Artificial intelligence / machine learning model update

Abstract

Embodiments of the present disclosure relate to artificial intelligence (AI) / machine learning (ML) model updates. In an aspect, a network device receives an AI / ML model and a first data set. The first data set includes a first set of data samples associated with the first set of tags. The network device obtains a second data set based on a plurality of unlabeled data samples, an AI / ML model, and a ratio of a number of tags indicating a line-of-sight (LOS) path to a number of tags indicating a non-line-of-sight (NLOS) path in the first tag set. The network device updates at least one model parameter of the AI / ML model based on the second data set. As provided in the disclosure, higher LOS / NLOS classification precision can be provided by the updated AI / ML model.

Description

Technical Field

[0001] Various example embodiments generally relate to the field of communications, and more particularly to network devices, terminal devices, methods, apparatuses and computer-readable storage media related to the updating of artificial intelligence (AI) / machine learning (ML) models. Background Technology

[0002] In communications technology, the scope of AI / ML-enabled positioning can be based on radio frequency (RF) fingerprints (e.g., channel impulse response (CIR), as discussed in the 3rd Generation Partnership Project (3GPP). In practice, some labeled data can be collected by positioning reference units (PRUs) deployed in the network. However, it is difficult to collect enough labeled data to enable large-scale model training for AI / ML-based positioning accuracy enhancement use cases. This has prompted research into AI / ML technologies with low dependence on labeled data.

[0003] Fortunately, unlabeled data containing only CIR is relatively easy to obtain. For example, one way to collect unlabeled data on the network side is for the User Equipment (UE) to report the CIR estimated from Location Reference Signal (PRS) measurements. In this way, a large amount of data without location labels but a relatively small amount of data with location labels will be obtained. This data can be used to train high-precision AI / ML models. Summary of the Invention

[0004] Generally, the exemplary embodiments of this disclosure provide network devices, terminal devices, methods, apparatuses, and computer-readable storage media for AI / ML updates. For example, the solutions provided by the exemplary embodiments of this disclosure can allow network devices to support updating AI localization models deployed at terminal devices with pseudo-labels, thereby providing higher line-of-sight (LOS) / non-line-of-sight (NLOS) classification accuracy.

[0005] In a first aspect, a network device is provided. The network device may include at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the network device to at least: receive an AI / ML model and a first dataset from an end device, wherein the AI / ML model has the capability to provide inference results for LOS / NLOS classification with confidence levels, and the first dataset includes a first set of data samples associated with a first label set, the first label set being true labels indicating LOS paths or NLOS paths; acquire a second dataset based on a plurality of unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set, wherein the second dataset includes a second set of data samples associated with a second label set, the second label set being pseudo labels indicating LOS paths or NLOS paths; and update at least one model parameter of the AI / ML model based on the second dataset.

[0006] In a second aspect, a terminal device is provided. The terminal device may include at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the terminal device to at least: receive a query message from a network device, the query message being used to query whether an AI / ML model at the terminal device has the capability to provide inference results for LOS / NLOS classification with confidence levels; based on determining that the AI / ML model has this capability, send the AI / ML model and a dataset to the network device, wherein the dataset includes a set of data samples associated with a label set, the label set being real labels indicating LOS paths or NLOS paths; and receive at least one model parameter of the updated AI / ML model obtained by updating the AI / ML model from the network device.

[0007] In a third aspect, a method is provided. This method may include: receiving, at a network device, an AI / ML model and a first dataset from an end device, wherein the AI / ML model has the ability to provide inference results for LOS / NLOS classification with confidence levels, and the first dataset includes a first set of data samples associated with a first label set, the first label set being true labels indicating LOS paths or NLOS paths; obtaining a second dataset based on multiple unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set, wherein the second dataset includes a second set of data samples associated with a second label set, the second label set being pseudo-labels indicating LOS paths or NLOS paths; and updating at least one model parameter of the AI / ML model based on the second dataset.

[0008] In a fourth aspect, a method is provided. This method may include: receiving a query message at a terminal device from a network device, the query message being used to query whether an AI / ML model at the terminal device has the ability to provide inference results for LOS / NLOS classification with a confidence level; based on determining that the AI / ML model has this ability, sending the AI / ML model and a dataset to the network device, wherein the dataset includes a set of data samples associated with a label set, the label set being real labels indicating LOS or NLOS paths; and receiving at least one model parameter of the updated AI / ML model obtained by updating the AI / ML model from the network device.

[0009] In a fifth aspect, an apparatus is provided. The apparatus may include: components for receiving an AI / ML model and a first dataset from an end device at a network device, wherein the AI / ML model has the capability to provide inference results for LOS / NLOS classification with confidence levels, and the first dataset includes a first set of data samples associated with a first set of labels, the first set of labels being true labels indicating LOS paths or NLOS paths; components for acquiring a second dataset based on a plurality of unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first set of labels, wherein the second dataset includes a second set of data samples associated with a second set of labels, the second set of labels being pseudo-labels indicating LOS paths or NLOS paths; and components for updating at least one model parameter of the AI / ML model based on the second dataset.

[0010] In a sixth aspect, an apparatus is provided. The apparatus may include: components for receiving a query message from a network device at a terminal device, the query message being for querying whether an AI / ML model at the terminal device has the capability to provide inference results for LOS / NLOS classification with confidence levels; components for sending an AI / ML model and a dataset to the network device based on determining that the AI / ML model has this capability, wherein the dataset includes a set of data samples associated with a label set, the label set being real labels indicating LOS paths or NLOS paths; and components for receiving at least one model parameter of an updated AI / ML model obtained by updating the AI / ML model from the network device.

[0011] In a seventh aspect, a non-transitory computer-readable medium is provided, including program instructions for causing a device to perform the method according to at least the third or fourth aspect.

[0012] In an eighth aspect, a computer program is provided, including instructions that, when executed by a device, cause the device to at least: receive an AI / ML model and a first dataset from a terminal device, wherein the AI / ML model has the capability to provide inference results for LOS / NLOS classification with confidence levels, and the first dataset includes a first set of data samples associated with a first set of labels, the first set of labels being true labels indicating LOS paths or NLOS paths; acquire a second dataset based on a plurality of unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first set of labels, wherein the second dataset includes a second set of data samples associated with a second set of labels, the second set of labels being pseudo labels indicating LOS paths or NLOS paths; and update at least one model parameter of the AI / ML model based on the second dataset.

[0013] In a ninth aspect, a computer program is provided, including instructions that, when executed by a device, cause the device to at least: receive a query message from a network device, the query message being used to query whether an AI / ML model at a terminal device has the capability to provide inference results for LOS / NLOS classification with confidence levels; based on determining that the AI / ML model has this capability, send the AI / ML model and a dataset to the network device, wherein the dataset includes a set of data samples associated with a label set, the label set being real labels indicating LOS paths or NLOS paths; and receive at least one model parameter of the updated AI / ML model obtained by updating the AI / ML model from the network device.

[0014] In a tenth aspect, a network device is provided. The network device may include a receiving circuitry configured to receive from an end device an AI / ML model and a first dataset, wherein the AI / ML model has the capability to provide inference results for LOS / NLOS classification with confidence levels, and the first dataset includes a first set of data samples associated with a first label set, the first label set being true labels indicating LOS paths or NLOS paths; an acquisition circuitry configured to acquire a second dataset based on a plurality of unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set, wherein the second dataset includes a second set of data samples associated with a second label set, the second label set being pseudo-labels indicating LOS paths or NLOS paths; and an updating circuitry configured to update at least one model parameter of the AI / ML model based on the second dataset.

[0015] In an eleventh aspect, a terminal device is provided. The terminal device may include a first receiving circuitry configured to receive a query message from a network device, the query message being used to query whether an AI / ML model at the terminal device has the capability to provide inference results for LOS / NLOS classification with confidence levels; a transmitting circuitry configured to transmit an AI / ML model and a dataset to the network device based on determining that the AI / ML model has this capability, wherein the dataset includes a set of data samples associated with a label set, the label set being real labels indicating LOS paths or NLOS paths; and a second receiving circuitry configured to receive from the network device at least one model parameter of an updated AI / ML model obtained by updating the AI / ML model.

[0016] It should be understood that the summary portion is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0017] Some exemplary embodiments will now be described with reference to the accompanying drawings, in which:

[0018] Figure 1 An example network environment in which example embodiments of this disclosure may be implemented is shown;

[0019] Figure 2 An example signaling process for AI / ML model updating according to some example embodiments of this disclosure is shown;

[0020] Figure 3 Example architectures of AI / ML models according to some example embodiments of this disclosure are shown;

[0021] Figure 4 Example functions of the UE and Location Management Function (LMF) according to some example embodiments of this disclosure are shown;

[0022] Figure 5 Another example signaling process for AI / ML model updates according to some example embodiments of this disclosure is shown;

[0023] Figure 6 An example flowchart of the AI / ML model update process according to some example embodiments of this disclosure is shown;

[0024] Figure 7 Another example flowchart of the AI / ML model update process according to some example embodiments of this disclosure is shown;

[0025] Figure 8 Simplified example block diagrams of devices suitable for implementing embodiments of the present disclosure are shown; and

[0026] Figure 9 Example block diagrams of example computer-readable media according to some example embodiments of the present disclosure are shown.

[0027] In all the accompanying drawings, the same or similar reference numerals denote the same or similar elements. Detailed Implementation

[0028] The principles of this disclosure will now be described with reference to some exemplary embodiments. It should be understood that these embodiments are described for illustrative purposes only and to assist those skilled in the art in understanding and implementing this disclosure, and do not imply any limitation on the scope of this disclosure. The disclosure described herein can be implemented in various ways other than those described below.

[0029] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0030] References to "an embodiment," "embodiment," "example embodiment," etc., in this disclosure indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment must include that particular feature, structure, or characteristic. Furthermore, these phrases do not necessarily refer to the same embodiment. Moreover, when a particular feature, structure, or characteristic is described in connection with an embodiment, whether explicitly described or not, those skilled in the art can combine such feature, structure, or characteristic with other embodiments to achieve the same result.

[0031] It is understood that while the terms “first” and “second” may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.

[0032] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments. As used herein, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” are intended to also include the plural forms. It will be further understood that the terms “comprising,” “including,” “having,” “containing,” and / or “containing,” when used herein, specify the presence of stated features, elements, and / or components, etc., but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof. As used herein, “at least one of the following: ” and “at least one of ” and similar wording, wherein the list of two or more elements is connected by “and” or “or”, means at least any one of these elements, or at least any two or more of these elements, or at least all of these elements.

[0033] As used in this application, the term "circuit system" may refer to one or more of the following: (a) Pure hardware circuit implementations (such as implementations only in analog and / or digital circuit systems) and (b) A combination of hardware circuitry and software, such as (if applicable): (i) A combination of analog and / or (multiple) digital hardware circuits and software / firmware, and (ii) Any part of the (multiple) hardware processors works in conjunction with software (including (multiple) digital signal processors), software, and (multiple) memories to enable the device (such as a mobile phone or server) to perform various functions, and (c) (a) multiple) hardware circuits and / or (a) multiple) processors, such as (a) multiple) microprocessors or a portion thereof, which require software (e.g., firmware) for operation, but may be absent when not required for operation.

[0034] This definition of circuit system applies to all uses of the term in this application, including in any claim. As a further example, as used in this application, the term circuit system also covers only hardware circuitry or a processor (or processors) or a portion of hardware circuitry or a processor and its accompanying software and / or firmware implementation. The term circuit system also covers (e.g., and if applicable to a particular claim element) baseband integrated circuits or processor integrated circuits for mobile devices or similar integrated circuits in servers, cellular network devices, or other computing or network devices.

[0035] As used herein, the term "communication network" refers to a network that conforms to any suitable communication standard, such as Long Term Evolution (LTE), LTE-A, Wideband Code Division Multiple Access (WCDMA), High-Speed ​​Packet Access (HSPA), Narrowband Internet of Things (NB-IoT), etc. Furthermore, communication between terminal devices and network devices in a communication network can be performed according to any suitable generation of communication protocols, including but not limited to third-generation (3G), fourth-generation (4G), 4.5G, fifth-generation (5G) communication protocols and / or higher. Embodiments of this disclosure can be applied to various communication systems. Given the rapid development of communications, there will naturally be communication technologies and systems that embody future types of this disclosure. The scope of this disclosure should not be construed as limited to the systems described above.

[0036] As used herein, the term "network device" refers to a node in a communication network through which terminal devices access the network and receive services. Network devices can refer to base stations (BS) or access points (APs), such as Node B (NodeB or NB), evolved Node B (eNodeB or eNB), NR NB (also known as gNB), Remote Radio Unit (RRU), Radio Header Terminal (RH), Remote Radio Header Terminal (RRH), relays, low-power nodes (such as femtocells, picocells, etc.), depending on the terminology and technology used.

[0037] The term "terminal device" refers to any terminal device capable of wireless communication. By way of example and not limitation, a terminal device may also be referred to as a communication device, user equipment (UE), subscriber station (SS), portable subscriber station, mobile station (MS), or access terminal (AT). Terminal devices may include, but are not limited to, mobile phones, cellular phones, smartphones, Voice over IP (VoIP) phones, wireless local loop phones, tablets, wearable terminal devices, personal digital assistants (PDAs), portable computers, desktop computers, image capture terminal devices (such as digital cameras), gaming terminal devices, music storage and playback devices, in-vehicle wireless terminal devices, wireless endpoints, mobile stations, laptop embedded devices (LEEs), laptop-mounted devices (LMEs), USB dongles, smart devices, wireless customer premises equipment (CPEs), Internet of Things (IoT) devices, watches or other wearable devices, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in industrial and / or automated processing chain environments), consumer electronics devices, commercially operating equipment, relay nodes, integrated access and backhaul (IAB) nodes, and / or industrial wireless networks, etc. In the following description, the terms "terminal device," "communication device," "terminal," "user equipment," and "UE" are used interchangeably.

[0038] As used herein, the terms “resource,” “transmission resource,” “resource block,” “physical resource block” (PRB), “uplink (UL) resource,” or “downlink (DL) resource” can refer to any resource used to perform communication, such as communication between terminal devices and network devices, including resources in the time domain, frequency domain, spatial domain, code domain, a combination of more than one domain, or any other resource enabling communication. In the following, resources in the time domain (such as subframes) will be used as examples of transmission resources to describe some exemplary embodiments of this disclosure. It should be noted that the exemplary embodiments of this disclosure are equally applicable to other resources in other domains.

[0039] As mentioned above, the scope of machine learning (ML) enabled localization can be based on RF fingerprints (e.g., CIR, etc.), which is one of the three research projects (SIDs) in 3GPP Release 18, as shown in Table 1. Table 1

[0040] Semi-supervised learning (SSL) can also be an effective way to solve this challenging task. At the 3GPP RAN1#110 meeting, SSL emerged as a hot new topic of research because it can provide pseudo-labels for large-scale unlabeled data. It is a promising solution for AI localization tasks where there is a small amount of labeled data (e.g., {CIR / PDP, location labels}) and enough available unlabeled data {CIR / PDP; N / A locations}. This is illustrated in Table 2. Table 2

[0041] AI / ML localization is data-driven. Training AI / ML models heavily relies on large amounts of labeled data. However, in the real world, collecting large amounts of labeled data is difficult and costly, while unlabeled data containing only CIRs is relatively easy to obtain. Therefore, using a combination of a small amount of labeled data and a large amount of unlabeled data to train the model's SSL becomes a reasonable solution to address this challenge.

[0042] In semi-supervised learning, one approach is to first train a model on a small labeled dataset, and then use that model to generate pseudo-labels on unlabeled data. The model can be fine-tuned / retrained by combining the original labeled data with unlabeled data of high confidence. These processes can be iteratively performed to find the optimal model.

[0043] However, using only unlabeled data with high confidence may not yield deterministic inference improvements, and in some cases may even lead to worse inference results, due to ignoring the data distribution at the UE location, such as the ratio between LOS and NLOS. When the selected unlabeled dataset differs from the actual UE distribution, the model will be trained to overfit on the selected unlabeled dataset, resulting in performance degradation on the actual dataset at the UE location.

[0044] Therefore, a solution is needed for training AI / ML models using the LOS / NLOS ratio, taking into account the actual deployment environment of the terminal device. An example embodiment of this disclosure provides a solution for SSL updates for AI. It allows network devices (such as LMF / over-the-top (OTT) devices) to support updating the UE's AI localization model with manually labeled data. The proposed solution includes two ideas. The first idea is that the AI / ML-assisted localization model provides an estimated confidence level as a secondary output in addition to its primary output of the LOS / NLOS classification index. The second idea is that, not only based on the estimated confidence level on unlabeled data, the LMF / OTT can adjust the LOS / NLOS ratio of the CIR / PDP data in the selected unlabeled dataset according to the actual deployment environment and reselect a dataset with pseudo-labels for SSL training. Therefore, it involves new functionality in the UE / LMF / OTT and message exchange between entities.

[0045] The proposed solution defines a framework for SSL training methods for AI / ML model updates. This framework includes a method within AI-assisted localization. In addition to its primary LOS / NLOS classification output, the AI ​​model on the UE side can provide estimated confidence as a secondary output. LMF can assist the UE in selecting appropriate unlabeled data from OTT based on the AI ​​model's estimated confidence. Furthermore, the framework also includes the ability to adjust the LOS / NLOS ratio of CIR / PDP data according to the actual deployment environment based on the AI ​​model's estimated LOS / NLOS confidence, and to reselect a dataset with pseudo-labels (also known as artificial labels) for SSL training.

[0046] It should be understood that the above process steps may work together, partially work together, or be independent of each other in the operational flow described in the following section. By implementing embodiments of this disclosure, network devices can support updating the UE's AI / ML model with pseudo-labels, thereby providing higher LOS / NLOS classification accuracy.

[0047] For illustrative purposes, references will be made below. Figures 1 to 9This disclosure describes the principles and exemplary embodiments of AI / ML updates. However, it should be noted that these embodiments are intended to enable those skilled in the art to understand the inventive concept of this disclosure and implement the solutions presented herein, and are not intended to limit the scope of this application in any way.

[0048] refer to Figure 1 The illustration shows an example network environment 100 in which exemplary embodiments of the present disclosure may be implemented. The network environment 100 (which may be part of a communication network) includes a terminal device 102 and a network device 104.

[0049] like Figure 1 As shown, terminal device 102 may also be referred to as user equipment 102 or UE 102. Network device 104 may also be referred to as LMF 104. Network device 104 may also include OTT. Alternatively, LMF and OTT together may be referred to as network device 104. Network device 104 may also be referred to as network node 104. Terminal device 102 and network device 104 can communicate with each other. In some example embodiments, network device 104 may also be referred to as an access network device including at least one of LMF or OTT devices. In some example embodiments, network device 104 may also be referred to as a core network device including at least one of LMF or OTT devices.

[0050] refer to Figure 2 This illustrates an example signaling process 200 for AI / ML model updating according to some example embodiments of this disclosure. Reference will be made to... Figure 1 describe Figure 2 .

[0051] like Figure 2 As shown, network device 104 sends query message 206 (204), which queries whether AI / ML model 224 at terminal device 102 has the ability to provide inference results for LOS / NLOS classification with confidence levels. Terminal device 102 receives query message 206 (202) from network device 104.

[0052] Terminal device 102 determines (208) that AI / ML model 224 has this capability. Terminal device 102 sends (210) AI / ML model 224 and dataset (hereinafter also referred to as the first dataset) to network device 104. The first dataset includes a set of data samples associated with a label set, which is the real labels indicating LOS or NLOS paths. Network device 104 receives (212) AI / ML model 224 and the first dataset.

[0053] Network device 104 obtains (214) a second dataset based on multiple unlabeled data samples, AI / ML model 224, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set. The second dataset includes a second set of data samples associated with the second label set, which is a set of pseudo-labels indicating LOS or NLOS paths.

[0054] Network device 104 updates (216) at least one model parameter of AI / ML model 224 based on the second dataset. Network device 104 sends (220) the updated AI / ML model 224 at least one model parameter to terminal device 102. Terminal device 102 receives (218) the updated AI / ML model 224 at least one model parameter.

[0055] Details of the acquisition (214) process and the update (216) process will be referred to Figure 3-5 They will not be described here for the sake of simplicity.

[0056] By implementing Figure 2 In one embodiment, network devices can support updating the UE's AI / ML model with pseudo-labels. LMF can adjust the LOS / NLOS data ratio according to the actual deployment environment and reselect the dataset with pseudo-labels for SSL training, thereby providing higher LOS / NLOS classification accuracy.

[0057] refer to Figure 3 This illustrates an example architecture 300 of an AI / ML model according to some example embodiments of the present disclosure. As an example, Figure 3 An SSL model training framework for AI / ML-assisted localization to classify LOS / NLOS signals using CIR / PDP as model input is shown.

[0058] like Figure 3 As shown, architecture 300 includes the CIP / PDP dataset 302, the SSL AI / ML model 304, and the LOS-NLOS inference results 306. The input to the SSL AI / ML model 304 is the CIR / PDP vector collected from each TRP, with a dimension of [missing information]. The output of the SSL AI / ML model 304 is a LOS / NLOS confidence score of 306, not a hard value for the LOS / NLOS classification. Table 3 illustrates the internal architecture of the AI / ML-assisted localization model. Cross-entropy is used as the loss function. Table 3 Architecture of AI / ML-assisted localization model

[0059] refer to Figure 4This illustrates example functionality 400 of the UE and LMF according to some example embodiments of this disclosure. For example... Figure 4 As shown, block 402 illustrates the functionality on the UE side. A pre-trained AI / ML-assisted localization model 304 resides on the UE side, and the UE also stores a small amount of data 404 with real labels for model lifecycle management (LCM). Storage and computing power are limited on the UE side.

[0060] As another example, block 406 illustrates the functionality on the LMF side. The LMF collects unlabeled data from the UE and maintains a large-scale unlabeled dataset 408 to obtain an overview of the wireless environment. Since obtaining sufficient real labels is difficult, and considering the limited storage and computing power on the UE side, we can leverage the large-scale unlabeled dataset on the LMF side, along with sufficient storage and computing power, for model updates. By using SSL-based model training, the LMF can update the model with higher LOS / NLOS classification accuracy.

[0061] refer to Figure 5 This illustrates another example signaling process 500 for AI / ML model updating according to some example embodiments of this disclosure. Figure 5 This illustrates a scenario where the AI / ML model on the UE side has been pre-trained in a supervised manner using a labeled dataset, and the LMF504 maintains an unlabeled CIR / PDP dataset. Figure 5 UE 502 in the text can correspond to Figure 1 Terminal device 102 in the LMF 504, together with OTT 506, can correspond to Figure 1 Network device 104.

[0062] As an example, UE 502 can be pre-trained in a supervised manner using labeled datasets (508). LMF 504 can be prepared to monitor and manage (510) AI / ML models. OTT 506 can maintain (512) a dataset with unlabeled CIR / PDP samples.

[0063] Alternatively or concurrently, UE 502 may send message (514) (518) to LMF 504 to request an update of the model using an SSL method. LMF 504 may receive message (516) (518) from UE 502. Alternatively, LMF 504 may trigger (520) an update of the AI / ML model using an SSL method.

[0064] LMF 504 can send (524) query (526) to inquire whether the AI / ML model of UE 502 has the ability to provide AI / ML-assisted localization inference results (such as LOS / NLOS classification) with confidence. UE 502 can receive (522) query (526) from LMF 504.

[0065] If the AI / ML model of UE 502 can provide AI / ML-assisted localization inference results with confidence, then UE 502 can send (528) ACK (532) to LMF 504 along with its AI / ML-assisted localization model and a small-scale labeled dataset (e.g., CIR / PDP+ labels). LMF 504 can receive (530) ACK (532), the AI / ML model, and the small-scale labeled dataset from UE 502.

[0066] If the AI / ML model of UE 502 cannot provide AI / ML-assisted localization inference results with confidence, UE 502 may send (534) negative acknowledgment (NACK) (538) to LMF 504. LMF 504 may receive (536) NACK (538) from UE 502. UE 502 may terminate the process of updating the AI / ML model using the SSL method.

[0067] LMF 504 can refer to the unlabeled data selection criteria to select (542) unlabeled CIR / PDP data from the database in the OTT, for example, inference confidence for category LOS and category NLOS respectively.

[0068] Additionally or alternatively, LMF 504 can compute inference confidence and assign pseudo-labels to unlabeled CIR / PDP samples. For example, LOS / NLOS confidence for an unlabeled dataset can be obtained by feeding unlabeled data into a pre-trained AI / ML-assisted model with LOS / NLOS classification capabilities. Figure 3 As discussed earlier, the model's output can be a vector containing two values: the first being the confidence level of LOS (Low-Solution Scale) and the second being the confidence level of NLOS (Non-Low-Solution Scale). For example, if the output is (a=98%, b=2%), this means the path is likely a LOS path. Pseudo-labels can then be added to the unlabeled data regarding the output LOS / NLOS confidence levels.

[0069] LMF 504 can sort the unlabeled data selected at position 542 (544) separately for both category LOS and category NLOS. LMF 504 can select the top samples from each category. LMF 504 can construct new datasets using data from both categories combined with original CIR / PDP and artificial LOS / NLOS labels.

[0070] In some example embodiments, the LOS / NLOS ratio in the labeled dataset can indicate the radio environment of the UE location (e.g., the location of UE502). Maintaining the same LOS / NLOS ratio in the selected unlabeled dataset can help preserve the characteristics of the UE 502's radio environment. Otherwise, there will be a drift between the two environments represented by the two datasets, and the LOS / NLOS accuracy will decrease due to this drift.

[0071] Therefore, let This indicates that samples selected from the unlabeled dataset can have their LOS / NLOS ratio obtained from the labeled dataset sent from UE 502 to LMF 504 / OTT 506. Here, the LOS path ratio is expressed as... And the proportion of the NLOS dataset is expressed as Then, you should choose... Unlabeled samples with LOS pseudo-labels and An unlabeled sample with NLOS pseudo-labels.

[0072] Based on the difference between the output LOS / NLOS confidence values, LMF 504 can rank samples to obtain the most confident LOS and NLOS paths. For example, for two outputs (a=98%, b=2%) and (a=60%, b=40%), the former output (a=98%, b=2%) has a higher probability of being a LOS path, meaning that samples with a larger difference between the two output values ​​have higher inference confidence.

[0073] Therefore, in general, the output LOS / NLOS confidence value can be expressed as ,but and ,in This represents the inference confidence of the LOS path. This represents the inference confidence of the NLOS path. Regarding the LOS / NLOS ratio in the labeled dataset, LMF 504 can be selected based on the highest confidence level. Value The number of unlabeled samples and those with the highest Value This is used to construct an artificial dataset of SSL with pseudo-labels from unlabeled data.

[0074] LMF 504 may use an SSL method to update (546) at least one model parameter of the model with the selected unlabeled dataset. After completing the update (546), LMF 504 may send (550) at least one model parameter of the updated AI / ML assisted localization model back to UE 502. UE 502 may receive (548) at least one model parameter of the updated AI / ML assisted localization model.

[0075] The dataset is trained using the selected SSL model with pseudo-labels. The original model can be trained using the same supervision method mentioned at the beginning of this section. This training process can be iterated on the LMF side to obtain higher LOS / NLOS classification accuracy.

[0076] Some embodiments of this disclosure also provide simulation evaluations of the proposed solution. In these simulations, an AI / ML-assisted LOS / NLOS classification model is trained in a supervised manner with one thousand labeled data samples. The same model is trained using three pseudo-label-based semi-supervised methods. The first method is SSL Method 1: unlabeled data is randomly selected without considering any confidence level or LOS / NLOS ratio. The second method is SSL Method 2: unlabeled data is selected only with respect to the output inference confidence level (i.e., the absolute value of α), without considering the LOS / NLOS ratio. The third method is SSL Method 3: unlabeled data is selected with respect to both the output inference confidence level and the LOS / NLOS ratio (i.e., α and β). The simulation results are shown in Table 4. Table 4. Comparison of error rates among different semi-supervised training methods

[0077] Table 4 shows that simulation results indicate no improvement in classification accuracy for Method 1, trained with randomly selected data, validating the importance of unlabeled data selection. For Methods 2 and 3, improvements in classification accuracy are observed. Method 3 exhibits a larger performance increment and validates the importance of confidence ranking by category (LOS and NLOS) and adjusting the LOS / NLOS ratio in SSL artificial dataset reconstruction. Therefore, it can be seen that the proposed solution can provide higher LOS / NLOS classification accuracy.

[0078] refer to Figure 6 It illustrates an example flowchart of an AI / ML model update process 600 according to some example embodiments of this disclosure. (See also...) Figure 1 describe Figure 6 .

[0079] At point 602, network device 104 receives an AI / ML model and a first dataset from terminal device 102. The AI / ML model has the ability to provide inference results for LOS / NLOS classification with confidence levels. The first dataset includes a first set of data samples associated with a first set of labels, which are the true labels indicating LOS or NLOS paths.

[0080] At 604, network device 104 acquires a second dataset based on multiple unlabeled data samples, an AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set. The second dataset includes a second set of data samples associated with the second label set, which consists of pseudo-labels indicating either LOS or NLOS paths. At 606, network device 104 updates at least one model parameter of the AI / ML model based on the second dataset.

[0081] In some example embodiments, network device 104 may send a query message to inquire about the capabilities of the AI / ML model at terminal device 102. The AI / ML model and a first dataset may be sent together with an ACK in response to the query message.

[0082] In some example embodiments, the sending of a query message may be based on receiving a request message from terminal device 102, the request message being used to request an update of the AI / ML model at network device 104 based on SSL.

[0083] In some example embodiments, the sending of the query message may be based on determining at network device 104 that an SSL-based update of the AI / ML model has been triggered at network device 104. In some example embodiments, network device 104 may send the updated AI / ML model to terminal device 102.

[0084] In some example embodiments, network device 104 may obtain a second dataset by: inputting multiple unlabeled data samples into an AI / ML model, determining multiple pseudo-labels and multiple confidence levels for the multiple unlabeled data samples; and selecting a second data sample set from the multiple unlabeled data samples based on the multiple pseudo-labels, multiple confidence levels, and ratios.

[0085] In some example embodiments, network device 104 may select a second data sample set by: determining multiple LOS pseudo-labels from multiple pseudo-labels that indicate LOS paths; sorting the first unlabeled data sample set based on the LOS confidence level of a first unlabeled data sample set associated with the multiple LOS pseudo-labels; and selecting a first number of unlabeled data samples from the sorted first unlabeled data sample set as a first part of the second data sample set, wherein the first number of unlabeled data samples has a higher confidence level than other unlabeled data samples in the sorted first unlabeled data sample set.

[0086] In some example embodiments, network device 104 may select a second data sample set by: determining a plurality of NLOS pseudo-labels from a plurality of pseudo-labels that indicate an NLOS path; sorting the second unlabeled data sample set based on the NLOS confidence level of the second unlabeled data sample set associated with the plurality of NLOS pseudo-labels; and selecting a second number of unlabeled data samples from the sorted second unlabeled data sample set as a second part of the second data sample set, wherein the second number of unlabeled data samples has a higher confidence level than other unlabeled data samples in the sorted second unlabeled data sample set.

[0087] In some example embodiments, the ratio of the first quantity to the second quantity may be substantially equal to the ratio of the number of tags indicating LOS paths to the number of tags indicating NLOS paths in the first tag set. In some example embodiments, network device 104 may determine the first quantity and the second quantity based on a predetermined quantity and ratio.

[0088] In some example embodiments, the input to the AI / ML model may include data samples. The output of the AI / ML model may include at least one of a first value indicating the confidence level of a data sample associated with a LOS path or a second value indicating the confidence level of a data sample associated with an NLOS path. The confidence level of a data sample associated with a LOS path may be determined based on the first value minus the second value. The confidence level of a data sample associated with an NLOS path may be determined based on the second value minus the first value.

[0089] In some example embodiments, the network device may be a first network device, and multiple unlabeled data samples may be stored in a second network (such as...). Figure 5 In an OTT 506 device or a first network device. In some example embodiments, at least one of the first data sample set or the second data sample set may include a CIR data sample or a PDP data sample. In some example embodiments, the network device may perform LMF.

[0090] refer to Figure 7This illustrates another example flowchart of an AI / ML model update process 700 according to some example embodiments of this disclosure. (See also...) Figure 1 describe Figure 7 .

[0091] At 702, terminal device 102 receives a query message from network device 104. The query message is used to query whether the AI / ML model at terminal device 102 has the ability to provide inference results for LOS / NLOS classification with confidence levels.

[0092] At 704, terminal device 102 sends the AI / ML model and dataset to network device 104 based on the determination that the AI / ML model possesses this capability. The dataset includes a set of data samples associated with a label set, which is a set of real labels indicating LOS or NLOS paths. At 706, terminal device 102 receives from network device 104 at least one model parameter of the updated AI / ML model, obtained by updating the AI / ML model.

[0093] In some example embodiments, the AI / ML model and dataset may be sent together with the ACK. In some example embodiments, the terminal device may send a NACK to the network device 104 based on the determination that the AI / ML model does not have this capability.

[0094] In some example embodiments, before receiving a query message, the terminal device 104 may send a request message to the network device 104, the request message being used to request an update of the AI / ML model at the network device based on SSL.

[0095] In some example embodiments, the dataset may be a first dataset. At least one model parameter of the updated AI / ML model may be updated based on a second dataset. The second dataset may be obtained based on multiple unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the label set.

[0096] In some example embodiments, the input to the AI / ML model may include data samples. The output of the AI / ML model may include at least one of a first value indicating the confidence level of a data sample associated with a LOS path or a second value indicating the confidence level of a data sample associated with an NLOS path. The confidence level of a data sample associated with a LOS path may be determined based on the first value minus the second value. The confidence level of a data sample associated with an NLOS path may be determined based on the second value minus the first value.

[0097] In some example embodiments, at least one of the first data sample set or the second data sample set may include a CIR data sample or a PDP data sample. In some example embodiments, the network device may perform LMF.

[0098] By implementing embodiments of methods 600 and 700, network devices can support updating the AI / ML model of the UE with pseudo-labels, thereby providing higher LOS / NLOS classification accuracy.

[0099] In some example embodiments, the apparatus capable of performing method 600 (e.g., network device 104) may include components for performing the corresponding steps of method 600. The components may be implemented in any suitable form. For example, the components may be implemented in a circuit system or a software module.

[0100] In some example embodiments, the apparatus may include: components for receiving an AI / ML model and a first dataset from an end device at a network device, wherein the AI / ML model has the ability to provide inference results for LOS / NLOS classification with confidence levels, and the first dataset includes a first set of data samples associated with a first set of labels, the first set of labels being true labels indicating LOS paths or NLOS paths; components for obtaining a second dataset based on a plurality of unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first set of labels, wherein the second dataset includes a second set of data samples associated with a second set of labels, the second set of labels being pseudo labels indicating LOS paths or NLOS paths; and components for updating at least one model parameter of the AI / ML model based on the second dataset.

[0101] In some example embodiments, the apparatus may include a component for sending a query message to a terminal device, the query message being used to query the capabilities of an AI / ML model at the terminal device, wherein the AI / ML model and a first dataset are sent together with an ACK in response to the query message.

[0102] In some example embodiments, the sending of the query message may be based on one of the following: receiving a request message from a terminal device, the request message being used to request an update of the AI / ML model at the network device based on SSL; or determining at the network device that an update of the AI / ML model at the network device based on SSL is triggered.

[0103] In some example embodiments, the apparatus may also include components for sending updated AI / ML models to terminal devices.

[0104] In some example embodiments, the components for obtaining the second dataset may include: components for determining multiple pseudo-labels and multiple confidence levels for multiple unlabeled data samples by inputting multiple unlabeled data samples into an AI / ML model; and components for selecting a second data sample set from the multiple unlabeled data samples based on the multiple pseudo-labels, multiple confidence levels, and ratios.

[0105] In some example embodiments, the components for selecting a second data sample set may include: components for determining a plurality of LOS pseudo-labels from a plurality of pseudo-labels that indicate a LOS path; components for sorting a first unlabeled data sample set based on the LOS confidence level of a first unlabeled data sample set associated with the plurality of LOS pseudo-labels; and components for selecting a first number of unlabeled data samples from the sorted first unlabeled data sample set as a first part of the second data sample set, wherein the first number of unlabeled data samples have a higher confidence level than other unlabeled data samples in the sorted first unlabeled data sample set.

[0106] In some example embodiments, the components for selecting a second data sample set may include: components for determining a plurality of NLOS pseudo-labels from a plurality of pseudo-labels that indicate an NLOS path; components for sorting the second unlabeled data sample set based on the NLOS confidence level of the second unlabeled data sample set associated with the plurality of NLOS pseudo-labels; and components for selecting a second number of unlabeled data samples from the sorted second unlabeled data sample set as a second part of the second data sample set, wherein the second number of unlabeled data samples has a higher confidence level than other unlabeled data samples in the sorted second unlabeled data sample set.

[0107] In some example embodiments, the ratio of the first quantity to the second quantity may be substantially equal to the ratio of the number of tags indicating LOS paths to the number of tags indicating NLOS paths in the first tag set.

[0108] In some example embodiments, the apparatus may also include components for determining a first quantity and a second quantity based on a predetermined quantity and ratio.

[0109] In some example embodiments, the input to the AI / ML model may include data samples. The output of the AI / ML model may include at least one of a first value indicating the confidence level of a data sample associated with a LOS path or a second value indicating the confidence level of a data sample associated with an NLOS path. The confidence level of a data sample associated with a LOS path may be determined based on the first value minus the second value. The confidence level of a data sample associated with an NLOS path may be determined based on the second value minus the first value.

[0110] In some example embodiments, at least one of the first data sample set or the second data sample set may include a CIR data sample; or a PDP data sample.

[0111] In some example implementations, the network device may perform LMF.

[0112] In some example embodiments, the apparatus may also include components for performing other steps in some example embodiments of method 600. In some example embodiments, the components include at least one processor and at least one memory including computer program code, the at least one memory and the computer program code being configured, together with the at least one processor, to enable the execution of the apparatus.

[0113] In some example embodiments, the apparatus capable of performing method 700 (e.g., terminal device 102) may include components for performing the corresponding steps of method 700. The components may be implemented in any suitable form. For example, the components may be implemented in a circuit system or a software module.

[0114] In some example embodiments, the apparatus may include: a component for receiving a query message from a network device at a terminal device, the query message being for querying whether an AI / ML model at the terminal device has the ability to provide inference results for LOS / NLOS classification with confidence levels; a component for sending an AI / ML model and a dataset to the network device based on determining that the AI / ML model has this ability, wherein the dataset includes a set of data samples associated with a label set, the label set being real labels indicating LOS paths or NLOS paths; and a component for receiving at least one model parameter of the updated AI / ML model obtained by updating the AI / ML model from the network device.

[0115] In some example implementations, the AI / ML model and dataset may be sent together with the ACK.

[0116] In some example embodiments, the apparatus may also include a component for sending a NACK to the network device based on the determination that the AI / ML model does not have the capability.

[0117] In some example embodiments, the apparatus may also include a component for sending a request message to the network device before receiving a query message, the request message being used to request an update of the AI / ML model at the network device based on SSL.

[0118] In some example embodiments, the dataset may be a first dataset. At least one model parameter of the updated AI / ML model may be updated based on a second dataset. The second dataset may be obtained based on multiple unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the label set.

[0119] In some example embodiments, the input to the AI / ML model may include data samples. The output of the AI / ML model may include at least one of a first value indicating the confidence level of a data sample associated with a LOS path or a second value indicating the confidence level of a data sample associated with an NLOS path. The confidence level of a data sample associated with a LOS path may be determined based on the first value minus the second value. The confidence level of a data sample associated with an NLOS path may be determined based on the second value minus the first value.

[0120] In some example embodiments, at least one of the first data sample set or the second data sample set may include a CIR data sample; or a PDP data sample.

[0121] In some example implementations, the network device may perform LMF.

[0122] In some example embodiments, the apparatus may also include components for performing other steps in some example embodiments of method 700. In some example embodiments, the components include at least one processor and at least one memory including computer program code, the at least one memory and the computer program code being configured, together with the at least one processor, to enable the execution of the apparatus.

[0123] refer to Figure 8 The diagram illustrates an example simplified block diagram of a device 800 suitable for implementing embodiments of the present disclosure. Device 800 can be provided to implement a communication device, such as... Figure 1 The terminal device 102 shown is illustrated. As shown, device 800 includes one or more processors 810, one or more memories 820 that can be coupled to processor 810, and one or more communication modules 840 that can be coupled to processor 810.

[0124] The communication module 840 is used for bidirectional communication. The communication module 840 has at least one antenna to facilitate communication. The communication interface can represent any interface necessary for communication with other network elements; for example, the communication interface can be a wireless or wired interface to other network elements, or a software-based interface for communication.

[0125] Processor 810 can be any type suitable for a local technology network and can include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture, as non-limiting examples. Device 800 can have multiple processors, such as application-specific integrated circuit chips that are time-dependent on a clock synchronized with the main processor.

[0126] Memory 820 may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, read-only memory (ROM) 824, electrically programmable read-only memory (EPROM), flash memory, hard disk, optical disc (CD), digital video disc (DVD), and other magnetic and / or optical storage. Examples of volatile memories include, but are not limited to, random access memory (RAM) 822 and other volatile memories that do not persist during power outages.

[0127] Computer program 830 includes computer-executable instructions that are executed by the associated processor 810. Program 830 may be stored in ROM 824. Processor 810 may perform any appropriate actions and processes by loading program 830 into RAM 822.

[0128] The embodiments of this disclosure can be implemented by a program, enabling device 800 to perform as described in the reference. Figures 2 to 7 Any process discussed in this disclosure. Embodiments of this disclosure may also be implemented by hardware or by a combination of software and hardware.

[0129] In some example embodiments, program 830 may be tangibly contained in a computer-readable medium, which may be included in device 800 (such as in memory 820) or in other storage devices accessible to device 800. Device 800 may load program 830 from the computer-readable medium into RAM 822 for execution. The computer-readable medium may include any type of tangible non-volatile storage, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc. Figure 9 An example of a computer-readable medium 900 in the form of a CD or DVD is shown. The computer-readable medium has a program 830 stored thereon.

[0130] In general, the various embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are illustrated and described as block diagrams, flowcharts, or other illustrated representations, it should be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof, as non-limiting examples.

[0131] This disclosure also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as those included in program modules, which are executed in a device on a target real or virtual processor to perform the functions described above. Figure 6 or Figure 7 Method 600 or 700 is described. Generally, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. The functionality of a program module can be combined or split among program modules as needed in various embodiments. The machine-executable instructions of a program module can be executed on a local or distributed device. In a distributed device, a program module can reside on both local and remote storage media.

[0132] Program code used to perform the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0133] In the context of this disclosure, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc.

[0134] Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination thereof. More specific examples of computer-readable storage media will include electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. The term “non-transient” as used herein is a limitation on the medium itself (i.e., tangible, not signaling), and not a limitation on the persistence of data storage (e.g., RAM and ROM).

[0135] Furthermore, although the operations are depicted in a specific order, this should not be construed as requiring these operations to be performed in the specific order or sequential sequence shown, or to perform all illustrated operations to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features described in the context of individual embodiments may also be combined and implemented in a single embodiment. Conversely, the various features described in the context of a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination.

[0136] Although this disclosure has been described in language specific to structural features and / or method actions, it should be understood that this disclosure as defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as examples of implementing the claims.

Claims

1. A network device, comprising: At least one processor; as well as At least one memory storing instructions that, when executed by the at least one processor, cause the network device to at least: The terminal device receives an artificial intelligence (AI) / machine learning (ML) model and a first dataset, wherein the AI / ML model has the ability to provide inference results for line-of-sight (LOS) / non-line-of-sight (NLOS) classification with confidence levels, and the first dataset includes a first set of data samples associated with a first set of labels, the first set of labels being true labels indicating LOS paths or NLOS paths; A second dataset is obtained based on multiple unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set. The second dataset includes a second set of data samples associated with the second label set, which is a set of pseudo-labels indicating LOS or NLOS paths. as well as Update at least one model parameter of the AI / ML model based on the second dataset.

2. The network device according to claim 1, wherein the network device is further configured to: A query message is sent to the terminal device to query the capabilities of the AI / ML model at the terminal device, wherein the AI / ML model and the first dataset are sent together with an acknowledgment (ACK) in response to the query message.

3. The network device according to claim 2, wherein the sending of the query message is based on one of the following: Receive a request message from the terminal device, the request message being used to request an update of the AI / ML model at the network device based on semi-supervised learning (SSL); or At the network device, a trigger is determined to update the AI / ML model at the network device based on the SSL.

4. The network device according to any one of claims 1 to 3, wherein the network device is further configured to: The updated AI / ML model is sent to the terminal device.

5. The network device according to any one of claims 1 to 4, wherein the network device is configured to acquire the second dataset by: By inputting the multiple unlabeled data samples into the AI / ML model, multiple pseudo-labels and multiple confidence levels are determined for the multiple unlabeled data samples; and Based on the plurality of pseudo-labels, the plurality of confidence levels, and the ratio, the second data sample set is selected from the plurality of unlabeled data samples.

6. The network device of claim 5, wherein the network device is configured to select the second data sample set in the following manner: From the plurality of pseudo-labels, determine a plurality of LOS pseudo-labels that indicate LOS paths; Based on the LOS confidence level of the first unlabeled data sample set associated with the plurality of LOS pseudo-labels, the first unlabeled data sample set is sorted; and A first number of unlabeled data samples are selected from the sorted first unlabeled data sample set as the first part of the second data sample set, wherein the first number of unlabeled data samples have a higher confidence level than other unlabeled data samples in the sorted first unlabeled data sample set.

7. The network device according to claim 5 or 6, wherein the network device is configured to select the second data sample set in the following manner: From the plurality of pseudo-labels, determine a plurality of NLOS pseudo-labels that indicate NLOS paths; The second unlabeled data sample set is sorted based on the NLOS confidence level associated with the plurality of NLOS pseudo-labels; and A second number of unlabeled data samples are selected from the sorted second unlabeled data sample set as the second part of the second data sample set, wherein the second number of unlabeled data samples have a higher confidence level than the other unlabeled data samples in the sorted second unlabeled data sample set.

8. The network device of claim 7, wherein the ratio of the first quantity to the second quantity is substantially equal to the ratio of the number of tags indicating LOS paths to the number of tags indicating NLOS paths in the first tag set.

9. The network device according to claim 7 or 8, wherein the network device is further configured to: The first quantity and the second quantity are determined based on the predetermined quantity and the ratio.

10. The network device according to any one of claims 1 to 9, wherein: The input to the AI / ML model includes data samples; The output of the AI / ML model includes at least one of the following: a first value indicating the confidence level of the data sample associated with the LOS path, or a second value indicating the confidence level of the data sample associated with the NLOS path; The confidence level of the data sample associated with the LOS path is determined based on the first value minus the second value; and The confidence level of the data sample associated with the NLOS path is determined based on the second value minus the first value.

11. The network device according to any one of claims 1 to 10, wherein the network device is a first network device, and the plurality of unlabeled data samples are stored in a second network device or the first network device.

12. The network device according to any one of claims 1 to 11, wherein at least one of the first data sample set or the second data sample set comprises: Channel impulse response (CIR) data samples; or Power delay distribution (PDP) data sample.

13. The network device according to any one of claims 1 to 12, wherein the network device performs a location management function (LMF).

14. A terminal device, comprising: At least one processor; as well as At least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the terminal device to at least: Receive a query message from a network device, the query message being used to query whether the artificial intelligence (AI) / machine learning (ML) model at the terminal device has the ability to provide inference results for line-of-sight (LOS) / non-line-of-sight (NLOS) classification with confidence levels; Based on the determination that the AI / ML model has the aforementioned capability, the AI / ML model and dataset are sent to the network device, wherein the dataset includes a set of data samples associated with a tag set, the tag set being real labels indicating LOS paths or NLOS paths; as well as Receive at least one model parameter from the network device, obtained by updating the AI / ML model.

15. The terminal device of claim 14, wherein the AI / ML model and the dataset are sent together with an acknowledgment (ACK).

16. The terminal device according to claim 14 or 15, wherein the terminal device is further configured to: Based on the determination that the AI / ML model does not have the aforementioned capability, a negative acknowledgment (NACK) is sent to the network device.

17. The terminal device according to any one of claims 14 to 16, wherein the terminal device is further configured to: Before receiving the query message, a request message is sent to the network device, the request message being used to request that the AI / ML model be updated at the network device based on semi-supervised learning (SSL).

18. The terminal device according to any one of claims 14 to 17, wherein the dataset is a first dataset, and wherein: The at least one model parameter of the updated AI / ML model is updated based on the second dataset; as well as The second dataset was obtained based on multiple unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the label set.

19. The terminal device according to any one of claims 14 to 18, wherein: The input to the AI / ML model includes data samples; The output of the AI / ML model includes at least one of the following: a first value indicating the confidence level of the data sample associated with the LOS path, or a second value indicating the confidence level of the data sample associated with the NLOS path; The confidence level of the data sample associated with the LOS path is determined based on the first value minus the second value; and The confidence level of the data sample associated with the NLOS path is determined based on the second value minus the first value.

20. The terminal device according to any one of claims 14 to 19, wherein the data sample set comprises: Channel impulse response (CIR) data samples; or Power delay distribution (PDP) data sample.

21. The terminal device according to any one of claims 14 to 20, wherein the network device performs a location management function (LMF).

22. A method comprising: At the network device, an artificial intelligence (AI) / machine learning (ML) model and a first dataset are received from a terminal device, wherein the AI / ML model has the ability to provide inference results for line-of-sight (LOS) / non-line-of-sight (NLOS) classification with confidence levels, and the first dataset includes a first set of data samples associated with a first set of labels, the first set of labels being true labels indicating LOS paths or NLOS paths; A second dataset is obtained based on multiple unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set. The second dataset includes a second set of data samples associated with the second label set, which is a set of pseudo-labels indicating LOS or NLOS paths. as well as Update at least one model parameter of the AI / ML model based on the second dataset.

23. A method comprising: At the terminal device, a query message is received from the network device, the query message being used to query whether the artificial intelligence (AI) / machine learning (ML) model at the terminal device has the ability to provide inference results for line-of-sight (LOS) / non-line-of-sight (NLOS) classification with confidence levels; Based on the determination that the AI / ML model has the aforementioned capability, the AI / ML model and dataset are sent to the network device, wherein the dataset includes a set of data samples associated with a tag set, the tag set being real labels indicating LOS paths or NLOS paths; as well as Receive at least one model parameter from the network device, obtained by updating the AI / ML model.

24. An apparatus comprising: A component for receiving an artificial intelligence (AI) / machine learning (ML) model and a first dataset from a terminal device at a network device, wherein the AI / ML model has the ability to provide inference results for line-of-sight (LOS) / non-line-of-sight (NLOS) classification with confidence levels, and the first dataset includes a first set of data samples associated with a first set of labels, the first set of labels being true labels indicating LOS paths or NLOS paths; A component for obtaining a second dataset based on multiple unlabeled data samples, the AI / ML model, and the ratio of the number of labels indicating LOS paths to the number of labels indicating NLOS paths in the first label set, wherein the second dataset includes a second set of data samples associated with a second label set, the second label set being pseudo-labels indicating LOS paths or NLOS paths. as well as A component for updating at least one model parameter of the AI / ML model based on the second dataset.

25. An apparatus comprising: A component for receiving a query message from a network device at a terminal device, the query message being used to query whether an artificial intelligence (AI) / machine learning (ML) model at the terminal device has the ability to provide inference results for line-of-sight (LOS) / non-line-of-sight (NLOS) classification with confidence levels; A component for sending the AI / ML model and dataset to the network device based on determining that the AI / ML model has the capability, wherein the dataset includes a set of data samples associated with a tag set, the tag set being real labels indicating LOS paths or NLOS paths; as well as A component for receiving at least one model parameter of the updated AI / ML model obtained by updating the AI / ML model from the network device.

26. A non-transitory computer-readable medium comprising program instructions for causing a device to perform at least the method according to claim 22 or 23.

Images