Apparatuses, methods, and computer programs for a network node and for user equipment for ai / ML model inference reporting description
By using delta inference reporting based on reference results, the overhead and power consumption issues in AI/ML model inference are mitigated, allowing efficient lifecycle management with reduced signaling and power usage.
Patent Information
- Application Number
- PCT/EP2026/063931
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-16
- Filing Date
- 2026-05-13
- Publication Date
- 2026-11-19
Smart Images

Figure EP2026063931_19112026_PF_FP_ABST
Abstract
Description
[0001] 202502837 1
[0002] Apparatuses, Methods, and Computer Programs for a Network Node and for User Equipment for AI / ML Model Inference Reporting
[0003] Description
[0004] The present disclosure relates to apparatuses, methods, and computer programs for a network node and for user equipment for artificial intelligence / machine learning (AI / ML) reporting in a mobile communication system, and more particularly, but not exclusively, to a concept for efficiently managing information for inference reporting by reporting inference deviation in relation to reference inference results.
[0005] The Third Generation Partnership Project (3GPP) has been studying the integration of artificial intelligence (Al) and machine learning (ML) into mobile communication systems, in particular into Fifth Generation New Radio (5G NR) and future Sixth Generation (6G) systems.
[0006] 3GPP Technical Report (TR) 38.843 V19.0.0 (2025-09) Technical Report, 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Study on Artificial Intelligence (AI)ZMachine Learning (ML) for NR air interface (Release 19), describes a framework for AI / ML model lifecycle management in the context of the NR air interface, covering aspects such as model training, model deployment, model inference, model monitoring, and model updating. AI / ML model inference refers to the process of using a trained AI / ML model to produce a set of outputs based on a set of inputs. Use cases studied in 3GPP include AI / ML-based beam management, AI / ML-based positioning, and AI / ML-based channel state information (CSI) compression. In these use cases, a user equipment (UE) may perform AI / ML model inference and report the inference results to the network. The inference results may include, for example, predicted beam indices, predicted Layer 1 reference signal received power (L1-RSRP) values, timing estimations, line-of-sight / non-line-of-sight (LOS / NLOS) indicators, or compressed CSI feedback. The network may use the reported inference results to configure or optimize radio resource management (RRM) parameters, positioning procedures, or beam selection.
[0007] Internal202502837 2
[0008] AI / ML model inference reporting serves a central role in the AI / ML model lifecycle management framework described in 3GPP TR 38.843. The network relies on inference reports received from the UE to monitor the performance of deployed AI / ML models and to determine whether corrective actions are needed. Such corrective actions may include model activation, model deactivation, model switching, or model fallback to a legacy non-AI / ML-based procedure. For example, if the inference results reported by the UE indicate that a deployed beam management model no longer predicts beam indices with sufficient accuracy, the network may trigger a model switch to an alternative AI / ML model or initiate a fallback to conventional beam sweeping procedures. Similarly, in AI / ML-based positioning, the network may use inference reports to assess whether the positioning accuracy achieved by the UE-side model meets the requirements of the requested quality of service (QoS) level, and if not, may reconfigure the positioning method or request additional measurements. Inference reporting thus provides the network with the observability necessary to maintain AI / ML model performance over time and across varying radio conditions.
[0009] Beyond model monitoring, inference reporting also supports data collection for model training and model updating. The network may aggregate inference reports received from one or more UEs to build or refine training data sets used for retraining AI / ML models, either at the network side or for distribution back to UEs as updated model parameters. In addition, inference reports may be used for performance evaluation during a model testing or validation phase prior to full deployment. The granularity and frequency of inference reporting therefore directly affect the network's ability to manage the AI / ML model lifecycle efficiently. However, as each inference report consumes uplink signaling resources and contributes to UE power consumption, a trade-off exists between the richness of the reported information and the associated overhead.
[0010] Achieving this trade-off is particularly important in scenarios involving multiple simultaneously active AI / ML functionalities at the UE, where each functionality may independently generate inference results requiring reporting to the network.
[0011] Further details can be found in
[0012] US11310687B2;
[0013] US20230413152A1;
[0014] US20220116838A1 ; and
[0015] Internal202502837 3
[0016] WO2021259492A1.
[0017] The present disclosure is based on the finding that when the UE performs AI / ML model inference and reports the inference results to the network, the resulting signaling constitutes a full inference report. Transmitting a full inference report to the network causes signaling overhead on the uplink control information (UCI) or other reporting channels and increases the UE's power consumption. This overhead becomes particularly significant when inference reporting is performed frequently, for example in scenarios with high mobility or dense network deployments where rapid model monitoring and adaptation are required. The signaling overhead associated with full inference reporting thus presents a challenge for efficient AI / ML integration in mobile communication systems, in particular for power-constrained UEs such as battery-operated devices or UEs operating in a discontinuous reception (DRX) mode. There is therefore a need for a concept that reduces the overhead of AI / ML model inference reporting while still enabling the network to obtain sufficient information about the UE's inference results for effective AI / ML model lifecycle management.
[0018] The present disclosure is based on the finding that when a UE performs AI / ML model inference in a mobile communication system, the UE conventionally transmits a full inference report comprising the complete inference result to the network, and that such full inference reporting causes signaling overhead and increases UE power consumption. The present disclosure is further based on the finding that the signaling overhead and UE power consumption associated with AI / ML model inference reporting can be reduced by configuring the UE to perform inference reporting relative to a reference inference result, such that the UE transmits a delta inference report indicating only a deviation from the reference inference result instead of the complete inference result.
[0019] According to a first aspect, a method for a network node of a mobile communication system is provided. The method comprises configuring a UE to perform AI / ML model inference reporting relative to a reference inference result. The method further comprises receiving, from the UE, a delta inference report indicating a deviation of an inference result obtained by the UE from the reference inference result, wherein the delta inference report has a reduced size compared to a full inference report comprising
[0020] Internal202502837 4
[0021] the complete inference result. By configuring the UE to report inference results relative to a reference inference result and by receiving a delta inference report instead of a full inference report, the network node can obtain the information necessary for AI / ML model lifecycle management while reducing the uplink signaling overhead and the UE power consumption associated with inference reporting.
[0022] In one example, configuring the UE comprises transmitting, to the UE, a mapping associating one or more reference inference results with respective model identifiers, functionality identifiers, and / or data set identifiers. By providing a structured mapping that associates reference inference results with model identifiers, functionality identifiers, and / or data set identifiers, the network node enables the UE to unambiguously identify the correct reference inference result for each AI / ML model and functionality, thereby ensuring that the delta inference report accurately reflects the deviation from the intended reference and avoiding misalignment between the network node and the UE.
[0023] In another example, the mapping is UE-specific and is transmitted via a Radio Resource Control (RRC) signaling message, and / or the mapping is cell-specific and is transmitted via a system information message. By supporting both UE-specific delivery via RRC signaling and cell-specific delivery via system information, the network node can flexibly choose between individually tailored mappings for specific UEs and common mappings shared across all UEs in a cell, thereby balancing configuration granularity against signaling efficiency depending on the deployment scenario.
[0024] In a further example, the delta inference report is received via at least one of Layer 1 (L1) signaling, Layer 2 (L2) signaling, or Layer 3 (L3) signaling. By supporting reception of the delta inference report via L1, L2, or L3 signaling, the network node can receive delta inference reports at different latency and reliability levels depending on the urgency and the use case, for example using L1 signaling for low-latency beam management scenarios and L3 signaling for less time-critical model monitoring scenarios.
[0025] In yet a further example, the method further comprises configuring or updating one or more AI / ML model parameters for the UE based on the received delta inference report.
[0026] Internal202502837 5
[0027] By configuring or updating AI / ML model parameters based on the received delta inference report, the network node can adapt the AI / ML model deployed at the UE in response to observed deviations from expected inference results, thereby enabling closed-loop AI / ML model lifecycle management with reduced reporting overhead.
[0028] In yet another example, the delta inference report comprises an identifier of the reference inference result used by the UE and a difference value representing the deviation of the obtained inference result from the reference inference result. By including both an identifier of the reference inference result and a difference value in the delta inference report, the network node can unambiguously reconstruct the UE's inference result from the compact delta representation, thereby enabling accurate model performance monitoring without requiring transmission of the full inference result.
[0029] In a still further example, the reference inference result is associated with at least one of a specific AI / ML model identified by a model identifier, a specific AI / ML functionality identified by a functionality identifier, or a specific data set identified by a data set identifier. By associating the reference inference result with at least one of a model identifier, a functionality identifier, or a data set identifier, the network node can maintain and manage reference inference results at different levels of granularity, for example per model, per functionality, or per data set, thereby supporting flexible deployment of delta inference reporting across different AI / ML use cases and model configurations.
[0030] According to a second aspect, a method for a UE in a mobile communication system is provided. The method comprises receiving, from a network node, a configuration for performing AI / ML model inference reporting relative to a reference inference result. The method further comprises performing an AI / ML model inference to obtain an inference result. The method further comprises determining a delta inference report based on a deviation of the obtained inference result from the reference inference result. The method further comprises transmitting the delta inference report to the network node, wherein the delta inference report has a reduced size compared to a full inference report comprising the complete inference result. By receiving a configuration for inference reporting relative to a reference inference result, performing the inference, determining the delta, and transmitting only the delta inference report, the UE reduces its uplink signaling overhead and power consumption compared to transmitting a full
[0031] Internal202502837 6
[0032] inference report, which is particularly beneficial for power-constrained UEs such as battery-operated devices or UEs operating in a discontinuous reception (DRX) mode.
[0033] In one example, receiving the configuration comprises receiving, from the network node, a mapping associating one or more reference inference results with respective model identifiers, functionality identifiers, and data set identifiers. By receiving a structured mapping from the network node, the UE can locally identify the correct reference inference result for each AI / ML model and functionality without requiring additional signaling exchanges at the time of inference, thereby enabling efficient and autonomous determination of the delta inference report.
[0034] In another example, the mapping is UE-specific and is received via an RRC signaling message, and / or the mapping is cell-specific and is received via a system information message. By supporting reception of the mapping via RRC signaling for UE-specific configurations and via system information for cell-specific configurations, the UE can obtain the mapping through established signaling procedures without requiring new signaling channels, thereby minimizing implementation complexity.
[0035] In a further example, the delta inference report is transmitted via at least one of L1 signaling, L2 signaling, or L3 signaling. By supporting transmission of the delta inference report via L1 , L2, or L3 signaling, the UE can select the signaling layer that best matches the latency and reliability requirements of the specific AI / ML use case, thereby enabling efficient integration of delta inference reporting into existing uplink signaling frameworks.
[0036] In yet a further example, determining the delta inference report comprises identifying, from the mapping, the reference inference result corresponding to the AI / ML model used for the inference, and computing the deviation between the obtained inference result and the identified reference inference result. By identifying the reference inference result from the mapping and computing the deviation, the UE performs a structured comparison that yields a compact and unambiguous delta representation, thereby ensuring consistency between the UE's delta inference report and the network node's interpretation thereof.
[0037] Internal202502837 7
[0038] In yet another example, the delta inference report comprises an identifier of the reference inference result used and a difference value representing the deviation of the obtained inference result from the reference inference result. By including both the identifier and the difference value, the UE provides the network node with sufficient information to reconstruct the full inference result from the delta inference report, thereby enabling accurate model monitoring while keeping the report size low or even minimal.
[0039] According to a third aspect, a computer program is provided having a program code for performing the method for the network node according to any of the examples of the first aspect, or for performing the method for the UE according to any of the examples of the second aspect, when the computer program is executed on a computer, a processor, or a programmable hardware component. By providing the concept as a computer program, the delta inference reporting methods can be implemented in software and deployed flexibly on existing hardware platforms in both network nodes and UE devices without requiring dedicated hardware modifications.
[0040] According to a fourth aspect, an apparatus for a network node of a mobile communication system is provided. The apparatus comprises one or more interfaces configured to communicate in the mobile communication system, and one or more processing devices configured to perform the method according to any of the examples of the first aspect. By providing the concept as a network node apparatus with interfaces and processing devices, the delta inference reporting concept can be integrated into existing network infrastructure, enabling deployment of reduced-overhead AI / ML model inference monitoring within established network architectures.
[0041] According to a fifth aspect, an apparatus for a UE of a mobile communication system is provided. The apparatus comprises one or more interfaces configured to communicate in the mobile communication system, and one or more processing devices configured to perform the method according to any of the examples of the second aspect. By providing the concept as a UE apparatus with interfaces and processing devices, the UE can be equipped to receive inference reporting configurations from the network and to generate and transmit delta inference reports, thereby benefiting from reduced uplink signaling overhead and reduced power consumption in a hardware implementation.
[0042] Internal202502837 8
[0043] Some aspects of the present disclosure relate to a mobile communication system comprising a network node with an apparatus for the network node and a UE with an apparatus for the UE, both as described herein.
[0044] Some examples of apparatuses, computer programs and / or methods will be described in the following by way of example only, and with reference to the accompanying figures, in which
[0045] Fig. 1 shows an example of a flowchart of a method for a network node;
[0046] Fig. 2 depicts an example of a flowchart of a method for user equipment;
[0047] Fig. 3 illustrates block diagrams of examples of apparatuses for a network node, user equipment, and a communication network;
[0048] Fig. 4 shows another example of a flowchart of a method for a network node; and
[0049] Fig. 5 depicts an example of a flowchart of a method for a UE.
[0050] Some examples are now described in more detail with reference to the enclosed figures. However, other possible examples are not limited to the features of these embodiments described in detail. Other examples may include modifications of the features, as well as equivalents and alternatives to the features. Furthermore, the terminology used herein to describe certain examples should not be restrictive of further possible examples.
[0051] Throughout the description of the figures, same or similar reference numerals refer to same or similar elements and / or features, which may be identical or implemented in a modified form while providing the same or a similar function. The thickness of lines, layers, and / or areas in the figures may also be exaggerated for the sake of clarification.
[0052] When two elements A and B are combined using an "or", this is to be understood as disclosing all possible combinations, i.e. , only A, only B, as well as A and B, unless
[0053] Internal202502837 9
[0054] expressly defined otherwise in the individual case. As an alternative wording for the same combinations, "at least one of A and B" or "A and / or B" may be used. This applies equivalently to combinations of more than two elements.
[0055] If a singular form, such as "a", "an", and "the" is used and the use of only a single element is not defined as mandatory either explicitly or implicitly, further examples may also use several elements to implement the same function. If a function is described below as implemented using multiple elements, further examples may implement the same function using a single element or a single processing entity. It is further understood that the terms "include", "including", "comprise", and / or "comprising", when used, describe the presence of the specified features, integers, steps, operations, processes, elements, components, and / or a group thereof, but do not exclude the presence or addition of one or more other features, integers, steps, operations, processes, elements, components, and / or a group thereof.
[0056] Fig. 1 shows an example of a flowchart of a method 10 for a network node of a mobile communication system. The method 10 comprises configuring 11 a user equipment (UE) to perform artificial intelligence / machine learning (AI / ML) model inference reporting relative to a reference inference result. Method 10 further comprises receiving 12, from the UE, a delta inference report indicating a deviation of an inference result obtained by the UE from the reference inference result, wherein the delta inference report has a reduced size compared to a full inference report comprising the complete inference result.
[0057] A network node is any entity of the mobile communication system that provides radio communication services to one or more UEs or that controls, manages, or coordinates such services. The network node may be a base station, a gNB (next generation NodeB), an eNodeB, a NodeB, a BTS (Base Transceiver Station), an access point, a transmission / reception point (TRP), or any other network entity capable of communicating with UEs over a radio interface. The network node may belong to a radio access network (RAN), such as a Next Generation RAN (NG-RAN), an evolved UTRAN (E-UTRAN), or a UTRAN, or it may belong to a core network entity involved in AI / ML model lifecycle management (LCM), such as a Location Management Function (LMF) or a Network Data Analytics Function (NWDAF).
[0058] Internal202502837 10
[0059] The mobile communication system may correspond to any 3GPP-standardized mobile communication network, for example a 6th Generation (6G) system, a 5th Generation (5G) system, a New Radio (NR) system, Long-Term Evolution (LTE), an LTE-Advanced (LTE-A) system, or any successor or evolution thereof. The term mobile communication system is used synonymously with mobile communication network. More generally, the network node may be any entity in a wireless communication system that is capable of configuring a UE and receiving reports from the UE.
[0060] Configuring the UE means that the network node transmits, signals, or otherwise provides to the UE one or more configuration parameters, instructions, or indications that cause the UE to adopt a particular reporting behavior for AI / ML model inference results. Configuring may also be referred to as instructing, setting up, establishing, activating, or enabling the UE for the specified reporting mode. The configuration may be transmitted via Radio Resource Control (RRC) signaling, for example in an RRC reconfiguration message, via a Medium Access Control - Control Element (MAC CE), via Downlink Control Information (DCI), via a system information message (e.g., SIB), or via LTE (Long Term Evolution) Positioning Protocol (LPP) signaling, depending on the network architecture and the AI / ML use case. AI / ML model inference refers to the process by which the UE executes, runs, or evaluates a deployed AI / ML model on input data, such as channel measurements, beam measurements, positioning measurements, or channel state information (CSI), to produce an inference result, i.e. , an output of the AI / ML model.
[0061] The AI / ML model may relate to any AI / ML-enabled use case in the mobile communication system, including but not limited to CSI feedback enhancement, beam management, positioning accuracy enhancement, or any other use case studied in the context of 3GPP AI / ML for NR air interface. The reference inference result is a predetermined, preconfigured, or previously communicated inference result that serves as a baseline, anchor, or expected value against which the UE compares its own locally obtained inference result.
[0062] The reference inference result may be associated with a specific dataset, a specific AI / ML model identified by a model identifier (model ID), and / or a specific AI / ML functionality identified by a functionality identifier (functionality ID). The association or
[0063] Internal202502837 11
[0064] mapping between the reference inference result and the corresponding dataset, model ID, and / or functionality ID may be provided by the network node to the UE as part of the configuration, and the mapping may be UE-specific and / or cell-specific. By configuring the UE to report relative to a reference inference result rather than requiring the UE to transmit the complete inference result, the network node enables a reporting mode in which only the difference between the UE's actual inference result and the reference inference result needs to be communicated.
[0065] Receiving the delta inference report means that the network node obtains, acquires, decodes, or extracts from a signaling message or data element transmitted by the UE a report that encodes the difference, deviation, offset, residual, or discrepancy between the inference result that the UE has obtained by executing its AI / ML model and the reference inference result that was previously configured or communicated to the UE. The delta inference report may also be referred to as a differential inference report, a residual inference report, an incremental inference report, or a relative inference report.
[0066] The deviation may be expressed as an arithmetic difference, a vector of differences, a set of quantized residuals, a compressed representation of the difference, or any other encoding that represents the change from the reference inference result to the UE's actual inference result. The UE obtains the inference result by performing model inference locally, i.e. , by executing the deployed AI / ML model on locally available input data such as received reference signals, channel measurements, or other observation data.
[0067] The UE then computes the deviation by comparing its locally obtained inference result against the reference inference result and encodes the deviation into the delta inference report. The delta inference report may be transmitted from the UE to the network node via Layer 1 (L1) signaling (e.g., on PUCCH or PUSCH(Physical Uplink Control or Shared Channel)), Layer 2 (L2) signaling (e.g., via MAC CE), or Layer 3 (L3) signaling (e.g., via RRC or LPP), depending on the latency requirements, the payload size, and the AI / ML use case. The network node may reconstruct the full inference result by combining the received delta inference report with the known reference inference result, for example by adding the reported deviation to the reference inference result, thereby
[0068] Internal202502837 12
[0069] recovering the UE's actual inference result without requiring the UE to transmit the complete inference result.
[0070] The delta inference report having a reduced size compared to a full inference report comprising the complete inference result: The reduced size of the delta inference report means that the number of bits, the payload size, or the signaling overhead required to transmit the delta inference report is smaller than the number of bits, payload size, or signaling overhead that would be required to transmit a full inference report containing the complete, uncompressed inference result of the AI / ML model.
[0071] A full inference report is a report that encodes the entire output of the AI / ML model inference, for example the complete set of predicted beam indices and associated L1-RSRPs, the full CSI feedback matrix, the complete estimated UE position coordinates, or the full set of timing and angle measurements, without reference to any baseline or prior value. The size reduction arises because, in many practical scenarios, the deviation between the UE's actual inference result and the reference inference result is smaller in magnitude, has lower entropy, or is sparser than the absolute inference result itself, and can therefore be encoded using fewer bits. The reduced size may be achieved by any suitable encoding, compression, or quantization technique applied to the deviation, including but not limited to differential encoding, delta coding, entropy coding, run-length encoding, vector quantization of the residual, or truncation of insignificant deviation components. The reduced size of the delta inference report provides at least two technical advantages: first, the signaling overhead on the uplink air interface is reduced, thereby freeing radio resources for other transmissions and improving spectral efficiency; and second, the UE's power consumption for report transmission is reduced, which is particularly beneficial for battery-operated or power-constrained UEs.
[0072] Fig. 2 depicts an example of a flowchart of a method 20 for a UE in a mobile communication system. Method 20 comprises receiving 21, from a network node, a configuration for performing AI / ML model inference reporting relative to a reference inference result. Method 20 further comprises performing 22 an AI / ML model inference to obtain an inference result and determining 23 a delta inference report based on a deviation of the obtained inference result from the reference inference result. Method 20
[0073] Internal202502837 13
[0074] further comprises transmitting 24 the delta inference report to the network node 300. The delta inference report has a reduced size compared to a full inference report comprising the complete inference result.
[0075] Performing an AI / ML model inference to obtain an inference result means that the UE executes a deployed AI / ML model on input data to produce an output, i.e. , the inference result. The AI / ML model may be, for example, a neural network, a decision tree, a support vector machine, or any other machine learning model that has been trained and deployed on the UE for a specific functionality such as beam management, positioning, channel state information (CSI) feedback, or radio link monitoring. The input data may comprise measurements performed by the UE, such as reference signal received power (RSRP) measurements, channel estimates, or sensor data. Performing the inference may also be referred to as executing the model, running the model, evaluating the model, or computing the model output. The inference result is the output produced by the AI / ML model, which may comprise, for example, a predicted beam index, a set of CSI parameters, a positioning estimate, or any other model-specific output quantity. The inference result may also be referred to as a model output, a prediction result, a model prediction, or an inference output.
[0076] Determining a delta inference report based on a deviation of the obtained inference result from the reference inference result means that the UE computes a difference, offset, or residual between its own inference result and the reference inference result previously provided or indicated by the network node, and encodes this difference into the delta inference report. The determining may comprise subtracting the reference inference result from the obtained inference result, computing a relative difference, identifying which components of the inference result differ from the corresponding components of the reference inference result, or applying a differential encoding scheme to represent only the changed or deviating portions of the inference result. The determining may also be referred to as computing, calculating, generating, deriving, or constructing the delta inference report. In some examples, the UE determines the delta inference report only when the deviation exceeds a predefined threshold, and otherwise refrains from reporting, thereby further reducing signaling overhead. In other examples, the UE always determines and transmits the delta inference report, regardless of the magnitude of the deviation. The deviation may be expressed as an absolute difference,
[0077] Internal202502837 14
[0078] a relative difference, a vector difference, a set of changed indices, or any other representation that captures the discrepancy between the obtained inference result and the reference inference result.
[0079] Transmitting the delta inference report to the network node means that the UE sends the determined delta inference report to the network node via an uplink signaling channel of the mobile communication system. The transmitting may be performed via Layer 1 (L1) signaling, such as uplink control information (UCI) carried on the Physical Uplink Control Channel (PUCCH) or the Physical Uplink Shared Channel (PUSCH), via Layer 2 (L2) signaling, such as a Medium Access Control - Control Element (MAC CE), or via Layer 3 (L3) signaling, such as a Radio Resource Control (RRC) message. The transmitting may also be referred to as sending, reporting, forwarding, or uploading the delta inference report. The choice of signaling layer may depend on the size of the delta inference report and the latency requirements of the inference reporting, where L1 signaling provides the lowest latency and L3 signaling provides the highest reliability and flexibility. Because the delta inference report has a reduced size compared to a full inference report comprising the complete inference result, the transmission consumes fewer uplink resources and less UE transmission power compared to transmitting the full inference report, thereby contributing to reduced signaling overhead and improved UE power efficiency.
[0080] Fig. 3 illustrates block diagrams of examples of apparatuses 30, 40 for a network node 300, user equipment 400, and a mobile communication system 500. Fig. 3 illustrates an apparatus 30 for a network node 300 of a mobile communication system 500. The apparatus 30 comprises one or more interfaces 32 configured to communicate in the mobile communication system 500. The one or more interfaces 32 are coupled to one or more processing devices 34, which are configured to perform one of the methods 10 for the network node 300 as described herein with reference to Fig. 1.
[0081] Fig. 3 also shows an apparatus 40 for a UE 400 of the mobile communication system 500. The apparatus 40 comprises one or more interfaces 42 configured to communicate in the mobile communication system 500. The one or more interfaces 42 are coupled to one or more processing devices 44, which are configured to perform one of the methods 20 for the UE 400 as described herein with reference to Fig. 2.
[0082] Internal202502837 15
[0083] As optional components (dashed lines), optional from the perspective of apparatuses 30, 40, Fig. 3 also shows a network node 300 of the mobile communication system 500 comprising the apparatus 30 and a UE 400 of the mobile communication system 500 comprising the apparatus 40. Fig. 3 also illustrates an example of the mobile communication system 500 comprising the network node 300 and the UE 400.
[0084] As illustrated in Fig. 3, the respective one or more processing devices 34, 44 are coupled to the respective one or more interfaces 32, 42. The one or more interfaces 32, 42 may serve as an interface for communicating in the mobile communication system 500. The one or more interfaces 32, 42 may correspond to one or more inputs and / or outputs for receiving and / or transmitting information, which may be in digital (bit) values or analog according to a specified code or protocol, within a module, between modules, or between modules of different entities. For example, an interface may comprise interface circuitry configured to receive and / or transmit information. In examples, an interface may comprise any means for obtaining, receiving, transmitting, or providing analog or digital signals or information, e.g., any connector, contact, pin, register, input port, output port, conductor, lane, etc., which allows providing or obtaining a signal or information. The one or more interfaces 32, 42 may be configured to communicate (transmit, receive, or both) in a wireless and / or wired manner, and they may be configured to communicate, i.e. , transmit and / or receive signals or information with further internal or external components.
[0085] The one or more interfaces 32, 42 or the apparatuses 30, 40 may comprise further components to enable communication in the mobile communication system 500; such components may include transceiver (transmitter and / or receiver) components, such as one or more Low-Noise Amplifiers (LNAs), one or more Power Amplifiers (PAs), one or more duplexers, one or more diplexers, one or more filters or filter circuitry, one or more converters, one or more mixers, accordingly adapted radio frequency components, one or more antennas, etc. For example, the respective one or more interfaces 32, 42 may enable radio communication with UEs and communication between base stations, which can be direct and / or indirect, wired and / or wireless, respectively.
[0086] Internal202502837 16
[0087] The one or more (signal) processing devices 34, 44 may be implemented using one or more processing units, one or more circuitries, or any means for processing, such as a processor, a computer, or a programmable hardware component being operable with accordingly adapted software. In other words, the described function of the one or more processing devices 34, 44 may as well be implemented in software, which is then executed on one or more programmable hardware components. Such hardware components may comprise a general-purpose processor, a Digital Signal Processor (DSP), a micro-controller, etc.
[0088] In examples, network node 300 may also be referred to as a base station, server, network device, etc. It may belong to an access network or to a core network. A network entity / node may correspond to a remote radio head, a transmission point, an access point, a macro cell, a small cell, a micro cell, a pico cell, a femto cell, or a metro cell. The term small cell may refer to any cell smaller than a macro cell, e.g., a micro cell, a pico cell, a femto cell, or a metro cell. A network node / base station can be a wireless interface of a wired network, which enables transmission and reception of radio signals to a communication device. For example, the network node 300 may be a gNB (next generation NodeB), an eNodeB, a NodeB, a BTS (Base Transceiver Station), or an access point, all of which may be implemented in a terrestrial station, a satellite, a high-altitude platform (HAPS), a plane, a drone, etc. .
[0089] In general, the UE 400 is capable of communicating wirelessly, e.g., smartphone, tablet, laptop, vehicle-mounted device, wearable, loT device, etc. In particular, however, the UE 400 may be a mobile communication device, e.g., a communication device that is suitable for being carried around by a user. For example, the UE 400 may be a mobile phone, such as a smartphone, a network access device embedded in a vehicle, ship, or airplane, or another type of mobile communication device, such as a computer, a laptop computer, a tablet computer, and so on. For example, the UE 400 may be a User Terminal (UT) or UE within the meaning of the respective communication standards being used for mobile communication.
[0090] Accordingly, the UE 400 and the network node 300 may be configured to communicate in a cellular mobile communication system, for example in a Sub-6GHz-based cellular mobile communication system (covering frequency bands between 400 MHz and 7
[0091] Internal202502837 17
[0092] GHz), in a mmWave-based cellular mobile communication system (covering frequency bands between 24 GHz and 71 GHz), or in the so-called mid-bands (covering frequency bands between 7 GHz and 24 GHz). In general, the mobile communication system 500 may, for example, correspond to one of the 3GPP-standardized mobile communication networks, where the term mobile communication system is used synonymously with mobile communication network. The mobile communication system 500 may correspond to, for example, a 6th Generation system (6G), a 5th Generation system (5G), a New Radio (NR) system, Long-Term Evolution (LTE), an LTE-Advanced (LTE-A) system, High Speed Packet Access (HSPA), a Universal Mobile Telecommunication System (UMTS), or a UMTS Terrestrial Radio Access Network (UTRAN), an evolved-UTRAN (e-UTRAN), a Global System for Mobile communication (GSM) or Enhanced Data rates for GSM Evolution (EDGE) network, a GSM / EDGE Radio Access Network (GERAN), or mobile communication networks with different standards, for example, generally an Orthogonal Frequency Division Multiple Access (OFDMA) network, a Time Division Multiple Access (TDMA) network, a Code Division Multiple Access (CDMA) network, a Wideband-CDMA (WCDMA) network, a Frequency Division Multiple Access (FDMA) network, a Spatial Division Multiple Access (SDMA) network, etc.. The mobile communication system 500 may hence be cellular. The term cell refers to a coverage area of radio services provided by a transmission point, a remote unit, a remote head, a remote radio head, a communication device, a network entity, or a NodeB, an eNodeB, an ngNB, a gNB, a beam, or a satellite, respectively.
[0093] As further illustrated in Fig. 3, the apparatus 30 in the network node 300 configures apparatus 40 of the UE 400 to perform the AI / ML model inference reporting relative to a reference inference result (step 11 in Fig. 1 , step 21 in Fig. 2). The UE then performs the AI / ML model inference to obtain an inference result (step 22 in Fig.2) and determines the delta inference report (step 23 in Fig. 2). The UE 400 then transmits the delta inference report, which is received by the network node 300 (step 12 in Fig.1 , step 24 in Fig. 2).
[0094] Fig. 4 shows another example of a flowchart of a method for a network node. The network node transmits / provides a mapping between inference and the associated dataset, model ID and functionality ID to the UE in step 412.
[0095] Internal202502837 18
[0096] In some examples, the network node transmits the mapping in step 412 upon initial deployment of one or more AI / ML models at the UE, for example as part of an initial model configuration procedure. In other examples, the network node transmits or updates the mapping upon detecting a change in the radio environment, upon a handover of the UE to a new cell, upon deployment of an additional AI / ML model at the UE, or upon reconfiguration of an existing AI / ML model. In yet further examples, the network node periodically refreshes the mapping, for example at a configurable interval, to account for temporal variations in the expected inference results.
[0097] The network node may, upon receiving the delta inference report, determine whether the reported deviation exceeds a predefined deviation threshold. If the deviation exceeds the threshold, the network node may take one or more corrective actions, such as configuring or updating one or more AI / ML model parameters for the UE, instructing the UE to switch to a different AI / ML model, triggering a model retraining or fine-tuning procedure at the UE, updating the reference inference result in the mapping, or initiating a fallback to a legacy non-AI / ML-based procedure. If the deviation is below the threshold, the network node may continue operating with the current AI / ML model configuration without modification.
[0098] The network node may configure the UE to transmit a full inference report instead of a delta inference report when the deviation exceeds a further threshold that is larger than the deviation threshold, or when the network node explicitly requests a full inference report, for example for model validation or data collection purposes. The network node may signal the request for a full inference report via DCI, MAC CE, or RRC signaling. After receiving the full inference report, the network node may update the reference inference result in the mapping and instruct the UE to resume delta inference reporting relative to the updated reference inference result.
[0099] In some examples, the network node receives delta inference reports from a plurality of UEs and aggregates the reported deviations to monitor the overall performance of a deployed AI / ML model across the cell or across a group of UEs. Based on the aggregated deviations, the network node may determine whether the AI / ML model requires retraining, whether the reference inference results in the mapping require
[0100] Internal202502837 19
[0101] updating, or whether a model switch should be triggered for all UEs in the cell or for a subset thereof.
[0102] Fig. 5 depicts an example of a flowchart of a method for a UE. The UE transmits or provides delta inference to the network (node) in step 512. The UE may determine, prior to step 512, whether the computed deviation exceeds a predefined reporting threshold. If the deviation exceeds the reporting threshold, the UE transmits the delta inference report in step 512. If the deviation does not exceed the reporting threshold, the UE refrains from transmitting a delta inference report, thereby further reducing the uplink signaling overhead and UE power consumption. The reporting threshold may be configured by the network node, for example via RRC signaling or MAC CE, or may be a predefined value stored at the UE.
[0103] In some examples, the UE performs step 512 periodically, at a reporting interval configured by the network node. In other examples, the UE performs step 512 upon occurrence of a triggering event, such as when the deviation exceeds the reporting threshold, when a timer expires, when the UE detects a change in channel conditions, or when the UE receives an explicit reporting request from the network node. In yet further examples, the UE performs step 512 both periodically and upon occurrence of a triggering event, with the event-triggered report supplementing the periodic reports.
[0104] If the deviation between the obtained inference result and the reference inference result exceeds a further threshold or if the UE receives a request from the network node, the UE may transmit a full inference report comprising the complete inference result instead of the delta inference report, at least in some exam les. After transmitting the full inference report, the UE may receive an updated mapping from the network node and resume delta inference reporting relative to the updated reference inference result.
[0105] The process of AI / ML model inference reporting at UE 400 is part of a broader framework for AI / ML model lifecycle management in mobile communication system 500, which encompasses model training, deployment, inference, monitoring, and updating. In 3GPP TR 38.843, signaling procedures for model inference reporting involve several key steps. First, UE 400 initiates the model inference process based on predefined triggers or conditions, such as a timer expiry, a change in channel conditions, a request
[0106] Internal202502837 20
[0107] from network node 300, or an event-based trigger configured by network node 300. Second, UE 400 performs the inference using the deployed AI / ML model to produce an inference result. Third, UE 400 reports the inference results to network node 300, which can include various metrics and outcomes derived from the model. Fourth, based on the reported inference results, network node 300 may take specific actions, such as adjusting resource allocations, optimizing network parameters, triggering a model update, or selecting a different AI / ML model for subsequent inference. The AI / ML model lifecycle management framework may also be referred to as a model management framework, an AI / ML model governance framework, or an inference management procedure.
[0108] In some examples, configuring 11 UE 400 comprises transmitting, to UE 400, a mapping associating one or more reference inference results with respective model identifiers, functionality identifiers, and / or data set identifiers. A mapping is a data structure that establishes a correspondence between entries, where each entry links a reference inference result to at least one of a model identifier, a functionality identifier, and a data set identifier. The mapping may be represented as a table, a list of tuples, a set of information elements within a signaling message, or any other structured data format that enables UE 400 to look up the reference inference result associated with a given model, functionality, or data set. The model identifier uniquely identifies a specific AI / ML model deployed at UE 400, and may also be referred to as a model index, a model ID, or a model reference number. The functionality identifier uniquely identifies a specific AI / ML functionality for which the model is used, such as beam management, channel state information feedback, positioning, or radio link monitoring, and may also be referred to as a functionality index, a functionality ID, or a use-case identifier. The data set identifier uniquely identifies a specific data set or input configuration associated with the inference, and may also be referred to as a data set index, a data set ID, a training data reference, or an input configuration identifier. By providing the mapping, network node 300 enables UE 400 to identify the correct reference inference result against which to compare its obtained inference result when determining the delta inference report, thereby ensuring that the delta is computed with respect to the appropriate baseline for the specific model, functionality, and data set combination in use.
[0109] Internal202502837 21
[0110] The following table illustrates an example of a mapping that may be provided by network node 300 to UE 400.
[0111] Model ID Functionality ID Data set inference
[0112]
[0113] M1 F1 D1 11
[0114] M2 F2 D2 I2
[0115] M3 F3 D3 I3
[0116]
[0117]
[0118] The table comprises four columns: a model identifier column (Model ID), a functionality identifier column (Functionality ID), a data set identifier column (Data set), and a reference inference result column (inference). Each row of the table defines a complete mapping entry. In the first row, model identifier M1 is associated with functionality identifier F1 , data set identifier D1 , and reference inference result 11. In the second row, model identifier M2 is associated with functionality identifier F2, data set identifier D2, and reference inference result I2. In the third row, model identifier M3 is associated with functionality identifier F3, data set identifier D3, and reference inference result I3. When UE 400 performs an inference using, for example, model M2 on data set D2 for functionality F2, UE 400 looks up the corresponding reference inference result I2 from the mapping and computes the delta inference report based on the deviation of its obtained inference result from reference inference result I2.
[0119] The table may contain any number of rows and may be extended or updated by network node 300 during the connection, for example upon deployment of additional AI / ML models or upon reconfiguration of existing models. The table may also be referred to as a reference inference table, an inference mapping table, a model-inference association table, or a reference result lookup table.
[0120] Internal202502837 22
[0121] In some examples, the mapping is UE-specific and is transmitted via a Radio Resource Control, RRC, signaling message, for example an RRC reconfiguration message. A UE-specific mapping is a mapping that is individually configured for a particular UE 400 and may differ from mappings configured for other UEs in mobile communication system 500. By transmitting the mapping via RRC signaling, network node 300 can tailor the reference inference results and their associations with model identifiers, functionality identifiers, and data set identifiers to the specific models deployed at UE 400 and to the specific operational context of UE 400. In other examples, the mapping is cell-specific and is transmitted via a system information message, such as a System Information Block (SIB). A cell-specific mapping is a mapping that is common to all UEs within a cell served by network node 300. By transmitting the mapping via a system information message, network node 300 can efficiently provide the same reference inference mapping to all UEs in the cell without requiring individual RRC signaling to each UE, thereby reducing signaling overhead in scenarios where all UEs in the cell use the same set of AI / ML models and reference inference results. In further examples, the mapping may be partly UE-specific and partly cell-specific, for instance where a cell-specific mapping provides a base set of reference inference results and a UE-specific mapping provides additional or overriding entries for models that are specific to a particular UE 400.
[0122] In some examples, the delta inference report is received by network node 300 via at least one of: Layer 1 , L1 , signaling; Layer 2, L2, signaling; or Layer 3, L3, signaling. L1 signaling refers to physical layer signaling, such as uplink control information (UCI) transmitted on the Physical Uplink Control Channel (PUCCH) or the Physical Uplink Shared Channel (PUSCH), and is suitable for transmitting compact delta inference reports with low latency. L2 signaling refers to Medium Access Control (MAC) layer signaling, such as a MAC Control Element (MAC CE), and provides a balance between latency and payload size for transmitting delta inference reports of moderate size. L3 signaling refers to Radio Resource Control (RRC) layer signaling, such as an RRC message or a measurement report, and is suitable for transmitting larger delta inference reports or for reporting that is part of an RRC-configured measurement procedure. The choice of signaling layer may depend on the size of the delta inference report, the urgency of the report, and the configuration provided by network node 300. By supporting reception of the delta inference report via L1 , L2, or L3 signaling, network
[0123] Internal202502837 23
[0124] node 300 can flexibly adapt the reporting mechanism to the requirements of the specific AI / ML use case. The signaling layer used for transmitting the delta inference report may also be referred to as the reporting channel, the feedback signaling layer, or the inference report transport mechanism.
[0125] In some examples, method 10 further comprises configuring or updating one or more AI / ML model parameters for UE 400 based on the received delta inference report.
[0126] Configuring or updating AI / ML model parameters means that network node 300, upon receiving the delta inference report from UE 400, adjusts one or more parameters of the AI / ML model deployed at UE 400. The model parameters may include, for example, model weights, bias values, hyperparameters, activation thresholds, quantization levels, or any other configurable parameter that affects the behavior or output of the AI / ML model. The configuring or updating may comprise transmitting updated model parameters to UE 400 via RRC signaling or MAC CE, instructing UE 400 to switch to a different pre-stored model version, triggering a fine-tuning or re-training procedure at UE 400, or adjusting the reference inference result in the mapping for subsequent delta inference reporting. By configuring or updating the model parameters based on the delta inference report, network node 300 can close the loop of the AI / ML model lifecycle management, ensuring that deviations detected through the delta inference reporting are addressed and that the AI / ML model at UE 400 remains aligned with the expected behavior. The configuring or updating may also be referred to as adapting, modifying, recalibrating, or refining the model parameters.
[0127] In some examples, the delta inference report comprises an identifier of the reference inference result used by UE 400 and a difference value representing the deviation of the obtained inference result from the reference inference result. The identifier of the reference inference result may be an index into the mapping table, a model identifier, a functionality identifier, a data set identifier, or a combination thereof that unambiguously identifies which reference inference result was used by UE 400 as the baseline for computing the delta. By including the identifier, UE 400 enables network node 300 to determine against which reference inference result the deviation was computed, which is particularly relevant when the mapping contains multiple entries. The difference value is a numerical or structured representation of the deviation and may take the form of an absolute difference, a relative percentage, a vector of element-wise differences, a set of
[0128] Internal202502837 24
[0129] indices and corresponding delta values for changed elements, or a compressed representation of the deviation using differential encoding. The difference value may also be referred to as a delta value, a residual value, an offset value, or a deviation measure. In some examples, the reference inference result is associated with at least one of: a specific AI / ML model identified by a model identifier; a specific AI / ML functionality identified by a functionality identifier; or a specific data set identified by a data set identifier. This association enables both UE 400 and network node 300 to unambiguously determine the context in which the reference inference result applies and to compute or interpret the delta inference report correctly for the specific model, functionality, and data set combination in use.
[0130] On the UE side, in some examples, receiving 21 the configuration comprises receiving, from network node 300, a mapping associating one or more reference inference results with respective model identifiers, functionality identifiers, and data set identifiers, as described above with reference to method 10. UE 400 stores the received mapping and uses it as a lookup resource when performing AI / ML model inference and determining the delta inference report. The mapping received by UE 400 corresponds to the mapping transmitted by network node 300 as described above, and may be received as part of an RRC configuration message, a system information message, or a combination thereof.
[0131] In some examples, the mapping received by UE 400 is UE-specific and is received via an RRC / LPP signaling message, and / or the mapping is cell-specific and is received via a system information message, as described above with reference to method 10. When the mapping is received via RRC signaling, UE 400 applies the mapping as a UE-specific configuration that remains valid until a subsequent RRC reconfiguration replaces or updates the mapping. When the mapping is received via a system information message, UE 400 applies the mapping as a cell-specific configuration that is common to all UEs in the cell. UE 400 may combine a cell-specific mapping received via system information with a UE-specific mapping received via RRC signaling, for example by using the UE-specific mapping to override or supplement entries of the cellspecific mapping.
[0132] Internal202502837 25
[0133] In some examples, the delta inference report is transmitted by UE 400 via at least one of: L1 signaling; L2 signaling; or L3 signaling. The signaling layer used for transmitting the delta inference report from UE 400 corresponds to the signaling layer on which network node 300 receives the delta inference report, as described above. UE 400 may be configured by network node 300 to use a specific signaling layer for transmitting the delta inference report, or UE 400 may autonomously select the signaling layer based on the size of the delta inference report or the latency requirements. Because the delta inference report has a reduced size compared to a full inference report, the delta inference report can in many cases be transmitted via L1 signaling, which would not be feasible for a full inference report due to the limited payload capacity of UCI on PUCCH. This enables faster reporting and reduced UE power consumption compared to L3-based full inference reporting.
[0134] In some examples, determining 23 the delta inference report comprises identifying, from the mapping, the reference inference result corresponding to the AI / ML model used for the inference, and computing the deviation between the obtained inference result and the identified reference inference result. Identifying the reference inference result from the mapping means that UE 400 looks up the mapping entry whose model identifier, functionality identifier, and / or data set identifier matches the AI / ML model, functionality, and data set used for the current inference, and retrieves the reference inference result from that mapping entry. For example, if UE 400 performs an inference using model M2 for functionality F2 on data set D2, UE 400 identifies reference inference result I2 from the mapping table. Computing the deviation means that UE 400 calculates a difference, offset, or residual between the obtained inference result and the identified reference inference result I2. The computing may comprise element-wise subtraction, relative difference computation, identification of changed components, or any other mathematical operation that quantifies the discrepancy between the two inference results. The identifying and computing may also be referred to as matching and differencing, looking up and subtracting, or resolving and comparing. By structuring the determining step as a two-phase process of identification and computation, UE 400 ensures that the delta is computed against the correct baseline, which is particularly important when UE 400 has multiple AI / ML models deployed simultaneously for different functionalities.
[0135] Internal202502837 26
[0136] In some examples, the delta inference report transmitted by UE 400 comprises an identifier of the reference inference result used and a difference value representing the deviation of the obtained inference result from the reference inference result. The identifier and the difference value are as described above with reference to method 10. By including both the identifier and the difference value in the delta inference report, UE 400 provides network node 300 with sufficient information to reconstruct the complete inference result by combining the identified reference inference result with the reported difference value, without requiring UE 400 to transmit the full inference result. This reconstruction capability at network node 300 further confirms that the delta inference report achieves a reduced size compared to a full inference report while preserving the informational content necessary for network node 300 to take appropriate network actions.
[0137] The aspects and features described in relation to a particular one of the previous examples may also be combined with one or more of the further examples to replace an identical or similar feature of that further example or to additionally introduce the features into the further example.
[0138] Examples may further be or relate to a (computer) program including a program code to execute one or more of the above methods when the program is executed on a computer, processor or other programmable hardware component. Thus, steps, operations or processes of different ones of the methods described above may also be executed by programmed computers, processors or other programmable hardware components. Examples may also cover program storage devices, such as digital data storage media, which are machine-, processor- or computer-readable and encode and / or contain machine-executable, processor-executable or computer-executable programs and instructions. Program storage devices may include or be digital storage devices, magnetic storage media such as magnetic disks and magnetic tapes, hard disk drives, or optically readable digital data storage media, for example. Other examples may also include computers, processors, control units, (field) programmable logic arrays ((F)PLAs), (field) programmable gate arrays ((F)PGAs), graphics processor units (GPUs), application-specific integrated circuits (ASICs), integrated circuits (ICs) or system-on-a-chip (SoCs) systems programmed to execute the steps of the methods described above.
[0139] Internal202502837 27
[0140] It is further understood that the disclosure of several steps, processes, operations or functions disclosed in the description or claims shall not be construed to imply that these operations are necessarily dependent on the order described, unless explicitly stated in the individual case or necessary for technical reasons. Therefore, the previous description does not limit the execution of several steps or functions to a certain order. Furthermore, in further examples, a single step, function, process or operation may include and / or be broken up into several sub-steps, -functions, -processes or -operations.
[0141] If some aspects have been described in relation to a device or system, these aspects should also be understood as a description of the corresponding method. For example, a block, device or functional aspect of the device or system may correspond to a feature, such as a method step, of the corresponding method. Accordingly, aspects described in relation to a method shall also be understood as a description of a corresponding block, a corresponding element, a property or a functional feature of a corresponding device or a corresponding system.
[0142] The following claims are hereby incorporated in the detailed description, wherein each claim may stand on its own as a separate example. It should also be noted that although in the claims a dependent claim refers to a particular combination with one or more other claims, other examples may also include a combination of the dependent claim with the subject matter of any other dependent or independent claim. Such combinations are hereby explicitly proposed, unless it is stated in the individual case that a particular combination is not intended. Furthermore, features of a claim should also be included for any other independent claim, even if that claim is not directly defined as dependent on that other independent claim.
[0143] Internal202502837 28
[0144] Reference numerals
[0145] 10 method for network node
[0146] 11 configuring AI / ML inference reporting
[0147] 12 receiving delta inference report
[0148] 20 method for UE
[0149] 21 receiving configuration
[0150] 22 performing AI / ML model inference
[0151] 23 determining delta inference report
[0152] 24 transmitting delta inference report
[0153] 30 apparatus for network node
[0154] 32 one or more interfaces
[0155] 34 one or more processing devices
[0156] 40 apparatus for UE
[0157] 42 one or more interfaces
[0158] 44 one or more processing devices
[0159] 300 network node
[0160] 400 UE
[0161] 412 transmit mapping
[0162] 500 mobile communication system
[0163] 512 transmit delta report
[0164] Internal
Claims
202502837 29ClaimsWhat is claimed is:
1. A method (10) for a network node (300) of a mobile communication system (500), the method (10) comprising:configuring (11) a user equipment (400), UE, to perform artificial intelligence I machine learning, AI / ML, model inference reporting relative to a reference inference result; andreceiving (12), from the UE (400), a delta inference report indicating a deviation of an inference result obtained by the UE (400) from the reference inference result,wherein the delta inference report has a reduced size compared to a full inference report comprising the complete inference result.
2. The method (10) of claim 1, wherein configuring (11) the UE (400) comprises transmitting, to the UE (400), a mapping associating one or more reference inference results with respective model identifiers, functionality identifiers, and / or data set identifiers.
3. The method (10) of claim 2, wherein the mapping is UE-specific and is transmitted via a Radio Resource Control, RRC, signaling message, or wherein the mapping is cell-specific and is transmitted via a system information message.
4. The method (10) of any one of claims 1 to 3, wherein the delta inference report is received via at least one of: Layer 1 , L1 , signaling; Layer 2, L2, signaling; or Layer 3, L3, signaling.
5. The method (10) of any one of claims 1 to 4, further comprising:configuring or updating one or more AI / ML model parameters for the UE (400) based on the received delta inference report.Internal202502837 306. The method (10) of any one of claims 1 to 5, wherein the delta inference report comprises an identifier of the reference inference result used by the UE (400) and a difference value representing the deviation of the obtained inference result from the reference inference result and / or wherein the reference inference result is associated with at least one of: a specific AI / ML model identified by a model identifier; a specific AI / ML functionality identified by a functionality identifier; or a specific data set identified by a data set identifier.
7. A method (20) for a user equipment (400), UE, in a mobile communication system (500), the method (20) comprising:receiving (21), from a network node (300), a configuration for performing artificial intelligence I machine learning, AI / ML, model inference reporting relative to a reference inference result;performing (22) an AI / ML model inference to obtain an inference result;determining (23) a delta inference report based on a deviation of the obtained inference result from the reference inference result; andtransmitting (24) the delta inference report to the network node (300),wherein the delta inference report has a reduced size compared to a full inference report comprising the complete inference result.
8. The method (20) of claim 7, wherein receiving (21) the configuration comprises receiving, from the network node (300), a mapping associating one or more reference inference results with respective model identifiers, functionality identifiers, and data set identifiers.
9. The method (20) of claim 8, wherein the mapping is UE-specific and is received via a Radio Resource Control, RRC, signaling message, and / or the mapping is cellspecific and is received via a system information message.Internal202502837 3110. The method (20) of any one of claims 7 to 9, wherein the delta inference report is transmitted via at least one of: Layer 1, L1, signaling; Layer 2, L2, signaling; or Layer 3, L3, signaling.
11. The method (20) of any one of claims 7 to 10, wherein determining (23) the delta inference report comprises:identifying, from the mapping, the reference inference result corresponding to the AI / ML model used for the inference; andcomputing the deviation between the obtained inference result and the identified reference inference result.
12. The method (20) of any one of claims 7 to 11 , wherein the delta inference report comprises an identifier of the reference inference result used and a difference value representing the deviation of the obtained inference result from the reference inference result.
13. A computer program having a program code for performing the method (10; 20) of any one of claims 1 to 12, when the computer program is executed on a computer, a processor, or a programmable hardware component.
14. An apparatus (30) for a network node (300) of a mobile communication system (500), the apparatus (30) comprising:one or more interfaces (32) configured to communicate in the mobile communication system (500); andone or more processing devices (34) configured to perform one of the methods (10) of claims 1 to 7.
15. An apparatus (40) for a user equipment (400), UE, of a mobile communication system (500), the apparatus (40) comprising:Internal202502837 32one or more interfaces (42) configured to communicate in the mobile communication system (500); andone or more processing devices (44) configured to perform one of the methods (20) of claims 8 to 12.Internal
