Methods and apparatus of machine learning training client selection in a communication system

Abstract

The disclosure relates to a 5G or 6G communication system for supporting higher data transmission rates and satisfying various service requirements. The disclosure provides a method performed by an FL server in a communication system. The method may include: receiving, from a consumer for a machine learning (ML) training, a request for the ML training including information on a requirement for an FL, wherein the requirement for the FL includes information on FL client selection criteria for the FL; selecting at least one FL client based on the information on the FL client selection criteria; generating an FL report for the at least one FL client based on a result of the ML training for the at least one FL client; and transmitting, to the consumer, an ML model training report including the FL report.

Description

METHODS AND APPARATUS OF MACHINE LEARNING TRAINING CLIENT SELECTION IN A COMMUNICATION SYSTEM

[0001] The disclosure relates to the field of artificial intelligence (AI) and machine learning (ML). More specifically, the disclosure relates to a method and apparatus of machine learning training client selection.

[0002] 5G mobile communication technologies define broad frequency bands such that high transmission rates and new services are possible, and can be implemented not only in "Sub 6GHz" bands such as 3.5GHz, but also in "Above 6GHz" bands referred to as mmWave including 28GHz and 39GHz. In addition, it has been considered to implement 6G mobile communication technologies (referred to as Beyond 5G systems) in terahertz bands (for example, 95GHz to 3THz bands) in order to accomplish transmission rates fifty times faster than 5G mobile communication technologies and ultra-low latencies one-tenth of 5G mobile communication technologies.

[0003] At the beginning of the development of 5G mobile communication technologies, in order to support services and to satisfy performance requirements in connection with enhanced Mobile BroadBand (eMBB), Ultra Reliable Low Latency Communications (URLLC), and massive Machine-Type Communications (mMTC), there has been ongoing standardization regarding beamforming and massive MIMO for mitigating radio-wave path loss and increasing radio-wave transmission distances in mmWave, supporting numerologies (for example, operating multiple subcarrier spacings) for efficiently utilizing mmWave resources and dynamic operation of slot formats, initial access technologies for supporting multi-beam transmission and broadbands, definition and operation of BWP (BandWidth Part), new channel coding methods such as a LDPC (Low Density Parity Check) code for large amount of data transmission and a polar code for highly reliable transmission of control information, L2 pre-processing, and network slicing for providing a dedicated network specialized to a specific service.

[0004] Currently, there are ongoing discussions regarding improvement and performance enhancement of initial 5G mobile communication technologies in view of services to be supported by 5G mobile communication technologies, and there has been physical layer standardization regarding technologies such as V2X (Vehicle-to-everything) for aiding driving determination by autonomous vehicles based on information regarding positions and states of vehicles transmitted by the vehicles and for enhancing user convenience, NR-U (New Radio Unlicensed) aimed at system operations conforming to various regulation-related requirements in unlicensed bands, NR UE Power Saving, Non-Terrestrial Network (NTN) which is UE-satellite direct communication for providing coverage in an area in which communication with terrestrial networks is unavailable, and positioning.

[0005] Moreover, there has been ongoing standardization in air interface architecture / protocol regarding technologies such as Industrial Internet of Things (IIoT) for supporting new services through interworking and convergence with other industries, IAB (Integrated Access and Backhaul) for providing a node for network service area expansion by supporting a wireless backhaul link and an access link in an integrated manner, mobility enhancement including conditional handover and DAPS (Dual Active Protocol Stack) handover, and two-step random access for simplifying random access procedures (2-step RACH for NR). There also has been ongoing standardization in system architecture / service regarding a 5G baseline architecture (for example, service based architecture or service based interface) for combining Network Functions Virtualization (NFV) and Software-Defined Networking (SDN) technologies, and Mobile Edge Computing (MEC) for receiving services based on UE positions.

[0006] As 5G mobile communication systems are commercialized, connected devices that have been exponentially increasing will be connected to communication networks, and it is accordingly expected that enhanced functions and performances of 5G mobile communication systems and integrated operations of connected devices will be necessary. To this end, new research is scheduled in connection with eXtended Reality (XR) for efficiently supporting AR (Augmented Reality), VR (Virtual Reality), MR (Mixed Reality) and the like, 5G performance improvement and complexity reduction by utilizing Artificial Intelligence (AI) and Machine Learning (ML), AI service support, metaverse service support, and drone communication.

[0007] Furthermore, such development of 5G mobile communication systems will serve as a basis for developing not only new waveforms for providing coverage in terahertz bands of 6G mobile communication technologies, multi-antenna transmission technologies such as Full Dimensional MIMO (FD-MIMO), array antennas and large-scale antennas, metamaterial-based lenses and antennas for improving coverage of terahertz band signals, high-dimensional space multiplexing technology using OAM (Orbital Angular Momentum), and RIS (Reconfigurable Intelligent Surface), but also full-duplex technology for increasing frequency efficiency of 6G mobile communication technologies and improving system networks, AI-based communication technology for implementing system optimization by utilizing satellites and AI (Artificial Intelligence) from the design stage and internalizing end-to-end AI support functions, and next-generation distributed computing technology for implementing services at levels of complexity exceeding the limit of UE operation capability by utilizing ultra-high-performance communication and computing resources.

[0008] The AI / ML techniques and relevant applications are being increasingly adopted by the wider industries and proved to be successful. These are now being applied to the telecommunication industry, including mobile networks. Although AI / ML techniques in general are quite mature nowadays, some of the relevant aspects of the technology are still evolving while new complementary techniques are frequently emerging. The learning methods include supervised learning, semi-supervised learning, unsupervised learning, and reinforcement learning. Each learning method fits one or more specific categories of inference (e.g., prediction) and requires a specific type of training data.

[0009] The lifecycle management of the AI / ML model is being defined in the 3GPP SA5 working group. The lifecycle stages include ML model training, ML testing, ML emulation, ML entity loading, and the inference phase. ML model training includes initial training and re-training of an ML model or a group of ML models. It also includes validation of the ML entity to evaluate the performance when the ML entity performs on the training data and validation data. If the validation result does not meet the expectation (e.g., the variance is not acceptable), the ML model associated with that entity needs to be re-trained. The ML model training is the initial phase of the workflow. ML testing includes testing the validated ML entity to evaluate the performance of the trained ML model when it performs on testing data. If the testing result meets the expectation, the ML entity may proceed to the next phase; otherwise, the ML model associated with that entity may need to be re-trained. ML emulation includes running an ML entity for inference in an emulation environment. The purpose is to evaluate the inference performance of the ML entity in the emulation environment prior to applying it to the target network or system. ML entity loading is the process (aka a sequence of atomic actions) of making a trained ML entity available for use at the target AI / ML inference function. AI / ML inference includes performing inference using a trained ML entity by the AI / ML inference function.

[0010] Federated Learning (FL) is a distributed machine learning approach that allows multiple FL clients to collaboratively train an ML model on local datasets contained in each FL client without explicitly exchanging data samples. A group of FL clients supports FL and FL server, wherein the FL client keeps the data localized and private and trains the ML model directly on the local nodes (client) where the data is generated or stored. For managing the FL, the ML training MnS consumer needs to know the FL clients and FL server included in the FL so that the consumer understands the impact of each one of them and can manage it correspondingly.

[0011] It is important to note that while the use of AI / ML techniques can lead to significant improvements in network performance, these benefits may come at the cost of increased energy consumption, especially increased non-renewable energy consumption and carbon emission. AI / ML-based solutions, due to their complex computational requirements, may consume additional energy compared to traditional non-AI / ML solutions. So, in order to have a sustainable AI / ML approach, it is important to govern energy type usage, green energy usage, carbon emission control, etc., in the overall AI / ML system, including the ML model training step.

[0012] When receiving an ML training request, the ML training MnS producer should evaluate whether the FL process needs to be started according to the training requirements provided by the ML training consumer. Based on the received requirements, the FL server may select appropriate FL clients. FL client selection is a crucial component of FL that directly impacts the performance, efficiency, and fairness of the learning process. However, selecting the optimal set of clients to participate in each round of FL poses several challenges. While some of the requirements are studied in 3GPP to select an FL client, the existing work is not exhaustive enough, especially when it comes to selecting FL clients based on criteria related to green energy, renewable energy usage, carbon emission, fault tolerance, etc. So, it is desirable to consider such aspects also while selecting FL clients so that a sustainable FL-based ML approach can be maintained and the expected performance of the final aggregated ML model can be ensured accordingly.

[0013] Thus, it is desired to address the above-mentioned disadvantages or other shortcomings or at least provide a useful alternative.

[0014] The disclosure relates to a method performed by at least one of a consumer, a producer, a server, a client in a communication system and a device for supporting ML training with federated learning (FL).

[0015] Accordingly, an aspect of the disclosure is to provide a method and apparatus for machine learning training client selection. The disclosure trains a ML model utilizing federated learning.

[0016] Another aspect of the disclosure is to introduce a 'is renewable source availability attribute' for ensuring renewable energy source-based FL client selection in the FL approach.

[0017] Another aspect of the disclosure is to introduce a renewable energy information attribute for ensuring a certain percentage of renewable energy usage in the FL client.

[0018] Another aspect of the disclosure is to introduce a renewable source type attribute for ensuring a particular type of renewable energy source usage by the FL client.

[0019] Another aspect of the disclosure is to introduce a carbon emission information attribute for ensuring reduced carbon emissions during the FL-based ML approach by selecting FL clients which have carbon emissions below a certain amount.

[0020] Another aspect of the disclosure is to introduce an FL client redundancy attribute for ensuring an FL client which has some redundancy for more reliability.

[0021] The technical problems to be achieved in the various examples of the disclosure are not limited to those mentioned above, and other technical problems not mentioned can be considered by a person having ordinary skill in the art from the various examples of the disclosure described below.

[0022] According to an aspect of the disclosure, a method performed by an FL server in a communication system is provided. The method may include: receiving, from a consumer for a machine learning (ML) training, a request for the ML training including information on a requirement for an FL, wherein the requirement for the FL includes information on FL client selection criteria for the FL; selecting at least one FL client based on the information on the FL client selection criteria; generating an FL report for the at least one FL client based on a result of the ML training for the at least one FL client; and transmitting, to the consumer, an ML model training report including the FL report.

[0023] According to an aspect of the disclosure, a method performed by a consumer in a communication system is provided. The method may include: generating information on a requirement for a federated learning (FL), wherein the requirement for the FL includes information on FL client selection criteria for the FL; transmitting, to an FL server, a request for a machine learning (ML) training including the information on the requirement for the FL; and receiving, from the FL server, an ML model training report including an FL report for at least one FL client associated with the FL client selection criteria, wherein the FL report is associated with a result of the ML training for the at least one FL client.

[0024] According to an aspect of the disclosure, a FL server in a communication system is provided. The FL server may include: a transceiver; a processor coupled to the transceiver; and memory coupled to the processor and storing instructions executable by the processor, wherein the instructions cause the FL server to: receive, from a consumer for a machine learning (ML) training, a request for the ML training including information on a requirement for an FL, wherein the requirement for the FL includes information on FL client selection criteria for the FL, select at least one FL client based on the information on the FL client selection criteria, generate an FL report for the at least one FL client based on a result of the ML training for the at least one FL client, and transmit, to the consumer, an ML model training report including the FL report.

[0025] According to an aspect of the disclosure, a consumer in a communication system is provided. The consumer may include: a transceiver; a processor coupled to the transceiver; and memory coupled to the processor and storing instructions executable by the processor, wherein the instructions cause the consumer to: generate information on a requirement for a federated learning (FL), wherein the requirement for the FL includes information on FL client selection criteria for the FL, transmit, to an FL server, a request for a machine learning (ML) training including the information on the requirement for the FL, and receive, from the FL server, an ML model training report including an FL report for at least one FL client associated with the FL client selection criteria, and wherein the FL report is associated with a result of the ML training for the at least one FL client.

[0026] In an aspect, the objects are achieved by providing a method for machine learning training client selection. The method may include receiving by a Management Services (MnS) producer in an FL server an ML training request message for training the ML model using FL from an MnS consumer device, where the ML training request message includes an FL client selection criteria for selecting FL clients from a plurality of FL clients for training the ML model, where the FL client selection criteria includes at least one of a 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and an FL client redundancy attribute. The method further includes sending by the MnS producer in the FL server a response message to the MnS consumer device indicating creation of a model object instance (MOI) and selecting by the MnS producer in the FL server a set of FL clients from the plurality of FL clients based on at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the FL client redundancy attribute.

[0027] In an aspect, the objects are achieved by providing a method for machine learning training client selection. The method may include generating by an MnS consumer device an FL client selection criteria for selecting FL clients for training the ML model using FL, where the FL client selection criteria includes at least one of a 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and a client redundancy attribute, and sending by the MnS consumer device an ML training request message for training the ML model to an MnS producer in an FL server, where the request message includes the FL client selection criteria for selecting FL clients for training of the ML model. The method further includes receiving by the MnS consumer device the response message from the MnS producer in the FL server indicating creation of a MOI and receiving by the MnS consumer device a notification message from the MnS producer in an FL server notifying the MnS consumer device about the availability of a consolidated ML training report in the FL server.

[0028] In an aspect, the objects are achieved by providing an FL server for machine learning training client selection. The FL server may include a memory including an MnS producer, a processor, and an FL client selection controller connected to the memory and the processor, where the FL client selection controller receives by the MnS producer in the FL server an ML training request message for training the ML model using FL from an MnS consumer device, where the ML training request message includes an FL client selection criteria for selecting FL clients from a plurality of FL clients for training the ML model, where the FL client selection criteria includes at least one of a 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and an FL client redundancy attribute. The FL client selection controller further sends by the MnS producer in the FL server the response message to the MnS consumer indicating creation of a MOI and selects by the MnS producer in the FL server a set of FL clients from the plurality of FL clients based on at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the FL client redundancy attribute.

[0029] In an aspect, the objects are achieved by providing an MnS consumer device for machine learning training client selection. The MnS consumer device may include a memory including an FL server, a processor, and an FL training request criteria controller connected to the memory and the processor. The FL training request criteria controller generates an FL client selection criteria for selecting FL clients for training the ML model using FL, where the FL client selection criteria includes at least one of a 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and a client redundancy attribute, sends an ML training request message for training the ML model to an MnS producer in an FL server, where the request message includes the FL client selection criteria for selecting FL clients for training of the ML model. The FL training request criteria controller further receives the response message from the MnS producer in the FL server indicating creation of a MOI and receives a notification message from the MnS producer in an FL server notifying the MnS consumer device about the availability of a consolidated ML training report in the FL server.

[0030] According to the examples of the disclosure, a sustainable FL based ML approach can be maintained and the expected performance of the final aggregated ML model can be ensured by selecting FL clients based on criteria related to green energy, renewable energy usage, carbon emission, fault tolerance etc.

[0031] The effects that can be obtained from the disclosure are not limited to the effects mentioned in the various examples, and other effects that are not mentioned can be clearly understood by a person having ordinary skill in the art to which the disclosure belongs from the description below.

[0032] The disclosure is illustrated in the accompanying drawings, throughout which like reference letters indicate corresponding parts in the various figures (FIGs). The examples herein will be better understood from the following description with reference to the drawings, in which:

[0033] Figure 1 is a block diagram that illustrates hardware components associated with the FL server according to examples as disclosed herein.

[0034] Figure 2 is a block diagram that illustrates hardware components associated with the MnS consumer device according to examples as disclosed herein.

[0035] Figure 3 is a flow chart that illustrates a proposed method, implemented by a FL server, for machine learning training client selection, according to examples as disclosed herein.

[0036] Figure 4 is a flow chart that illustrates a proposed method, implemented by an MnS consumer device, for training a ML model using FL according to examples as disclosed herein.

[0037] Figure 5 is a sequence diagram that illustrates an example scenario for training a ML model using FL according to examples as disclosed herein.

[0038] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. Also, the various embodiments described herein are not necessarily mutually exclusive, as some embodiments can be combined with one or more other embodiments to form new embodiments. The term "or" as used herein, refers to a non-exclusive or, unless otherwise indicated. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein can be practiced and to further enable those skilled in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.

[0039] As is existing in the field, embodiments can be described and illustrated in terms of blocks which carry out a described function or functions.　These blocks, which can be referred to herein as managers, units, modules, hardware components or the like, are physically implemented by analog and / or digital circuits such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, hardwired circuits and the like, and can optionally be driven by firmware and software. The circuits may, for example, be embodied in one or more semiconductor chips, or on substrate supports such as printed circuit boards and the like.　The circuits constituting a block can be implemented by dedicated hardware, or by a processor (e.g., one or more programmed microprocessors and associated circuitry), or by a combination of dedicated hardware to perform some functions of the block and a processor to perform other functions of the block.　Each block of the embodiments can be physically separated into two or more interacting and discrete blocks without departing from the scope of the disclosure.　Likewise, the blocks of the embodiments can be physically combined into more complex blocks without departing from the scope of the disclosure.

[0040] The accompanying drawings facilitate understanding of various technical features. The embodiments described are not limited by these drawings; the present disclosure includes any alterations, equivalents, and substitutes beyond those depicted. Terms like first, second, etc., are used for distinction and should not limit the elements they describe.

[0041] It is important to note that while the use of (artificial intelligence / machine learning) AI / ML techniques can lead to significant improvements in network performance, these benefits may come at the cost of increased energy consumption, especially increased non-renewable energy consumption and carbon emissions. The AI / ML-based solutions, due to their complex computational requirements, may consume additional energy compared to traditional non-AI / ML solutions. So, in order to have a sustainable AI / ML approach, it is important to govern energy type usage, green energy usage, carbon emission control, etc., in the overall AI / ML system, including the ML model training steps.

[0042] When receiving a ML training request, the ML training management service (MnS) producer should evaluate whether the federated learning (FL) process needs to be started according to the training requirements provided by the ML training consumer. Based on the received requirements, the FL server may select appropriate FL Clients. The FL client selection is a crucial component of FL that directly impacts the performance, efficiency, and fairness of the learning process. However, selecting the optimal set of clients to participate in each round of FL poses several challenges. While some of the requirements are studied in 3GPP to select an FL client, the existing work is not exhaustive enough, especially when it comes to selecting FL clients based on FL client selection criteria related to green energy, renewable energy usage, carbon emission, fault tolerance, etc. So, it is desirable to consider such aspects also while selecting FL clients so that a sustainable FL-based ML approach can be maintained and the expected performance of the final aggregated ML model can be ensured accordingly.

[0043] In an example, the disclosure includes an MnS consumer providing to the MnS producer in a FL server a set of requirements that includes the FL client selection criteria related to but not limited to renewable source availability attribute, a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and a FL client redundancy attribute, etc. to select FL clients for collaboratively training the ML model. Following the FL client selection criteria is provided as part of requirements to select FL clients:

[0044] - Is renewable source (IsRenewableSource) availability attribute: This defines that an FL client can be selected by an FL server depending upon if the client is using a renewable energy source. Its values can be TRUE or FALSE, where TRUE means the FL client must be using a renewable energy source, and FALSE means the energy source of the FL client may or may not be a renewable energy source.

[0045] - Renewable source type (RenewableSourceType) attribute: This defines that an FL client can be selected by an FL server only if it is using a particular type of renewable energy source (fully or partially). It can have values such as wind, solar, biogas, biofuel, aerothermal, geothermal, hydrothermal, and ocean energy, hydropower, biomass, landfill gas, sewage treatment plant gas, and biogases.

[0046] - RenewableEnergyInfo attribute: This defines a certain amount of renewable-sourced energy that at least the FL client should be using for its operation in order to be selected as an FL client. It can be defined either in terms of percentage or in terms of a ratio (e.g., ratio of renewable energy consumed to total energy consumed).

[0047] - Carbon emission information (CarbonEmissionInfo) attribute: This defines the maximum carbon emission amount below which the FL client should emit in order to be chosen by the FL server for training an ML model as an FL client. Its value can be in terms of kilograms of CO2 equivalent to consumed energy (in kWh or joules).

[0048] - FL client redundancy (ClientRedundancy) attribute: This defines that the FL client must have some type of redundancy for fault tolerance to handle client dropouts in order to be selected by the FL server to train an ML model. Its values can be TRUE or FALSE, where TRUE means the FL client must have some type of redundancy, and FALSE means the FL client may or may not have redundancy in order to be selected as an FL client.

[0049] The solution proposes following attributes to enhance network resource model (NRM) for AL / ML management in 3GPP TS 28.105:

[0050] The attributes can be present in MLTrainingRequest IOC. Alternatively, the attribute can be present in a new information object class (IOC) name contained in MLTrainingFunction. The examples of Attributes for AL / ML management are shown in Table 1 but not limited to the Table 1.

[0051] Attribute nameSupport QualifierisReadableisWritableisInvariantisNotifyableAttribute DescriptionAttribute PropertiesIsRenewableSourceMTTFTThis allows that a FL client can be selected by FL server depending upon, if the client is using a renewable energy source. Its values can be TRUE or FALSE, where TRUE means the FL client must be using a renewable energy source and FALSE means the energy source of FL client does not matter for its selection.type: Booleanmultiplicity: 1isOrdered: N / AisUnique: N / AdefaultValue: NoneisNullable: FalseRenewableSourceTypeMTTFTThis defines that a FL client can be selected by FL server only if it is using a particular type of renewable energy source (fully or partially). It can have values as wind, solar, biogas, biofuel, aerothermal, geothermal, hydrothermal and ocean energy, hydropower, biomass, landfill gas, sewage treatment plant gas and biogasestype: Stringmultiplicity: 1isOrdered: N / AisUnique: N / AdefaultValue: NoneisNullable: TrueRenewableEnergyInfoMTTFTThis defines a certain amount of renewable sourced energy, which at least, the FL client should be using for its operation in order to be selected as FL client. It can be defined either in terms of percentage or in terms of a ratio (e.g. ratio of renewable energy consumed by total energy consumed)type: Realmultiplicity: 1isOrdered: N / AisUnique: N / AdefaultValue: NoneisNullable: FalseCarbonEmissionInfoMTTFTThis defines the threshold carbon emission amount value, below which the FL client should emit in order to be chosen by FL server for training a ML model as FL client. Its value can in terms of kilogram of CO2 equivalent to consumed energy (in kWh or joule).type: Realmultiplicity: 1isOrdered: N / AisUnique: N / AdefaultValue: NoneisNullable: FalseClientRedundancyMTTFTThis allows that the FL client must have some type of redundancy for fault tolerance to handle client dropouts in order to be selected by FL server to train a ML model. Its values can be TRUE or FALSE, where TRUE means the FL client must have some type of redundancy and FALSE means that the redundancy of the FL client does not matter for its selection.type: Booleanmultiplicity: 1isOrdered: N / AisUnique: N / AdefaultValue: NoneisNullable: False

[0052] Referring now to the drawings, and more particularly to Figures 1 through 5, there are shown preferred examples but not limited hereto.

[0053] Figure 1 illustrates exemplary hardware components of an FL server (100) according to examples disclosed herein. The FL server (100) comprises at least one processor (101), a memory (102), a communicator (103), and an FL client selection controller (104). The components of the FL server (100) are operatively coupled to one another via one or more internal buses, interfaces, or interconnect mechanisms.

[0054] In one or more examples, the FL server (100) corresponds to or is implemented within a 3GPP-compliant management and control entity's, such as a gNodeB, an eNodeB, a centralized unit (CU), a distributed unit (DU), or any other 3GPP-compliant network function, network element, or managed entity capable of participating in federated machine learning procedures or an MnS producer entity. The processor (101) is configured to orchestrate federated machine learning operations, including management of FL client selection, model distribution, and aggregation of locally trained model updates. The processor (101) executes instructions stored in the memory (102) and controls interactions among the memory (102), the communicator (103), and the FL client selection controller (104). The processor (101) may include one or more processing units such as a central processing unit (CPU), an application processor (AP), a graphics processing unit (GPU), a neural processing unit (NPU), or other hardware accelerators or any combination thereof to support scalable and latency-tolerant ML processing in a 3GPP network environment.

[0055] The memory (102) is configured to store an operating system, virtualization or container runtime components, application programs, configuration data, and operational data used during the execution of federated learning procedures. The memory (102) stores instructions that, when executed by the processor (101), cause the FL server (100) to perform one or more FL-related operations in compliance with 3GPP management and analytics frameworks. The memory (102) may comprise one or more volatile and / or non-volatile computer-readable storage media, including RAM, ROM, flash memory, magnetic or optical storage, EPROM, EEPROM, or any combination thereof, and may be implemented as a non-transitory computer-readable storage medium. The memory (102) may further store federated learning policies, ML model parameters, training round information, client capability information, and ML training request messages received from one or more MnS consumer devices (200).

[0056] The communicator (103) is configured to support communication between the FL server (100), one or more MnS producer entities, and one or more MnS consumer devices (200) over standardized 3GPP management interfaces. The communicator (103) enables communication via service-based interfaces (SBI) and supports one or more communication protocols, including but not limited to hypertext transfer protocol (HTTP / HTTPS), transmission control protocol / internet protocol (TCP / IP), user datagram protocol (UDP), and other protocols defined or referenced in 3GPP specifications. In some examples, the communicator (103) further supports communication over non-3GPP access networks or satellite and broadcast systems, including digital video broadcasting by satellite (DVB-S2). The communicator (103) may include one or more transceivers, network interface controllers, protocol stacks, or virtualized communication functions implemented in hardware, software, or a combination thereof.

[0057] In an example, the FL client selection controller (104) is implemented as a dedicated integrated circuit or as a hardware logic block fabricated on a semiconductor substrate within the FL server (100). The FL client selection controller (104) comprises one or more hardware processing units, control logic circuits, state machines, registers, and scheduling logic configured to autonomously manage federated learning client selection and coordination without reliance on general-purpose software execution. The FL client selection controller (104) is operatively coupled to the processor (101), the memory (102), and the communicator (103) via one or more hardware interfaces, buses, or interconnects. The FL client selection controller (104) further manages coordination of federated training rounds and enforcement of training constraints, thereby enabling efficient, reliable, and scalable execution of federated machine learning within a 3GPP-compliant network management architecture.

[0058] The FL client selection controller (104) receives the ML training request message for training the ML model using FL from the MnS consumer device (200). The ML training request message includes the FL client selection criteria for selecting FL clients (300) from the plurality of FL clients (300) for training the ML model. The FL client selection criteria include at least one of a 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and an FL client redundancy attribute.

[0059] Further, the FL client selection controller (104) sends a response message to the MnS consumer device (200).

[0060] In an example, the ML training request message is or a part of a create model object instance (MOI) request message or a modify MOI attributes request message. Similarly, the response message is a create MOI response message or a modify MOI attributes response message having modified MOI attributes.

[0061] In an embodiment, the FL client selection controller (104) selects a set of FL clients (300) from the plurality of FL clients (300) based on at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the FL client redundancy attribute.

[0062] Further, the FL client selection controller (104) collaborates with the set of selected FL clients (300) to train the ML model based on the ML training request message. The FL client selection controller (104) generates the consolidated ML training report after the completion of training of the ML model by the set of selected FL clients (300). The FL client selection controller (104) sends a notification message to the MnS consumer device (200) notifying the MnS consumer device (200) about the availability of the consolidated ML training report in the FL server (100).

[0063] Furthermore, the FL client selection controller (104) selects the set of FL clients (300) from the plurality of FL clients (300) based on the FL client selection criteria. This includes determining by the MnS producer in the FL server (100) whether the set of FL clients (300) from the plurality of FL clients (300) is using a renewable energy source or not based on the 'is renewable source availability attribute'. The 'is renewable source availability attribute' includes values set to TRUE, indicating that the FL client must be using a renewable energy source to be selected as an FL client, and to FALSE, indicating that the energy source of the FL client does not matter for the selection. The FL client selection controller (104) determines that a particular type of renewable energy source is to be used fully or partially by the set of FL clients (300) from the plurality of FL clients (300) based on the renewable source type attribute. The renewable source type attribute includes values set to at least one of wind, solar, biogas, biofuel, aerothermal, geothermal, hydrothermal, ocean energy, hydropower, biomass, landfill gas, sewage treatment plant gas, and biogases. The FL client selection controller (104) also determines that a certain amount of renewable sourced energy, which at least the FL clients (300) should be using for their operation, in order to be selected as the set of FL clients (300) from the plurality of FL clients (300) by the FL server (100) based on the renewable energy information attribute. The renewable energy information attribute includes values set to one of the percentage amount of renewable energy consumed by an FL client or a ratio of renewable energy consumed to total energy consumed. Further, the FL client selection controller (104) determines that the carbon emission amount of the set of FL clients (300) from the plurality of FL clients (300) is below a defined threshold carbon emission amount value in order to be chosen by the FL server (100) as FL clients (300) based on the carbon emission information attribute. The carbon emission information includes values set to kilograms of carbon dioxide (CO2) equivalent to the consumed energy measured in kWh or joules. The FL client selection controller (104) also determines that the set of FL clients (300) from the plurality of FL clients (300) must have some type of redundancy for fault tolerance to handle client dropouts to be selected as FL clients (300) by the FL server (100) based on the client redundancy attribute. The client redundancy attribute includes values set to TRUE, indicating that the FL client must have some type of redundancy for fault tolerance to be selected as an FL client, and FALSE, indicating that the redundancy of the FL client does not matter for the selection. The MnS producer in the FL server (100) selects the set of FL clients (300) from the plurality of FL clients (300) for collaborative training of the ML model based on the values of at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the client redundancy attribute.

[0064] In an example, the FL client selection criteria is included in an MLTrainingRequest Information Object Class (IOC) or a new IOC name contained in the MLTrainingFunction IOC.

[0065] In an example, to train the ML model, The FL client selection controller (104) generates the ML training request message to be sent to the set of selected FL clients (300) based on the ML training request message received from the MnS consumer device (200), sends the ML training request message to an MnS producer in each selected FL client from the set of selected FL clients (300) for training the ML model, receives a response message to the ML training request message from the MnS producer of each of the set of selected FL clients (300). Each of the set of selected FL clients (300) performs training of the ML model based on the ML training request message received from the MnS consumer in the FL server (100), and creates an ML training report upon completion of training of the ML model. The FL client selection controller (104) receives a notification message from each of the set of selected FL clients (300). The notification message indicates availability of the ML training report in the FL clients (300).

[0066] In an example, to generate the consolidated ML training report after the completion of training of the ML model by the set of selected FL clients (300), the FL client selection controller (104) receives the ML training reports from the set of FL clients (300) upon completion of training of the ML model by the set of FL clients (300) and aggregates all the ML training reports received from each of the set of selected FL clients (300) to create the consolidated ML training report.

[0067] Figure 2 is a block diagram illustrating exemplary hardware components of an MnS consumer device (200) according to examples disclosed herein.

[0068] As illustrated, the MnS consumer device (200) comprises at least one processor (201), a memory (202), a communicator (203), and an FL training request criteria controller (204). The components of the MnS consumer device (200) are operatively coupled to one another via one or more internal buses, interfaces, or interconnects.

[0069] Examples of the MnS consumer device (200) include but are not limited to non-real-time radio access network intelligent controller (Non-RT RIC), an operations administration and maintenance (OAM) system. The processor (201) is configured to manage local participation in ML training using FL, including execution of local training operations and interaction with the FL server (100). The processor (201) executes instructions stored in the memory (202) and controls interactions among the memory (202), the communicator (203), and the FL training request criteria controller (204). The processor (201) may comprise one or more processing units, including but not limited to a central processing unit (CPU), an application processor (AP), a graphics processing unit (GPU), a visual processing unit (VPU), a neural processing unit (NPU), or any combination thereof to support efficient on-device ML processing.

[0070] The memory (202) is configured to store an operating system, application programs, configuration parameters, and temporary or persistent data used by the processor (201) during the execution of federated learning operations. The memory (202) stores instructions that, when executed by the processor (201), cause the MnS consumer device (200) to perform one or more local ML training, inference, aggregation-preparation, or reporting operations in support of FL. The memory (202) may include one or more volatile and / or non-volatile computer-readable storage media, including RAM, ROM, flash memory, magnetic storage devices, optical storage devices, EPROM, EEPROM, or any combination thereof, and may be implemented as a non-transitory computer-readable storage medium.

[0071] The communicator (203) is configured to facilitate communication between the MnS consumer device (200) and the FL server (100). The communicator (203) supports communication over standardized 3GPP management interfaces, including service-based interfaces (SBI), and one or more communication protocols such as hypertext transfer protocol (HTTP / HTTPS), transmission control protocol / internet protocol (TCP / IP), user datagram protocol (UDP), and other protocols defined or referenced in 3GPP specifications. In some examples, the communicator (203) further supports communication over non-3GPP access networks or satellite and broadcast systems, including digital video broadcasting by satellite (DVB-S2). The communicator (203) may include one or more transceivers, network interface controllers, protocol stacks, or virtualized communication functions implemented in hardware, software, or a combination thereof.

[0072] In an example, the FL training request criteria controller (204) is implemented as a dedicated integrated circuit or as a hardware logic block fabricated on a semiconductor substrate within the MnS consumer device (200). The FL training request criteria controller (204) comprises one or more hardware processing units, control logic circuits, state machines, registers, and hardware schedulers configured to autonomously manage local participation in federated learning without reliance on general-purpose software execution. The FL training request criteria controller (204) is operatively coupled to the processor (201), the memory (202), and the communicator (203) via one or more hardware interfaces, buses, or interconnects.

[0073] In an example, the FL training request criteria controller (204) generates the FL client selection criteria for selecting FL clients (300) for training the ML model using FL. The FL client selection criteria include at least one of a 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and a client redundancy attribute.

[0074] Further, the FL training request criteria controller (204) sends the ML training request message for training the ML model to the MnS producer in the FL server (100). The request message includes the FL client selection criteria for selecting FL clients (300) for training of the ML model. The FL training request criteria controller (204) receives the response message from the MnS producer in the FL server (100) indicating the creation of the MOI. The FL training request criteria controller (204) receives the notification message from the MnS producer in the FL server (100). This message notifies the MnS consumer device (200) about the availability of the consolidated ML training report in the FL server (100).

[0075] In an example, the FL training request criteria controller (204) generates the FL client selection criteria by setting values of the 'is renewable source availability attribute' to TRUE, indicating that the FL client must be using a renewable energy source to be selected as an FL client, and to FALSE, indicating that the energy source of the FL client does not matter for the selection. The FL client selection criteria can include setting a particular type of renewable energy source to be used fully or partially by the FL client. The renewable source type attribute includes values set to at least one of wind, solar, biogas, biofuel, aerothermal, geothermal, hydrothermal, ocean energy, hydropower, biomass, landfill gas, sewage treatment plant gas, and biogases. Further, the FL client selection criteria include setting a certain amount of renewable sourced energy that the FL clients (300) at least should be using for their operation to be selected as FL clients (300). The renewable energy information attribute includes values set to one of the percentage amount of renewable energy consumed by an FL client or a ratio of renewable energy consumed to total energy consumed. Furthermore, the FL client selection criteria include setting a threshold carbon emission amount value below which the FL client should emit to be chosen by the FL server (100) as an FL client. The carbon emission information includes values set to kilograms of CO2 equivalent to the consumed energy measured in kWh or joules. Furthermore, the FL client selection criteria include setting some type of redundancy for fault tolerance to handle client dropouts in order to be selected as FL clients (300) by the FL server. The client redundancy attribute includes values set to TRUE, indicating that the FL client must have some type of redundancy for fault tolerance to be selected as an FL client, and FALSE, indicating that the redundancy of the FL client does not matter for the selection.

[0076] In an example, the FL client selection criteria include at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the client redundancy attribute. These criteria are included in an MLTrainingRequest Information Object Class (IOC) or a new IOC name contained in the MLTrainingFunction IOC.

[0077] In an example, the ML training request message is or a part of a create MOI request message or a modify MOI attributes request message. Similarly, the response message is a create MOI response message or a modify MOI attributes response message having modified MOI attributes.

[0078] Figure 3 is a flow chart that illustrates a proposed method, implemented by the FL server (100), for machine learning training client selection, according to examples as disclosed herein.

[0079] At step 301, the method includes receiving by the MnS producer in the FL server (100) the ML training request message for training the ML model using FL from the MnS consumer device (200). The ML training request message includes the FL client selection criteria for selecting FL clients (300) from the plurality of FL clients (300) for training the ML model. The FL client selection criteria include at least one of a 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and an FL client redundancy attribute.

[0080] At step 302, the method includes sending by the MnS producer in the FL server (100) the response message to the MnS consumer device (200) indicating the creation of the MOI. At step 303, the method includes selecting by the MnS producer in the FL server (100) the set of FL clients (300) from the plurality of FL clients (300) based on at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the FL client redundancy attribute.

[0081] At step 304, the method includes generating by the MnS producer in the FL server (100) the consolidated ML training report after completion of training of the ML model by the set of selected FL clients (300) and at step 305, the method includes sending by the MnS producer in the FL server (100) the notification message to the MnS consumer device (200) notifying the MnS consumer device (200) about the availability of the consolidated ML training report in the FL server (100).

[0082] In an example, the method includes collaborating by the MnS producer in the FL server (100) with the set of selected FL clients (300) to train the ML model based on the ML training request message.

[0083] The collaborating include generating, by the MnS producer in the FL server (100), a ML training request message to be sent to the set of selected FL clients based on the ML training request message received from the MnS consumer device (200) and sending, by the MnS producer in the FL server (100), acting as an MnS consumer, the ML training request message to an MnS producer in each selected FL client from the set of selected FL clients for training the ML model based on the ML training request message. The collaborating further includes receiving, by the MnS consumer in the FL server (100), acting as an MnS consumer, a response message to the ML training request message from the MnS producer of each of the set of selected FL clients, where each selected FL client from the set of selected FL clients performs training of the ML model based on the ML training request message received from the MnS consumer in the FL server (100) and creates an ML training report upon completion of training of the ML model, receiving, by the MnS producer in the FL server (100), acting as an MnS consumer, a notification message notifying availability of the ML training report from each of the set of selected FL clients and sending, by the MnS Producer in each selected FL client from the set of selected FL clients, a notification message to the MnS consumer in the FL server (100) about availability of the ML training report in FL client"

[0084] In an example, the method includes selecting by the MnS producer in the FL server (100) the set of FL clients (300) from the plurality of FL clients (300) based on the FL client selection criteria. This includes determining by the MnS producer in the FL server (100) whether the set of FL clients (300) from the plurality of FL clients (300) is using the renewable energy source or not based on the 'is renewable source availability attribute'. The 'is renewable source availability attribute' includes values set to TRUE indicating that the FL client must be using the renewable energy source to be selected as the FL client and to FALSE indicating that the energy source of the FL client does not matter for the selection.

[0085] Further, the method includes determining that the particular type of renewable energy source to be used fully or partially by the set of FL clients (300) from the plurality of FL clients (300) based on the renewable source type attribute. The renewable source type attribute includes values set to at least one of wind, solar, biogas, biofuel, aerothermal, geothermal, hydrothermal, ocean energy, hydropower, biomass, landfill gas, sewage treatment plant gas, and biogases.

[0086] Furthermore, the method includes determining that the certain amount of renewable sourced energy which at least the FL clients (300) should be using for their operation in order to be selected as the set of FL clients (300) from the plurality of FL clients (300) by the FL server (100) based on the renewable energy information attribute. The renewable energy information attribute includes values set to one of the percentage amount of renewable energy consumed by an FL client or the ratio of renewable energy consumed to total energy consumed.

[0087] Determining by the MnS producer in the FL server (100) further includes determining that the carbon emission amount of the set of FL clients (300) from the plurality of FL clients (300) is below the defined threshold carbon emission amount value in order to be chosen by the FL server (100) as an FL client based on the carbon emission information attribute. The carbon emission information includes values set to kilograms of CO2 equivalent to the consumed energy measured in kWh or joules.

[0088] Further determining by the MnS producer in the FL server (100) includes determining that the set of FL clients (300) from the plurality of FL clients (300) must have some type of redundancy for fault tolerance to handle client dropouts in order to be selected as FL clients (300) by the FL server (100) based on the client redundancy attribute. The client redundancy attribute includes values set to TRUE indicating that the FL client must have some type of redundancy for fault tolerance in order to be selected as an FL client and FALSE indicating that redundancy of the FL client does not matter for the selection.

[0089] Selecting by the MnS producer in the FL server (100) the set of FL clients (300) from the plurality of FL clients (300) for collaborative training of the ML model is based on the values of at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the client redundancy attribute.

[0090] In an example, the FL client selection criteria include at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the client redundancy attribute which are included in an MLTrainingRequest Information Object Class (IOC) or the new IOC name contained in MLTrainingFunction IOC.

[0091] In an example, the method includes collaborating by the MnS producer in the FL server (100) with the set of selected FL clients (300) to train the ML model. This includes generating by the MnS producer in the FL server (100) the ML training request message to be sent to the set of selected FL clients (300) based on the ML training request message received from the MnS consumer device (200) and sending by the MnS producer in the FL server (100) acting as an MnS consumer device (200) the ML training request message to an MnS producer in each selected FL client from the set of selected FL clients (300) for training the ML model based on the ML training request message.

[0092] The method further includes receiving by the MnS consumer device (200) in the FL server (100) the response message to the ML training request message from the MnS producer of each of the set of selected FL clients (300) where each selected FL client from the set of selected FL clients (300) performs training of the ML model based on the ML training request message received from the second MnS consumer in the FL server (100) device and creates an ML training report upon completion of training of the ML model. Receiving by the MnS producer in the FL server (100) acting as an MnS consumer the notification message notifying availability of the ML training report from each of the set of selected FL clients (300) and sending by the MnS producer in each selected FL client from the set of selected FL clients (300) the notification message to the MnS consumer device (200) in the FL server (100) about the availability of the ML training report in the FL client.

[0093] In an example, the method includes generating by the MnS producer in the FL server (100) the consolidated ML training report after completion of training of the ML model by the set of selected FL clients (300). This includes receiving by the MnS producer in the FL server (100) ML training reports from the set of FL clients (300) upon completion of training of the ML model by the set of FL clients (300) and aggregating by the MnS producer in the FL server (100) all the ML training reports received from each of the set of selected FL clients (300) to create the consolidated ML training report.

[0094] In an example, the ML training request message is or a part of a create MOI request message or a modify MOI attributes request message. Similarly, the response message is a create MOI response message or a modify MOI attributes response message having modify MOI attributes.

[0095] Figure 4 is a flow chart that illustrates a proposed method, implemented by the MnS consumer device (200), for training a ML model using FL according to examples as disclosed herein.

[0096] At step 401, the method includes generating by the MnS consumer device (200) the FL client selection criteria for selecting FL clients (300) for training the ML model using FL where the FL client selection criteria include at least one of the 'is renewable source availability attribute', a renewable source type attribute, a renewable energy information attribute, a carbon emission information attribute, and a client redundancy attribute. At step 402, the method includes sending by the MnS consumer device (200) the ML training request message for training the ML model to the MnS producer in the FL server (100) where the request message includes the FL client selection criteria for selecting FL clients (300) for training of the ML model. At step 403, the method includes receiving by the MnS consumer device (200) the response message from the MnS producer in the FL server (100) indicating the creation of the MOI. At step 404, the method includes generating, by the MnS producer in the FL server, a consolidated ML training report after completion of training of the ML model by the set of selected FL clients (300). At step 405, the method includes receiving by the MnS consumer device (200) the notification message from the MnS producer in the FL server (100) notifying the MnS consumer device (200) about the availability of the consolidated ML training report in the FL server (100).

[0097] In an example, the method includes generating by the MnS consumer device (200) the FL client selection criteria including setting values of the 'is renewable source availability attribute' to TRUE indicating that the FL client must be using the renewable energy source in order to be selected as the FL client and to FALSE indicating that an energy source of the FL client does not matter for the selection and setting the particular type of the renewable energy source to be used fully or partially by the FL client where the renewable source type indicates that the FL client is selected by the FL server (100) only if it is using the given particular type of renewable energy source fully or partially. The renewable source type attribute includes values set to at least one of wind, solar, biogas, biofuel, aerothermal, geothermal, hydrothermal, ocean energy, hydropower, biomass, landfill gas, sewage treatment plant gas, and biogases. The method further includes setting the certain amount of renewable sourced energy which at least the FL clients (300) at least should be using for their operation in order to be selected as FL clients (300) where the renewable energy information attribute includes values set to one of the percentage amount of renewable energy consumed by an FL client or the ratio of renewable energy consumed to total energy consumed, setting the threshold carbon emission amount value below which the FL client should emit in order to be chosen by the FL server (100) as an FL client where the carbon emission information includes values set to kilograms of CO2 equivalent to the consumed energy measured in kWh or joules, and setting some type of redundancy for fault tolerance to handle client dropouts in order to be selected as FL clients (300) by the FL server (100) where the client redundancy attribute includes values set to TRUE indicating that the FL client must have some type of redundancy for fault tolerance in order to be selected as an FL client and FALSE indicating that redundancy of the FL client does not matter for the selection.

[0098] In an example, the FL client selection criteria include at least one of the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the client redundancy attribute are included in an MLTrainingRequest Information Object Class (IOC) or the new IOC name contained in MLTrainingFunction IOC.

[0099] In an example, the ML training request message is or a part of a create MOI request message or a modify MOI attributes request message. Similarly, the response message is a create MOI response message or a modify MOI attributes response message having modify MOI attributes.

[0100] Figure 5 is a sequence diagram that illustrates an example scenario for training a ML model using FL, according to examples disclosed herein.

[0101] In an example, at step 1 the MnS consumer device (200) sends a request to the MnS producer in the FL server (100) for training the model using Federated Learning. The request contains the FL client selection criteria to be used by the FL server (100) to select FL clients (300). These client selection requirements can include IsRenewableSource, RenewableSourceType, RenewableEnergyInfo, CarbonEmissionInfo, and ClientRedundancy.

[0102] At step 2, the MnS producer of the FL server (100) sends the response of the createMOI request to the MnS consumer device (200). At step 3, the MnS producer in the FL server (100) selects the FL clients (300) for collaborative model training as per the requested FL client selection criteria. At step 4, the MnS producer in the FL server (100) then behaving as the MnS consumer device (200) sends a request to the MnS producer in FL client1 for training the ML model as per the intended training requirements. This request is formed using the information received in step 1.

[0103] At step 5, the MnS producer of the FL client sends the response of the createMOI request to the MnS consumer device (200) in the FL server (100). At step 6, step 4 is repeated for all selected FL clients (300). At step 7, step 5 is performed by all selected FL clients (300). At step 8, model training is done in all selected FL clients (300) as per the FL requirements sent by the MnS consumer device (200) following the Federated Learning mechanisms.

[0104] At step 9, upon training completion, the MnS producer in FL client1 creates the ML training report. At step 10, assuming that the MnS consumer device (200) in the FL server (100) has subscribed for the creation notification, the producer in FL client1 sends the notifyMOIcreation notification to the MnS consumer device (200) notifying about the availability of the ML training report. At step 11, step 9 is repeated for all selected FL clients (300). At step 12, step 10 is performed by all selected FL clients (300).

[0105] At step 13, after receiving all ML training reports from all FL clients (300), the MnS producer in the FL server (100) consolidates the final ML training report and the final ML model is aggregated / created / produced in the FL server (100). At step 14, assuming that the MnS consumer device (200) has subscribed for the creation notification, the MnS producer in the FL server (100) sends the notifyMOIcreation notification to the MnS consumer device (200) notifying about the availability of the final consolidated ML training report and reference to the trained ML model.

[0106] In an example, the disclosure introduces a new attribute for ensuring renewable energy source-based FL client selection in the FL approach. Another example introduces a new attribute for ensuring a certain percentage of renewable energy usage in the FL client. Further, an example introduces a new attribute for ensuring a particular type of renewable energy source usage by the FL client.

[0107] In an example, the disclosure introduces a new attribute for ensuring reduced carbon emission during the FL-based ML approach by selecting FL clients (300) which have carbon emissions below a certain amount. An example also introduces a new attribute for ensuring an FL client which has some redundancy for more reliability.

[0108] The disclosure enables sustainable FL training for an ML model as it promotes the selection of FL clients (300) based on energy-saving principles including green energy, carbon emission, and renewable energy sources. One or more combinations of the given attributes such as the 'is renewable source availability attribute', the renewable source type attribute, the renewable energy information attribute, the carbon emission information attribute, and the FL client redundancy attribute can be used by the FL server (100) to select the FL clients (300). Such selection of FL clients (300) by an MnS producer in the FL server (100) can result in overall decreased carbon emissions and reduced fossil fuel-based energy consumption, thus promoting overall sustainable development of AI / ML in 5G systems. Further, the disclosure enables the selection of FL clients (300) based on redundancy which in turn makes the overall FL mechanism more fault-tolerant and reliable.

[0109] By incorporating renewable energy sources into the client selection process, the disclosure ensures that the energy consumed during FL training is derived from sustainable and environmentally friendly sources. This not only reduces the carbon footprint associated with the training process but also encourages the use of renewable energy infrastructure. For instance, clients located in regions with abundant solar or wind energy can be prioritized, thereby leveraging the natural energy resources available in those areas. This strategic selection not only optimizes energy consumption but also aligns with global efforts to transition towards greener energy solutions.

[0110] Further, the inclusion of the FL client redundancy attribute enhances the robustness and reliability of the FL system. By selecting clients with redundant capabilities, the system can better handle potential failures or dropouts during the training process. This redundancy ensures that the training process can continue smoothly even if some clients become unavailable, thereby maintaining the integrity and continuity of the ML model training. This fault-tolerant approach is particularly crucial in 5G systems, where the reliability and efficiency of AI / ML applications are paramount. Thus, the disclosure represents a significant advancement in sustainable and resilient FL training methodologies, contributing to the broader goals of environmental sustainability and technological reliability.

[0111] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and / or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of preferred embodiments, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the scope of the embodiments as described herein.

Claims

1.A method performed by a federated learning (FL) server in a communication system, the method comprising:receiving, from a consumer for a machine learning (ML) training, a request for the ML training including information on a requirement for an FL, wherein the requirement for the FL includes information on FL client selection criteria for the FL;selecting at least one FL client based on the information on the FL client selection criteria;generating an FL report for the at least one FL client based on a result of the ML training for the at least one FL client; andtransmitting, to the consumer, an ML model training report including the FL report.2.The method of claim 1,wherein the information on FL client selection criteria for the FL includes at least one of:information on a redundancy for fault tolerance to handle client dropouts;information on an energy source; andinformation on a carbon emission,wherein, in case that the information on the redundancy is configured to true, the at least one FL client has the redundancy, andwherein, in case that the information on the redundancy is configured to false, the redundancy is not included in the FL client selection criteria.3.The method of claim 1, further comprising:transmitting, to the consumer, a response of the request; andreceiving, from the at least one FL client, information on the result of the ML training for the at least one FL client,wherein data samples for the ML training are not exchanged among the FL server and the at least one FL client.4.A method performed by a consumer in a communication system, the method comprising:generating information on a requirement for a federated learning (FL), wherein the requirement for the FL includes information on FL client selection criteria for the FL;transmitting, to an FL server, a request for a machine learning (ML) training including the information on the requirement for the FL; andreceiving, from the FL server, an ML model training report including an FL report for at least one FL client associated with the FL client selection criteria,wherein the FL report is associated with a result of the ML training for the at least one FL client.5.The method of claim 4,wherein the information on FL client selection criteria for the FL includes at least one of:information on a redundancy for fault tolerance to handle client dropouts;information on an energy source; andinformation on a carbon emission,wherein, in case that the information on the redundancy is configured to true, the at least one FL client has the redundancy, andwherein, in case that the information on the redundancy is configured to false, the redundancy is not included in the FL client selection criteria.6.The method of claim 5, further comprising:receiving, from the FL server, a response of the request,wherein data samples for the ML training are not exchanged among the FL server and the at least one FL client.7.A federated learning (FL) server in a communication system, the FL server comprising:a transceiver;a processor coupled to the transceiver; andmemory coupled to the processor and storing instructions executable by the processor,wherein the instructions cause the FL server to:receive, from a consumer for a machine learning (ML) training, a request for the ML training including information on a requirement for an FL, wherein the requirement for the FL includes information on FL client selection criteria for the FL,select at least one FL client based on the information on the FL client selection criteria,generate an FL report for the at least one FL client based on a result of the ML training for the at least one FL client, andtransmit, to the consumer, an ML model training report including the FL report.8.The FL server of claim 7,wherein the information on FL client selection criteria for the FL includes at least one of:information on a redundancy for fault tolerance to handle client dropouts;information on an energy source; andinformation on a carbon emission,wherein, in case that the information on the redundancy is configured to true, the at least one FL client has the redundancy, andwherein, in case that the information on the redundancy is configured to false, the redundancy is not included in the FL client selection criteria.9.The FL server of claim 7,wherein the instructions further cause the FL server to:transmit, to the consumer, a response of the request, andreceive, from the at least one FL client, information on the result of the ML training for the at least one FL client, andwherein data samples for the ML training are not exchanged among the FL server and the at least one FL client.10.A consumer in a communication system, the consumer comprising:a transceiver;a processor coupled to the transceiver; andmemory coupled to the processor and storing instructions executable by the processor,wherein the instructions cause the consumer to:generate information on a requirement for a federated learning (FL), wherein the requirement for the FL includes information on FL client selection criteria for the FL,transmit, to an FL server, a request for a machine learning (ML) training including the information on the requirement for the FL, andreceive, from the FL server, an ML model training report including an FL report for at least one FL client associated with the FL client selection criteria, andwherein the FL report is associated with a result of the ML training for the at least one FL client.11.The consumer of claim 10,wherein the information on FL client selection criteria for the FL includes at least one of:information on a redundancy for fault tolerance to handle client dropouts;information on an energy source; andinformation on a carbon emission,wherein, in case that the information on the redundancy is configured to true, the at least one FL client has the redundancy, andwherein, in case that the information on the redundancy is configured to false, the redundancy is not included in the FL client selection criteria.12.The consumer of claim 10,wherein the instructions further cause the consumer to receive, from the FL server, a response of the request, andwherein data samples for the ML training are not exchanged among the FL server and the at least one FL client.

Images