Methods to enable separate training for two-sided ai / ML models using nominal models

Abstract

A method in a user equipment (UE) configured to train and validate performance of one or more two-sided models is described. The UE is configured to communicate with a network node (NW). The method includes training a second UE-part model of one or more two-sided models based on a first union of a first data set and a second data set, and a first assumption by the UE that the NW uses a first NW-part model. The first data set is usable to train a first UE-part model of the one or more two-sided models. The method also includes validating performance of the second UE-part model based on a test data set and the first NW-part model and performing one or more actions based on the validation of the performance of the second UE-part model.

Description

[0001] METHODS TO ENABLE SEPARATE TRAINING FOR TWO-SIDED AI / ML MODELS USING NOMINAL MODELS

[0002] TECHNICAL FIELD

[0003] The present disclosure relates to wireless communications, and in particular, to separate training for two-sided models using nominal models.

[0004] BACKGROUND

[0005] The Third Generation Partnership Project (3GPP) has developed and is developing standards for Fourth Generation (4G) (also referred to as Long Term Evolution (LTE)) and Fifth Generation (5G) (also referred to as New Radio (NR)) wireless communication systems. Such systems provide, among other features, broadband communication between network nodes, such as base stations, and mobile user equipments (UE), as well as communication between network nodes and between WDs. The 3GPP is also developing standards for Sixth Generation (6G) wireless communication networks.

[0006] Artificial Intelligence (Al) and Machine Learning (ML) have been investigated, both in academia and industry, as promising tools to optimize the design of the air interface in wireless communication networks. Example use cases include: using autoencoders for Channel State Information (CSI) compression to reduce the feedback overhead and improve channel prediction accuracy; using deep neural networks for classifying Line-of-Sight (LOS) and Non-LOS (NLOS) conditions to enhance the positioning accuracy; and using reinforcement learning for beam selection at the network side and / or the UE side to reduce the signaling overhead and beam alignment latency; and using deep reinforcement learning to leam an optimal precoding policy for complex Multiple Input Multiple Output (MIMO) precoding problems.

[0007] An NR air interface may support AI / ML-based algorithms for enhanced performance and / or reduced complexity / overhead, e.g., as described in the 3GPP NR standardization works for 3GPP Release-18 (Rel-18), or to be described in works for 3GPP Release-19 (Rel-19). Through studying and specifying a few selected use cases (CSI feedback, beam management, and positioning), the works aim to design the mechanisms to accommodate AI / ML into the 3 GPP standard.

[0008] General aspects for 3 GPP NR Rel-18 AI / ML for NR air interface

[0009] LCM operations for AI / ML for NR air interface Al development and operation may include lifecycle management (LCM) of the AI / ML model (e.g., model training, model deployment, model inference, model monitoring, model updating) and AI / ML functionality.

[0010] In a study item associated with NR Rel-18 AI / ML for NR air interface, the LCM procedure is described for the case that an AI / ML model has a model ID with associated information and / or for the case that a given functionality is provided by some AI / ML operations. Two types of LCM operations are described: (A) functionality -based LCM; and (B) model-ID-based LCM.

[0011] Functionality refers to an AI / ML-enabled Feature / FG enabled by configuration(s), where configured on(s) is(are) supported based on conditions indicated by UE capability. Correspondingly , functionality-based LCM operates based on, at least, one configuration of an AI / ML-enabled Feature / FG or specific configurations of an AI / ML-enabled Feature / FG. In functionality-based LCM, the network indicates activation and / or deactivation and / or fallback and / or switching of AI / ML functionality via 3 GPP signaling (e.g., Radio Resource Control (RRC), Medium Access Control (MAC)- Control Element (CE), Downlink Control Information (DCI)). Models may not be identified at the network, and UE may perform model-level LCM. Whether and how much awareness / interaction network should have about model-level LCM requires further study. For functionality identification, there may be either one or more than one functionality defined within an AI / ML-enabled feature, whereby AI / ML-enabled Feature refers to a Feature where AI / ML may be used.

[0012] In model-ID-based LCM, models are identified at the network, and network / UE may activate / deactivate / select / switch individual AI / ML models via model ID. A model may be associated with specific configurations / conditions associated with the UE capability of an AI / ML-enabled Feature / FG and additional conditions (e.g., scenarios, sites, and datasets) as determined / identified between the UE side and network side. An AI / ML model identified by a model ID may be logical, and how it maps to physical AI / ML model(s) may be up to implementation.

[0013] Functional framework for AI / ML model LCM

[0014] FIG. 1 shows an example functional framework for AI / ML for NR Air Interface. More specifically, a functional framework that can be used for studying model LCM aspects for different Al for PHY use cases is shown. The general framework includes the following: - Data Collection may refer to a function that provides input data to the Model Training, Management, and Inference functions. o Training Data: Data needed as input for the AI / ML Model Training function. o Monitoring Data: Data needed as input for the Management of AI / ML models or AI / ML functionalities. o Inference Data: Data needed as input for the AI / ML Inference function.

[0015] - Model Training may refer to a function that performs AI / ML model training, validation, and testing which may generate model performance metrics that can be used as part of the model testing procedure. The Model Training function may be also responsible for data preparation (e.g., data pre-processing and cleaning, formatting, and transformation) based on Training Data delivered by a Data Collection function if required. o Trained / Updated Model: In case of having a Model Storage function, this is used to deliver trained, validated, and tested AI / ML models to the Model Storage function, or to deliver an updated version of a model to the Model Storage function.

[0016] - Management may refer to a function that oversees the operation (e.g., selection, (de)activation, switching, fallback) and monitoring (e.g., performance) of AI / ML models or AI / ML functionalities. This function is also responsible for making decisions to ensure the proper inference operation based on data received from the Data Collection function and the Inference function. o Management Instruction: Information needed as input to manage the Inference function. Concerning information may include selection / (de)activation / switching of AI / ML models or AI / ML-based functionalities, fallback to non- AI / ML operation (i.e., not relying on inference process), etc. o Model Transfer / Delivery Request: Used to request model(s) to the Model Storage function. o Performance Feedback / Retraining Request: Information needed as input for the Model Training function, e.g., for model (re)training or updating purposes.

[0017] - Inference may refer to a function that provides outputs from the process of applying AI / ML models or AI / ML functionalities, using the data that is provided by the Data Collection function (i.e., Inference Data) as an input. The Inference function is also responsible for data preparation (e.g., data preprocessing and cleaning, formatting, and transformation) based on Inference Data delivered by a Data Collection function, if required. o Inference Output: Data used by the Management function to monitor the performance of AI / ML models or AI / ML functionalities.

[0018] - Model Storage may refer to a function responsible for storing trained / updated models that can be used to perform the Inference function. o Note: The Model Storage function in FIG. 1 may be used as a reference point for protocol terminations, model transfer / delivery, and related processes. o Model Transfer / Delivery: Used to deliver an AI / ML model to the Inference function.

[0019] UE-NW collaboration levels for one-and two-sided AI / ML models

[0020] The AI / ML models (e.g.., discussed in the Rel-18 study item on AI / ML for the NR air interface) can be categorized into the following two types:

[0021] - One-sided AI / ML model, which can be a UE-sided AI / ML model whose inference is performed entirely at the UE, or a network node sided (NW-sided), e.g., network node, AI / ML model whose inference is performed entirely at the network node (NW) or network.

[0022] - Two-sided AI / ML model, which refers to a paired AI / ML Model(s) over which joint inference is performed across the UE and the NW, i.e., the first part of the inference is performed by UE and then the remaining part is performed by network node (e.g., gNB), or vice versa. As an example of a two-sided AI / ML model, FIG. 2 shows a use case of autoencoder (AE)-based CSI feedback / report, where an encoder (UE-part of the two-sided AE model) is operated at a UE to compress the estimated wireless channel, and the output of the encoder (the compressed wireless channel information estimates) is reported from the UE to a network node (e.g., gNB). The network node may use a decoder (NW-part of the two-sided AE model) to reconstruct the estimated wireless channel information. The two-sided AI / ML model may include an encoder at the UE side and a decoder at the network node side (e.g., the base station (BS) side, a gNB side, etc.). In the case of a two-sided model, the code is generated by the encoder and only interpretable by a jointly trained decoder. The situation is different from running an AI / ML model in the UE, reporting the output over the air in a fully standardized format, and running a separate AI / ML model at the network node (e.g., BS).

[0023] When applying AI / ML on air interface use cases, different levels of collaboration between network nodes and UEs can be considered:

[0024] - No collaboration between network nodes and UEs. In this case, a proprietary ML model operating with the existing standard air-interface is applied at one end of the communication chain (e.g., at the UE side), and the model life cycle management (e.g., model selection / training, model monitoring, model retraining, model update) is done at this node without inter-node assistance (e.g., assistance information provided by the network node).

[0025] - Limited collaboration between network nodes and UEs for one-sided models. In this case, an ML model is operating at one end of the communication chain (e.g., at the UE side), but this node gets assistance from the node(s) at the other end of the communication chain (e.g., a next generation Node B (gNB)) for its Al model life cycle management to some extent (e.g., for training / retraining the Al model, model update, model monitoring, model selection / fallback / switching).

[0026] - Joint ML operation between network nodes and UEs for two-sided models. In this case, it may be assumed that the Al model is split with one part located at the network node (NW) side (i.e., network side) and the other part located at the UE side. Hence, the Al model may require joint inference between the network node and UE, and the Al model life cycle management involves both ends of a communication chain.

[0027] Model training for one- and two-sided models

[0028] For two-sided models, the model training process may require sharing of data from one side to the other side since the input and output of a two-sided model reside within different vendor’s domain. Different model training types can be considered for two-sided model training between M network vendors and N UE / chip-set vendors, where M>=1 and N>=1:

[0029] - Type 1 : Joint training of the two-sided model at a single side / entity, e.g., the UE side or the NW side. For instance, a two-sided model (UE-part model and NW-part model) is trained at the NW side (e.g., by a NW vendor), then, the UE-part of the trained model (e.g., encoder for the AE-based CSI compression use case) is transferred / delivered from the NW-side to the UE-side, and vice versa.

[0030] - Type 2: Joint training of the two-sided model at network side and UE side, respectively. Joint training can be done simultaneously at the network and UE sides or be performed in a sequential way. In case of Type 2 simultaneous joint training, the UE-part model (trained at the UE side) and the NW-part model (trained at the NW side) are jointly trained in the same loop through exchanging forward propagation values and backward propagation values between NW and UE. In case of Type 2 sequential joint training, one side (UE- side or NW-side) starts its model training first, it then opens an API to facilitate the other side to do the model training. For instance, the NW side trains its model first (thus also obtaining what is sometimes known as a nominal encoder, but that is not used at the UE), and then the UE side can train its encoder by using an Application Programming Interface (API). The API may accept, e.g., a CSI report and a target CSI, both which are derived by the UE side based on the data (note that the CSI report is generated, at least partially, by the UE encoder under training and may thus not be an efficient CSI report at each step in the training). The API may return gradients of the decoder and a loss function, with respect to the variables in the CSI report. Thus, the UE may train an encoder that is matched to the decoder.

[0031] - Type 3: Sequential training starting with UE side training or sequential training starting with NW side training, where the UE-part model and the NW-part model are trained by UE side and network side, respectively. Take sequential training with NW-first training approach for AE-based CSI compression as an example, the NW can firstly train the UE-part and NW-part models jointly using training data (e.g., target CSI samples), and then share a dataset consisting of UE-part model output (e.g., latent space variables) associated to the ground-truth / labels (e.g., target CSI) for the UE side to train its UE-part model (e.g., an encoder). Alternatively, the NW can share a dataset consisting of gradients of the NW-part model (e.g., the gradients of the decoder) together with loss function value indicating the discrepancy of the NW-part model output (e.g., the decoder output) and the ground-truth / labels (e.g., target CSI) with respect to the UE-part model output (e.g., latent space variables), based on which the UE-side trains its UE-part model (e.g., an encoder). At least from Rel-18 RAN 1 ’s discussion on inter-vendor training collaboration and RAN4’s discussion on encoder testing options, both model design flexibility and interoperability may be supported for the two-sided CSI compression use case. In cases where two-sided models are used, if UE vendors or network vendors each define their own models, then there may be a strong risk that this inter-operability will be undermined. Networks would need to be trained with (and operators may need to run) a large number of decoder models in parallel and in real time, and UEs may need to be trained for and be able to load different AI / ML models depending on to which network the UEs are connected. With these many models, ensuring robust performance meeting Radio Access Network 4 (RAN4) requirements becomes extremely complex. There is also a risk that some UEs may not be able to run their models in certain networks and vice-versa, hence the interoperability benefits cannot be achieved.

[0032] Further, interoperability can be maintained if one-part of the two-sided model is specified. However, it is also acknowledged that defining a model is a challenging task, particularly when there is no access to predetermined quality field data during the process of standardizing that model. If the data assumed for training the standardized model does not sufficiently capture different channel conditions or aspects that can be faced in real deployment, the model will fail to generalize or provide good performance, thereby severely limiting the utilization of two-sided models and two-sided CSI compression.

[0033] SUMMARY

[0034] Some embodiments advantageously provide methods, systems, and apparatuses for separate training for two-sided models using nominal models.

[0035] One or more embodiments provide inter-vendor collaboration to train two-sided models that maintains interoperability in a multi-vendor eco-system, while at the same time providing engineering freedom to both UE (and / or associated components or devices) and the network node (and / or associated components or nodes) to optimize their own model independently. Interoperability may be achieved at a nominal performance level by defining a nominal UE-part and / or NW-part model trained using a first data set, which is selected to achieve a nominal performance level.

[0036] Performance that exceed the performance of existing technology may be achieved by single sided embodiments, while still maintaining interoperability. In one or more embodiments, UE / UE-side and / or network node / network node side may obtain a more accurate training dataset by matching field data distribution. The dataset that may be based on field measurements may be used at the UE-side and / or NW-side to enhance its own nominal model independently from the other side, and give better model performance, with low standardization complex! ty / effort to support the inter-vendor collaboration to train two-sided models.

[0037] According to one aspect, a method in a user equipment (UE) configured to train and validate performance of one or more two-sided models is described. The UE is configured to communicate with a network node (NW). The method includes training a second UE-part model of one or more two-sided models based on a first union of a first data set and a second data set, and a first assumption by the UE that the NW uses a first NW-part model. The first data set is usable to train a first UE-part model of the one or more two-sided models. The method also includes validating performance of the second UE-part model based on a test data set and the first NW-part model and performing one or more actions based on the validation of the performance of the second UE-part model.

[0038] In some embodiments, the method further includes training the first UE-part model using the first data set and based on the first assumption.

[0039] In some other embodiments, one or both of: the first UE-part model is a nominal UE-part model or a reference UE-part model; and the first NW-part model is a nominal NW-part model or a reference NW-part model.

[0040] In some embodiments, validating the performance of the second UE-part model includes determining whether the second UE-part model fulfills one or more performance requirements, and the one or more actions include transmitting, to the NW, a first indication indicating an availability of the second UE-part model at the UE.

[0041] In some other embodiments, the method further includes receiving, from the NW, a second indication indicating the UE to one or more of: (A) activate use the second UE-part model by the UE; (B) deactivate use of the second UE-part model by the UE; and (C) fall back to the first UE-part model.

[0042] In some embodiments, the one or more actions include determining a model inference output using the second UE-part model and transmitting, to the NW, a third indication indicating the model inference output.

[0043] In some other embodiments, the model inference output includes an artificial intelligence based channel state information report.

[0044] In some embodiments, the method further includes: (A) training a nominal test NW-part model using the first data set and the first UE-part model; (B) training the second UE-part model of one or more two-sided models based on a second union of the first data set and the second data set, and the nominal test NW-part model; and (C) validating performance of the second UE-part model using the test data and the nominal test NW- part model.

[0045] In some embodiments, the method further includes receiving, from the NW, the second data set including information usable to reconstruct the second UE-part model and reconstructing the second UE-part model based on information included in the second data set.

[0046] In some other embodiments, the method further includes receiving, from the NW, a joint data set usable by the UE to train the second UE-part model and by the NW to train a second-NW part model.

[0047] According to another aspect, a user equipment (UE) configured to train and validate performance of one or more two-sided models is described. The UE is configured to communicate with a network node (NW) and to perform any one of the steps of any one of the embodiments implemented in the UE.

[0048] According to one aspect, a method in a network node (NW) configured to train and validate performance of one or more two-sided models is described. The NW is configured to communicate with a user equipment (UE). The method includes training a second NW-part model of one or more two-sided models based on a third union of a first data set and a third data set, and a first NW-part model. The first data set and the first NW- part model are usable to train a nominal test UE-part model. The method also includes validating performance of the second NW-part model based on a test data set and the nominal test UE-part model and performing one or more actions based on the validation of the performance of the second NW-part model.

[0049] In some embodiments, the method further includes training the nominal test UE- part model using the first data set and the first NW-part model.

[0050] In some other embodiments, the first NW-part model is a nominal NW-part model or a reference NW-part model.

[0051] In some embodiments, the method further includes receiving, from the UE, a first indication indicating an availability of a second UE-part model at the UE.

[0052] In some other embodiments, validating the performance of the second NW-part model includes determining whether the second NW-part model fulfills one or more performance requirements, and the one or more actions include transmitting, to the UE, a second indication indicating the UE to one or more of: (A) activate use the second UE-part model by the UE; (B) deactivate use of the second UE-part model by the UE; and (C) fall back to the first UE-part model. In some embodiments, the method further includes receiving, from the UE, a third indication indicating a model inference output, the model inference output being based on a second UE-part model.

[0053] In some other embodiments, the model inference output includes an artificial intelligence based channel state information report.

[0054] In some embodiments, the method further includes transmitting, to the UE, a second data set including information usable by the UE to reconstruct the second UE-part model.

[0055] In some other embodiments, the method further includes transmitting, to the UE, a joint data set usable by the UE to train the second UE-part model and by the NW to train a second-NW part model.

[0056] In some embodiments, the method further includes determining whether to fall back to using the first NW-part model based on the validation.

[0057] According to one aspect, a network node (NW) configured to train and validate performance of one or more two-sided models is described. The NW is configured to communicate with a user equipment (UE) and to perform any one of the steps of any one of the method implemented in the NW.

[0058] BRIEF DESCRIPTION OF THE DRAWINGS

[0059] A more complete understanding of the present embodiments, and the attendant advantages and features thereof, will be more readily understood by reference to the following detailed description when considered in conjunction with the accompanying drawings wherein:

[0060] FIG. 1 shows an example functional framework for AI / ML for NR air interface;

[0061] FIG. 2 shows an example AE-based CSI compression using two-sided AI / ML model use case;

[0062] FIG. 3 is a schematic diagram of an example network architecture illustrating a communication system connected via an intermediate network to a host computer according to the principles in the present disclosure;

[0063] FIG. 4 is a block diagram of a host computer communicating via a network node with a user equipment over an at least partially wireless connection according to some embodiments of the present disclosure;

[0064] FIG. 5 is a flowchart of an example process in a user equipment according to some embodiments of the present disclosure; FIG. 6 is a flowchart of an example process in a network node according to some embodiments of the present disclosure;

[0065] FIG. 7 is a flowchart of an example process in a user equipment according to some embodiments of the present disclosure; and

[0066] FIG. 8 is a flowchart of an example process in a network node according to some embodiments of the present disclosure.

[0067] DETAILED DESCRIPTION

[0068] Before describing in detail example embodiments, it is noted that the embodiments reside primarily in combinations of apparatus components and processing steps related to separate training for two-sided models using nominal models. Accordingly, components have been represented where appropriate by conventional symbols in the drawings, showing only those specific details that are pertinent to understanding the embodiments so as not to obscure the disclosure with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein. Like numbers refer to like elements throughout the description.

[0069] As used herein, relational terms, such as “first” and “second,” “top” and “bottom,” and the like, may be used solely to distinguish one entity or element from another entity or element without necessarily requiring or implying any physical or logical relationship or order between such entities or elements. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the concepts described herein. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes” and / or “including” when used herein, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0070] In embodiments described herein, the joining term, “in communication with” and the like, may be used to indicate electrical or data communication, which may be accomplished by physical contact, induction, electromagnetic radiation, radio signaling, infrared signaling or optical signaling, for example. One having ordinary skill in the art will appreciate that multiple components may interoperate, and modifications and variations are possible of achieving the electrical and data communication. In some embodiments described herein, the term “coupled,” “connected,” and the like, may be used herein to indicate a connection, although not necessarily directly, and may include wired and / or wireless connections.

[0071] The term “network node” (or NW) used herein can be any kind of network node comprised in a radio network which may further comprise any of base station (BS), radio base station, base transceiver station (BTS), base station controller (BSC), radio network controller (RNC), g Node B (gNB), evolved Node B (eNB or eNodeB), Node B, multistandard radio (MSR) radio node such as MSR BS, multi-cell / multicast coordination entity (MCE), integrated access and backhaul (IAB) node, relay node, donor node controlling relay, radio access point (AP), transmission points, transmission nodes, Remote Radio Unit (RRU) Remote Radio Head (RRH), a core network node (e.g., mobile management entity (MME), self-organizing network (SON) node, a coordinating node, positioning node, MDT node, etc.), an external node (e.g., 3rd party node, anode external to the current network), nodes in distributed antenna system (DAS), a spectrum access system (SAS) node, an element management system (EMS), etc. The network node may also comprise test equipment. The term “radio node” used herein may be used to also denote a user equipment (UE) such as a wireless device (WD) or a radio network node. In some embodiments, the term network node is referred to as “network” or “NW”.

[0072] In some embodiments, the non-limiting terms wireless device (WD) or a user equipment (UE) are used interchangeably. The UE herein can be any type of wireless device capable of communicating with a network node or another UE over radio signals, such as a wireless device (WD). The UE may also be a radio communication device, target device, device to device (D2D) UE, machine type UE or UE capable of machine to machine communication (M2M), low-cost and / or low-complexity UE, a sensor equipped with UE, Tablet, mobile terminals, smart phone, laptop embedded equipped (LEE), laptop mounted equipment (LME), USB dongles, Customer Premises Equipment (CPE), an Internet of Things (loT) device, or a Narrowband loT (NB-IOT) device, etc.

[0073] Also, in some embodiments the generic term “radio network node” is used. It can be any kind of a radio network node which may comprise any of base station, radio base station, base transceiver station, base station controller, network controller, RNC, evolved Node B (eNB), Node B, gNB, Multi-cell / multicast Coordination Entity (MCE), IAB node, relay node, access point, radio access point, Remote Radio Unit (RRU) Remote Radio Head (RRH). In one or more embodiments, the term “network” refers to network node (or NW) and / or any of other network node components. In some embodiments, the term networkside or NW-side or network node side is used and may refer to the network node (and / or any of its components). In some other embodiments, the term UE-side is used and may refer to the UE (and / or any of its components).

[0074] In some embodiments, the term “model” is used and may refer to an AI / ML model or any other model. Further, a model may be a one-sided model (e.g., a one-sided AI / ML model), a two-sided model (e.g., a two-sided AI / ML model), or an N-sided model (e.g., where more than two entities are involved in, engage with, or associated with the model). The “side” associated with the model may refer to a system component, device, node, etc. For example, a UE-sided model may be a model associated with the UE (where inferences and other actions associated with the model are performed entirely at (or a predetermined portion of the actions are performed by) the UE). Similarly, a network node sided (NW- sided) may be associated with the network node (where inferences and other actions associated with the model are performed entirely at (or a predetermined portion of the actions are performed by) the network node).

[0075] A two-sided model may refer to a model for which at least some actions associated with the model are performed jointly by more than one entity (e.g., by the network node and the UE). In a nonlimiting example, a two-sided model may refer to a paired AI / ML model over which one or more joint actions are performed by the UE and the NW. The joint actions may include at least one part of an inference being performed by UE and at least another part of the inference being performed by network node. A two-sided model may be associated with one or more data sets that may be independently used or jointly used by each side, i.e., the UE, the network node, etc. However, the two-sided model is not limited to being associated with data sets and any other resource may be independently or jointly used. Further, the embodiments are not limited to a two-sided model and the features described herein with respect to any model such as the two-sided model may apply to any other model such as an N-sided model.

[0076] In some embodiments, an N-sided model or two-sided model may include part models such as a network node part model, a UE part model, etc. In one or more embodiments, the term “network node part model” or “NW-part model” may refer to a model associated with the network node, e.g., stored in the network node, obtained by the network node, generated by the network node, modified by the network node, structured (or re-structured) by the network node, trained by the network node, activated / deactivated by the network node, etc. In one or more embodiments, the term “UE-part model” may refer to a model associated with the UE, e.g., stored in the UE, obtained by the UE, generated by the UE, modified by the UE, structured (or re-structured) by the UE, trained by the UE, activated / deactivated by the UE, etc.

[0077] In some embodiments, the term “union” is used and may refer to one or more elements that are common to two or more sets. For example, a first set and a second set may have elements A and B in common, i.e., the union of the first set and the second set is A and B. Further, the term “union” of the first dataset and the second dataset may not necessarily include the entire first dataset. The term “union” may include elements from only one set, from two or more sets, or elements that are not part of a set. In addition, sets may have more than one union. For example, a first set and a second set may have a first union and a second union, where the first union may be a subset of an overall union between the two sets, and the second union may be another subset of the overall union. In another example, a first set may have elements A, B, C, D, and E, and a second set may have elements B, C, D, E, and F. The union of the two sets may include the common elements, i.e., B, C, D, E, where a first union may be B and C, and the second union may be D and E. Any other combinations of elements, sets, and subsets are possible.

[0078] Note that although terminology from one particular wireless system, such as, for example, 3GPP LTE and / or New Radio (NR), may be used in this disclosure, this should not be seen as limiting the scope of the disclosure to only the aforementioned system. Other wireless systems, including without limitation Wide Band Code Division Multiple Access (WCDMA), Worldwide Interoperability for Microwave Access (WiMax), Ultra Mobile Broadband (UMB) and Global System for Mobile Communications (GSM), may also benefit from exploiting the ideas covered within this disclosure.

[0079] Note further, that functions described herein as being performed by a user equipment or a network node may be distributed over a plurality of user equipments and / or network nodes. In other words, it is contemplated that the functions of the network node and user equipment described herein are not limited to performance by a single physical device and, in fact, can be distributed among several physical devices.

[0080] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. It will be further understood that terms used herein should be interpreted as having a meaning that is consistent with their meaning in the context of this specification and the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0081] Referring again to the drawing figures, in which like elements are referred to by like reference numerals, there is shown in FIG. 3 a schematic diagram of a communication system 10, according to an embodiment, such as a 3 GPP-type cellular network that may support standards such as LTE and / or NR (5G), which comprises an access network 12, such as a radio access network (RAN) 12, and a core network 14. The access network 12 comprises a plurality of network nodes 16a, 16b, 16c (referred to collectively as network nodes 16), such as NBs, eNBs, gNBs or other types of wireless access points, each defining a corresponding coverage area 18a, 18b, 18c (referred to collectively as coverage areas 18). Each network node 16a, 16b, 16c is connectable to the core network 14 over a wired or wireless connection 20. A first user equipment (UE) 22a located in coverage area 18a is configured to wirelessly connect to, or be paged by, the corresponding network node 16a. A second UE 22b in coverage area 18b is wirelessly connectable to the corresponding network node 16b. While a plurality of UEs 22a, 22b (collectively referred to as user equipments 22) are illustrated in this example, the disclosed embodiments are equally applicable to a situation where a sole UE is in the coverage area or where a sole UE is connecting to the corresponding network node 16. Note that although only two UEs 22 and three network nodes 16 are shown for convenience, the communication system may include many more UEs 22 and network nodes 16.

[0082] Also, it is contemplated that a UE 22 can be in simultaneous communication and / or configured to separately communicate with more than one network node 16 and more than one type of network node 16. For example, a UE 22 can have dual connectivity with a network node 16 that supports LTE and the same or a different network node 16 that supports NR. As an example, UE 22 can be in communication with an eNB for LTE / E-UTRAN and a gNB for NR / NG-RAN.

[0083] A network node 16 is configured to include a node management unit 24 which is configured to perform any step and / or task and / or process and / or method and / or feature described in the present disclosure, e.g., network node functions. A user equipment 22 is configured to include a UE management unit 26 which is configured to perform any step and / or task and / or process and / or method and / or feature described in the present disclosure, e.g., UE functions.

[0084] Further, any of the radio access network 12 (and / or its components such as network nodes 16 and / or UEs 22) and / or core network 14 may be in communication with any other network such as a cloud network. Although not shown, core network 14 may include one or more network nodes 16 (and / or UEs 22).

[0085] Example implementations, in accordance with an embodiment, of the UE 22 and network node 16 discussed in the preceding paragraphs will now be described with reference to FIG..

[0086] The communication system 10 includes a network node 16 provided in a communication system 10. Network node 16 includes hardware 28 enabling it to communicate with the UE 22. The hardware 28 may include a radio interface 30 for setting up and maintaining at least a wireless connection 32 with a UE 22 located in a coverage area 18 served by the network node 16. The radio interface 30 may be formed as or may include, for example, one or more RF transmitters, one or more RF receivers, and / or one or more RF transceivers. The radio interface 30 includes an array of antennas 34 to radiate and receive signal(s) carrying electromagnetic waves. In some embodiments, radio interface 30 may be configured for setting up and maintaining at least a wireless / wired connection with other network nodes 16.

[0087] In the embodiment shown, the hardware 28 of the network node 16 further includes processing circuitry 36. The processing circuitry 36 may include a processor 38 and a memory 40. In particular, in addition to or instead of a processor, such as a central processing unit, and memory, the processing circuitry 36 may comprise integrated circuitry for processing and / or control, e.g., one or more processors and / or processor cores and / or FPGAs (Field Programmable Gate Array) and / or ASICs (Application Specific Integrated Circuitry) adapted to execute instructions. The processor 38 may be configured to access (e.g., write to and / or read from) the memory 40, which may comprise any kind of volatile and / or nonvolatile memory, e.g., cache and / or buffer memory and / or RAM (Random Access Memory) and / or ROM (Read-Only Memory) and / or optical memory and / or EPROM (Erasable Programmable Read-Only Memory).

[0088] Thus, the network node 16 further has software 42 stored internally in, for example, memory 40, or stored in external memory (e.g., database, storage array, network storage device, etc.) accessible by the network node 16 via an external connection. The software 42 may include application 44 which may include software application configured to provide application functions, such as a functions associated with a service provided to UE 22.

[0089] The software 42 may be executable by the processing circuitry 36. The processing circuitry 36 may be configured to control any of the methods and / or processes described herein and / or to cause such methods, and / or processes to be performed, e.g., by network node 16. Processor 38 corresponds to one or more processors 38 for performing network node 16 functions described herein. The memory 40 is configured to store data, programmatic software code and / or other information described herein. In some embodiments, the software 42 may include instructions that, when executed by the processor 38 and / or processing circuitry 36, causes the processor 38 and / or processing circuitry 36 to perform the processes described herein with respect to network node 16. For example, processing circuitry 36 of the network node 16 may include anode management unit 24 which is configured to perform any step and / or task and / or process and / or method and / or feature described in the present disclosure, e.g., network node functions.

[0090] The communication system 10 further includes the UE 22 already referred to. The UE 22 may have hardware 46 that may include a radio interface 48 configured to set up and maintain a wireless connection 32 with a network node 16 serving a coverage area 18 in which the UE 22 is currently located. The radio interface 48 may be formed as or may include, for example, one or more RF transmitters, one or more RF receivers, and / or one or more RF transceivers. The radio interface 48 includes an array of antennas 50 to radiate and receive signal(s) carrying electromagnetic waves.

[0091] The hardware 46 of the UE 22 further includes processing circuitry 52. The processing circuitry 52 may include a processor 54 and memory 56. In particular, in addition to or instead of a processor, such as a central processing unit, and memory, the processing circuitry 52 may comprise integrated circuitry for processing and / or control, e.g., one or more processors and / or processor cores and / or FPGAs (Field Programmable Gate Array) and / or ASICs (Application Specific Integrated Circuitry) adapted to execute instructions. The processor 54 may be configured to access (e.g., write to and / or read from) memory 56, which may comprise any kind of volatile and / or nonvolatile memory, e.g., cache and / or buffer memory and / or RAM (Random Access Memory) and / or ROM (Read-Only Memory) and / or optical memory and / or EPROM (Erasable Programmable Read-Only Memory).

[0092] Thus, the UE 22 may further comprise software 58, which is stored in, for example, memory 56 at the UE 22, or stored in external memory (e.g., database, storage array, network storage device, etc.) accessible by the UE 22. The software 58 may be executable by the processing circuitry 52. The software 58 may include an application 60. The application 60 may be operable to provide a service to a human or non-human user via the UE 22 and / or be configured to provide application client functions, e.g., associated with application 44.

[0093] The processing circuitry 52 may be configured to control any of the methods and / or processes described herein and / or to cause such methods, and / or processes to be performed, e.g., by UE 22. The processor 54 corresponds to one or more processors 54 for performing UE 22 functions described herein. The UE 22 includes memory 56 that is configured to store data, programmatic software code and / or other information described herein. In some embodiments, the software 58 and / or the application 60 may include instructions that, when executed by the processor 54 and / or processing circuitry 52, causes the processor 54 and / or processing circuitry 52 to perform the processes described herein with respect to UE 22. For example, the processing circuitry 52 of the user equipment 22 may include UE management unit 26 which is configured to perform any step and / or task and / or process and / or method and / or feature described in the present disclosure, e.g., UE functions.

[0094] In some embodiments, the inner workings of the network node 16 and UE 22 may be as shown in FIG. 4 and independently, the surrounding network topology may be that of FIG. 3.

[0095] The wireless connection 32 between the UE 22 and the network node 16 is in accordance with the teachings of the embodiments described throughout this disclosure. More precisely, the teachings of some of these embodiments may improve the data rate, latency, and / or power consumption and thereby provide benefits such as reduced user waiting time, relaxed restriction on file size, better responsiveness, extended battery lifetime, etc. In some embodiments, a measurement procedure may be provided for the purpose of monitoring data rate, latency and other factors on which the one or more embodiments improve.

[0096] Although FIGS. 3 and 4 show various “units” such as node management unit 24 and UE management unit 26 as being within a respective processor, it is contemplated that these units may be implemented such that a portion of the unit is stored in a corresponding memory within the processing circuitry. In other words, the units may be implemented in hardware or in a combination of hardware and software within the processing circuitry.

[0097] FIG. 5 is a flowchart of an example process in a user equipment 22 according to some embodiments of the present disclosure. One or more blocks described herein may be performed by one or more elements of user equipment 22 such as by one or more of user equipment 22 such as by one or more of processing circuitry 52 (including the UE management unit 26), processor 54, and / or radio interface 48. User equipment 22 such as via processing circuitry 52 and / or processor 54 and / or radio interface 48 is configured to determine (Block SI 00) a first nominal part of a two-sided model usable to validate a second part model of the two-sided model and validate (Block SI 02) the second part model of the two-sided model using the first nominal part. The first nominal part includes at least one of a nominal network node (NW)-part model and a nominal UE-part model, and the second part model includes at least one of a second NW-part model and a second UE-part model.

[0098] In some embodiments, determining the first nominal part includes defining one or more of an exact structure of the two-sided model including model parameters, weights and biases, nominal acceptable performance levels, and a first data set, by specific channel conditions considered for training the corresponding nominal model.

[0099] In some other embodiments, when the nominal NW-part of the two-sided model is determined, the method further includes training the second UE-part model at least by one or more of: (A) training the second UE-part model based on a of a second dataset and a first data set and using the nominal NW-part model; (B) validating the performance of the second UE-part model using a test data set and the nominal NW-part model; and (C) if the second UE-part model fulfills performance requirements, optionally indicating the availability of the second UE-part model to the network node 16.

[0100] In some embodiments, when the nominal UE-part of the two sided model is determined, the method further includes training the second UE-part model at least by one or more of: (A) training a nominal test NW-part model based on a first data set and using the nominal UE-part model; (B) training the second UE-part model based on a union of a second dataset and the first data set and using the nominal test NW-part model; (C) validating performance of the second UE-part model using a test data set and the nominal test NW-part model; and (D) if the second UE-part model fulfills performance requirements, optionally indicating the availability of the second UE-part model to the network node 16.

[0101] FIG. 6 is a flowchart of an example process in a network node 16. One or more blocks described herein may be performed by one or more elements of network node 16 such as by one or more of processing circuitry 36 (including the node management unit 24), processor 38, and / or radio interface 30. Network node 16 such as via processing circuitry 36 and / or processor 38 and / or radio interface 30 is configured to determine (Block S104) a first nominal part of a two-sided model usable to validate a second part model of the two-sided model and validate (Block SI 06) the second part model of the two- sided model using the first nominal part. The first nominal part includes at least one of a nominal network node (NW)-part model and a nominal UE-part model, and the second part model includes at least one of a second NW-part model and a second UE-part model.

[0102] In some embodiments, determining the first nominal part includes defining one or more of an exact structure of the two-sided model including model parameters, weights and biases, nominal acceptable performance levels, and a first data set, by specific channel conditions considered for training the corresponding nominal model.

[0103] In some other embodiments, when the nominal NW-part of the two-sided model is determined, the method further includes training the second NW-part model at least by one or more of: (A) training a nominal test UE-part model based on a first data set and using the nominal NW-part model; (B) training the second NW-part model based on the union of the first data set and a third dataset, and using the nominal test UE-part model; (C) testing whether the second NW-part model allows the UE to fulfil a nominal acceptable performance level by testing against the nominal test UE-part model and by using the test data set; and (D) if the second NW-part model allows the UE to fulfil a nominal acceptable performance level, optionally testing whether the second NW-part model allows for improved performance as compared to the nominal NW-part model when using the third dataset.

[0104] In some embodiments, when the nominal UE-part of the two sided model is determined, the method further includes training the second NW-part model at least by one or more of: (A) training a nominal test NW-part model based on a first data set and using the nominal UE-part model; (B) training the second NW-part model based on the union of the first data set and a third dataset, and using the nominal UE-part model; (C) testing whether the second NW-part model allows the UE to fulfil a nominal acceptable performance level by testing against the nominal UE-part model and by using the test data set; and (D) if the second NW-part model allows the UE to fulfil a nominal acceptable performance level, optionally testing whether the second NW-part model allows for improved performance as compared to the nominal test NW-part model when using the third dataset.

[0105] FIG. 7 is a flowchart of an example process in a UE 22 according to some embodiments of the present disclosure. UE 22 is configured to train and validate performance of one or more two-sided models and to communicate with a NW 16. One or more blocks described herein may be performed by one or more elements of user equipment 22 such as by one or more of processing circuitry 52 (including the UE management unit 26), processor 54, and / or radio interface 48. User equipment 22 such as via processing circuitry 52 and / or processor 54 and / or radio interface 48 is configured to train (Block SI 08) a second UE-part model of one or more two-sided models based on a first union of a first data set and a second data set, and a first assumption by the UE 22 that the NW 16 uses a first NW-part model. The first data set is usable to train a first UE-part model of the one or more two-sided models. The UE 22 is also configured to validate (Block SI 10) performance of the second UE-part model based on a test data set and the first NW-part model and perform (Block SI 12) one or more actions based on the validation of the performance of the second UE-part model.

[0106] In some embodiments, the method further includes training the first UE-part model using the first data set and based on the first assumption.

[0107] In some other embodiments, one or both of: the first UE-part model is a nominal UE-part model or a reference UE-part model; and the first NW-part model is a nominal NW-part model or a reference NW-part model.

[0108] In some embodiments, validating the performance of the second UE-part model includes determining whether the second UE-part model fulfills one or more performance requirements, and the one or more actions include transmitting, to the NW 16, a first indication indicating an availability of the second UE-part model at the UE 22.

[0109] In some other embodiments, the method further includes receiving, from the NW 16, a second indication indicating the UE to one or more of: (A) activate use the second UE-part model by the UE 22; (B) deactivate use of the second UE-part model by the UE 22; and (C) fall back to the first UE-part model.

[0110] In some embodiments, the one or more actions include determining a model inference output using the second UE-part model and transmitting, to the NW 16, a third indication indicating the model inference output.

[0111] In some other embodiments, the model inference output includes an artificial intelligence based channel state information report.

[0112] In some embodiments, the method further includes: (A) training a nominal test NW-part model using the first data set and the first UE-part model; (B) training the second UE-part model of one or more two-sided models based on a second union of the first data set and the second data set, and the nominal test NW-part model; and (C) validating performance of the second UE-part model using the test data and the nominal test NW- part model.

[0113] In some embodiments, the method further includes receiving, from the NW, the second data set including information usable to reconstruct the second UE-part model and reconstructing the second UE-part model based on information included in the second data set.

[0114] In some other embodiments, the method further includes receiving, from the NW 16, a joint data set usable by the UE 22 to train the second UE-part model and by the NW to train a second-NW part model.

[0115] FIG. 8 is a flowchart of an example process in a network node 16. The network node (NW) 16 is configured to train and validate performance of one or more two-sided models and to communicate with a UE 22. One or more blocks described herein may be performed by one or more elements of network node 16 such as by one or more of processing circuitry 36 (including the node management unit 24), processor 38, and / or radio interface 30. Network node 16 such as via processing circuitry 36 and / or processor 38 and / or radio interface 30 is configured to train (Block SI 14) a second NW-part model of one or more two-sided models based on a third union of a first data set and a third data set, and a first NW-part model. The first data set and the first NW-part model are usable to train a nominal test UE-part model. The NW 16 is also configured to validate (Block SI 16) performance of the second NW-part model based on a test data set and the nominal test UE-part model and perform (Block SI 18) one or more actions based on the validation of the performance of the second NW-part model.

[0116] In some embodiments, the method further includes training the nominal test UE- part model using the first data set and the first NW-part model.

[0117] In some other embodiments, the first NW-part model is a nominal NW-part model or a reference NW-part model.

[0118] In some embodiments, the method further includes receiving, from the UE 22, a first indication indicating an availability of a second UE-part model at the UE 22.

[0119] In some other embodiments, validating the performance of the second NW-part model includes determining whether the second NW-part model fulfills one or more performance requirements, and the one or more actions include transmitting, to the UE 22, a second indication indicating the UE 22 to one or more of: (A) activate use the second UE-part model by the UE 22; (B) deactivate use of the second UE-part model by the UE 22; and (C) fall back to the first UE-part model. In some embodiments, the method further includes receiving, from the UE 22, a third indication indicating a model inference output, the model inference output being based on a second UE-part model.

[0120] In some other embodiments, the model inference output includes an artificial intelligence based channel state information report.

[0121] In some embodiments, the method further includes transmitting, to the UE 22, a second data set including information usable by the UE 22 to reconstruct the second UE- part model.

[0122] In some other embodiments, the method further includes transmitting, to the UE 22, a joint data set usable by the UE 22 to train the second UE-part model and by the NW 16 to train a second-NW part model.

[0123] In some embodiments, the method further includes determining whether to fall back to using the first NW-part model based on the validation.

[0124] Having described the general process flow of arrangements of the disclosure and having provided examples of hardware and software arrangements for implementing the processes and functions of the disclosure, the sections below provide details and examples of arrangements for separate training for two-sided models using nominal models. In some embodiments, the term NW is used and may refer to network node 16 (and / or any other components of system 10 associated with network node 16). In some other embodiments, the term NW side / part refers to the network node side / part.

[0125] Further, the term “network” or “NW” may refer to a network node such as a gNB, base station, unit within the base station, relay node, core network node, a core network node, or a device supporting D2D communication. The network node may be deployed in a 5G network, or a 6G network. Moreover, although the term AI / ML model uses a single form, it should be well understood that it should not prevent the implementation of more than one AI / ML model. The UE 22 may either be configured or autonomously switch between the models depending on certain conditions and / or proprietary implementations.

[0126] In some embodiments, the term “two-sided CSI compression” may be used as an example of a feature that has two Al models - one on the UE side (also called as UE-part model) and another on the network node side (also called as NW-part model). The UE-part Al model may communicate directly with the NW-part Al model, which may refer to the models being jointly or sequentially trained in some manner. The embodiment may also be applicable for other AI / ML features that require / benefit from two-sided models that has training dependencies. Further, the term “two-sided models” may refer to models (e.g., Al models) that are associated with one or more sides of the network such as a UE and a network node. For example, two-sided models may refer to models that may be generated, modified, trained, determined, activated, deactivated by different network entities.

[0127] The UE 22 (and the network node 16 for two-sided model cases) is assumed to have the capability of running AI / ML models supporting the AI / ML-enabled features and its respected data collection procedures. In addition, the term “training” may represent training, retraining, fine-tuning, or any other type of training. The term may also include preprocessing of the dataset before it serves as the input of the AI / ML model, e.g., converting the channel to eigenvector, selecting beams and taps, etc.

[0128] In addition, the term dataset may refer to a set of data used to train a UE-part and / or NW-part model. However, one dataset may be used to train more than one UE-part model and / or NW-part model. For example, from the same dataset, the UE 22 and the network node 16 may train different models that handle different UE-part input sizes, puncturing rates, latent space quantization sizes, etc.

[0129] In an embodiment, a method for defining a first nominal part of the two-sided model is described. The two-sided model may be used to validate the actual second-part model of the two-sided model case that may be developed / optimized without exposing the actual first part model implemented at the network node 16 or UE 22.

[0130] In some embodiments, the method for defining a nominal part or nominal-part model may include the network node 16 and / or the UE 22 being configured to one or more of:

[0131] • Define the exact structure of the model including model parameters, weights and biases.

[0132] • Define the nominal acceptable performance level(s).

[0133] • Define a first data set by specific channel conditions considered for training the nominal model.

[0134] For the case in which a nominal a NW part or a nominal NW-part model of the two-sided model is defined, the method for training the second UE-part model may include one or more of:

[0135] • UE 22 trains a second UE-part model, based on a union of the second dataset and the first data set and using the nominal NW-part model.

[0136] • UE 22 validates the performance of the second UE-part model using a test data set and the nominal NW-part model. • If the second UE-part model fulfills the performance requirements, the UE 22 indicates the availability of the second UE-part model to the network node 16.

[0137] For the case for which a nominal NW part of the two-sided model is defined, the method for training the second NW-part model may include one or more of:

[0138] • Network node 16 trains a nominal test UE-part model, based on the first data set and using the nominal NW part.

[0139] • Network node 16 trains a second NW-part model, based on the union of the first data set and third dataset and using the nominal test UE-part model.

[0140] • Network node 16 tests whether the second NW-part model still allows the UE 22 to fulfil nominal acceptable performance level by testing against the nominal test UE-part model and by using the test data set.

[0141] • If the above condition is fulfilled, the network node 16 can test whether the second NW-part model allows for improved performance as compared to the nominal NW-part model when using the third dataset.

[0142] For the case for which a nominal UE part of the two-sided model is defined, the method for training the second UE part model may include one or more of:

[0143] • UE 22 trains a nominal test NW-part model, based on the first data set and using the nominal UE part model.

[0144] • UE 22 trains a second UE-part model, based on a union of the second dataset and the first data set and using the nominal test NW-part model.

[0145] • The performance of the second UE-part model is validated using a test data set and the nominal test NW-part model.

[0146] • If the second UE-part model fulfills the performance requirements, the UE 22 may indicate the availability of the second UE-part model to the network node 16.

[0147] For the case of defining nominal UE part of the two-sided model, the method for training the second NW-part model covers:

[0148] • Network node 16 trains a nominal test NW-part model, based on the first data set and using the nominal UE part model.

[0149] • Network node 16 trains a second NW part model, based on the union of the first data set and third dataset and using the nominal UE model.

[0150] • Network node 16 tests if the second NW-part model still allows the UE to fulfil nominal acceptable performance level by testing against the nominal UE-part model and by using the test data set. • If the above condition is fulfilled, the network node 16 can test whether the second NW-part model allows for improved performance as compared to the nominal test NW-part model when using the third dataset.

[0151] Defining one side of the two-sided model

[0152] A step of this embodiment is to define one part of the two models, either the NW- part model or the UE-part model, which may be referred to as a nominal NW-part model or nominal UE-part model, respectively.

[0153] Interoperability may be achieved at a nominal performance level by defining a nominal UE-part and / or nominal NW-part model trained using a first data set known to both the UE 22 and the network node 16. The first data set is selected such that a nominal acceptable performance level is achieved when one or more nominal models are implemented by UE 22 and / or network node 16.

[0154] Examples of specified one or more nominal acceptable performance level(s) include:

[0155] System level Key Performance Indicators (KPIs), e.g. minimum achieved UE throughput while the feature is activated; and / or

[0156] - Intermediate KPIs, e.g. squared generalized cosine similarity (SGCS) statistics such as 5-percential / mean SGCS, and / or normalized mean square error (NMSE) statistics such as 5-percential / mean NMSE, shall not be lower than a threshold(s); and / or

[0157] - Loss function, e.g., loss value statistics.

[0158] The value from the above performance level(s) may be determined by the expected performance of the model (e.g., for the CSI compression case, the expected performance is obtained by comparing the decoder output and the ground truth). A tolerance factor may be further introduced, e.g., the performance is considered acceptable when a model has a maximum of X% lower performance compared to the expected performance. In another alternative, the acceptable performance may be derived from the performance of the non- AI mechanism. For example, the performance of a model is considered acceptable when the model has Y% higher performance compared to the non-AI mechanism.

[0159] In an embodiment, if the nominal NW-part model is defined, using the knowledge of the nominal NW-part model, the nominal acceptable performance, and the corresponding first data set used to train the NW-part model, the UE 22 may independently develop and train a UE-part model, assuming that the NW-side implements the nominal NW-part model. In another embodiment, if the nominal UE-part model is defined, using the knowledge of the nominal UE-part model, the nominal acceptable performance, and the corresponding first data set used to train the UE-part model, the network node 16 may independently develop and train aNW-part model, assuming that the UE-side implements the nominal UE-part model.

[0160] Following the approach, the UE-part and NW-part models that are independently developed at the UE 22 and network node 16 are interoperable without the specific needs to accommodate for any vendor specific considerations, and when used together, can fulfill a nominal performance level (i.e. performance similar or better than the nominal performance level).

[0161] Accordingly, the UE 22 and the network node 16 can independently develop their models by training on the first data set. The first data set can be defined by specific channel conditions considered for training the nominal model. As one nonlimiting example, the first dataset is defined by a certain specified configuration of a 3GPP channel model TS 38.901. As another example, the first dataset is defined by a certain set of multiple specified configurations of multiple channel models. The first dataset can also be defined by 3GPP using field dataset(s).

[0162] The defined nominal model (nominal UE-part model and / or nominal NW-part model), nominal performance level including performance metric(s), and conditions to obtain the first data set may be corresponding to the ones specified in 3GPP, and by that available to both the network node 16 and UE 22.

[0163] In the following embodiment, a nominal NW-part model is assumed to be defined. Steps can be performed to optimize both the NW-part and the UE-part models. Further, similar steps can be followed if the nominal UE-part model is defined.

[0164] When referring to UE 22 performing a certain action, the procedure may be performed on-device, or by another entity that performs actions related to training and preparing the UE-part model (e.g., external server responsible for training the UE-part model based on measurements collected from one or more UEs). A test data set may be a data set that match the characteristics of the first data set, but not part of the first data set. It is used to validate that model fulfills the nominal performance level requirement.

[0165] Example Method 1 : UE-sided process

[0166] The example method may include one or more of the following:

[0167] • UE 22 may train a first UE-part model that is trained using a first data set and using the defined nominal NW-part model. • UE 22 collects / obtains a second data set and trains a second UE-part model, based on a union of the second data set and the first data set. o The UE 22 may assume that the NW-part model is frozen during its UE-part model training (e.g. network node 16 is using the nominal NW model). The term “frozen” may refer to as the NW-part model not changing or not being changed during the training. o The term “union” of the first dataset and the second dataset may not necessarily include the entire first dataset but may also be part of the first dataset. That is, the term “union” may refer to elements that are common to more than one set or that overlap. However, the term “union” is not limited as such and may include elements from only one set, from two or more sets, or elements that are not part of a set.

[0168] • The performance of the second UE-part model may be validated using a test data set. For example: o The UE-part model that is trained on the union of the first and the second data sets, fulfills the minimal performance level when tested jointly with the nominal NW-part model.

[0169] • If the second UE-part model fulfills the performance requirements, the UE 22 may: o Indicate the availability of the second UE-part model to the network node 16.

[0170] ■ The indication may further include information about which (sub) test set the UE 22 passed a conformance test on, e.g., for the case where the UE 22 has more than one UE-part model to handle different scenarios / configurations. o The UE 22 may also indicate the availability of a model that is trained using the first dataset (i.e. first UE-part model).

[0171] • The network node 16 may configure the UE 22 to use the second UE-part model. If so, the UE 22 uses the second UE-part model to provide its model inference output (e.g., Al based CSI report) to the network node 16.

[0172] • The network node 16 may monitor the performance when the second UE-part model is activated and may instruct the UE to fall back to the first UE-part model (if supported) or deactivate the model / functionality.

[0173] Example Case 7: The second UE-part model may be trained by a UE server (e.g., a network node 16 or UE 22) for a new release of UEs 22, these new release of UEs 22 (deployed with the second UE-part model) may be required to pass the RAN4 lab testing before being released to the market. The trained second UE-part model is deployed at the test UE 22, and its performance is verified by using a test equipment implemented with nominal NW- part model and using the test dataset. If the tested model performance meets the minimum performance requirement (e.g., set by RAN4), then the conformance testing is passed. When connecting to a network, the new release UE 22 indicates in its UE capability report that it supports this two-sided Al-enabled feature, network node 16 can activate this feature for these UEs 22.

[0174] Example Case 2.

[0175] The second UE-part model may be trained by a UE server for "legacy" UEs 22 that are already released to the market and deployed with the first UE-part model or UEs 22 configured to support a predetermined radio access technology. For instance, the UE side of the network (e.g., operator) want to improve the model performance of its legacy UEs 22 with a new field data set.

[0176] The UE server (network node 16 or UE 22) can test the second UE-part model using the test dataset in addition to a nominal NW-part model, to determine whether the model performance can meet the requirements (e.g., RAN4 defined minimum performance requirements). If so, the UE server may deploy / deliver the second-part model to these “legacy” UEs.

[0177] In some embodiments, at least two options can be considered:

[0178] Option 1: The UE 22 deployed with the second UE-part model indicates the availability of this model to the network node 16. The network node 16 may perform a performance check of this second UE-part model using model performance monitoring mechanisms (or using RAN4 testing methods) before activating this second UE-part model. The network node 16 can also perform model-level LCM, e.g., configure UE 22 to switch between the first UE-part and second UE-part models.

[0179] Option 2: The UE-side delivers the trained second UE-part model to these “legacy” UEs 22 over the top without 3GPP signaling, and the UEs 22 do not indicate the availability of this second UE-part model to the network node 16. In this case, the NW- side may not be aware of which model (the first or the second UE-part model) is used / activated / selected by a “legacy” UE. Hence, the UE-part model update / s witching is transparent to the network node 16. The network node 16 can still perform functionality based LCM, e.g., activate / deactivate this Al-based functionality / feature.

[0180] Example Case 3:

[0181] The second UE-part model may be trained on-device, i.e., the UE 22 itself uses collected measurements to train the second UE-part model or finetune / update its first UE- part model. If the UE 22 also stores the nominal NW-part model, test dataset, etc., then, the performance verification of the retained / updated / finetuned UE-part model can be verified on-device (i.e., by the UE itself). Similar to the Example Case 2, two options can be considered here, model-ID-based LCM or functionality-based LCM.

[0182] Example Method 2: NW sided process

[0183] The example method may include one or more of the following:

[0184] • For testing purposes, network node 16 trains a test UE-part model using the defined nominal NW-part and using a first data set.

[0185] • Network node 16 collects / obtains a third data set and trains a second NW-part model, based on a union of the third and first data sets. o The network node 16 may assume that the UE-part model is frozen during training (e.g. UE 22 is using the test UE-part model). The terms “UE-part model is frozen” may be interpreted as the UE-part model does not change during the training. The training itself, in one alternative, may involve the union of the first and the third data set and UE-part model. In another alternative, the training may involve the union of the first and the third data set and the output of the UE-part model when the UE-part model is given the union of the first and the third data set as inputs.

[0186] • Network node 16 tests or determine whether the second NW-part model still allows the UE 22 to fulfil one or more requirements (e.g., the RAN4 test requirements) by testing against the test UE-part model and by using the test data set. o If the above condition is fulfilled (i.e., the one or more requirements are met), network node 16 may test whether the second NW-part model allows for improved performance as compared to the nominal NW-part model when using the third dataset. o If the above condition is fulfilled (i.e., the one or more requirements are met), network node 16 may activate the second NW-part model for model inference (e.g., the network node 16 runs the second NW-part model to reconstruct the channel measurement from the Al-based CSI report (i.e. output of the UE-part model).

[0187] Further, the example method may include one or more of the following steps:

[0188] Step 1: Network node 16 trains a pair of the nominal NW-part model and nominal test UE-part model using the first dataset. The test dataset may include only ground truth labels associated with the testing conditions. Network node 16 may use the defined nominal NW-part model directly as its initial NW-part model.

[0189] Step 2: Network node 16 configures the UE 22 to perform channel measurements and report the collected measurements to the network node 16. Network node 16 (e.g., a gNB) may collect data samples (e.g., channel measurements) from multiple UEs 22. Further, network node 16 may create a data set (called as the third dataset) using these collected data samples. The collected data samples can also be labels with the associated UE-condition / configuration information. Some of the UE-condition / configuration used as a label may be provided by the UE 22 at the time of reporting the collected data measurements, wherein the UE-condition / configuration represents the UE- condition / configuration at the time of performing the data measurement.

[0190] In some embodiments, some of the UE-condition / configuration to be used as label are instead derived by the network node 16, such as the RRC configuration of the UE 22 at the time of collecting the measurements. For example, network node 16 upon receiving the collected data samples from the UE 22 may label them with the UE-specific RRC configuration the UE 22 applied during the data measurement collection. In one embodiment, the labels used by the network node 16 for the collected data samples may be associated with the UE-condition / configuration provided by the UE 22 and UE- condition / configuration derived by the network node 16.

[0191] Step 3: Network node 16 checks the performance of the nominal NW-part model and the nominal test UE-part model using the third dataset and decides whether to perform NW-part model update / retraining. For example, if the model performance is above the RAN4 defined minimum model performance level (e.g., the mean SGCS / throughput is larger than the RAN4 defined threshold, or the mean SGCS / throughput is larger than the RAN4 defined threshold plus a delta value), then, the network node 16 may keep using the nominal NW-part model without performing any NW-part model update / retraining.

[0192] Step 4: If network node 16 decides to update / retain the UW-part model, then, the NW node trains the second NW-part model using the union of third dataset and first data set. Taking the CSI compression use case as an example, the third dataset may include, e.g., H_target samples (i.e. ground truth measurements), eigenvector, etc., collected from UEs 22. The network node 16 can use this data set to train the second NW-part model, i.e. a retained / updated / improved version of the initial nominal NW side model.

[0193] Step 5: Network node 16 tests if the second NW side model still allows the UE to fulfil the minimum performance requirements by testing against the nominal test UE-part model and by using the test data set. The network node 16 may do a performance check of this second NW-part model using model performance monitoring mechanisms (or using RAN4 testing methods) before activating this second NW-part model.

[0194] If the performance check passes, network node 16 may test whether the second NW-part model allows for improved performance as compared to the nominal NW-part model. Network node 16 may also run the second NW-part model and may continue to monitor the performance of the second-part model and the real UE-part model pair during operation. If the network node 16 observes that the performance is not sufficient, the network node 16 may decide to fall back to using the first NW model (i.e. the nominal NW-part model) or deactivate the model / functionality.

[0195] Example Method 3: independent NW and UE sided process

[0196] This method represents a combination of both Methods 1 and 2 performed independently by the network node 16 and UE 22. A separate data set is used for separate training of the second NW model and the second UE model. In some embodiments, one side of the model is enhanced assuming that the other side model is frozen to be a nominal model (i.e. either the defined nominal NW-part model, or nominal test UE-part model). However, the second NW-part model can be trained using a combination of both the first and third data set. As compared to example “Method 2,” the third dataset may include both ground-truth samples (e.g., H_target samples) and the latent space variables (e.g., associated AI-CSI), collected from UEs 22. In this case, network node 16 can use these {AI-CSI, H target} samples to refine / improve its nominal test UE-part model and then retain / update / improve version of the initial nominal NW-part model with sufficient knowledge of the actual UE side model.

[0197] The second UE-part model can be trained using a combination of both the first and second data set. The second data set collected / obtained by the UE 22 may include information that allows the UE 22 to train a second UE model with sufficient knowledge of the actual NW sided model. For example, the UE 22 can be configured to report the channel measurement to the network node 16, after which, the network node 16 delivers the second data set to the UE 22, which contains both H_target samples and the associated AI-CSI (latent space variables) that can be used by the UE to reconstruct the UE sided model, i.e. the second UE model.

[0198] Network node 16 may test if the second NW-part model still allows the UE-part model used at UEs to fulfil the minimum performance requirements by testing against nominal test UE-part model and by using the test data set.

[0199] Also, the performance of the second UE-part model may be validated by the UE- side using the test data set and the defined nominal NW-part model.

[0200] The network node 16 may continue to monitor the performance of the second UE / NW-part model (s) during operation. If the network node 16 observes that the performance is not sufficient, network node 16 may decide to fall back to using the first NW-part model and / or first UE-part model or deactivate the model / functionality.

[0201] Example Method 4: joint data set and independent NW and UE sided process This method represents a combination of both Methods 1 and 2 performed independently by the network node 16 and UE 22. However, a joint data set may be used for separate training of the second NW model and the second UE model. In some embodiments, the joint data set may be a set that is different from the first data set and / or the second data set and / or third data set.

[0202] Example steps to obtain the joint data set may include one or more of the following:

[0203] - UE 22 performs measurements and reports them back to the network node 16. The UE 22 may also be configured implicitly and / or explicitly with indications of when those measurements are to be used for training / retraining / updating the second NW-part model. For example, the indication is received by the UE 22 as part of a configuration for performing radio measurements. In another embodiment, the UE 22 receives the indication as part of a configuration for data collection for UE-side model training.

[0204] - Network node 16 creates the data set to be used by both the network node 16 and UE 22 to train their second models separately. The data set may be transmitted from the network node 16 to the UE 22. The dataset may include ground-truth samples only or ground-truth samples (e.g., H target samples) and the latent space variables or other variables. - UE 22 creates the data set to be used by both the network node 16 and UE 22 to train their second models separately. The data set may be transmitted from the UE 22 to the network node 16. The dataset may include ground-truth samples only or ground-truth samples (e.g., H target samples) and the latent space variables or other variables.

[0205] The following is a nonlimiting list of example embodiments:

[0206] Embodiment Al. A method in a user equipment (UE) configured to communicate with a network node, the method comprising: determining a first nominal part of a two-sided model usable to validate a second part model of the two-sided model, the first nominal part including at least one of a nominal network node (NW)-part model and a nominal UE-part model, the second part model including at least one of a second NW-part model and a second UE-part model; and validating the second part model of the two-sided model using the first nominal part.

[0207] Embodiment A2. The method of Embodiment Al, wherein determining the first nominal part includes defining one or more of: an exact structure of the two-sided model including model parameters, weights and biases; nominal acceptable performance levels; and a first data set, by specific channel conditions considered for training the corresponding nominal model.

[0208] Embodiment A3. The method of any one of Embodiments Al and A2, wherein when the nominal NW-part of the two-sided model is determined, the method further includes training the second UE-part model at least by one or more of: training the second UE-part model based on a union of a second dataset and a first data set and using the nominal NW-part model; validating the performance of the second UE-part model using a test data set and the nominal NW-part model; and if the second UE-part model fulfills performance requirements, optionally indicating the availability of the second UE-part model to the network node.

[0209] Embodiment A4. The method of any one of Embodiments Al -A3, wherein when the nominal UE-part of the two sided model is determined, the method further includes training the second UE-part model at least by one or more of: training a nominal test NW-part model based on a first data set and using the nominal UE-part model; training the second UE-part model based on a union of a second dataset and the first data set and using the nominal test NW-part model; validating performance of the second UE-part model using a test data set and the nominal test NW-part model; and if the second UE-part model fulfills performance requirements, optionally indicating the availability of the second UE-part model to the network node.

[0210] Embodiment Bl . A user equipment (UE) configured to communicate with a network node, the UE being configured to, and / or comprising a radio interface and / or processing circuitry configured to: determine a first nominal part of a two-sided model usable to validate a second part model of the two-sided model, the first nominal part including at least one of a nominal network node (NW)-part model and a nominal UE-part model, the second part model including at least one of a second NW-part model and a second UE-part model; and validate the second part model of the two-sided model using the first nominal part.

[0211] Embodiment B2. The UE of Embodiment B 1 , wherein determining the first nominal part includes defining one or more of: an exact structure of the two-sided model including model parameters, weights and biases; nominal acceptable performance levels; and a first data set, by specific channel conditions considered for training the corresponding nominal model.

[0212] Embodiment B3. The UE of any one of Embodiments Bl and B2, wherein when the nominal NW-part of the two-sided model is determined, the UE and / or the radio interface and / or the processing circuitry is further configured to train the second UE-part model at least by one or more of: training the second UE-part model based on a union of a second dataset and a first data set and using the nominal NW-part model; validating the performance of the second UE-part model using a test data set and the nominal NW-part model; and if the second UE-part model fulfills performance requirements, optionally indicating the availability of the second UE-part model to the network node. Embodiment B4. The UE of any one of Embodiments B1-B3, wherein when the nominal UE-part of the two sided model is determined, the UE is further configured to train the second UE-part model at least by one or more of: training a nominal test NW-part model based on a first data set and using the nominal UE-part model; training the second UE-part model based on a union of a second dataset and the first data set and using the nominal test NW-part model; validating performance of the second UE-part model using a test data set and the nominal test NW-part model; and if the second UE-part model fulfills performance requirements, optionally indicating the availability of the second UE-part model to the network node.

[0213] Embodiment Cl . A method in a network node configured to communicate with a user equipment (UE), the method comprising: determining a first nominal part of a two-sided model usable to validate a second part model of the two-sided model, the first nominal part including at least one of a nominal network node (NW)-part model and a nominal UE-part model, the second part model including at least one of a second NW-part model and a second UE-part model; and validating the second part model of the two-sided model using the first nominal part.

[0214] Embodiment C2. The method of Embodiment Cl, wherein determining the first nominal part includes defining one or more of: an exact structure of the two-sided model including model parameters, weights and biases; nominal acceptable performance levels; and a first data set, by specific channel conditions considered for training the corresponding nominal model.

[0215] Embodiment C3. The method of any one of Embodiments Cl and C2, wherein when the nominal NW-part of the two-sided model is determined, the method further includes training the second NW-part model at least by one or more of: training a nominal test UE-part model based on a first data set and using the nominal NW-part model; training the second NW-part model based on the union of the first data set and a third dataset, and using the nominal test UE-part model; testing whether the second NW-part model allows the UE to fulfil a nominal acceptable performance level by testing against the nominal test UE-part model and by using the test data set; and if the second NW-part model allows the UE to fulfil a nominal acceptable performance level, optionally testing whether the second NW-part model allows for improved performance as compared to the nominal NW-part model when using the third dataset.

[0216] Embodiment C4. The method of any one of Embodiments C1-C3, wherein when the nominal UE-part of the two sided model is determined, the method further includes training the second NW-part model at least by one or more of: training a nominal test NW-part model based on a first data set and using the nominal UE-part model; training the second NW-part model based on the union of the first data set and a third dataset, and using the nominal UE-part model; testing whether the second NW-part model allows the UE to fulfil a nominal acceptable performance level by testing against the nominal UE-part model and by using the test data set; and if the second NW-part model allows the UE to fulfil a nominal acceptable performance level, optionally testing whether the second NW-part model allows for improved performance as compared to the nominal test NW-part model when using the third dataset.

[0217] Embodiment DI . A network node configured to communicate with a user equipment (UE), the network node being configured to, and / or comprising a radio interface and / or processing circuitry configured to: determine a first nominal part of a two-sided model usable to validate a second part model of the two-sided model, the first nominal part including at least one of a nominal network node (NW)-part model and a nominal UE-part model, the second part model including at least one of a second NW-part model and a second UE-part model; and validate the second part model of the two-sided model using the first nominal part.

[0218] Embodiment D2. The network node of Embodiment D 1 , wherein determining the first nominal part includes defining one or more of: an exact structure of the two-sided model including model parameters, weights and biases; nominal acceptable performance levels; and a first data set, by specific channel conditions considered for training the corresponding nominal model.

[0219] Embodiment D3. The network node of any one of Embodiments DI and D2, wherein when the nominal NW-part of the two-sided model is determined, the network node and / or the radio interface and / or the processing circuitry being further configured to train the second NW-part model at least by one or more of: training a nominal test UE-part model based on a first data set and using the nominal NW-part model; training the second NW-part model based on the union of the first data set and a third dataset, and using the nominal test UE-part model; testing whether the second NW-part model allows the UE to fulfil a nominal acceptable performance level by testing against the nominal test UE-part model and by using the test data set; and if the second NW-part model allows the UE to fulfil a nominal acceptable performance level, optionally testing whether the second NW-part model allows for improved performance as compared to the nominal NW-part model when using the third dataset.

[0220] Embodiment D4. The network node of any one of Embodiments D1-D3, wherein when the nominal UE-part of the two sided model is determined, the network node being further configured to train the second NW-part model at least by one or more of: training a nominal test NW-part model based on a first data set and using the nominal UE-part model; training the second NW-part model based on the union of the first data set and a third dataset, and using the nominal UE-part model; testing whether the second NW-part model allows the UE to fulfil a nominal acceptable performance level by testing against the nominal UE-part model and by using the test data set; and if the second NW-part model allows the UE to fulfil a nominal acceptable performance level, optionally testing whether the second NW-part model allows for improved performance as compared to the nominal test NW-part model when using the third dataset.

[0221] As will be appreciated by one of skill in the art, the concepts described herein may be embodied as a method, data processing system, computer program product and / or computer storage media storing an executable computer program. Accordingly, the concepts described herein may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects all generally referred to herein as a “circuit” or “module.” Any process, step, action and / or functionality described herein may be performed by, and / or associated to, a corresponding module, which may be implemented in software and / or firmware and / or hardware. Furthermore, the disclosure may take the form of a computer program product on a tangible computer usable storage medium having computer program code embodied in the medium that can be executed by a computer. Any suitable tangible computer readable medium may be utilized including hard disks, CD-ROMs, electronic storage devices, optical storage devices, or magnetic storage devices.

[0222] Some embodiments are described herein with reference to flowchart illustrations and / or block diagrams of methods, systems and computer program products. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer (to thereby create a special 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 and / or block diagram block or blocks.

[0223] These computer program instructions may also be stored in a computer readable memory or storage medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instruction means which implement the function / act specified in the flowchart and / or block diagram block or blocks.

[0224] The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. It is to be understood that the functions / acts noted in the blocks may occur out of the order noted in the operational illustrations. 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 / acts involved. Although some of the diagrams include arrows on communication paths to show a primary direction of communication, it is to be understood that communication may occur in the opposite direction to the depicted arrows.

[0225] Computer program code for carrying out operations of the concepts described herein may be written in an object oriented programming language such as Python, Java® or C++. However, the computer program code for carrying out operations of the disclosure may also be written in conventional procedural programming languages, such as the "C" programming language. The program 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. In the latter scenario, the remote computer may be connected to the user's computer through a local area network (LAN) 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).

[0226] Many different embodiments have been disclosed herein, in connection with the above description and the drawings. It will be understood that it would be unduly repetitious and obfuscating to literally describe and illustrate every combination and subcombination of these embodiments. Accordingly, all embodiments can be combined in any way and / or combination, and the present specification, including the drawings, shall be construed to constitute a complete written description of all combinations and subcombinations of the embodiments described herein, and of the manner and process of making and using them, and shall support claims to any such combination or subcombination.

[0227] It will be appreciated by persons skilled in the art that the embodiments described herein are not limited to what has been particularly shown and described herein above. In addition, unless mention was made above to the contrary, it should be noted that all of the accompanying drawings are not to scale. A variety of modifications and variations are possible in light of the above teachings and following claims.

Claims

What is claimed is:

1. A method in a user equipment, UE, (22) configured to train and validate performance of one or more two-sided models, the UE (22) being configured to communicate with a network node, NW, (16) the method comprising: training (S108) a second UE-part model of one or more two-sided models based on a first union of a first data set and a second data set, and a first assumption by the UE (22) that the NW (16) uses a first NW-part model, the first data set being usable to train a first UE-part model of the one or more two-sided models; validating (SI 10) performance of the second UE-part model based on a test data set and the first NW-part model; and performing (SI 12) one or more actions based on the validation of the performance of the second UE-part model.

2. The method of Claim 1, wherein the method further includes: training the first UE-part model using the first data set and based on the first assumption.

3. The method of any one of Claims 1 and 2, wherein one or both of: the first UE-part model is a nominal UE-part model or a reference UE-part model; and the first NW-part model is a nominal NW-part model or a reference NW-part model.

4. The method of any one of Claims 1-3, wherein validating the performance of the second UE-part model includes determining whether the second UE-part model fulfills one or more performance requirements, and the one or more actions include: transmitting, to the NW (16), a first indication indicating an availability of the second UE-part model at the UE (22).

5. The method of any one of Claims 1-4, wherein the method further includes: receiving, from the NW (16), a second indication indicating the UE (22) to one or more of: activate use the second UE-part model by the UE (22); deactivate use of the second UE-part model by the UE (22); andfall back to the first UE-part model.

6. The method of any one of Claims 1-5, wherein the one or more actions include: determining a model inference output using the second UE-part model; and transmitting, to the NW (16), a third indication indicating the model inference output.

7. The method of Claim 6, wherein the model inference output includes an artificial intelligence based channel state information report.

8. The method of any one of Claims 1-7, wherein the method further includes: training a nominal test NW-part model using the first data set and the first UE-part model; training the second UE-part model of one or more two-sided models based on a second union of the first data set and the second data set, and the nominal test NW-part model; and validating performance of the second UE-part model using the test data and the nominal test NW-part model.

9. The method of any one of Claims 1-8, wherein the method further includes: receiving, from the NW (16), the second data set including information usable to reconstruct the second UE-part model; and reconstructing the second UE-part model based on information included in the second data set.

10. The method of any one of Claims 1-8, wherein the method further includes: receiving, from the NW (16), a joint data set usable by the UE (22) to train the second UE-part model and by the NW (16) to train a second-NW part model.

11. A user equipment, UE, (22) configured to train and validate performance of one or more two-sided models, the UE (22) being configured to communicate with a network node, NW, (16) and to perform any one of the steps of any one of Claims 1-10.

12. A method in a network node, NW, (16) configured to train and validate performance of one or more two-sided models, the NW (16) being configured to communicate with a user equipment, UE, (22) the method comprising: training (S 114) a second NW-part model of one or more two-sided models based on a third union of a first data set and a third data set, and a first NW-part model, the first data set and the first NW-part model being usable to train a nominal test UE-part model; validating (SI 16) performance of the second NW-part model based on a test data set and the nominal test UE-part model; and performing (S 118) one or more actions based on the validation of the performance of the second NW-part model.

13. The method of Claim 12, wherein the method further includes: training the nominal test UE-part model using the first data set and the first NW- part model.

14. The method of any one of Claims 12 and 13, wherein the first NW-part model is a nominal NW-part model or a reference NW-part model.

15. The method of any one of Claims 12-14, wherein the method further includes: receiving, from the UE (22), a first indication indicating an availability of a second UE-part model at the UE (22).

16. The method of any one of Claims 12-15, wherein validating the performance of the second NW-part model includes determining whether the second NW- part model fulfills one or more performance requirements, and the one or more actions include: transmitting, to the UE (22), a second indication indicating the UE (22) to one or more of: activate use the second UE-part model by the UE (22); deactivate use of the second UE-part model by the UE (22); and fall back to the first UE-part model.

17. The method of any one of Claims 12-16, wherein the method further includes: receiving, from the UE (22), a third indication indicating a model inference output, the model inference output being based on a second UE-part model.

18. The method of Claim 17, wherein the model inference output includes an artificial intelligence based channel state information report.

19. The method of any one of Claims 17 and 18, wherein the method further includes: transmitting, to the UE (22), a second data set including information usable by the UE (22) to reconstruct the second UE-part model.

20. The method of any one of Claims 17-19, wherein the method further includes: transmitting, to the UE (22), a joint data set usable by the UE (22) to train the second UE-part model and by the NW (16) to train a second-NW part model.

21. The method of any one of Claims 12-20, wherein the method further includes: determining whether to fall back to using the first NW-part model based on the validation.

22. A network node, NW, (16) configured to train and validate performance of one or more two-sided models, the NW (16) being configured to communicate with a user equipment, UE, (22) and to perform any one of the steps of any one of Claims 12-21.

Images