Communication device and method for generalized AI / ML model alignment

Abstract

The invention discloses a method for aligning a generalized artificial intelligence (AI) / machine learning (ML) model by a first node, and the method comprises the following steps: monitoring the generalized AI / ML model of the first node; and reporting alignment information to the second node through alignment signaling, the alignment information comprising setting information related to at least one configuration identifier (ID), state, scenario and / or configuration of the generalized AI / ML model.

Description

Technical Field

[0001] The present application relates to the field of wireless communication systems, and more specifically, to communication devices and methods for artificial intelligence (AI) / machine learning (ML) model replication for air interfaces. For example, the present application relates to the study item description (SID) of Release 18 on new radio (NR) air interface AI / ML established in the 3rd generation partnership project (3GPP) radio access network (RAN) plenary meeting 94e (December 2022). The discussion was led by RAN1 and started in May 2022. In particular, the present application covers one-sided models and two-sided models. Background Art

[0002] At the RAN 94e meeting, machine learning was established as a research project, with RAN1 likely to take the lead. After much discussion, it was decided to study three application scenarios: channel state information (CSI) feedback enhancement, beam prediction, and positioning.

[0003] The discussion follows a general framework and is discussed specifically for each application scenario. A generalized ML model means that the ML model is trained and tested using a dataset from at least one setting, but can run in settings that are different from the training and test datasets. Specifically, a generalized ML model can be tested in the following ways. This consensus was reached at the RAN1110 meeting. Generalization can be achieved in the following cases.

[0004] According to the chairperson's minutes of the RAN1 110 meeting, the consensus reached is as follows.

[0005] As a starting point, the following scenarios are considered to validate the generalized performance of AI / ML models in different scenarios / configurations:

[0006] Case 1: The AI / ML model is trained based on a training dataset from Scenario #A / Configuration #A, and then the AI / ML model performs inference / testing on the same Scenario #A / Configuration #A dataset.

[0007] Case 2: The AI / ML model is trained based on a training dataset from scenario #A / configuration #A, and then the AI / ML model performs inference / testing on a dataset different from scenario #A / configuration #A, such as scenario #B / configuration #B or scenario #A / configuration #B.

[0008] Case 3: The AI / ML model is trained on a training dataset constructed by mixing datasets from multiple scenarios / configurations, including scenario #A / configuration #A and other datasets different from scenario #A / configuration #A, such as scenario #B / configuration #B, scenario #A / configuration #B. The AI / ML model then performs reasoning / testing on a dataset from a single scenario / configuration among the multiple scenarios / configurations, such as scenario #A / configuration #A, scenario #B / configuration #B, scenario #A / configuration #B.

[0009] Note: Enterprises are required to report the ratio of dataset mixing. Note: The number of multiple scenarios / configurations can be greater than two. For Further Study (FFS): A specific set of scenarios / configurations. For Further Study (FFS): Other cases of generalized validation, such as Case 2A: The AI / ML model is trained based on the training dataset of scenario #A / configuration #A, and then updated based on a fine-tuning dataset different from scenario #A / configuration #A (e.g., scenario #B / configuration #B, scenario #A / configuration #B). Afterwards, the AI / ML model is tested on a dataset different from scenario #A / configuration #A, such as scenario #B / configuration #B, scenario #A / configuration #B. At the same time, determine the indicators for model performance monitoring. These indicators can be used as a reference for generalized performance evaluation.

[0010] According to the chairman's record of the RAN1 110b-e meeting, in the CSI compression application scenario using the bilateral model, at least the following performance monitoring indicator options were further studied: reuse of intermediate KPIs as monitoring indicators (direct indicators, such as SGCS); indirect other indicators (such as BLER, NACK / ACK); input-based monitoring, such as data drift between training data sets and observation data sets; monitoring based on traditional CSI; other options are not excluded.

[0011] Issue 1: The understanding of the generalized model in a specific setting should be consistent between the gNB and the UE. The gNB and the UE may have different understandings of whether the generalized model works properly in certain settings. Model monitoring can indicate that the generalized ML model fails in certain settings. This indication may occur at the gNB, the UE, or a third node. Therefore, in this setting, the gNB and the UE may have different understandings of the input / output of the generalized model, and such issues should be avoided.

[0012] Question 2: When to deactivate / activate / fine-tune / retrain a generalized model? A generalized model can operate in multiple settings. If an ML model fails in a specific setting, whether the relevant operation needs to be performed in all settings supported by the ML model is still an open question. There are currently few relevant prior art references in 3GPP.

[0013] As can be seen from the above problems, there are many possible settings for evaluating generalized ML models. Some companies suggest assigning an ID to each setting to follow the CSI framework. Therefore, communication devices and methods are urgently needed to align generalized artificial intelligence (AI) / machine learning (ML) models to solve the problems in the existing technology and provide a set of methods including model alignment, deactivation, activation, fine-tuning and retraining to reduce overhead, improve communication performance and / or provide high reliability. Summary of the invention

[0014] The purpose of this application is to propose a communication device and method for aligning a generalized artificial intelligence (AI) / machine learning (ML) model, which can solve the problems in the prior art and provide a method for aligning, deactivating, activating, fine-tuning, and retraining the model to reduce overhead, improve communication performance and / or provide high reliability.

[0015] In a first aspect of the present application, a method for aligning a generalized artificial intelligence (AI) / machine learning (ML) model by a first node is provided, comprising monitoring the generalized AI / ML model of the first node; and reporting alignment information to a second node through alignment signaling, wherein the alignment information includes setting information related to at least one configuration identifier (ID), state, scenario and / or configuration of the generalized AI / ML model.

[0016] In a second aspect of the present application, a first node is provided, comprising a memory; a transceiver; and a processor coupled to the memory and the transceiver, wherein the processor is configured to execute the above method.

[0017] In a third aspect of the present application, a method for aligning a generalized artificial intelligence (AI) / machine learning (ML) model by a second node is provided, comprising monitoring the generalized AI / ML model of the second node; and updating alignment information to the first node through alignment signaling, wherein the alignment information includes setting information related to at least one configuration identifier (ID), state, scenario and / or configuration of the generalized AI / ML model.

[0018] According to a fourth aspect of the present application, a second node is provided, comprising a memory; a transceiver; and a processor coupled to the memory and the transceiver, wherein the processor is configured to execute the above method.

[0019] In a fifth aspect of the present application, a non-temporary machine-readable storage medium is provided, in which instructions are stored, and when the instructions are executed by a computer, the computer executes the above method.

[0020] In a sixth aspect of the present application, a chip is provided, the chip comprising a processor, the processor being configured to call and run a computer program stored in a memory so that a device on which the chip is installed executes the above method.

[0021] In a seventh aspect of the present application, a computer-readable storage medium is provided, in which a computer program is stored, and the program can enable a computer to execute the above method.

[0022] In an eighth aspect of the present application, a computer program product is provided, which includes a computer program, and the program can enable a computer to execute the above method.

[0023] In a ninth aspect of the present application, a computer program is provided, which can enable a computer to execute the above method. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or related technologies, the following briefly introduces the drawings to be described in the embodiments. Obviously, the drawings are only some embodiments of the present application, and ordinary technicians in this field can obtain other drawings based on these drawings without paying any price.

[0025] Figure 1 The schematic diagram of the basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application is exemplarily illustrated.

[0026] Figure 2 It is a block diagram of a communication node in a communication network system according to an embodiment of the present application.

[0027] Figure 3The present invention exemplarily illustrates a flowchart of a method for aligning a generalized artificial intelligence (AI) / machine learning (ML) model by a first node according to an embodiment of the present application.

[0028] Figure 4 The flowchart of the method for aligning a generalized artificial intelligence (AI) / machine learning (ML) model by a second node according to an embodiment of the present application is exemplarily illustrated.

[0029] Figure 5 The schematic diagram exemplarily illustrates the functional framework of RAN intelligence according to an embodiment of the present application.

[0030] Figure 6 The flowchart of the monitoring process of the gNB for the ML model according to an embodiment of the present application is exemplified.

[0031] Figure 7 The flowchart of the monitoring process of the UE to the ML model according to an embodiment of the present application is exemplarily illustrated.

[0032] Figure 8 A block diagram of a communication device (eg, UE) according to an embodiment of the present application is exemplarily illustrated.

[0033] Fig. 9 A block diagram of a communication device (e.g., gNB) according to an embodiment of the present application is exemplarily illustrated.

[0034] Fig.10 A system block diagram for wireless communication according to an embodiment of the present application is exemplarily described. DETAILED DESCRIPTION

[0035] The technical content, structural features, objectives and effects of the present application are described in detail below in conjunction with the accompanying drawings. Specifically, the terms in the embodiments of the present application are only used to describe the purpose of specific embodiments, and are not intended to limit the present application.

[0036] AI / ML is introduced into the physical layer (PHY layer) and medium access control (MAC) layer to enhance system performance. Several use cases have been identified in 3GPP RAN1 for research, namely: CSI feedback compression; beam management; positioning. ML models can be trained online or offline.

[0037] For the machine learning model, training is required first. The training can be performed by a node, which can be a gNB (next generation base station), UE (user equipment), or a third node. The training method can be online training or offline training. After the ML model training is completed, it will be deployed. For enhanced CSI feedback, the ML model is a bilateral model, so the model is deployed on the UE and gNB respectively. The operation of the CSI model will be monitored to ensure that it works properly.

[0038] Some embodiments of the present application discuss mechanisms for replicating AI / ML models over the air. Figure 1 The schematic diagram of the basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application is exemplarily illustrated. Figure 1 In some embodiments, the basic model of the autoencoder is as follows: the encoder compresses the original CSI-RS value (referred to as the original CSI) / maximum eigenvector and reports its output to the gNB; the gNB receives the report and decompresses it; the new CSI report is a CSI report containing CSI feedback enhanced by the AI / ML model. At the UE, the input signal is compressed and output to the channel. The input to the encoder can be the maximum eigenvector or the channel matrix; the compressed output is used as the input to the decoder and reconstructed at the gNB.

[0039] Figure 2 An embodiment of the present application is schematically illustrated, wherein in a communication network system 40, at least one first node 10 is included, such as at least one user equipment (UE); a second node 20, such as a base station (such as a gNB); and at least one third node 30, such as at least one user equipment (UE) or other communication device. The communication network system 40 includes: at least one first node 10 (for example, at least one UE); a second node 20 (for example, a base station (gNB)); and at least one third node 30 (for example, at least one UE or other communication device). The first node 10 may include: a memory 12; a transceiver 13; and a processor 11, which is coupled to the memory 12 and the transceiver 13. The second node 20 may include: a memory 22; a transceiver 23; and a processor 21, which is coupled to the memory 22 and the transceiver 23. The third node 30 may include: a memory 32; a transceiver 33; and a processor 31, which is coupled to the memory 32 and the transceiver 33. The processor 11, 21, or 31 may be configured to perform the functions, processes, and / or methods described in the present application. Each layer of the wireless interface protocol may be implemented on the processor 11, 21 or 31. The memory 12, 22 or 32 is operatively coupled to the processor 11, 21 or 31 and stores various information to support the operation of the processor 11, 21 or 31. The transceiver 13, 23 or 33 is operatively coupled to the processor 11, 21 or 31 and is used to send and / or receive wireless signals.

[0040] The processor 11, 21 or 31 may include an application-specific integrated circuit (ASIC); other chipsets; logic circuits; data processing devices. The memory 12, 22 or 32 may include a read-only memory (ROM); a random access memory (RAM); flash memory; a memory card; a storage medium and / or other storage devices. The transceiver 13, 23 or 33 may include a baseband circuit for processing radio frequency signals. When the embodiment is implemented in software form, the technology may be implemented by modules (such as processes, functions, etc.) that perform corresponding functions. These modules may be stored in the memory 12, 22 or 32 and executed by the processor 11, 21 or 31. The memory 12, 22 or 32 may be integrated inside the processor 11, 21 or 31; an external memory, which is communicatively connected to the processor 11, 21 or 31 in various known ways.

[0041] In some embodiments, the processor 11 is configured to monitor the generalized AI / ML model, and the transceiver 13 is configured to report alignment information containing configuration information (identifier, ID), status, scenario and / or configuration related to at least one of the generalized AI / ML models to the second node through alignment signaling. This method can solve the problems in the prior art, provide methods for aligning, deactivating, activating, fine-tuning, and retraining generalized models, reduce overhead, and provide good communication performance and / or high reliability. In addition, the transceiver 11 is also configured to receive updated alignment information from the second node.

[0042] In some embodiments, the processor 21 is configured to monitor the generalized AI / ML model and is configured to update the first node with alignment information including configuration information (identifier, ID), state, scenario and / or configuration related to at least one of the generalized AI / ML models through alignment signaling. The method can solve the problems in the prior art, provide a method for aligning, deactivating, activating, fine-tuning, and retraining the generalized model, reduce overhead, and provide good communication performance and / or high reliability.

[0043] Figure 3A method 300 for aligning a generalized artificial intelligence (AI) / machine learning (ML) model by a first node (e.g., UE) is shown, according to an embodiment of the present application. In some embodiments, the method 300 includes: step 302, monitoring the generalized AI / ML model at the first node; step 304, reporting to the second node through alignment signaling, alignment information containing configuration information (identifier, ID), status, scenario and / or configuration related to at least one of the generalized AI / ML models. The method can solve the problems in the prior art, provide methods for aligning, deactivating, activating, fine-tuning, and retraining generalized models, reduce overhead, and provide good communication performance and / or high reliability. In addition, the method 300 further includes receiving updated alignment information from the second node.

[0044] Figure 4 A method 400 for a second node (e.g., a gNB) to perform artificial intelligence (AI) / machine learning (ML) model alignment is shown, according to an embodiment of the present application. In some embodiments, the method 400 includes: step 402, monitoring a generalized AI / ML model at a second node; step 404, updating alignment information containing configuration information (identifier, ID), status, scenario and / or configuration related to at least one of the generalized AI / ML models to the first node via alignment signaling. The method can solve the problems in the prior art, provide methods for aligning, deactivating, activating, fine-tuning, and retraining generalized models, reduce overhead, and provide good communication performance and / or high reliability.

[0045] Figure 5 The schematic diagram exemplarily illustrates the functional framework of RAN intelligence according to an embodiment of the present application. Figure 5 It is shown that in some embodiments, the ML model needs to be monitored during the inference process. The functional framework of RAN intelligence is provided by RAN3 and can be further modified to apply to RAN1. After the ML model is deployed, it needs to be continuously monitored to check whether it is working properly. Typically, the performance of the ML model will be compared with the set standard. If the ML model fails to work properly, the UE will switch to another ML model; or fall back to non-AI mode. The monitored ML model will be retrained to ensure that it returns to normal.

[0046] The present application mainly relates to aligning the knowledge of a generalized model between two nodes, and to the working / operation mode of a generalized machine learning (ML) model. In some embodiments of the present application, a method for aligning the working state of a generalized ML model is provided. The present application provides signaling, an alignment process, and a method for activating, deactivating, fine-tuning, and / or retraining the working state of a generalized ML model for alignment. The effects of some embodiments of the present application may include at least one of the following: the UE and the gNB have the same understanding of the working state of the ML model under a specific setting, which helps the UE and the gNB to be consistent when exchanging information. For example, in a bidirectional ML model for enhanced CSI feedback, the model can be generalized between different ranks, but at certain ranks, the generalized model may not work properly due to changes in the wireless environment, so a non-generalized model should be used at the specific rank. Therefore, the gNB and the UE should align the understanding. In addition, a method for deactivating, activating, fine-tuning, and / or retraining a generalized model is also provided.

[0047] Example 1: Signaling design:

[0048] In some examples, the generalized ML model is configured to have multiple configurations. Each configuration contains a set of scenarios and a description of the configuration.

[0049] In some examples, the generalized ML model is configured to have multiple configurations and configuration IDs. Each configuration contains a description of a set of scenarios and configurations and is identified by a configuration ID.

[0050] Management and alignment of broad scope / dimensions / configurations. Through model monitoring, broad models can be run in a range of settings (scenarios, configurations).

[0051] Scenario A / Configuration X Scenario A / Configuration Y Scenario B / Configuration X Scenario B / Configuration Y 1 1 1 1

[0052] Note: 0: Fault; 1: Normal operation.

[0053] In some examples, the UE's monitoring in scenario A / configuration X is marked with the configuration ID:

[0054] Configuration ID 1 Configuration ID 2 Configuration ID 3 Configuration ID 4 1 1 1 1

[0055] Note: 0: Fault; 1: Normal operation.

[0056] The signaling format can be:

[0057] {configuration ID: status}, where status indicates whether the ML model works properly in the setup associated with that configuration ID.

[0058] In some examples, the signaling is in bit vector format and does not include a configuration ID, for example: {0101011}, where "1" indicates that the model works properly in a specific scenario / configuration, and "0" indicates that the model does not work properly.

[0059] In some examples, the alignment information may be in RRC signaling, MAC control element (MAC CE), or DCI field. In some cases, if the state of a configuration ID is not configured, it is assumed that the ML model works normally in the setting associated with the configuration ID.

[0060] In some examples, instead of introducing configuration IDs, the settings of the scenarios / configurations are labeled, numbered, or indexed. In this case, the index replaces the configuration ID, for example: {1, state, 2, state, 3, state...}, where "1", "2", "3" are indexes, and the state indicates whether the generalized ML model works normally in the corresponding setting. In some examples, the normal working state is represented as "1", otherwise it is "0". In some examples, the index may not be explicitly indicated.

[0061] In some examples, the alignment signaling does not include state information. For example, when the ML model works normally under the settings of configuration ID 1 and configuration ID 2, the alignment signaling can directly include {configuration ID 1, configuration ID 2}.

[0062] The absence of a configuration ID indicates that the ML model cannot work properly under the setting. In other words, the implicitly indicated configuration ID indicates that the ML model works properly under the setting corresponding to the configuration ID. In some examples, this set of configuration IDs can dynamically form an alignment information list. If the ML model is judged to be faulty under the setting corresponding to a certain configuration ID, the configuration ID can be deleted from the alignment signaling. On the contrary, if the ML model is judged to work properly, the new configuration ID can be added to the alignment signaling.

[0063] In some examples, the alignment signaling is located or included under the model description item. In some examples, the alignment signaling is a child of the model ID, and the configuration ID is subordinate to the model ID. For example, {model ID X: {configuration ID 1, configuration ID 2}}, where the presence of configuration ID 1 indicates that the ML model with model ID X works normally under the settings associated with configuration ID 1.

[0064] Example 2: Process of generalized model alignment:

[0065] Alignment refers to the UE and / or gNB performing specific processes to reach a consistent understanding of whether the generalized ML model works properly under specific settings. For example, when the model monitoring is on the gNB side, the alignment can be performed based on the gNB configuration; when the model monitoring is on the gNB side, the alignment can also be performed based on the UE's request; when the model monitoring is on the UE side, the alignment can be performed based on the UE's report. In some examples, the UE can directly report the alignment information and perform the configuration of the alignment information. In some examples, the UE reports the monitoring results and / or the recommended status configuration information, such as {scenario A / configuration X}, and then the gNB sends the configuration signaling of the alignment information.

[0066] The ML model can be deployed on the gNB or UE (single-sided model) or on both (double-sided model). The following examples involve single-sided models, and the alignment process is also applicable to double-sided models. Monitoring can be performed by the UE, the gNB, or both. Figure 6 As shown, the ML model is deployed on the gNB side in some examples and monitored by the gNB. The monitoring results are updated in the alignment signaling, and the configuration ID and its related status are configured to the UE through the alignment signaling.

[0067] like Figure 7 As shown, the UE monitors the ML model. In some examples, the UE reports alignment signaling to the gNB, and the gNB confirms the UE's report. First, the UE reports updated alignment information to the gNB, and the gNB can update the alignment information on its own side. Subsequently, the gNB confirms the alignment information to the UE.

[0068] Example 3: Transmission of alignment information:

[0069] In some examples, alignment signaling is triggered by monitoring results, and alignment can be initiated by the gNB or the UE. In some examples, alignment signaling is reported periodically. If the alignment information is not updated, the UE or gNB may not send alignment signaling. In some examples, alignment signaling is transmitted only when at least one of its fields is updated.

[0070] Example 4: Activation / deactivation of ML models:

[0071] This embodiment discusses when to deactivate, activate, retrain or fine-tune a generalized ML model. In some examples, if the generalized ML model is determined to work properly under a supported setting / configuration ID, the ML model is activated. In some examples, if the generalized ML model is determined to work properly under all supported setting / configuration IDs, the ML model is activated. In some examples, if the generalized ML model is determined to work properly under some supported setting / configuration IDs, the ML model is activated. The determination of whether the generalized ML model works properly is performed by model monitoring. In some examples, if the generalized ML model is determined to not work properly under a supported setting / configuration ID, the ML model is deactivated / retrained / fine-tuned. In some examples, if the generalized ML model is determined to not work properly under all supported setting / configuration IDs, the ML model is deactivated / retrained / fine-tuned. In some examples, if the generalized ML model is determined to not work properly under some supported setting / configuration IDs, the ML model is deactivated / retrained / fine-tuned. In some examples, if the generalized ML model is determined to not work properly under some supported setting / configuration IDs, the ML model is deactivated / retrained / fine-tuned.

[0072] Example 5: Data collection:

[0073] In some examples, alignment of generalized model states may trigger data collection. Alignment signaling may be sent when an ML model is deactivated or judged to be faulty. At the same time, the transmission of alignment signaling may implicitly trigger data collection. For example, for a two-sided model, if the model is judged to be faulty under a certain setting, it means that new data is needed to train or fine-tune the model, so data collection is triggered under this setting. For example, after the alignment signaling is transmitted, the UE reports the real data to the gNB.

[0074] In some examples, a bit may be appended to the end of the alignment signaling to indicate whether data collection is triggered at the same time, "0" indicating that data collection is not triggered, and "1" indicating that data collection is triggered.

[0075] Figure 8 A block diagram of a communication device 1000 according to an embodiment of the present application is exemplarily illustrated. The communication device 1000 may be a first node, such as a UE. The second node may be a base station. The communication device 1000 includes: a monitoring module 1001, configured to monitor a generalized AI / ML model; a transceiver 1002, configured to report alignment information containing configuration information (ID), status, scenario and / or configuration related to at least one generalized AI / ML model to the second node through alignment signaling. The method can solve the problems in the prior art, provide methods for aligning, deactivating, activating, fine-tuning, and retraining generalized models, reduce overhead, and provide good communication performance and / or high reliability. In addition, the transceiver 1002 is also configured to receive updated alignment information from the second node.

[0076] Fig. 9 A block diagram of a communication device 1100 according to an embodiment of the present application is exemplarily illustrated. The communication device 1100 may be a second node, such as a base station. The first node and the third node may be UEs. The communication device 1100 includes a monitoring module 1101, configured to monitor a generalized AI / ML model at a second node; an update module 1102, configured to update alignment information containing configuration information (ID), status, scenario and / or configuration related to at least one of the generalized AI / ML models to the first node through alignment signaling. The method can solve the problems in the prior art, provide methods for aligning, deactivating, activating, fine-tuning, and retraining generalized models, reduce overhead, and provide good communication performance and / or high reliability.

[0077] In some examples, the generalized AI / ML model is configured to have multiple configurations, each configuration including a description of a set of scenarios and configuration information. In some examples, the generalized AI / ML model is configured to have multiple configurations and configuration IDs, each configuration including a description of a set of scenarios and configuration information, and is identified by a configuration ID. In some examples, the alignment signaling indicates whether the generalized AI / ML model works properly under settings related to the configuration ID, state, scenario, and / or configuration, or according to the default settings, the alignment signaling indicates that the generalized AI / ML model works properly.

[0078] In some examples, a set of configuration IDs are dynamically configured and listed. If the generalized AI / ML model is judged to be faulty under the setting corresponding to a configuration ID, the configuration ID will be deleted from the alignment signaling; if the generalized AI / ML model is judged to be working properly, the configuration ID will be added to the alignment signaling.

[0079] In some examples, the alignment signaling can be at least one of the following: radio resource configuration (RRC) signaling, media access control-control element (MAC-CE) or downlink control information (DCI). In some examples, the alignment signaling is located under the model description item or as a sub-item of the model ID, and the configuration ID is subordinate to the model ID. In some examples, the alignment signaling is triggered by the monitoring results and can be reported periodically or transmitted only when at least one field in the alignment information is updated.

[0080] In some examples, the alignment signaling is a broadcast-like signaling that can be transmitted to multiple UEs. The broadcast-like signaling includes at least one of the following: a model ID, a type of ML model used for alignment (e.g., a pre-processing type, a post-processing type), a configuration ID, and a working status (i.e., whether the ML model is working properly). After receiving the alignment signaling, the multiple UEs report their respective alignment information to the gNB or the third node. The broadcast-like signaling can be at least one of the following: DCI, a reference signal (RS), MAC-CE, RRC signaling, ACK / NACK (e.g., whether the new data indicator is switched). As another way, in some examples, the alignment information is notified to multiple UEs by the gNB through a broadcast-like signaling. The broadcast-like signaling includes at least one of the following: a model ID, a type of ML model used for alignment (e.g., a pre-processing type, a post-processing type), a configuration ID, and a working status (i.e., whether the ML model is working properly). The broadcast-like signaling can be at least one of the following: DCI, RS, MAC-CE, RRC signaling, ACK / NACK (e.g., whether the new data indicator is switched).

[0081] In some examples, when the general AI / ML model is determined to be working normally in one, some or all supported setting / configuration IDs, the general AI / ML model is activated. In some examples, when the general AI / ML model is determined to be working abnormally in one, some or all supported setting / configuration IDs, the general AI / ML model is deactivated, retrained and / or fine-tuned. In some examples, alignment signaling triggers data collection. In some examples, after transmitting the alignment signaling, the first node reports real data to the second node. In some examples, an additional bit is appended to the alignment signaling to indicate whether data collection is triggered.

[0082] In summary, in some embodiments of the present application, a method is provided to align the understanding of the working state of a general model. Signaling for alignment is provided, an alignment process is provided, and activation, deactivation, fine-tuning and / or retraining of the working state of a general ML model is provided. The inventive effects of some embodiments of the present application may include at least one of the following. The UE and the gNB have the same understanding of the working state of the ML model under specific settings, which helps the UE and the gNB to reach a consistent understanding in information exchange. For example, if a bidirectional ML model generalizes between multiple ranks, and at a certain specific rank, the general model does not work properly, a non-general model should be used. Therefore, the gNB and the UE need to align this understanding. At the same time, a method for deactivating, activating, fine-tuning and / or retraining the general model is also provided.

[0083] Fig.107 is a block diagram of an exemplary wireless communication system 700 according to an embodiment of the present application. The embodiments described herein may be implemented into a system using any appropriately configured hardware and / or software. Fig.10 A system 700 is shown, which includes a radio frequency (RF) circuit 710, a baseband circuit 720, an application circuit 730, a memory / storage device 740, a display 750, a camera 760, a sensor 770, and an input / output (I / O) interface 780, which are coupled to each other at least as shown. The application circuit 730 may include circuits, such as but not limited to one or more single-core or multi-core processors. The processor may include any combination of general-purpose processors and special-purpose processors, such as a graphics processor, an application processor. The processor may be coupled to a memory / storage device and configured to execute instructions stored in the memory / storage device to enable various applications and / or operating systems to run on the system.

[0084] Monitoring of the ML model is not limited to the UE or gNB. Monitoring can be performed on a third node, and relevant signaling and data need to be reported to the third node. The third node can be a UE, a gNB, or a server. The method in the embodiment of the present invention is applicable. In this way, the signaling overhead between the gNB and the UE can be reduced.

[0085] While the present disclosure has been described in connection with what is considered to be the most practical and preferred embodiment, it is to be understood that the present disclosure is not limited to the disclosed embodiment, but is intended to cover various arrangements made without departing from the scope of the broadest interpretation of the appended claims.

Claims

1. A method for aligning a generalized artificial intelligence AI / machine learning ML model, performed by a first node, characterized in that: include: Monitor the generalized AI / ML model of the first node; as well as Alignment information is reported to the second node via alignment signaling, wherein the alignment information includes setting information related to at least one configuration identifier ID, state, scenario and / or configuration of the generalized AI / ML model.

2. The method according to claim 2, characterized in that: Also included is receiving updated alignment information from the second node.

3. The method according to claim 1, characterized in that The generalized AI / ML model is configured with multiple configurations, each configuration including a set of scenarios and configuration descriptions.

4. The method according to claim 1, characterized in that: The generalized AI / ML model is configured with multiple configurations and configuration IDs, each configuration includes a set of scenarios and configuration descriptions and is identified by a configuration ID.

5. The method according to any one of claims 1 to 4, characterized in that The alignment signaling indicates whether the generalized AI / ML model works properly under settings related to configuration ID, state, scenario and / or configuration, or under default settings, the alignment signaling indicates that the generalized AI / ML model works properly.

6. The method according to claim 4 or 5, characterized in that: The set of configuration IDs is dynamic. If the generalized AI / ML model is judged to be faulty under the setting corresponding to the configuration ID, the configuration ID is deleted from the alignment signaling; if the generalized AI / ML model is judged to be working normally, the configuration ID is added to the alignment signaling.

7. The method according to any one of claims 1 to 6, characterized in that The alignment signaling is at least one of the following: radio resource control RRC signaling, media access control-control element MAC-CE, or downlink control information DCI.

8. The method according to any one of claims 1 to 7, characterized in that The alignment signaling belongs to an entry of a model description or a sub-entry of a model ID, and the configuration ID is attached to the model ID.

9. The method according to any one of claims 1 to 8, characterized in that The alignment signaling is triggered by a monitoring result, and the alignment signaling is reported periodically or transmitted only when at least one field in the alignment information is updated.

10. The method according to any one of claims 1 to 9, characterized in that If the generalized AI / ML model is determined to be working properly under one, some or all of the supported setting / configuration IDs, the generalized AI / ML model is activated.

11. The method according to any one of claims 1 to 10, characterized in that When the generalized AI / ML model is determined to be operating abnormally under one, some or all supported setting / configuration IDs, the generalized AI / ML model is disabled, retrained and / or fine-tuned.

12. The method according to any one of claims 1 to 11, characterized in that The alignment signaling triggers data collection.

13. The method according to claim 12, characterized in that After the alignment signaling is transmitted, the first node reports real data to the second node.

14. The method according to claim 12 or 13, characterized in that An extra bit is appended at the end of the alignment signaling to indicate whether data collection is triggered.

15. A method for aligning the generalized artificial intelligence AI / machine learning ML model by the second node, characterized in that: include: monitoring a generalized AI / ML model of the second node; as well as Alignment information is updated to the first node through the alignment signaling, where the alignment information includes setting information related to at least one configuration identifier ID, state, scene and / or configuration of the generalized AI / ML model.

16. The method according to claim 15, characterized in that The generalized AI / ML model is configured with multiple configurations, each configuration including a set of scenarios and configuration descriptions.

17. The method according to claim 15, characterized in that The generalized AI / ML model is configured with multiple configurations and configuration IDs, each configuration includes a set of scenarios and configuration descriptions and is identified by the configuration ID.

18. The method according to any one of claims 15 to 17, characterized in that The alignment signaling indicates whether the generalized AI / ML model works properly under settings related to configuration ID, state, scenario and / or configuration, or under default settings, the alignment signaling indicates that the generalized AI / ML model works properly.

19. The method according to claim 17 or 18, characterized in that The set of configuration IDs is dynamic. If the generalized AI / ML model is judged to be faulty under the setting corresponding to the configuration ID, the configuration ID is deleted from the alignment signaling; if the generalized AI / ML model is judged to be working normally, the configuration ID is added to the alignment signaling.

20. The method according to any one of claims 15 to 19, characterized in that The alignment signaling is at least one of the following: radio resource control RRC signaling, media access control-control element MAC-CE, or downlink control information DCI.

21. The method according to any one of claims 15 to 20, characterized in that The alignment signaling belongs to an entry of a model description or a sub-entry of a model ID, and the configuration ID is attached to the model ID.

22. The method according to any one of claims 15 to 21, characterized in that The alignment signaling is triggered by a monitoring result, and the alignment signaling is reported periodically or transmitted only when at least one field in the alignment information is updated.

23. The method according to any one of claims 15 to 22, characterized in that If the generalized AI / ML model is determined to be working properly under one, some or all of the supported setting / configuration IDs, the generalized AI / ML model is activated.

24. The method according to any one of claims 15 to 23, characterized in that When the generalized AI / ML model is determined to be operating abnormally under one, some or all supported setting / configuration IDs, the generalized AI / ML model is disabled, retrained and / or fine-tuned.

25. The method according to any one of claims 15 to 24, characterized in that The alignment signaling triggers data collection.

26. The method according to claim 25, characterized in that After the alignment signaling is transmitted, the first node reports real data to the second node.

27. The method according to claim 25 or 26, characterized in that An extra bit is appended at the end of the alignment signaling to indicate whether data collection is triggered.

28. A first node, characterized in that: include: Memory; Transceiver; as well as a processor coupled to the memory and the transceiver; The processor is configured to execute the method according to any one of claims 1 to 14.

29. A second node, characterized in that: include: Memory; Transceiver; as well as a processor coupled to the memory and the transceiver; The processor is configured to execute the method according to any one of claims 15 to 27.

Images