Signaling for switching between phases in artificial intelligence / machine learning

Abstract

A method is provided that includes receiving from a network multiple phase configurations for a machine learning (ML) functionality. The method includes activating a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations. The method includes sending to the network a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode. And the method includes receiving from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

Description

SIGNALING FOR SWITCHING BETWEEN PHASES IN ARTIFICIAL INTELLIGENCE I MACHINE LEARNINGTECHNOLOGICAL FIELD

[0001] The present disclosure relates generally to telecommunications and, in particular, to artificial intelligence (Al) / machine learning (ML) in a telecommunications system.BACKGROUND

[0002] A telecommunications system can be seen as a facility that enables communication sessions between two or more entities such as user terminals, base stations and / or other nodes by providing carriers between the various entities involved in the communications path. A telecommunications system can be provided for example by means of a communication network and one or more compatible communication devices. The communication sessions may comprise, for example, communication of data for carrying communications such as voice, video, electronic mail (email), text message, multimedia and / or content data and so on. Non-limiting examples of services provided comprise two-way or multi-way calls, data communication or multimedia services and access to a data network system, such as the Internet.

[0003] In a wireless telecommunications system, at least a part of a communication session between at least two stations occurs over a wireless link. Examples of wireless telecommunications systems comprise public land mobile networks (PLMN), satellite based communication systems and different wireless local networks, for example wireless local area networks (WLAN). Some wireless systems can be divided into cells, and are therefore often referred to as cellular systems.

[0004] A user can access the telecommunications system by means of an appropriate communication device or terminal. A communication device of a user may be referred to as user equipment (UE) or user device. A communication device is provided with an appropriate signal receiving and transmitting apparatus for enabling communications, for example enabling access to a communication network or communications directly with other users. The communication device may access a carrier provided by a station, forexample a base station of a cell, and transmit and / or receive communications on the carrier.

[0005] The telecommunications system and associated devices typically operate in accordance with a given standard or specification which sets out what the various entities associated with the communication system are permitted to do and how operations should be achieved. Communication protocols and / or parameters which shall be used for connection of the various entities are also typically defined. One example of a telecommunications system is the Universal Mobile Telecommunications System (UMTS). Other examples of telecommunications systems are Long-Term Evolution (LTE), LTE Advanced and the so-called 5G or New Radio (NR) networks. NR is being standardized by the 3rd Generation Partnership Project (3GPP).BRIEF SUMMARY

[0006] Example implementations of the present disclosure are directed to telecommunications and, in particular, to artificial intelligence (Al) / machine learning (ML) in a telecommunications system. The present disclosure includes, without limitation, the following example implementations.

[0007] Some example implementations provide an apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to at least: receive from a network multiple phase configurations for a machine learning (ML) functionality; activate a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; send to the network a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and receive from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0008] Some example implementations provide a method comprising: receiving from a network multiple phase configurations for a machine learning (ML) functionality;activating a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; sending to the network a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and receiving from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0009] Some example implementations provide an apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to at least: configure a user equipment (UE) with multiple phase configurations for a machine learning (ML) functionality; receive from the UE an indication of activation of a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; receive from the UE a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and send to the UE an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0010] Some example implementations provide a method comprising: configuring a user equipment (UE) with multiple phase configurations for a machine learning (ML) functionality; receiving from the UE an indication of activation of a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; receiving from the UE a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and sending to the UE an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0011] Some example implementations provide an apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to at least: determine non-applicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the ML functionality is deployed for inference operation; send to a network a report that indicates non-applicability of the one or more inference configurations; receive from the network an indication to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode; and based on the indication from the network, switch to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or finetune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0012] Some example implementations provide a method comprising: determining non-applicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the ML functionality is deployed for inference operation; sending to a network a report that indicates non-applicability of the one or more inference configurations; receiving from the network an indication to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode; and based on the indication from the network, switching to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or fine-tune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0013] Some example implementations provide an apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to at least: receive from a user equipment (UE) a report that indicates non-applicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the MLfunctionality is deployed for inference operation; make a determination to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode, based on the non-applicability of the one or more inference configurations; and based on the determination, send to the UE an indication to switch to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or fine-tune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0014] Some example implementations provide a method comprising: receiving from a user equipment (UE) a report that indicates non-applicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the ML functionality is deployed for inference operation; making a determination to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode, based on the non-applicability of the one or more inference configurations; and based on the determination, sending to the UE an indication to switch to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or finetune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0015] These and other features, aspects, and advantages of the present disclosure will be apparent from a reading of the following detailed description together with the accompanying figures, which are briefly described below. The present disclosure includes any combination of two, three, four or more features or elements set forth in this disclosure, regardless of whether such features or elements are expressly combined or otherwise recited in a specific example implementation described herein. The present disclosure is intended to be read holistically such that any separable features or elements of the disclosure, in any of its aspects and example implementations, should be viewed as combinable unless the context of the disclosure clearly dictates otherwise.

[0016] It will therefore be appreciated that this Brief Summary is provided merely for purposes of summarizing some example implementations so as to provide a basicunderstanding of some aspects of the disclosure. Accordingly, it will be appreciated that the above described example implementations are merely examples and should not be construed to narrow the scope or spirit of the disclosure in any way. Other example implementations, aspects and advantages will become apparent from the following detailed description taken in conjunction with the accompanying figures which illustrate, by way of example, the principles of some described example implementations.BRIEF DESCRIPTION OF THE FIGURE(S)

[0017] Having thus described example implementations of the disclosure in general terms, reference will now be made to the accompanying figures, which are not necessarily drawn to scale, and wherein:

[0018] FIG. 1 illustrates a telecommunications system that includes one or more public land mobile networks (PLMNs) coupled to one or more external data networks, according to some example implementations of the present disclosure;

[0019] FIG. 2 illustrates a deployment of a PLMN, according to some example implementations;

[0020] FIG. 3 illustrates different phases in a reinforcement learning (RL) model, according to some example implementations;

[0021] FIG. 4 illustrates a functional framework for artificial intelligence (Al) / machine learning (ML) with a number of lifecycle management (LCM) procedures, according to some example implementations;

[0022] FIG. 5 illustrates different approaches to trigger a configuration change for phase switching for user equipment (UE)-side models, according to some example implementations;

[0023] FIG. 6 is a signaling chart for an implementation-specific approach to trigger a configuration change, according to some example implementations;

[0024] FIG. 7 is a flowchart of procedures for RL model phase switching between a training phase (phase 1) and an inference phase (phase 2), according to some example implementations;

[0025] FIG. 8 is a flowchart of procedures for RL model phase switching between the training phase, the inference phase, and a retraining phase (phase 3), according to some example implementations;

[0026] FIG. 9 is a signaling chart of procedures for initial configuration and applicability reporting after transition from the training phase to the inference phase, according to some example implementations;

[0027] FIG. 10 is a signaling chart of procedures for applicability reporting after transitions between the inference phase and the retraining phase, according to some example implementations;

[0028] FIGS. HAand 11B are flowcharts illustrating various steps in a method according to various example implementations;

[0029] FIG. 12 is a flowchart illustrating various steps in a method according to various example implementations;

[0030] FIGS. 13A, 13B and 13C are flowcharts illustrating various steps in a method according to various example implementations;

[0031] FIGS. 14Aand 14B are flowcharts illustrating various steps in a method according to various example implementations; and

[0032] FIG. 15 illustrates an apparatus according to some example implementations.DETAILED DESCRIPTION

[0033] Some implementations of the present disclosure will now be described more fully hereinafter with reference to the accompanying figures, in which some, but not all implementations of the disclosure are shown. Indeed, various implementations of the disclosure may be embodied in many different forms and should not be construed as limited to the implementations set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. Like reference numerals refer to like elements throughout.

[0034] Unless specified otherwise or clear from context, references to first, second or the like should not be construed to imply a particular order. A feature described as being above another feature (unless specified otherwise or clear from context) may instead bebelow, and vice versa; and similarly, features described as being to the left of another feature else may instead be to the right, and vice versa. Also, while reference may be made herein to quantitative measures, values, geometric relationships or the like, unless otherwise stated, any one or more if not all of these may be absolute or approximate to account for acceptable variations that may occur, such as those due to engineering tolerances or the like.

[0035] As used herein, unless specified otherwise or clear from context, the “or” of a set of operands is the “inclusive or” and thereby true if and only if one or more of the operands is true, as opposed to the “exclusive or” which is false when all of the operands are true. Thus, for example, “[A] or [B]” is true if [A] is true, or if [B] is true, or if both [A] and [B] are true. Further, the articles “a” and “an” mean “one or more,” unless specified otherwise or clear from context to be directed to a singular form. Furthermore, it should be understood that unless otherwise specified, the terms “data,” “content,” “digital content,” “information,” and similar terms may be at times used interchangeably. The term “network” may refer to a group of interconnected computers including clients and servers; and within a network, these computers may be interconnected directly or indirectly by various means including via one or more switches, routers, gateways, access points or the like.

[0036] The present disclosure discusses systems and architectures that, while specific terms may be used, are broadly applicable across various technologies. For instance, while the present disclosure may reference technologies from 3 GPP such as Global System for Mobile Communications (GSM), UMTS, LTE, LTE Advanced, 5GNR, 5G Advanced, and 6G, the present disclosure is equally relevant to non-3GPP technologies like IEEE 802, Bluetooth, and Bluetooth Low Energy. Example implementations of the present disclosure described herein also mention public land mobile networks (PLMNs) and mobile network operators (MNOs), but example implementations are similarly applicable to standalone non-public networks (SNPNs) and the private entities operating these networks. Furthermore, although some examples and figures focus on radio access networks (RANs) and 3 GPP access, example implementations are applicable to any type of network access. This includes not only 5G or 6G 3GPP access but also non-3GPP access, such as wireline access, untrusted non-3GPP access, and trusted non-3GPP accessusing wireless access gateway function (W-AGF), non-3GPP interworking function (N3IWF), or trusted non-3GPP gateway function (TNGF) to connect to a 5G or 6G core network.

[0037] Further, as used in this application, the term “circuitry” may refer to one or more or all of the following: (a) hardware-only circuit implementations (such as implementations in only analog and / or digital circuitry); (b) combinations of hardware circuits and software, such as (as applicable): (i) a combination of analog and / or digital hardware circuit(s) with software / firmware and (ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions); or (c) hardware circuit(s) and / or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (e.g., firmware) for operation, but the software may not be present when it is not needed for operation.

[0038] The above definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.

[0039] FIG. 1 illustrates a telecommunications system 100 according to various example implementations of the present disclosure. The telecommunications system generally includes one or more telecommunications networks. As shown, for example, the system includes one or more PLMNs 102 coupled to one or more other external data networks 104 - notably including a wide area network (WAN) such as the Internet. As will be appreciated, a PLMN may be deployed in a number of different manners. Some deployments of 4GLTE and 5GNR in particular are considered standalone (SA) deployments. Other deployments combine 4G LTE and 5G technologies, and are referred to as non-standalone (NSA) deployments.

[0040] Each of the PLMNs 102 includes a core network (CN) 106 backbone, such as the Evolved Packet Core (EPC) of 4G LTE, the 5G core network (5GC) (at times referred to as the NGC) of 5G NR, and the 6G core network (6GC) of 6G; and each of the core networks and the Internet are coupled to one or more RANs 108, air interfaces or the like that implement one or more radio access technologies (RATs). Examples of these RANs include the evolved UMTS terrestrial radio access network (E-UTRAN) of 4G LTE, the next generation (NG) radio access network (NG-RAN) of 5G NR, and the 6G RAN. As used herein, a “network device” refers to any suitable device at a network side of a telecommunications network. Examples of suitable network devices are described in greater detail below.

[0041] Examples of RATs include 3GPP radio access technologies such as GSM, CDMA2000 IxEV-DO (HRPD), CDMA2000 lx (IxRTT), UTRA, E-UTRA, 5GNR, 5G Advanced, and 6G. Other examples of RATs include IEEE 802 technologies such as IEEE 802.11 (Wi-Fi), IEEE 802.15 (including 802.15.1 (WPAN / Bluetooth), 802.15.4 (Zigbee) and 802.15.6 (WBAN)), Bluetooth, Bluetooth Low Energy (BLE), ultra wideband (UWB), and the like. Generally, a RAT may refer to any 2G, 3G, 4G, 5G, 6G or higher generation RAT and their different versions, as well as to any other RAT that may be arranged to interwork with such a mobile communication technology to provide access to the CN 106 of a MNO.

[0042] The telecommunications system 100 also includes one or more radio units that may be varyingly known as user equipment (UE) 110, terminal device, terminal equipment, mobile station or the like. The UE is generally a device configured to communicate with a network device or a further UE in a telecommunications network. The UE may be a portable computer (e.g., laptop, notebook, tablet computer), mobile phone (e.g., cell phone, smartphone), wearable computer (e.g., smartwatch), or the like. In other examples, the UE may be an Internet of things (loT) device, an industrial loT (IIoT device), a vehicle equipped with a vehicle-to-everything (V2X) communication technology, or the like. In some examples, as referenced by 3 GPP, the UE may be a narrowband loT (NB-IoT) device, an enhanced machine-type communication (eMTC) device, a reduced capability (RedCap) device, an ambient loT device, or the like.

[0043] In operation, these UEs 110 may connect to one or more of theRANs 108 according to their particular RATs to thereby access a particular CN 106 of a PLMN 102, or to access one or more of the external data networks 104 (e.g., the Internet). The external data network may provide Internet access, operator services, 3rd party services, etc. For example, the International Telecommunication Union (ITU) has classified 5G mobile network services into three categories: enhanced mobile broadband (eMBB), ultra-reliable and low-latency communications (URLLC), and massive machine type communications (mMTC) or massive internet of things (MIoT).

[0044] In various examples, a RAN 108 may be configured as one or more macrocells, microcells, picocells, femtocells or the like. The RAN may generally include one or more RAN nodes that interact with UEs 110. In various examples, a RAN node may be referred to as a base station (BS), access point (AP), base transceiver station (BTS), Node B (NB), evolved NB (eNB), macro BS, NB (MNB) or eNB (MeNB), home BS, NB (HNB) or eNB (HeNB), next generation NB (gNB), enhanced gNB (en-gNB), next generation eNB (ng-eNB), 6GNB (6gNB), or the like. The term ‘gNB’ in 5GNR may correspond to the eNB in 4G LTE. Also, a NG-RAN node may refer to a gNB or a ng-eNB. And unless otherwise specified, a gNB in 5G NR or a 6gNB in 6G may at times be more generally referred to as a gNB.

[0045] The RAN 108 may include some type of network controlling / governing entity responsible for control of the RAN nodes. The network controlling / governing entity and RAN node may be separate or integrated into a single apparatus. The network controlling / governing entity may include processing circuity configured to carry out various management functions, etc. The processing circuity may be associated with a memory, computer-readable storage medium or database for maintaining information required in the management functions.

[0046] FIG. 2 illustrates a deployment of a PLMN 102, such as a 5G NR deployment or a 6G deployment. As shown, the RAN 108 (e.g., NG-RAN, 6G RAN) includes one or more RAN nodes 202 (e.g., gNBs, 6gNBs) configured to connect one or more UEs 110 to the RAN to thereby access the CN 106 (e.g., 5GC, 6GC). In some deployments, operations of a gNB or other a RAN node may be distributed or functionally split into components including one or more remote radio head (RRHs) or radio units (RUs), and abaseband unit (BBU); and in some architectures, the BBU may be split into a central / centralized unit (CU) (central node) and a distributed unit (DU) (distributed node). The CU may be, for example, a server, host or node. In some architectures, the RRH / RU and DU may be collocated. It is also possible that node operations may be distributed among a plurality of servers, hosts or nodes.

[0047] It should also be understood that the distribution of work between core network operations and RAN node operations may vary depending on implementation. A 5G network architecture may be based on a so-called CU-DU split. One gNB-CU (a CU) may control one or more gNB-DUs (DUs). The gNB-CU may control a plurality of spatially separated gNB-DUs, acting at least as transmit / receive (Tx / Rx) nodes. In some example implementations, however, the gNB-DUs may include, for example, a radio link control (RLC), medium access control (MAC) layer and a physical (PHY) layer, whereas the gNB-CU may include the layers above the RLC layer, such as a packet data convergence protocol (PDCP) layer, a radio resource control (RRC), and an internet protocol (IP) layer. Other functional splits are also possible. It is considered that skilled person is familiar with the OSI model and the functionalities within each layer.

[0048] In some example implementations, the server or CU may generate a virtual network through which the server communicates with the radio node. In general, virtual networking may involve a process of combining hardware and software network resources and network functionality into a single, software-based administrative entity, a virtual network. Such virtual network may provide flexible distribution of operations between the server and the radio head / node. In practice, any digital signal processing task may be performed in either the CU or the DU, and the boundary where the responsibility is shifted between the CU and the DU may be selected according to implementation.

[0049] In 3 GPP, an investigation is ongoing regarding the use of artificial intelligence(Al) / machine learning (ML) technologies and algorithms for enhanced performance and / or reduced complexity / overhead by cementing the foundation for future air-interface use cases. Initial use cases under investigation include channel state information (CSI) feedback enhancement (e.g., overhead reduction, improved accuracy), beam management (e.g., beam prediction in time, and / or spatial domain for overhead and latency reduction), positioning accuracy enhancements, and mobility.

[0050] A general AI / ML framework has been provided that identifies common notation and terminology for AI / ML-related functions, procedures and interfaces. The stages of AI / ML include exploration or model generation (e.g., model training, model validation, model testing, etc.) and exploitation or inference operation (e.g., input / output, pre- / post-process, etc.). In this regard, life cycle management (LCM) of an AI / ML model may include, for example, data collection and exploration, model training, model deployment, model inference, model monitoring and model updating. For an LCM procedure, it may be the case that an AI / ML model (at times more generally referred to as a “ML” model, or more simply a “model”) has a model identifier (ID) with associated information. Additionally or alternatively, it may be the case that a given functionality is provided by some AI / ML operations.

[0051] For UE-side models and the UE-part of two-sided models, UE capability reporting may be taken as a starting point for the identification of AI / ML functionality (at times more simply referred to as ML functionality). For AI / ML functionality identification and functionality-based LCM of UE-side models and / or UE-part of two- sided models, functionality refers to an AI / ML-enabled feature or feature group (FG) enabled by one or more configurations supported based on conditions indicated by UE capability. Likewise, functionality-based LCM operates based on at least one configuration of AI / ML-enabled feature / FG or specific configurations of an AI / ML- enabled feature / FG. For AI / ML model identification and model-ID-based LCM of UE- side models and / or UE-part of two-sided models, model-ID-based LCM operates based on identified models. In this regard, a model may be associated with specific configurations / conditions associated with UE capability of an AI / ML-enabled feature / FG and additional conditions (e.g., scenarios, sites, and datasets) as determined / identified between UE-side and network (NW)-side.

[0052] In the general AI / ML framework, functionalities may be distinguished between supported functionalities, applicable functionalities and activated functionalities. Supported functionalities refer to functionalities that a UE 110 can indicate to the network (e.g., RAN node 202 of RAN 108) by using UE capability information (via RRC signaling / LTE positioning protocol (LPP) signaling). These are the AI / ML-enabled features / FGs that a UE is capable of supporting. Applicable functionalities refer tofunctionalities that the UE is ready to apply for inference. Applicable functionalities are a subset of the supported functionalities that are relevant and suitable for current network conditions or use case. Activated functionalities refer to functionalities already enabled for performing inference. The activated functionalities are a subset of the applicable functionalities that are currently in use or enabled on the UE.

[0053] Conditions that identify AL / ML features / functionalities may be signaled through a UE capability message (in the RRC configuration), and are currently under the refinement within the 3GPP. Some proposals have been made on their structure (e.g., general-level conditions that are applicable regardless of the enabled use case, and usecase specific conditions) and contents. In particular, some proposals have been made for general conditions, as well as for CSI compression enhancement, beam management, and positioning specific conditions. As these proposals have demonstrated, the number of conditions and their potential content / values are extensive, and they are expected to only progress as new AI / ML use cases and functionalities are introduced. This will make it even more laborious to signal conditions through the UE capability message and the corresponding synchronization from the network side to process.

[0054] The general AI / ML framework defines additional conditions for an AI / ML- enabled feature / FG. In this regard, additional conditions refer to any aspects that are assumed for the training of the model but are not a part of UE capability for the AI / ML- enabled feature / FG. It does not imply that additional conditions are necessarily specified. Additional conditions can be divided into two categories: NW-side additional conditions and UE-side additional conditions. For inference for UE-side models, to ensure consistency between training and inference regarding NW-side additional conditions (if identified), a number of options may be taken as potential approaches (when feasible and necessary). In one of these options, information and / or indication on NW-side additional conditions is provided to the UE 110 by the network.

[0055] In some contexts, the UE 110 may need to determine whether a supported functionality is applicable (an applicable functionality) for a given condition. The UE may determine an applicable functionality based on NW-side additional conditions (if identified), UE-side additional conditions (the UE’s internal condition), model availability and / or performance monitoring. One example of a NW-side additionalcondition is an associated ID. The associated ID serves to align the assumptions made across training and at least inference, and may be monitoring of a model. Through this ID, the UE can detect the consistency across training and inference. For instance, the associated ID may impact the UE’s assumption on beams of set A / set B in beam management use case 1 and case 2. The format of the associated ID may be a sequence of bits representing network’s codebook configurations during training, such as antenna direction. Once UE is able to determine the applicable functionality with this information, the UE may need to report an indication of the applicable functionality(ies) to the network.

[0056] A number of AI / ML approaches are based on reinforcement learning (RL) or supervised learning (SL). One key difference between RL-based approaches and SL- based approaches is, depending on how well the RL model is converged, the model behavior can be different. FIG. 3 illustrates one example characterization of the behavior based on how well the model is converged. In phase 1 (a training phase), the model actively learns and acquires training data via exploration. In phase 2 (an inference phase, also referred to as a converged phase), the agent may consider being converged and start to exploit the acquired knowledge. During this phase, the RL agent may still perform no exploration or in a limited manner. Later, the agent may re-enter a learning phase to retrain, e.g., due to changes in the radio environment making the prior training data obsolete or incomplete. This phase may be referred to as phase 3 (a retraining phase).

[0057] From RL point of view there are no defined rules to identify these phases of a RL model. Furthermore, the phase of the RL model changes when the model is trained. The table below provides characteristics of the models in different phases.Characteristics of RL algorithms in different phases

[0058] ML management procedures such as LCM procedures may define procedures which are performed to accelerate delivery and maintain and ensure the quality of the AI / ML-based application. A similar approach with well-defined LCM procedures is studied in 3 GPP New-Radio (NR) Release 18 and Release 19 to manage an AI / ML model or functionality deployed in the air interface. However, in Release 18-19 AI / ML, it assumed that the models are trained offline and LCM procedures are not studied for RL use cases where the models are trained online while deployed in air interface.Furthermore, the LCM is categorized based on which level the LCM procedures are performed, on functionality or on logical model level based on model ID. It is often assumed that the model ID based LCM is not mandatory as not all use cases require the logical model specific details to perform (functionality-based) LCM procedures.

[0059] FIG. 4 illustrates a functional framework for AI / ML with a number of LCM procedures. In RL, since model is trained online based on the data collected from itsinteractions with the environment, the characteristic of the model evolves while it’s deployed in air interface. Hence, feature configurations need to consider the requirements during each RL phase identified in the table above. Furthermore, if the RL model is capable of being trained offline, the offline data collection requirements would be similar to that of SL use cases. The table below includes example feature configuration requirements based on the phase of the RL model for the initial training and inference phases. The retraining phase may be similar to the training phase.Example Feature Configuration Requirements for LCM Procedures in Different RLModel Phases

[0060] RL models may require online training to update the model as it gathers new experiences by interacting with the environment. However, in 3 GPP NR Release 18 and Release 19 AI / ML framework studies, online training and model update still exist as for future study (FFS) items without formal definitions. Hence, online training of RL model may be a procedure in which data is collected when the model is deployed or “online,” and the ML model may be trained / fine-tuned using that data. In general, online learning refers to algorithms that uses the new data samples for training, and then the data samples are discarded. Therefore, there may be some (near) real-time requirement for the data collection and training. This applies in some cases of RL, but not in all cases. For instance, in one example, the new training samples may be stored in a memory, and training may take place by using the memory every now and then (up to implementation), but there is no requirement on sequential or real-time use of the data. In practice, may bebetter to sample non-sequential data points for training, to avoid correlations in time domain.

[0061] As described, a phase of a model may be implemented in a phase mode of operation. In this regard, a training phase mode (at times more simply be referred to as a training mode) may implement the training phase. An inference phase mode (also referred to as an inference or converged mode) may implement the inference or converged phase, and a retraining phase mode (also referred to as a retraining mode) may implement the retraining phase. As described herein, the “phase” of a model may refer to an “operating mode” of the model, and the terms may be used interchangeably. And as indicated above, term “operating phase mode” may also be used to refer to the phase / operating mode of the model, or otherwise refer to the mode of operation in which the phase is implemented.

[0062] Due to ability to train the model online, the phase of the RL model may change while the model is deployed in air interface. A transition from phase 1 (training phase) to phase 2 (inference phase) may occur during operation in the training phase mode in which the model is trained. And a transition from phase 2 to phase 3 (retraining phase) may occur during operation in the inference phase mode, such as due to changes in the environment. In this case, the model may be trained online instead of fallback or switching-off the functionality.

[0063] Functionality-based LCM may operate based on, at least, one configuration of AI / ML-enabled feature / FG or specific configurations of an AI / ML-enabled feature / FG. Based on the phase of RL model, the configuration of the feature may be updated or changed. But in functionality-based LCM, the network (e.g., RAN node 202) is unaware of the physical model. In general, online training may be a factor that may trigger a change of the configuration pertaining the ML feature, yet it is not the only factor. Other factors may be the change of model due to training which leads to different requirements set by data collection configuration between UE 110 and RAN node 202 (e.g., safe exploration).

[0064] As shown in FIG. 5, there may be different approaches to trigger a configuration change for phase switching for UE-side models. In an implementationbased approach (option A), RL model phase definitions are implementation specific andtransparent to 3 GPP. The UE 110 may request a configuration due to change of RL model phase or any other condition. The configurations of training phase and converged phase may be completely transparent to the network. If additional information about the UE may be needed for some configuration changes, e.g., battery status of UE when enabling the exploration when idling, the UE may indicate such changes. The network may then deactivate the functionality.

[0065] Another approach is a model ID-based approach (option B), which is a hybrid implementation-specific approach. In model ID-based phase switching, the UE 110 may indicate a change of phase of a model based on an implementation-specific model phase definitions. The network may then verify the indicated change of phase based on the network’s own implementation-specific criteria before any feature reconfiguration / change.

[0066] Yet another approach is a functionality-based approach (option C), which is a standardized approach. In functionality-based phase switching, the model phase definitions may be standardized and thereby understood by both the UE 110 and the network (e.g., RAN node 202). The UE may trigger a change of phase of a model based on requirements and a request for a new configuration, and the UE may respond by providing the configuration. In the standardized approach, as well as the hybrid implementation-specific approach, the model phase is not transparent to 3 GPP signaling as both the UE 110 and the network need a common understanding of what each phase means.

[0067] In 3 GPP NR Release 19, the CSI framework may be expected to configure data collection for UE-sided model training. It has also been agreed that, for UE-side models in beam management, at least the associated ID can be configured within the CSI framework. The UE may assume similar properties of a downlink (DL) transmit beam or beam set / list associated with the same associated ID. It is also a working assumption that the UE 110 may assume that NW-side additional conditions with the same associated ID are consistent at least within a cell. If the associated ID is cell-specific, it means that the UE requires to train a cell-specific ML model because UE is not aware of the underlying network implementation behind the ID (e.g., network’s beam design). Thus, once the UEmoves to another cell, the UE may have to re-initiate the training procedure of the ML model or re-download the required ML model.

[0068] Regarding beam management, there were a few proposals in 3 GPP NR Release 18-19 on evaluating the performance of spatial beam prediction targeting improvements on throughput and latency via feeding RL model with measured reference signal received powers (RSRPs) and beam IDs of the best M beams and estimating a quality of service (QoS)-based metric online for each beam by exploring different beams and observing the QoS based metric as reward. Given the measured RSRP values of the best M beams along with beam usage values as a context vector to the RL model, the RL model may estimate the QoS-based metric for each action, namely, the beam ID. The QoS -based metric may be an amount of data, in bytes, transmitted during a single measurement interval. As described further below, some example implementations of the present disclosure that focus on beam prediction may apply to this approach on QoSbased beam prediction.

[0069] Briefly returning to PIG. 5, the different approaches to trigger a configuration change for phase switching may each have a number of features. In implementationbased phase switching (option A), the switching between phases are transparent to the network, signaling proposed in Release 19 may be reused, and nothing RL specific may be required for indication of applicability reporting.

[0070] In implementation-based phase switching, functionality configuration may provide both training and inference phase configurations with a single RRC message. The UE 110 may switch between phases autonomously, without network configuration, control, or indication / report. Applicability reporting may provide applicable and non- applicable functionalities, without indication of the phase to the network. Whether the functionalities are for the inference phase mode (phase 2) or the training phase mode (phase 1) is up to UE implementation. Non-applicability, however, may trigger a RRC reconfiguration for both training and inference phase modes. The indication of the modes is up to UE implementation. EIG. 6 is a signaling chart 100 for an implementationspecific approach to trigger a configuration change.

[0071] Model ID-based phase switching (option B) may be similar to implementation-based phase switching but with additional information. The switchingbetween operating modes for respective phases (within the same functionality) may be transparent to the network. Signaling proposed in Release 19 may be extended to model ID which may be used to enable operating mode changes. And QoS beam management may also be realized by this approach.

[0072] In functionality-based phase switching (option C), the switching between operating modes may be fully visible to the network. And, again, QoS beam management may also be realized by this approach.

[0073] In Release 19, the training phase (e.g., initial training) and the retraining phase (e.g., retraining or fine-tuning) are implementation-specific. In some scenarios, the network is only aware of functionality that may use inference operations. The network may be aware of training of the models within a functionality (option B), and also aware of or control switching from training to inference for a given functionality (option C). In this case, however, the signaling mechanisms are still missing. The examples can be best realized by RL-based problems where RL agents’ states, rewards, actions are visible and / or control by the network. Because, in the training phase, the UE 110 may require additional measurement configurations, response from the network for reward computation or the actions considered, as well as encountering new states.

[0074] Example implementations of the present disclosure focus on model ID-based phase switching (option B) and functionality-based phase switching (option C) to provide solutions with the information to be exchanged to distinguish between the training phase and the inference phase, as well as between the inference phase and the retraining phase. The solutions of some examples also enable determination of the applicability of the inference phase. In a combination of options, between the training phase and the inference phase, the configuration may be updated since the active functionality (or implicitly models) may be converged. And between the inference phase and the retraining phase, there may be an optional change in the configuration since the converged functionality may have a model which has been already fine-tuned.

[0075] According to some example implementations, in model ID-based phase switching (option B), a functionality configuration may provide both training and inference configurations for the training and inference phase modes with a single RRC message, and may also include a retraining configuration for the retraining phase mode.Applicability reporting may provide applicable and / or non-applicable functionality(ies) for the inference phase mode (and perhaps also the training and / or retraining phase modes). Non-applicability may trigger a RRC reconfiguration for one or more of the phases. In some examples, this may explicitly switch operating modes only with a model ID.

[0076] In functionality-based phase switching (option C), according to some example implementations, a AI / ML functionality configuration may provide both training and inference configurations for the training and inference phase modes with a single RRC message, and may also include a retraining configuration for the retraining phase mode. A switch between operating modes by the UE 110 may be visible to the network. Applicability reporting may provide applicable and / or non-applicable functionality(ies) for the inference phase mode (and perhaps also the training and / or retraining phase modes). The applicability reporting may therefore indicate applicable and / or non- applicable configuration(s) for the functionality(ies) for the inference phase modes (and perhaps also the other phase modes). Non-applicability may trigger a RRC reconfiguration for one or more of the phases. And based on the measurement reporting, the network may explicitly indicate a phase switch.

[0077] FIG. 7 is a flowchart 700 of procedures for RL model phase switching between a training phase (phase 1) and an inference phase (phase 2), according to some example implementations. As shown at blocks 702, 704, the UE 110 may receive configurations for training (phase 1) and inference (phase 2), and determine the applicability of a training configuration to activate phase 1 based on the received configurations. The UE may also receive configurations for retraining (phase 3). The UE may activate phase 1 after determining that the configuration of phase 1 is applicable, as shown at block 706.

[0078] As shown at block 708, the UE 110 may determine that a change is required from the training configuration to an inference configuration. Based on the determination or detection of change, the UE may again determine the applicability of the inference configuration to activate phase 2, as shown at block 710. In some examples, the UE may evaluate the applicability of phase 2 considering the changes in NW-side additional conditions, UE-side additional conditions, model availability and / or performancemonitoring. And the UE may activate phase 2 after determining the configuration of phase 2 is applicable, as shown at block 712.

[0079] FIG. 8 is a flowchart 800 of procedures for RL model phase switching between the training phase mode (phase 1), the inference phase mode (phase 2), and a retraining phase mode (phase 3), according to some example implementations. The procedures may include the same or similar operations for configuration, activation and transition between phase 1 and phase 2, as described above and shown at blocks 702-712. As shown at block 814, then, the UE 110 may determine that a change is required from the inference configuration to a retraining configuration. Based on the determination or detection of change, the UE may again determine the applicability of the retraining configuration to activate phase 3, as shown at block 816. In some examples, the UE may evaluate the applicability of phase 3 considering the changes in NW-side additional conditions, UE-side additional conditions, model availability and / or performance monitoring. And the UE may activate phase 3 after determining the configuration of phase 3 is applicable, as shown at block 818.

[0080] In functionality-based phase switching (Option C), the switching of different phases may be visible to, configured and controlled by the network (e.g., RAN node 202). The network may provide multiple phase configurations to the UE 110 in a single RRC message, such as based on the UE capabilities information. The multiple phase configurations may include initial training configurations (phase 1), inference configurations to be applied when initial training is completed and the UE detects that the functionality is in the inference phase mode (phase 2), and / or additional retraining configurations (phase 3) if the UE detects any change in the environment. The phase switching may be either initiated by the UE (request by UE) or the network may directly trigger the UE to change the phase without UE request.

[0081] FIG. 9 is a signaling chart 900 of procedures for initial configuration and applicability reporting after transition from the training phase mode to the inference phase, according to some example implementations. As shown at steps 901, 902, UE capability information may be exchanged between the UE 110 and the network (e.g. RAN node 202). The network may at step 903 configure the UE with multiple functionality configurations for at least phase 1 (training phase) and phase 2 (inferencephase), and perhaps also phase 3 (retraining phase). The network may also configure the UE with UE assistance information (UAI) or RCC reconfiguration complete to report applicability of different phases.

[0082] The UE 110 may at step 904 receive the phase 1 configuration and determine that whether the UE requires any initial training of ML model(s) associated with the functionality. In the RL case, an RL agent may make this determination by examining the states, actions, rewards and observations. Whether the phase 1 configuration provided by the network is applicable, the UE may consider NW-side additional conditions (e.g. associated ID), UE-side additional conditions (e.g. UE’s hardware, UE’s power state), and / or the availability of the model(s) to determine the applicability of the phase 1 configurations. The UE may then activate the phase 1 corresponding to provided configurations to allow training of model(s) for the functionality.

[0083] The UE 110 may at step 905 indicate the activation of phase 1 to the network (e.g., RAN node 202), and the UE may provide the network with which phase 1 configurations are applicable. In the case of beam prediction, for example, the UE may provide ReportConfiglD(s). The UE may configure UAI or RCC reconfiguration complete to indicate the phase 1. If the UE is configured with UAI, the UE may provide the information via UAI following a delay imposed by prohibit timer. The UAI provides no demand on when the UE transmit the report exactly so that the UE may in some examples complete the phase 1 in a longer duration. Otherwise, the UE may provide the information in a shorter time duration via an acknowledgement RRC message (e.g., RCC reconfiguration complete). In this case, the UE may be conditioned to provide phase 1 switching implicitly via applicability reporting allowing UE to switch autonomously or event-based triggering. In a UE-explicit request, the UE may request that the network provide phase 1 activation.

[0084] After the activation of phase 1, the UE 110 may at step 906 perform data collection exploration, training and monitoring. Since models are transparent to the network (e.g., RAN node 202), any LCM operations related to training and monitoring of the model(s) are internal to UE. However, the network may provide additional reference signal (RS) configurations for data collection at the UE-side.

[0085] The UE 110 may at steps 907, 908 determine when to switch from phase 1 to phase 2. In one example, UE may use an epsilon-greedy strategy. In the epsilon-greedy strategy, the UE may use a randomly-generated probability from a uniform distribution. If the value is below a threshold (epsilon) defined by the UE, the UE may decide to use phase 1 ; otherwise, the UE may decide to use phase 2. It may be possible for the UE to choose other complicated strategies. In some examples, the strategy(ies) may be UE implementation-specific. In other examples, the network (e.g., RAN node 202) may be in control and provide additional configurations. The strategies (events) may be included in the initial configuration if the network allows UE autonomous switching without UE request. In another example, the UE may request switching, and the network may trigger activation of one of the strategies.

[0086] The determination of phase 2 (inference phase mode) applicability by the UE 110 may be based on NW-side additional conditions (e.g., associated ID), UE-side additional conditions (e.g., UE’s hardware, UE’s power state), model availability, and / or performance monitoring of the available model(s). This monitoring may be internal to the UE to match the accuracy of the model(s) for the AI / ML feature (e.g., beam prediction accuracy in case of beam prediction) with a predefined accuracy. For instance, when the model(s) reach their highest estimated reward most of the time, the model(s) may be converged and ready for inference. It should be noted, however, that this performance monitoring may be UE-specific and internal to UE implementation for determining applicability, and may not have direct impact on monitoring operation of the functionality performance as configured by the network (e.g., RAN node 202).

[0087] After the determination, the UE 110 may at step 909 perform applicability reporting concerning the phase 2 provided configuration, and may indicate at least configuration IDs (e.g., ReportConfiglD(s) in case of beam prediction). Similar to step 905, UAI and RCC reconfiguration complete may be used for the reporting. In one example, the network (e.g., RAN node 202) may configure the UE to assess applicability without any additional delay (or non-applicability) immediately via RCC reconfiguration complete. In another example, UAI can be configured to report applicable or non- applicable configurations.

[0088] The network (e.g., RAN node 202) may at steps 910, 911 select one of the phase 2 (inference phase mode) configurations (e.g., CSI-ReconfinlD(s) in case of beam prediction) provided in step 903 for activation and activate the inference. This may be done through RRC reconfiguration, MAC CE or DCI. The UE may then operate in an inference phase mode to implement the inference phase based on one of the phase 2 configurations.

[0089] At some time after the initial configuration, the UE’s conditions may change, such as when the UE 110 moves to another cell, or the UE moves from outdoor to indoor, or NW-side additional conditions (e.g., codebook configurations) within a cell change. The change in the UE’s conditions change may impact the inference phase mode operation, and the model(s) being used in phase 2 may no longer be applicable. In these cases, the functionality may need to be switched to another configuration, or switched to the retraining phase to retrain the model(s) with new scenario.

[0090] FIG. 10 is a signaling chart 1000 of procedures for applicability reporting after transitions between the inference phase and the retraining phase, according to some example implementations. As shown at steps 1012, 1013, the UE 110 may evaluate the applicability of phase 2 considering changes in NW-side additional conditions, UE-side additional conditions, model availability and / or performance monitoring. This may trigger non-applicability of phase 2 configurations. RRC does not put any demand on when the UAI is transmitted by the UE on one hand, and any changes applied to NW-side additional requires an explicit reconfiguration message to the UE prompting the UE to return a reconfiguration complete. The network (e.g., RAN node 202) may therefore configure the UE to report its applicability or non-applicability (step 1013) using RRC reconfiguration complete even though UAI may still be an option.

[0091] As shown at steps 1014, 1015, 1016, upon receiving the non-applicability of phase 2, if at least one phase 3 configuration is not provided, the network (e.g., RAN node 202) may provide phase 3 (retraining) configuration(s) with RRC reconfiguration message. If phase 3 configuration(s) is already provided to UE 110, the network may explicitly request that the UE to switch to phase 3 with RRC, MAC control element (CE) or downlink control information (DCI) messages. If more than one configuration isapplicable in phase 3, the network may indicate switching via configuration ID (e.g., ReportConfigID in case of beam prediction).

[0092] The UE 110 may at steps 1017, 1018, 1019 switch from phase 2 to phase 3 upon receiving any of the indication (step 1014, 1015 or 1016) from the network (e.g., RAN node 202). If phase 3 is applicable, the UE may respond with RRC reconfiguration complete message (similar to step 905 in FIG. 9). After the activation of phase 3, the UE may perform data collection exploration, training and monitoring. Since models are transparent to the network, any LCM operations related to training and monitoring of the model(s) are internal to UE. However, the network may provide additional RS configurations (in case of beam prediction) for the data collection (similar to step 906 in FIG. 9).

[0093] The UE 110 may at steps 1020, 1021 determine when to switch from phase 3 to phase 2. Then UE may evaluate the determination of applicability of phase 2 again. These steps may be performed in a manner the same as or similar to steps 907 and 908. The UE 110 may at step 1022 perform applicability reporting with the indication of phase 2 and related ReportConfiglD(s). And at step 1023, the UE 110 may continue steps 1010, 1011

[0094] The signaling charts 900, 1000 are described above in the case of functionality-based phase switching (option C). Example implementations may also be equally applicable to model ID-based phase switching (option B). In these examples, the phase switching from the training phase to the inference phase for same functionality may include the explicit exchange of model ID(s) between the network (e.g., RAN node 202) and the UE 110. In this regard, the network may explicitly switch from one model ID to another model ID which may indicate a change between phase 1 and phase 2, or between phase 2 and phase 3, for the same functionality.

[0095] In some examples in the case of model ID-based phase switching, the UE 110 may provide (e.g., in the UE capability information) a mapping on which model ID(s) refer to the training phase and which model ID(s) refer to the inference phase. Additionally or alternatively, after configuration, the UE may send the network (e.g., RAN node 202) information that indicates a selected mapping between phases and model ID(s). The mapping and indication may be applied during step 905 (for training phaseactivation) and step 909 (for applicability reporting indication) in FIG. 9. And in some examples, during runtime, the UE may send the network information that indicates model ID(s) in each request to switch between phases.

[0096] FIGS. 11 A and 1 IB are flowcharts illustrating various steps in a method 1100 according to various example implementations. The method includes receiving from a network multiple phase configurations for a machine learning (ML) functionality, as shown at block 1102 of FIG. 11 A. The method includes activating a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations, as shown at block 1104. The method includes sending to the network a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode, as shown at block 1106. And the method includes receiving from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration, as shown at block 1108.

[0097] In some examples, the method 1100 further includes activating the inference phase mode based on the indication from the network.

[0098] In some examples, the method 1100 further includes sending to the network an indication of activation of the training phase mode and the applicable training configuration.

[0099] In some examples, the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode. In some of these examples, the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.

[0100] In some examples, the method 1100 further includes making a determination to switch from the training phase mode to the inference phase mode, as shown at block 1110 of FIG. 1 IB. And in some of these examples, the method also includes determining the at least one applicable inference configuration for the switch, as shown at block 1112.

[0101] In some examples, the method is performed by a user equipment (UE), and the at least one applicable inference configuration is determined based on at least one of one or more UE-side additional conditions, one or more network-side additional conditions, ML model availability, or performance monitoring.

[0102] In some examples, the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode. In some of these examples, the report includes the at least one model ID for the inference phase mode.

[0103] In some examples, the indication from the network to activate the inference phase mode includes an indication of the applicable inference configuration selected by the network from the at least one applicable inference configuration.

[0104] FIG. 12 is a flowchart illustrating various steps in a method 1200 according to various example implementations. The method includes configuring a user equipment (UE) with multiple phase configurations for a machine learning (ML) functionality, as shown at block 1202. The method includes receiving from the UE an indication of activation of a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations, as shown at block 1204. The method includes receiving from the UE a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode, as shown at block 1206. And the method includes sending to the UE an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration, as shown at block 1208.

[0105] In some examples, the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode. In some of these examples, the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.

[0106] In some examples, the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phasemode and the inference phase mode. In some of these examples, the report includes the at least one model ID for the inference phase mode.

[0107] In some examples, the method 1200 further includes selecting the applicable inference configuration of the at least one applicable inference configuration. In some of these examples, the indication to activate the inference phase mode includes an indication of the applicable inference configuration.

[0108] FIGS. 13A - 13C are flowcharts illustrating various steps in a method 1300 according to various example implementations. The method includes determining nonapplicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the ML functionality is deployed for inference operation, as shown at block 1302 of FIG. 13 A. The method includes sending to a network a report that indicates non-applicability of the one or more inference configurations, as shown at block 1304. The method includes receiving at block 1306 from the network an indication to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode; and based on the indication from the network, switching at block 1308 to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or fine-tune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0109] In some examples, the method 1300 further includes sending to the network a report that indicates at least one applicable retraining configuration. In some of these examples, the switch is to the retraining phase mode, and the applicable retraining configuration for the ML functionality is selected from the at least one applicable retraining configuration.

[0110] In some examples, the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes the applicable retraining configuration.

[0111] In some examples, the method 1300 further includes receiving from the network multiple phase configurations for the ML functionality, and the multiple phase configurations include the applicable retraining configuration.

[0112] In some examples, the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes an indication of the applicable retraining configuration of the multiple phase configurations.

[0113] In some examples, the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the inference phase mode and the retraining phase mode. In some of these examples, the indication of the applicable retraining configuration includes the at least one model ID for the retraining phase mode.

[0114] In some examples, the switch is to the retraining phase mode, and the method 1300 further includes sending to the network an indication of activation of the retraining phase mode and the applicable retraining configuration.

[0115] In some examples, the switch is to the retraining phase mode, and the method 1300 further includes sending to the network another report that indicates at least one applicable inference configuration of the one or more inference configurations for a switch from the retraining phase mode to the inference phase mode, as shown at block 1310 of FIG. 13B. In some of these examples, the method also includes receiving from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for the inference operation based on an applicable inference configuration of the at least one applicable inference configuration, as shown at block 1312

[0116] In some examples, the method 1300 further includes making a determination to switch from the retraining phase mode to the inference phase mode, as shown at block 1314 of FIG. 13C. And in some of these examples, the method also includes determining the at least one applicable inference configuration for the switch, as shown at block 1316.

[0117] In some examples, the method 1300 is performed by a user equipment (UE), and the at least one applicable inference configuration is determined based on at least one of one or more UE-side additional conditions, one or more network-side additional conditions, ML model availability, or performance monitoring.

[0118] In some examples, the indication from the network to activate the inference phase mode includes an indication of the applicable inference configuration selected by the network from the at least one applicable inference configuration.

[0119] FIGS. 14A and 14B are flowcharts illustrating various steps in a method 1400 according to various example implementations. The method includes receiving from a user equipment (UE) a report that indicates non-applicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the ML functionality is deployed for inference operation, as shown at block 1402 of FIG. 14A. The method includes making at block 1404 a determination to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode, based on the non-applicability of the one or more inference configurations; and based on the determination, sending at block 1406 to the UE an indication to switch to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or fine-tune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0120] In some examples, the method 1400 further includes receiving from the UE a report that indicates at least one applicable retraining configuration. In some of these examples, the switch is to the retraining phase mode, and the applicable retraining configuration for the ML functionality is selected from the at least one applicable retraining configuration.

[0121] In some examples, the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes the applicable retraining configuration.

[0122] In some examples, the method 1400 further includes configuring the UE with multiple phase configurations for the ML functionality, and the multiple phase configurations include the applicable retraining configuration.

[0123] In some examples, the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes an indication of the applicable retraining configuration of the multiple phase configurations.

[0124] In some examples, the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the inference phase mode and the retraining phase mode. In some of these examples, the indication of theapplicable retraining configuration includes the at least one model ID for the retraining phase mode.

[0125] In some examples, the switch is to the retraining phase mode, and the method 1400 further includes receiving from the UE an indication of activation of the retraining phase mode and the applicable retraining configuration.

[0126] In some examples, the switch is to the retraining phase mode, and the method 1400 further includes receiving from the UE another report that indicates at least one applicable inference configuration of the one or more inference configurations for a switch from the retraining phase mode to the inference phase mode, as shown at block 1408 of FIG. 14B. In some of these examples, the method also includes sending to the UE an indication to activate the inference phase mode in which the at least one ML model is deployed for the inference operation based on an applicable inference configuration of the at least one applicable inference configuration, as shown at block 1410

[0127] In some examples, the method 1400 further includes selecting the applicable inference configuration of the at least one applicable inference configuration. In some of these examples, the indication to activate the inference phase mode includes an indication of the applicable inference configuration.

[0128] According to example implementations of the present disclosure, a telecommunications system 100 or PLMN 102, and its components such as a UE 110, CN 106, RAN 108 and / or RAN node 202, may be implemented by various means. Means for implementing the system and its components may include hardware, firmware, software, or combinations thereof. In some examples, one or more apparatuses may be configured to function as or otherwise implement the system and its components shown and described herein. In examples involving more than one apparatus, the respective apparatuses may be connected to or otherwise in communication with one another in a number of different manners, such as directly or indirectly via a wired or wireless network or the like.

[0129] According to some example implementations, at least some of the method 1100 described with respect to FIGS. 11 A and 1 IB may be carried out by an apparatus comprising means for performing functions corresponding steps of the method. Similarly,at least some of the method 1200 described with respect to FIG. 12 may be carried out by an apparatus comprising means for performing functions corresponding steps of the method. At least some of the method 1300 described with respect to FIGS. 13A - 13C may be carried out by an apparatus comprising means for performing functions corresponding steps of the method. And least some of the method 1400 described with respect to FIGS. 14A and 14B may be carried out by an apparatus comprising means for performing functions corresponding steps of the method. Examples of a suitable apparatus may include a user equipment, user device, user terminal or the like. Other examples of a suitable apparatus may include a RAN node (e.g., ng-eNB, gNB, 6gNB, DU, CU) or any suitable apparatus, such as a server, host or node.

[0130] FIG. 15 illustrates an apparatus 1500 in which means for performing various functions includes hardware, alone or under direction of one or more computer programs from a computer-readable storage medium or other memory, such as computer memory, according to some example implementations of the present disclosure. The apparatus may include one or more of each of a number of components such as, for example, processing circuitry 1502 connected to computer-readable storage medium or other memory 1504.

[0131] The processing circuitry 1502 may be composed of one or more processors alone or in combination with one or more computer-readable storage media. The processing circuitry is generally any piece of computer hardware that is capable of processing information such as, for example, data, computer programs and / or other suitable electronic information. The processing circuitry is composed of a collection of electronic circuits some of which may be packaged as an integrated circuit or multiple interconnected integrated circuits (an integrated circuit at times more commonly referred to as a “chip”). The processing circuitry may be configured to execute computer programs, which may be stored onboard the processing circuitry or otherwise stored in the memory 1504 (of the same or another apparatus).

[0132] The processing circuitry 1502 may be a number of processors, a multi-core processor or some other type of processor, depending on the particular implementation. Further, the processing circuitry may be implemented using a number of heterogeneous processor systems in which a main processor is present with one or more secondary processors on a single chip. As another illustrative example, the processing circuitry maybe a symmetric multi-processor system containing multiple processors of the same type. In yet another example, the processing circuitry may be embodied as or otherwise include one or more ASICs, FPGAs or the like. Thus, although the processing circuitry may be capable of executing a computer program to perform one or more functions, the processing circuitry of various examples may be capable of performing one or more functions without the aid of a computer program. In either instance, the processing circuitry may be appropriately programmed to perform functions or operations according to example implementations of the present disclosure.

[0133] The memory 1504 is generally any piece of computer hardware that is capable of storing information such as, for example, data, computer programs, instructions 1506 (e.g., computer-readable program code) and / or other suitable information either on a temporary basis and / or a permanent basis. The memory may include volatile and / or nonvolatile memory, and may be fixed or removable. Examples of suitable memory include recording media, random access memory (RAM), read-only memory (ROM), a hard drive, a flash memory, a thumb drive, a removable computer diskette, an optical disk or some combination thereof.

[0134] The memory 1504 is a non-transitory device capable of storing information. One example of a suitable memory is a computer-readable storage medium, which is distinguishable from a computer-readable transmission medium capable of carrying information from one location to another. Examples of suitable computer-readable transmission media comprise electronic carrier signals, telecommunications signals, or some combination thereof. As used herein, the term “non-transitory” is a limitation of the medium itself (i.e., tangible, not a signal) as opposed to a limitation on data storage persistency (e.g., RAM versus ROM). A computer-readable medium as described herein generally refers to a computer-readable storage medium or computer-readable transmission medium. A computer-readable medium is any entity or device capable in which information, such as one or more computer programs or portions thereof, may be stored and carried.

[0135] In addition to the memory 1504 (e.g., computer-readable storage medium), the processing circuitry 1502 may also be connected to one or more interfaces for displaying, transmitting and / or receiving information. The interfaces may include a communicationsinterface 1508 and / or one or more user interfaces. The communications interface may be configured to transmit and / or receive information, such as to and / or from other apparatus(es), network(s) or the like. The communications interface may be configured to transmit and / or receive information by physical (wired) and / or wireless communications links. Examples of suitable communication interfaces include a network interface controller (NIC), wireless NIC (WNIC) or the like.

[0136] The user interfaces may include a display 1510 and / or one or more user input interfaces 1512. The display may be configured to present or otherwise display information to a user, suitable examples of which include a liquid crystal display (LCD), light-emitting diode (LED) display, organic LED (OLED) display, active-matrix OLED (AMOLED) or the like. The user input interfaces may be wired or wireless, and may be configured to receive information from a user into the apparatus, such as for processing, storage and / or display. Suitable examples of user input interfaces include a microphone, image or video capture device, keyboard or keypad, joystick, touch-sensitive surface (separate from or integrated into a touchscreen), biometric sensor or the like. The user interfaces may further include one or more interfaces for communicating with peripherals such as printers, scanners or the like.

[0137] Execution of the instructions 1506 by the processing circuitry 1502, or storage of the instructions in the memory 1504, supports combinations of operations for implementing example implementations of the present disclosure. In this manner, an apparatus 1500 may comprise at least one processing circuitry and at least one memory coupled to the at least one processing circuitry, where the at least one processing circuitry is configured to execute instructions stored in the at least one memory. It will also be understood that one or more functions, and combinations of functions, may be implemented by special purpose hardware-based computer systems and / or processing circuitry which perform the specified functions, or combinations of special purpose hardware and program code instructions.

[0138] Some example implementations of the present disclosure may also be carried out in the form of a computer process defined by one or more computer programs or portions thereof. Example implementations of the present disclosure may be carried out by executing at least one portion of a computer program comprising instructions. Thecomputer program may be in source code form, object code form, or in some intermediate form. The computer program may be stored in a computer-readable medium that is readable by a computer, processing circuitry or other suitable apparatus. As indicated above, for example, the computer program may be stored in a memory, such as a computer-readable storage medium. Additionally or alternatively, for example, the computer program may be stored in a computer-readable transmission medium. The coding of software for carrying out example implementations of the present disclosure is well within the scope of a person of ordinary skill in the art.

[0139] As will be appreciated, any suitable instructions may be loaded onto a computer, a processing circuitry or other programmable apparatus from a memory or a computer-readable medium (e.g., computer-readable storage medium, computer-readable transmission medium) to produce a particular machine, such that the particular machine becomes a means for implementing the functions specified herein. The instructions may also be stored in a computer-readable medium that can direct a computer, a processing circuitry or other programmable apparatus to function in a particular manner to thereby generate a particular machine or particular article of manufacture. In some examples, the instructions stored in the computer-readable medium may produce an article of manufacture, where the article of manufacture becomes a means for implementing functions described herein. The instructions may be retrieved from a computer-readable medium and loaded into a computer, processing circuitry or other programmable apparatus to configure the computer, processing circuitry or other programmable apparatus to execute operations to be performed on or by the computer, processing circuitry or other programmable apparatus.

[0140] Retrieval, loading and execution of instructions comprising program code instructions may be performed sequentially such that one instruction is retrieved, loaded and executed at a time. In some example implementations, retrieval, loading and / or execution may be performed in parallel such that multiple instructions are retrieved, loaded, and / or executed together. Execution of the program code instructions may produce a computer-implemented process such that the instructions executed by the computer, processing circuitry or other programmable apparatus provide operations for implementing functions described herein.

[0141] As explained above and reiterated below, the present disclosure includes, without limitation, the following example implementations.

[0142] Clause 1. A method comprising: receiving from a network multiple phase configurations for a machine learning (ML) functionality; activating a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; sending to the network a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and receiving from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0143] Clause 2. The method of clause 1 , wherein the method further comprises activating the inference phase mode based on the indication from the network.

[0144] Clause 3. The method of clause 1 or clause 2, wherein the method further comprises sending to the network an indication of activation of the training phase mode and the applicable training configuration.

[0145] Clause 4. The method of clause 3, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.

[0146] Clause 5. The method of any of clauses 1 to 4, wherein the method further comprises: making a determination to switch from the training phase mode to the inference phase mode; and determining the at least one applicable inference configuration for the switch.

[0147] Clause 6. The method of clause 5, wherein the method is performed by a user equipment (UE), and the at least one applicable inference configuration is determined based on at least one of one or more UE-side additional conditions, one or more networkside additional conditions, ML model availability, or performance monitoring.

[0148] Clause 7. The method of any of clauses 1 to 6, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the report includes the at least one model ID for the inference phase mode.

[0149] Clause 8. The method of any of clauses 1 to 7, wherein the indication from the network to activate the inference phase mode includes an indication of the applicable inference configuration selected by the network from the at least one applicable inference configuration.

[0150] Clause 9. An apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to perform the method of any of clauses 1 to 8.

[0151] Clause 10. An apparatus comprising means for performing the method of any of clauses 1 to 8.

[0152] Clause 11. A computer-readable medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 1 to 8.

[0153] Clause 12. A computer-readable storage medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 1 to 8.

[0154] Clause 13. A computer program comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 1 to 8.

[0155] Clause 14. A method comprising: configuring a user equipment (UE) with multiple phase configurations for a machine learning (ML) functionality; receiving from the UE an indication of activation of a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; receiving from the UE a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and sending to the UE an indication to activate the inference phase mode in which the at leastone ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0156] Clause 15. The method of clause 14, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.

[0157] Clause 16. The method of clause 14 or clause 15, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the report includes the at least one model ID for the inference phase mode.

[0158] Clause 17. The method of any of clauses 14 to 16, wherein the method further comprises selecting the applicable inference configuration of the at least one applicable inference configuration, and wherein the indication to activate the inference phase mode includes an indication of the applicable inference configuration.

[0159] Clause 18. An apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to perform the method of any of clauses 14 to 17.

[0160] Clause 19. An apparatus comprising means for performing the method of any of clauses 14 to 17.

[0161] Clause 20. A computer-readable medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 14 to 17.

[0162] Clause 21. A computer-readable storage medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 14 to 17.

[0163] Clause 22. A computer program comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 14 to 17.

[0164] Clause 23. A method comprising: determining non-applicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the ML functionality is deployed for inference operation; sending to a network a report that indicates nonapplicability of the one or more inference configurations; receiving from the network an indication to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode; and based on the indication from the network, switching to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or fine-tune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0165] Clause 24. The method of clause 23, wherein the method further comprises sending to the network a report that indicates at least one applicable retraining configuration, and wherein the switch is to the retraining phase mode, and the applicable retraining configuration for the ML functionality is selected from the at least one applicable retraining configuration.

[0166] Clause 25. The method of clause 23 or clause 24, wherein the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes the applicable retraining configuration.

[0167] Clause 26. The method of any of clauses 23 to 25, wherein the method further comprises receiving from the network multiple phase configurations for the ML functionality, and the multiple phase configurations include the applicable retraining configuration.

[0168] Clause 27. The method of clause 26, wherein the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes an indication of the applicable retraining configuration of the multiple phase configurations.

[0169] Clause 28. The method of clause 27, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the inference phase mode and the retraining phase mode, and wherein the indication of the applicable retraining configuration includes the at least one model ID for the retraining phase mode.

[0170] Clause 29. The method of any of clauses 23 to 28, wherein the switch is to the retraining phase mode, and the method further comprises sending to the network an indication of activation of the retraining phase mode and the applicable retraining configuration.

[0171] Clause 30. The method of any of clauses 23 to 29, wherein the switch is to the retraining phase mode, and the method further comprises: sending to the network another report that indicates at least one applicable inference configuration of the one or more inference configurations for a switch from the retraining phase mode to the inference phase mode; and receiving from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for the inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0172] Clause 31. The method of clause 30, wherein the method further comprises: making a determination to switch from the retraining phase mode to the inference phase mode; and determining the at least one applicable inference configuration for the switch.

[0173] Clause 32. The method of clause 30 or clause 31, wherein the method is performed by a user equipment (UE), and the at least one applicable inference configuration is determined based on at least one of one or more UE-side additional conditions, one or more network-side additional conditions, ML model availability, or performance monitoring.

[0174] Clause 33. The method of any of clauses 30 to 32, wherein the indication from the network to activate the inference phase mode includes an indication of the applicable inference configuration selected by the network from the at least one applicable inference configuration.

[0175] Clause 34. An apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to perform the method of any of clauses 23 to 33.

[0176] Clause 35. An apparatus comprising means for performing the method of any of clauses 23 to 33.

[0177] Clause 36. A computer-readable medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 23 to 33.

[0178] Clause 37. A computer-readable storage medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 23 to 33.

[0179] Clause 38. A computer program comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 23 to 33.

[0180] Clause 39. A method comprising: receiving from a user equipment (UE) a report that indicates non-applicability of one or more inference configurations for a machine learning (ML) functionality in an inference phase mode in which at least one ML model associated with the ML functionality is deployed for inference operation; making a determination to switch to an applicable inference configuration, or switch from the inference phase mode to a retraining phase mode, based on the non-applicability of the one or more inference configurations; and based on the determination, sending to the UE an indication to switch to the applicable inference configuration for the inference phase mode, or from the inference phase mode to the retraining phase mode in which the retraining phase mode is activated to retrain or fine-tune the at least one ML model based on an applicable retraining configuration for the ML functionality.

[0181] Clause 40. The method of clause 39, wherein the method further comprises receiving from the UE a report that indicates at least one applicable retraining configuration, and wherein the switch is to the retraining phase mode, and the applicable retraining configuration for the ML functionality is selected from the at least one applicable retraining configuration.

[0182] Clause 41. The method of clause 39 or clause 40, wherein the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes the applicable retraining configuration.

[0183] Clause 42. The method of any of clauses 39 to 41, wherein the method further comprises configuring the UE with multiple phase configurations for the MLfunctionality, and the multiple phase configurations include the applicable retraining configuration.

[0184] Clause 43. The method of clause 42, wherein the indication is to switch from the inference phase mode to the retraining phase mode, and the indication includes an indication of the applicable retraining configuration of the multiple phase configurations.

[0185] Clause 44. The method of clause 43, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the inference phase mode and the retraining phase mode, and wherein the indication of the applicable retraining configuration includes the at least one model ID for the retraining phase mode.

[0186] Clause 45. The method of any of clauses 39 to 44, wherein the switch is to the retraining phase mode, and the method further comprises receiving from the UE an indication of activation of the retraining phase mode and the applicable retraining configuration.

[0187] Clause 46. The method of any of clauses 39 to 45, wherein the switch is to the retraining phase mode, and the method further comprises: receiving from the UE another report that indicates at least one applicable inference configuration of the one or more inference configurations for a switch from the retraining phase mode to the inference phase mode; and sending to the UE an indication to activate the inference phase mode in which the at least one ML model is deployed for the inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

[0188] Clause 47. The method of clause 46, wherein the method further comprises selecting the applicable inference configuration of the at least one applicable inference configuration, and wherein the indication to activate the inference phase mode includes an indication of the applicable inference configuration.

[0189] Clause 48. An apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to perform the method of any of clauses 39 to 47.

[0190] Clause 49. An apparatus comprising means for performing the method of any of clauses 39 to 47.

[0191] Clause 50. A computer-readable medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 39 to 47.

[0192] Clause 51. A computer-readable storage medium comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 39 to 47.

[0193] Clause 52. A computer program comprising instructions that, in response to execution by at least one processing circuitry, causes an apparatus to perform the method of any of clauses 39 to 47.

[0194] Many modifications and other implementations of the disclosure set forth herein will come to mind to one skilled in the art to which the disclosure pertains having the benefit of the teachings presented in the foregoing description and the associated figures. Therefore, it is to be understood that the disclosure is not to be limited to the specific implementations disclosed and that modifications and other implementations are intended to be included within the scope of the appended claims. Moreover, although the foregoing description and the associated figures describe example implementations in the context of certain example combinations of elements and / or functions, it should be appreciated that different combinations of elements and / or functions may be provided by alternative implementations without departing from the scope of the appended claims. In this regard, for example, different combinations of elements and / or functions than those explicitly described above are also contemplated as may be set forth in some of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

Claims

WHAT IS CLAIMED IS:

1. An apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to at least: receive from a network multiple phase configurations for a machine learning (ML) functionality; activate a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; send to the network a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and receive from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

2. The apparatus of claim 1, wherein the at least one processing circuitry is configured to execute the instructions to cause the apparatus to further activate the inference phase mode based on the indication from the network.

3. The apparatus of claim 1, wherein the at least one processing circuitry is configured to execute the instructions to cause the apparatus to further send to the network an indication of activation of the training phase mode and the applicable training configuration.

4. The apparatus of claim 3, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.-47-5. The apparatus of claim 1, wherein the at least one processing circuitry is configured to execute the instructions to cause the apparatus to further at least: make a determination to switch from the training phase mode to the inference phase mode; and determine the at least one applicable inference configuration for the switch.

6. The apparatus of claim 5, wherein the apparatus is to implement a user equipment (UE), and the at least one applicable inference configuration is determined based on at least one of one or more UE-side additional conditions, one or more networkside additional conditions, ML model availability, or performance monitoring.

7. The apparatus of claim 1, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the report includes the at least one model ID for the inference phase mode.

8. The apparatus of claim 1, wherein the indication from the network to activate the inference phase mode includes an indication of the applicable inference configuration selected by the network from the at least one applicable inference configuration.

9. A method comprising: receiving from a network multiple phase configurations for a machine learning (ML) functionality; activating a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations;-48-sending to the network a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and receiving from the network an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

10. The method of claim 9, wherein the method further comprises activating the inference phase mode based on the indication from the network.

11. The method of claim 9, wherein the method further comprises sending to the network an indication of activation of the training phase mode and the applicable training configuration.

12. The method of claim 11, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.

13. The method of claim 9, wherein the method further comprises: making a determination to switch from the training phase mode to the inference phase mode; and determining the at least one applicable inference configuration for the switch.

14. The method of claim 13, wherein the method is performed by a user equipment (UE), and the at least one applicable inference configuration is determined based on at least one of one or more UE-side additional conditions, one or more networkside additional conditions, ML model availability, or performance monitoring.-49-15. The method of claim 9, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the report includes the at least one model ID for the inference phase mode.

16. The method of claim 9, wherein the indication from the network to activate the inference phase mode includes an indication of the applicable inference configuration selected by the network from the at least one applicable inference configuration.

17. An apparatus comprising: at least one memory configured to store instructions; and at least one processing circuitry configured to access the at least one memory, and execute the instructions to cause the apparatus to at least: configure a user equipment (UE) with multiple phase configurations for a machine learning (ML) functionality; receive from the UE an indication of activation of a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; receive from the UE a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and send to the UE an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

18. The apparatus of claim 17, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and-SO-wherein the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.

19. The apparatus of claim 17, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the report includes the at least one model ID for the inference phase mode.

20. The apparatus of claim 17, wherein the at least one processing circuitry is configured to execute the instructions to cause the apparatus to further select the applicable inference configuration of the at least one applicable inference configuration, and wherein the indication to activate the inference phase mode includes an indication of the applicable inference configuration.

21. A method comprising: configuring a user equipment (UE) with multiple phase configurations for a machine learning (ML) functionality; receiving from the UE an indication of activation of a training phase mode in which at least one ML model associated with the ML functionality is trained based on an applicable training configuration of the multiple phase configurations; receiving from the UE a report that indicates at least one applicable inference configuration of the multiple phase configurations for a switch from the training phase mode to an inference phase mode; and sending to the UE an indication to activate the inference phase mode in which the at least one ML model is deployed for inference operation based on an applicable inference configuration of the at least one applicable inference configuration.

22. The method of claim 21, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the indication of the activation of the training phase mode includes the at least one model ID for the training phase mode.

23. The method of claim 21, wherein the at least one ML model associated with the ML functionality includes at least one model identifier (ID) for each of the training phase mode and the inference phase mode, and wherein the report includes the at least one model ID for the inference phase mode.

24. The method of claim 21, wherein the method further comprises selecting the applicable inference configuration of the at least one applicable inference configuration, and wherein the indication to activate the inference phase mode includes an indication of the applicable inference configuration.

Images