Method and wireless network for managing data

Abstract

The disclosure relates to a 5G or 6G communication system for supporting a higher data transmission rate. Accordingly, the embodiments herein provide a method performed by a management service (MnS) producer. The method includes receiving, from an MnS consumer, a machine learning (ML) training request information for an ML model training, wherein the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split. The method includes identifying a preference of the MnS consumer as to whether a training data is to be split, based on the data split indication. The method includes transmitting, to the MnS consumer, a response associated with the ML training request information.

Description

METHOD AND WIRELESS NETWORK FOR MANAGING DATA

[0001] Embodiments disclosed herein relate to a data management system and method, and more particularly to a method and a wireless network for managing data.

[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 principal object of the embodiments herein is to disclose a method and a system (or wireless network) for managing data in the wireless network.

[0009] Accordingly, the embodiments herein provide a method performed by a management service (MnS) producer. The method includes receiving, from an MnS consumer, a machine learning (ML) training request information for an ML model training. In an embodiment, the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split. The method includes identifying a preference of the MnS consumer as to whether the training data is to be split, based on the data split indication. The method includes transmitting, to the MnS consumer, a response associated with the ML training request information.

[0010] Accordingly, the embodiments herein provide a management service (MnS) producer. The MnS producer includes a controller. The MnS producer includes a memory storing instructions. The MnS producer includes at least one processor communicatively coupled to the controller and the memory. The at least one processor configured to execute the instructions stored in the memory to cause the MnS producer to receive, from an MnS consumer, a machine learning (ML) training request information for an ML model training. In an embodiment, the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split. The at least one processor configured to execute the instructions stored in the memory to cause the MnS producer to identify a preference of the MnS consumer as to whether a training data is to be split, based on the data split indication. The at least one processor configured to execute the instructions stored in the memory to cause the MnS producer to transmit, to the MnS consumer, a response associated with the ML training request information.

[0011] Accordingly, the embodiments herein provide a performed by a management service (MnS) consumer. The method includes determining whether a training data is to be split. The method includes transmitting, to an MnS producer, a machine learning (ML) training request information for an ML model training, wherein the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split. The method includes receiving, from the MnS producer, a response associated with the ML training request information

[0012] Accordingly, the embodiments herein provide a management service (MnS) consumer. The MnS consumer includes a controller. The MnS consumer includes a memory storing instructions. The MnS consumer includes at least one processor communicatively coupled to the controller and the memory. The at least one processor configured to execute the instructions stored in the memory to cause the MnS consumer to determine whether a training data is to be split. The at least one processor configured to execute the instructions stored in the memory to cause the MnS consumer to transmit, to an MnS producer, transmit, to an MnS producer, a machine learning (ML) training request information for an ML model training, wherein the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split. The at least one processor configured to execute the instructions stored in the memory to cause the MnS consumer to receive, from the MnS producer, a response associated with the ML training request information.

[0013] These and other aspects of the embodiments herein will be better appreciated and understood when considered in conjunction with the following description and the accompanying drawings. It should be understood, however, that the following descriptions, while indicating at least one embodiment and numerous specific details thereof, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the embodiments herein without departing from the scope thereof, and the embodiments herein include all such modifications.

[0014] The embodiments disclosed herein are illustrated in the accompanying drawings, throughout which like reference letters indicate corresponding parts in the various figures. The embodiments herein will be better understood from the following description with reference to the drawings, in which:

[0015] FIG. 1 illustrates an operations of a GenAI in a wireless network, according to prior art.

[0016] FIG. 2 is a sequence diagram illustrating a method for training a data management, according to embodiments as disclosed herein.

[0017] FIG. 3 illustrates sequence diagram indicating distributed training management, in accordance with the embodiment of the present disclosure.

[0018] FIG. 4 shows various hardware components of a first producer, according to embodiments as disclosed herein.

[0019] FIG. 5 shows various hardware components of a consumer, according to embodiments as disclosed herein.

[0020] FIG. 6 is a flow chart illustrating a method, implemented by the first producer, for managing the data in the wireless network, according to embodiments as disclosed herein.

[0021] FIG. 7 is a flow chart illustrating a method, implemented by the consumer, for managing the data in the wireless network, according to embodiments as disclosed herein.

[0022] 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. 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 of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.

[0023] For the purposes of interpreting this specification, the definitions (as defined herein) will apply and whenever appropriate the terms used in singular will also include the plural and vice versa. It is to be understood that the terminology used herein is for the purposes of describing particular embodiments only and is not intended to be limiting. The terms "comprising", "having" and "including" are to be construed as open-ended terms unless otherwise noted.

[0024] The words / phrases "exemplary", "example", "illustration", "in an instance", "and the like", "and so on", "etc.", "etcetera", "e.g.," , "i.e.," are merely used herein to mean "serving as an example, instance, or illustration." Any embodiment or implementation of the present subject matter described herein using the words / phrases "exemplary", "example", "illustration", "in an instance", "and the like", "and so on", "etc.", "etcetera", "e.g.," , "i.e.," is not necessarily to be construed as preferred or advantageous over other embodiments.

[0025] Embodiments herein may be described and illustrated in terms of blocks which carry out a described function or functions.　These blocks, which may 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 may optionally be driven by a firmware. 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 may 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 may 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 may be physically combined into more complex blocks without departing from the scope of the disclosure.

[0026] It should be noted that elements in the drawings are illustrated for the purposes of this description and ease of understanding and may not have necessarily been drawn to scale. For example, the flowcharts / sequence diagrams illustrate the method in terms of the steps required for understanding of aspects of the embodiments as disclosed herein. Furthermore, in terms of the construction of the device, one or more components of the device may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the present embodiments so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein. Furthermore, in terms of the system, one or more components / modules which comprise the system may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the present embodiments so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein.

[0027] The accompanying drawings are used to help easily understand various technical features and it should be understood that the embodiments presented herein are not limited by the accompanying drawings. As such, the present disclosure should be construed to extend to any modifications, equivalents, and substitutes in addition to those which are particularly set out in the accompanying drawings and the corresponding description. Usage of words such as first, second, third etc., to describe components / elements / steps is for the purposes of this description and should not be construed as sequential ordering / placement / occurrence unless specified otherwise.

[0028] The embodiments herein achieve a method and system (or wireless network) for data management.

[0029] Artificial Intelligence (AI) / Machine Learning (ML) techniques and relevant applications are being increasingly adopted by the wider industries and proved to be successful. These are now being applied to telecommunication industry including mobile networks. Although, the 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 machine learning techniques in general include supervised learning, semi-supervised learning, unsupervised learning and reinforcement learning. Each learning method fits one or more specific category of inference (e.g., prediction), and requires specific type of training data.

[0030] A lifecycle management of an AI / ML model is being defined in a Third Generation Partnership Project (3GPP) Service & System Aspects Working Group 5 (SA WG5). The lifecycle stages include a ML model training stage, a ML testing stage, a ML emulation stage, a ML entity loading stage and an inference phase stage. The ML model training stage includes an initial training and re-training, of an ML model or a group of ML models. Further, the ML model training stage also includes validation of an ML entity to evaluate the performance when the ML entity performs on the training data and validation data. If a 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 stage is an initial phase of the workflow. The ML testing stage evaluates 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. The ML emulation stage runs the ML entity for inference in an emulation environment. The purpose is to evaluate an inference performance of the ML entity in the emulation environment prior to applying it to a target network or a target system. The ML entity loading stage is a process (also known as a sequence of atomic actions) of making the trained ML entity available for use at a target AI / ML inference function. The AI / ML inference performs inference using the trained ML entity by the AI / ML inference function.

[0031] Further, Generative artificial intelligence (GenAI) is an emerging category of the AI techniques that are capable of generating new content. The generated content could be in various formats like images, text, video, audio, code etc. Large Language Models (LLM) and Deep representation learning are key foundation elements in GenAI based techniques. The LLMs are very deep neural networks that are trained on huge amount of data, typically to handle language centric problems / scenarios, like text prediction of next word given a sequence of words framing sentence. The deep representation learning (e.g., Generative model) is about using the deep neural networks for learning the distribution of the input data to generate new content that maintain a statistical consistency with an original input data. These representations are learnt from the input data during learning process. For instance, with thousands of facial images as input data, the generative model could learn to produce new images on non-existing people from the input data. Generative Adversarial Networks (GAN) are one such neural network architecture and techniques used for learning the data representations. The GANs consist of two neural networks, such as generator (110) and discriminator (120) that work with contradicting learning goals. Generator's goal is to generate new data which has statistical consistency with input data, the discriminator's goal is to evaluate and discriminate the generated content by the generator (110) against the real input data. During training, the generator (110) learns to create new content that is close to real input data and deceive the discriminator (120). Synthetic data generation is one utility of the generative model. The synthetic data generation is about artificially generating the data than from a real world data (observations). The artificially generated data has the similar statistical patterns as the real world data. The synthetic data generation is used for various applications to overcome the imbalance in the real data, handle insufficient real world data pertaining to certain scenarios, improve efficiency of ML models with more accurate predictions that are trained on the synthetic data. FIG. 1 illustrates an operations of the GenAI in a wireless network (100).

[0032] The training data is a backbone of the entire AI and ML project and without that it is not possible to train the model that learns and predict for humans requirements. The ample amount of the training data should be available for a successful and efficient training process resulting in the ML model with optimal performance and more precise accuracies. However, at times the training data available is not sufficient for various ML problems, because of factors like imbalance in the available data, insufficient number of required data samples, available data is not representing all scenarios / contexts, to achieve required level of accuracy etc. This results in the requirement for the data mimicking the real data, to be generated before it can be used for the ML training.

[0033] A model that has to perform in a particular context (e.g., location, time, node or the like) should be trained with the data specific for the same context. Hence, a consumer may specify the context for the data to be generated using the generative AI concept. As per the generative AI concept, the data generated need to be checked / discriminated for its genuineness / statistical consistency (with the input training data) before it is considered a successful generation. This process is called the discriminator process. The consumer would like to provide a criteria (e.g., confidence value) to be satisfied while checking the genuineness for that data. This will help producer to efficiently perform the discriminator functionality.

[0034] Further, a distributed machine learning is the process of distributing the machine learning workload across different compute nodes. The functionality is called DT (Distributed Training). The workload distribution can be split of the ML model or split of the data set used for training or combination of both. Basically, the distributed training is a distributed computing approach to the machine learning training and inference. This process is used for handling the workload related to large model training with large data sets. During distributed machine learning process, weights and intermediate results from various compute nodes are aggregated by a central server to determine or update the parameters of the model.

[0035] The distributed training is a model training paradigm that involves spreading training workload across multiple training functions, to accelerate the training process and / or reduce the required computational resources. The distributed training may be used for traditional machine learning models, as well as for large models. In a fifth generation system (5GS), the ML training function may be located within a management system or in a network function (NF) (e.g. gNB or Network Data Analytics Function (NWDAF)), i.e. a worker node for training. Each node has different computing resources and storage capacity based on physical infrastructure such as Central Processing Unit (CPU) / Graphics processing unit, (GPU) / data processing unit (DPU), memory, storage, and network bandwidth. In order to obtain load balance between the nodes and maximize the efficiency of resource utilization, splitting up the training may be necessary and involving multiple training functions according to the actual situation of the nodes may be needed. Thus, aspects of distributed training need to be supported in the management system.

[0036] A crucial part of enabling distributed training is providing training requirement to a producer so that the producer implements training functionality fulfilling the consumer's requirements. The existing mechanism lacks on providing sufficient requirement to the producer. The training data provided by the consumer may pertain to a particular location. Hence, the consumer may wish to train the model for a particular purpose (e.g., handover optimization, radio link failure avoidance or the like) pertaining to a same location. The sample training data provided by the consumer may not be enough. Hence, the consumer may not want the producer to split the training data, making it sparser, while doing distributed training.

[0037] The above information is presented as background information only to help the reader to understand the present disclosure. Applicants have made no determination and make no assertion as to whether any of the above might be applicable as prior art with regard to the present application.

[0038] In an embodiment, an ML training request includes information enabling training data generation using a generative AI. In an embodiment, an ML training report includes a reference to the ML model trained using a generated ML data. This will allow generation of the training data that is comprehensive enough to enable efficient ML model training. The GenAI machine learning technique are a new and trending class for generating new content, with a lot of successful implementations in industry. With proven success of the techniques, these techniques can be used for synthetic data generation in mobile networks, for creating better ML models improving efficiency of various network operations. The solution involves training an ML model using Generative AI (GenAI) concept to generate a training data.

[0039] The present disclosure relates to training management of machine learning models. Particularly, to a method and system for enhanced ML training management. In 5GS, the ML training function may be located within the management system or in the NF (e.g. gNB or NWDAF), i.e. the worker node for training. Each node has different computing resources and storage capacity based on physical infrastructure such as CPU / GPU / DPU, memory, storage, and network bandwidth. In order to obtain load balance between nodes and maximize the efficiency of resource utilization, splitting up the training may be necessary and involving multiple training functions according to the actual situation of nodes may be needed. Thus, aspects of distributed training need to be supported in the management systems. Thus, the consumer to provide, to the producer, a set of requirements related to DT that should be supported and taken care of by the DT process created by the producer. The information such as target inference location and data split indication may be provided as the requirements.

[0040] In an embodiment, the present disclosure involves providing, to a producer, from a consumer a set of requirements related to a DT that may be supported and taken care of by a DT process created by the producer. The following information shall be provided as the requirements. The information pertaining to target inference location which specifies the location of the inference function that will host the trained ML Model. The target inference location may be defined as geographical coordinate, geographical area as a convex polygon with a geo coordinate specifying a corner of the convex polygon. Further, the information related to data split indication may be indicated that is a Boolean attribute specifying whether the provided training data should be split among the worker node or not. The value "true" specify that the training data can be spilt. Whereas value "false" specify that the training data shall not be split. The default value is false. Thus, the present disclosure suggests Network Resource Model (NRM) change for 3GPP TS28.105 indicating new attribute in MLTrainingRequest Information Object Class (IOC) to capture the requirements indicated above.

[0041] Referring now to the drawings, and more particularly to FIGs. 1 to 7, where similar reference characters denote corresponding features consistently throughout the figures, there are shown at least one embodiment.

[0042] FIG. 2 is a sequence diagram illustrating a method for training a data management in a wireless network (100), according to embodiments as disclosed herein. The wireless network (100) can be, for example, but not limited to a fourth generation (4G) network, a fifth generation (5G) network, a sixth generation (6G) network, an Open Radio Access Network (ORAN) or the like.

[0043] At step 1, the consumer (210) sends a request (e.g., create MOI (MLTrainingRequest{genAIInfo})) to train a model to generate new training data. The request is included with the information (e.g., genAIInfo or the like) enabling training data generation using the generative AI. The consumer (210) can be, a Management Service (MnS) consumer (210a). In an embodiment, the consumer (210) sends the training request to a first producer (220) (e.g., MnS producer (GenAI MLTraining Function) (220a)) for training a model that can generate the training data. The request includes the following information, in addition to what is already defined in MLTrainingRequest [3GPP TS 28.105], related to the GenAI.

[0044] 1. trainingDataGeneration: This defines whether the request to generate training data to be used for training an ML model for the analytics type provided using inferenceType attribute.

[0045] 2. genAIDataContext: This defines the context for which the data is to be generated

[0046] a. Location: This defines the geographical location for which the training data is to be generated.

[0047] b. Time: This defines the time duration for which the training data is to be generated.

[0048] 3. genuinenessConfidenceThreshold: It is the numerical value (in unit of percentage) that represents the required threshold to be met for asserting the genuineness / statistical consistency (with the input training data) of the generated data. This is provided by the consumer (210) with the ML Training request.

[0049] At step 2, the first producer (220) creates the training process (e.g., MLTraining Process creation). At step 3, the first producer (220) sends a response (e.g., createMOI() Response) to the consumer (210).

[0050] At step 4 and step 5, after the training complete the first producer (220) instantiate the MLTrainingReport IOC capturing the training result. This will include the training data generated using generative AI. As part of this steps 4 and 5, the first producer (220) generates the model. As part of this steps, the model will be trained to generate new training data for various purpose (indicated by infereneType in the training request). In an embodiment, the first producer (220) trains the model and instantiate the ML training report with the following information, in addition to what is already defined in MLTrainingReport [3GPP TS 28.105]:

[0051] 1. ML Model: This defines the reference to the generated ML model.

[0052] 2. genAIGenuinenessConfidence: It is the numerical value (in unit of percentage) that represents the genuineness / statistical consistency (with the input training data) of the data that can be generated using the trained model. This is provided by the producer with the ML Training report.

[0053] The first producer (220) instantiates ML model training using the generated data and provides the reference to the trained ML model. In an embodiment, gENAImLModelRef defines the reference to the generated ML model trained using data generated utilizing GenAI.

[0054] At step 6, the consumer (210) is notified about the training result including the generated data using notifyMOICreation(GeneratedData) (for example). At step 7, the first producer (220) sends a request (e.g., createMOI (MLTrainingRequest{GeneratedData})) to train a ML model to a second producer (230) using the generated data. The request may also include the requirement for DT. The processing of such information is described as part of FIG.3. The second producer (230) can be, for example, but not limited to an MnS producer (ML training provider) (230a).

[0055] At step 8, the training process is created at the second producer (230). At step 9, the first producer (220) is provided with an acknowledgment (e.g., createMOI() Response) from the second producer (230).

[0056] At step 10, the first producer (220) updates the report for initial training with the details of the trained model using an MLTrainingReport Update (for example). At step 11, the consumer (210) is notified about the trained model using the notifyMOIChange(ML Model).

[0057] FIG. 3 illustrates a sequence diagram indicating a distributed training management procedure in the wireless network (100). The wireless network (100) includes the consumer (210) (e.g., MnS consumer (210a)), the first producer 220 (e.g., MnS producer ( Distributed MLTraining Function) (220b)), and the second producer 230 (e.g., MnS producer (AI / ML Inference Function) (230b)).

[0058] The distributed training describes dividing the model and training data among several computers and training each split separately. Data parallelism and model parallelism are the two basic approaches that may be used to implement distributed model training. The data parallelism defines that the data is divided depending on the number of worker nodes present in the system (or wireless network (100)). All workers may use the same procedure on various data partitions. A single coherent output results from having the exact model available to all worker nodes. Thus, the data samples are distributed independently or identically, which is valid for most ML techniques. Further, the model parallelism is a machine learning approach for distributing a neural network model over several computers or computing devices. In model parallelism, the neural network model's parameters are divided across several machines, enabling each machine to process a piece of the input data and determine the appropriate output.

[0059] The ML training function may be located within the management system or in the NF (e.g. gNB or NWDAF), i.e. the worker node for training. Each node has different computing resources and storage capacity based on physical infrastructure such as CPU / GPU / DPU, memory, storage, and network bandwidth. In order to obtain load balance between nodes and maximize the efficiency of resource utilization, splitting up the training may be necessary and involving multiple training functions according to the actual situation of nodes may be needed. Thus, aspects of distributed training need to be supported in the management systems. A crucial part of enabling distributed training is providing training requirement to the producer so that the first producer (220) implements training functionality fulfilling the consumer's requirements.

[0060] Thus, the present disclosure may provide solutions that involves the consumer (210) providing a set of requirements related to the DT to the first producer (220) that may be supported and taken care of by the DT process created by the first producer (220). The target inference location and the data spilt indication may be provided as requirements. The target inference location may specify the location of the inference function that will host the trained ML model. The target inference location information may be defined as geographical coordinate, geographical area as the convex polygon with a geo coordinate specifying the corner of the convex polygon. Further, the data split indication information indicates the Boolean attribute specifying whether the provided training data should be split among the worker node or not. The value "true" specify that the training data can be spilt. Whereas value "false" specify that the training data shall not be split.

[0061] The solution proposes following NRM change for 3GPP TS 28.105.

[0062] New attribute in MLTrainingRequest IOC to capture the requirements as follows:

[0063]

[0064] Initially, the MnS consumer (210) may send a request to the MnS producer (220) for training the model using distributed training. The request contains the requirements to be fulfilled by the trained ML model. Based on the received request, the first producer (220) creates the ML training process and initiate the training. The training may include selecting the worker node as part the requirements submitted by the consumer (210). Further, the first producer (220) may send the response to the consumer (210) which may be the acknowledgement from the first producer (220). Particularly, the model training may be performed to ensure that resultant ML model satisfies all the requirements submitted by the consumer (210). This involves the first producer (220) to determine and apply the distribution strategy for ML training process, that is, split of ML model and / or data sets across the worker nodes, considering satisfying the training requirements from the consumer (210). Based on the location information from the consumer (210), the first producer (220) selects the worker nodes related to the given location only. If data split indication is present, the first producer (220) may use the indication to do data set distribution across the worker nodes.

[0065] Once the training is completed, the first producer (220) creates a ML training report. Assuming the consumer (210) may be subscribed for the creation notification, the first producer (220) send a notify managed object instance (MOI) creation notification to the consumer (210) notifying the completion of the ML training process. Further, the consumer (210) then decides to load the received ML model in the appropriate inference function hosted on a network node (e.g. gNB). To achieve the above-mentioned aspect, the consumer (210) may send a ML model loading request to the second producer (230). Then the inference producer may load the ML model and the second producer (230) may send a response to the consumer (210).

[0066] In other words, the distributed training management may follow the below steps:

[0067] At step 1: The MnS consumer (210) may send a request (e.g., create MOI (MLTrainingRequest{DTReq}) to the MnS Producer (220) for training a model using distributed training. The request contains the requirements to be fulfilled by the trained ML model.

[0068] At step 2: The first producer (220) may create the ML training process and initiate the training. The training may include selecting the worker node as part the requirements submitted by the consumer (210). At step 3: The first producer (220) may send a response to the consumer (210).

[0069] Step 4: the operation indicates that the model training is done to ensure that resultant ML Model satisfies all the requirements submitted by the consumer (210). This step involves that the first producer (220) determines and applies the distribution strategy for ML training process, that is, split of ML model and / or data sets across the worker nodes, considering to satisfy the training requirements from the consumer (210). Based on the location information from the consumer (210), the first producer (220) selects the worker nodes related to the given location only. If data split indication is present, the first producer (220) uses this indication to do data set distribution across worker nodes.

[0070] At step 5: On training completion, the first producer (220) creates a ML training report.

[0071] At step 6: Assuming the consumer (210) may be subscribed for the creation notification, the first producer (220) may send a notify MOI creation notification to the consumer (210) notifying the completion of the ML training process.

[0072] At steps 7-9: The consumer (210) then decides to load the received ML model in the appropriate inference function hosted on the network node (e.g. gNB or the like). To achieve this the consumer (210) may send a ML model loading request to the second producer (230) and the inference producer (230) may load the ML model where the second producer (230) may send a response to the consumer (210).

[0073] The proposed method enables efficient and timely execution of ML training for the large ML model requiring ample amount of time and resources.

[0074] FIG. 4 shows various hardware components of the first producer (220), according to embodiments as disclosed herein. In an embodiment, the first producer (220) includes a processor (410), a communicator (420), a memory (430), and a data handling controller (440). The processor (410) is coupled with the communicator (420), the memory (430), the data handling controller (440).

[0075] The data handling controller (440) receives the create MOI request for ML Training Request IOC (Information Object Class) from a consumer (210). The create MOI request includes at least one of: a first parameter and a second parameter. The first parameter includes at least one of: a training data generation indication, a generative AI data context and a genuineness confidence threshold. The second parameter includes at least one of: a target inference location and a data split allowed input. The first parameter provides information related to the requirements of generative artificial intelligence, and where in the second parameter provides information related to the expectations of distributed training.

[0076] In an embodiment, the training data generation indication defines whether the create MOI request is to generate the training data to be used for training an ML model for an analytics type provided using an inference name attribute. In an embodiment, the generative AI data context defines the context for which the data is to be generated based on the location and the time. The location defines the geographical location for which the training data is to be generated and the time defines the time duration for which the training data is to be generated. In an embodiment, the genuineness confidence threshold represents a required threshold to be met for asserting at least one of: genuineness of the generated data and a statistical consistency of the generated data, wherein the genuineness confidence threshold is provided by the consumer (210) with the create MOI request.

[0077] In an embodiment, the target inference location specifies the location of the inference function that hosts the trained ML model. The target inference location is defined as a geographical coordinate, a geographical area as a convex polygon with the geo coordinate specifying the corner of the convex polygon. In an embodiment, the data split allowed input specifies whether the provided training data should be split not using a first value and a second value, wherein the first value specifies that the training data is split and the second value specifies that the training data is not split.

[0078] Further, the data handling controller (440) creates an ML training process based on the create MOI request. Further, the data handling controller (440) sends the create MOI response to the consumer (210). The create MOI response includes the ML training process.

[0079] In an embodiment, the data handling controller (440) trains the ML model using the generative AI model. Further, the data handling controller (440) creates the ML training report. The ML training report includes the generative AI genuineness confidence representing at least one of: genuineness of the data and the statistical consistency of the data that is generated using the trained model. The generative AI genuineness confidence is created by the first producer (220) with the ML training report.

[0080] In an embodiment, the data handling controller (440) notifies the creation of ML training report including the generated data.

[0081] In an embodiment, the data handling controller (440) sends the create MOI including the ML training request having a generated data to the second producer (230). The second producer (230) creates the ML training process. Further, the data handling controller (440) receives the create MOI response from the second producer (230) based on the create MOI comprising the ML training request having the generated data.

[0082] In an embodiment, the data handling controller (440) updates the ML training report upon receiving the create MOI response from the second producer (230) based on the create MOI comprising a ML training request having a generated data. Further, the data handling controller (440) sends a notification about the MOI creation change corresponding to the ML model.

[0083] In an embodiment, the data handling controller (440) sends the notify MOI creation notification to the consumer (210) notifying the completion of the ML training report, when the consumer (210) has subscribed for a creation notification corresponding the ML training report.

[0084] The data handling controller (440) is 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 may optionally be driven by firmware.

[0085] The processor (410) may include one or a plurality of processors. The one or the plurality of processors may be a general-purpose processor, such as a central processing unit (CPU), an application processor (AP), or the like, a graphics-only processing unit such as a graphics processing unit (GPU), a visual processing unit (VPU), and / or an AI-dedicated processor such as a neural processing unit (NPU). The processor (410) may include multiple cores and is configured to execute the instructions stored in the memory (430).

[0086] Further, the processor (410) is configured to execute instructions stored in the memory (430) and to perform various processes. The communicator (420) is configured for communicating internally between internal hardware components and with external devices via one or more networks. The memory (430) also stores instructions to be executed by the processor (410). The memory (430) may include non-volatile storage elements. Examples of such non-volatile storage elements may include magnetic hard discs, optical discs, floppy discs, flash memories, or forms of electrically programmable memories (EPROM) or electrically erasable and programmable (EEPROM) memories. In addition, the memory (430) may, in some examples, be considered a non-transitory storage medium. The term "non-transitory" may indicate that the storage medium is not embodied in a carrier wave or a propagated signal. However, the term "non-transitory" should not be interpreted that the memory (430) is non-movable. In certain examples, a non-transitory storage medium may store data that can, over time, change (e.g., in Random Access Memory (RAM) or cache).

[0087] Further, at least one of the plurality of modules / controller may be implemented through the data driven model (e.g., AI model / the ML model) using a data driven controller (not shown). The data driven controller can be an ML model based controller and an AI model based controller. A function associated with the AI model may be performed through the non-volatile memory, the volatile memory, and the processor (410). The processor (410) may include one or a plurality of processors. At this time, one or a plurality of processors may be a general purpose processor, such as a central processing unit (CPU), an application processor (AP), or the like, a graphics-only processing unit such as a graphics processing unit (GPU), a visual processing unit (VPU), and / or an AI-dedicated processor such as a neural processing unit (NPU).

[0088] The one or a plurality of processors control the processing of the input data in accordance with a predefined operating rule or AI model stored in the non-volatile memory and the volatile memory. The predefined operating rule or artificial intelligence model is provided through training or learning.

[0089] Here, being provided through learning means that a predefined operating rule or AI model of a desired characteristic is made by applying a learning algorithm to a plurality of learning data. The learning may be performed in a device itself in which AI according to an embodiment is performed, and / o may be implemented through a separate server / system.

[0090] The AI model may include of a plurality of neural network layers. Each layer has a plurality of weight values, and performs a layer operation through calculation of a previous layer and an operation of a plurality of weights. Examples of neural networks include, but are not limited to, convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN), restricted Boltzmann Machine (RBM), deep belief network (DBN), bidirectional recurrent deep neural network (BRDNN), generative adversarial networks (GAN), and deep Q-networks.

[0091] The learning algorithm is a method for training a predetermined target device (for example, a robot) using a plurality of learning data to cause, allow, or control the target device to make a determination or prediction. Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0092] Although FIG. 4 shows various hardware components of the first producer (220) but it is to be understood that other embodiments are not limited thereon. In other embodiments, the first producer (220) may include less or more number of components. Further, the labels or names of the components are used only for illustrative purposes and does not limit the scope of the disclosure. One or more components can be combined together to perform the same or substantially similar function in the first producer (220).

[0093] FIG. 5 shows various hardware components of the consumer (210), according to embodiments as disclosed herein. In an embodiment, the consumer (210) includes a processor (510), a communicator (520), a memory (530), and a data handling controller (540). The processor (510) is coupled with the communicator (520), the memory (530), the data handling controller (540).

[0094] The data handling controller (540) sends the create MOI request for ML Training Request IOC to the first producer (220). The create MOI request includes at least one of: the first parameter and the second parameter. The first parameter includes at least one of: the training data generation indication, the generative artificial intelligence (AI) data context and the genuineness confidence threshold. The second parameter includes at least one of: the target inference location and the data split allowed input. Further, the data handling controller (540) receives the create MOI response from the first producer (220) based on the create MOI request, where the create MOI response includes the ML training process created at the first producer (220). The first parameter provides information related to the requirements of generative artificial intelligence, and the second parameter provides information related to the expectations of distributed training.

[0095] In an embodiment, the data handling controller (540) receives the notification about a creation of ML training report including the generated data. In an embodiment, the data handling controller (540) receives the notification about the MOI creation change corresponding to the ML model, from the producer, upon updating the ML training report on receiving a create MOI response from the second producer (230) based on the create MOI comprising the ML training request having the generated data.

[0096] In an embodiment, the data handling controller (540) receives the notify MOI creation notification, from the first producer (220), notifying the completion of the training report at the first producer (220), when the consumer (210) has subscribed for a creation notification corresponding the ML training report.

[0097] In an embodiment, the data handling controller (540) sends the ML model loading request to the second producer (230). Further, the data handling controller (540) receives the response from the second producer (230) based on the ML model loading request.

[0098] The data handling controller (540) is 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 may optionally be driven by firmware.

[0099] The processor (510) may include one or a plurality of processors. The one or the plurality of processors may be a general-purpose processor, such as a central processing unit (CPU), an application processor (AP), or the like, a graphics-only processing unit such as a graphics processing unit (GPU), a visual processing unit (VPU), and / or an AI-dedicated processor such as a neural processing unit (NPU). The processor (510) may include multiple cores and is configured to execute the instructions stored in the memory (530).

[0100] Further, the processor (510) is configured to execute instructions stored in the memory (530) and to perform various processes. The communicator (520) is configured for communicating internally between internal hardware components and with external devices via one or more networks. The memory (530) also stores instructions to be executed by the processor (510). The memory (530) may include non-volatile storage elements. Examples of such non-volatile storage elements may include magnetic hard discs, optical discs, floppy discs, flash memories, or forms of electrically programmable memories (EPROM) or electrically erasable and programmable (EEPROM) memories. In addition, the memory (530) may, in some examples, be considered a non-transitory storage medium. The term "non-transitory" may indicate that the storage medium is not embodied in a carrier wave or a propagated signal. However, the term "non-transitory" should not be interpreted that the memory (530) is non-movable. In certain examples, a non-transitory storage medium may store data that can, over time, change (e.g., in Random Access Memory (RAM) or cache).

[0101] Further, at least one of the plurality of modules / controller may be implemented through the data driven model (e.g., AI model / the ML model) using a data driven controller (not shown). The data driven controller can be an ML model based controller and an AI model based controller. A function associated with the AI model may be performed through the non-volatile memory, the volatile memory, and the processor (510). The processor (510) may include one or a plurality of processors. At this time, one or a plurality of processors may be a general purpose processor, such as a central processing unit (CPU), an application processor (AP), or the like, a graphics-only processing unit such as a graphics processing unit (GPU), a visual processing unit (VPU), and / or an AI-dedicated processor such as a neural processing unit (NPU).

[0102] The one or a plurality of processors control the processing of the input data in accordance with a predefined operating rule or AI model stored in the non-volatile memory and the volatile memory. The predefined operating rule or artificial intelligence model is provided through training or learning.

[0103] Here, being provided through learning means that a predefined operating rule or AI model of a desired characteristic is made by applying a learning algorithm to a plurality of learning data. The learning may be performed in a device itself in which AI according to an embodiment is performed, and / o may be implemented through a separate server / system.

[0104] The AI model may include of a plurality of neural network layers. Each layer has a plurality of weight values, and performs a layer operation through calculation of a previous layer and an operation of a plurality of weights. Examples of neural networks include, but are not limited to, convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN), restricted Boltzmann Machine (RBM), deep belief network (DBN), bidirectional recurrent deep neural network (BRDNN), generative adversarial networks (GAN), and deep Q-networks.

[0105] The learning algorithm is a method for training a predetermined target device (for example, a robot) using a plurality of learning data to cause, allow, or control the target device to make a determination or prediction. Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0106] Although FIG. 5 shows various hardware components of the consumer (210) but it is to be understood that other embodiments are not limited thereon. In other embodiments, the consumer (210) may include less or more number of components. Further, the labels or names of the components are used only for illustrative purposes and does not limit the scope of the disclosure. One or more components can be combined together to perform the same or substantially similar function in the consumer (210).

[0107] FIG. 6 is a flow chart (S600) illustrating a method, implemented by the first producer (220), for managing the data in the wireless network (100). The operations (S602-S604) are handled by the data handling controller (440).

[0108] At S602, the method includes receiving the create MOI request for ML Training Request IOC from the consumer (210). The create MOI request includes at least one of: the first parameter and the second parameter. The first parameter includes at least one of: the training data generation indication, the generative AI data context and the genuineness confidence threshold. The second parameter includes at least one of: the target inference location and the data split allowed input. At S604, the method includes creating the ML training process based on the create MOI request.

[0109] FIG. 7 is a flow chart (S700) illustrating a method, implemented by the consumer (210), for managing the data in the wireless network (100), according to embodiments as disclosed herein. The operations (S702-S704) are handled by the data handling controller (540). At S702, the method includes sending the create MOI request for ML Training Request IOCto the first producer (220). The create MOI request includes at least one of: the first parameter and the second parameter. The first parameter includes at least one of: the training data generation indication, the generative AI data context and the genuineness confidence threshold. The second parameter includes at least one of: the target inference location and the data split allowed input. At S704, the method includes receiving the create MOI response from the first producer (220) based on the create MOI request. The create MOI response includes the ML training process created at the first producer (220).

[0110] The various actions, acts, blocks, steps, or the like in the flow charts (S600 and S700) may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some of the actions, acts, blocks, steps, or the like may be omitted, added, modified, skipped, or the like without departing from the scope of the disclosure.

[0111] The specific examples provided to explain the embodiments according to the present disclosure are merely a combination of each standard, method, detail method, and operation, and the various embodiments described herein can be performed through a combination of at least two or more techniques among the various techniques described. In addition, at this time, it can be performed according to a method determined through a combination of one or at least two or more of the aforementioned techniques. For example, it may be possible to perform a combination of parts of the operation of one embodiment with parts of the operation of another embodiment.

[0112] The embodiments disclosed herein can be implemented through at least one software program running on at least one hardware device and performing network management functions to control the elements. The elements shown in figures can be at least one of a hardware device, or a combination of hardware device and software module.

[0113] 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 at least one embodiment, 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

A method performed by a management service (MnS) producer, the method comprising:receiving, from an MnS consumer, a machine learning (ML) training request information for an ML model training, wherein the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split;identifying a preference of the MnS consumer as to whether the training data is to be split, based on the data split indication; andtransmitting, to the MnS consumer, a response associated with the ML training request information.The method of claim 1, wherein in case that the data split indication has a first value, the data split indication indicates that the training data is to be split, andwherein in case that the data split indication has a second value, the data split indication indicates that the training data is not to be split.The method of claim 2, wherein the data split indication has a boolean attribute, andwherein the first value is 'true' and the second value is 'false'.The method of claim 1, wherein the ML training request information includes a distributed training expectation information, wherein the distributed training expectation information includes the data split indication.The method of claim 1, wherein the data split indication is at least one of a readable attribute, a writeable attribute, a variant attribute, or a notifyable attribute.A management service (MnS) producer, comprising:a controller;a memory storing instructions; andat least one processor communicatively coupled to the controller and the memory, configured to execute the instructions stored in the memory to cause the MnS producer to:receive, from an MnS consumer, a machine learning (ML) training request information for an ML model training, wherein the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split;identify a preference of the MnS consumer as to whether the training data is to be split, based on the data split indication; andtransmit, to the MnS consumer, a response associated with the ML training request information.The MnS producer of claim 6, wherein in case that the data split indication has a first value, the data split indication indicates that the training data is to be split, andwherein in case that the data split indication has a second value, the data split indication indicates that the training data is not to be split.The MnS producer of claim 7, wherein the data split indication has a boolean attribute, andwherein the first value is 'true' and the second value is 'false'.The MnS producer of claim 6, wherein the ML training request information includes a distributed training expectation information, wherein the distributed training expectation information includes the data split indication.The MnS producer of claim 6, wherein the data split indication is at least one of a readable attribute, a writeable attribute, a variant attribute, or a notifyable attribute.A method performed by a management service (MnS) consumer, the method comprising:determining whether a training data is to be split;transmitting, to an MnS producer, a machine learning (ML) training request information for an ML model training, wherein the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split; andreceiving, from the MnS producer, a response associated with the ML training request information.The method of claim 11, wherein the ML training request information includes a distributed training expectation information, wherein the distributed training expectation information includes the data split indication.The method of claim 11, wherein in case that the data split indication has a first value, the data split indication indicates that the training data is to be split, andwherein in case that the data split indication has a second value, the data split indication indicates that the training data is not to be split.A management service (MnS) consumer, comprising:a controller;a memory storing instructions; andat least one processor communicatively coupled to the controller and the memory, configured to execute the instructions stored in the memory to cause the MnS consumer to:determine whether a training data is to be split;transmit, to an MnS producer, a machine learning (ML) training request information for an ML model training, wherein the ML training request information includes a data split indication indicating whether a training data for the ML model training is to be split; andreceive, from the MnS producer, a response associated with the ML training request information.The MnS consumer of claim 14, wherein the ML training request information includes a distributed training expectation information, wherein the distributed training expectation information includes the data split indication.

Images