Device, method and system for machine learning

Abstract

Disclosed are a device, method and system for machine learning. An exemplary method comprises: obtaining a local distribution parameter set of a second device, wherein the local distribution parameter set characterizes data distribution characteristics of a local data set of the second device; and sending a trained specific model to the second device, wherein the specific model is determined at least in part on the basis of the local distribution parameter set.

Description

Devices, methods, and systems for machine learning

[0001] Priority Declaration

[0002] This application claims priority to the Chinese patent application filed on January 31, 2024, with application number 202410141992.3, and invention name “Device, method and system for machine learning”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present disclosure relates to the field of communications, and in particular, to devices, methods, and systems for machine learning. Background Art

[0004] Artificial Intelligence (AI) technology is increasingly being used. AI models can be trained through machine learning. Improved devices, methods, and systems are desirable for training and using AI models.

[0005] Summary of the Invention

[0006] The present disclosure provides devices, methods, and systems for machine learning.

[0007] One aspect of the present disclosure relates to an electronic device, comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, which, when executed by the at least one processing unit, causes the electronic device to perform the following operations: obtain a local distribution parameter set of a second device, the local distribution parameter set characterizing data distribution characteristics of a local data set of the second device; and send a trained specific model to the second device, wherein the specific model is determined at least in part based on the local distribution parameter set.

[0008] Another aspect of the present disclosure relates to an electronic device, comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, wherein the computer program code, when executed by the at least one processing unit, causes the electronic device to perform the following operations: sending a local distribution parameter set to a first device, the local distribution parameter set characterizing data distribution characteristics of a local data set of the electronic device; receiving a trained specific model from the first device, wherein the specific model is determined at least in part based on the local distribution parameter set.

[0009] Another aspect of the present disclosure relates to an electronic device, comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, wherein the computer program code, when executed by the at least one processing unit, causes the electronic device to perform the following operations: receive one or more local distribution parameter sets, each local distribution parameter set in the one or more local distribution parameter sets characterizing data distribution characteristics of a local data set of a corresponding client in one or more clients; send a task identifier to a model library, the task identifier identifying a training task associated with the one or more clients; receive one or more trained candidate models corresponding to the task identifier from the model library; and send a model list, the model list comprising a matching model selected from the one or more candidate models for each client in the one or more clients, the matching model being determined at least in part based on the corresponding local distribution parameter set in the one or more local distribution parameter sets.

[0010] Another aspect of the present disclosure relates to a computer-readable storage medium storing one or more instructions, which, when executed by one or more processing circuits of an electronic device, causes the electronic device to perform any method as described in the present disclosure.

[0011] Another aspect of the present disclosure relates to a computer program product, comprising a computer program, which implements any method as described in the present disclosure when executed by a processor.

[0012] Another aspect of the present disclosure relates to an apparatus comprising means for performing any of the methods described in the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The above and other objects and advantages of the present disclosure will be further described below in conjunction with specific embodiments and with reference to the accompanying drawings. In the accompanying drawings, identical or corresponding technical features or components will be represented by identical or corresponding reference numerals.

[0014] FIG1 shows a schematic diagram of an exemplary system for federated learning according to some embodiments of the present disclosure.

[0015] FIG2 shows an exemplary block diagram of an electronic device according to some embodiments of the present disclosure.

[0016] FIG3 shows an exemplary block diagram of an electronic device according to some embodiments of the present disclosure.

[0017] FIG4 shows an exemplary block diagram of an electronic device according to some embodiments of the present disclosure.

[0018] FIG5 shows a flowchart of a method according to some embodiments of the present disclosure.

[0019] FIG6 shows a flowchart of a method according to some embodiments of the present disclosure.

[0020] FIG7 shows a flowchart of a method according to some embodiments of the present disclosure.

[0021] FIG8 shows a schematic diagram of a process according to some embodiments of the present disclosure.

[0022] FIG8A shows a schematic diagram of a process according to some embodiments of the present disclosure.

[0023] FIG9 illustrates a correspondence between a local data set and a local distribution parameter set according to some embodiments of the present disclosure.

[0024] FIG10A illustrates an example process of calculating the similarity between a local dataset and a training dataset according to some embodiments of the present disclosure.

[0025] FIG10B shows an example of calculating a feature distance between two samples of a local dataset and a training dataset according to some embodiments of the present disclosure.

[0026] FIG11 is a block diagram illustrating a first example of an exemplary configuration of a gNB to which the techniques of this disclosure may be applied.

[0027] FIG12 is a block diagram illustrating a second example of an exemplary configuration of a gNB to which the techniques of this disclosure may be applied.

[0028] FIG. 13 is a block diagram illustrating an example of an exemplary configuration of a communication device to which the technology of the present disclosure may be applied.

[0029] FIG. 14 is a block diagram illustrating an example of an exemplary configuration of a car navigation device to which the technology of the present disclosure can be applied.

[0030] Although the embodiments described in this disclosure may be susceptible to various modifications and alternative forms, specific embodiments thereof are shown as examples in the drawings and described in detail in this disclosure. It should be understood that the drawings and detailed description thereof are not intended to limit the embodiments to the particular forms disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the claims. DETAILED DESCRIPTION

[0031] Exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings. For the sake of clarity and conciseness, not all features of the embodiments are described in the specification. However, it should be understood that many implementation-specific settings must be made in the process of implementing the embodiments in order to achieve the developer's specific goals, such as meeting those restrictions related to equipment and services, and these restrictions may vary depending on the implementation. In addition, it should be understood that although the development work may be very complex and time-consuming, it is only a routine task for those skilled in the art who benefit from the contents of this disclosure.

[0032] It should also be noted here that in order to avoid obscuring the present disclosure due to unnecessary details, only the processing steps and / or equipment structures that are closely related to at least the scheme according to the present disclosure are shown in the accompanying drawings, while other details that are not very relevant to the present disclosure are omitted.

[0033] Artificial intelligence technology is increasingly being used. Artificial intelligence models can be trained through machine learning. Generally speaking, machine learning is performed iteratively based on a training dataset. A complete training dataset typically includes an input data component and an output data component. The input data component may represent the raw data, while the output data component may include labels, predictions, or decisions corresponding to the raw data. For example, in the example of training a classification model, the output data component may be the classification label corresponding to the raw data. In another example, in the example of training a prediction model, the output data component may be the prediction result corresponding to the raw data. In another example, in the example of training a decision model, the output data component may be the decision result corresponding to the raw data. By providing a training dataset consisting of both the input and output data components to the model being trained, the model's behavior and performance can be iteratively corrected, thereby obtaining a trained model. The trained model can then be used to generate new labels, predictions, or decisions based on new input data.

[0034] A dataset that includes both input data and a complete set of output data is called a labeled dataset. A dataset that includes input data but lacks the complete set of output data is called an unlabeled dataset or a partially unlabeled dataset. Using an unlabeled or partially unlabeled dataset as a training dataset to train a model can lead to inadequate model training, thereby reducing the performance of the trained model.

[0035] The inventors of the present disclosure have discovered that the training data sets held by the model training party are not always complete. Obtaining the corresponding output data part based on the input data part of the data set usually requires sufficient expertise and experience (for example, image labeling experts, etc.) and sufficient processing time, and costs accordingly. In some cases, due to limitations in knowledge, time or cost, it is difficult for the model training party to obtain a complete output data part. This challenge is particularly prominent in the Federated Learning (FL) process described below. According to the technology of the present disclosure, the output data part corresponding to the input data part can be completed to form a complete training data set. The data completion process can be automatic, fast, accurate and / or low-cost.

[0036] Furthermore, the inventors of this disclosure have discovered that the input data in a training dataset may partially depend on the specific scenario. For example, a training image set associated with a first scenario may have different characteristics than a training image set associated with a second scenario, resulting in differences between models trained on these two sets of images. When applying a trained model to a new local image set, it is advantageous to find the trained model that best matches the local image set.

[0037] 1. Exemplary System

[0038] This section describes an example system for federated learning. Federated learning is a distributed learning technique. In a federated learning system, each of multiple training devices trains a corresponding local model using its local dataset as the training dataset. Each training device can upload the trained local model to a server. The server aggregates the local models received from multiple training devices to generate a global model.

[0039] FIG1 illustrates a schematic diagram of an exemplary system 100 for federated learning according to some embodiments of the present disclosure. System 100 may include a server 110. Server 110 may serve as a federated learning server. System 100 may also include multiple clients 120-1 through 120-5. The multiple clients may be collectively represented by 120. Each client 120 may be communicatively coupled to server 110. Each client 120 may serve as a training device for federated learning.

[0040] During federated learning, the training of the artificial intelligence model can be an iterative process including multiple training rounds. In a training round, each client 120 can perform training using the client's local dataset as the training dataset, thereby obtaining a corresponding trained local model 121. The local datasets of different clients 120 may be different, and therefore the obtained local models 121 may be different. For example, clients 120-1, 120-2, 120-3, 120-4, and 120-5 can obtain trained local models 121-1, 121-2, 121-3, 121-4, and 121-5, respectively. Each client 120 can upload the trained local model 121 to the server 110. The server 110 can aggregate the multiple local models 121 received from each client 120 to generate a global model 111. The global model 111 can then be distributed to each client 120 via the downlink between the server 110 and each client 120. Based on the received global model 111 , each client 120 may update its local model 121 .

[0041] In the next training round, each client 120 can continue to train the updated local model 121 of the client, thereby obtaining a trained local model 121. Once again, each client 120 can upload the trained local model 121 obtained in the current training round to the server 110. The server 110 can aggregate the multiple local models 121 received from the various clients 120 to generate a global model 111. The global model 111 can be distributed to the various clients 120 again for subsequent training rounds. This process can be repeated until the end condition is reached. The performance of the global model 111 can be improved by iteratively going through multiple training rounds.

[0042] During federated learning, each client uploads a local model without having to exchange local training datasets with a server or other training devices. Therefore, federated learning can leverage the training capabilities and training datasets of a large number of clients to train AI models while ensuring data security and user privacy. This helps improve training efficiency and enhance the performance of the resulting AI models.

[0043] In order to improve the training efficiency of federated learning and the generalization ability of the trained model, it is hoped that as many clients as possible will participate in the federated learning process. However, it is difficult to require all clients participating in the federated learning process to have sufficient experience, knowledge, processing power or cost to obtain a complete training data set (specifically, the complete output data portion of the client's local data set) in a conventional manner (for example, equipped with a labeling expert with specialized knowledge). Therefore, the local data set of one or more of these clients may not be a complete training data set because some input data portions of the local data set do not have corresponding output data portions. When a local data set with a missing output data portion is used as a training data set, the client's training of the local model may be insufficient, and the resulting global model is not what is expected.

[0044] On the other hand, the input data portion held by the client may vary from scenario to scenario. The input data portion obtained in the first scenario may not be accurately processed based on inherent expertise and experience (e.g., a labeling expert trained based on the second scenario) to generate an accurate output data portion. If the client uses a fixed mechanism to obtain the output data portion, the output data portion given by the fixed mechanism for the changing input data portion in the changing scenario may lack accuracy (e.g., labeling accuracy). In addition, if a single mechanism is applied to all clients participating in the federated learning process, it may not be possible to accurately adapt the input data portion of each client in the corresponding scenario. The obtained output data portion may be inaccurate. When a local dataset with an inaccurate output data portion is used as a training dataset, the client's training of the local model may be inaccurate, and the resulting global model is not expected.

[0045] It should be understood that the federated learning process described in conjunction with FIG1 is only an example of a machine learning process that faces one or more technical issues. Similar technical issues may exist in other machine learning processes. Therefore, the technology described in this disclosure is not limited to the federated learning process. For example, during model use (rather than model training), the model user also faces the problem of selecting a matching model for the local data set in a specific scenario. In addition, there are one or more other technical issues.

[0046] 2. Exemplary Equipment

[0047] Figure 2 shows an exemplary block diagram of an electronic device 200 according to some embodiments of the present disclosure. In some embodiments, the electronic device 200 can be implemented on the server side. Therefore, the electronic device 200 can be referred to as a server or a server device. For example, in the system 100 described in Figure 1, the electronic device 200 can be implemented at the server 110. The electronic device 200 can be implemented as the server itself, as a part of the server, or as a control device for controlling the server. For example, the electronic device 200 can be implemented as a chip for controlling the server. In some embodiments of the present disclosure, the electronic device 200 is implemented as the server itself, which is merely for the convenience of description and is not intended to be limiting. In other embodiments, the electronic device 200 can alternatively be implemented as a device other than a server.

[0048] According to some embodiments of the present disclosure, the electronic device 200 may include a communication unit 210 , a storage unit 220 , and a processing circuit 230 .

[0049] The communication unit 210 of the electronic device 200 can be used to receive or send wired or radio transmissions. The communication unit 210 can perform functions such as up-conversion and digital-to-analog conversion on transmitted signals, and / or perform functions such as down-conversion and analog-to-digital conversion on received signals. The communication unit 210 can be implemented using various technologies. For example, the communication unit 210 can be implemented as communication interface components such as an antenna device, radio frequency circuitry, and a portion of baseband processing circuitry. The communication unit 210 is depicted with dashed lines because it can alternatively be located within the processing circuitry 230 or external to the electronic device 200.

[0050] The storage unit 220 of the electronic device 200 can store information generated by the processing circuit 230, information received from other devices via the communication unit 210 or information to be transmitted to other devices, programs, machine code, and data used for the operation of the electronic device 200, etc. The storage unit 220 can be a volatile memory and / or a non-volatile memory. For example, the storage unit 220 can include, but is not limited to, random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), read-only memory (ROM), and flash memory. The storage unit 220 is drawn with a dotted line because it can alternatively be located within the processing circuit 230 or outside the electronic device 200.

[0051] The processing circuit 230 of the electronic device 200 can be configured to perform one or more operations, thereby providing various functions of the electronic device 200. The functions of the elements disclosed herein can be implemented using a circuit or processing circuit 230, which includes a general-purpose processor, a special-purpose processor, an integrated circuit, an ASIC ("application-specific integrated circuit"), a conventional circuit and / or a combination thereof configured or programmed to perform the disclosed functions. Processors are considered processing circuits or circuits because they include transistors and other circuits therein. In the present disclosure, a circuit, unit or device is hardware that performs or is programmed to perform the functions described. The hardware can be any hardware disclosed herein or otherwise known that is programmed or configured to perform the functions described. When the hardware is a processor that can be considered a type of circuit, the circuit, device or unit is a combination of hardware and software, with the software being used to configure the hardware and / or processor. According to some embodiments, in response to the processing circuit 230 executing the computer program code contained in the storage unit 220, the electronic device 200 can be configured to perform one or more steps performed by the server device in the present disclosure.

[0052] According to some embodiments, the electronic device 200 may be configured to obtain a local distribution parameter set of a second device, which characterizes the data distribution characteristics of the local data set of the second device. The second device may be another electronic device different from the electronic device 200. For example, when the electronic device 200 is implemented as a server for machine learning, the second device may be a client associated with the machine learning. In some embodiments, the client may be a user of a model associated with the machine learning. In other embodiments, the client may be a trainer of a model associated with the machine learning. For example, the client may be a training device associated with federated learning.

[0053] According to some embodiments, the electronic device 200 may be further configured to transmit a trained specific model to the second device, where the specific model is determined at least in part based on the local distribution parameter set.

[0054] The distribution parameter set can characterize the data distribution characteristics of the data set. The inventors of the present disclosure have found that different distribution parameter sets can be used to distinguish different data sets (for example, parts of input data obtained in different scenarios). By selecting a specific model based on the local distribution parameter set of the local data set of the second device, the selected specific model can be matched with the local data set of the second device. Such a specific model can be called a matching model. By applying the matched specific model to the local data set of the second device, the result output by the model has high accuracy. In addition, by transmitting the local distribution parameter set instead of the local data set itself, the communication burden between electronic devices can be advantageously saved, because the local data set typically contains a large amount of data. Additionally, avoiding the transmission of the local data set can maintain data security and user privacy.

[0055] In some embodiments, the processing circuit 230 may include a parameter acquisition unit 231 and a model distribution unit 232, each of which may include corresponding hardware circuits or software modules. The parameter acquisition unit 231 may be configured to perform the step of obtaining a local distribution parameter set. The model distribution unit 232 may be configured to perform the step of sending a specific model. According to some embodiments of the present disclosure, the electronic device 200 may be further configured to perform one or more additional steps performed by the server device, as described in further detail below. Accordingly, the processing circuit 230 may include one or more additional units (not shown) for performing the one or more additional steps.

[0056] Figure 3 shows an exemplary block diagram of an electronic device 300 according to some embodiments of the present disclosure. In some embodiments, the electronic device 300 can be implemented on the client side. Therefore, the electronic device 300 can be referred to as a client or a client device. For example, in the system 100 described in Figure 1, the electronic device 300 can be implemented at the client 120. The electronic device 300 can be implemented as the client itself, as a part of the client, or as a control device for controlling the client. For example, the electronic device 300 can be implemented as a chip for controlling the client. In some embodiments of the present disclosure, the electronic device 300 is implemented as the client itself, which is merely for convenience of description and is not intended to be limiting. In other embodiments, the electronic device 300 can alternatively be implemented as a device other than a client.

[0057] According to some embodiments of the present disclosure, electronic device 300 may include a communication unit 310, a storage unit 320, and a processing circuit 330. The specific implementations of communication unit 310, storage unit 320, and processing circuit 330 may be similar to the communication unit 210, storage unit 220, and processing circuit 230 described above with respect to electronic device 200, and are not further described herein. According to some embodiments, in response to processing circuit 330 executing computer program code contained in storage unit 320, electronic device 300 may be configured to perform one or more steps performed by a client device in implementing various methods according to some embodiments of the present disclosure.

[0058] According to some embodiments, the electronic device 300 may be configured to send a local distribution parameter set to the first device. The local distribution parameter set may characterize the data distribution characteristics of the local data set of the electronic device 300. The first device may be another electronic device different from the electronic device 300. In some embodiments, when the electronic device 300 is implemented as a client associated with machine learning, the first device may include a server for the machine learning. The first device may be implemented by the electronic device 200. In other embodiments, the first device may alternatively include other devices besides the server for machine learning.

[0059] According to some embodiments, the electronic device 300 may be further configured to receive a trained specific model from the first device, wherein the specific model is determined at least in part based on the local distribution parameter set of the local data set of the electronic device 300 .

[0060] In some embodiments, the processing circuit 330 may include a parameter sending unit 331 and a model receiving unit 332, each of which may include corresponding hardware circuits or software modules. The parameter sending unit 331 may be configured to perform the step of sending a local distribution parameter set. The model receiving unit 332 may be configured to perform the step of receiving a trained specific model. According to some embodiments of the present disclosure, the electronic device 300 may be further configured to perform one or more additional steps performed by the server device, as described in further detail below. Accordingly, the processing circuit 330 may include one or more additional units (not shown) for performing the one or more additional steps.

[0061] Figure 4 shows an exemplary block diagram of an electronic device 400 according to some embodiments of the present disclosure. The electronic device 400 can be configured to operate in collaboration with other electronic devices (e.g., the electronic device 200) to implement the technology described in the present disclosure. In some embodiments, the electronic device 400 can be implemented at a network element device in a cellular network, so that the technology described in the present disclosure can be performed in the environment of the cellular network. As a specific example, the electronic device 400 can be implemented at a network data analysis function (NWDAF). In other embodiments, the electronic device 400 can be implemented at other devices of other types of networks without limitation.

[0062] According to some embodiments of the present disclosure, electronic device 400 may include a communication unit 410, a storage unit 420, and a processing circuit 430. The specific implementations of communication unit 410, storage unit 420, and processing circuit 430 may be similar to the communication unit 210, storage unit 220, and processing circuit 230 described above with respect to electronic device 200, and are not further described here. According to some embodiments, in response to processing circuit 430 executing computer program code contained in storage unit 420, electronic device 400 may be configured to perform one or more steps of various methods according to some embodiments of the present disclosure.

[0063] According to some embodiments, the electronic device 400 may be configured to receive one or more local distribution parameter sets, each of which may characterize a data distribution characteristic of a local data set of a corresponding client among one or more clients.

[0064] According to some embodiments, the electronic device 400 may be further configured to send a task identifier to the model library. The task identifier may identify a training task associated with the one or more clients.

[0065] According to some embodiments, the electronic device 400 may be further configured to receive one or more trained candidate models corresponding to the task identifier from a model library.

[0066] According to some embodiments, the electronic device 400 may be further configured to send a model list. The model list may include a matching model selected from the one or more candidate models for each of the one or more clients, the matching model being determined at least in part based on a corresponding local distributed parameter set in the one or more local distributed parameter sets.

[0067] In some embodiments, the processing circuit 430 may include a parameter management unit 431 and a model matching unit 432, each of which may include corresponding hardware circuits or software modules. The parameter management unit 431 may be configured to perform the step of obtaining the one or more local distribution parameter sets. The model matching unit 432 may be configured to send a task identifier, receive a candidate model, and send a model list. In some embodiments, the model matching unit 432 may include a subunit (not shown) for performing each of these steps.

[0068] According to some embodiments of the present disclosure, the electronic device 400 may be further configured to perform one or more additional steps, as described in further detail below with respect to NWDAF. Accordingly, the processing circuit 430 may include one or more additional units (not shown) for performing the one or more additional steps.

[0069] 3. Exemplary Methods

[0070] FIG5 illustrates a flow chart of a method 500 according to some embodiments of the present disclosure. According to some embodiments, the method 500 may be performed at a server for machine learning. For example, the method 500 may be performed by the server 110. The server 110 may be implemented by the aforementioned electronic device 200. Accordingly, the method 500 may be performed by the processing circuit 230 of the electronic device 200.

[0071] Method 500 may start from step 510. In step 510, the server may be configured to obtain a local distribution parameter set of a client. The local distribution parameter set characterizes data distribution characteristics of a local data set of the client.

[0072] In some embodiments, the server may be configured to receive the local distribution parameter set directly from the client. In other embodiments, the server may be configured to receive the local distribution parameter set from another device associated with the client (e.g., an intermediate device between the server and the client). In some embodiments, if a certain degree of loss of data security and user privacy can be tolerated, the server may receive a local dataset (or a sample of a portion of the local dataset) from the client and calculate the local distribution parameter set.

[0073] The obtained local distribution parameter set may include one or more distribution parameters. The number and type of the one or more distribution parameters may vary depending on the local dataset. As an example, the local dataset may follow a Gaussian distribution, a gamma distribution, a lognormal distribution, or any other type of distribution. Accordingly, the local distribution parameter set may include distribution parameters corresponding to these distribution types. Optionally, the local distribution parameter set may additionally include a type indicator indicating the distribution type.

[0074] Method 500 may continue to step 520. In step 520, the server may be configured to send the trained specific model to the client, wherein the specific model is determined at least in part based on the local distribution parameter set.

[0075] In this disclosure, a trained model refers to a model that has completed a training process based on a training dataset. A trained model can be used to generate an output data portion (e.g., a label, prediction, decision, etc.) based on an input data portion. Therefore, a particular trained model can be applied by a client device to at least a portion of its local dataset to generate an output data portion corresponding to the at least a portion.

[0076] According to some embodiments of the present disclosure, the specific model determined for the client may be selected from one or more candidate models. Among the one or more training data sets used to train the one or more candidate models, the specific training data set used to train the specific model may have the highest similarity with the client's local data set. Selecting a specific model based on the similarity between the training data set and the local data set can make the selected specific model adapt to the client's local data set. Accordingly, compared with other models in the one or more candidate models, the output data portion generated by the specific model based on the local data set will be more accurate.

[0077] According to some embodiments of the present disclosure, the similarity between a local dataset and a specific training dataset may be determined based, at least in part, on: (1) a local distribution parameter set for the local dataset, and (2) a specific distribution parameter set for the specific training dataset. An example calculation process for determining the similarity between two datasets based on two distribution parameter sets is described further below.

[0078] In some embodiments, the server may be configured to implement a model provider, while the client may be configured to implement a model consumer. In this case, the specific model determined in step 520 may be applied by the client to the client's local dataset to generate an output data portion. The client may use the output data portion to perform one or more actions.

[0079] For example, the client can be an access control device that uses facial recognition, the server is maintained by a facial recognition service provider, and the specific model is a specific facial recognition model provided by the facial recognition service provider. The access control device can apply the specific facial recognition model to the captured user image and determine whether to allow the user to enter based on the output result of the facial recognition model. When the same access control device is deployed in different scenes (for example, in different regions, times, lighting conditions, or shooting angles), the resulting local data sets (for example, pixel data in the user image set) may have different data distribution characteristics. Through method 500, the facial recognition service provider can provide the access control device with a specific facial recognition model that best matches the local data set in the current scene, because the specific facial recognition model is trained using a training data set with similar data distribution characteristics. In this way, the accuracy of the access control device in the current scene can be improved.

[0080] It should be understood that face recognition is only one use case of the model described in this disclosure. The technology described in this disclosure is not limited to this specific use case.

[0081] In other embodiments, the client may be configured to implement a model training process. For example, the server may be configured to implement a federated learning process, and the client may be configured to implement one of multiple clients of the federated learning process. In this case, the server may be configured to receive a trained local model from the client. The local model is trained by the client based on a padded local dataset. The padded local dataset includes a padded output data portion. The padded output data portion is generated by the specific model based on at least a portion of the client's local dataset.

[0082] Specifically, the client can provide at least a portion of the local dataset (e.g., at least a portion that lacks a corresponding output data portion) as input data to a specific model to generate the supplemented output data portion. The output data portion can be used to complete the client's local dataset. The supplemented local dataset can form a complete training dataset for training the client's local model.

[0083] The server may be further configured to generate a global model based at least in part on the aggregation of local models received from the clients. For example, the server may aggregate the local model with trained local models from other clients participating in federated learning, as described with respect to FIG1 . The trained local models from the other clients may be trained based on the local datasets of the other clients. The local datasets of the other clients may or may not be padded local datasets.

[0084] According to some embodiments of the present disclosure, one or more candidate models trained from which a specific model is selected may be maintained in a model library. The model library may also maintain one or more training data sets corresponding to the one or more candidate models. The model library may reside in another device different from the server (e.g., a data analysis repository function (ADRF) in a cellular network). In this case, the server may also be configured to send the local distribution parameter set obtained in step 510 together with the client identifier of the client. The server may then receive a list of models that may include the specific model selected for the client.

[0085] In some embodiments, the local distribution parameter set, along with the client identifier, may be sent to a remote device (e.g., electronic device 400) that communicates with the model library. The server may then receive a model list from the remote device. In some embodiments, the model list may include one or more matching models determined for one or more clients. The server may be configured to identify a matching model (e.g., the specific model described above) from the model list for sending to each client based on the client identifier of the client. The identified matching model may then be sent to the corresponding client, as described in step 520.

[0086] According to some embodiments of the present disclosure, the server may be additionally configured to send the local distribution parameter set obtained in step 510 together with a task identifier. The task identifier can be used to identify a specific training task of the client in the federated learning process. The specific training task may correspond to a specific service provided by the client to the user. The server may then be configured to receive a specific model, wherein the specific model is determined at least in part based on the task identifier.

[0087] In some embodiments, the local distribution parameter set and the task identifier may be sent by the server to a remote device (e.g., electronic device 400) that communicates with the model library. The server may then receive the specific model associated with the client from the remote device. For example, the remote device may obtain one or more candidate models corresponding to the task identifier from the model library. The remote device may select a specific model that matches the client from the one or more candidate models.

[0088] In an embodiment of federated learning, each task identifier can be used to identify a specific training task for a group of clients in the federated learning process. For example, a first group of clients may be configured to perform a first training task (e.g., training a face recognition model based on a local image set) and be associated with a first task identifier, while a second group of clients may be configured to perform a second training task (e.g., training a channel prediction model based on a local signal dataset) and be associated with a second task identifier. Each client can be associated with a task identifier. The server can identify different tasks for different clients and send the corresponding task identifiers along with the distribution parameter set associated with the client to a remote device, thereby obtaining a matching model corresponding to the training task. For example, the server can obtain a set of trained face recognition models for the first group of clients. Each model in the set of face recognition models is distributed by the server to a corresponding matching client in the first group of clients to complete the client's local image set. The server can obtain a set of trained channel prediction models for the second group of clients. Each model in the set of channel prediction models is distributed by the server to a corresponding matching client in the second group of clients to complete the client's local signal dataset. It should be understood that the above two tasks are merely exemplary and the technology of this disclosure is not limited thereto.

[0089] According to some embodiments, the technology of the present disclosure may be performed in a cellular network. By way of example and not limitation, a server may be deployed with an application function (AF) of a cellular network. A remote device may implement a network data analysis function (NWDAF) of a cellular network. The model library may reside at a data analysis repository function (ADRF) of the cellular network. In this case, if the AF deployed at the server is untrusted, the server may send a set of local distribution parameters (along with an optional task identifier or client identifier) ​​to a network exposure function (NEF) of the cellular network. The NEF may forward this information to the NWDAF. Accordingly, the NWDAF may return a model list containing a specific model to the server via the NEF. If the AF deployed at the server is trusted, the server may send a set of local distribution parameters (along with an optional task identifier or client identifier) ​​directly to the NWDAF. Accordingly, the NWDAF may directly return a model list containing a specific model to the server without passing through the NEF.

[0090] According to some embodiments, the server may send one or more encoder models to the client, which may be used to determine data distribution characteristics or other auxiliary information of the client's local data set, as further described below with respect to FIG. 8A .

[0091] Specifically, in some embodiments, the server may send one or more encoder models to the client. The one or more encoder models may be configured to determine one or more characteristics associated with the client. The server may receive the determined one or more characteristics from the client. The one or more characteristics may serve as auxiliary information from the client. Further, the server may perform one or more tasks based on the one or more characteristics. In some embodiments, the one or more tasks may be associated with an AI model.

[0092] In some embodiments, the one or more encoder models may include a dataset encoder model. The dataset encoder model may be configured to determine data distribution characteristics of the local dataset based on the client's local dataset. The data distribution characteristics may be used as described above. In some embodiments, the one or more encoder models may include a preference encoder model. The preference encoder model may be configured to determine preferences associated with the client. In some embodiments, the one or more encoder models may include an intent encoder model. The intent encoder model may be configured to determine intent associated with the client. Other types of encoder models for determining other characteristics associated with the client are also possible.

[0093] In some embodiments, the server may be configured to retrieve the one or more encoder models from an encoder model repository.For example, the encoder model repository may be implemented in a data analysis repository function ADRF in a cellular network.

[0094] In some embodiments, the server may be configured to send a configuration associated with the client's transmission mode to the client. The configured transmission mode may include a first mode associated with enabling an AI model or a second mode associated with disabling an AI model. Preferably, the configuration associated with the first mode may further specify the client's transmission timing or parameters of one or more AI models used by the client.

[0095] It should be understood that the above description is merely an exemplary embodiment of the method 500. Details of each step of the method 500 as well as additional or optional steps will be further described below.

[0096] FIG6 illustrates a flow chart of a method 600 according to some embodiments of the present disclosure. According to some embodiments, method 600 may be performed at a client for machine learning. For example, method 600 may be performed by client 120. Client 120 may be implemented by the aforementioned electronic device 300. Accordingly, method 600 may be performed by processing circuit 330 of electronic device 300.

[0097] Method 600 may start from step 610. In step 610, the client may be configured to send a local distribution parameter set to the server, where the local distribution parameter set characterizes data distribution characteristics of a local data set of the client.

[0098] In the present disclosure, a local dataset of a client refers to a collection of data held by or accessible to the client individually. The local dataset may include various types of data. For example, the local dataset may include data captured by the client's sensors, user data associated with the client's users, communication data generated by the client's communications, and so on. The client may be configured to generate a corresponding local distribution parameter set based on the local dataset (or a sample of a portion thereof). The client may have multiple local datasets, and a local distribution parameter set may be generated for each local dataset.

[0099] Method 600 may continue to step 620. In step 620, the client may be configured to receive a trained specific model from the server, wherein the specific model is determined at least in part based on a local distribution parameter set of the client.

[0100] As previously described, the received specific model may be selected from one or more candidate models. The specific model may be matched with the client's local dataset. Specifically, among the one or more training datasets used to train the one or more candidate models, the specific training dataset used to train the specific model may have the highest similarity with the client's local dataset. The similarity may be determined at least in part based on a specific distribution parameter set of the specific training dataset and a local distribution parameter set of the client's local dataset.

[0101] According to some embodiments of the present disclosure, a client may use a received specific model to generate an output data portion based on at least a portion of a local dataset. For example, the client may provide the at least a portion of the local dataset as input data to the received specific model to generate an output data portion corresponding to the at least a portion. Depending on the functionality to be implemented by the specific model, the output data portion may include at least one of a label, a prediction, and / or a decision generated based on the at least a portion of the local dataset.

[0102] According to some embodiments of the present disclosure, a client may be configured to implement a model consumer. The model consumer may be configured to use the output data portion generated by a specific model to perform one or more actions, such as described above with respect to the example of the access control device.

[0103] According to other embodiments of the present disclosure, the client may be configured to implement a model training party. For example, the server may be configured as a server that implements a federated learning process, and the client may be configured as one of multiple clients that implement the federated learning process. In this case, the client may use the output data portion generated by the specific model to complete the local dataset. Compared to the original local dataset (e.g., the local dataset used to generate the local distribution parameter set in step 610), the completed local dataset additionally includes the output data portion generated by the specific model received in step 620. The client may be further configured to use the completed local dataset as a training dataset to train the client's local model, and send the trained local model to the server for aggregation to generate a global model for the federated learning process.

[0104] The inventors of the present disclosure have recognized that using the padded local dataset as the training dataset can ensure that the local model is fully trained. Therefore, the performance of the global model that aggregates the local models is also improved.

[0105] The inventors of the present disclosure also recognized that, for multiple clients participating in a federated learning process, each client's padded local dataset is adaptively padded based on the data distribution characteristics of the client's local dataset. This improves the accuracy of the padded local dataset (specifically, the output data portion), thereby improving the accuracy of the trained local models. Consequently, the performance of the global model that aggregates the local models is also improved.

[0106] The data completion process described in the present disclosure is automatic, fast, accurate, and / or cost-effective. It does not require the client to possess prior experience or knowledge. Furthermore, the process can be adaptively updated based on the changing scenarios faced by the client (e.g., adaptively updating the matching model sent to the client).

[0107] As previously described, the output data portion generated by the specific model may include a label, prediction, or decision generated based on at least a portion of the local data set. For example, for an input data portion that lacks a label, the client may input the input data portion into the received specific model to generate a corresponding label. The client may associate the input data portion with the generated label to form labeled data. For an input data portion that lacks a prediction result, the client may input the input data portion into the received specific model to generate a corresponding prediction result. The client may associate the input data portion with the generated prediction result. For an input data portion that lacks a decision result, the client may input the input data portion into the received specific model to generate a corresponding decision result. The client may associate the input data portion with the generated decision result.

[0108] According to some embodiments of the present disclosure, while completing the local dataset, the client may not participate in one or more rounds of the federated learning process. For example, while generating the completed local dataset, the client may not train the local model, or the client may not send the trained local model to the server, or the client may instruct the server not to aggregate the local models received from the client. After generating the completed local dataset, the client may send a completion indication to the server. The client may re-participate in subsequent rounds of the federated learning process. In this way, it is possible to avoid aggregating the local model trained by the client based on the incomplete training dataset into the global model.

[0109] According to some alternative embodiments of the present disclosure, while completing a local dataset, a client may participate in the federated learning process but may not use at least the uncompleted portion of the local dataset to train the local model. For example, the training dataset used by the client to train the local model may include a first input data portion that has a corresponding output data portion in the local dataset, but may not include a second input data portion that does not have a corresponding output data portion. The client may then use the second input data portion to train the local model only after completing the output data portion corresponding to the second input data portion.

[0110] According to some embodiments, the client may receive one or more encoder models that may be used to determine data distribution characteristics or other auxiliary information of the client's local dataset, as further described below with respect to FIG. 8A .

[0111] Specifically, in some embodiments, a client may be configured to receive one or more encoder models from a base station or server. The client may execute the one or more encoder models. These models may be configured to determine one or more characteristics associated with the client. Different models may be used to determine different characteristics associated with the client. The client may transmit the determined one or more characteristics to the server or base station. For example, data representing the one or more characteristics may be encoded for transmission.

[0112] In some embodiments, the one or more encoder models may include a dataset encoder model that is configured to determine data distribution characteristics of the local dataset based on the client's local dataset. The data distribution characteristics can be used as described above to select a suitable trained model for the client. In some embodiments, the one or more encoder models may include a preference encoder model that is configured to determine preferences associated with the client. In some embodiments, the one or more encoder models may include an intent encoder model that is configured to determine intents associated with the client. Other types of encoder models are also possible.

[0113] In some embodiments, the one or more encoder models may be maintained in an encoder model repository. The encoder model repository may be located at a data analysis repository function (ADRF) in the cellular network. A server or base station may retrieve a suitable encoder model from the ADRF and send it to the corresponding client.

[0114] In some embodiments, a client may be configured to receive a configuration associated with a transmission mode of the client from a base station or a server. The transmission mode may include a first mode associated with enabling an AI model or a second mode associated with not enabling the AI ​​model. Optionally, the configuration associated with the first mode may further specify the client's transmission timing or parameters of one or more AI models used by the client.

[0115] It should be understood that the above description is merely an exemplary embodiment of the method 600. Details of each step of the method 600 as well as additional or optional steps will be further described below.

[0116] 7 shows a flow chart of a method 700 according to some embodiments of the present disclosure. According to some embodiments, the method 700 may be implemented by the aforementioned electronic device 400. Accordingly, the method 700 may be executed by the processing circuit 430 of the electronic device 400.

[0117] Method 700 may begin at step 720. In step 710, electronic device 400 may be configured to receive one or more local distribution parameter sets, each of which characterizes data distribution characteristics of a local data set of a corresponding client among one or more clients.

[0118] Method 700 may continue to step 720. In step 720, electronic device 400 may be configured to send a task identifier to a model library. The task identifier may identify a specific training task associated with the one or more clients. The model library maintains one or more trained models. The one or more models may be trained for different machine learning tasks.

[0119] For example, the one or more clients may include a first group of clients configured to perform a first training task (e.g., training a face recognition model based on a local image set) and be associated with a first task identifier. Accordingly, the electronic device 400 may be configured to send the first task identifier associated with the first training task to the model library.

[0120] The method 700 may continue to step 730. In step 730, the electronic device 400 may be configured to receive one or more trained candidate models corresponding to the task identifier from the model library.

[0121] The trained one or more candidate models may be retrieved from the model library based on the task identifier and returned to the electronic device 400. For example, for the aforementioned first group of clients, the trained one or more candidate models may include a group of trained face recognition models.

[0122] Method 700 may continue to step 740. In step 740, electronic device 400 may be configured to send a model list. The model list may include a matching model selected from the one or more candidate models for each of the one or more clients, wherein the matching model is determined at least in part based on a corresponding local distribution parameter set in the one or more local distribution parameter sets.

[0123] For example, for a first client among the one or more clients, the model list may include a first matching model selected based on a first set of local distribution parameters for a first local dataset of the first client. For a second client among the one or more clients, the model list may include a second matching model selected based on a second set of local distribution parameters for a second local dataset of the second client. If no corresponding matching model exists for a particular client among the one or more clients, the model list may include a default model associated with the task identifier.

[0124] According to some embodiments of the present disclosure, each of the one or more candidate models received by the electronic device 400 is trained using a corresponding training dataset in one or more training datasets. To determine a matching model for each client, the electronic device 400 may determine, from the one or more training datasets, a specific training dataset that has the highest similarity to the local dataset of the client. The electronic device 400 may then determine the candidate model corresponding to the specific training dataset from the one or more candidate models as the matching model for the client.

[0125] According to some embodiments of the present disclosure, for example, the similarity can be determined at least in part based on the local distribution parameter set of the local data set and the specific distribution parameter set of the specific training data set. For example, for a first client among the one or more clients, the electronic device 400 can select a first training data set based on the first local distribution parameter set of the first local data set of the first client and the distribution parameter set of the training data set of each candidate model, and the first training data set has the highest similarity with the first local data set. Then, the electronic device 400 can determine the first candidate model trained based on the first training data set as the first matching model for the first client. For a second client among the one or more clients, the electronic device 400 can select a second training data set based on the second local distribution parameter set of the second local data set of the second client and the distribution parameter set of the training data set of each candidate model, and the second training data set has the highest similarity with the second local data set. Then, the electronic device 400 can determine the second candidate model trained based on the second training data set as the second matching model for the second client.

[0126] According to some embodiments of the present disclosure, in step 730, the electronic device 400 may additionally receive one or more training data sets associated with the trained one or more candidate models from the model library. Each of the one or more candidate models is trained based on a corresponding training data set in the one or more training data sets. The electronic device 400 may calculate a distribution parameter set for each training data set based on the received one or more training data sets. The calculated distribution parameter set may be used to determine the similarity between the training data set and the local data set of the client using a local distribution parameter set instrument. In some embodiments, the electronic device 400 may receive a sample (instead of all) of each training data set in the one or more training data sets from the model library, thereby saving communication and computing overhead.

[0127] According to other embodiments of the present disclosure, the distribution parameter sets of corresponding training data sets in the one or more training data sets are calculated and maintained by the model library. The electronic device 400 can receive the one or more distribution parameter sets of the one or more training data sets from the model library. In this case, the electronic device 400 does not need to receive the training data sets and does not need to calculate the distribution parameter sets of the training data sets.

[0128] The distribution parameter sets described herein (e.g., the local distribution parameter set for a local data set of a client, or the distribution parameter set for a training data set) may include one or more distribution parameters. The number and type of the one or more distribution parameters may vary from data set to data set. As an example, a data set may obey a Gaussian distribution, a gamma distribution, a lognormal distribution, or any other type of distribution. Accordingly, the distribution parameter set may include distribution parameters corresponding to these distribution types. In some embodiments, the data set obeys a Gaussian distribution, and the distribution parameter set may include a mean μ and a standard deviation δ. In some embodiments, the data set obeys a gamma distribution, and the distribution parameter set may include a shape parameter α and an inverse scale parameter β. In some embodiments, the data set obeys a lognormal distribution, and the distribution parameter set may include a mean μ and a variance δ. In other embodiments, the data set may obey other types of distributions, and the distribution parameter set may include corresponding distribution parameters.

[0129] According to some embodiments of the present disclosure, the local distribution parameter set of the client may further include a type indicator indicating the distribution type. The electronic device 400 may calculate the distribution parameter set of the training data set associated with the candidate model based on the distribution type indicated by the type indicator.

[0130] According to some embodiments, in step 710, the electronic device 400 may receive the one or more local distribution parameter sets from a server for machine learning. The server may be associated with the one or more clients. For example, the server may be a server for federated learning, and the one or more clients may be training devices participating in the federated learning.

[0131] According to some embodiments, the task identifier sent by the electronic device 400 in step 720 may be received from the server for machine learning. For example, the task identifier may be received from the server together with the local distribution parameter set.

[0132] According to some embodiments, the electronic device 400 may send the matching model of each of the one or more clients together with the client identifier of the client to a recipient (e.g., a server) in step 740. For example, the model list sent by the electronic device 400 may additionally include the identifier of each of the one or more clients. Each identifier may be stored in the model list in association with the matching model for the client of the identifier. This allows the recipient of the model list to identify the corresponding matching model for each client.

[0133] The technology of the present disclosure can be performed in a cellular network. For example, the electronic device 400 can be implemented at the network data analysis function (NWDAF) of the cellular network. The model library can reside at the data analysis repository function (ADRF) of the cellular network. The server for machine learning can be deployed with the application function (AF) of the cellular network. In some embodiments, the electronic device 400 at the NWDAF can communicate directly with the server. In other embodiments, the electronic device 400 at the NWDAF can communicate with the server via the network open function (NEF) of the cellular network. The specific receiving path can be determined based on the AF deployed at the server. For example, if the AF is trusted, the electronic device 400 can receive the distribution parameter set and task identifier directly from the server. Otherwise, the electronic device 400 can receive the distribution parameter set and task identifier from the server via the NEF.

[0134] It should be understood that the above description is merely an exemplary embodiment of the method 700. Details of each step of the method 700 as well as additional or optional steps will be further described below.

[0135] 4. Example Process

[0136] FIG8 is a schematic diagram of a process 800 according to some embodiments of the present disclosure. Process 800 is performed in a cellular network environment. The cellular network may include ARDF, NWDAF, NEF, etc. It should be understood that a cellular network is merely an example of a communication environment in which the techniques of the present disclosure may be performed. In other embodiments, the techniques of the present disclosure may be performed in other communication environments without limitation.

[0137] Process 800 may involve a machine learning server and client 1-N. The server and client may be implemented by the aforementioned electronic device 200 and electronic device 300, respectively. The NWDAF may be implemented by the aforementioned electronic device 400. In the example of the cellular network, the model library may reside at the ADRF, and the server may be deployed with an AF. The client may reside at one or more user equipment (UE). Each UE may access the cellular network through a base station (e.g., a gNB) and communicate with other network element devices. In other examples, one or more of the model library, server, and client may reside in other suitable locations.

[0138] Process 800 may begin at step 801. In step 801, the server may send a trigger notification to the client. The trigger notification may be used to initiate execution of process 800. In some embodiments, the server may periodically send the trigger notification. In other embodiments, the server may send the trigger notification based on a trigger request from a client. For example, if more than a threshold proportion (e.g., 50%) of clients 1-N send a trigger request, the server may send the trigger notification. Alternatively, the server may send a trigger notification to a single client based on the trigger request of the client.

[0139] In some embodiments, when the client is a training device for federated learning, the client can send a trigger request based on checking a local dataset. For example, if the local dataset is missing a portion of the output data, the client can send a trigger request to trigger data completion. Specifically, the trigger request can be sent when the completeness of the local dataset falls below a threshold. In other embodiments, when the client is a model consumer, the client can send a trigger request based on the lack of a locally available model or the performance of a locally available model falling below a threshold.

[0140] In step 802 , in response to receiving a trigger request, the client may generate a local distribution parameter set for the local data set.

[0141] The data sets described in the present disclosure (e.g., local data sets or training data sets) may include various types of data. In some embodiments, the data sets may include media data, which may include but are not limited to text data, image data, audio data, video data, tactile data, and the like. In some embodiments, the data sets may include sensor data, which may include data collected by one or more sensors, which may describe one or more physical or chemical parameters, including but not limited to temperature, humidity, speed, acceleration, altitude, electrical parameters, magnetic parameters, purity, pH, concentration, and the like. In some embodiments, the data sets may include communication data, which may include but are not limited to measurement data or indicator data related to the communication environment, communication quality, and communication status. In some embodiments, the data sets may include user data, which may include but are not limited to user behavior data, user preference data, user analysis data, and the like. In some embodiments, the data sets may include a combination of these data. In some embodiments, the data sets may include data after preprocessing these data.

[0142] As previously described, the local data set may obey various types of distributions. For example, the data set may obey a Gaussian distribution, a gamma distribution, a lognormal distribution, or any other type of distribution. Accordingly, the local distribution parameter set may include distribution parameters corresponding to these distribution types. In some embodiments, the local distribution parameter set may include a mean μ and a standard deviation δ for a Gaussian distribution. In some embodiments, the local distribution parameter set may include a shape parameter α and an inverse scale parameter β for a gamma distribution. In some embodiments, the local distribution parameter set may include a mean μ and a variance δ for a lognormal distribution. In other embodiments, the local data set may obey other types of distributions, and the local distribution parameter set may include corresponding distribution parameters. The client may generate a local distribution parameter set that characterizes the local data set based on the entire local data set or a sample of a portion thereof.

[0143] In some embodiments, a client may use a dataset encoder model to determine a set of local distribution parameters. The dataset encoder model may be applied to the client's local dataset or a sample portion thereof. The dataset encoder model may analyze and process the input local dataset to extract data distribution characteristics of the local dataset. The data distribution characteristics may be represented as a set of local distribution parameters.

[0144] In some embodiments, a client may receive a dataset encoder model from a base station (e.g., a gNB) or a server. As further described below with respect to FIG. 8A , the client may receive one or more encoder models from the base station or server. The one or more encoder models may be configured to determine one or more characteristics of the client. The one or more encoder models may include a dataset encoder model. The determined one or more characteristics may include data distribution characteristics of a local dataset of the client. The client may provide the local dataset as input to the dataset encoder model. The dataset encoder model may analyze the local dataset and extract a set of local distribution parameters that characterize its data distribution characteristics.

[0145] In some embodiments, the client may not rely on the dataset encoder model received from the base station or server. For example, the client may use a local dataset encoder model to generate a local distribution parameter set. For another example, the client may use an algorithm other than the dataset encoder model to process the local dataset to generate a local distribution parameter set.

[0146] In step 803, the client may send a distribution parameter set response to the server. Each distribution parameter set response includes a corresponding local distribution parameter set of the corresponding client. In some embodiments, the corresponding local distribution parameter set also includes a type indicator corresponding to the distribution type.

[0147] In some alternative embodiments, in response to the trigger notification, instead of executing steps 802 and 803 , the client may send the local data set (or a portion thereof) to the server, and the server may generate a corresponding local distribution parameter set.

[0148] In step 804, the server may send the distribution parameter list together with the task identifier to the NWDAF. For example, the server may integrate the obtained local distribution parameter sets to generate a distribution parameter list. The distribution parameter list includes one or more local distribution parameter sets associated with one or more clients. The server may also associate each local distribution parameter set with the corresponding client, thereby recording the client from which each local distribution parameter set originates. The distribution parameter list may also include a client identifier associated with one or more clients, thereby associating each client identifier with each local distribution parameter set. In an embodiment of federated learning, the server may further associate each distribution parameter set with a specific training task (e.g., a specific task identifier) ​​in which the corresponding client participates, because a single server may serve multiple groups of clients performing multiple learning tasks.

[0149] In an embodiment of a cellular network, the server may be deployed with an AF. Depending on the attributes of the AF, step 804 may have different sub-steps. For example, if the AF is untrusted, sub-steps 804a and 804b may be executed. In sub-step 804a, the server sends the distribution parameter list along with the task identifier to the NEF. Then, in sub-step 804b, the NEF sends the distribution parameter list along with the task identifier to the NWDAF. If the AF is trusted, the distribution parameter list along with the task identifier does not need to be forwarded via the NEF. Therefore, instead of executing sub-steps 804a and 804b, sub-step 804c is executed. In step 804c, the AF sends the distribution parameter list along with the task identifier directly to the NWDAF.

[0150] In step 805, the NWDAF may send the task identifier received in step 804 to the model library. The model library may reside at the ADRF. The model library maintains one or more trained models. The one or more models may be trained for different machine learning tasks. For example, the one or more models may include dozens, hundreds, or more models. In some embodiments, the model library may also maintain one or more training data sets for training the one or more models. In some embodiments, the model library may also maintain one or more distribution parameter sets for the one or more training data sets.

[0151] In step 806, the model library may retrieve one or more candidate models corresponding to the task identifier based on the task identifier received in step 805. The one or more candidate models may have been trained for a task that is the same as or similar to the training task indicated by the task identifier. For example, if the task identifier indicates a task of training a face recognition model, the one or more candidate models may include a set of trained face recognition models maintained in the model library. In some embodiments, the model library may also retrieve one or more training data sets associated with the retrieved one or more candidate models, each of the one or more candidate models being trained based on a corresponding training data set in the one or more training data sets. In some embodiments, the model library may also retrieve one or more distribution parameter sets associated with the one or more candidate models (more specifically, the one or more training data sets), each of the one or more distribution parameter sets indicating data distribution characteristics of the corresponding training data set.

[0152] In step 807, the model library may send a candidate model response to the NWDAF. The candidate model response may include the one or more candidate models determined in step 806. In some embodiments, the candidate model response may include one or more training data sets (or a portion of samples) associated with the one or more candidate models. In other embodiments, the candidate model response may include one or more distribution parameter sets associated with the one or more candidate models (more specifically, the one or more training data sets), rather than the training data sets themselves.

[0153] In step 808, NWDAF may perform model matching. The model matching may be performed based on the distribution parameter list associated with the client received in step 804 and the one or more training data sets (or one or more distribution parameter sets of the one or more training data sets) received in step 807. Specifically, for the local distribution parameter set for a specific client in the distribution parameter list, NWDAF may calculate the similarity between the local data set of the specific client and the training data set based on the local distribution parameter set and the distribution parameter set of each training data set in the one or more training data sets received. Then, NWDAF may select a specific training data set in the one or more training data sets that has the highest similarity with the local data set of the specific client. NWDAF may determine a specific candidate model corresponding to the specific training data set in the one or more candidate models as the matching model for the specific client. The specific candidate model is trained based on the specific training data set. NWDAF may perform model matching on each client associated with the distribution parameter list, thereby obtaining a matching model for each client.

[0154] In some embodiments, NWDAF receives a training data set from a model library. NWDAF can calculate a distribution parameter set of the data set based on the received training data set. The calculation method can be the same as the method in which the client calculates the local distribution parameter set of the local data set. In some embodiments, NWDAF can determine the distribution type based on a type indicator associated with the distribution parameter set of the local data set, such as Gaussian distribution, gamma distribution, lognormal distribution, etc. NWDAF can then assume that the training data set has the distribution type and calculate one or more distribution parameters corresponding to the distribution type accordingly. In other embodiments, each data set has a consistent distribution type. Therefore, it is not necessary to rely on the type indicator.

[0155] In some other embodiments, NWDAF receives the distribution parameter set of the training dataset from the model library. Therefore, NWDAF does not need to calculate the distribution parameter set of the training dataset.

[0156] In step 809, the obtained one or more matching models may be organized into a model list response to be sent to the server. In some embodiments, each model list response may include one or more matching models associated with a single task identifier. The NWDAF may include the obtained one or more matching models in the model list response in association with the corresponding one or more client identifiers. In other embodiments, each model list response may include one or more matching models associated with multiple task identifiers. The NWDAF may include the obtained one or more matching models in the model list response in association with one or more client identifiers and one or more task identifiers. The one or more client identifiers or one or more task identifiers may be received by the NWDAF in step 804.

[0157] Depending on the properties of the AF deployed at the server, step 809 may have different sub-steps. For example, if the AF is untrusted, sub-steps 809a and 809b may be executed. In sub-step 809a, the NWDAF sends the Model List Response to the NEF. Then, in sub-step 809b, the NEF sends the Model List Response to the server. If the AF is trusted, the Model List Response does not need to be forwarded via the NEF. Therefore, instead of executing sub-steps 809a and 809b, sub-step 809c is executed. In step 809c, the NWDAF sends the Model List Response directly to the server.

[0158] In step 810, the server may send each matching model in the received model list response to the corresponding client. For example, the server may send the matching model in the model list response to the client corresponding to the client identifier associated with the matching model. The server may determine the matching model for each client based on the client identifier (and optionally the task identifier). In some embodiments, sending the matching model may include sending a set of model parameters for the model. In other embodiments, sending the matching model may include sending a storage location associated with the set of model parameters for the model, so that the client can retrieve the set of model parameters from the storage location.

[0159] After receiving the matching model, the client may apply the matching model to the local data set in step 811 to generate an output data portion.

[0160] When the client is a model consumer, the output data portion generated by matching the model can be used to take one or more actions, as described above with respect to the access control device.

[0161] When the client is a model training party participating in federated learning, the output data portion can be used to complete the client's local data set. The completed local data set can be used as a training data set for training the local model. Specifically, in step 812, each client can apply the received matching model to at least a portion of the local data set (for example, the input data portion that lacks the corresponding output data portion) to generate an output data portion. The output data portion can be integrated with the at least a portion of the local data set to complete the at least a portion. Optionally, the client can send a completion indication to the server after completing the completion. Optionally, the server can send an indication to the client allowing the client to participate in training after receiving the completion indication.

[0162] In step 813, the client can use the completed local data set as a complete training data set to train the local model. The local model is associated with the task identifier described above, for example. That is, the training task identified by the client's task identifier includes training the local model. In the present disclosure, the trained local model can be a variety of appropriate types of models, including machine learning models that have been developed and are to be developed. Each artificial intelligence model can be represented by a set of model parameters that characterize the model. In addition, the method for training the local model can be a variety of appropriate methods. The technology of the present disclosure does not limit the type of model and the training method. Example model types include, but are not limited to, models trained by convolutional neural networks (CNNs), deep neural networks (DNNs), recurrent neural networks (RNNs), long short-term memory (LSTMs), generative models, random forests, and their deformation and enhancement algorithms.

[0163] After training the local model, in step 814 , the client may send the trained local model to the server in step 814 .

[0164] The server may then aggregate the one or more trained local models received from the one or more clients to generate a global model in step 815. The generated global model may be sent to the one or more clients for use or to execute the next training round.

[0165] In process 800, the data completion process of the client may span one or more training rounds of federated learning. In some embodiments, during the data completion process, a specific client participating in the process may skip the one or more training rounds of federated learning. For example, the specific client may not train a local model or upload a local model. Additionally or alternatively, the server may not aggregate the local model from the specific client before receiving a completion indication. In other embodiments, the specific client may participate in federated learning based only on a first part of the local data set (e.g., the first part is complete). The incomplete second part of the local data set may be used for federated learning after the data completion process.

[0166] Table 1 shows example input information of the NWDAF according to some embodiments of the present disclosure. In these embodiments, the client is implemented as a user equipment (UE) associated with a user.

[0167] Table 1

[0168] Table 2 shows example output information of NWDAF according to some embodiments of the present disclosure.

[0169] Table 2

[0170] It should be understood that what is described above is merely an exemplary embodiment of process 800. The technology of the present disclosure may include more or fewer steps and may be implemented in a communication environment different from a cellular environment. For example, in some embodiments, the server, NWDAF, and model library may reside in the same device, so that the technology of the present disclosure can be implemented in a simple server-client architecture. For another example, in some embodiments, there are intermediate elements not shown between the various elements of process 800 for relaying and forwarding. For another example, in some embodiments, the server may be implemented at a network element other than the AF, or the model library may be implemented at a network element other than the ADRF. Alternatively, the operations performed by the NWDAF may be performed by other network elements.

[0171] Figure 8A shows a schematic diagram of process 800A according to some embodiments of the present disclosure. Process 800A can be performed in a cellular network environment. Process 800A can involve clients 1-N and a base station (e.g., gNB) or an associated AF. In some embodiments, the AF can implement a server associated with an AI model. The client can reside at one or more user equipment (UEs). Each UE can access the cellular network through a base station (e.g., gNB). The client can be associated with the AI ​​model, for example, to participate in training or use the AI ​​model.

[0172] At step 821, the base station may be configured to send an encoder model request to the encoder model repository. In some embodiments, the encoder model request may include a specified encoder model type. In other embodiments, the encoder model request may include information associated with the desired encoder model type. This information may be analyzed to determine the appropriate encoder model type for each client. For example, in a federated learning scenario, this information may include a task identifier for the federated learning process in which the client participates. The task identifier may be used to determine the appropriate encoder model type (e.g., an appropriate dataset encoder model). In other scenarios, this information may include other information associated with the client.

[0173] The encoder model repository can be configured to store or maintain one or more candidate encoder models. Optional encoder types include, but are not limited to, dataset encoder models, preference encoder models, intent encoder models, or any other suitable type of encoder.

[0174] In some embodiments, the encoder model repository may be implemented at a data analysis repository function ADRF in a cellular network. In other embodiments, the encoder model repository may be implemented at any other location accessible to a base station or server.

[0175] At step 822, the encoder model repository may issue one or more encoder models in response to the encoder model request. The encoder model repository may determine one or more suitable encoder models. In some embodiments, the one or more encoder models may be determined based on an analysis of the information in the encoder model request. For example, in a federated learning scenario, an appropriate dataset encoder model may be determined based at least in part on a task identifier in the information. The dataset encoder model may be used to obtain data distribution characteristics of the client's local dataset. In some scenarios, information associated with the desired encoder model type may be associated with the client's preferences. Accordingly, an appropriate preference encoder model may be determined based at least in part on the information. In some scenarios, information associated with the desired encoder model type may be associated with the client's intent. Accordingly, an appropriate intent encoder model may be determined based at least in part on the information.

[0176] At step 823, the base station may send a configuration associated with the client's transmission mode to the client. This configuration may be associated with whether the client has an AI model enabled. For example, the transmission mode specified by the configuration may be selected from one of the following modes: (i) a first mode associated with enabling the AI ​​model, or (ii) a second mode associated with not enabling the AI ​​model. When configuring the first mode for the client, the configuration associated with the first mode may further specify a first transmission timing for the client. In some embodiments, the first transmission timing may specify a schedule for one or more communication transmissions of the client in the first mode. Additionally, the configuration associated with the first mode may also specify parameters of one or more AI models to be used by the client. The one or more AI models may include the AI ​​model associated with federated learning described above or any other AI model to be used by the client. When configuring the second mode for the client, the configuration associated with the second mode may specify a second transmission timing for the client. In some embodiments, the second transmission timing may specify a schedule for one or more communication transmissions of the client in the second mode. The one or more communication transmissions may be the same as or different from the one or more communication transmissions described above for the first mode. In some embodiments, the second transmission timing associated with the second mode may be different from the first transmission timing associated with the first mode. That is, the client's transmission timing may vary based on whether the AI ​​model is enabled.

[0177] In some embodiments, in the first mode, the configured first transmission timing may also vary depending on the different AI models enabled by the client. For example, the first transmission timing configured when the client enables a first AI model may be different from the first transmission timing configured when the client enables a second AI model.

[0178] In step 824 , the base station may send the one or more encoder models received in step 822 to corresponding clients.

[0179] In step 825, the client may upload one or more encoding results to the base station. Specifically, the client may locally run the received one or more encoder models. The one or more encoder models may analyze the client's data and behavior and generate corresponding encoding results. For example, a dataset encoder model may analyze the client's local dataset to generate data distribution characteristics. The data distribution characteristics may be encoded as part of the encoding result. For another example, a preference encoder model may collect and analyze data associated with user behavior and extract preference data representing the user's preferences. The preference data may be encoded as part of the encoding result. For another example, an intent encoder model may collect and analyze data associated with user behavior and extract intent data representing the user's intent. The intent data may be encoded as part of the encoding result. In other embodiments, other types of encoder models may generate other types of encoded data. The encoded data may be considered an abstraction of the client's local data or user behavior, providing insights into the client's local data or user behavior without leaking user privacy or compromising the client's security.

[0180] In step 826, the base station may perform one or more tasks. Specifically, the base station may perform one or more tasks based on the encoded data received in step 825. In some embodiments, the one or more tasks may be associated with an AI model. For example, the encoded data may be used as auxiliary information from the client for training, updating, selecting or using the AI ​​model. In the aforementioned federated learning scenario, the encoded data may include a local distribution parameter set of the client. The local distribution parameter set may be used to select a specific trained model suitable for the client, as described above. In other scenarios, the encoded data may include data associated with the user's preferences or intentions, which may be used to customize the model associated with the client. For example, an intelligent driving model that matches the user's driving preferences may be trained and deployed based on the user's driving preferences. Performing the one or more tasks based on the auxiliary information from the client can improve the performance and experience associated with the AI ​​model.

[0181] Optionally, in step 827, the client may determine localized updates to the received one or more encoder models. The encoder models retrieved from the encoder model repository are not necessarily adapted for every client. Therefore, the client may evaluate the model performance while running the encoder model. The client may determine updates to the encoder model that are tailored to the client's local environment.

[0182] In step 828, the client can upload the determined update or updated encoder model to the base station. In some embodiments, the received update or updated model can be evaluated. The evaluation can be performed by one or both of the base station and the encoder model repository (e.g., ADRF). For example, the performance improvement degree, scope of application, model size, etc. of the updated encoder model can be evaluated. Based on the result of the evaluation, it can be determined whether to accept the model update proposed by the client. If it is determined to accept the model update, the updated encoder model can be stored in the encoder model repository for subsequent use. If it is determined not to accept the model update, the updated encoder model can be abandoned. In some embodiments, the base station can feed back an indication of whether to accept or not accept the model update to the corresponding client.

[0183] It should be understood that the above description is merely an exemplary embodiment of process 800A. The techniques of the present disclosure may include more or fewer steps, and these steps may be performed in a different order. For example, the configuration in step 823 may be performed at any appropriate time. Furthermore, in alternative embodiments, process 800A may be implemented in a communication environment different from a cellular environment.

[0184] Process 800A provides one or more encoder models to the client to obtain auxiliary information associated with the client. This auxiliary information can facilitate one or more tasks (including but not limited to tasks associated with the AI ​​model). In some embodiments, process 800A can be performed in conjunction with the method or process described in the present disclosure (e.g., process 800). However, it should be understood that process 800A is not limited to the aforementioned application scenarios associated with the data distribution characteristics of the local data set. As described above, process 800A can be performed for the purpose of obtaining other auxiliary information (e.g., user preferences, intentions, etc.). Therefore, in an alternative embodiment, process 800A can be performed independently of the method or process described in the present disclosure (e.g., process 800) without limitation.

[0185] 5. Exemplary Calculation Method

[0186] This section describes one or more exemplary computational methods that can be used to compute a set of distribution parameters and to compute similarity between two data sets based on the set of distribution parameters. It should be understood that the data sets and computational methods used in this section are merely exemplary. In other embodiments, alternative data sets and computational methods can be used without departing from the scope of this disclosure.

[0187] According to an embodiment of the present disclosure, the local data set of the client is represented as a data set LD. The data set LD may include N samples. The data set LD may be represented as follows:

[0188] Where N represents the number of samples, M represents the number of data points contained in each sample, and Q represents the number of feature dimensions for each data point. In expression (1), N samples are listed from left to right. Within each sample, M data points are listed from top to bottom. Within each data point, Q values ​​for that data point on Q feature dimensions are listed from left to right.

[0189] A local dataset can contain various types of data and be organized accordingly. For example, if dataset LD is a radar measurement dataset, N represents the number of measurement samples, M represents the number of measurement points contained in each measurement sample, and Q represents the number of characteristic dimensions of each measurement point. If each measurement point represents a coordinate in three-dimensional space, Q can be equal to 3. For another example, if dataset LD is an image dataset, N represents the number of extracted image samples, M represents the number of channels per image sample, and Q represents the number of dimensions of the vector obtained by flattening the data for each channel. Flattening indicates that the end of each row of the data matrix is ​​concatenated with the beginning of the next row to form a vector.

[0190] It should be understood that the above data organization method for the local dataset is merely an example. Any other suitable data organization method may be adopted. For example, although the local dataset LD is represented as a high-order tensor, it can be represented as a vector or scalar without departing from the scope of this disclosure.

[0191] In some embodiments, to reduce processing load, the original local dataset may be preprocessed to obtain the local dataset LD. For example, the original local dataset may be downsampled to reduce the value of N. Additionally or alternatively, the original local dataset may be subjected to dimensionality reduction to reduce the value of M or Q. Other processing is also possible, as are combinations of these processing methods.

[0192] The data distribution characteristics can be calculated for each feature dimension of each sample of the local dataset LD. The data distribution characteristics can be represented by a distribution parameter set including one or more distribution parameters. In the example of this section, it is assumed that each feature dimension of the sample of the local dataset LD obeys a Gaussian distribution. Accordingly, the mean (μ) and standard deviation (δ) can be used to characterize the distribution characteristics of the data of the local dataset LD. The mean and standard deviation of data obeying the Gaussian distribution can be calculated by known methods. For the aforementioned dataset LD, the calculated distribution parameter set LX can be expressed as follows:

[0193] Among them, each pair of mean and standard deviation (μ, δ) represents the data distribution characteristics of a sample in a feature dimension in the dataset LD.

[0194] FIG9 shows the correspondence between the local data set and the local distribution parameter set according to some embodiments of the present disclosure. As shown in FIG9, a pair of mean and standard deviation This example illustrates the distribution characteristics of the data of the second sample in the local dataset LD on the second feature dimension (indicated by the bold box). The means and standard deviations of other pairs can similarly correspond to the distribution characteristics of the data of the corresponding samples in the dataset LD on the corresponding feature dimensions.

[0195] It should be understood that the Gaussian distribution used in the present disclosure is merely exemplary. In practical applications, it may be assumed that the dataset LD obeys other types of distributions, and the distribution parameter set may be calculated based on this assumption. Other types of distributions may include, but are not limited to, gamma distribution, lognormal distribution, and the like. When the dataset LD obeys a gamma distribution, the shape parameter α and the inverse scale parameter β may be used instead. When the dataset LD obeys a lognormal distribution, the mean μ and the variance δ may be used instead. Furthermore, these distributions may be one-dimensional or multi-dimensional.

[0196] For the training dataset TD in the model library, the distribution parameter set TX of the training dataset can be calculated in the same way. As mentioned above, the training dataset TD retrieved from the model library and the local dataset LD of the client are both associated with the same training task because the training dataset TD is retrieved based on the task identifier associated with the client. Therefore, the training dataset TD and the local dataset LD may adopt similar data organization methods. If the data organization methods of the training dataset retrieved from the model library and the local dataset LD are different, the training dataset can be reorganized (for example, by NWDAF) to obtain the training dataset TD. Accordingly, the distribution parameter set TX of the training dataset TD can have a form similar to the distribution parameter set LX.

[0197] Before calculating the distribution parameter set TX, the training dataset TD may also be preprocessed (e.g., by downsampling, dimensionality reduction, etc.). In the example in this section, the training dataset TD contains the same N, M, and Q values ​​as the local dataset LD. In other examples, the size of the training dataset TD may differ from that of the local dataset LD. Accordingly, one or more of the N, M, and Q values ​​of the training dataset TD may differ from those of the local dataset LD.

[0198] After obtaining the distribution parameter set LX of the local dataset LD and the distribution parameter set TX of the training dataset TD, the similarity between the local dataset LD and the training dataset TD can be measured based on the distribution parameter set LX and the distribution parameter set TX. Figures 10A and 10B illustrate an example process for calculating the similarity between the local dataset and the training dataset according to an embodiment of the present disclosure.

[0199] Figure 10A illustrates an example process for calculating the similarity between a local dataset and a training dataset, according to some embodiments of the present disclosure. As shown, each sample in the local dataset LD and the training dataset TD can be placed in a left column and a right column, respectively. For example, the local dataset LD is placed in the left column, and each sample in the training dataset TD is placed in the right column. Each sample in the left column can be connected to each sample in the right column. Each connection between each sample in the left column and each sample in the right column can be assigned a corresponding weight, thereby creating a bipartite graph. The distance between the local dataset LD and the training dataset TD can be calculated based on this bipartite graph.

[0200] In some embodiments, the KM (Kuhn-Munkres) algorithm can be used to calculate the minimum distribution distance between the local dataset LD and the training dataset TD. The minimum distribution distance is calculated based on the weight of each connection. This minimum distribution distance can represent the similarity between the local dataset LD and the training dataset TD. The larger the minimum distribution distance, the lower the similarity between the local dataset LD and the training dataset TD. In other embodiments, other appropriate algorithms besides the KM algorithm can be used to calculate the distance between the local dataset LD and the training dataset TD, and the calculated distance can be used as an alternative measure of the similarity between the two datasets.

[0201] In some embodiments, the weight assigned to each connection can be calculated based at least in part on the distribution parameter set TX of the distribution parameter set LX. As an example, the weight can be associated with the average sample distance between two samples. In the example of this section, the weight can be equal to the average sample distance. In other examples, the weight can be a function of the average sample distance. In other examples, the weight can also be associated with one or more additional factors in addition to the distribution parameter set TX of the distribution parameter set LX.

[0202] In some embodiments, the average sample distance between two samples can be calculated based at least in part on a set of feature distances d between the respective feature dimensions of the two samples. In the examples of this section, the average sample distance can be expressed as the average of the set of feature distances d. In other examples, the average sample distance between two samples can be measured in other ways. For example, for the hth sample in the local dataset LD and the kth sample in the training dataset TD, the average sample distance between the two samples can be expressed as:

[0203] Where h represents the index of the sample in the local dataset LD in the left column, k represents the index of the sample in the training dataset TD in the right column, and q represents the index of the feature dimension (q is less than or equal to Q). hk(q) represents the feature distance between the h-th sample in the local dataset LD and the k-th sample in the training dataset TD on the q-th feature dimension.

[0204] In some embodiments, the characteristic distance d can be calculated based at least in part on the distribution parameter set LX and the distribution parameter set TX. For example, the characteristic distance d hk (q) can be based on the distribution parameters of the h-th sample in the q-th feature dimension in the local dataset LD (∈LX) and the distribution parameter of the kth sample of the training dataset TD in the qth feature dimension (∈TX) is used to calculate. For example, the feature distance d hk (q) can be calculated as the Euclidean distance between the two-dimensional coordinate points represented by the two distribution parameters.

[0205] FIG10B shows an example of calculating a feature distance between two samples of a local dataset and a training dataset according to some embodiments of the present disclosure.

[0206] As shown in the figure, each element in the distribution parameter set LX of the local dataset LD can be placed in a two-dimensional coordinate system with μ as the horizontal coordinate and δ as the vertical coordinate. Each element in the distribution parameter set TX of the training dataset TD can also be placed in the two-dimensional coordinate system. Each element can be represented as a point in the two-dimensional coordinate system Where n represents the index of the sample, and q represents the index of the feature dimension. For example, n is a positive integer not greater than N, and q is a positive integer not greater than Q. In Figure 10B, the dots can represent elements from the distribution parameter set TX (with subscript i), and the squares can represent elements from the distribution parameter set LX (with subscript j). The distance d can be expressed as the Euclidean distance between two elements (one element from the parameter set TX and the other element from the parameter set LX) in a two-dimensional coordinate system. The distance d can be used to represent the feature distance between two data samples represented by these two elements (one sample from the data set TD and the other sample from the data set LD) on a certain feature dimension. As an example, the distance shown is It represents the feature distance between the first sample of dataset LD and the third sample of dataset TD for the qth feature dimension.

[0207] This distance d can be calculated for each element of the distribution parameter set LX and each element of the distribution parameter set TX. Based on the calculated set of distances d, the average feature distance between two samples of the local dataset LD and the training dataset TD can be determined, as shown in Expression (3). The calculated average feature distance can be used as a weight to calculate the similarity between the local dataset LD and the training dataset TD, as previously described with respect to FIG. 10A .

[0208] As previously described, when there are multiple training datasets TD, a specific training dataset among the multiple training datasets TD that has the highest similarity to the local dataset LD can be selected. A specific model trained based on this specific training dataset can then be selected as the matching model for the local dataset LD (or, in other words, the client that holds this dataset).

[0209] It should be understood that the calculation method described above is merely one or more example embodiments of the technology disclosed herein, and one or more steps of the calculation method described above may be modified or replaced without departing from the scope of the present disclosure.

[0210] 6. Application product examples

[0211] The technology disclosed herein can be applied to various products.

[0212] For example, the control device / base station mentioned in this disclosure can be implemented as any type of base station, such as an eNB, such as a macro eNB and a small eNB. A small eNB can be an eNB that covers a cell smaller than a macro cell, such as a pico eNB, a micro eNB, and a home (femto) eNB. For another example, it can be implemented as a gNB, such as a macro gNB and a small gNB. A small gNB can be a gNB that covers a cell smaller than a macro cell, such as a pico gNB, a micro gNB, and a home (femto) gNB. Alternatively, the base station can be implemented as any other type of base station, such as a NodeB and a base transceiver station (BTS). A base station can include: a main body configured to control wireless communications (also called a base station device); and one or more remote radio heads (RRHs) located at a location different from the main body. In addition, the various types of terminals described below can all operate as a base station by temporarily or semi-permanently performing base station functions. For example, the terminal devices mentioned in the present disclosure may be implemented as mobile terminals (such as smart phones, tablet personal computers (PCs), notebook PCs, portable game terminals, portable / dongle-type mobile routers, and digital camera devices) or vehicle-mounted terminals (such as car navigation devices) in some embodiments. The terminal device may also be implemented as a terminal that performs machine-to-machine (M2M) communication (also known as a machine-type communication (MTC) terminal). In addition, the terminal device may be a wireless communication module (such as an integrated circuit module comprising a single chip) installed on each of the above-mentioned terminals.

[0213] Application examples according to the present disclosure will be described below with reference to the accompanying drawings.

[0214] [Example about base stations]

[0215] It should be understood that the term "base station" in the present disclosure has the full breadth of its usual meaning and includes at least a wireless communication station used as part of a wireless communication system or radio system to facilitate communication. Examples of base stations may include, but are not limited to, the following: a base station may be one or both of a base transceiver station (BTS) and a base station controller (BSC) in a GSM system, one or both of a radio network controller (RNC) and a Node B in a WCDMA system, an eNB in ​​an LTE and LTE-Advanced system, or a corresponding network node in a future communication system (such as a gNB, eLTE eNB, etc. that may appear in a 5G communication system). Some of the functions in the base station of the present disclosure may also be implemented as an entity that has a control function for communication in D2D, M2M, and V2V communication scenarios, or as an entity that plays a spectrum coordination role in a cognitive radio communication scenario.

[0216] First example

[0217] FIG11 is a block diagram illustrating a first exemplary configuration of a gNB to which the techniques of this disclosure may be applied. The gNB 2100 includes multiple antennas 2110 and a base station device 2120. The base station device 2120 and each antenna 2110 may be connected to each other via an RF cable. In one implementation, the gNB 2100 (or base station device 2120) herein may correspond to the control-side electronic device described above.

[0218] Each antenna 2110 includes a single or multiple antenna elements (such as multiple antenna elements included in a multiple-input multiple-output (MIMO) antenna) and is used for base station device 2120 to transmit and receive wireless signals. As shown in Figure 11, gNB 2100 may include multiple antennas 2110. For example, multiple antennas 2110 may be compatible with multiple frequency bands used by gNB 2100.

[0219] The base station device 2120 includes a controller 2121 , a memory 2122 , a network interface 2123 , and a wireless communication interface 2125 .

[0220] The controller 2121 may be, for example, a CPU or a DSP, and operates various functions of the higher layers of the base station device 2120. For example, the controller 2121 determines the location information of a target terminal device in at least one terminal device based on the positioning information of at least one terminal device on the terminal side in the wireless communication system acquired by the wireless communication interface 2125 and the specific location configuration information of at least one terminal device. The controller 2121 may have a logical function of performing the following controls: the control may be, for example, radio resource control, radio bearer control, mobility management, access control, and scheduling. The control may be performed in conjunction with a nearby gNB or core network node. The memory 2122 includes RAM and ROM, and stores programs executed by the controller 2121 and various types of control data (such as a terminal list, transmission power data, and scheduling data).

[0221] The network interface 2123 is a communication interface for connecting the base station device 2120 to the core network 2124. The controller 2121 can communicate with the core network node or another gNB via the network interface 2123. In this case, the gNB 2100 and the core network node or other gNB can be connected to each other via a logical interface (such as an S1 interface and an X2 interface). The network interface 2123 can also be a wired communication interface or a wireless communication interface for wireless backhaul. If the network interface 2123 is a wireless communication interface, the network interface 2123 can use a higher frequency band for wireless communication than the frequency band used by the wireless communication interface 2125.

[0222] The wireless communication interface 2125 supports any cellular communication scheme, such as Long Term Evolution (LTE) and LTE-Advanced, and provides wireless connectivity to terminals located in the gNB 2100 cell via the antenna 2110. The wireless communication interface 2125 may typically include, for example, a baseband (BB) processor 2126 and RF circuitry 2127. The BB processor 2126 can perform various signal processing functions, such as encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing, and performs various signal processing for layers such as Layer 1 (L1), Medium Access Control (MAC), Radio Link Control (RLC), and Packet Data Convergence Protocol (PDCP). In place of the controller 2121, the BB processor 2126 may perform some or all of the aforementioned logical functions. The BB processor 2126 may be a memory storing communication control programs, or a module including a processor configured to execute programs and associated circuitry. Program updates can modify the functionality of the BB processor 2126. This module may be a card or blade inserted into a slot in the base station device 2120. Alternatively, it may be a chip mounted on the card or blade. Meanwhile, the RF circuit 2127 may include, for example, a mixer, a filter, and an amplifier, and transmits and receives wireless signals via the antenna 2110. Although FIG11 shows an example in which one RF circuit 2127 is connected to one antenna 2110, the present disclosure is not limited to this illustration, and one RF circuit 2127 may be connected to multiple antennas 2110 at the same time.

[0223] As shown in Figure 11 , the wireless communication interface 2125 may include multiple BB processors 2126. For example, multiple BB processors 2126 may be compatible with multiple frequency bands used by the gNB 2100. As shown in Figure 11 , the wireless communication interface 2125 may include multiple RF circuits 2127. For example, multiple RF circuits 2127 may be compatible with multiple antenna elements. While Figure 11 illustrates an example in which the wireless communication interface 2125 includes multiple BB processors 2126 and multiple RF circuits 2127, the wireless communication interface 2125 may also include a single BB processor 2126 or a single RF circuit 2127.

[0224] Second example

[0225] FIG12 is a block diagram illustrating a second exemplary configuration of a gNB to which the techniques of this disclosure can be applied. gNB 2200 includes multiple antennas 2210, RRHs 2220, and base station equipment 2230. RRHs 2220 and each antenna 2210 can be connected to each other via an RF cable. Base station equipment 2230 and RRHs 2220 can be connected to each other via a high-speed line such as an optical fiber cable. In one implementation, gNB 2200 (or base station equipment 2230) herein may correspond to the control-side electronic device described above.

[0226] Each antenna 2210 includes a single or multiple antenna elements (such as multiple antenna elements included in a MIMO antenna) and is used for RRH 2220 to transmit and receive wireless signals. As shown in Figure 12, gNB 2200 may include multiple antennas 2210. For example, multiple antennas 2210 may be compatible with multiple frequency bands used by gNB 2200.

[0227] Base station device 2230 includes a controller 2231, a memory 2232, a network interface 2233, a wireless communication interface 2234, and a connection interface 2236. Controller 2231, memory 2232, and network interface 2233 are the same as controller 2121, memory 2122, and network interface 2123 described with reference to FIG.

[0228] The wireless communication interface 2234 supports any cellular communication scheme (such as LTE and LTE-Advanced) and provides wireless communication to terminals located in the sector corresponding to the RRH 2220 via the RRH 2220 and the antenna 2210. The wireless communication interface 2234 may generally include, for example, a BB processor 2235. The BB processor 2235 is identical to the BB processor 2126 described with reference to FIG. 11 , except that the BB processor 2235 is connected to the RF circuit 2222 of the RRH 2220 via the connection interface 2236. As shown in FIG. 12 , the wireless communication interface 2234 may include multiple BB processors 2235. For example, the multiple BB processors 2235 may be compatible with multiple frequency bands used by the gNB 2200. Although FIG. 12 illustrates an example in which the wireless communication interface 2234 includes multiple BB processors 2235, the wireless communication interface 2234 may also include a single BB processor 2235.

[0229] The connection interface 2236 is an interface for connecting the base station device 2230 (wireless communication interface 2234) to the RRH 2220. The connection interface 2236 may also be a communication module for connecting the base station device 2230 (wireless communication interface 2234) to the RRH 2220 for communication in the high-speed line.

[0230] The RRH 2220 includes a connection interface 2223 and a wireless communication interface 2221 .

[0231] The connection interface 2223 is an interface for connecting the RRH 2220 (wireless communication interface 2221) to the base station device 2230. The connection interface 2223 may also be a communication module for communication in the above-mentioned high-speed line.

[0232] The wireless communication interface 2221 transmits and receives wireless signals via the antenna 2210. The wireless communication interface 2221 may generally include, for example, an RF circuit 2222. The RF circuit 2222 may include, for example, a mixer, a filter, and an amplifier, and transmits and receives wireless signals via the antenna 2210. Although FIG12 illustrates an example in which one RF circuit 2222 is connected to one antenna 2210, the present disclosure is not limited to this illustration, and one RF circuit 2222 may be connected to multiple antennas 2210 simultaneously.

[0233] As shown in FIG12 , the wireless communication interface 2221 may include multiple RF circuits 2222. For example, the multiple RF circuits 2222 may support multiple antenna elements. Although FIG12 shows an example in which the wireless communication interface 2221 includes multiple RF circuits 2222, the wireless communication interface 2221 may also include a single RF circuit 2222.

[0234] [Example of User Equipment / Terminal Equipment]

[0235] First example

[0236] 13 is a block diagram illustrating an example of an exemplary configuration of a communication device 2300 (e.g., a smart phone, a contact, etc.) to which the technology of the present disclosure may be applied. The communication device 2300 includes a processor 2301, a memory 2302, a storage device 2303, an external connection interface 2304, a camera 2306, a sensor 2307, a microphone 2308, an input device 2309, a display device 2310, a speaker 2311, a wireless communication interface 2312, one or more antenna switches 2315, one or more antennas 2316, a bus 2317, a battery 2318, and an auxiliary controller 2319. In one implementation, the communication device 2300 (or processor 2301) herein may correspond to the aforementioned transmitting device or terminal-side electronic device.

[0237] The processor 2301 may be, for example, a CPU or a system on a chip (SoC), and controls the functions of the application layer and other layers of the communication device 2300. The memory 2302 includes RAM and ROM, and stores data and programs executed by the processor 2301. The storage device 2303 may include storage media such as semiconductor memories and hard disks. The external connection interface 2304 is an interface for connecting an external device (such as a memory card and a universal serial bus (USB) device) to the communication device 2300.

[0238] The camera 2306 includes an image sensor (such as a charge coupled device (CCD) and a complementary metal oxide semiconductor (CMOS)) and generates a captured image. The sensor 2307 may include a group of sensors such as a measurement sensor, a gyroscope sensor, a geomagnetic sensor, and an acceleration sensor. The microphone 2308 converts the sound input to the communication device 2300 into an audio signal. The input device 2309 includes, for example, a touch sensor, a keypad, a keyboard, a button, or a switch configured to detect a touch on the screen of the display device 2310, and receives an operation or information input from the user. The display device 2310 includes a screen (such as a liquid crystal display (LCD) and an organic light emitting diode (OLED) display) and displays the output image of the communication device 2300. The speaker 2311 converts the audio signal output from the communication device 2300 into sound.

[0239] The wireless communication interface 2312 supports any cellular communication scheme (such as LTE and LTE-Advanced) and performs wireless communication. The wireless communication interface 2312 may generally include, for example, a BB processor 2313 and an RF circuit 2314. The BB processor 2313 may perform, for example, encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing, and perform various types of signal processing for wireless communication. Meanwhile, the RF circuit 2314 may include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via an antenna 2316. The wireless communication interface 2312 may be a chip module on which the BB processor 2313 and the RF circuit 2314 are integrated. As shown in FIG13 , the wireless communication interface 2312 may include multiple BB processors 2313 and multiple RF circuits 2314. Although FIG13 shows an example in which the wireless communication interface 2312 includes multiple BB processors 2313 and multiple RF circuits 2314, the wireless communication interface 2312 may also include a single BB processor 2313 or a single RF circuit 2314.

[0240] In addition, in addition to the cellular communication scheme, the wireless communication interface 2312 can support other types of wireless communication schemes, such as a short-range wireless communication scheme, a near-field communication scheme, and a wireless local area network (LAN) scheme. In this case, the wireless communication interface 2312 can include a BB processor 2313 and an RF circuit 2314 for each wireless communication scheme.

[0241] Each of the antenna switches 2315 switches the connection destination of the antenna 2316 between a plurality of circuits (eg, circuits for different wireless communication schemes) included in the wireless communication interface 2312 .

[0242] Each of the antennas 2316 includes a single or multiple antenna elements (such as multiple antenna elements included in a MIMO antenna) and is used for transmitting and receiving wireless signals via the wireless communication interface 2312. As shown in FIG13 , the communication device 2300 may include multiple antennas 2316. Although FIG13 shows an example in which the communication device 2300 includes multiple antennas 2316, the communication device 2300 may also include a single antenna 2316.

[0243] In addition, the communication device 2300 may include an antenna 2316 for each wireless communication scheme. In this case, the antenna switch 2315 may be omitted from the configuration of the communication device 2300.

[0244] The bus 2317 connects the processor 2301, the memory 2302, the storage device 2303, the external connection interface 2304, the camera 2306, the sensor 2307, the microphone 2308, the input device 2309, the display device 2310, the speaker 2311, the wireless communication interface 2312, and the auxiliary controller 2319. The battery 2318 supplies power to the various blocks of the communication device 2300 shown in FIG13 via a feeder line, which is partially shown as a dotted line in the figure. The auxiliary controller 2319 operates the minimum necessary functions of the communication device 2300, for example, in sleep mode.

[0245] Second example

[0246] Figure 14 is a block diagram showing an example of an exemplary configuration of a car navigation device 2400 to which the technology of the present disclosure can be applied. The car navigation device 2400 includes a processor 2401, a memory 2402, a global positioning system (GPS) module 2404, a sensor 2405, a data interface 2406, a content player 2407, a storage medium interface 2408, an input device 2409, a display device 2510, a speaker 2411, a wireless communication interface 2413, one or more antenna switches 2416, one or more antennas 2417, and a battery 2418. In one implementation, the car navigation device 2400 (or processor 2401) herein may correspond to a transmitting device or a terminal-side electronic device.

[0247] The processor 2401 may be, for example, a CPU or an SoC, and controls a navigation function and other functions of the car navigation apparatus 2400. The memory 2402 includes a RAM and a ROM, and stores data and programs executed by the processor 2401.

[0248] The GPS module 2404 uses GPS signals received from GPS satellites to measure the position (such as latitude, longitude, and altitude) of the car navigation device 2400. The sensor 2405 may include a group of sensors such as a gyroscope sensor, a geomagnetic sensor, and an air pressure sensor. The data interface 2406 is connected to, for example, the vehicle network 2421 via a terminal not shown, and acquires data generated by the vehicle (such as vehicle speed data).

[0249] The content player 2407 reproduces content stored in a storage medium (such as a CD or DVD) inserted into the storage medium interface 2408. The input device 2409 includes, for example, a touch sensor, button, or switch configured to detect a touch on the screen of the display device 2510, and receives an operation or information input from the user. The display device 2510 includes a screen such as an LCD or OLED display and displays an image of a navigation function or reproduced content. The speaker 2411 outputs the sound of the navigation function or the reproduced content.

[0250] The wireless communication interface 2413 supports any cellular communication scheme (such as LTE and LTE-Advanced) and performs wireless communication. The wireless communication interface 2413 may generally include, for example, a BB processor 2414 and an RF circuit 2415. The BB processor 2414 may perform, for example, encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing, and perform various types of signal processing for wireless communication. Meanwhile, the RF circuit 2415 may include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via an antenna 2417. The wireless communication interface 2413 may also be a chip module on which the BB processor 2414 and the RF circuit 2415 are integrated. As shown in Figure 14, the wireless communication interface 2413 may include multiple BB processors 2414 and multiple RF circuits 2415. Although Figure 14 shows an example in which the wireless communication interface 2413 includes multiple BB processors 2414 and multiple RF circuits 2415, the wireless communication interface 2413 may also include a single BB processor 2414 or a single RF circuit 2415.

[0251] In addition, in addition to the cellular communication scheme, the wireless communication interface 2413 can support other types of wireless communication schemes, such as short-range wireless communication schemes, near field communication schemes, and wireless LAN schemes. In this case, for each wireless communication scheme, the wireless communication interface 2413 can include a BB processor 2414 and an RF circuit 2415.

[0252] Each of the antenna switches 2416 switches the connection destination of the antenna 2417 between a plurality of circuits included in the wireless communication interface 2413 , such as circuits for different wireless communication schemes.

[0253] Each of the antennas 2417 includes a single or multiple antenna elements (such as multiple antenna elements included in a MIMO antenna) and is used for transmitting and receiving wireless signals with the wireless communication interface 2413. As shown in Figure 14, the car navigation device 2400 may include multiple antennas 2417. Although Figure 14 shows an example in which the car navigation device 2400 includes multiple antennas 2417, the car navigation device 2400 may also include a single antenna 2417.

[0254] In addition, the car navigation device 2400 may include an antenna 2417 for each wireless communication scheme. In this case, the antenna switch 2416 may be omitted from the configuration of the car navigation device 2400.

[0255] The battery 2418 supplies power to the respective blocks of the car navigation device 2400 shown in Fig. 14 via a feeder line, which is partially shown as a dotted line in the figure. The battery 2418 accumulates the power supplied from the vehicle.

[0256] The technology of the present disclosure can also be implemented as an in-vehicle system (or vehicle) 2420 including a car navigation device 2400, an in-vehicle network 2421, and one or more blocks of a vehicle module 2422. The vehicle module 2422 generates vehicle data (such as vehicle speed, engine speed, and fault information) and outputs the generated data to the in-vehicle network 2421.

[0257] The exemplary embodiments of the present disclosure are described above with reference to the accompanying drawings, but the present disclosure is certainly not limited to the above examples. Those skilled in the art may obtain various changes and modifications within the scope of the appended claims, and it should be understood that these changes and modifications will naturally fall within the technical scope of the present disclosure.

[0258] It should be understood that the machine-executable instructions in the machine-readable storage medium or program product according to the embodiments of the present disclosure can be configured to perform operations corresponding to the above-mentioned device and method embodiments. When referring to the above-mentioned device and method embodiments, the embodiments of the machine-readable storage medium or program product are clear to those skilled in the art and are therefore not described again. Machine-readable storage media and program products for carrying or including the above-mentioned machine-executable instructions also fall within the scope of the present disclosure. Such storage media may include, but are not limited to, floppy disks, optical disks, magneto-optical disks, memory cards, memory sticks, and the like.

[0259] In addition, it should be understood that the above series of processes and devices can also be implemented by software and / or firmware. In the case of implementation by software and / or firmware, the storage medium of the relevant device stores the corresponding program constituting the corresponding software, and when the program is executed, various functions can be performed.

[0260] For example, a plurality of functions included in one unit in the above embodiments may be implemented by separate devices. Alternatively, a plurality of functions implemented by a plurality of units in the above embodiments may be implemented by separate devices, respectively. In addition, one of the above functions may be implemented by a plurality of units. Needless to say, such a configuration is included in the technical scope of the present disclosure.

[0261] In this specification, the steps described in the flowchart include not only processing that is performed in order in time series, but also processing that is performed in parallel or individually rather than necessarily in time series. In addition, even in the steps processed in time series, it goes without saying that the order can be appropriately changed.

[0262] 7. Example embodiments of the present disclosure

[0263] 1. An electronic device, comprising: at least one processing unit; and at least one storage unit, wherein the at least one storage unit comprises computer program code, which, when executed by the at least one processing unit, causes the electronic device to perform the following operations: obtain a local distribution parameter set of a second device, wherein the local distribution parameter set characterizes data distribution characteristics of a local data set of the second device; and send a trained specific model to the second device, wherein the specific model is determined at least in part based on the local distribution parameter set.

[0264] 2. The electronic device of embodiment 1, wherein the operations further comprise: sending the local distribution parameter set together with a client identifier of the second device; and receiving a model list, the model list including the specific model.

[0265] 3. The electronic device of embodiment 1, wherein the electronic device is configured as a server implementing a federated learning process, and the second device is configured as one of a plurality of clients implementing the federated learning process.

[0266] 4. An electronic device as described in Example 3, wherein the operation further includes: sending the local distribution parameter set together with a task identifier, wherein the task identifier identifies a specific training task associated with the second device in the federated learning process; and receiving the specific model, wherein the specific model is determined at least in part based on the task identifier.

[0267] 5. An electronic device as described in Example 3, wherein the operation further includes: receiving a trained local model from the second device, wherein the local model is trained by the second device based on a padded local dataset, wherein the padded local dataset is generated by the specific model based on at least a portion of the local dataset; and generating a global model based at least in part on an aggregation of the local models.

[0268] 6. An electronic device as described in Example 1, wherein the electronic device is configured to implement an application function AF of a cellular network, and the operation further includes: if the AF is not trusted, sending the local distribution parameter set to a network data analysis function NWDAF of the cellular network via a network exposure function NEF of the cellular network; and if the AF is trusted, sending the local distribution parameter set directly to the NWDAF.

[0269] 7. An electronic device as described in Example 1, wherein the specific model is selected from one or more candidate models, and among one or more training data sets used to train the one or more candidate models, the specific training data set used to train the specific model has the highest similarity with the local data set of the second device.

[0270] 8. An electronic device as described in Example 7, wherein the similarity is determined at least in part based on the following items: (1) the local distribution parameter set of the local data set, and (2) the specific distribution parameter set of the specific training data set.

[0271] 9. The electronic device according to embodiment 1, wherein the operation further comprises:

[0272] sending one or more encoder models to the second device, the one or more encoder models configured to determine one or more characteristics associated with the second device;

[0273] receiving the determined one or more characteristics from the second device; and

[0274] Based on the one or more characteristics, one or more tasks are performed.

[0275] 10. The electronic device of embodiment 9, wherein the one or more encoder models include at least one of the following:

[0276] a dataset encoder model configured to determine the data distribution characteristic based on the local dataset of the second device;

[0277] a preference encoder model configured to determine preferences associated with the second device;

[0278] An intent encoder model is configured to determine an intent associated with the second device.

[0279] 11. The electronic device according to embodiment 9, wherein the operation further comprises:

[0280] The one or more encoder models are retrieved from a data analysis repository function ADRF in the cellular network.

[0281] 12. The electronic device of embodiment 1, wherein the operation further comprises: sending a configuration associated with a transmission mode of the second device to the second device, the transmission mode comprising one of the following:

[0282] a first mode associated with enabling an artificial intelligence model; or

[0283] A second mode associated with not enabling the artificial intelligence model.

[0284] 13. The electronic device of embodiment 12, wherein the configuration associated with the first mode further specifies:

[0285] The transmission timing of the second device when the artificial intelligence model is enabled; or

[0286] Parameters of one or more artificial intelligence models used by the second device. 14. An electronic device comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, wherein the computer program code, when executed by the at least one processing unit, causes the electronic device to: send a local distribution parameter set to a first device, the local distribution parameter set characterizing data distribution characteristics of a local data set of the electronic device; and receive a trained specific model from the first device, wherein the specific model is determined at least in part based on the local distribution parameter set.

[0287] 15. The electronic device of embodiment 14, wherein the operation further comprises: using the received specific model to generate an output data portion based on at least a portion of the local data set.

[0288] 16. The electronic device of embodiment 15, wherein the output data portion includes labels, predictions, and / or decisions generated based on at least a portion of the local dataset.

[0289] 17. An electronic device as described in Example 15, wherein the operation further includes: using the output data portion to complete the local data set; using the completed local data set as a training data set to train a local model of the electronic device; and sending the trained local model to the first device for aggregation to generate a global model for a federated learning process.

[0290] 18. The electronic device as described in Example 17, wherein the operation further includes: not participating in the federated learning process during the completion of the local data set.

[0291] 19. The electronic device as described in Example 17, wherein the operation further includes: during the completion of the local data set, not using the at least one portion of the local data set that is not completed to train the local model.

[0292] 20. An electronic device as described in Example 14, wherein the specific model is selected from one or more candidate models, and among one or more training data sets used to train the one or more candidate models, the specific training data set used to train the specific model has the highest similarity with the local data set of the second device.

[0293] 21. An electronic device as described in embodiment 20, wherein the similarity is determined at least in part based on the following items: (1) the local distribution parameter set of the local data set, and (2) the specific distribution parameter set of the specific training data set.

[0294] 22. The electronic device according to embodiment 14, wherein the operation further comprises:

[0295] receiving one or more encoder models;

[0296] executing the one or more encoder models to determine one or more characteristics associated with the electronic device;

[0297] The determined one or more characteristics are sent to the first device.

[0298] 23. The electronic device of embodiment 22, wherein the one or more encoder models include at least one of the following:

[0299] a data set encoder model configured to determine the data distribution characteristic based on the local data set of the electronic device;

[0300] a preference encoder model configured to determine preferences associated with the electronic device;

[0301] An intent encoder model is configured to determine an intent associated with the electronic device.

[0302] 24. The electronic device of embodiment 22, wherein the one or more encoder models are maintained at a data analysis repository function (ADRF) in a cellular network.

[0303] 25. The electronic device of embodiment 14, wherein the operations further comprise: receiving a configuration associated with a transmission mode of the electronic device, the transmission mode comprising one of the following:

[0304] a first mode associated with enabling an artificial intelligence model; or

[0305] A second mode associated with not enabling the artificial intelligence model.

[0306] 26. The electronic device of embodiment 25, wherein the configuration associated with the first mode further specifies:

[0307] The transmission timing of the electronic device when the artificial intelligence model is enabled; or

[0308] Parameters of one or more artificial intelligence models used by the electronic device.

[0309] 27. An electronic device, comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, wherein the computer program code, when executed by the at least one processing unit, causes the electronic device to perform the following operations: receive one or more local distribution parameter sets, each local distribution parameter set in the one or more local distribution parameter sets characterizing data distribution characteristics of a local data set of a corresponding client in one or more clients; send a task identifier to a model library, the task identifier identifying a training task associated with the one or more clients; receive one or more trained candidate models corresponding to the task identifier from the model library; and send a model list, the model list comprising a matching model selected from the one or more candidate models for each client in the one or more clients, the matching model being determined at least in part based on the corresponding local distribution parameter set in the one or more local distribution parameter sets.

[0310] 28. An electronic device as described in Example 27, wherein the operation further includes: for each of the one or more clients: determining a specific training data set in one or more training data sets that has the highest similarity with the local data set of the client; and determining a candidate model in the one or more candidate models corresponding to the specific training data set as a matching model for the client.

[0311] 29. An electronic device as described in Example 28, wherein the similarity between the local dataset of the client and the specific training dataset is determined at least in part based on the following items: (1) a local distribution parameter set of the local dataset, and (2) a specific distribution parameter set of the specific training dataset.

[0312] 30. An electronic device as described in Example 29, wherein: the specific distribution parameter set is received from the model library; or the specific distribution parameter set is determined based on the specific training data set in the one or more training data sets received from the model library.

[0313] 31. The electronic device of embodiment 27, wherein the electronic device is configured to implement a network data analysis function (NWDAF) in a cellular network, and the model library resides at a data analysis repository function (ADRF) in the cellular network.

[0314] 32. The electronic device of embodiment 27, wherein the one or more local distribution parameter sets and the task identifier are received from a server for machine learning.

[0315] 33. An electronic device as described in Example 32, wherein receiving the one or more distributed parameter sets and the task identifier includes: receiving the one or more distributed parameter sets and the task identifier directly from the server; or receiving the one or more distributed parameter sets and the task identifier from the server via a network exposure function NEF of a cellular network.

[0316] 34. The electronic device of embodiment 27, wherein sending the model list comprises sending the matching model of each of the one or more clients together with the client identifier of the client to the server.

[0317] 35. An electronic device as described in Example 28, wherein each of the one or more local distribution parameter sets includes at least one of the following: a mean and a standard deviation for a Gaussian distribution; a shape parameter and an inverse scale parameter for a gamma distribution; or a mean and a variance for a lognormal distribution.

[0318] 36. A method comprising the following steps: obtaining a local distribution parameter set of a second device, wherein the local distribution parameter set characterizes data distribution characteristics of a local data set of the second device; and sending a trained specific model to the second device, wherein the specific model is determined at least in part based on the local distribution parameter set.

[0319] 37. A method comprising the steps of: sending a local distribution parameter set to a first device, wherein the local distribution parameter set characterizes data distribution characteristics of a local data set of the electronic device; and receiving a trained specific model from the first device, wherein the specific model is determined at least in part based on the local distribution parameter set.

[0320] 38. A method comprising the following steps: receiving one or more local distribution parameter sets, each of the one or more local distribution parameter sets characterizing data distribution characteristics of a local data set of a corresponding client in one or more clients; sending a task identifier to a model library, the task identifier identifying a training task associated with the one or more clients; receiving one or more trained candidate models corresponding to the task identifier from the model library; and sending a model list, the model list including a matching model selected from the one or more candidate models for each client in the one or more clients, the matching model being determined at least in part based on the corresponding local distribution parameter set in the one or more local distribution parameter sets.

[0321] 39. A computer-readable storage medium storing one or more instructions, which, when executed by one or more processing circuits of an electronic device, cause the electronic device to perform the method of any one of embodiments 36-38.

[0322] 40. A computer program product comprising a computer program, which, when executed by a processor, performs the method of any one of embodiments 36-38.

[0323] 41. An apparatus comprising means for performing the method of any one of embodiments 36-38.

Claims

1. An electronic device comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, the computer program code, when executed by the at least one processing unit, causing the electronic device to perform the following operations: Obtaining a local distribution parameter set of a second device, where the local distribution parameter set represents a data distribution characteristic of a local data set of the second device; as well as The trained specific model is sent to the second device, wherein the specific model is determined at least in part based on the local distribution parameter set.

2. The electronic device according to claim 1, wherein The operations further include: sending the locally distributed parameter set together with a client identifier of the second device; and A model list is received, the model list including the specific model.

3. The electronic device according to claim 1, wherein The electronic device is configured as a server implementing a federated learning process, and the second device is configured as one of a plurality of clients implementing the federated learning process.

4. The electronic device according to claim 3, wherein: The operations further include: sending the set of local distribution parameters along with a task identifier, the task identifier identifying a specific training task associated with the second device during the federated learning process; and The specific model is received, wherein the specific model is determined based at least in part on the task identifier.

5. The electronic device according to claim 3, wherein: The operations further include: receiving a trained local model from the second device, the local model being trained by the second device based on the padded local dataset, wherein the padded local dataset is generated by the specific model based on at least a portion of the local dataset; and A global model is generated based at least in part on the aggregation of the local models.

6. The electronic device according to claim 1, wherein The electronic device is configured to implement an application function AF of a cellular network, and the operations further include: If the AF is not trusted, sending the local distribution parameter set to a network data analysis function NWDAF of the cellular network via a network exposure function NEF of the cellular network; and If the AF is trusted, the local distribution parameter set is sent directly to the NWDAF.

7. The electronic device according to claim 1, wherein The specific model is selected from one or more candidate models, and among one or more training data sets used to train the one or more candidate models, the specific training data set used to train the specific model has the highest similarity with the local data set of the second device.

8. The electronic device according to claim 7, wherein: The similarity is determined based at least in part on: (1) the local distribution parameter set of the local dataset, and (2) the specific distribution parameter set of the specific training dataset.

9. The electronic device according to claim 1, wherein The operations further include: sending one or more encoder models to the second device, the one or more encoder models configured to determine one or more characteristics associated with the second device; receiving the determined one or more characteristics from the second device; and Based on the one or more characteristics, one or more tasks are performed.

10. The electronic device according to claim 9, wherein The one or more encoder models include at least one of: a dataset encoder model configured to determine the data distribution characteristic based on the local dataset of the second device; a preference encoder model configured to determine preferences associated with the second device; An intent encoder model is configured to determine an intent associated with the second device.

11. The electronic device according to claim 9, wherein The operations further include: The one or more encoder models are retrieved from a data analysis repository function ADRF in the cellular network.

12. The electronic device according to claim 1, wherein The operations further include sending, to the second device, a configuration associated with a transmission mode of the second device, the transmission mode comprising one of the following: a first mode associated with enabling an artificial intelligence model; or A second mode associated with not enabling the artificial intelligence model.

13. The electronic device according to claim 12, wherein: The configuration associated with the first mode further specifies: The transmission timing of the second device when the artificial intelligence model is enabled; or Parameters of one or more artificial intelligence models used by the second device.

14. An electronic device comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, wherein the computer program code, when executed by the at least one processing unit, causes the electronic device to perform the following operations: Sending a local distribution parameter set to a first device, where the local distribution parameter set represents a data distribution characteristic of a local data set of the electronic device; as well as A trained specific model is received from a first device, wherein the specific model is determined at least in part based on the local distribution parameter set.

15. The electronic device according to claim 14, wherein The operations further include: An output data portion is generated based on at least a portion of the local data set using the received specific model.

16. The electronic device according to claim 15, wherein The output data portion includes labels, predictions, and / or decisions generated based on the at least a portion of the local dataset.

17. The electronic device according to claim 15, wherein: The operations further include: Using the output data portion to complete the local data set; Using the padded local data set as a training data set to train a local model of the electronic device; and The trained local models are sent to the first device for aggregation to generate a global model for the federated learning process.

18. The electronic device according to claim 17, wherein: The operations further include: While completing the local dataset, it does not participate in the federated learning process.

19. The electronic device according to claim 17, wherein: The operations further include: During the completion of the local dataset, the at least one uncompleted portion of the local dataset is not used to train the local model.

20. The electronic device according to claim 14, wherein The specific model is selected from one or more candidate models, and among one or more training data sets used to train the one or more candidate models, the specific training data set used to train the specific model has the highest similarity with the local data set of the second device.

21. The electronic device according to claim 20, wherein: The similarity is determined based at least in part on: (1) the set of local distribution parameters of the local data set, and (2) A specific distribution parameter set of the specific training data set.

22. The electronic device according to claim 14, wherein The operations further include: receiving one or more encoder models; executing the one or more encoder models to determine one or more characteristics associated with the electronic device; The determined one or more characteristics are sent to the first device.

23. The electronic device according to claim 22, wherein: The one or more encoder models include at least one of: a data set encoder model configured to determine the data distribution characteristic based on the local data set of the electronic device; a preference encoder model configured to determine preferences associated with the electronic device; An intent encoder model is configured to determine an intent associated with the electronic device.

24. The electronic device according to claim 22, wherein The one or more encoder models are maintained at a data analysis repository function ADRF in the cellular network.

25. The electronic device according to claim 14, wherein The operations further include: Receiving a configuration associated with a transmission mode of the electronic device, the transmission mode comprising one of the following: a first mode associated with enabling an artificial intelligence model; or A second mode associated with not enabling the artificial intelligence model.

26. The electronic device according to claim 25, wherein The configuration associated with the first mode further specifies: The transmission timing of the electronic device when the artificial intelligence model is enabled; or Parameters of one or more artificial intelligence models used by the electronic device.

27. An electronic device comprising: at least one processing unit; and at least one storage unit, the at least one storage unit comprising computer program code, wherein the computer program code, when executed by the at least one processing unit, causes the electronic device to perform the following operations: receiving one or more local distribution parameter sets, each of the one or more local distribution parameter sets characterizing a data distribution characteristic of a local data set of a corresponding client among the one or more clients; Sending a task identifier to a model repository, where the task identifier identifies a training task associated with the one or more clients; receiving, from the model library, one or more trained candidate models corresponding to the task identifier; as well as A model list is sent, the model list including a matching model selected from the one or more candidate models for each of the one or more clients, the matching model being determined based at least in part on a corresponding local distributed parameter set of the one or more local distributed parameter sets.

28. The electronic device according to claim 27, wherein The operations further include: For each of the one or more clients: determining a specific training data set among the one or more training data sets that has the highest similarity to the local data set of the client; and A candidate model corresponding to the specific training dataset among the one or more candidate models is determined as a matching model for the client.

29. The electronic device according to claim 28, wherein The similarity between the client's local dataset and the specific training dataset is determined based at least in part on: (1) a local distribution parameter set of the local dataset, and (2) a specific distribution parameter set of the specific training dataset.

30. The electronic device of claim 29, wherein: The specific distribution parameter set is received from the model library; or The specific distribution parameter set is determined based on the specific training data set among the one or more training data sets received from the model library.

31. The electronic device according to claim 27, wherein The electronic device is configured to implement a network data analysis function NWDAF in a cellular network, and the model library resides at a data analysis repository function ADRF in the cellular network.

32. The electronic device according to claim 27, wherein The one or more local distribution parameter sets and the task identifier are received from a server for machine learning.

33. The electronic device according to claim 32, wherein: Receiving the one or more distribution parameter sets and the task identifier includes: receiving the one or more distributed parameter sets and the task identifier directly from the server; or The one or more distributed parameter sets and the task identifier are received from the server via a Network Exposure Function (NEF) of a cellular network.

34. The electronic device according to claim 27, wherein Sending the model list includes sending the matching model of each client among the one or more clients together with the client identifier of the client to the server.

35. The electronic device according to claim 28, wherein Each of the one or more local distribution parameter sets includes at least one of the following: for the mean and standard deviation of a Gaussian distribution; shape parameter and inverse scale parameter for the gamma distribution; or Mean and variance for the lognormal distribution.

36. A method comprising the steps of: Obtaining a local distribution parameter set of a second device, where the local distribution parameter set represents a data distribution characteristic of a local data set of the second device; as well as The trained specific model is sent to the second device, wherein the specific model is determined at least in part based on the local distribution parameter set.

37. A method comprising the steps of: Sending a local distribution parameter set to a first device, where the local distribution parameter set represents a data distribution characteristic of a local data set of the electronic device; as well as A trained specific model is received from a first device, wherein the specific model is determined at least in part based on the local distribution parameter set.

38. A method comprising the steps of: receiving one or more local distribution parameter sets, each of the one or more local distribution parameter sets characterizing a data distribution characteristic of a local data set of a corresponding client among the one or more clients; Sending a task identifier to a model repository, where the task identifier identifies a training task associated with the one or more clients; receiving, from the model library, one or more trained candidate models corresponding to the task identifier; as well as A model list is sent, the model list including a matching model selected from the one or more candidate models for each of the one or more clients, the matching model being determined based at least in part on a corresponding local distributed parameter set of the one or more local distributed parameter sets.

39. A computer-readable storage medium storing one or more instructions, which, when executed by one or more processing circuits of an electronic device, cause the electronic device to perform the method according to any one of claims 36 to 38.

40. A computer program product comprising a computer program which, when executed by a processor, performs the method of any one of claims 36 to 38.

41. An apparatus comprising means for performing the method of any one of claims 36-38.

Images