WO2024197929A1

WO2024197929A1 - Methods, devices, and computer readable medium for communication

Info

Publication number: WO2024197929A1
Application number: PCT/CN2023/085768
Authority: WO
Inventors: Peng Guan; Zhen He; Wei Chen; Gang Wang
Original assignee: NEC Corp
Current assignee: NEC Corp
Priority date: 2023-03-31
Filing date: 2023-03-31
Publication date: 2024-10-03
Anticipated expiration: 2025-09-30

Abstract

Embodiments of the present disclosure provide a solution for model updating. In a solution, a terminal device obtains similarity information between a first model and a second model. The terminal device determines a processing on the second model based on the similarity information. The processing includes one or more of: a model training, a model monitoring, or a model delivery. In this way, it can avoid unnecessary processing of the model, thereby saving power at the terminal device. Further, it can also avoid unnecessary transmissions of model, thereby saving signaling overhead.

Description

METHODS, DEVICES, AND COMPUTER READABLE MEDIUM FOR COMMUNICATION

FIELDS

Example embodiments of the present disclosure generally relate to the field of communication techniques and in particular, to methods, devices, and computer readable medium for communication.

BACKGROUND

Several technologies have been proposed to improve communication performances. For example, communication devices may employ a data processing model to improve communication qualities. The data processing model can be applied to different scenarios to achieve better performances. Thus, how to properly apply the data processing model is worth studying, in order to ensure satisfying communication performances.

SUMMARY

In general, example embodiments of the present disclosure provide a solution for communication.

In a first aspect, there is provided a terminal device comprising: a processor, configured to cause the terminal device to: obtain similarity information between a first model and a second model, wherein the similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device, and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model is deduced from a same original model; the second model is trained after the first model in a timeline; or at least one parameter corresponding to the first model is same with the second model.

In a second aspect, there is provided a network device comprising: a processor, configured to cause the network device to: transmit a similarity determination configuration to a terminal device; obtain similarity information between a first model and a second model, wherein the similarity information comprises a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device; and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model are deduced from a same original model; the second model is trained after the first model in a timeline; or at least one parameter corresponding to the first model is same with the second model.

In a third aspect, there is provided a communication method performed by a terminal device. The method comprises obtaining similarity information between a first model and a second model, wherein the similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determining, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device, and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model is deduced from a same original model; the second model is trained after the first model in a timeline; or at least one parameter corresponding to the first model is same with the second model.

In a fourth aspect, there is provided a communication method performed by a network device. The method comprises transmitting a similarity determination configuration to a terminal device; obtaining similarity information between a first model and a second model, wherein the similarity information comprises a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determining, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device; and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model are deduced from a same original model; the second model is trained after the first model in a timeline; or at least one parameter corresponding to the first model is same with the second model.

In a fifth aspect, there is provided a computer readable medium having instructions stored thereon, the instructions, when executed on at least one processor, causing the at least one processor to carry out the method according to the third, or fourth aspect.

Other features of the present disclosure will become easily comprehensible through the following description.

BRIEF DESCRIPTION OF THE DRAWINGS

Through the more detailed description of some example embodiments of the present disclosure in the accompanying drawings, the above and other objects, features and advantages of the present disclosure will become more apparent, wherein:

FIG. 1 illustrates an example communication environment in which example embodiments of the present disclosure can be implemented;

FIG. 2 illustrates a schematic diagram of a framework 200 for a determination of model similarity;

FIG. 3 illustrates a schematic diagram of a framework for a determination of dataset similarity;

FIG. 4 illustrates a schematic diagram of cosine similarity;

FIG. 5 illustrates a signaling flow of model processing in accordance with some embodiments of the present disclosure;

FIG. 6 illustrates a flowchart of a method implemented at a network device according to some example embodiments of the present disclosure;

FIG. 7 illustrates a flowchart of a method implemented at a terminal device according to some example embodiments of the present disclosure; and

FIG. 8 illustrates a simplified block diagram of an apparatus that is suitable for implementing example embodiments of the present disclosure.

Throughout the drawings, the same or similar reference numerals represent the same or similar element.

DETAILED DESCRIPTION

Principle of the present disclosure will now be described with reference to some example embodiments. It is to be understood that these embodiments are described only for the purpose of illustration and help those skilled in the art to understand and implement the present disclosure, without suggesting any limitation as to the scope of the disclosure. Embodiments described herein can be implemented in various manners other than the ones described below.

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

As used herein, the term ‘terminal device’ refers to any device having wireless or wired communication capabilities. Examples of the terminal device include, but not limited to, user equipment (UE) , personal computers, desktops, mobile phones, cellular phones, smart phones, personal digital assistants (PDAs) , portable computers, tablets, wearable devices, internet of things (IoT) devices, Ultra-reliable and Low Latency Communications (URLLC) devices, Internet of Everything (IoE) devices, machine type communication (MTC) devices, devices on vehicle for V2X communication where X means pedestrian, vehicle, or infrastructure/network, devices for Integrated Access and Backhaul (IAB) , Space borne vehicles or Air borne vehicles in Non-terrestrial networks (NTN) including Satellites and High Altitude Platforms (HAPs) encompassing Unmanned Aircraft Systems (UAS) , eXtended Reality (XR) devices including different types of realities such as Augmented Reality (AR) , Mixed Reality (MR) and Virtual Reality (VR) , the unmanned aerial vehicle (UAV) commonly known as a drone which is an aircraft without any human pilot, devices on high speed train (HST) , or image capture devices such as digital cameras, sensors, gaming devices, music storage and playback appliances, or Internet appliances enabling wireless or wired Internet access and browsing and the like. The ‘terminal device’ can further has ‘multicast/broadcast’ feature, to support public safety and mission critical, V2X applications, transparent IPv4/IPv6 multicast delivery, IPTV, smart TV, radio services, software delivery over wireless, group communications and IoT applications. It may also incorporate one or multiple Subscriber Identity Module (SIM) as known as Multi-SIM. The term “terminal device” can be used interchangeably with a UE, a mobile station, a subscriber station, a mobile terminal, a user terminal or a wireless device.

The term “network device” refers to a device which is capable of providing or hosting a cell or coverage where terminal devices can communicate. Examples of a network device include, but not limited to, a Node B (NodeB or NB) , an evolved NodeB (eNodeB or eNB) , a next generation NodeB (gNB) , a transmission reception point (TRP) , a remote radio unit (RRU) , a radio head (RH) , a remote radio head (RRH) , an IAB node, a low power node such as a femto node, a pico node, a reconfigurable intelligent surface (RIS) , and the like.

The terminal device or the network device may have Artificial intelligence (AI) or Machine learning capability. It generally includes a model which has been trained from numerous collected data for a specific function, and can be used to predict some information.

The terminal or the network device may work on several frequency ranges, e.g., FR1 (e.g., 450 MHz to 6000 MHz) , FR2 (e.g., 24.25GHz to 52.6GHz) , frequency band larger than 100 GHz as well as Tera Hertz (THz) . It can further work on licensed/unlicensed/shared spectrum. The terminal device may have more than one connection with the network devices under Multi-Radio Dual Connectivity (MR-DC) application scenario. The terminal device or the network device can work on full duplex, flexible duplex and cross division duplex modes.

The embodiments of the present disclosure may be performed in test equipment, e.g., signal generator, signal analyzer, spectrum analyzer, network analyzer, test terminal device, test network device, channel emulator. In some embodiments, the terminal device may be connected with a first network device and a second network device. One of the first network device and the second network device may be a master node and the other one may be a secondary node. The first network device and the second network device may use different radio access technologies (RATs) . In some embodiments, the first network device may be a first RAT device and the second network device may be a second RAT device. In some embodiments, the first RAT device is eNB and the second RAT device is gNB. Information related with different RATs may be transmitted to the terminal device from at least one of the first network device or the second network device. In some embodiments, first information may be transmitted to the terminal device from the first network device and second information may be transmitted to the terminal device from the second network device directly or via the first network device. In some embodiments, information related with configuration for the terminal device configured by the second network device may be transmitted from the second network device via the first network device. Information related with reconfiguration for the terminal device configured by the second network device may be transmitted to the terminal device from the second network device directly or via the first network device.

As used herein, the singular forms ‘a’ , ‘an’ and ‘the’ are intended to include the plural forms as well, unless the context clearly indicates otherwise. The term ‘includes’ and its variants are to be read as open terms that mean ‘includes, but is not limited to. ’ The term ‘based on’ is to be read as ‘at least in part based on. ’ The term ‘one embodiment’ and ‘an embodiment’ are to be read as ‘at least one embodiment. ’ The term ‘another embodiment’ is to be read as ‘at least one other embodiment. ’ The terms ‘first, ’ ‘second, ’ and the like may refer to different or same objects. Other definitions, explicit and implicit, may be included below.

In some examples, values, procedures, or apparatus are referred to as ‘best, ’ ‘lowest, ’ ‘highest, ’ ‘minimum, ’ ‘maximum, ’ or the like. It will be appreciated that such descriptions are intended to indicate that a selection among many used functional alternatives can be made, and such selections need not be better, smaller, higher, or otherwise preferable to other selections.

As used herein, the term “resource, ” “transmission resource, ” “uplink resource, ” or “downlink resource” may refer to any resource for performing a communication, such as a resource in time domain, a resource in frequency domain, a resource in space domain, a resource in code domain, or any other resource enabling a communication, and the like. In the following, unless explicitly stated, a resource in both frequency domain and time domain will be used as an example of a transmission resource for describing some example embodiments of the present disclosure. It is noted that example embodiments of the present disclosure are equally applicable to other resources in other domains.

As mentioned above, the data processing model can be applied to different scenarios to achieve better performances. In some embodiments, the data processing model can be implemented at the network device side. Alternatively, the data processing model can be implemented at the terminal device side. In other embodiments, the data processing model can be implemented at both the network device and the terminal device.

In some solutions, a model configuration may be transmitted between devices. However, it may cause severe delay due to the extremely large payload size. Further, when the terminal device performs model training or model monitoring, it may consume quite a large amount of power.

In order to solve at least part of the above problems, a solution on updating a model is needed. According to embodiments of the present disclosure, solutions on updating a model based on similarity information related to the model are proposed. In particular, a terminal device obtains similarity information between a first model and a second model. The terminal device determines a processing on the second model based on the similarity information. The processing includes one or more of: a model training, a model monitoring, or a model delivery. In this way, it can avoid unnecessary processing of the model, thereby saving power at the terminal device. Further, it can also avoid unnecessary transmissions of model, thereby saving signaling overhead.

In this context, the term “similarity information” used herein may refer to information that indicates how similar/difference two objects are. The term “model” used herein may refer to a data driven algorithm that applies artificial intelligence/machine learning (AI/ML) techniques to generate a set of outputs based on a set of inputs. The terms “model” and “AI/ML model” may be used interchangeable. The model may comprise a set of weights values that may be learned during training for a specific architecture or configuration, where a set of weights values may also be called a parameter set or a dataset. The term “dataset” used herein may refer to a group of data used for model training. The term “performance metric” used herein may refer to any metrics that reflect how well the model is. The term “dataset for training” used herein may refer to an initial set of data used to help a program understand how to apply technologies like neural networks to learn and produce sophisticated results. It may be complemented by subsequent sets of data called validation and testing sets. The terms “dataset for training” , “training data” , “training set” , “training dataset” and “learning set” may be used interchangeable.

The term “data collection” may refer to a process of collecting data by the network nodes, management entity, or UE for the purpose of model training, data analytics and inference. The term “model training” used herein may refer to a process to train a model [by learning the input/output relationship] in a data driven manner and obtain the trained model for inference. The term “model inference” used herein can refer to a process of using a trained model to produce a set of outputs based on a set of inputs. The term “model monitoring” used herein may refer to a procedure that monitors the inference performance of the model. The term “model delivery” used herein may refer to a procedure that transfer the model between devices. The term “model validation” used herein may refer to a subprocess of training, to evaluate the quality of a model using a dataset different from one used for model training, that helps selecting model parameters that generalize beyond the dataset used for model training.

The term “UE-side model” used herein may refer to a model of which inference is performed entirely at the UE. The term “network-side model” used herein may refer to a model of which inference is performed entirely at the network. The term “one-side model” used herein may refer to a UE-side model or a network-side model. The term “two Two-sided model” used herein may refer to a paired model (s) over which joint inference is performed, where joint inference includes inference whose inference is performed jointly across the UE and the network, i.e, the first part of inference is firstly performed by UE and then the remaining part is performed by gNB, or vice versa.

The term “model activation” used herein may refer to enabling a model for a specific function. The term “model deactivation” used herein may refer to disabling a model for a specific function. The term “model switching” used herein may refer to deactivating a currently active model and activating a different model for a specific function. The term “model management” used herein may refer to a more general term includes one or more of following functions/procedures: model activation, deactivation, selection, switching, fallback, and update (including re-training) .

Principles and implementations of the present disclosure will be described in detail below with reference to the figures.

FIG. 1 illustrates a schematic diagram of an example communication environment 100 in which example embodiments of the present disclosure can be implemented. In the communication environment 100, a plurality of communication devices, including a terminal device 110 and a network device 120, can communicate with each other.

In the example of FIG. 1, the terminal device 110 may be a UE and the network device 120 may be a base station serving the UE. The serving area of the network device 120 may be called a cell 102.

It is to be understood that the number of devices and their connections shown in FIG. 1 are only for the purpose of illustration without suggesting any limitation. The communication environment 100 may include any suitable number of devices configured to implementing example embodiments of the present disclosure. Although not shown, it would be appreciated that one or more additional devices may be located in the cell 102, and one or more additional cells may be deployed in the communication environment 100. It is noted that although illustrated as a network device, the network device 120 may be another device than a network device. The network device 130 may be another device than a network device. Although illustrated as a terminal device, the terminal device 110 may be other device than a terminal device.

In the following, for the purpose of illustration, some example embodiments are described with the terminal device 110 operating as a UE and the network device 120 operating as a base station. However, in some example embodiments, operations described in connection with a terminal device may be implemented at a network device or other device, and operations described in connection with a network device may be implemented at a terminal device or other device.

In some example embodiments, if the terminal device 110 is a terminal device and the network device 120 is a network device, a link from the network device 120 to the terminal device 110 is referred to as a downlink (DL) , while a link from the terminal device 110 to the network device 120 is referred to as an uplink (UL) . In DL, the network device 120 is a transmitting (TX) device (or a transmitter) and the terminal device 110 is a receiving (RX) device (or a receiver) . In UL, the terminal device 110 is a TX device (or a transmitter) and the network device 120 is a RX device (or a receiver) .

The communications in the communication environment 100 may conform to any suitable standards including, but not limited to, Global System for Mobile Communications (GSM) , Long Term Evolution (LTE) , LTE-Evolution, LTE-Advanced (LTE-A) , New Radio (NR) , Wideband Code Division Multiple Access (WCDMA) , Code Division Multiple Access (CDMA) , GSM EDGE Radio Access Network (GERAN) , Machine Type Communication (MTC) and the like. The embodiments of the present disclosure may be performed according to any generation communication protocols either currently known or to be developed in the future. Examples of the communication protocols include, but not limited to, the first generation (1G) , the second generation (2G) , 2.5G, 2.75G, the third generation (3G) , the fourth generation (4G) , 4.5G, the fifth generation (5G) communication protocols, 5.5G, 5G-Advanced networks, or the sixth generation (6G) networks.

With reference to FIG. 2 and FIG. 3, a model 220-1 (also referred to as “second model” ) and a model 220-2 (also referred to as “first model” ) may be implemented in the communication environment 100. The model 220-1 and the model 220-2 are associated with parameter related to a communication between the terminal device 110 and the network device 120. The parameter related to the communication may be any suitable parameter. For example, the parameter related to the communication may be associated with a use case of the model 220-1 and the model 220-2. By way of example, the parameter may be a positioning parameter. Alternatively, or in addition, the parameter may be a beam related parameter. In some other embodiments, the parameter may be a channel state information (CSI) related parameter.

In some embodiments, the model 220-1 and the model 220-2 may have a same identifier. For example, the model 220-1 and the model 220-2 may have the same model ID. Alternatively, or in addition, the model 220-1 and the model 220-2 may have the same dataset for input. By way of example, as shown in FIG. 2, both the model 220-1 and the model 220-2 may use the input sample dataset 210. In some other embodiments, the model 220-1 and the model 220-2 may be deduced from a same original model. For example, the model 220-1 model may be trained at the terminal device 110 based on an original model. The model 220-2 may be trained at the network device 120 based on the original model. Alternatively, the original model may be trained using two different datasets to obtain the model 220-1 and the model 220-2. In some embodiments, the model 220-1 may be trained after the model 220-2 in timeline. For example, after delivery the model 220-2 to the terminal device 110, the network device 120 may continue training to obtain the model 220-1. In some other embodiments, one or more parameters corresponding to the model 220-1 may be same with the model 220-2. For example, the one or more parameters may include an index corresponding to a set of weight values. As another example, the one or more parameters may include a use case of the model. It is noted that the one or more parameters may include any parameters related to model. Embodiments of the present disclosure are not limited to this aspect.

FIG. 2 illustrates a schematic diagram of a framework 200 for a determination of model similarity. As shown in FIG. 2, an input sample dataset 210 may be used as inputs of the model 220-1 and the model 220-2. For example, the input sample dataset 210 may be a sample dataset which is used to control inputs for various models. The sample dataset can minimize other factors which may influence outputs, so that the outputs can reflect the similarity in a more reasonable way. In some embodiments, the sample dataset may be generated by the network device 120. Alternatively, the sample dataset may be generated by other devices. In some embodiments, the model 220-1 may be implemented at the terminal device 110 and the model 220-2 may be implemented at the network device 120. Alternatively, the model 220-1 may be implemented at the network device 120 and the model 220-2 may be implemented at the terminal device 110. Alternatively, both the model 220-1 and the model 220-2 may be implemented at the network device 120. Alternatively, both the model 220-1 and the model 220-2 may be implemented at the terminal device 110.

The model 220-1 may output an output dataset 230-1 and the model 220-2 may output an output dataset 230-2. A similarity determination algorithm 240 may be applied to the output datasets 230-1 and 230-2 to obtain the model similarity 250 between the models 220-1 and 220-2.

FIG. 3 illustrates a schematic diagram of a framework 300 for a determination of dataset similarity. As shown in FIG. 3, a dataset 310-1 may be used for training the model 220-1 and a dataset 310-2 may be used for training the model 220-2. In other words, during training the model 220-1, the dataset 310-1 may be used as examples given to the model 220-1 to analyze and learn. During training the model 220-2, the dataset 310-2 may be used as examples given to the model 220-2 to analyze and learn. The similarity determination algorithm 240 may be applied to the datasets 310-1 and 310-2 to obtain a dataset similarity between the models 220-1 and 220-2. Alternatively, the weight similarity between the models 220-1 and 220-2 may be obtained by comparing a first dataset including a part or all of weight parameters of the model 220-1 and a second dataset including a part or all of weight parameters of the model 220- 2. For example, the similarity determination algorithm 240 may be applied to a first set of weight parameters corresponding to the model 220-1 and a second set of weight parameters corresponding to the model 220-2 to obtain a weight similarity between the models 220-1 and 220-2. It is noted that similarity determination algorithm 240 may any proper kinds of similarities between the models 220-1 and 220-2.

In some embodiments, the similarity determination algorithm 240 may be a cross-entropy algorithm. The cross-entropy may be used as a loss function in classification models and can be impressed by the two-classification cross-entropy used by the cat. The cross-entropy may be divided into two parts, “cross” and “entropy” . The formula for entropy may include discrete variable: Entropy=-∑_iP (i) log P (i) , where i represents a discrete variable, ; and continuous variable: Entropy=-∫P (x) logP (x) dx. The formula for probability distribution P of entropy may include H (P) =Entropy=E_x～P [-logP (x) ] , where x represents a continuous variable, x～P means that x belongs to P, P represents a model (for example, the model 220-1) , P (x) means that x is a parameter of P The formula for estimated probability distribution Q of entropy may include H (Q) =Entropy=E_x～Q [-logQ (x) ] , , where x represents a continuous variable, x～Q means that x belongs to Q, Q represents a model (for example, the model 220-2) , Q (x) means that x is a parameter of Q Cross-entropy can be expressed using H (P, Q) , meaning that P is used to calculate expectations and Q is used to calculate the encoding length: H (P, Q) =E_x～P [-logQ (x) ] =-∑_iP (i) logQ (i) , where H (P, Q) the cross-entropy which is represents a similarity between the parameter value P output by a first model and the parameter value Q output by a second model, and P (i) represents an estimated value of the first model, and Q (i) represents an estimated value of the second model .

Alternatively, the similarity determination algorithm 240 may be a mean-squared error (MSE) algorithm. MSE may measure how well predicted values match some true values. MSE is often used as a loss function for regression problem. For example, MSE is a commonly used statistical measure and loss function in ML regression models, such as linear regression. The formula of MSE may bewhere n represents the sample size, Y_i represents the real value andrepresents the predicted value, i represents the i-th sample in the sample set. By way of example, the i-th sample in the output dataset 230-1 may be used as Y_i and the i-th sample in the output dataset 230-2 may be used asIn this way, the model similarity between the models 220-1 and 220-2 may be

In some other embodiments, the similarity determination algorithm 240 may be a cosine similarity algorithm. Cosine Similarity may be to use the angle between two vectors in space to judge the similarity of these two vectors. When the angle between two vectors is lager and the distance is longer, the maximum distance is 180 degrees between the two vectors. The smaller the angle, the closer the distance, and the minimum distance is 0 degree between the two vectors, which coincides completely. As shown in FIG. 4, the cosine of the angle may bewhere θ represent the angle between the two vectors, a represents a vector with respect to the origin point in the coordinate, and b represents a vector with respect to the origin point in the coordinate, ab represents a product of the two vectors ‘a’ and ‘b’ , ||a|| and ||b|| represent lengths of the two vectors ‘a’ and ‘b’ , respectively. The cosine similarity may not be sensitive to the value, instead, it is quite sensitive to the direction. The parameter a may be associated with the first model and the parameter b may be associated with the second model. By way of example, the output dataset 230-1 may be treated as the vector a and the output dataset 230-2 may be treated as the vector b. Only as an example, if the output dataset 230-1 includes: 3, 2, 0, 5, and the output dataset 230-2 includes: 1, 0, 0, 0, the vector a may be {3, 2, 0, 5 } and the vector b may be {1, 0, 0, 0} . In this case, ab = 3*1 + 2*0 + 0*0 + 5*0 = 3, ||a|| =6.16, ||b|| =1, and the cosine similarity may be (3/ (6.16*1) ) which equals to 0.49.

In some embodiments, a model differentiation may be sued to evaluate the similarity. In this case, a cumulative distribution function (CDF) may be configured as the similarity determination algorithm 240.

The models 220-1 and 220-2 may be used for different use cases. For example, the models 220-1 and 220-2 may be applied in a use case of quality of experience (QoE) . Alternatively, or in addition, the models 220-1 and 220-2 may be applied in a use case of mobility. In some other embodiments, the models 220-1 and 220-2 may be applied in a use case of load balancing. In a yet example embodiment, the models 220-1 and 220-2 may be applied in a use case of energy saving. Different AI/ML models used for different use cases can be configured with different similarity computation algorithms.

Reference is made to FIG. 5, which illustrates a signaling flow 500 of updating model in accordance with some embodiments of the present disclosure. As shown in FIG. 5, the signaling flow 400 involves a device 510 and a device 520. For example, the device 510 may be implemented at the terminal device 110 and the device 520 may be implemented at the network device 120. Alternatively, the device 520 may be implemented at a core network device, for example, a location management function (LMF) entity.

The device 520 may transmit (5010) a similarity determination configuration to the device 510. In other words, the device 510 may receive the similarity determination configuration from the device 520. In some embodiments, the similarity determination configuration may be transmitted in downlink control information. Alternatively, the similarity determination configuration may be transmitted in medium access control control element (MAC CE) . In some other embodiments, the similarity determination configuration may be transmitted in a radio resource control (RRC) message.

The similarity determination configuration may include one or more parameters that are used for determining the similarity information. For example, the similarity determination configuration may include the similarity determination algorithm (for example, the similarity determination algorithm 240) . The similarity determination configuration may also include a measurement and reporting criteria of the similarity.

In some embodiments, the similarity determination configuration may include model identification information. For example, the similarity determination configuration may include a model ID (for example, the model ID of the model 220-1) . The similarity determination configuration may include other information that identifying the model.

Alternatively, or in addition, the similarity determination configuration may include a type of the similarity information. For example, the type of the similarity information may include a model similarity. In some embodiments, the type of the similarity information may include a dataset difference. As another example, the type of the similarity information may include a model difference. The terms “model difference” , “model distance” and “model dis-similarity” may be used interchangeable. Details of different types of similarity information are described later. It is noted that the term “similarity” described in the present disclosure can be replaced by the term “difference” or other similar terms. A determination condition for the similarity and the determination condition for the difference may be opposite. By way of example, an action can be performed, when the similarity is above or equal to a certain threshold or the difference is below or equal to the certain threshold.

In some other embodiments, the similarity determination configuration may include a mode use case. For example, as mentioned above, the similarity determination configuration may indicate the use case of quality of experience (QoE) . Alternatively, or in addition, the similarity determination configuration may indicate a use case of mobility. As another example, the similarity determination configuration may include a use case of beam prediction.

Alternatively, or in addition, the similarity determination configuration may include a sample dataset (for example, the sample dataset 210) for training the model (for example, the first model and/or the second model) . In some embodiments, a configuration the sample dataset can include a periodical configuration that indicates when and how to update the sample dataset. For example, the periodical configuration may include a starting timestamp that indicates when to update the sample dataset. The periodical configuration may also include a periodicity that indicates a period of updating the sample dataset.

In some embodiments, the device 510 may transmit a request for the similarity determination configuration to the device 520. In this case, the similarity determination configuration may be transmitted based on the request. Alternatively, the transmission of the similarity determination configuration may be triggered by the device 520. In other words, the device 520 may transmit the similarity determination configuration without the request. In some embodiments, the device 520 may transmit the similarity determination configuration periodically. In other words, the device 510 may receive the similarity determination configuration periodically.

The device 510 obtains (5020) similarity information between the model 220-1 and the model 220-2. The similarity information between the models 220-1 and 220-2 may indicate how similar the models 220-1 and 220-2 are. In some embodiments, the similarity information may indicate differences between the models 220-1 and 220-2. In some embodiments, the device 510 may determine a model similarity between an output of the model 220-1 and an output the model 220-2. For example, the model similarity may be determined based on the similarity computation algorithm 240 and the output datasets 230-1 and 230-2. Alternatively, the device 510 may determine a model difference between the output of the model 220-1 and the output of the model 220-2. By way of example, the model difference may be determined based on the similarity computation algorithm 240 and the output datasets 230-1 and 230-2. Alternatively, or in addition, the device 510 may determine a dataset similarity between a dataset (referred to as “second dataset” ) corresponding to the model 220-1 and a dataset (referred to as “first dataset” ) corresponding to the model 220-2. By way of example, the dataset similarity may be determined based on the similarity computation algorithm 240 and the datasets 310-1 and 310-2.

In some embodiments, the similarity information may be transmitted from the device 520 to the device 510. In some embodiments, the similarity information may be transmitted in DCI. Alternatively, the similarity information may be transmitted in MAC CE. In some other embodiments, the similarity information may be transmitted in a RRC message.

In some embodiments, the device 510 may receive a trigger of similarity determination. For example, the device 510 may receive the trigger of similarity determination from the device 520. In some embodiments, the device 510 may obtain the similarity in response to the reception of the trigger of similarity determination. For example, if the device 510 receives an explicit indication, the device 510 may stop training the model 220-1 and start to determine the similarity between the models 220-1 and 220-2. Alternatively, if the device 510 receives a message carrying a similarity determination configuration, the device 510 may stop training the model 220-1 and start to determine the similarity between the models 220-1 and 220-2. In other words, the device 510 may stop training the model 220-1 and start to determine the similarity based on an implicit indication.

In some embodiments, the trigger of similarity determination may include a periodicity for performing the similarity determination. In this case, the device 510 can perform the similarity determination/measurement in a periodical manner. Alternatively, or in addition, the trigger of similarity determination may include a starting time at which the similarity determination starts. For example, the trigger of similarity determination may include s time stamp that is used to indicate the starting time for the similarity determination. Since the model is always being trained constantly, so if the stating time is different, it cannot really reflect the similarity, meanwhile, the network device needs to record the model at that time being.

The device 510 determines (5030) a processing on the model 220-1 based at least one on the similarity information. The processing includes one or more of: a model training, a model monitoring or a model delivery. In some embodiments, the processing may be determined based on the similarity information and a model performance of the model 220-1. The model delivery may include a model uploading or a model downloading. Details of determining the processing are described later.

In some embodiments, the model 220-1 may be trained at the device 510 and monitored at the device 520. In this case, in some embodiments, the similarity determination configuration may include a periodical configuration for the model delivery of the model 220-1. For example, the periodical configuration may include a periodicity for the model delivery of the model 220-1. The periodical configuration may also include a starting time at which the model 220-1 is delivered. In addition, the periodical configuration may include an indication (referred to as “first indication” ) for enabling or disabling delivering the model 220-1. In some embodiments, the first indication may be transmitted in DCI. Alternatively, the first indication may be transmitted in MAC CE. In some other embodiments, the first indication may be transmitted in a RRC message.

In some other embodiments, the periodical configuration may include an event-based configuration for the model delivery. For example, the event-based configuration may include a similarity threshold. The similarity threshold may include a model similarity threshold and/or a dataset similarity threshold. In this case, if a similarity (for example, the model similarity and/or the dataset similarity) is below or equal to the similarity threshold, the device 510 may deliver the model 220-1 to the device 520. If the similarity is is above or equal to the similarity threshold, the device 510 may not deliver the model 220-1.

Alternatively, the event-based configuration may include a difference threshold. The difference threshold may include a model difference threshold and/or a dataset difference threshold. In this case, if a difference (for example, the model difference and/or the dataset difference) is above or equal to the difference threshold, the device 510 may deliver the model 220-1 to the device 520. If the difference is below or equal to the difference threshold, the device 510 may not deliver the model 220-1.

In some embodiments, the device 510 may deliver the model 220-1 to the device 520 according to a periodical configuration for model delivery. The periodical configuration may indicate when to deliver the model 220-1.

The device 510 may stop the model training on the model 220-1. For example, after delivering the model 220-1 to the device 520, the device 510 may stop training the model 220-1. For example, after delivering the model 220-1, the device 510 may stop the model training and wait for network indication on the model performance measurement. Alternatively, the device 510 may stop training the model 220-1, before delivering the model 220-1 to the device 520.

In some embodiments, the device 520 may transmit an indication (referred to as “second indication” ) regarding whether the device 510 is allowed to continue training the model 220-1 to the device 510. In other words, the device 510 may receive the second indication from the network device 110. In this case, if the second indication indicates that the device 510 is allowed to continue training the model 220-1, the device 510 may perform the model training on the model 220-1 until a next periodicity for model delivery. Alternatively, if the second indication indicates the terminal device is not allowed to continue training the model, the device 510 may continue stopping the model training on the model 220-1 for a predetermined duration. In some embodiments, the device 510 may start a timer (referred to as “first timer” ) upon receiving the second indication. In this case, after the first timer expires, the device 510 may continue to perform the model training on the model 220-1. The first timer may be configured by the device 520 or may be pre-configured. In some embodiments, the second indication may be transmitted in DCI. Alternatively, the second indication may be transmitted in MAC CE. In some other embodiments, the second indication may be transmitted in a RRC message.

Alternatively, the device 510 may determine whether a similarity is above a configured similarity threshold. For example, when the device 510 needs to deliver the model 220-1, the device 510 may periodically determine the similarity information between a current trained model (i.e., the model 220-1) and a previous uploaded model (i.e., the model 220-2) . In some embodiments, if the similarity is above or equal to the configured similarity threshold (which means that the model 220-1 and the model 220-2 are quite similar) , the device 510 may skip delivering the model 220-1 to the device 520 to be skipped. In this case, the device 510 may transmit to the device 520 an indication (referred to as “third indication” ) indicating that the delivering of the model 220-1 to be skipped for a predetermined period. Alternatively, if the similarity is below or equal to the configured similarity threshold, the device 510 may deliver the model 220-2 to the device 520 according to a periodical configuration for model delivery. The third indication may be transmitted in one of: a RRC message, uplink control information (UCI) , or an uplink MAC CE.

In some other embodiments, the device 510 may determine whether a difference is above or equal to a configured difference threshold. In this case, if the difference is below or equal to the configured difference threshold, the device 510 may skip delivering the model 220-1 to the device 520. In this case, the device 510 may transmit to the device 520 a third indication indicating that the delivering of the model 220-2 to be skipped for the predetermined period. Alternatively, if the difference is above or equal to the configured difference threshold, the device 510 may deliver the model 220-2 to the device 120 according to a periodical configuration for model delivery.

In some embodiments, the device 510 may determine whether a similarity in the similarity information is above or equal to a configured similarity threshold. In this case, if the similarity is above or equal to the configured similarity threshold and a performance metric for the model 220-1 is above or equal to a configured performance threshold, the device 510 may stop the model training on the model 220-1 for a determined duration. If the similarity is above or equal to the configured similarity threshold and a performance metric for the model 220-1 is below or equal to a configured performance threshold, the device 510 may continue training the model 220-1. Alternatively, the device 510 may determine whether a difference is above or equal to a configured difference threshold. In this case, if the difference is below or equal to the configured difference threshold and a performance metric for the model 220-1 is above or equal to the configured performance threshold, the device 510 may stop the model training on the model 220-1 for a determined duration. If the difference is below or equal to the configured difference threshold and a performance metric for the model 220-1 is below or equal to the configured performance threshold, the device 510 may continue training the model 220-1. In some embodiments, if a timer (referred to as “third timer” ) starting from a time point when the similarity information is obtained expires, the device 510 may continue to performing the model training on the model 220-1. For example, the determined duration may be running time of the third timer.

In some embodiments, the model 220-1 is trained and monitored at the device 510. In this case, the similarity determination configuration may further comprise a model training related configuration. For example, the model training related configuration comprises a performance metric threshold. In this case, if a monitored performance metric of the model 220-1 is above or equal to the performance metric threshold, the device 510 may stop the model training on the model 220-1. Alternatively, if the monitored performance metric of the model 220-1 is below or equal to the performance metric threshold, the device 510 may continue training the model 220-1. By way of example, the performance metric may be an inference accuracy of the model. It is noted that the monitored performance metric may be any metrics that reflect how well the model is.

In some other embodiments, the model training related configuration comprises a similarity threshold. In this case, if the similarity is above or equal to the similarity threshold, the device 510 may stop the model training on the model 220-1. If the similarity is below or equal to the similarity threshold, the device 510 may deliver the model 220-2 to the device 520.

Alternatively, the model training related configuration comprises a difference threshold. In this case, if the difference is below or equal to the difference threshold, the device 510 may stop the model training on the model 220-1. If the difference is above or equal to the difference threshold, the device 510 may deliver the model 220-1 to the device 520.

In some embodiments, the device 510 may transmit, to the device 520, an indication (referred to as “fourth indication” ) one or more of: the performance metric threshold is met, the performance metric threshold is met, the similarity threshold is not met, the similarity threshold is met, the performance metric threshold is not met while the similarity threshold is met, or the performance metric threshold is met while the similarity threshold is not met. By way of example, the fourth indication may be transmitted in one of: UCI, uplink MAC CE, or uplink RRC message.

In some embodiments, the model 220-1 is trained at the device 520 and monitored at the device 510. In this case, in some embodiments, the similarity determination configuration includes a periodical configuration for the model delivery of the model 220-1. For example, the periodical configuration may include one or more of: a periodicity for the model delivery of the model 220-1, a starting time at which the is delivered, or an indication (referred to as “fifth indication” ) for enabling or disabling downloading the second model. In some other embodiments, the similarity determination configuration includes an event-based configuration for the model delivery of the model 220-1.

In some embodiments, if a similarity is below or equal to a first similarity threshold in the event-based configuration, the device 520 may transmit to the device 510, an indication (referred to as “sixth indication” ) indicating that the second model is to be delivered. In other words, the device 510 may receive the sixth indication from the device 520. The sixth indication may be transmitted in one of: DCI, MAC CE, or RRC message. In some other embodiments, if the similarity is below or equal to a second similarity threshold in the event-based configuration, the device 510 may perform the model monitoring on the model 220-1. If the similarity is above or equal to the second similarity threshold, the device 510 may transmit to the device 520, an indication (referred to as “seventh indication” ) for further model training on the model 220-1. In other words, the device 520 may receive the seventh indication from the device 510. The seventh indication may be transmitted in one of: UCI, uplink MAC CE, or uplink RRC message.

Alternatively, if the difference is above or equal to a first difference threshold in the event-based configuration, the device 520 may transmit to the device 510 an indication (referred to as “eighth indication” ) indicating that the model 220-1 is to be delivered. In other words, the device 510 may receive the eighth indication from the device 520. The eighth indication may be transmitted in one of: DCI, MAC CE or RRC message.

In some embodiments, if the difference is above or equal to a second difference threshold in the event-based configuration, the device 510 may perform the model monitoring on the model 220-1. Alternatively, if the difference is below or equal to the second difference threshold, the device 510 may transmit to the device 520, an indication (referred to as “ninth indication” ) for further model training on the model 220-1. In other words, the device 520 may receive the ninth indication from the device 510. The ninth indication may be transmitted in one of: UCI, uplink MAC CE or uplink RRC message.

In some embodiments, the device 520 may transmit the model 220-1 to the device 510 according to a periodical configuration for model delivery. In other words, the device 510 may receive the model 220-1 from the device 520. In this case, after receiving the model 220-1, the device 510 may perform a performance monitoring on the model 220-1. The device 510 may transmit to the device 520 a result of the performance monitoring. For example, the result of the performance monitoring may indicate a prediction accuracy of the model 220-1. The result of the performance monitoring may be transmitted in one of: UCI, uplink MAC CE, or uplink RRC message.

The device 520 may transmit to the device 510 at least one of the following: an indication (referred to as “tenth indication” ) on similarity computation, or the model 220-1. In other words, the device 510 may receive the tenth indication and/or the model 220-1 from the device 520. In some embodiments, the tenth indication on similarity computation may indicate one of: whether a similarity in the similarity information is above or equal to a configured similarity threshold, or an exact value of the similarity, or whether a difference in the similarity information is below or equal to a configured difference, or an exact value of the difference.

In some embodiments, if the similarity is below or equal to a configured similarity threshold or a difference in the similarity information is above or equal to a configured difference threshold, the device 510 may receive the model 220-1 from the device 520. In this case, after receiving the model 220-1, the device 510 may perform a performance monitoring on the model 220-1. The device 510 may transmit to the device 520 a result of the performance monitoring. In other words, the device 520 may receive the result of the performance monitoring from the device 510. The result of the performance monitoring may be transmitted in one of: UCI, uplink MAC CE, or uplink RRC message.

Alternatively, the device 520 may transmits the model 220-1 to the device 510 according to a periodical configuration for model delivery. In other words, the device 510 may receive the model 220-1 from the device 520. In this case, the device 510 may determine the similarity information between the model 220-1 and the model 220-2. The device 510 may report the similarity information to the device 520. The device 520 may transmit to the device 510 an indication (referred to as “eleventh indication” ) regarding whether the terminal device is to monitor performance of the model 220-1. In other words, the device 510 may receive the eleventh indication from the device 520. In some embodiments, if the device 510 monitors the model 220-1 based on the ninth indication, the device 510 may report a performance measurement to the device 520. In some embodiments, the device 510 may start a timer (referred to as “fourth timer” ) upon a reception of the ninth indication. In this case, the device 510 may stop the fourth timer upon a transmission of the performance measurement. For example, based on the downlink indication, the device 510 may decide whether to trigger model monitoring and feedback the performance measurement towards the device 510. By way of example, the fourth timer can be configured towards the device 510. When the device 510 receive the indication for triggering of model monitoring, the device 510 may start the timer. When the device 510 feedbacks the performance measurement towards the network or feedbacks that radio link failure (RLF) happened, the device 510 may stop the fourth timer and the device 520 may continue to perform model training. Once the fourth timer expires, the device 520 may re-send the indication to trigger the device 510 to perform model monitoring.

In some embodiments, the device 520 may transmit an indication (referred to as “twelfth indication” ) regarding whether the device 510 is allowed to continue monitoring the model 220-1. In other words, the device 510 may receive the twelfth indication from the device 520. The twelfth indication may be transmitted in one of: DCI, MAC CE or RRC message.

In some embodiments, the device 510 may transmit (5040) the similarity information to the device 520. In other words, the device 520 may receive the similarity information from the device 510. The similarity information may be transmitted in one of: UCI, uplink MAC CE, or uplink RRC message. In some embodiments, the reporting of the similarity information may be one side reporting. For example, the similarity information may be reported from the device 510 to the device 520. In some embodiments, the device 520 may transmit a request for the similarity information to the device 510. In this case, the similarity information may be reported based on the request. The device 520 may transmit to the device 510, a configuration message indicating that the similarity information is received.

The device 520 determines (5050) the similarity information. The device 520 determines (5060) a processing on the model 220-1. Determination (5050) of the similarity information at the device 520 may be similar as determination (5020) at the device 510. Determination (5060) of the processing at the device 520 may be similar as determination (5030) at the device 510. Details of the determination (5050) of the similarity information and the determination (5060) of the processing are omitted for clarity purpose. In some embodiments, the device 520 may transmit the determined (5060) the similarity information to the device 510.

FIG. 6 illustrates a flowchart of a communication method 600 implemented at a terminal device in accordance with some embodiments of the present disclosure. The method 600 may be implemented at the terminal device 110.

At block 610, the terminal device obtains similarity information between a first model and a second model. The similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model. The first model and the second model are associated with a parameter related to a communication between the terminal device and the network device. A relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model are deduced from the same original model; the second model is trained after the first model in the timeline; or at least one parameter corresponding to the first model is the same with the second model.

At block 620, the terminal device determines, based at least on the similarity information, a processing on the second model. The processing comprises at least one of: a model training, a model monitoring, or a model delivery.

In some embodiments, the terminal device may receive, from a network device, a similarity determination configuration. In some embodiments, the terminal device may determine the similarity information based on the similarity determination configuration.

In some embodiments, the similarity determination configuration comprises at least one of: model identification information, a similarity computation algorithm, a type of the similarity information, a sample dataset for training a model, or a model use case. In some embodiments, the similarity determination configuration is received periodically. In some embodiments, the similarity determination configuration further comprises at least one of: a periodicity for updating the sample dataset, or a starting time at which the sample dataset is updated.

In some embodiments, the second model is trained at the terminal device and monitored at a network device. In some embodiments, the similarity determination configuration further comprises one of: a periodical configuration for the model delivery of the second model, or an event-based configuration for the model delivery of the second model.

In some embodiments, the periodical configuration comprises at least one of: a periodicity for the model delivery of the second model, a starting time at which the second model is delivered, or a first indication for enabling or disabling delivering the second model.

In some embodiments, the event-based configuration comprises a similarity threshold. In some embodiments, the terminal device may in accordance with a determination that a similarity in the similarity information is below or equal to the similarity threshold, deliver the second model to the network device. In some embodiments, the event-based configuration comprises a difference threshold. In some embodiments, the terminal device may in accordance with a determination that a difference in the similarity information is above or equal to the difference threshold, deliver the second model to the network device.

In some embodiments, the terminal device may deliver the second model to the network device according to a periodical configuration for model delivery. In some embodiments, the terminal device may cause the model training on the second model to be stopped. In some embodiments, the terminal device may receive, from the network device, a second indication regarding whether the terminal device is allowed to continue training the second model. In some embodiments, the terminal device may in accordance with a determination that the second indication indicating the terminal device is allowed to continue training the second model, performing the model training on the second model until a next periodicity for model delivery. In some embodiments, the terminal device may in accordance with a determination that the second indication indicating the terminal device is not allowed to continue training the second model, continue causing the model training on the second model to be stopped for a predetermined duration.

In some embodiments, the terminal device may start a first timer upon a reception of the second indication. In some embodiments, the terminal device may after the first timer expires, continue to perform the model training on the second model.

In some embodiments, the terminal device may after delivering the second model to the network device, cause the model training on the second model to be stopped. In some embodiments, the terminal device may before delivering the second model to the network device, cause the model training on the second model to be stopped.

In some embodiments, the terminal device may determine whether a similarity in the similarity information a configured similarity threshold. In some embodiments, the terminal device may in accordance with a determination that the similarity is above or equal to the configured similarity threshold, cause delivering the second model to a network device to be skipped; and transmit to the network device a third indication indicating that the delivering of the second model to be skipped for a predetermined period.

In some embodiments, the terminal device may in accordance with a determination that the similarity is below or equal to the configured similarity threshold, deliver the second model to the network device according to a periodical configuration for model delivery. In some embodiments, the terminal device may determine whether a difference in the similarity information is above or equal to a configured difference threshold. In some embodiments, the terminal device may in accordance with a determination that the difference is below or equal to the configured difference threshold, cause delivering the second model to a network device to be skipped. In some embodiments, the terminal device may transmit to the network device a third indication indicating that the delivering of the second model to be skipped for the predetermined period. In some embodiments, the terminal device may in accordance with a determination that the difference is above or equal to the configured difference threshold, deliver the second model to the network device according to a periodical configuration for model delivery.

In some embodiments, the terminal device may determine whether a similarity in the similarity information is above or equal to a configured similarity threshold. In some embodiments, the terminal device may in accordance with a determination that the similarity is above or equal to the configured similarity threshold and a performance metric for the second model is above or equal to a configured performance threshold, cause the model training on the second model to be stopped for a determined duration. In some embodiments, the terminal device may determine whether a difference in the similarity information is above or equal to a configured difference threshold. In some embodiments, the terminal device may in accordance with a determination that the difference is below or equal to the configured difference threshold and a performance metric for the second model is above the configured performance threshold, cause the model training on the second model to be stopped for a determined duration. In some embodiments, the terminal device may in accordance with a determination that a third timer starting from a time point when the similarity information is obtained expires, continue to performing the model training on the second model.

In some embodiments, the second model is trained and monitored at the terminal device. In some embodiments, the similarity determination configuration further comprises a model training related configuration.

In some embodiments, the model training related configuration comprises a performance metric threshold. In some embodiments, the terminal device may in accordance with a determination that a monitored performance metric of the second model is above or equal to the performance metric threshold, cause the model training on the second model to be stopped.

In some embodiments, the model training related configuration comprises a similarity threshold. In some embodiments, the terminal device may in accordance with a determination that a similarity in the similarity information is above or equal to the similarity threshold, cause the model training on the second model to be stopped. In some embodiments, the terminal device may in accordance with a determination that the similarity is below or equal to the similarity threshold, deliver the second model to the network device.

In some embodiments, the model training related configuration comprises a difference threshold. In some embodiments, the terminal device may in accordance with a determination that a difference in the similarity information is below or equal to the difference threshold, cause the model training on the second model to be stopped. In some embodiments, the terminal device may in accordance with a determination that the difference is above or equal to the difference threshold, deliver the second model to the network device.

In some embodiments, the terminal device may transmit, to the network device, a fourth indication indicating at least one of: the performance metric threshold is met, the performance metric threshold is met, the similarity threshold is not met, the similarity threshold is met, the performance metric threshold is not met while the similarity threshold is met, or the performance metric threshold is met while the similarity threshold is not met.

In some embodiments, the second model is trained at a network device and monitored at the terminal device. In some embodiments, the similarity determination configuration further comprises one of: a periodical configuration for the model delivery of the second model, or an event-based configuration for the model delivery of the second model. In some embodiments, the periodical configuration comprises at least one of: a periodicity for the model delivery of the second model, a starting time at which the second model is delivered, or a fourth indication for enabling or disabling downloading the second model.

In some embodiments, the terminal device may in accordance with a determination that a similarity in the similarity information is below or equal to a first similarity threshold in the event-based configuration, receive, from the network device, a sixth indication indicating that the second model is to be delivered. In some embodiments, the terminal device may in accordance with a determination that the similarity is below or equal to a second similarity threshold in the event-based configuration, perform the model monitoring on the second model. In some embodiments, the terminal device may in accordance with a determination that the similarity is above or equal to the second similarity threshold, transmit, to the network device, a seventh indication for further model training on the second model.

In some embodiments, the terminal device may in accordance with a determination that a difference in the similarity information is above or equal to a first difference threshold in the event-based configuration, receive, from the network device, an eighth indication indicating that the second model is to be delivered. In some embodiments, the terminal device may in accordance with a determination that the difference is above or equal to a second difference threshold in the event-based configuration, performing the model monitoring on the second model. In some embodiments, the terminal device may in accordance with a determination that the difference is below or equal to the second difference threshold, transmit, to the network device, a ninth indication for further model training on the second model.

In some embodiments, the terminal device may receive the second model from the network device according to a periodical configuration for model delivery. In some embodiments, the terminal device may after receiving the second model, perform a performance monitoring on the second model. In some embodiments, the terminal device may transmit to the network device a result of the performance monitoring.

In some embodiments, the terminal device may receive, from the network device, at least one of the following: a tenth indication on similarity computation, or the second model. In some embodiments, the tenth indication on similarity computation indicates one of: whether a similarity in the similarity information is above or equal to a configured similarity threshold, or an exact value of the similarity, or whether a difference in the similarity information is below or equal to a configured difference, or an exact value of the difference.

In some embodiments, the terminal device may in accordance with a determination that a similarity in the similarity information is below or equal to a configured similarity threshold or a determination that a difference in the similarity information is above or equal to a configured difference threshold, receive the second model from the network device. In some embodiments, the terminal device may after receiving the second model, perform a performance monitoring on the second model. In some embodiments, the terminal device may transmit to the network device a result of the performance monitoring.

In some embodiments, the terminal device may receive the second model from the network device according to a periodical configuration for model delivery. In some embodiments, the terminal device may determine the similarity information between the first model and the second model. In some embodiments, the terminal device may report the similarity information to the network device. In some embodiments, the terminal device may receive, from the network device, an eleventh indication regarding whether the terminal device is to monitor performance of the second model. In some embodiments, the terminal device may in accordance with a determination that the terminal device monitors the second model based on the ninth indication, report a performance measurement to the network device.

In some embodiments, the terminal device may start a fourth timer upon a reception of the ninth indication. In some embodiments, the terminal device may stop the fourth timer upon a transmission of the performance measurement.

In some embodiments, the terminal device may receive, from the network device, a twelfth indication regarding whether the terminal device is allowed to continue monitoring the second model. In some embodiments, the terminal device may report the similarity information to the network device.

FIG. 7 illustrates a flowchart of a communication method 700 implemented at a terminal device in accordance with some embodiments of the present disclosure. For the purpose of discussion, the method 700 will be described from the perspective of the terminal device 110 in FIG. 1.

At block 710, the network device transmits a similarity determination configuration to a terminal device. In some embodiments, the similarity determination configuration comprises at least one of: model identification information, a similarity computation algorithm, a type of the similarity information, a sample dataset for training a model, or a model use case. In some embodiments, the similarity determination configuration is transmitted periodically. In some embodiments, the similarity determination configuration further comprises at least one of: a periodicity for updating the sample dataset, or a starting time at which the sample dataset is updated.

At block 720, the network device obtains similarity information between a first model and a second model. The similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model. The first model and the second model are associated with a parameter related to a communication between the terminal device and the network device. A relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model are deduced from a same original model; the second model is trained after the first model in the timeline; or at least one parameter corresponding to the first model is same with the second model.

At block 730, the network device determines, based at least on the similarity information, a processing on the second model. The processing comprises at least one of: a model training, a model monitoring, or a model delivery.

In some embodiments, the event-based configuration comprises a similarity threshold. In some embodiments, the event-based configuration comprises a difference threshold.

In some embodiments, the network device may receive the second model from the terminal device according to a periodical configuration for model delivery. In some embodiments, the network device may transmit, to the terminal device, a second indication regarding whether the terminal device is allowed to continue training the second model.

In some embodiments, the network device may determine whether a similarity in the similarity information between the first model and the second model is above or equal to a configured similarity threshold. In some embodiments, the network device may in accordance with a determination that the similarity is above or equal to the configured similarity threshold, receive from the terminal device a third indication indicating that the delivering of the second model to be skipped for a predetermined period.

In some embodiments, the network device may determine whether a difference in the similarity information between the first model and the second model is above or equal to a configured difference threshold. In some embodiments, the network device may in accordance with a determination that the difference is below or equal to the configured difference threshold, receive from the terminal device the third indication indicating that the delivering of the second model to be skipped for a predetermined period.

In some embodiments, the second model is trained and monitored at the terminal device. In some embodiments, the similarity determination configuration further comprises a model training related configuration. In some embodiments, the model training related configuration comprises a performance metric threshold.

In some embodiments, the model training related configuration comprises a similarity threshold. In some embodiments, the network device may in accordance with a determination that the similarity is below or equal to the similarity threshold, receive the second model from the terminal device.

In some embodiments, the model training related configuration comprises a difference threshold. In some embodiments, the network device may in accordance with a determination that a difference in the similarity information is above or equal to the difference threshold, receive the second model from the terminal device.

In some embodiments, the network device may receive, from the terminal device, a fourth indication indicating at least one of: the performance metric threshold is met, the performance metric threshold is met, the similarity threshold is not met, the similarity threshold is met, the performance metric threshold is not met while the similarity threshold is met, or the performance metric threshold is met while the similarity threshold is not met.

In some embodiments, the second model is trained at a network device and monitored at the terminal device. In some embodiments, the similarity determination configuration further comprises one of: a periodical configuration for the model delivery of the first, or an event-based configuration for the model delivery of the first. In some embodiments, the periodical configuration comprises at least one of: a periodicity for the model delivery of the second model, a starting time at which the second model is delivered, or a fourth for enabling or disabling downloading the second model.

In some embodiments, the network device may in accordance with a determination that the similarity is below or equal to a first similarity threshold in the event-based configuration, transmit to the terminal device a sixth indication indicating that the second model is to be downloaded. In some embodiments, the network device may in accordance with a determination that the similarity is above or equal to the second similarity threshold, receive from the terminal device a seventh indication for further model training on the second model.

In some embodiments, the network device may in accordance with a determination that a difference in the similarity information is above or equal to a first difference threshold in the event-based configuration, transmit to the terminal device, an eighth indication indicating that the second model is to be delivered. In some embodiments, the network device may in accordance with a determination that the difference is above or equal to a second difference threshold in the event-based configuration and in accordance with a determination that the difference is below or equal to the second difference threshold, receive, from the terminal device, a ninth indication for further model training on the second model.

In some embodiments, the network device may receive from the terminal device a result of a performance monitoring on the second model. In some embodiments, the network device may transmit to the terminal device at least one of the following: a tenth indication on similarity computation, or the second model.

In some embodiments, the tenth indication on similarity computation indicates one of: whether a similarity in the similarity information is above or equal to a configured similarity threshold, or an exact value of the similarity, or whether a difference in the similarity information is below or equal to a configured difference, or an exact value of the difference.

In some embodiments, the network device may in accordance with a determination that a similarity in the similarity information is below or equal to a configured similarity threshold or a determination that a difference in the similarity information is above or equal to a configured difference threshold, transmit the second model to the terminal device. In some embodiments, the network device may receive from the terminal device a result of performance monitoring on the second model.

In some embodiments, the network device may transmit the second model to the terminal device according to a periodical configuration for model delivery. In some embodiments, the network device may receive the similarity from the terminal device. In some embodiments, the network device may transmit to the terminal device an eleventh indication regarding whether the terminal device is to monitor performance of the second model based on the similarity information.

In some embodiments, the network device may transmit to the terminal device a twelfth indication regarding whether the terminal device is allowed to continue monitoring the second model. In some embodiments, the network device may receive the similarity information from the terminal device.

FIG. 8 is a simplified block diagram of a device 800 that is suitable for implementing embodiments of the present disclosure. The device 800 can be considered as a further example implementation of any of the devices as shown in FIG. 1. Accordingly, the device 800 can be implemented at or as at least a part of the terminal device 110, or the network device 120.

As shown, the device 800 includes a processor 810, a memory 820 coupled to the processor 810, a suitable transceiver 840 coupled to the processor 810, and a communication interface coupled to the transceiver 840. The memory 810 stores at least a part of a program 830. The transceiver 840 may be for bidirectional communications or a unidirectional communication based on requirements. The transceiver 840 may include at least one of a transmitter 842 and a receiver 844. The transmitter 842 and the receiver 844 may be functional modules or physical entities. The transceiver840 has at least one antenna to facilitate communication, though in practice an Access Node mentioned in this application may have several ones. The communication interface may represent any interface that is necessary for communication with other network elements, such as X2/Xn interface for bidirectional communications between eNBs/gNBs, S1/NG interface for communication between a Mobility Management Entity (MME) /Access and Mobility Management Function (AMF) /SGW/UPF and the eNB/gNB, Un interface for communication between the eNB/gNB and a relay node (RN) , or Uu interface for communication between the eNB/gNB and a terminal device.

The program 830 is assumed to include program instructions that, when executed by the associated processor 810, enable the device 800 to operate in accordance with the embodiments of the present disclosure, as discussed herein with reference to FIGS. 1 to 7. The embodiments herein may be implemented by computer software executable by the processor 810 of the device 800, or by hardware, or by a combination of software and hardware. The processor 810 may be configured to implement various embodiments of the present disclosure. Furthermore, a combination of the processor 810 and memory 820 may form processing means 850 adapted to implement various embodiments of the present disclosure.

The memory 820 may be of any type suitable to the local technical network and may be implemented using any suitable data storage technology, such as a non-transitory computer readable storage medium, semiconductor based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory and removable memory, as non-limiting examples. While only one memory 820 is shown in the device 800, there may be several physically distinct memory modules in the device 800. The processor 810 may be of any type suitable to the local technical network, and may include one or more of general purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs) and processors based on multicore processor architecture, as non-limiting examples. The device 800 may have multiple processors, such as an application specific integrated circuit chip that is slaved in time to a clock which synchronizes the main processor.

According to embodiments of the present disclosure, a terminal device comprising a circuitry is provided. The circuitry is configured to: at least one memory storing instructions that, when executed by the at least one processor, cause the terminal device at least to: obtain similarity information between a first model and a second model, wherein the similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device, and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model is deduced from a same original model; the second model is trained after the first model in a timeline; or at least one parameter corresponding to the first model is the same with the second model. According to embodiments of the present disclosure, the circuitry may be configured to perform any method implemented by the terminal device as discussed above.

According to embodiments of the present disclosure, a network device comprising a circuitry is provided. The circuitry is configured to: at least one memory storing instructions that, when executed by the at least one processor, cause the network device at least to: transmit a similarity determination configuration to a terminal device; obtain similarity information between a first model and a second model, wherein the similarity information comprises a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device; and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model are deduced from a same original model; the second model is trained after the first model in a timeline; or at least one parameter corresponding to the first model is the same with the second model. According to embodiments of the present disclosure, the circuitry may be configured to perform any method implemented by the first network device as discussed above.

The term “circuitry” used herein may refer to hardware circuits and/or combinations of hardware circuits and software. For example, the circuitry may be a combination of analog and/or digital hardware circuits with software/firmware. As a further example, the circuitry may be any portions of hardware processors with software including digital signal processor (s) , software, and memory (ies) that work together to cause an apparatus, such as a terminal device or a network device, to perform various functions. In a still further example, the circuitry may be hardware circuits and or processors, such as a microprocessor or a portion of a microprocessor, that requires software/firmware for operation, but the software may not be present when it is not needed for operation. As used herein, the term circuitry also covers an implementation of merely a hardware circuit or processor (s) or a portion of a hardware circuit or processor (s) and its (or their) accompanying software and/or firmware.

In summary, embodiments of the present disclosure provide the following aspects.

In an aspect, it is proposed a terminal device comprising: a processor, configured to cause the terminal device to: obtain similarity information between a first model and a second model, wherein the similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device, and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model is deduced from a same original model; the second model is trained after the first model in a timeline; or at least one parameter corresponding to the first model is same with the second model.

In some solutions, the processor is further configured to cause the terminal device to: receive, from a network device, a similarity determination configuration; and determine the similarity information based on the similarity determination configuration.

In some solutions, the similarity determination configuration comprises at least one of: model identification information, a similarity computation algorithm, a type of the similarity information, a sample dataset for training a model, or a model use case.

In some solutions, the similarity determination configuration is received periodically.

In some solutions, the similarity determination configuration further comprises at least one of: a periodicity for updating the sample dataset, or a starting time at which the sample dataset is updated.

In some solutions, the second model is trained at the terminal device and monitored at a network device.

In some solutions, the similarity determination configuration further comprises one of: a periodical configuration for the model delivery of the second model, or an event-based configuration for the model delivery of the second model.

In some solutions, the periodical configuration comprises at least one of: a periodicity for the model delivery of the second model, a starting time at which the second model is delivered, or a first indication for enabling or disabling delivering the second model.

In some solutions, the event-based configuration comprises a similarity threshold, and wherein the processor is further configured to cause the terminal device to: in accordance with a determination that a similarity in the similarity information is below or equal to the similarity threshold, deliver the second model to the network device; or wherein the event-based configuration comprises a difference threshold, and wherein the processor is further configured to cause the terminal device to: in accordance with a determination that a difference in the similarity information is above or equal to the difference threshold, deliver the second model to the network device.

In some solutions, the processor is further configured to cause the terminal device to: deliver the second model to the network device according to a periodical configuration for model delivery.

In some solutions, the processor is further configured to cause the terminal device to: cause the model training on the second model to be stopped; receive, from the network device, a second indication regarding whether the terminal device is allowed to continue training the second model; and in accordance with a determination that the second indication indicating the terminal device is allowed to continue training the second model, performing the model training on the second model until a next periodicity for model delivery; or in accordance with a determination that the second indication indicating the terminal device is not allowed to continue training the second model, continue causing the model training on the second model to be stopped for a predetermined duration.

In some solutions, the processor is further configured to cause the terminal device to: start a first timer upon a reception of the second indication; and after the first timer expires, continue to perform the model training on the second model.

In some solutions, the processor is further configured to cause the terminal device to: after delivering the second model to the network device, cause the model training on the second model to be stopped; or before delivering the second model to the network device, cause the model training on the second model to be stopped.

In some solutions, the processor is further configured to cause the terminal device to: determine whether a similarity in the similarity information a configured similarity threshold; in accordance with a determination that the similarity is above or equal to the configured similarity threshold, cause delivering the second model to a network device to be skipped; and transmit to the network device a third indication indicating that the delivering of the second model to be skipped for a predetermined period; or in accordance with a determination that the similarity is below or equal to the configured similarity threshold, deliver the second model to the network device according to a periodical configuration for model delivery; or wherein the processor is further configured to cause the terminal device to: determine whether a difference in the similarity information is above or equal to a configured difference threshold; in accordance with a determination that the difference is below or equal to the configured difference threshold, cause delivering the second model to a network device to be skipped; and transmit to the network device a third indication indicating that the delivering of the second model to be skipped for the predetermined period; or in accordance with a determination that the difference is above or equal to the configured difference threshold, deliver the second model to the network device according to a periodical configuration for model delivery.

In some solutions, the processor is further configured to cause the terminal device to: determine whether a similarity in the similarity information is above or equal to a configured similarity threshold; and in accordance with a determination that the similarity is above or equal to the configured similarity threshold and a performance metric for the second model is above or equal to a configured performance threshold, cause the model training on the second model to be stopped for a determined duration; or wherein the processor is further configured to cause the terminal device to: determine whether a difference in the similarity information is above or equal to a configured difference threshold; and in accordance with a determination that the difference is below or equal to the configured difference threshold and a performance metric for the second model is above the configured performance threshold, cause the model training on the second model to be stopped for a determined duration.

In some solutions, the processor is further configured to cause the terminal device to: in accordance with a determination that a third timer starting from a time point when the similarity information is obtained expires, continue to performing the model training on the second model.

In some solutions, the second model is trained and monitored at the terminal device.

In some solutions, the similarity determination configuration further comprises a model training related configuration.

In some solutions, the model training related configuration comprises a performance metric threshold, and wherein the processor is further configured to cause the terminal device to: in accordance with a determination that a monitored performance metric of the second model is above or equal to the performance metric threshold, cause the model training on the second model to be stopped.

In some solutions, the model training related configuration comprises a similarity threshold, and wherein the processor is further configured to cause the terminal device to: in accordance with a determination that a similarity in the similarity information is above or equal to the similarity threshold, cause the model training on the second model to be stopped; and in accordance with a determination that the similarity is below or equal to the similarity threshold, deliver the second model to the network device; or wherein the model training related configuration comprises a difference threshold, and wherein the processor is further configured to cause the terminal device to: in accordance with a determination that a difference in the similarity information is below or equal to the difference threshold, cause the model training on the second model to be stopped; and in accordance with a determination that the difference is above or equal to the difference threshold, deliver the second model to the network device.

In some solutions, the processor is further configured to cause the terminal device to: transmit, to the network device, a fourth indication indicating at least one of: the performance metric threshold is met, the performance metric threshold is met, the similarity threshold is not met, the similarity threshold is met, the performance metric threshold is not met while the similarity threshold is met, or the performance metric threshold is met while the similarity threshold is not met.

In some solutions, the second model is trained at a network device and monitored at the terminal device.

In some solutions, the periodical configuration comprises at least one of: a periodicity for the model delivery of the second model, a starting time at which the second model is delivered, or a fourth indication for enabling or disabling downloading the second model.

In some solutions, the processor is further configured to cause the terminal device to: in accordance with a determination that a similarity in the similarity information is below or equal to a first similarity threshold in the event-based configuration, receive, from the network device, a sixth indication indicating that the second model is to be delivered; or wherein the processor is further configured to cause the terminal device to: in accordance with a determination that the similarity is below or equal to a second similarity threshold in the event-based configuration, performing the model monitoring on the second model; and in accordance with a determination that the similarity is above or equal to the second similarity threshold, transmit, to the network device, a seventh indication for further model training on the second model; or wherein the processor is further configured to cause the terminal device to: in accordance with a determination that a difference in the similarity information is above or equal to a first difference threshold in the event-based configuration, receive, from the network device, an eighth indication indicating that the second model is to be delivered; or wherein the processor is further configured to cause the terminal device to: in accordance with a determination that the difference is above or equal to a second difference threshold in the event-based configuration, performing the model monitoring on the second model; and in accordance with a determination that the difference is below or equal to the second difference threshold, transmit, to the network device, a ninth indication for further model training on the second model.

In some solutions, the processor is further configured to cause the terminal device to: receive the second model from the network device according to a periodical configuration for model delivery; after receiving the second model, perform a performance monitoring on the second model; and transmit to the network device a result of the performance monitoring.

In some solutions, the processor is further configured to cause the terminal device to: receive, from the network device, at least one of the following: a tenth indication on similarity computation, or the second model.

In some solutions, the tenth indication on similarity computation indicates one of: whether a similarity in the similarity information is above or equal to a configured similarity threshold, or an exact value of the similarity, or whether a difference in the similarity information is below or equal to a configured difference, or an exact value of the difference.

In some solutions, the processor is further configured to cause the terminal device to: in accordance with a determination that a similarity in the similarity information is below or equal to a configured similarity threshold or a determination that a difference in the similarity information is above or equal to a configured difference threshold, receive the second model from the network device; after receiving the second model, perform a performance monitoring on the second model; and transmit to the network device a result of the performance monitoring.

In some solutions, the processor is further configured to cause the terminal device to: receive the second model from the network device according to a periodical configuration for model delivery; determine the similarity information between the first model and the second model; report the similarity information to the network device; receive, from the network device, an eleventh indication regarding whether the terminal device is to monitor performance of the second model; and in accordance with a determination that the terminal device monitors the second model based on the ninth indication, report a performance measurement to the network device.

In some solutions, the processor is further configured to cause the terminal device to: start a fourth timer upon a reception of the ninth indication; and stop the fourth timer upon a transmission of the performance measurement.

In some solutions, the processor is further configured to cause the terminal device to: receive, from the network device, a twelfth indication regarding whether the terminal device is allowed to continue monitoring the second model.

In some solutions, the processor is further configured to cause the terminal device to: report the similarity information to the network device.

In an aspect, it is proposed a network device comprising: a processor, configured to cause the network device to: transmit a similarity determination configuration to a terminal device; obtain similarity information between a first model and a second model, wherein the similarity information comprises a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and the network device; and wherein a relationship between the first model and the second model comprises at least one of the followings: the first model and the second model have a same identifier; the first model and the second model have a same dataset for input; the first model and the second model are deduced from a same original model; the second model is trained after the first model in the timeline; or at least one parameter corresponding to the first model is same with the second model.

In some solutions, the similarity determination configuration is transmitted periodically.

In some solutions, the event-based configuration comprises a similarity threshold, or wherein the event-based configuration comprises a difference threshold.

In some solutions, the processor is further configured to cause the network device to: receive the second model from the terminal device according to a periodical configuration for model delivery.

In some solutions, the processor is further configured to cause network terminal device to: transmit, to the terminal device, a second indication regarding whether the terminal device is allowed to continue training the second model.

In some solutions, the processor is further configured to cause the network device to: determine whether a similarity in the similarity information between the first model and the second model is above or equal to a configured similarity threshold; and in accordance with a determination that the similarity is above or equal to the configured similarity threshold, receive from the terminal device a third indication indicating that the delivering of the second model to be skipped for a predetermined period; or wherein the processor is further configured to cause the network device to: determine whether a difference in the similarity information between the first model and the second model is above or equal to a configured difference threshold; and in accordance with a determination that the difference is below or equal to the configured difference threshold, receive from the terminal device the third indication indicating that the delivering of the second model to be skipped for a predetermined period.

In some solutions, wherein the second model is trained and monitored at the terminal device.

In some solutions, wherein the similarity determination configuration further comprises a model training related configuration.

In some solutions, wherein the model training related configuration comprises a performance metric threshold.

In some solutions, wherein the model training related configuration comprises a similarity threshold, and wherein the processor is further configured to cause the network device to: in accordance with a determination that the similarity is below or equal to the similarity threshold, receive the second model from the terminal device; or wherein the model training related configuration comprises a difference threshold, and wherein the processor is further configured to cause the network device to: in accordance with a determination that a difference in the similarity information is above or equal to the difference threshold, receive the second model from the terminal device.

In some solutions, wherein the processor is further configured to cause the network device to: receive, from the terminal device, a fourth indication indicating at least one of: the performance metric threshold is met, the performance metric threshold is met, the similarity threshold is not met, the similarity threshold is met, the performance metric threshold is not met while the similarity threshold is met, or the performance metric threshold is met while the similarity threshold is not met.

In some solutions, wherein the second model is trained at a network device and monitored at the terminal device.

In some solutions, wherein the similarity determination configuration further comprises one of: a periodical configuration for the model delivery of the first, or an event-based configuration for the model delivery.

In some solutions, wherein the periodical configuration comprises at least one of:a periodicity for the model delivery of the second model, a starting time at which the second model is delivered, or a fourth for enabling or disabling downloading the second model.

In some solutions, wherein the processor is further configured to cause the network device to: in accordance with a determination that the similarity is below or equal to a first similarity threshold in the event-based configuration, transmit to the terminal device a sixth indication indicating that the second model is to be downloaded; or wherein the processor is further configured to cause the terminal device to: in accordance with a determination that the similarity is above or equal to the second similarity threshold, receive from the terminal device a seventh indication for further model training on the second model; or wherein the processor is further configured to cause the terminal device to: in accordance with a determination that a difference in the similarity information is above or equal to a first difference threshold in the event-based configuration, transmit to the terminal device, an eighth indication indicating that the second model is to be delivered; or wherein the processor is further configured to cause the terminal device to: in accordance with a determination that the difference is above or equal to a second difference threshold in the event-based configuration and in accordance with a determination that the difference is below or equal to the second difference threshold, receive, from the terminal device, a ninth indication for further model training on the second model.

In some solutions, wherein the processor is further configured to cause the network device to: receive from the terminal device a result of a performance monitoring on the second model.

In some solutions, wherein the processor is further configured to cause the network device to: transmit to the terminal device at least one of the following: a tenth indication on similarity computation, or the second model.

In some solutions, wherein the tenth indication on similarity computation indicates one of: whether a similarity in the similarity information is above or equal to a configured similarity threshold, or an exact value of the similarity, or whether a difference in the similarity information is below or equal to a configured difference, or an exact value of the difference.

In some solutions, wherein the processor is further configured to cause the network device to: in accordance with a determination that a similarity in the similarity information is below or equal to a configured similarity threshold or a determination that a difference in the similarity information is above or equal to a configured difference threshold, transmit the second model to the terminal device; and receive from the terminal device a result of performance monitoring on the second model.

In some solutions, wherein the processor is further configured to cause the network device to: transmit the second model to the terminal device according to a periodical configuration for model delivery; receive the similarity from the terminal device; and transmit to the terminal device an eleventh indication regarding whether the terminal device is to monitor performance of the second model based on the similarity information.

In some solutions, wherein the processor is further configured to cause the network device to: transmit to the terminal device a twelfth indication regarding whether the terminal device is allowed to continue monitoring the second model.

In some solutions, wherein the processor is further configured to cause the network device to: receive the similarity information from the terminal device.

In an aspect, a network device comprises means for performing the method implemented by the network device discussed above.

In an aspect, a terminal device comprises means for performing the method implemented by the terminal device discussed above.

In an aspect, a computer readable medium having instructions stored thereon, the instructions, when executed on at least one processor, causing the at least one processor to perform the method implemented by the network device discussed above.

In an aspect, a computer readable medium having instructions stored thereon, the instructions, when executed on at least one processor, causing the at least one processor to perform the method implemented by the terminal device discussed above.

In an aspect, a computer program comprising instructions, the instructions, when executed on at least one processor, causing the at least one processor to perform the method implemented by the network device discussed above.

In an aspect, a computer program comprising instructions, the instructions, when executed on at least one processor, causing the at least one processor to perform the method implemented by the terminal device discussed above.

Generally, various embodiments of the present disclosure may be implemented in hardware or special purpose circuits, software, logic or any combination thereof. Some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device. While various aspects of embodiments of the present disclosure are illustrated and described as block diagrams, flowcharts, or using some other pictorial representation, it will be appreciated that the blocks, apparatus, systems, techniques or methods described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.

The present 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, being executed in a device on a target real or virtual processor, to carry out the process or method as described above with reference to FIGS. 1 to 8. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, or the like that perform particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or split between program modules as desired in various embodiments. Machine-executable instructions for program modules may be executed within a local or distributed device. In a distributed device, program modules may be located in both local and remote storage media.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowcharts and/or block diagrams to be implemented. The program code may execute entirely on a machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

The above program code may be embodied on a machine readable medium, which may be any tangible medium that may contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine readable medium may be a machine readable signal medium or a machine readable storage medium. A machine readable medium may include but not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine readable storage medium would include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM) , a read-only memory (ROM) , an erasable programmable read-only memory (EPROM or Flash memory) , an optical fiber, a portable compact disc read-only memory (CD-ROM) , an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

Further, while operations are depicted in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Likewise, while several specific implementation details are contained in the above discussions, these should not be construed as limitations on the scope of the present disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features that are described in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment may also be implemented in multiple embodiments separately or in any suitable sub-combination.

Although the present disclosure has been described in language specific to structural features and/or methodological acts, it is to be understood that the present disclosure defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.

Claims

A terminal device, comprising:

a processor, configured to cause the terminal device to:

obtain similarity information between a first model and a second model, wherein the similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and

determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and

wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device, and

wherein a relationship between the first model and the second model comprises at least one of the followings:

the first model and the second model have a same identifier;

the first model and the second model have a same dataset for input;

the first model and the second model are deduced from a same original model;

the second model is trained after the first model in a timeline; or

at least one parameter corresponding to the first model is same with the second model.
The terminal device of claim 1, wherein the processor is further configured to cause the terminal device to:

receive, from the network device, a similarity determination configuration; and

determine the similarity information based on the similarity determination configuration.
The terminal device of claim 2, wherein the similarity determination configuration comprises at least one of:

model identification information,

similarity computation algorithm information,

a type of the similarity information,

a sample dataset for training a model, or

a model use case.
The terminal device of claim 3, wherein the similarity determination configuration comprises at least one of:

a periodicity for updating the sample dataset, or

a starting time at which the sample dataset is updated.
The terminal device of claim 1, wherein the processor is further configured to cause the terminal device to:

cause the model training on the second model to be stopped;

receive, from the network device, an indication regarding whether the terminal device is allowed to continue training the second model; and

in accordance with a determination that the indication indicates the terminal device is allowed to continue training the second model, perform the model training on the second model until a next periodicity for model delivery; or

in accordance with a determination that the indication indicates the terminal device is not allowed to continue training the second model, continue causing the model training on the second model to be stopped for a predetermined duration.
The terminal device of claim 1, wherein the processor is further configured to cause the terminal device to:

determine whether a similarity in the similarity information is above or equal to a configured similarity threshold;

in accordance with a determination that the similarity is above or equal to the configured similarity threshold,

cause delivering the second model to the network device to be skipped; and

transmit to the network device an indication indicating that the delivering of the second model to be skipped for a predetermined period; or

in accordance with a determination that the similarity is below the configured similarity threshold, deliver the second model to the network device according to a periodical configuration for model delivery.
The terminal device of claim 1, wherein the processor is further configured to cause the terminal device to:

determine whether a similarity in the similarity information is above or equal to a configured similarity threshold; and

in accordance with a determination that the similarity is above or equal to the configured similarity threshold and a performance metric for the second model is above or equal to a configured performance threshold, cause the model training on the second model to be stopped for a determined duration.
The terminal device of claim 1, wherein the model training related configuration comprises a similarity threshold, and

wherein the processor is further configured to cause the terminal device to:

in accordance with a determination that a similarity in the similarity information is above or equal to the similarity threshold, cause the model training on the second model to be stopped; and

in accordance with a determination that the similarity is below the similarity threshold, deliver the second model to the network device.
The terminal device of claim 8, wherein the processor is further configured to cause the terminal device to:

transmit, to the network device, an indication indicating at least one of:

the performance metric threshold is met,

the performance metric threshold is not met,

the similarity threshold is not met,

the similarity threshold is met,

the performance metric threshold is not met while the similarity threshold is met, or

the performance metric threshold is met while the similarity threshold is not met.
The terminal device of claim 1, wherein the processor is further configured to cause the terminal device to:

in accordance with a determination that a similarity in the similarity information is below a first similarity threshold in an event-based configuration, receive, from the network device, an indication indicating that the second model is to be delivered; or

wherein the processor is further configured to cause the terminal device to:

in accordance with a determination that the similarity is below a second similarity threshold in the event-based configuration, performing the model monitoring on the second model; and

in accordance with a determination that the similarity is above or equal to the second similarity threshold, transmit, to the network device, an indication for further model training on the second model.
The terminal device of claim 1, wherein the processor is further configured to cause the terminal device to:

in accordance with a determination that a similarity in the similarity information is below a configured similarity threshold, receive the second model from the network device;

after receiving the second model, perform a performance monitoring on the second model; and

transmit to the network device a result of the performance monitoring.
The terminal device of claim 1, wherein the processor is further configured to cause the terminal device to:

receive the second model from the network device according to a periodical configuration for model delivery;

determine the similarity information between the first model and the second model;

report the similarity information to the network device;

receive, from the network device, an indication regarding whether the terminal device is to monitor performance of the second model; and

in accordance with a determination that the terminal device monitors the second model based on the indication, report a performance measurement to the network device.
A method implemented by a terminal device, comprising:

obtaining similarity information between a first model and a second model, wherein the similarity information comprises at least one of: a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and

determining, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and

wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and a network device, and

wherein a relationship between the first model and the second model comprises at least one of the followings:

the first model and the second model have a same identifier;

the first model and the second model have a same dataset for input;

the first model and the second model are deduced from a same original model;

the second model is trained after the first model in a timeline; or

at least one parameter corresponding to the first model is the same with the second model.
The method of claim 13, further comprising:

receiving, from the network device, a similarity determination configuration; and

determining the similarity information based on the similarity determination configuration.
The method of claim 13, further comprising:

causing the model training on the second model to be stopped;

receiving, from the network device, an indication regarding whether the terminal device is allowed to continue training the second model; and

in accordance with a determination that the indication indicates the terminal device is allowed to continue training the second model, performing the model training on the second model until a next periodicity for model delivery; or

in accordance with a determination that the indication indicates the terminal device is not allowed to continue training the second model, continuing causing the model training on the second model to be stopped for a predetermined duration.
The method of claim 13, further comprising:

determining whether a similarity in the similarity information is above or equal to a configured similarity threshold;

in accordance with a determination that the similarity is above or equal to the configured similarity threshold,

causing delivering the second model to the network device to be skipped; and

transmitting to the network device an indication indicating that the delivering of the second model to be skipped for a predetermined period; or

in accordance with a determination that the similarity is below the configured similarity threshold, delivering the second model to the network device according to a periodical configuration for model delivery.
The method of claim 13, further comprising:

determining whether a similarity in the similarity information is above or equal to a configured similarity threshold; and

in accordance with a determination that the similarity is above or equal to the configured similarity threshold and a performance metric for the second model is above or equal to a configured performance threshold, causing the model training on the second model to be stopped for a determined duration.
The method of claim 13, wherein the model training related configuration comprises a similarity threshold, and

wherein the method further comprises:

in accordance with a determination that a similarity in the similarity information is above or equal to the similarity threshold, causing the model training on the second model to be stopped; and

in accordance with a determination that the similarity is below the similarity threshold, delivering the second model to the network device.
The method of claim 18, further comprising:

transmitting, to the network device, an indication indicating at least one of:

the performance metric threshold is met,

the performance metric threshold is not met,

the similarity threshold is not met,

the similarity threshold is met,

the performance metric threshold is not met while the similarity threshold is met, or

the performance metric threshold is met while the similarity threshold is not met.
A network device, comprising:

a processor, configured to cause the network device to:

transmit a similarity determination configuration to a terminal device;

obtain similarity information between a first model and a second model, wherein the similarity information comprises a model similarity between a first output of the first model and a second output of the second model, or a dataset similarity between a first dataset corresponding to the first model and a second dataset corresponding to the second model; and

determine, based at least on the similarity information, a processing on the second model, wherein the processing comprises at least one of: a model training, a model monitoring, or a model delivery, and

wherein the first model and the second model are associated with a parameter related to a communication between the terminal device and the network device; and

wherein a relationship between the first model and the second model comprises at least one of the followings:

the first model and the second model have a same identifier;

the first model and the second model have a same dataset for input;

the first model and the second model are deduced from a same original model;

the second model is trained after the first model in a timeline; or

at least one parameter corresponding to the first model is same with the second model.