDNs (e.g., where FLDNs in the subset may demonstrate similar tasks, objectives, and / or characteristics).

[0055] In some embodiments, an apparatus for performing FMTL training includes a federated learning aggregation (FLA) computing environment and a plurality of FLDNs. In some embodiments, the FLA computing environment includes means for determining possible or necessary personalized tasks for a machine learning model and one or more use cases of the machine learning model in one more iterations of FMTL training. In some embodiments, the FLA computing environment includes means for determining associations between personalized tasks and candidate FLDNs in one or more iterations of the FMTL training. For example, the FLA computing environment may include means for generating a mapping between one or more FLDNs of a plurality of candidate FLDNs and one or more personalized tasks determined for a machine learning model.

[0056] In some embodiments, the FLA computing environment includes means for generating personalized machine learning models for performing personalized tasks (e.g., and which may be further trained by FLDNs to perform the personalized task) and means for providing personalized machine learning models to corresponding FLDNs for training. In some embodiments, the means for generating a personalized machine learning model includes means for generating a subsequent iteration of the personalized machine learning model based on local model parameters received from and generated by FLDNs during training of the machine learning model at each FLDN. For example, the FLA computing environment may aggregate received local model parameters to generate the subsequent iteration of the personalized machine learning model (also referred to herein as a personalized global machine learning model).

[0057] In some embodiments, the FLA computing environment includes means for determining the performance and / or trustworthiness of the personalized machine learning model trained by a respective FLDN or a personalized global machine learning model generated based on one or more model training iterations performed by a group of FLDNs. For example, the FLA computing environment may include means for generating one or more training scores for a personalized machine learning model based on local model parameters obtained during one or more training iterations. The training score may include a performance score, a trustworthiness score, and / or the like. In some embodiments, the FLA computing environment includes means for receiving an indication that an FLDN intends to leave a group of FLDNs participating in an FMTL training process or that a new FLDN requests to join the group of FLDNs. In some embodiments, the FLA computing environment includes means for predicting an effect of a participating FLDN departure or the joining of an additional FLDN on performance and / or trustworthiness of a personalized global machine learning model being trained in an ongoing FMTL training process.

[0058] In some embodiments, the apparatus includes one or more computing devices associated with an operator entity, where the FLA computing environment may receive data, commands, requests, user inputs and / or the like from the operator entity via the computing device. In some embodiments, the FLA computing environment includes means for receiving, from the computing device of the operator entity, information on the intent and / or one or more use cases of a machine learning model. In some embodiments, the FLA computing environment includes means for generating or determining one or more personalized tasks of the machine learning model based on the information from the operator entity. In some embodiments, the FLA computing environment includes means for providing an indication of the personalized tasks to the computing device for presentation to and review by the operator entity. In some embodiments, the FLA computing environment includes means for receiving, from the computing device of the operator entity, a filtered subset of the personalized tasks, one or more additional personalized tasks, and / or the like. In some embodiments, the FLA computing environment includes means for generating a new personalized machine learning model based on a personalized task indicated by input from the computing device, where the FLA computing environment may provide the new personalized machine learning model to a current or newly determined plurality of FLDNs for training. In some embodiments, the FLA computing environment includes means for suspending training of a personalized machine learning model based on input from the computing device (e.g., input froman operator entity may cause the FLA computing environment to activate or deactivate personalized tasks and initiate or suspend training of corresponding personalized machine learning models).

[0059] In some embodiments, the FLA computing environment includes means for generating or retrieving from a data store one or more personalized machine learning models based on the determined personalized tasks, where a respective personalized machine learning model is configured for performing a particular personalized task. In some embodiments, the FLA computing environment includes means for generating a report that indicates a plurality of personalized tasks determined for the machine learning model, one or more use cases indicated for the machine learning model, one or more personalized machine learning models generated or retrieved based upon the plurality of personalized tasks, one or more training scores for a respective personalized machine learning model, a personalized global machine learning model, or one or more participating FLDNs, and / or the like. In some embodiments, the FLA computing environment includes means for providing the report to the computing device associated with the operator entity.Example Embodiments

[0060] As illustrated in FIG. 1, a network environment 100 for performing federated multi-task learning (FMTL) is provided in accordance with various embodiments of the present disclosure. In some embodiments, the network environment 100 includes a federated learning aggregation (FLA) computing environment 103 in communication with a plurality of federated learning distributed nodes (FLDNs), which are depicted for illustrative and descriptive purposes as FLDNs 105 A, 105B, and 105C. By way of example, the network environment 100 may be deployed within a radio access architecture based on long term evolution advanced (LTE Advanced, LIE -A) and / or new radio (NR, 5G). However, the network environment may be deployed in other network architectures including within other communication networks including, for example, other communication networks developed in the future, e.g., sixth generation (6G) networks, as well as any of a number of existing networks including a universal mobile telecommunications system (UMTS) radio access network (UTRAN, E-UTRAN or NG-RAN), wireless local area network (WLAN or WiFi™), worldwide interoperability for microwave access (WiMAX™), Bluetooth®, personal communications services (PCS), ZigBee®, wideband code division multiple access(WCDMA), systems using ultra-wideband (UWB) technology, sensor networks, mobile ad-hoc networks (MANETs) and Internet Protocol multimedia subsystems (IMS) or any combination thereof.

[0061] In some embodiments, the network environment 100 includes one or more computing devices 101 that communicate with the FLA computing environment 103 to provide data and / or requests associated with performing FMTL training processes or receive data associated with FMTL training processes. In some embodiments, a computing device 101 is associated with an operator entity. In some embodiments, the computing device 101 includes means for receiving an input, such as a user input, from an operator entity. In some embodiments, the input includes or indicates one or more intents, use cases, personalized tasks, and / or the like for a machine learning model. In some embodiments, the computing device 101 includes means for providing inputs to the FLA computing environment 103. For example, the computing device 101 may include means for indicating intents or use cases of a machine learning model to the FLA computing environment 103 (e.g., which in turn may translate to personalized tasks for FMTL training of one or more personalized global machine learning models). In another example, the computing device 101 may include means for providing information to the FLA computing environment 103 for determining one or more personalized tasks. The information may include, for example, a performance intent, a trustworthiness intent, one or more use cases, a service level agreement, thresholds for model performance, thresholds for model trustworthiness (e.g., including model explainability), and / or the like. In some embodiments, the computing device 101 includes means for configuring personalized tasks for which FMTL training will be performed. For example, the computing device 101 may include means for filtering, activating, or deactivating a subset of proposed personalized tasks determined by the FLA computing environment 103 and provided to the computing device 101 for presentation to an operating entity.

[0062] In some embodiments, as further depicted in FIGS. 3A-B, the network environment 100 includes or embodies one or more data analytic functions 107 including network data analytic functions (NWDAFs) (e.g., shown as NWDAF 107A in FIGS. 3A-B), one or more management data analytics functions (MDAFSs) (e.g., shown as MDAF 107B in FIGS. 3A-B), and / or the like. In some embodiments, the data analytic function 107 embodies any number of computing devices and / or other systems embodied in hardware, software, firmware, and / or the like that (i) collect data from FLDNs, other network functions, operations, administration, and maintenance (0AM)systems, and other network elements and / or (ii) generate analytic outputs based on collected data. For example, a NWDAF may receive, analyze, and report collected data from other analytic functions and generate data insights based thereon, such as compliance with service-level parameters, statuses of and changes to network device configurations and properties, and / or the like. In some embodiments, the FLA computing environment 103 may embody an NWDAF, such as by executing, via at least one processor, instructions for performing functionality of the FLA computing environment 103 as described herein. In another example, an MDAF may report network-related parameters, such as load level and resource utilization, and generate predictions and recommendations for network management based on historical network behaviors.

[0063] The FLA computing environment 103 may include any number of computing device(s) and / or other system(s) embodied in hardware, software, firmware, and / or the like that communicate with computing devices 101 and FLDNs (e.g., FLDNs 105A, 105B, 105C, and / or the like) over one or more networks, determine personalized tasks for machine learning models, determine associations between personalized tasks and FLDNS, generate personalized global machine learning models, determine training scores and other metrics associated with FMTL training processes, and / or the like. In some embodiments, the FLA computing environment 103 includes or embodies a cross-domain management domain. For example, in some contexts, the FLA computing environment 103 is a cross-domain management domain including means for performing FMTL training processes for a plurality of FLDNs that include or embody a plurality of individual management domains. In some embodiments a management domain refers to a computing environment including any number of computing device(s) and / or other system(s) embodied in hardware, software, firmware, and / or the like that govern and manages network services, functionalities, and topologies for a particular domain, such as a radio area network (RAN) domain, a transport domain, a fifth generation (5G) core domain, and / or the like. In some embodiments, a cross-domain management domain refers to number of computing device(s) and / or other system(s) embodied in hardware, software, firmware, and / or the like that aggregate data, analytics, and / or resource needs across individual management domains and coordinate data sharing to network performance for all or a subset of the individual management domains.

[0064] In some embodiments, the FLA computing environment 103 includes or embodies a radio access network (RAN) management domain. For example, in some contexts, the FLA computing environment 103 is a RAN management domain including means for performing FMTL trainingprocesses for a plurality of FLDNs that include or embody a plurality of network nodes (e.g., gNodeBs and / or the like). In some embodiments, the FLA computing environment 103 includes or embodies a core management domain. For example, in some contexts, the FLA computing environment 103 is a core management domain including means for performing FMTL training processes for a plurality of FLDNs that include or embody a plurality of network data analytic functions (NWDAFs). In some embodiments, the FLA computing environment 103 includes or embodies a core NWDAF. For example, in some contexts, the FLA computing environment 103 is a core NWDAF including means for performing FMTL training processes for a plurality of FLDNs that include or embody a plurality of distributed NWDAFs.

[0065] In some embodiments, the FLA computing environment 103 includes or embodies a network node. For example, in some contexts, the FLA computing environment 103 is a gNodeB, or other suitable network node, including means for performing FMTL training processes for a plurality of FLDNs that include or embody a plurality of user equipment (UEs). In some embodiments, the FLA computing environment 103 includes or embodies a non-real-time RAN intelligent controller (RIC). For example, in some contexts, the FLA computing environment 103 is a non-real-time RIC, such as an rApp, including means for performing FMTL training processes for a plurality of FLDNs that include or embody a plurality of real-time RICs (e.g., a plurality of xApps and / or the like). In some embodiments, the FLA computing environment 103 includes or embodies a real-time RIC. For example, in some contexts, the FLA computing environment 103 is a real-time RIC, such as an xApp, including means for performing FMTL training processes for a plurality of FLDNs that include or embody a plurality of network nodes, such as gNodeBs.

[0066] In some embodiments, the FLA computing environment 103 includes means for determining associations between various personalized tasks and FLDNs. For example, the FLA computing environment 103 may include means for determining, for at least a first personalized task and a second personalized task of a plurality of personalized tasks, (i) a first subset of a plurality of federated learning distributed nodes (FLDNs) associated with the first personalized task and (ii) a second subset of the plurality of FLDNs associated with the second personalized task. In some embodiments, the FLA computing environment 103 includes means for generating initial or subsequent iterations of personalized global machine learning models, where a respective personalized global machine learning model is configured for performing a particular personalized task. For example, the FLA computing environment 103 may include means for generating a firstpersonalized machine learning model for performing a first personalized task and a second personalized machine learning model for performing a second personalized task, where the respective personalized tasks are associated with one or more use cases for the machine learning models.

[0067] In some embodiments, the FLA computing environment 103 includes means for performing FMTL training including coordinating a plurality of model training iterations between the FLA computing environment 103 and various groupings of participating FLDNs, where each grouping of FLDNs may be associated with a different personalized task and locally training a corresponding personalized machine learning model. For example, the FLA computing environment 103 may include (i) means for providing a first personalized machine learning model, or a subsequent trained iteration of the first personalized machine learning model, to a first subset of a plurality of FLDNs, and (ii) means for providing a second personalized machine learning model, or a subsequent trained iteration of the second personalized machine learning model, to a second subset of the plurality of FLDNs (e.g., wherein the respective FLDNs separately train the corresponding personalized machine learning model to perform a first personalized task or a second personalized task). In some embodiments, the FLA computing environment includes means for receiving, from FLDNs, personalized local machine learning models that were independently and locally trained by respective FLDNs. In some embodiments, a personalized local machine learning model may include or embody a plurality of local model parameters. In one example, the FLA computing environment 103 may include means for receiving, from a first and second subset of a plurality of FLDNs, respective local model parameters associated with the training of the corresponding personalized global machine learning models by FLDNs of each subset (e.g., where a respective subset of FLDNs is associated with FMTL training of machine learning models for performing one of a plurality personalized tasks).

[0068] In some embodiments, the FLA computing environment 103 includes means for generating a subsequent iteration of a personalized global machine learning model based on an aggregation of local model parameters from a plurality of trained personalized local machine learning models (e.g., where a respective personalized local machine learning model is generated by an FLDN training a provided version of the personalized global machine learning model). For example, the FLA computing environment 103 may include (i) means for generating a subsequent trained iteration of a first personalized global machine learning model based on an aggregation of the localmodel parameters received from a first subset of a plurality of FLDNs performing FMTL training of the first personalized global machine learning model, and (ii) means for generating a subsequent trained iteration of a second personalized machine learning model based on an aggregation of local model parameters received from a second subset of the plurality of FLDNs performing FMTL training of the second personalized global machine learning model.

[0069] In some embodiments, the FLA computing environment 103 includes means for receiving, in a respective model training iteration, one or more training scores from each participating FLDN for each respective personalized global machine learning model being trained. For example, the FLA computing environment 103 may include means for receiving, from respective FLDNs of a first subset and respective FLDNs of a second subset, at least one training score associated with training of the corresponding personalized ML model by the FLDNs. In some embodiments, the training score includes a performance score, a trustworthiness score, and / or the like. In some embodiments, the FLA computing environment 103 includes means for generating one or more aggregated training scores for a personalized global machine learning model based on corresponding training scores. For example, the FLA computing environment 103 may include means for generating an aggregated performance score based on a plurality of locally achieved performance scores and an aggregated trustworthiness score based on a plurality of locally achieved trustworthiness scores. In some embodiments, the FLA computing environment 103 includes means for providing one or more aggregated training scores for one or more personalized global machine learning models to an operating entity. For example, the FLA computing environment 103 may include means for providing a respective aggregated performance score and a respective aggregated trustworthiness score for a first personalized global machine learning model and a second personalized global machine learning model to a computing device 101 and / or one or more proxy entities associated with an operator entity.

[0070] In some embodiments, the FLA computing environment 103 includes means for determining one or more personalized tasks based on a machine learning model use case received from the computing device 101 associated with the operator entity or more proxy entities associated with the operating entity. In some embodiments, the FLA computing environment 103 receives the machine learning model use case in the form of a user input. In one example, a machine learning model use case includes network performance improvement for mobility, network improvement for stability, model trustworthiness improvement, and / or the like. In someembodiments, the FLA computing environment 103 includes means for receiving, from a data store (e.g., memory 204 as shown in FIG. 2), one or more historical personalized tasks based on one or more one machine learning model use cases. In some embodiments, the FLA computing environment 103 includes means for determining one or more personalized tasks based at least in part on one or more historical personalized tasks.

[0071] In some embodiments, the FLA computing environment 103 includes means for generating a mapping of one or more one machine learning use case to subsets of a plurality of historical personalized tasks. In some embodiments, FLA computing environment 103 includes means for generating categories of candidate FLDNs based on the mapping. For example, the FLA computing environment 103 may include means for generating, based on the mapping, (i) a first category comprising a first subset of the plurality of FLDNs and (ii) a second category comprising a second subset of the plurality of FLDNs. In some embodiments, the FLA computing environment 103 includes means for determining one or more personalized tasks are based on one or more generated categories of FLDNs.

[0072] In some embodiments, the FLA computing environment 103 includes means for providing a plurality of historical personalized tasks to a computing device 101 associated with the operator entity. In some embodiments, the FLA computing environment 103 includes means for receiving from the operator entity (e.g. , via the computing device 101 or one or more proxy entities) a filtered subset of the plurality of historical personalized tasks or one or more additional personalized task. For example, the FLA computing environment 103 may include means for receiving a user input indicative of a subset of the plurality of historical personalized tasks or one or more additional personalized tasks. In some embodiments, the FLA computing environment 103 includes means for determining personalized tasks for FMTL training based on a received filtered subset of the plurality of historical personalized tasks and / or one or more additional personalized tasks.

[0073] In some embodiments, the FLA computing environment 103 includes means for determining one or more additional personalized tasks during or between iterations of FMTL training for other personalized tasks and personalized global machine learning models. In some embodiments, the FLA computing environment 103 includes means for generating one or more personalized global machine learning models for performing the additional determined personalized task. For example, the FLA computing environment 103 may include, between model training iterations for FMTL training of a first and a second personalized global machine learningmodel for a first personalized task and a second personalized task, (i) means for determining a third personalized task, (ii) means for determining a third subset of a plurality ofFLDNs associated with the third personalized task, (iii) means for generating a third personalized machine learning model to perform the third personalized task, and (iv) means for providing the third personalized machine learning model or a subsequent trained iteration of the third personalized machine learning model to the third subset ofFLDNs. The respective FLDNs of the third subset may train the third personalized machine learning model or the subsequent trained iteration of the third personalized machine learning model in a remaining subset of the plurality of model iterations. In a final model training iteration, a personalized global machine learning model for the third personalized task may be generated by the FLA computing environment 103 based on a corresponding plurality of local model parameters received from respective FLDNs of the third subset during the remaining subset of the plurality of model training iterations.

[0074] In some embodiments, the FLA computing environment 103 includes means for determining FLDNS for FMTL training based on or more properties or statuses of one or more FLDNs. In some embodiments, the properties or statuses include a computational resource availability, a power availability, a channel quality, a data quality score, a data trustworthiness score, location data, velocity data, a trust score, a quantity of available training data samples, and / or the like. In some embodiments, the FLA computing environment 103 includes means for receiving the properties or statuses, or indications of changes thereof, from an FLDN or an analytic function associated with the FLDN (e.g., NWDAF, MDAF, and / or the like). In some embodiments, the FLA computing environment 103 includes means for removing one or more participating FLDNs from further FMTL training of a personalized global machine learning model based on a change to the one or more properties or statuses of the FLDN. In some embodiments, the FLA computing environment 103 includes means for reassigning the removed FLDN to a second personalized task (e.g., for participating in FMTL training of a corresponding personalized global machine learning model) based on a change to the one or more properties or statuses of the FLDN.

[0075] In some embodiments, the FLA computing environment 103 includes means for providing a notification indicative of a plurality of personalized tasks to one or more FLDNs. In some embodiments, the FLA computing environment 103 includes means for receiving, from one or more FLDNs, a request to associate FLDN with an indicated personalized task. In some embodiments, the FLA computing environment 103 includes means for determining participantFLDNs for FMTL training based on such requests from FLDNs to associate the FLDN with a personalized task.

[0076] In some embodiments, the FLA computing environment 103 includes means for receiving, from a participant FLDN, a request for disassociation from a grouping of participant FLDNs associated with ongoing FMTL training of a personalized global machine learning model. In some embodiments, the FLA computing environment 103 includes means for determining that a current or historical training score associated with the requesting FLDN and a personalized machine learning model trained by the requesting FLDN meets one or more predetermined thresholds. In some embodiments, the FLA computing environment 103 includes means for, in response to the at least one training score meeting the predetermined threshold, providing, to the requesting FLDN, a request to retain the at FLDN in the corresponding grouping of FLDNs for a plurality of training iterations. In some embodiments, the request to retain the FLDN indicates a monetary retention incentive. The monetary retention incentive may include an amount or rate of monetary compensation for continuing to participate in the FTML training process. For example, the operator entity may provide a monetary amount to one or more UEs to participate (or continue participating) in FMTL training. In another example, the operator entity may allocate resources or services to the UE in exchange for the UE participating in FMTL training. In some embodiments, the request for disassociation of the participant FLDN includes one or more disassociation factors including a computational overload, a bandwidth limitation, a power limitation, and / or the like. In some embodiments, the request for disassociation includes a duration of unavailability of the participant FLDN.

[0077] In some embodiments, the FLA computing environment 103 includes means for receiving, during FMTL training, a request for inclusion of an additional FLDN in a grouping of participant FLDNs associated with the training. In some embodiments, the FLA computing environment 103 includes means for determining an estimated impact to performance and / or trustworthiness of a corresponding personalized global machine learning model if the requesting FLDN is added to the grouping of participant FLDNs for training the personalized global machine learning model. In some embodiments, the FLA computing environment 103 includes means for adding the requesting FLDN to the grouping of participant FLDNS in response to determining that the estimated impact to model performance and / or model trustworthiness meets a predetermined threshold.

[0078] The FLDN 105A, 105B, or 105C may include any number of computing device(s) and / or other system(s) embodied in hardware, software, firmware, and / or the like that communicate with the FLA computing environment 103 over one or more networks, provide data (e.g., local environment information, machine learning model data, and / or the like) to the FLA computing environment 103, store local training data, receive and train personalized global machine learning models (e.g., including generating local model parameters), generate training scores for machine learning models, request association with or disassociation from personalized tasks, and / or the like. In some embodiments, an FLDN includes or embodies an individual management domain, a network node, an NWDAF, a user equipment, a real-time RIC, and / or the like. For example, in some contexts, the FLDN is a network node, such as a gNodeB, including means for receiving and independently and locally training a personalized global machine learning model using local training data. In another example, in some contexts, the FLDN is a distributed NWDAF.

[0079] In some embodiments, the FLDN includes means for storing training data in memory accessible to the FLDN. For example, the FLDN includes means for storing local training data that may be used by the FLDN to train one or more iterations of a personalized global machine learning model. In some embodiments, the FLDN includes means for receiving from the FLA computing environment 103 a request to provide local environment data associated with the FLDN. In some embodiments, the FLDN includes means for providing local environment data to the FLA computing environment 103. In some embodiments, the local environment data includes resource- related information, training data-related information, and / or the like. In some embodiments, the local environment data includes one or more indications of computational resource availability, power availability, channel quality, data quality, data trustworthiness, and / or the like. In some embodiments, the FLDN includes means for providing to the FLA computing environment 103 indications of one or more properties or statuses of the FLDN and / or a change to one or more properties or statuses of the FLDN. In some embodiments, the FLDN includes means for receiving one or more predetermined thresholds for detecting particular changes, or magnitudes thereof, to one or more properties or statuses of the FLDN. In some embodiments, the FLDN includes means for determining whether a change to a property or status of the FLDN (including a change to local environment information) meets a predetermined threshold, where a positive determination may cause the FLDN to provide an indication of the change to the FLA computing environment 103. In some embodiments, properties or statuses of the FLDN include FLDN location, FLDN velocity,trust score, quantity of available training data samples, and / or the like. In some embodiments, an FLDN location is indicated by or determined based on location data associated with the FLDN, such as global positioning system (GPS) data and / or the like. In some embodiments, and FLDN velocity is indicated by or determined based on location data, sensor data, such as data from an accelerometer, and / or the like.

[0080] In some embodiments, the FLDN includes means for performing any number of iterations of FMTL training to train a personalized global machine learning model (e.g., including subsequent trained iterations thereof) and, thereby, generate a personalized local machine learning model including local model parameters. For example, the FLDN may include means for locally and independent training, using local training data, a personalized global machine learning model to generate a plurality of local model parameters, which may define a personalized local machine learning model. In some embodiments, the FLDN includes means for providing a personalized local machine learning model and / or corresponding local model parameters to the FLA computing environment 103. In some embodiments, the FLDN includes means for determining one or more training scores or a personalized global (or local) machine learning model. In some embodiments, the one or more training scores include a performance score, a trustworthiness score, and / or the like. In some embodiments, the training score includes metadata indicative of the associated personalized machine learning model and / or local training data used to train the personalized machine learning model. For example, the metadata may include or indicate a machine learning model identifier, a quantity of local training data samples, labels or features of the local training data, a distribution of the local training data, noisiness of the local training data, any omissions of information or samples in the local training data, and / or the like. In some embodiments, the FLDN includes means for providing one or more training scores to the FLA computing environment 103.

[0081] In some embodiments, the FLDN includes means for receiving from the FLA computing environment 103 a notification indicative of personalized tasks. In some embodiments, the FLDN includes means for providing to the FLA computing environment 103 a request to be associated with a particular personalized task, where association of the FLDN with the particular personalized task results in the FLDN being included in FMTL training of a personalized global machine learning model for performing the particular task. In some embodiments, the FLDN determines a personalized task for which to request association based on local environment information associated with the FLDN and / or other properties or statuses associated with the FLDN.

[0082] In some embodiments, the FLDN includes means for providing to the FLA computing environment 103 a request for disassociation from a personalized task with which the FLDN is currently associated at the FLA computing environment 103. In some embodiments, the request for disassociation causes the FLA computing environment 103 to remove the FLDN from further FMTL training iterations for a personalized global machine learning model for performing the corresponding personalized task. In some embodiments, the request for disassociation includes or indicates one or more disassociation factors including computational overload, bandwidth limitation, and / or the like. In some embodiments, the request for disassociation includes or indicates a duration of unavailability of the FLDN to participate in FMTL training for the currently associated personalized task and / or other personalized tasks. In some embodiments, the request for disassociation causes the FLA computing environment 103 to provide to the requesting FLDN a request to remain associated with the personalized task and continue participating in FMTL training of the corresponding personalized global machine learning model. In some embodiments, the request to retain the FLDN indicates or includes one or more retention incentives. For example, the request to retain the FLDN may indicate a monetary amount or other resource that will be provided to the FLDN in response to the FLDN remaining associated with the personalized task and continuing to participate in FMTL training of the personalized global machine learning model. In some embodiments, the FLDN includes means for providing, to the FLA computing environment 103, an acceptance or rejection of the request to retain the FLDN, where acceptance or rejection may cause the FLA computing environment 103 to retain or remove the FLDN as a participant in FMTL training for the corresponding personalized task.

[0083] FIG. 2 shows an example apparatus 200 according to one embodiment. In some embodiments, the federated learning aggregation (FLA) computing environment shown in FIG. 1 and described herein is embodied as one or more of the apparatus 200. In some embodiments, the FLDN shown in FIG. 1 and described herein is embodied as the apparatus 200 (e.g., where such embodiment of the apparatus 200 includes means for performing operations and functions of the FLDN described herein). In one example, the FLA computing environment 103 may be a first user equipment (UE) embodied as a first apparatus 200 configured to perform the process 500 (FIG. 5) and an FLDN 105 may be a second UE embodied as a second apparatus 200 configured to perform the process 600 (FIG. 6).

[0084] The apparatus 200 may be an embodiment of a network device or may be embodied by or otherwise associated with a network device. For example, the apparatus may be embodied by or otherwise associated with the application function (AF), the multicast and broadcast session management function (MB-SMF), and / or the multicast and broadcast user plane function (MB- UPF).

[0085] Regardless of the device that embodies the apparatus 200, the apparatus may include processor 202, memory 204, and network interface 206. The apparatus 200 may be configured to execute the operations described herein. For example, the processor 202, memory 204, and / or network interface 206A may perform actions and processes described herein for implementing and managing federated multi-task learning (FMTL)-based training of personalized global machine learning models. Although these components are described with respect to the performance of various functions, it should be understood that the particular implementations necessarily include the use of particular hardware. It should also be understood that certain of these components may include similar or common hardware. For example, two sets of circuitries may both leverage use of the same processor, network interface, storage medium, or the like to perform their associated functions, such that duplicate hardware is not required for each set of circuitries.

[0086] In some embodiments, the processor 202 (and / or co-processor or any other processing circuitry assisting or otherwise associated with the processor) may be in communication with the memory 204 via a bus for passing information among components of the apparatus. The memory 204 is non-transitory and may include, for example, one or more volatile and / or non-volatile memories. In other words, for example, the memory 204 may be an electronic storage device (e.g., a non-transitory computer-readable storage medium). The memory 204 may be configured to store information, data, content, applications, instructions, or the like for enabling the apparatus to carry out various functions in accordance with an example embodiment disclosed herein.

[0087] The processor 202 may be embodied in a number of different ways and may, for example, include one or more processing devices configured to perform independently. In some non-limiting embodiments, the processor 202 may include one or more processors configured in tandem via a bus to enable independent execution of instructions, pipelining, and / or multithreading. The use of the term “processor” may be understood to include a single core processor, a multi-core processor, multiple processors internal to the apparatus, and / or remote or “cloud” processors.

[0088] In some embodiments, the processor 202 may be configured to execute instructions stored in the memory 204 and / or circuitry otherwise accessible to the processor 202. In some embodiments, the processor 202 may be configured to execute hard-coded functionalities. As such, whether configured by hardware or software methods, or by a combination thereof, the processor 202 may represent an entity (e.g., physically embodied in circuitry) capable of performing operations according to an embodiment disclosed herein while configured accordingly. Alternatively, as another example, when the processor 202 is embodied as an executor of software instructions, the instructions may specifically configure the processor 202 to perform the algorithms and / or operations described herein when the instructions are executed.

[0089] In some embodiments, the apparatus 200 may optionally include input / output circuitry that may, in turn, be in communication with processor 202 to provide output to a user and / or other entity and, in some embodiments, to receive an indication of an input. The input / output circuitry may comprise a user interface and may include a display, and may comprise a web user interface, a mobile application, a query-initiating computing device, a kiosk, or the like. In some embodiments, the input / output circuitry may also include a keyboard, a mouse, a joystick, a touch screen, touch areas, soft keys, a microphone, a speaker, or other input / output mechanisms. The processor and / or user interface circuitry comprising the processor may be configured to control one or more functions of one or more user interface elements through computer program instructions (e.g., software and / or firmware) stored on a memory accessible to the processor (e.g., memory 204, and / or the like).

[0090] The network interface 206 may be any means such as a device or circuitry embodied in either hardware or a combination of hardware and software that is configured to receive and / or transmit data from / to a network and / or any other device, circuitry, or module in communication with the apparatus 200. In this regard, the network interface 206 may include, for example, a network interface for enabling communications with a wired or wireless communication network, such as the application function (AF), multicast and broadcast service function (MBSF), multicast and broadcast user plane function (MB-UPF), and / or multicast and broadcast session management function (MB-SMF). For example, the network interface 206 may include one or more network interface cards, antennae, buses, switches, routers, modems, and supporting hardware and / or software, or any other device suitable for enabling communications via a network. Additionally, or alternatively, the network interface 206 may include the circuitry for interacting with theantenna / antennae to cause transmission of signals via the antenna / antennae or to handle receipt of signals received via the antenna / antennae.

[0091] FIGS. 3A-3B illustrate a cross-functional diagram for federated multi-task learning (FMTL) in accordance with an example embodiment of the present disclosure.

[0092] As shown in FIG. 3 A, the FLA computing environment 103 a computing device 101 of an operating entity may perform a preparation sequence 300A for FMTL. In some embodiments, the preparation sequence 300A is performed prior to initiation of a plurality of training iterations of in an FMTL training process. In some embodiments, the operating entity may directly interact with the FLA computing environment 103 via the computing device 101. In some embodiments, the operating entity may indirectly interact with the FLA computing environment 103 via one or more proxy entities including an intent and / or policy manager, a network and service orchestrator, a machine learning orchestrator, a machine learning trust engine, and / or the like.

[0093] In some embodiments, via the computing device 101, the entity operator indicates the intent of the use case for the machine learning model to the FLA computing environment 103 in the form of one or more inputs provided to the FLA computing environment 103 (block 1). The input may be a user input, an input defined by an automated entity, such as a proxy entity, and / or the like. In some embodiments, the intent includes a performance intent, a trustworthiness intent, and / or the like. In some embodiments, intent is defined by a service level agreement (SLA) between the operator entity and a vendor entity, which may be associated with the FLA computing environment 103, one or more FLDNs 105, and / or the like. In some embodiments, the use case includes handover optimization in which FMTL training may be performed to train personalized global machine learning models for UEs associated with different intents (e.g., and, resultingly, different personalized tasks).

[0094] As one example, in a handover optimization use case, an intent may include or embody a network performance-related intent for different user equipment (UEs) associated with different speeds and, resultingly, different network performance thresholds, such as threshold limits for radio link failures (RLFs). For example, high-speed UEs may be associated with a higher threshold limit for RLFs, medium-speed UEs may be associated with a medium threshold limit for RLFs, and low-speed UEs may be associated with a lower threshold limit for RLFs. As another example in the handover optimization use case, the intent may include or embody a network performance- related intent for UEs associated with different cell coverages, which may be identified usingtracking area identifiers (TAIs). For example, UEs within a cell coverage of a first TAI (TAI 1) may be associated with a higher threshold limit for RLFs, UEs within a cell coverage of a second TAI (TAI 2) may be associated with a medium threshold limit, and UEs within a cell coverage of a third TAI (TAI 3) may be associated with a lower threshold limit for RLFs.

[0095] As another example, in a handover optimization use case, an intent may include or embody a machine learning trustworthiness-related intent. For example, UEs within a cell coverage of a TAI 1 may be associated with a higher threshold for machine learning model explainability, UEs within a cell coverage of TAI 2 may be associated with a medium threshold for machine learning model explainability, and UEs within a cell coverage of TAI 3 may be associated with a lower threshold for machine learning model explainability.

[0096] In some embodiments, the FLA computing environment 103 retrieves one or more previously stored personalized tasks and associated metadata from memory 204 to enhance the process of personalized tasks generation, leveraging past experiences, and increase the reusability of already determined personalized tasks (block 2). As described herein, in some embodiments, the memory 204 includes or embodies one or more data stores or repositories, which may be locally or remotely accessible to the FLA computing environment 103. In some embodiments, the FLA computing environment 103 determines one or more personalized tasks for FMTL for the considered use case based on the information provided by the operator entity in block 1. In some embodiments, the FLA computer environment 103 generates one or more mappings of intents to personalized tasks for FMTL, where a respective intent may be associated with a category of FLDNs, such as a category of UEs. As one example, the FLA computing environment 103 may categorize a plurality of UEs (e.g., FLDNs) based on their respective velocities (e.g., low-speed, medium-speed, high-speed), where the FLA computing environment 103 may map each category of UEs to a personalized task focused on network performance (e.g., RLFs). As another example, the FLA computing environment 103 may categorize the plurality of UEs based on their respective geographical locations (e.g., TAI 1, 2, 3, etc.,), where the FLA computing environment 103 may map each category of UEs to a personalized task focused on network performance. In another example, the FLA computing environment 103 may categorize the plurality of UEs based on their respective geographical locations with each category being mapped to a personalized task focused on machine learning trustworthiness (e.g., machine learning model explainability).

[0097] In some embodiments, the FLA computing environment 103 may report on the determined personalized tasks for FMTL from block 2 and / or block 3 to the computing device 101 for verification by the operating entity (e.g., personalized task filtering, activation, deactivation, addition) (block 4). In some embodiments, the FLA computing environment 103 stores the determined personalized tasks in memory 204. For example, the FLA computing environment 103 may store the determined personalized tasks in one or more data stores, which may be accessible to additional FLA computing environments.

[0098] As shown in FIG. 3B, the FLA computing environment 103, computing device 101 of the operating entity, plurality FLDNs 105, and, optionally, an NWDAF 107A and / or MDAF 107B may perform a training sequence 300B for FMTL. In some embodiments, the training sequence 300B, or a subset of blocks thereof, are performed repetitiously for a plurality of training iterations. In some embodiments, via the computing device, 101 an operator entity filters a subset of the personalized tasks from the overall personalized tasks reported at block 4 and provides the filtered subset of personalized tasks to the FLA computing environment 103 (block 5 A). In some embodiments , at block 5 A, the operating entity provides to the FLA computing environment 103 a recommendation for one or more new personalized tasks in addition to or in substitution of the overall personalized tasks reported at block 4. In some embodiments, the operating entity configures (e.g., creates, updates, deletes, disables, enables, and / or the like) the filtered and / or recommended personalized tasks in the FLA computing environment 103 via one or more proxy entities (block 5B).

[0099] In some embodiments, the FLA computing environment 103 provides requests to candidate FLDNs 105, one or more NWDAFs 107A, and / or one or more MDAFs 107B for reports on the local environment of the respective FLDN (block 6A). In some embodiments, the FLA computing environment 103 receives the reports from the FLDNs 105, NWDAFs 107A, and / or MDAFs 107B (block 6B) (e.g., which may be used as input in further determinations of personalized tasks, participant FLDNs, and / or the like). In some embodiments, the requests for information provided to the FLDNs 105 or analytic functions include a reporting deadline set by the FLA computing environment 103.

[0100] The FLA computing environment 103 may provide the one or more requests between each iteration of FMTL training or in a subset of model training iterations. In some embodiments, the FLA computing environment 103 indicates to the candidate FLDNs 105, NWDAF 107A, and / orMDAF 107B the reporting parameters for requested reports. In some embodiments, reporting parameters include computational resource availability, power availability, channel quality, data quality, data trustworthiness, and / or the like. In some embodiments, the FLA computing environment 103 provides requests for additional information to the candidate FLDNs 105 and / or one or more analytic functions, such as a network data analytic function (NWDAF) or management data analytic function (MDAF). In some embodiments, the additional information includes location information (e.g., current location of the FLDN 105, associations with tracking area identifiers, and / or the like), velocity information (e.g., current speed, acceleration, and / or the like), FLDN trust score, historical data, quantity of available training samples accessible to the FLDN 105, and / or the like.

[0101] In some embodiments, the FLA computing environment 103 subscribes to the FLDNs 105 and / or analytics functions for notifications regarding changes to an FLDN local environment and / or one or more properties or statuses of a respective FLDN 105. In some embodiments, the FLA computing environment 103 provides one or more threshold values or conditions to the FLDNs 105 and / or analytic functions, such as the NWDAF 107A or MDAF 107B. In some embodiments, the FLDN or analytic function determine whether a change to an aspect of the FLDN local environment or a change to one or more statuses or properties of the FLDN 105 satisfy a provided threshold. In some embodiments, in response to determining satisfaction of the threshold, the FLDN 105 or analytic function indicates the change in FLDN local environment and / or FLDN property or status to the FLA computing environment 103. In one example, in response to an FLDN 105, NWDAF 107A, or MDAF 107B determining that computational resource availability at the FLDN 105 falls below 2 central processing units (CPUs) (e.g., or another suitable threshold value), the FLDN 105, NWDAF 107A, or MDAF 107B provides an indication of the change in computational resource availability to the FLA environment 103. In another example, in response to an FLDN 105, NWDAF 107A, or MDAF 107B determining that a location of the FLDN 105 changes from a first TAI to a second TAI, the FLDN 105, NWDAF 107A, or MDAF 107B provides an indication of the change in TAI association to the FLA environment 103. In another example, in response to an FLDN 105, NWDAF 107A, or MDAF 107B determining that computational resource availability at the FLDN 105 exceeds or falls below a velocity threshold (e.g., 50 kilometers (km) / hour (hr), 60 km / hr, or another suitable threshold value), the FLDN 105, NWDAF 107A, or MDAF 107B provides an indication of the change in FLDN velocity to theFLA environment 103. In some embodiments, where a reporting deadline is specified and the FLDN 105 or analytic function determines that no threshold is satisfied, the FLDN or analytic function may indicate a lack of change (“no change”) to the FLA computing environment 103 (e.g., instead of sending a full report of the current FLDN local environment and / or properties or statuses).

[0102] In some embodiments, based on the collected information in blocks 5 and / or 6A-B, the FLA computing environment 103 further determines new personalized tasks in addition to tasks that were determined in block 2 (block 7). The FLA computing environment 103 may determine new personalized tasks by categorizing the FLDNs 105 based on the received information. In some embodiments, the FLA computing environment 103 may categorize FLDNs 105 based on their respective total number of local training samples. For example, the FLA computing environment 103 may categorize FLDNs having less than 10000 samples to a first personalized task and categorize FLDNs having more than 10000 to a second personalized task. In some embodiments, the FLA computing environment 103 categorizes FLDNs based on their trust score. For example, the FLA computing environment 103 may categorize FLDNs having a trust score within in a high threshold range to a first personalized task and categorize FLDNs having a trust score in a low threshold range to a second personalized task.

[0103] In some embodiments, the FLA computing environment 103 stores the determined personalized tasks, associations between personalized tasks and FLDNs, and corresponding metadata in memory 204. For example, FLA computing environment 103 may store the determined personalized tasks and corresponding metadata in a data repository (e.g., which may be associated with an analytics data repository function) such that other FLA computing environments may utilize the information by directly querying the data repository.

[0104] In some embodiments, based on the information gathered in blocks 4-6, the FLA computing environment 103 determines associations between candidate FLDNs and identified personalized tasks (e.g., also considering the intersection between personalized tasks identified in blocks 2 and 6) (block 8). In some embodiments, the FLA computing environment 103 provides indications of personalized tasks and corresponding metadata information to the FLDNs 105 such that the FLA computing environment 103 may receive self-selections from the FLDNs 105 to participate in FMTL training of corresponding personalized global machine learning models. For example, the FLA computing environment 103 may broadcast all the determined personalizedtasks to the candidate FLDNs 105 along with metadata information for each task. The FLDNs 105 may themselves select the personalized task that they would like to participate in and report the selection back to the FLA computing environment 103.

[0105] In some embodiments, the FLA computing environment 103 generates and / or or retrieves from memory 204 respective personalized global machine learning models that are mapped to corresponding personalized tasks (block 9). Also at block 9, in some embodiments, the FLA computing environment 103 provides the personalized global machine learning models to corresponding FLDNs 105 as determined in step 7. In some embodiments, the respective FLDN 105 locally and independently trains the personalized global machine learning model to perform the corresponding personalized task, thereby generated a personalized local machine learning model. In some embodiments, in addition to the personalized global machine learning model, the FLA computing environment 103 may also include metadata information related to the personalized task selected for an FLDN 105. In some embodiments, the FLDNs 105 use such information to customize their respective local training to improve the trained personalized model performance and trustworthiness. For example, based on the metadata information, an FLDN 105 may generate synthetic data to increase the number of training samples for local training, such as in instances where the personalized tasks are determined based on the total number of local training samples available in FLDNs.

[0106] In some embodiments, the FLA computing environment 103, following an iteration of FMTL training at the FLDNs 105, receives respective personalized local machine learning models from the FLDNs 105 (block 10), where the personalized local machine learning models may be provided in the form of local model parameters, one or more training scores (e.g., performance score, trustworthiness score, and / or the like), metadata (e.g., number of training samples, data quality indicators, etc.,), and / or the like. For example, after training their respective local models, the FLDNs 105 may send their local model parameters back to the FLA computing environment 103 within a reporting deadline set by the FLA computing environment 103.

[0107] In some embodiments, the FLA computing environment 103 aggregates local model parameters received from FLDNs 105 to obtain subsequent iterations of personalized global machine learning models corresponding to each personalized task (block 11). In some embodiments, the FLA computing environment 103 stores one or more versions of the personalized global machine learning models in memory 204. For example, the FLA computingenvironment 103 may store a time series record of versions of a personalized global machine learning model obtained throughout iterations of the workflow 300B.

[0108] In some embodiments, once a personalized global machine learning model converges and / or or after a predetermined quantity of model training iterations (e.g., 3, 5, or any suitable quantity), the FLA computing environment 103 receives an achieved personalized model performance and personalized model trustworthiness from each participant FLDN 105 (block 12). In some embodiments, the achieved personalized model performance and personalized model trustworthiness are provided in the form of one or more training scores, such as a personalized local model performance score and a personalized local model trustworthiness score. In some embodiments, the FLA computing environment 103 generates an average model performance and average model trustworthiness for each personalized task by averaging the individual model performance and individual model trustworthiness measures reported by each corresponding FLDN 105.

[0109] In some embodiments, based on the generated average model performance and average model trustworthiness for each personalized task, the FLA computing environment 103 may also calculate the “actual contribution” of each participant FLDN 105 to the corresponding personalized global machine learning model (block 12). In some embodiments, according to Equation 1, the FLA computing environment 103 generates an average over the personalized model performance / trustworthiness of all participant FLDNs 105 within a personalized task and generates a contribution score value for each participant FLDN 105.„ , „ , „ received personalized model performance / trustworthiness of each FLDNScore value of each FLDN= - - - - - - —!- - - - - averaged personalized model performance / trustworthiness(Equation 1)

[0110] In some embodiments, in response to determining that the score value of a particular FLDN 105 is low (e.g., falls below a predetermined threshold), the FLA computing environment 103 initiates a troubleshooting process by requesting additional information from the corresponding FLDN 105 and analytics functions (e.g., NWDAF 107A, MDAF 107B, and / or the like). In some embodiments, based on the score value and / or new information, the FLA computing environment 103 (i) reassigns the participant FLDN 105 to a new or alternative existing personalized task for the model training iteration, or (ii) removes the participant FLDN 105 from the candidate FLDNlist, thereby excluding the FLDN 105 from subsequent model training iterations. In some embodiments, the FLA computing environment 103 determines the individual achieved personalized model performance and achieved personalized model trustworthiness of the participant FLDN without the participant FLDN sending training scores for performance and trustworthiness. For example, the FLA computing environment 103 may independently determine the individual contribution (e.g., performance and trustworthiness) of the FLDN 105 in embodiments where the FLA computing environment 103 is able to access a common test dataset for the use case or receives a common test dataset from the computing device 101 of the operator entity to perform such evaluation on the received personalized global machine learning models.

[0111] In some embodiments, following model convergence or a predetermined quantity of model training iterations, the FLA computing environment 103 generates and provides to the operating entity the respective average model performance and average model trustworthiness for the final iteration of each personalized global machine learning model (block 13). In some embodiments, the FLA computing environment 103 provides the averaged model information to the operating entity via transmission to the computing device 101, one or more proxy entities, and / or the like.

[0112] FIG.4 shows an example schema 400 for managing federated multi-task learning (FMTL) in accordance with an example embodiment of the present disclosure. In some embodiments, the schema 400 includes an FLA consumer 401, a FLA computing environment 403, and a plurality of FLDN producers 405. The FLA consumer 401 is representative of the operator entity described herein, including computing device and / or proxy entities associated with the operating entity. In some embodiments, the plurality of FLDN producers 405 are representative of federated learning distributed nodes (FLDNs) as described herein. In some embodiments, as shown in FIG. 4, the FLA computing environment 403 communicates with the FLA consumer 401 and the plurality of FLDN producers to coordinate and manage FMTL training of personalized global machine learning models for performing respective personalized tasks (e.g., corresponding to one or more machine learning use cases).

[0113] In some embodiments of FMTL, FLDN producers 405 report information on availability of computation and radio resources at each model training iteration. In such approaches, for each training iteration, the participant FLDNs may be firstly selected. For example, the candidate FLDNs may report their computation resource availability for the training task to the FLA computing environment 403, such as a federated learning (FL) server. The FLA computingenvironment 403 may make the participant FLDN selection based on the reports from the candidate FLDNs and other conditions, such as wireless channel conditions. Previous approaches to federated learning may fail to collect and consider information for the available training data on the FLDNs. In some embodiments, the schema 400 includes enabling the FLA computing environment 403 to collect information about training data distributions, data quality and local datasets of the candidate FLDNs in order to detect potential biases and / or misbehaviors and promptly react. Such embodiments may address potential challenges with generating FLDN selections for federated; however, to, such approaches may fail to achieve high averaged statistical accuracy in output machine learning models across heterogenous FLDNs and corresponding local environments. Such challenges of federated learning may be overcome by implementing federated multi-task learning (FMTL) techniques including novel management mechanisms for FMTL training developed by the inventors.

[0114] In some embodiments, the schema 400 includes configuring and / or enabling the FLA consumer 401 to provide information on the intent of the use case of one or more machine learning models to assist the FLA computing environment 403 in determining personalized tasks for FMTL. In some embodiments, the FLA consumer 401 further includes means for discovering supported / available personalized tasks from FLA computing environment 403. In some embodiments, the FLA consumer 401 includes means for configuring activation and deactivation of personalized tasks at the FLA computing environment 403, thereby controlling personalized tasks for which FMTL training is performed. In some embodiments, the FLA consumer 401 includes means for receiving reports related to personalized global models from FLA computing environment 403.

[0115] In some embodiments, the schema 400 includes configuring and / or enabling the FLA computing environment 403 to determine possible or requisite personalized tasks for a use case in one or more iterations of the FMTL training process. In some embodiments, the FLA computing environment 403 includes means for discovering the mapping between the FLDN producers 405 (e.g., candidate FLDNs) and their associated personalized tasks in one or more iterations of FMTL training process. In some embodiments, the FLA computing environment 403 includes means for determining the performance and / or trustworthiness of personalized global model(s) from each participant FLDN producer 405, or grouping of FLDN producers (e.g., for which average performance and / or trustworthiness may be determined by the FLA computing environment 403).In some embodiments, the FLA computing environment 403 includes means to ensure the performance and trustworthiness of personalized global model(s) is not adversely affected beyond a threshold level if one or more participant FLDN producers 405 depart from or one or more candidate FLDN producers 405 determine to join the FMTL training process.

[0116] Referring now to FIG 5., shown is an example flowchart of a federated multi-task learning (FMTL) training process 500, which may be performed by an apparatus, such as the apparatus 200 of FIG. 2. The apparatus 200 may embody the FLA computing environment 103 shown in FIG. 1, including means for performing various functions and operations as described herein.

[0117] In some embodiments, at block 503, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for receiving, from a computing device of an operator entity, an intent of one or more use cases for a machine learning model. In one example, the intent may include improving network performance in device mobility. In another example, the intent may include improving network stability. In another example, the intent may include improving model trustworthiness for one or more global machine learning networks associated with providing automated network management. In some embodiments, the use case refers to improvement of the corresponding network aspect for a particular category of network devices. For example, use cases for improving network performance may include generating respective globalized machine learning models to minimize, at varying threshold limits, radio link failures (RLFs) for different categories of network devices based on device location (e.g., location of the device within cell coverage of a particular tracking area identifier) or device velocity. In another example, use cases for improving model trustworthiness may include generating respective globalized machine learning models having low, medium, and high levels of explainability for devices located in different cell coverages.

[0118] In some embodiments, at block 506, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for determining a plurality of personalized tasks based on the intent and / or use case of block 503, where the plurality of personalized task includes at least a first personalized task and a second personalized task. In some embodiments, the apparatus 200 determines the plurality of personalized tasks based on one or more historical personalized tasks. In some embodiments, the apparatus 200 determines the plurality of personalized tasks based on local environment data and / or other properties or statuses of respective candidate federated learning distributed nodes(FLDNs) (e.g., where such information may be provided by the FLDN or one or more network analytic functions). In some embodiments, the apparatus 200 determines one or more personalized tasks based on one or more modifications to personalized tasks obtained at block 512. For example, the apparatus 200 may remove one or more personalized tasks based on a filtered subset of the plurality of tasks received from an operating entity. In another example, the apparatus 200 may add one or more personalized tasks received from or indicated by an operating entity. In some embodiments, the apparatus 200 determines one or more additional personalized tasks based on data obtained during one or more model training iterations. For example, the apparatus 200 may determine one or more additional personalized tasks based on detected changes to local environment data and / or other properties or statuses of an FLDN (e.g., indications thereof being received from the FLDN or one or more network analytics functions).

[0119] In some embodiments, at block 509, the apparatus performing the process 500 optionally includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for providing a notification indicative of the plurality of personalized tasks to the computing device of the operating entity. In some embodiments, the notification includes a listing of personalized tasks. In some embodiments, the notification includes metadata associated with determination of the personalized tasks including use cases, one or more categories of FLDNs, local environment data associated with one or more FLDNs, one or more properties or statuses of respective FLDNs, and / or the like.

[0120] In some embodiments, at block 512, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for receiving one or more modifications to the plurality of personalized tasks from the computing device of the operator entity. For example, the apparatus 200 may receive, from a computing device 101, a filtered subset of the plurality of personalized tasks, an activation or deactivation of one or more personalized tasks (e.g., to initiate, suspend, or resume FMTL training processes associated therewith), or an indication of one or more additional personalized tasks for inclusion in FMTL training.

[0121] In some embodiments, at block 515, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for determining associations between candidate FLDNs and each personalized task. In some embodiments, the apparatus 200 determines the associations based on local environment dataand / or other properties or statuses of respective candidate FLDNs. In some embodiments, the apparatus 200 adjusts associations of participant FLDNs based on changes to local environment data and / or other properties or statuses of the FLDN. For example, the apparatus 200 may recategorize an FLDN from a first personalized task to a second personalized task based on a change in FLDN velocity or TAI association. In some embodiments, the apparatus 200 adjusts associations of participant FLDNs based on average performance scores determined at block 530. For example, the apparatus 200 may remove an association between a participant FLDN and a personalized task based on determining that an average performance score of the participant FLDN fails to meet a predetermined threshold. In some embodiments, the apparatus 200 determines a participant FLDN for the personalized task based on historical training scores associated with the FLDN. In some embodiments, the apparatus 200 determines a participant FLDN for the personalized task in response to receiving a request from a candidate FLDN to participate in the FMTL training.

[0122] In some embodiments, at block 518, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for generating a respective personalized global machine learning model for each of the plurality of personalized tasks. For example, apparatus 200 may generate, or retrieve from memory 204, a first personalized global machine learning model for performing a first personalized task and a second personalized global machine learning model for performing a second personalized task.

[0123] In some embodiments, at block 521, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for providing a respective personalized global machine learning model (e.g., including an initial or subsequent trained iteration thereof) to a corresponding set of FLDNs for training. At or following block 521, the respective FLDNs independently and locally train the corresponding personalized machine learning model to perform the corresponding personalized task, where the personalized machine learning model may be an initially received version of the personalized machine learning model or a subsequent trained iteration of the personalized machine learning model.

[0124] In some embodiments, at block 524, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for receiving a respective trained personalized local machine learning model, and / or data associated therewith, from corresponding set of FLDNs. In some embodiments, the means for receiving thetrained personalized local machine learning model includes means for receiving local model parameters that define a respective personalized local machine learning model trained by participant FLDNs. In some embodiments, the apparatus performing the process 500 includes means, such as the memory 204, for storing one or more versions of a personalized global machine learning model, including personalized local machine learning models associated with training of the personalized global machine learning model. For example, in one or more iterations of FMTL training, the apparatus 200 may store received local model parameters in memory 204, which may include a data store accessible to other apparatuses performing FMTL training processes (e.g., other FLA computing environments).

[0125] In some embodiments, at block 527, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for receiving one or more training scores for the respective personalized local machine learning models. In one example, a first set of FLDNs trains a first personalized global machine learning model for performing a first personalized task and a second set of FLDNs trains a second personalized global machine learning model for performing a second personalized task. The apparatus 200 may receive, from respective FLDNs of the first set and the second set, a trustworthiness score and a performance score for associated with the locally trained version of the personalized global machine learning model (e.g., also referred to as a personalized local machine learning model). In some embodiments, the apparatus performing the process 500 includes means, such as memory 204, for storing the training scores. For example, the apparatus 200 may store training scores in memory 204 in association with stored versions of a corresponding personalized global machine learning model, thereby providing a time series record of personalized global machine learning model versions and corresponding training scores. In some embodiments, block 527 occurs in each iteration of FMTL training or following a predetermined quantity of model training iterations (e.g., 3, 5, or any suitable quantity of model training iterations). In some embodiments, block 527 occurs in response to participant FLDN and / or the apparatus performing the process 500 determining that a personalized global machine learning model has achieved convergence (e.g., model error settles within an error range and / or model performance approaches a threshold value or limit).

[0126] In some embodiments, at block 530, the apparatus performing the process 500 further includes means, such as the processor 202, the memory 204, the network interface 206, or the like,for generating an average training score for each personalized task by averaging the individual training scores reported by each corresponding FLDN. For example, the apparatus 200 may generate an average model performance score and an average model trustworthiness score for each personalized task by averaging the individual model performance and individual model trustworthiness measures reported by each corresponding FLDN.

[0127] In some embodiments, at block 530, the apparatus performing the process 500 further includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for generating a respective average performance score for the plurality of participating FLDNs according to Equation 1. In some embodiments, the average performance scores for the FLDNs are used at block 515 to modify participant FLDNs. For example, at block 515, the apparatus performing the process 500 may remove or recategorize an FLDN based on a comparison of the performance score of the FLDN to a minimum performance threshold.

[0128] In some embodiments, the apparatus performing the process 500 utilizes the average performance scores for participant FLDN as an input determining associations between personalized tasks and FLDNs at block 515. For example, at block 515, the apparatus 200 may compare the average performance score of a participant FLDN for a first personalized task to a predetermined minimum performance threshold. In response to determining the average performance score fails to satisfy the minimum performance threshold, the apparatus 200 may exclude the FLDN as a participant in further iterations of FMTL training for the first personalized task, and potentially associate the FLDN with FMTL training for a second personalized task.

[0129] In some embodiments, at block 533, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for generating a subsequent iteration of the personalized global machine learning models. In some embodiments, the apparatus performing the process 500 includes means for generating a subsequent iteration of a personalized global machine learning model by aggregating the local model parameters of personalized local machine learning models for a respective personalized task (e.g., as reported to the apparatus by corresponding participant FLDNs). The subsequent iteration of the personalized global machine learning model may be provided to corresponding participant FLDNs for further FMTL training. For example, following block 533, the process 500 may proceed to a perform a subsequent iteration of block 521 including providing the subsequentiteration of respective personalized global machine learning models to corresponding sets of participant FLDNs.

[0130] In some embodiments, at block 536, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for providing a report to an operating entity, such as by providing the report to a computing device associated with the operating entity. In some embodiments, the report includes one or more average training scores for a final iteration of the respective personalized global machine learning model for each personalized task. For example, the report may include an average performance score and an average trustworthiness score for a final iteration of a personalized global machine learning model. In some embodiments, the report includes data associated with participant FLDNs including average performance scores, quantity of participant FLDNs, category of participant FLDNs, and / or the like.

[0131] Referring now to FIG 6., shown is an example flowchart of a federated multi-task learning (FMTL) training process 600, which may be performed by an apparatus, such as the apparatus 200 of FIG. 2. The apparatus 200 may embody the FLDN 105 shown in FIG. 1, including means for performing various functions and operations as described herein. For example, the apparatus performing the process 600 may be a user equipment configured to perform the various functions and operations of respective FLDNs described herein.

[0132] In some embodiments, at block 603, the apparatus performing the process 600 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for storing training data for performing local training of machine learning models. For example, the FLDN performing the process 600 may obtain and store training data for training machine learning models to generate optimal machine learning model architectures for various use cases, such as improving network stability, energy conversation, mobility handover, and / or network performance for various categories of network devices (e.g., velocity -based categories, cell coverage area-based categories, and / or the like). In some embodiments, the training data includes measurements for network device positions, channel state information (CSI) data, allocated power, and / or the like.

[0133] In some embodiments, at block 606, the apparatus performing the process 600 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for receiving a request to provide local environment data to a federated learning aggregation (FLA) computing environment. In one example, the FLDN may receive from the FLA computingenvironment 103 a request to report resource-related information and training data-related information. In some embodiments, the request indicates one or more reporting parameters that indicate requested information. For example, the reporting parameters may instruct the FLDN to report on computational resource availability, power availability, channel quality, data quality, data trustworthiness, and / or the like. In some embodiments, the request includes reporting parameters for other properties or statuses of the FLDN. For example, the request may include reporting parameters instructing the FLDN to report on location information, velocity information, trust score, historical information, quantity of available training samples, and / or the like.

[0134] In some embodiments, at block 609, the apparatus performing the process 600 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for providing local environment data (and potentially other requested information) to the FLA computing environment. For example, the FLDN may provide to the FLA computing environment 103 an indication of computational resource availability, power availability, channel quality, data quality, data trustworthiness, and / or the like. In another example, the FLDN may provide a current velocity, a location (e.g., association with a TAI), a quantity of local training data samples, and / or the like. In some embodiments, based on the received information, the FLA computing environment 103 determines one or more personalized tasks and an association between the FLDN and the personalized task for purposes of FMTL training.

[0135] In some embodiments, at block 612, the apparatus performing the process 600 optionally includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for receiving a notification of one or more personalized tasks for which the apparatus may participate in FMTL training. For example, the FLDN may receive a notification indicative of a plurality of personalized tasks from the FLA computing environment.

[0136] In some embodiments, at block 615, the apparatus performing the process 500 optionally includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for providing to the FLA computing environment a request for association with a personalized task. For example, the FLDN may provide to the FLA computing environment a request to associate the FLDN with one of a plurality of personalized tasks reported to the FLDN, where association with the personalized task includes the FLDN participating in FMTL training of a personalized global machine learning model for performing the task.

[0137] In some embodiments, at block 618, the apparatus performing the process 600 optionally includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for providing to the FLA computing environment a request for disassociation from a personalized task with which the apparatus is currently associated. For example, the FLDN may provide to the FLA computing environment a request to disassociate the FLDN from a currently associated personalized task, where disassociation may result in the FLDN being excluded as a participant in further FMTL training for the personalized task. In some embodiments, the request for disassociation includes one or more disassociation factors including a computational overload, a bandwidth limitation, a power limitation, and / or the like. In some embodiments, the apparatus performing the process 500 optionally includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for generating the request for disassociation and one or more disassociation factors based on monitoring or predicting changes to one or more properties, statuses, or other local environment data associated with the apparatus. For example, the FLDN may determine that a computational load, bandwidth, and / or power level meets a predetermined threshold, where such determination may cause the FLDN to generate and provide a request for disassociation from one or more personalized tasks to the FLA computing environment. In some embodiments, the request for disassociation indicates a duration of unavailability for the apparatus performing the process 600. The duration of unavailability may be generated by the apparatus performing the process 600 based on local environment data or other properties or status of the apparatus.

[0138] In various embodiments, following one or more of blocks 609-615, the FLA computing environment associates the apparatus performing the process 600 with a personalized task such that the apparatus participates in one or more model training iterations of an FMTL training process for training a personalized global machine learning model to perform the personalized task.

[0139] In some embodiments, at block 621, the apparatus performing the process 600 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for receiving an initial version of a personalized global machine learning model for performing the personalized task, or a subsequent iteration of the personalized global machine learning model following one or more model training iterations. For example, the FLDN receives from the FLA computing environment a personalized global machine learning model.

[0140] In some embodiments, at block 624, the apparatus performing the process 600 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for training the initial version or a subsequent iteration of the personalized global machine learning model to perform the corresponding personalized task. For example, the FLDN may use local training data including a plurality of local training samples to locally and independently train the personalized global machine learning model to perform the personalized task. In some embodiments, at block 624, the apparatus generates a personalized local machine learning model based on the personalized global machine learning model. In some embodiments, the apparatus generates local model parameters that the define the personalized local machine learning model. In some embodiments, the apparatus performing the process 600 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for generating synthetic data and generating additional local training samples based on the synthetic data to improve local model training by using the additional synthetic training samples.

[0141] In some embodiments, at block 627, the apparatus performing the process 600 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for providing the personalized local machine learning model to the FLA computing environment, where the FLA computing environment may generate a subsequent iteration of the personalized global machine learning model based on the personalized local machine learning models received from respective participant FLDN. For example, the FLDN provides local model parameters obtained via local training to the FLA computing environment. In some embodiments, the apparatus reports metadata associated with the local training, such as a quantity of training samples.

[0142] In some embodiments, at block 630, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, for generating one or more training scores for the iteration of the personalized global machine learning model trained by the apparatus. In some embodiments, the training score includes a performance score, a trustworthiness score, and / or the like. For example, during each iteration or a subset of iterations of the FMTL training process, the FLDN generates a performance score and a trustworthiness score for the trained model.

[0143] In some embodiments, at block 633, the apparatus performing the process 500 includes means, such as the processor 202, the memory 204, the network interface 206, or the like, forproviding the one or more training scores to the FLA computing environment. For example, in each iteration or a subset of iterations of the FMTL training process, the FLDN provides to the FLA computing environment a performance score and a trustworthiness score for a current trained iteration of the personalized global machine learning model (e.g., which may be a personalized local machine learning model at the FLDN).

[0144] In various embodiments, the method, apparatus and computer program product of the present disclosure provide management mechanisms for performing federated multi-task learning (FMTL) training for machine learning models. The method, apparatus, and computer program product provide improved solutions for integrating FMTL into training of machine learning models for network automation. In various embodiments, the method, apparatus, and computer program product improve network automation by overcoming technical challenges associated achieving high statistical accuracy and model trustworthiness in machine learning models that perform network management functions.

[0145] It will be understood that each block of the flowcharts and combination of blocks in the flowcharts show in the figures and described herein may be implemented by various means, such as hardware, firmware, processor, circuitry, and / or communication devices associated with execution of software including one or more program instructions. For example, one or more of the procedures or operations described above may be embodied by computer program instructions. In this regard, the computer program instructions which embody the procedures or operations described above may be stored by a memory 204 of an apparatus (e.g., user equipment (UE)) employing a disclosed embodiment and executed by a processor 202. As will be appreciated, any such computer program instructions may be loaded onto a computer or other programmable apparatus (for example, hardware) to produce a machine, such that the resulting computer or other programmable apparatus implements the functions specified in the flowchart blocks. These computer program instructions may also be stored in a computer-readable memory that may direct a computer or other programmable apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture the execution of which implements the function specified the flowchart blocks. The computer program instructions may also be loaded onto a computer or other programmable apparatus to cause a series of operations to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or otherprogrammable apparatus provide operations for implementing the functions specified in the flowchart blocks.

[0146] Many modifications and other embodiments of the present disclosure set forth herein will come to mind to one skilled in the art to which these embodiments pertain having the benefit of the teachings presented in the foregoing descriptions and the associated drawings. Therefore, it is to be understood that the disclosure is not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of the appended claims. Moreover, although the foregoing descriptions and the associated drawings describe example embodiments in the context of certain example combinations of elements and / or functions, it should be appreciated that different combinations of elements and / or functions can be provided by alternative embodiments without departing from the scope of the appended claims. In this regard, for example, different combinations of elements and / or functions than those explicitly described above are also contemplated as can be set forth in some of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

Claims

CLAIMS1. A method (500) of network device of a federated learning aggregation (FLA) computing environment of a communication network, the method comprising: determining (503), for at least a first personalized task and a second personalized task of a plurality of personalized tasks for performing network automation operations for the communication network , (i) a first set of federated learning distributed nodes (FLDNs) (105) from a plurality of FLDNs (105) associated with the first personalized task and (ii) a second set of FLDNs (105) from the plurality of FLDNs (105) associated with the second personalized task; generating (518) a first personalized machine learning (ML) model for performing the first personalized task and a second personalized ML model for performing the second personalized task; and performing (521) a respective plurality of model training iterations for training the first personalized ML model and the second personalized ML model, the respective plurality of model training iterations comprising: providing (521), in a respective initial model training iteration for the first personalized ML model and the second personalized ML model, (i) the first personalized ML model to respective FLDNs (105) of the first set and (ii) the second personalized ML model to respective FLDNs (105) of the second set; providing (521), in respective subsequent model training iterations for the first personalized ML model and the second personalized ML model, (i) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the first set and (ii) a subsequent trained iteration of the first personalized ML model to respective FLDNs (105) of the second set, wherein: in a respective model training iteration, respective FLDNs (105) of the first set and respective FLDNs (105) of the second set independently train the corresponding personalized ML model to perform the corresponding personalized task; receiving (524), in a respective model training iteration, local model parameters from the respective FLDNs (105), wherein the local model parameters are associatedwith the training of the corresponding personalized ML model by the respective FLDNs (105) of the first set or the second set; and generating (530) (i) the subsequent trained iteration of the first personalized ML model based on an aggregation of the local model parameters associated with the respective FLDNs (105) of the first set; and (ii) the subsequent trained iteration of the second personalized ML model based on an aggregation of the local model parameters associated with respective FLDNs (105) of the second set, wherein: an output of a final iteration of each of the respective plurality of training iterations comprises a first personalized global ML model for performing the first personalized task and a second personalized global ML model for performing the second personalized task.

2. The method as claimed in claim 1 , wherein: the plurality of FLDNs (105) comprise one or more individual management domains; and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a cross-domain management domain (103).

3. The method as claimed in claim 1, wherein: the plurality of FLDNs (105) comprise one or more network nodes; and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a radio access network (RAN) management domain (103).

4. The method as claimed in claim 1 , wherein: the plurality of FLDNs (105) comprise one or more network data analytic functions (NWDAFs); and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and theperforming the respective plurality of training iterations are performed using a core management domain (103).

5. The method as claimed in claim 1, wherein: the plurality of FLDNs (105) comprise one or more distributed NWDAFs; and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a central NWDAF (103).

6. The method as claimed in claim 1 , wherein: the plurality of FLDNs (105) comprise one or more user equipment (UE); and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a network node (103).

7. The method as claimed in claim 1 , wherein: the plurality of FLDNs (105) comprise one or more real-time RAN intelligent controllers (RICs); and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a non-real-time RIC (103).

8. The method as claimed in claim 1, wherein: the plurality of FLDNs (105) comprise one or more network nodes; and the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a real-time RIC (103).

9. The method as claimed in claim 1 , wherein: the determining the first set and the second set of the plurality of FLDNs (105), the generating the first personalized ML model and the second personalized ML model, and the performing the respective plurality of training iterations are performed using a first UE (103); and the plurality of FLDNs (105) comprise one or more additional UEs.

10. The method as claimed in any of claims 1 to 9, further comprising, in a respective model training iteration: receiving (527), from respective FLDNs (105) of the first set and respective FLDNs (105) of the second set, at least one training score associated with training of the corresponding personalized ML model by the FLDNs (105); and generating (530) a respective aggregated training score for the first personalized ML model and the second personalized ML model based on the training scores received from the corresponding FLDNs (105).

11. The method as claimed in claim 10, further comprising: providing (536) the at least one aggregated training score associated with the first personalized ML model and the at least one aggregated score associated with the second personalized ML model to a computing device associated with an operator entity.

12. The method as claimed in claim 11, wherein: a respective training score of the at least one training score comprises a performance score for performing the corresponding personalized task; and a respective training score of the at least one aggregated training score comprises an aggregated performance score for performing the corresponding personalized task.

13. The method as claimed in claim 12, wherein: the respective training score further comprises a trustworthiness score for performing the corresponding personalized task; andthe respective aggregated training score further comprises an aggregated trustworthiness score for performing the corresponding personalized task.

14. The method as claimed in any of claims 11 to 13, further comprising: determining (506, 512) the plurality of personalized tasks based on a user input received from the computing device associated with the operator entity, wherein the user input indicates at least one machine learning model use case.

15. The method as claimed in claim 14, further comprising: receiving (506), from a data store, a plurality of historical personalized tasks based on the at least one machine learning model use case, wherein the plurality of personalized tasks are determined based on the plurality of historical personalized tasks.

16. The method as claimed in claim 15, further comprising: generating (515) a mapping of the at least one machine learning use case to subsets of the plurality of historical personalized tasks; and generating (515), based on the mapping, at least (i) a first category comprising the first subset of the plurality of FLDNs (105) and (ii) a second category comprising the second subset of the plurality of FLDNs (105), wherein the plurality of personalized tasks are further determined based on the first category and the second category.

17. The method as claimed in claim 15, further comprising: providing (509) the plurality of historical personalized tasks to the computing device associated with the operator entity; receiving (512) a second user input from the computing device associated with the operator entity, wherein the second user input indicates at least one of a filtered subset of the plurality of historical personalized tasks or at least one additional personalized task; and determining (506) the plurality of personalized tasks further based on at least one of the filtered subset of the plurality of historical personalized tasks or the at least one additional personalized task.

18. The method as claimed in any of claims 1 to 16, further comprising, between one or more model training iterations of the plurality of model training iterations: determining (506) a third personalized task; determining (515) a third set of the plurality of FLDNs (105) associated with the third personalized task; generating (518) a third personalized machine learning model to perform the third personalized task; and providing (521) the third personalized machine learning model or a subsequent trained iteration of the third personalized machine learning model to the third set of FLDNs (105), wherein: respective FLDNs (105) of the third set train the third personalized machine learning model or the subsequent trained iteration of the third personalized machine learning model in a remaining subset of the plurality of model iterations; and in a final model training iteration, a personalized global machine learning model for the third personalized task is generated based on a corresponding plurality of local model parameters received from respective FLDNs (105) of the third set during the remaining subset of the plurality of model training iterations.

19. The method as claimed in claim 18, wherein: one or more of the first set of the plurality of FLDNs (105), the second set of the plurality of FLDNs (105), or the third set of the plurality of FLDNs (105) are determined based on one or more properties or statuses of the plurality of FLDNs (105).

20. The method as claimed in claim 17 or claim 18, further comprising: removing (506) at least one FLDN from the first set of the plurality of FLDNs (105) based on a change to the one or more properties or statuses of the at least one FLDN; and adding (506) the at least one FLDN to the second set of the plurality of FLDNs (105) or the third set of the plurality of FLDNs (105) based on the change to the one or more properties or statuses of the at least one FLDN.

21. The method as claimed in claim 19, wherein: the one or more properties or statuses of the plurality of FLDNs (105) comprises at least one of a computational resource availability, a power availability, a channel quality, a data quality score, or a data trustworthiness score.

22. The method as claimed in claim 19 or claim 20, wherein: the one or more properties or statuses of the plurality of FLDNs (105) comprises at least one of location data, velocity data, a trust score, or a quantity of available training data samples.

23. The method as claimed in any of claims 1 to 21, further comprising: providing (515) a notification indicative of the plurality of personalized tasks to the plurality of FLDNs (105); receiving (512), from at least one FLDN of the plurality of FLDNs (105), a request to associate the at least one FLDN with the first personalized task or the second personalized task; and determining (515) at least one of the first set of the plurality of FLDNs (105) or the second set of the plurality of FLDNs (105) based on the request from the at least one FLDN.

24. The method as claimed in any of claims 1 to 22, further comprising: receiving (512), from at least one FLDN, a request for disassociation from the first set of the plurality of FLDNs (105) or the second set of the plurality of FLDNs (105); determining (506) that at least one training score associated with the at least one FLDN and a personalized machine learning model trained by the at least one FLDN meets a predetermined threshold; and in response to the at least one training score meeting the predetermined threshold, providing (506), to the at least one FLDN, a request to retain the at least one FLDN in the corresponding set of the plurality of FLDNs (105) for the plurality of training iterations, wherein the request to retain the at least one FLDN indicates a monetary retention incentive.

25. The method as claimed in claim 24, wherein: the request for disassociation comprises at least one disassociation factor; andthe at least one disassociation factor is at least one of a computational overload, a bandwidth limitation, or a power limitation.

26. An apparatus for a federated learning aggregation (FLA) computing environment of a communication network, the apparatus comprising means for performing the method as claimed in any of claims 1 to 25.

27. A computer program comprising instructions, which, when executed by an apparatus for a federated learning aggregation (FLA) computing environment of a communication network, cause the apparatus to perform the method as claimed in any of claims 1 to 25.

Images