WO2025026970A1

WO2025026970A1 - Method of activating a candidate model

Info

Publication number: WO2025026970A1
Application number: PCT/EP2024/071452
Authority: WO
Inventors: Hojin Kim; Rikin SHAH; Andreas Andrae; Reuben GEORGE STEPHEN; Shravan Kumar KALYANKAR
Original assignee: Continental Automotive Technologies GmbH
Current assignee: Aumovio Germany GmbH
Priority date: 2023-08-02
Filing date: 2024-07-29
Publication date: 2025-02-06
Anticipated expiration: 2026-02-02

Abstract

The present disclosure describes methods of data-driven AI/ML model signaling for efficient model switching with machine learning model operation in wireless mobile communication systems including base stations (e.g., gNB) and mobile stations (e.g., UE). The pre-configured model modes with the associated candidate models are used to switch models based on model monitoring.

Description

TITLE

Method of activating a candidate model

TECHNNICAL FIELD

The present disclosure relates to AI/ML based applicable model update report signaling, where techniques for pre-configuring and signaling the model candidacy for the efficient switching of machine learning model operation are presented.

BACKGROUND

In 3GPP (3rd Generation Partnership Project), one of the selected study items as the approved Release 18 package is AI/ML (artificial intelligence/machine learning) as described in the related document (RP-213599) addressed in 3GPP TSG RAN (Technical Specification Group Radio Access Network) meeting #94e. The official title of the AI/ML study item is “Study on AI/ML for NR Air Interface”, and currently RAN WG1 and WG2 are actively working on a specification. The goal of this study item is to identify a common AI/ML framework and areas of obtaining gains using AI/ML based techniques with use cases.

According to 3GPP, the main objective of this study item is to study AI/ML frameworks for air-interfaces with target use cases by considering performance, complexity, and potential specification impacts. In particular, AI/ML models, terminology, and descriptions to identify common and specific characteristics for a framework will be one key work scope. Regarding AI/ML frameworks, various aspects are under consideration for investigation and one key item is about the lifecycle management of AI/ML models where multiple stages are included as mandatory for model training, model deployment, model inference, model monitoring, model updating, etc.

Earlier, in 3GPP TR 37.817 for Release 17 titled as “Study on enhancement for Data Collection for NR and EN-DC”, UE mobility was also considered as one of the AI/ML use cases and one of the scenarios for model training/inference is that both functions are located within a RAN node. Following, in Release 18, the new work item of “Artificial Intelligence (AI)ZMachine Learning (ML) for NG-RAN” was initiated to specify data collection enhancements and signaling support within existing NG-RAN interfaces and architectures.

For the above active standardization works, currently there is no specification defined for signaling methods or network (e.g., gNB)/mobile station (e.g., UE) behaviors about supporting an UE status update reporting about AI/ML model operation when multiple LCM (lifecycle management) operations are enabled on a device such as model training and inferencing and/or updating, etc., for one or more particular application scenarios with different set of ML features/functionalities. Since applicable conditions to support the enabled LCM operations can dynamically change due to in-device condition change and/or external condition change, a model drift can occur as model performance is degraded along with the related condition changes. To detect such a model drift, model monitoring is one of the key phases in LCM operations and the follow-up countermeasures such as model switching, model re-training or model de-activation are commonly used. For example, model transfer is needed from one node to another node when the existing model is replaced with an alternative one. Additional signaling overhead can be significant to execute such a model transfer.

US 2023 069 342 A1 discloses a computer system that detects an abnormality based on time series data, including: an abnormality diagnosis unit that diagnoses an abnormality of the time series data from a machine learning model created based on learning data; a model degradation detection unit that detects degradation in the machine learning model; a learning curve estimation unit that estimates a learning curve and predicts a number of errors per unit time; a model switch cost calculation unit that calculates a number of errors per unit time of a model in operation, a number of errors per unit time of a switch candidate model, a first total cost and a second total cost; and a model switch time prediction unit that compares the first total cost with the second total cost to calculate switch time of a machine learning model. US 2023 022 737 A1 discloses a generation support apparatus including: a storage device that stores component tables for storing a change history of a plurality of component groups for a machine learning model, and a model component management table for storing versions for the plurality of component groups; a model change recorder that inputs configuration information on an updated first machine learning model, adds the change history to the component tables, and records a version of an updated component group in the model component management table; and a derived model parameter synthesizer that generates a model parameter including a component corresponding to a version of an updated component group of the first machine learning model, for a second machine learning model including a component group of the same version as a component group before updating of the first machine learning model.

US 2019 012 876 A1 discloses a platform for providing projections, predictions, and recommendations for casino and gaming environments. The platform leverages machine learning and cognitive computing. Through a natural language interface, the platform presents this information in a way which is natural and timely for casino operational executives to understand and act upon. The platform can optimize gaming machine performance casino floor performance based on various metrics that are predicted by the platform.

US 2019 332 895 A1 discloses methods for optimizing a network comprising a core computing system (CCS) and a set of edge computing devices (ECDs), wherein each of the ECDs locally performs computations based on a trained machine learning (ML) model. A plurality of ML models are continually trained at the CCS, concurrently, based on data collected from the ECDs. One or more states of the network and/or components thereof are monitored. The monitored states are relied upon to decide (when) to change a trained ML model as currently used by any of the ECDs to perform said computations. It may be decided to change the model used by a given one of the ECDs to perform ML-based computations. One of the models as trained at the CCS is selected (based on the monitored states) and corresponding parameters are sent to this ECD. The latter can resume computations according to a trained model. EP 4 075 348 A1 discloses a computer-implemented method and a quality control system for quality control of a machine learning model, which can be based on a federated learning method collectively performed by nodes of a decentralized distributed database. A modified machine learning model is generated by training the machine learning model using a first local dataset of a first node of the distributed database. The modified machine learning model is sent to the other nodes of the distributed database and checked against a respective local dataset of each of the other nodes. The other nodes perform a joint evaluation of the check results, e.g., through a consensus method. The machine learning model is replaced by the modified machine learning model if the evaluation result indicates an improvement. The invention enables federated learning without the need of a central authority managing the global model.

US 2021 019 612 A1 discloses an end-to-end cloud-based machine learning platform providing personalized game player experiences. Data lineage is generated for all transformed data for generating feature ETLs, and for training machine learning models. That data is used to understand the performance of off-line and online recommender systems for the personalization of the game player experiences. To that end, the platform pipeline provides the life cycle of the transformed data to a self-healing system that compare it to the life cycle of the user interactions. By comparing the two life cycles, the self-healing system can automatically provide a diagnostic, and it can also automatically provide an action if the performance of the model predictions has changed over time.

BRIEF DESCRIPTION OF THE DRAWINGS

The disclosed invention will be further discussed in the following based on preferred embodiments presented in the attached drawings. However, the disclosed invention may be embodied in many different forms and should not be construed as limited to said preferred embodiments. Rather, said preferred embodiments are provided for thoroughness and completeness, and fully convey the scope of the invention to the skilled person. The following detailed description refers to the attached drawings, in which:

Figure 1 is an exemplary table of a type-1 for model state modes;

Figure 2 is an exemplary table of a type-2 for model state modes;

Figure 3 is an exemplary table of mapping relationship for candidate models and model state modes;

Figure 4 is an exemplary block diagram of model state mode transitions;

Figure 5 is a flowchart of network behavior for configuring mapping relationship for candidate models and model state modes;

Figure 6 is a flowchart of UE behavior for UE autonomous decision on model switching;

Figure 7 is a flowchart of UE behavior for network decision on model switching; and

Figure 8 is a flowchart of timer-based model state mode transition.

DETAILED DESCRIPTION

The detailed description set forth below, with reference to the annexed drawings, is intended as a description of various configurations and is not intended to represent the only configurations in which the concepts described herein may be practiced. The detailed description includes specific details for the purpose of providing a thorough understanding of the various concepts. However, it will be apparent to those skilled in the art that these concepts may be practiced without these specific details. In particular, although terminology from 3GPP 5G NR may be used in this disclosure to exemplify embodiments herein, this should not be seen as limiting the scope of the invention. Some of the embodiments contemplated herein will now be described more fully with reference to the accompanying drawings. Other embodiments, however, are contained within the scope of the subject matter disclosed herein, the disclosed subject matter should not be construed as limited to only the embodiments set forth herein; rather, these embodiments are provided by way of example to convey the scope of the subject matter to those skilled in the art.

Generally, all terms used herein are to be interpreted according to their ordinary meaning in the relevant technical field, unless a different meaning is clearly given and/or is implied from the context in which it is used. All references to a/an/the element, apparatus, component, means, step, etc. are to be interpreted openly as referring to at least one instance of the element, apparatus, component, means, step, etc., unless explicitly stated otherwise. The steps of any methods disclosed herein do not have to be performed in the exact order disclosed, unless a step is explicitly described as following or preceding another step and/or where it is implicit that a step must follow or precede another step. Any feature of any of the embodiments disclosed herein may be applied to any other embodiment, wherever appropriate. Likewise, any advantage of any of the embodiments may apply to any other embodiments, and vice versa. Other objectives, features and advantages of the enclosed embodiments will be apparent from the following description.

In some embodiments, a more general term “network node” may be used and may correspond to any type of radio network node or any network node, which communicates with a UE (directly or via another node) and/or with another network node. Examples of network nodes are NodeB, MeNB, ENB, a network node belonging to MCG or SCG, base station (BS), multi-standard radio (MSR) radio node such as MSR BS, eNodeB, gNodeB, network controller, radio network controller (RNC), base station controller (BSC), relay, donor node controlling relay, base transceiver station (BTS), access point (AP), transmission points, transmission nodes, RRU, RRH, nodes in distributed antenna system (DAS), core network node (e.g. Mobile Switching Center (MSC), Mobility Management Entity (MME), etc), Operations & Maintenance (O&M), Operations Support System (OSS), Self Optimized Network (SON), positioning node (e.g. Evolved- Serving Mobile Location Centre (E-SMLC)), Minimization of Drive Tests (MDT), test equipment (physical node or software), another UE, etc.

In some embodiments, the non-limiting term user equipment (UE) or wireless device may be used and may refer to any type of wireless device communicating with a network node and/or with another UE in a cellular or mobile communication system. Examples of UE are target device, device to device (D2D) UE, machine type UE or UE capable of machine to machine (M2M) communication, PDA, PAD, Tablet, mobile terminals, smart phone, laptop embedded equipped (LEE), laptop mounted equipment (LME), USB dongles, UE category Ml, UE category M2, ProSe UE, V2V UE, V2X UE, etc.

Additionally, terminologies such as base station/gNodeB and UE should be considered non-limiting and in particular do not imply a certain hierarchical relation between the two; in general, “gNodeB” could be considered as device 1 and “UE” could be considered as device 2 and these two devices communicate with each other over some radio channel. And in the following the transmitter or receiver could be either gNodeB (gNB), or UE.

As will be appreciated by one skilled in the art, aspects of the embodiments may be embodied as a system, apparatus, method, or computer program product.

Accordingly, embodiments may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects.

For example, the disclosed embodiments may be implemented as a hardware circuit comprising custom very-large-scale integration (“VLSI”) circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, or other discrete components. The disclosed embodiments may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices, or the like. As another example, the disclosed embodiments may include one or more physical or logical blocks of executable code which may, for instance, be organized as an object, procedure, or function.

Furthermore, embodiments may take the form of a computer program product embodied in one or more computer readable storage devices storing machine readable code, computer readable code, and/or program code, referred hereafter as code. The storage devices may be tangible, non- transitory, and/or non-transmission. The storage devices may not embody signals. In a certain embodiment, the storage devices only employ signals for accessing code.

Any combination of one or more computer readable medium may be utilized. The computer readable medium may be a computer readable storage medium. The computer readable storage medium may be a storage device storing the code. The storage device may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, holographic, micromechanical, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing.

More specific examples (a non-exhaustive list) of the storage device would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random-access memory (“RAM”), a read-only memory (“ROM”), an erasable programmable read-only memory (“EPROM” or Flash memory), a portable compact disc read-only memory (“CD-ROM”), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device.

Code for carrying out operations for embodiments may be any number of lines and may be written in any combination of one or more programming languages including an object-oriented programming language such as Python, Ruby, Java, Smalltalk, C++, or the like, and conventional procedural programming languages, such as the “C” programming language, or the like, and/or machine languages such as assembly languages. The code may execute entirely on the user’s computer, partly on the user’s computer, as a stand-alone software package, partly on the user’s computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user’s computer through any type of network, including a local area network (“LAN”), wireless LAN (“WLAN”), or a wide area network (“WAN”), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider (“ISP”)).

Furthermore, the described features, structures, or characteristics of the embodiments may be combined in any suitable manner. In the following description, numerous specific details are provided, such as examples of programming, software modules, user selections, network transactions, database queries, database structures, hardware modules, hardware circuits, hardware chips, etc., to provide a thorough understanding of embodiments. One skilled in the relevant art will recognize, however, that embodiments may be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures, materials, or operations are not shown or described in detail to avoid obscuring aspects of an embodiment. Reference throughout this specification to “one embodiment,” “an embodiment,” or similar language means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, appearances of the phrases “in one embodiment”, “in an embodiment”, and similar language throughout this specification may, but do not necessarily, all refer to the same embodiment, but mean “one or more but not all embodiments” unless expressly specified otherwise. The terms “including”, “comprising”, “having”, and variations thereof mean “including but not limited to”, unless expressly specified otherwise. An enumerated listing of items does not imply that any or all of the items are mutually exclusive, unless expressly specified otherwise. The terms “a”, “an”, and “the” also refer to “one or more” unless expressly specified otherwise.

Aspects of the embodiments are described below with reference to schematic flowchart diagrams and/or schematic block diagrams of methods, apparatuses, systems, and program products according to embodiments. It will be understood that each block of the schematic flowchart diagrams and/or schematic block diagrams, and combinations of blocks in the schematic flowchart diagrams and/or schematic block diagrams, can be implemented by code. This code may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart diagrams and/or block diagrams.

The code may also be stored in a storage device that can direct a computer, other programmable data processing apparatus, or other devices to function in a particular manner, such that the instructions stored in the storage device produce an article of manufacture including instructions which implement the function/act specified in the flowchart diagrams and/or block diagrams.

The code may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus, or other devices to produce a computer implemented process such that the code which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified in the flowchart diagrams and/or block diagrams.

The flowchart diagrams and/or block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of apparatuses, systems, methods, and program products according to various embodiments. In this regard, each block in the flowchart diagrams and/or block diagrams may represent a module, segment, or portion of code, which includes one or more executable instructions of the code for implementing the specified logical function(s).

It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. Other steps and methods may be conceived that are equivalent in function, logic, or effect to one or more blocks, or portions thereof, of the illustrated Figures.

Although various arrow types and line types may be employed in the flowchart and/or block diagrams, they are understood not to limit the scope of the corresponding embodiments. Indeed, some arrows or other connectors may be used to indicate only the logical flow of the depicted embodiment. For instance, an arrow may indicate a waiting or monitoring period of unspecified duration between enumerated steps of the depicted embodiment. It will also be noted that each block of the block diagrams and/or flowchart diagrams, and combinations of blocks in the block diagrams and/or flowchart diagrams, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and code.

The description of elements in each figure may refer to elements of proceeding figures. Like numbers refer to like elements in all figures, including alternate embodiments of like elements.

The following explanation will provide the detailed description of the mechanism about data-driven AI/ML model signaling for based ML model operation in wireless mobile communication system including base station (e.g., gNB) and mobile station (e.g., UE)

AI/ML lifecycle can be split into several stages such as data collection/pre-processing, model training, model testing/validation, model deployment/update, model monitoring, model switching/selection etc., where each stage is equally important to achieve target performance with any specific model(s). In applying AI/ML model for any use case or application, one of the challenging issues is to manage the lifecycle of AI/ML model. This is mainly because a data/model drift occurs during model deployment/inference which results in performance degradation of the AI/ML model. Model drift occurs, when dataset statistically change after the model is deployed and when model inference capability is impacted due to unseen data as input. In a similar aspect, the statistical property of a dataset and the relationship between input and output for the trained model can be changed with drift occurrence. Then, model adaptation is required to support operations such as model switching, re-training, fallback, etc.

According to a first aspect of the invention, a method of activating a candidate model out of a set of candidate models for an AI/ML model switching operation in a wireless network comprises a step of configuring, by a first node of the wireless network, a mapping relationship by allocating a model state mode to each of the candidate models, the model state mode indicating a readiness status for the model switching operation. Further, the method comprises a step of receiving, by a second node of the wireless network, of the mapping relationship. Further, the method comprises a step of selecting the candidate model for the model switching operation based on the received mapping relationship. Further, the method comprises a step of performing the model switching operation, during which the candidate model is actived.

Advantageously, the set of candidate models, the mapping relationship, and/or the model state modes are indexed.

Advantageously, the mapping relationship comprises conditions associated with the model state modes.

Advantageously, the model state modes are pre-configured.

Advantageously, a switching indication is transmitted from the first node to the second node, the switching indication indicating the selected candidate model for model switching operation.

Advantageously, the model switching operation is autonomously initiated by the second node by selecting the candidate model based on a monitored for switching conditions, and a model switching status update is transmitted to the first node. Advantageously, the conditions being characterized by indicating an activation time delay for the model switching operation, the conditions corresponding to one of the following:

• A non-applicable model state condition, indicating that the corresponding candidate model is available in a model registry but not available for activation,

• An applicable model state condition, indicating that the corresponding candidate model is available for activation and its readiness status is not in an executable format;

• A partially-loadable/in-active model state condition, indicating that the corresponding candidate model is in an executable format and not fully integrated with related applications;

• A fully-loadable/pre-active model state condition, indicating that the corresponding candidate model is pre-loaded for activation with full integration completion; Or

• An active model state mode condition, indicating that the corresponding candidate model is already activated.

Advantageously, the state mode conditions are obtained via measurements of applicable conditions.

Advantageously, the indexed model state modes are differentiated based on a time duration range of activating their corresponding candidate model with a pre-configured timing threshold value, the indexed order corresponding to one of the following:

• A lowest indexed model state mode has the highest amount of time to activate the corresponding candidate model;. and

• A highest indexed model state mode has the lowest amount of time to activate the corresponding candidate model.

Advantageously, the candidate models have multiple model state modes based on a combination of different parameters and/or an AI/ML applicable condition information, the mapping relationship being characterized by one of the following: • The mapping relationship is configured by the fist node or the second node or a joint effort of both nodes;

• The mapping relationship is provided via a RRC re-configuration message;

• about the mapping relationship is updated via L1 or L2 signaling;

• The mapping relationship information is broadcasted or multicasted in case of grouped nodes.

Advantageously, the multiple model state modes of a candidate model, mode transitions between the multiple model state modes are defined, the mode transitions being pre-configurable and applicable for different use cases, a mode transition information to configure the multiple mode state modes being signaled via RRC.

Advantageously, the candidate model is selected by the first node, the second node, or a joint effort between the first node and the second node.

Advantageously, the selection of the candidate model is based on a model collaboration type, a UE ML capability, a model meta dataset, or a ML applicable condition information.

Advantageously, the mode transitions are configured based on a timer, the timer being characterized by at least one of the following:

• The timer is set to allow a mode transition to another model state mode;

• A mode transition to another model state mode is initiated, before the timer expires;

• A mode transition to the previous model state is initiated, if the timer expires without a mode transition;

• Multiple timers are configured for different combinations of mode transitions.

According to a second aspect of the invention, an apparatus for activating a candidate model for an AI/ML model switching operation in a wireless network comprises a wireless transceiver, a processor coupled with a memory in which computer program instructions are stored, said instructions being configured to implement above-described steps.

According to a third aspect of the invention, a user equipment (UE) for activating a candidate model for an AI/ML model switching operation in a wireless network comprises the above-described apparatus, wherein the UE receives the mapping relationship, monitores the activated model for switching conditions, and, if the switching conditions are fulfilled, initiates the model switching, and transmits a model switching status update, and, if model switching conditions are no fulfilled, monitores the activated candidate model.

According to a fourth aspect of the invention, a base station (gNB) for activating a candidate model for an AI/ML model switching operation in a wireless network, comprises the above-described apparatus, wherein the configuration of the mapping relationship, transmitting of the mapping relationship to the second node, receiving the model switching status update and transmitting the switching indication, is proceeded.

According to a sixth aspect of the invention, a wireless communication system for activating a candidate model for an AI/ML model switching operation in a wireless network, comprising the above-described base station (gNB), the base sation comprising a processor coupled with a memory in which computer program instructions are stored, said instructions being configured to implement above-described steps, the wireless communication system comprising the above-described user equipment (UE), the UE comprising a processor coupled with a memory in which computer program instructions are stored, said instructions being configured to implement above-described steps.

In the method disclosed herein, firstly a finite set of candidate models to be used for model switching is configured to be indexed for a mapping relationship with pre-configured model state modes. Specifically, each candidate models are indicated with different model state modes that represent different levels of a readiness status for a model switching operation. For example, some candidate models can be in a state mode of performing fast model activation with a pre-loaded status and other candidate models can be in state mode of taking more time for activation by not being pre-loaded. Mapping of model state modes to candidate models is configured based on a combination of different parameters and/or ML applicable condition information. For example, when a UE device has little resources to execute additional models along with the already activated models, then, any candidate model might not be allowed to be pre-loaded for model switching in advance. If there is little chance for the activated model to drift or fail for any LCM operation, then, it would not be necessary to make candidate models ready for pre-loading, since a pre-active state consumes additional device resources. Therefore, when determining model state modes for candidate models, the related assistance information such as device ML capability/resource, model meta data, LCM phase, or ML use case/application, must be available to determine accurate model state modes for each candidate models.

In addition, model state modes and associated description information are pre-configured to be used for the mapping relationship tables of model state modes and candidate models. The pre-configured model state mode information can be signaled through system information and/or RRC message.

Figure 1 shows an exemplary table for type-1 model state modes. In this table, five indexed model state modes are configured with specific conditions and each state mode has different levels of activation time delay for a model switching operation. Specifically, a non-applicable model state mode indicates that a candidate model is available in model registry and is not supportable for activation at the time of a measurement for applicable conditions to activate a model. An applicable model state mode indicates that a candidate model is supportable for activation and its activation readiness is not in executable format. A partially-loadable model state mode indicates that a candidate model is in an executable format and not fully integrated with other related applications. A fully-loadable model state mode indicates that a candidate model is pre-loaded for activation with full integration completion. Among all indexed model state modes, the fully-loadable model state mode is the fastest way to activate a candidate model. An active model state mode indicates that a candidate model is already activated as an alternative after executing the model switching operation. This exemplary table can be further expanded with more indexes and different model state modes depending on implementation applications.

Figure 2 shows an exemplary table for type-2 model state modes. In this table, five indexed model state modes are configured and each state mode has a different level of an activation time delay for a model switching operation. This type-2 table is similar compared to the type-1 table of Fig.1 . However, model state modes are differently named based on a different level of activation readiness for the model switching operation and this table can be also expanded to have more indexed model state modes if necessary. Other types of model state mode tables can be possible. For example, with the indexed model state modes, each state mode is differentiated based on a time duration range for activating a candidate model with pre-configured timing threshold values and the criteria of splitting model state modes then depend on different time consumptions of alternative model activation/switching. In this approach of using time duration ranges for model state modes, the lowest indexed model state mode has the highest amount of time to activate a model and the highest indexed model state mode has the lowest amount of time to activate a model. Different orders of indexing with associated time duration ranges and/or timing threshold values are applicable.

Figure 3 shows an exemplary table of a mapping relationship for candidate models and model state modes. This table indicates that each candidate model has a different model state mode based on a combination of different parameters and/or ML applicable condition information (e.g., device ML capability/resource, model meta data, LCM phase, ML use case/application, etc.). The mapping relationship for candidate models and model state modes can be configured by either the network side or the UE side or can even be a joint effort of both sides. For the signaling aspect, the initial mapping relationship for the candidate models and model state modes can be provided through a RRC re-configuration message and any additional update/change information about this mapping relationship can be sent through L1 or L2 signaling. For UE group based model switching scenarios, the mapping relationship information can be broadcasted or multicasted for grouped UEs (e.g., system information or RRC signaling).

Figure 4 shows an exemplary block diagram of model state mode transitions. In these model state mode transitions, different combinations of model state mode transitions are applicable for different use cases and a finite set of model state mode transitions can be configured so that model state mode transitions can be limited to the pre-configured scenarios. The configured model state mode transition information can be provided through system information or RRC signaling.

Figure 5 shows a flowchart of a network behavior for configuring the mapping relationship for candidate models and model state modes. In this flowchart, the network side configures and transmits the mapping relationship information of candidate models and model state modes. Before this mapping relationship configuration, candidate models are determined with the associated index information and the list of candidate models can be created either by the network side or by the UE side or by a joint effort of both sides. The determination of candidate models is based on the model collaboration type, UE ML capability, model meta data, ML applicable condition information with applications, etc. After receiving a model monitoring report from the UE, the model switching operation can be initiated by the network side or by the UE side or by a joint effort of both sides. If the model switching operation is initiated by the network, an indication signal of activating candidate models is sent to the UE.

Figure 6 shows a flowchart of UE behavior for an UE autonomous decision on a model switching operation. In this flowchart, the model switching operation is based on an UE autonomous decision and a model switching update is reported to the network side after activating alternative models determined by the UE autonomously. The configured conditions of activating candidate model(s) are based on pre-configured information and measurement information, such as a combination of different parameters and/or ML applicable condition information (e.g., device ML capability/resource, model meta data, LCM phase, ML use case/application, etc.). Figure 7 shows a flowchart of UE behavior for a network decision on a model switching operation. In this flowchart, the model switching operation is determined by the network side and the UE receives an indication of candidate model information for activation from the network side.

When candidate models are pre-determined by the network side and activation of a candidate model is decided by the network side, a candidate model list and target applicable conditions for the candidate models are configured by the network side and an indication of activating candidate model(s) for the UE is transmitted. At the UE side, active model performance is monitored and a model monitoring status is reported to the network side. An indication of activating candidate model(s) is also received from the network side so that the indicated candidate model(s) can be activated. When candidate models are pre-determined by the network side and activation of candidate model is decided by the UE, the activation of candidate model(s) is determined by the UE autonomously and an update status is reported to the network. When candidate models are pre-determined by the UE and the activation of a candidate model is decided by the network side, a candidate model list is determined by the UE and the indicated candidate model(s) can be activated when activation of candidate model(s) is determined by the network side. When candidate models are pre-determined by the UE and activation of candidate model is decided by the UE, both candidate model list and activation of candidate model(s) are determined by the UE based on pre-configured conditions of model switching and/or alternative model activation.

Figure 8 shows a flowchart of a timer-based model state mode transition. For a transition of model state modes, a timer can be set to allow one model state mode to transition to another model state mode. For example, once the timer is set, the model state mode can be transitioned before the timer expires. If no model state mode transition occurs before the timer expires, the model state mode is transitioned to a previous state mode, (e.g., from pre-active to inactive, from inactive to applicable, etc.). Multiple timers can be configured for different combinations of model state mode transitions that can be preset. To use the mapping relationship of candidate models and model state modes, sidelink based UE-to-UE can apply the same method for model switching operations with different configuration content and/or model state mode descriptions/conditions.

Abbreviations:

Al Artificial intelligence BWP Bandwidth part CBG Code block group CLI Cross Link Interference CP Cyclic prefix CQI Channel quality indicator CPU CSI processing unit CRB Common resource block CRC Cyclic redundancy check CRI CSI-RS Resource Indicator CSI Channel state information CSI-RS Channel state information reference signal CSI-RSRP CSI reference signal received power CSI-RSRQ CSI reference signal received quality CSI-SINR CSI signal-to-noise and interference ratio CW Codeword DCI Downlink control information DL Downlink DM-RS Demodulation reference signals DRX Discontinuous Reception EPRE Energy per resource element IAB-MT Integrated Access and Backhaul - Mobile Terminal ID Identificator L1 -RSRP Layer 1 reference signal received power LI Layer Indicator LCM Life cycle management MCS Modulation and coding scheme ML Machine learning NW Network PDCCH Physical downlink control channel PDSCH Physical downlink shared channel PSS Primary Synchronisation signal PUCCH Physical uplink control channel QCL Quasi co-location PMI Precoding Matrix Indicator PRB Physical resource block PRG Precoding resource block group PRS Positioning reference signal PT-RS Phase-tracking reference signal RAN Radio Access Network RB Resource block RBG Resource block group Rl Rank Indicator RIV Resource indicator value RS Reference signal SCI Sidelink control information SLIV Start and length indicator value SR Scheduling Request SRS Sounding reference signal SS Synchronisation signal SSS Secondary Synchronisation signal SS-RSRP SS reference signal received power

SS-RSRQ SS reference signal received quality SS-SINR SS signal-to-noise and interference ratio TB Transport Block TCI Transmission Configuration Indicator TDM Time division multiplexing UE User equipment UL Uplink WG Work group

Claims

1 . A method of activating a candidate model out of a set of candidate models for an AI/ML model switching operation in a wireless network, the method comprising the steps:

• Configuring, by a first node of the wireless network, a mapping relationship by allocating a model state mode to each of the candidate models, the model state mode indicating a readiness status for the model switching operation,

• Receiving, by a second node of the wireless network, of the mapping relationship,

• Selecting the candidate model for the model switching operation based on the received mapping relationship, and

• Performing the model switching operation, during which the candidate model is activated.

2. The method according to claim 1 , characterized in that the set of candidate models, the mapping relationship, and/or the model state modes are indexed.

3. The method according to claim 1 or 2, characterized in that the mapping relationship comprises conditions associated with the model state modes.

4. The method according to any of the previous claims, characterized in that the model state modes are pre-configured.

5. The method according to any of the previous claims, characterized in that a switching indication is transmitted from the first node to the second node, the switching indication indicating the selected candidate model for model switching operation.

6. The method according to any of the previous claims, characterized in that the model switching operation is autonomously initiated by the second node by selecting the candidate model based on a monitored for switching conditions, and a model switching status update is transmitted to the first node.

7. The method according to claim 3, characterized in that the conditions being characterized by indicating an activation time delay for the model switching operation, the conditions corresponding to one of the following:

8. The method according to claim 7, characterized in that the state mode conditions are obtained via measurements of applicable conditions.

9. The method according to claim 2, characterized in that the indexed model state modes are differentiated based on a time duration range of activating their corresponding candidate model with a pre-configured timing threshold value, the indexed order corresponding to one of the following:

10. The method according to any of the previous claims, characterized in that the candidate models have multiple model state modes based on a combination of different parameters and/or an AI/ML applicable condition information, the mapping relationship being characterized by one of the following:

• The mapping relationship is configured by the fist node or the second node or a joint effort of both nodes;

• The mapping relationship is provided via a RRC re-configuration message;

• about the mapping relationship is updated via L1 or L2 signaling;

11 . The method according to claim 10, characterized in that for the multiple model state modes of a candidate model, mode transitions between the multiple model state modes are defined, the mode transitions being pre-configurable and applicable for different use cases, a mode transition information to configure the multiple mode state modes being signaled via RRC.

12. The method according to any previous claims, characterized in that the the candidate model is selected by the first node, the second node, or a joint effort between the first node and the second node.

13. The method according to claim 12, characterized in that the selection of the candidate model is based on a model collaboration type, a UE ML capability, a model meta dataset, or a ML applicable condition information.

14. The method according to claim 11 , characterized in that the mode transitions are configured based on a timer, the timer being characterized by at least one of the following:

The timer is set to allow a mode transition to another model state mode; • A mode transition to another model state mode is initiated, before the timer expires;

15. An apparatus for activating a candidate model for an AI/ML model switching operation in a wireless network comprises a wireless transceiver, a processor coupled with a memory in which computer program instructions are stored, said instructions being configured to implement steps of the claims 1 to 14.

16. A user equipment (UE) for activating a candidate model for an AI/ML model switching operation in a wireless network comprises the apparatus according to claim 15, wherein the UE receives the mapping relationship, monitores the activated model for switching conditions, and, if the switching conditions are fulfilled, initiates the model switching, and transmits a model switching status update, and, if model switching conditions are no fulfilled, monitores the activated candidate model.

17. A base station (gNB) for activating a candidate model for an AI/ML model switching operation in a wireless network, comprises the apparatus according to claim 15, wherein the configuration of the mapping relationship , transmitting of the mapping relationship to the second node, receiving the model switching status update and transmitting the switching indication, is proceeded.

18. A wireless communication system for activating a candidate model for an AI/ML model switching operation in a wireless network, comprising the base station (gNB) according to claim 17, the base sation comprising a processor coupled with a memory in which computer program instructions are stored, said instructions being configured to implement steps of claims 1 to 14, the wireless communication system comprising the user equipment (UE) according to claim 16, the UE comprising a processor coupled with a memory in which computer program instructions are stored, said instructions being configured to implement steps of the claims 1 to 14.