Training method and device of federal global model

Abstract

The application discloses a training method and device of a federal global model. When a target training request initiated by a target electronic device is acquired, a target device set to which the target electronic device belongs is determined according to a preset device set mapping relationship, wherein the device set mapping relationship comprises a mapping relationship between each electronic device and each device set; whether a total quantity of training requests that have passed in a current training period of electronic devices in the target device set exceeds a preset request quantity corresponding to the target device set is judged; if the total quantity of training requests does not exceed the preset request quantity corresponding to the target device set, passing indication information passing the target training request is sent, and the passing indication information is used to instruct the target electronic device to train the current federal global model. The application can improve the accuracy of a model obtained through federal learning.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a training method and apparatus for a federated global model. Background Technology

[0002] To address privacy concerns during joint model training, federated learning (FL) has been proposed. Federated learning allows individual electronic devices to train their own models. After training, each device uploads its model parameters to a server. The server aggregates these parameters and distributes new parameters back to the devices, which then apply them to their models to continue training. This process can be iterated repeatedly until the model converges or other conditions are met. Existing federated global model training systems typically consist of a selector, an aggregator, a coordinator, and client runtimes deployed on the devices. The coordinator coordinates the devices and aggregators for each task; the client connects to the selector, which publishes tasks, selects clients, and summarizes client information to the coordinator; client-supplied data is routed through the selector to the corresponding aggregator for aggregation. However, this approach suffers from inconsistent data volumes and sampling skew during training iterations, leading to training instability and reduced accuracy of the federated learning model.

[0003] In other words, the accuracy of models obtained through federated learning in existing technologies is relatively low. Summary of the Invention

[0004] This application provides a training method and apparatus for a federated global model, which can improve the accuracy of the model obtained by federated learning, which is relatively low.

[0005] Firstly, the training method for the federated global model provided in this application, applied to a server, includes:

[0006] When a target training request initiated by a target electronic device is received, the target device set to which the target electronic device belongs is determined according to a preset device set mapping relationship, wherein the device set mapping relationship includes the mapping relationship between each electronic device and each device set;

[0007] Determine whether the total number of training requests that electronic devices in the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set;

[0008] If the total number of requests does not exceed the preset number of requests corresponding to the target device set, then a pass instruction message for passing the target training request is issued. The pass instruction message is used to instruct the target electronic device to train the current federated global model.

[0009] Secondly, the training method for the federated global model provided in this application, when applied to electronic devices, includes:

[0010] Send a target training request to the server;

[0011] When the server returns the pass indication information, the current federated global model on the server is obtained. The pass indication information is obtained by the server when it receives the target training request initiated by the electronic device, and the server determines the target device set to which the electronic device belongs according to the preset device set mapping relationship. The server issues the pass indication information when it determines that the total number of training requests that each electronic device in the target device set has passed in the current training cycle does not exceed the preset number of requests corresponding to the target device set.

[0012] The current federated global model is trained locally to obtain the client model parameters;

[0013] The client model parameters are returned to the server.

[0014] Thirdly, the training apparatus for the federal global model provided in this application includes:

[0015] The determination module is used to determine the target device set to which the target electronic device belongs based on a preset device set mapping relationship when a target training request initiated by the target electronic device is obtained. The device set mapping relationship includes the mapping relationship between each electronic device and each device set.

[0016] The judgment module is used to determine whether the total number of training requests that electronic devices in the target device set have passed in the current training period exceeds the preset number of requests corresponding to the target device set.

[0017] The sending module is configured to send a pass indication information for the target training request if the total number of requests does not exceed the preset number of requests corresponding to the target device set. The pass indication information is used to instruct the target electronic device to train the current federated global model.

[0018] Fourthly, the training apparatus for the federal global model provided in this application includes:

[0019] The sending module is used to send target training requests to the server;

[0020] The acquisition module is used to acquire the current federated global model on the server when the server returns the pass indication information. The pass indication information is obtained by the server when it receives the target training request initiated by the electronic device, determines the target device set to which the electronic device belongs according to the preset device set mapping relationship, and determines that the total number of training requests that each electronic device in the target device set has passed in the current training period does not exceed the preset number of requests corresponding to the target device set.

[0021] The training module is used to train the current federated global model locally to obtain client model parameters;

[0022] The return module is used to return the client model parameters to the server.

[0023] Fifthly, the server provided in this application includes a memory and a processor. The memory stores a computer program, and the processor is used to run the computer program in the memory to implement the steps in the training method of the federated global model provided in the first aspect of this application.

[0024] Sixthly, the electronic device provided in this application includes a memory and a processor. The memory stores a computer program, and the processor is used to run the computer program in the memory to implement the steps in the training method of the federated global model provided in the second aspect of this application.

[0025] Seventhly, the computer-readable storage medium provided in this application stores multiple instructions that are adapted for loading by a processor to implement the steps in the training method of the federated global model provided in this application.

[0026] Eighthly, the computer program product provided in this application includes a computer program or instructions that, when executed by a processor, implement the steps in the training method of the federated global model provided in this application.

[0027] In this application, compared with related technologies, when a target training request initiated by a target electronic device is obtained, the target device set to which the target electronic device belongs is determined. Then, the target electronic device is instructed to perform model training only when the total number of training requests that have been approved within the target device set does not exceed the preset number of requests corresponding to the target device set. This can avoid too many electronic devices in a certain device set participating in training, thereby ensuring that the data of electronic devices in each device set are sampled evenly, making the amount of data in each iteration more consistent, and making the training more stable. This can improve the accuracy of the model obtained by federated learning. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of a training system for the federated global model provided in this application embodiment;

[0030] Figure 2 This is a flowchart illustrating one embodiment of the training method for the federated global model provided in this application.

[0031] Figure 3 This is a flowchart illustrating the steps before determining whether the total number of training requests passed by electronic devices in the target device set during the current training period exceeds the preset number of requests corresponding to the target device set, in one embodiment of the training method for the federated global model provided in this application.

[0032] Figure 4 This is a flowchart illustrating another embodiment of the training method for the federated global model provided in this application.

[0033] Figure 5 This is a flowchart illustrating yet another embodiment of the training method for the federated global model provided in this application.

[0034] Figure 6 This is a flowchart illustrating yet another embodiment of the training method for the federated global model provided in this application.

[0035] Figure 7 This is a schematic diagram of the structure of an embodiment of the training device for the federated global model provided in this application;

[0036] Figure 8 This is a schematic diagram of another embodiment of the training device for the federated global model provided in this application;

[0037] Figure 9 This is a schematic diagram of the structure of the electronic device / server provided in the embodiments of this application. Detailed Implementation

[0038] It should be noted that the principles of this application are illustrated by example in a suitable computing environment. The following description is based on the specific embodiments of this application that are illustrated, and should not be regarded as limiting other specific embodiments not detailed herein.

[0039] In the following description of this application, "some embodiments" are referred to, which describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subset of all possible embodiments, and may be combined with each other without conflict.

[0040] In the following description of this application, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0042] To improve the effectiveness of federated learning, embodiments of this application provide a method for training a federated global model, a training device for a federated global model, an electronic device, a computer-readable storage medium, and a computer program product. The training method for the federated global model can be executed by the training device for the federated global model, or by an electronic device integrating the training device for the federated global model; the training device for the federated global model can be executed by the training device for the federated global model, or by a server integrating the training device for the federated global model.

[0043] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0044] Please refer to Figure 1 This application also provides a training system for a federated global model, such as Figure 1 As shown, the training system for the federated global model includes a server 200 and multiple electronic devices 100. Each electronic device 100 integrates a training device for the federated global model provided in this application, and the server 200 also integrates a training device for the federated global model provided in this application. The server 200 is network-connected to each electronic device 100. The number of electronic devices 100 can be set according to specific circumstances.

[0045] Among them, electronic device 100 can be any device equipped with a processor and having processing capabilities, such as mobile electronic devices with processors such as smartphones, tablets, PDAs, laptops, and smart speakers, or fixed electronic devices with processors such as desktop computers, televisions, servers, and industrial equipment. Server 200 can be any device equipped with a processor and having processing capabilities, such as fixed electronic devices with processors such as servers and industrial equipment.

[0046] In addition, the training system for this federated global model may also include a memory for storing raw data, intermediate data, and result data.

[0047] In this embodiment of the application, the storage device can be a cloud storage device. Cloud storage is a new concept that is extended and developed from the concept of cloud computing. A distributed cloud storage system (hereinafter referred to as a storage system) refers to a storage system that uses cluster applications, grid technology and distributed storage file system functions to bring together a large number of storage devices of various types in the network (storage devices are also called storage nodes) through application software or application interfaces to work together to provide data storage and business access functions to the outside world.

[0048] Currently, the storage method of storage systems is as follows: Logical volumes are created. During the creation of a logical volume, physical storage space is allocated to each logical volume. This physical storage space may consist of a single storage device or the disks of several storage devices. Clients store data on a logical volume, which means storing the data on the file system. The file system divides the data into many parts, each part being an object. Each object contains not only the data but also additional information such as a data identifier (ID, ID entity). The file system writes each object to the physical storage space of that logical volume and records the storage location information of each object. Therefore, when a client requests access to data, the file system can allow the client to access the data based on the storage location information of each object.

[0049] The process by which a storage system allocates physical storage space to a logical volume is as follows: the physical storage space is pre-divided into strips according to the capacity estimate of the objects stored in the logical volume (this estimate often has a large margin relative to the actual capacity of the objects to be stored) and the grouping of Redundant Array of Independent Disks (RAID). A logical volume can be understood as a strip, thus allocating physical storage space to the logical volume.

[0050] It should be noted that, Figure 1The schematic diagram of the training system for the federated global model shown is merely an example. The training system and scenario of the federated global model described in this application are for the purpose of more clearly illustrating the technical solutions of this application and do not constitute a limitation on the technical solutions provided in this application. As those skilled in the art will know, with the evolution of the training system for the federated global model and the emergence of new business scenarios, the technical solutions provided in this application are also applicable to similar technical problems.

[0051] The following sections provide detailed descriptions of each example. It should be noted that the sequence numbers of the following embodiments are not intended to limit the preferred order of the embodiments.

[0052] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating one embodiment of the training method for the federated global model provided in this application, as shown below. Figure 2 As shown, the training method of the federated global model provided in this application is applied to a server. The process of the training method of the federated global model provided in this application is as follows:

[0053] 201. When a target training request initiated by the target electronic device is obtained, the target device set to which the target electronic device belongs is determined according to the preset device set mapping relationship.

[0054] The device set mapping relationship includes the mapping relationship between each electronic device and each device set.

[0055] In this embodiment, the mapping relationship between individual electronic devices and device sets can be pre-defined. For example, electronic devices located in the same time zone can be placed into the same device set. The formal time zone division includes 24 time zones, each represented by a single letter. Each time zone is divided every 15° of longitude, with one exception: each time zone has a central meridian. For example, based on the time zones, there are 24 device sets: Device Set 1, Device Set 2, ..., Device Set 24. Each device set includes at least one electronic device.

[0056] In one specific embodiment, when a target training request initiated by a target electronic device is obtained, the location of the target electronic device is obtained, the time zone of the target electronic device is determined based on the location of the target electronic device, and the target device set to which the target electronic device belongs is determined based on the time zone of the target electronic device.

[0057] In this embodiment, at the start of the current training cycle, each electronic device sends a training request to the server. The server then identifies each electronic device as the target electronic device based on the order in which the training requests are received. One training cycle is one iterative training iteration, and the number of training cycles can be set according to specific circumstances.

[0058] 202. Determine whether the total number of training requests that electronic devices in the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set.

[0059] In this embodiment of the application, at the start of the current training cycle, the server initializes the total number of training requests that have passed to a preset initial value, for example, the preset initial value is zero.

[0060] Specifically, when a target training request initiated by a target electronic device is received, it is determined whether the target electronic device is the first electronic device to initiate a training request in the current training cycle. If the target electronic device is the first electronic device to initiate a training request in the current training cycle, the total number of training requests that have been passed is determined by incrementing a preset initial value by one. If the target electronic device is not the first electronic device to initiate a training request in the current training cycle, the total number of training requests that have been passed is obtained, and it is determined whether the total number of training requests that have been passed by electronic devices in the target device set in the current training cycle exceeds the preset number of requests corresponding to the target device set.

[0061] The preset number of requests corresponding to the target device set can be set as needed, for example, 100 or other values. The preset number of requests for different target device sets can be different or the same.

[0062] 203. If the total number of requests does not exceed the preset number of requests corresponding to the target device set, then issue a pass instruction message for passing the target training request.

[0063] The instruction information is used to instruct the target electronic device to train the current federal global model.

[0064] If the total number of requests does not exceed the preset number of requests corresponding to the target device set, it means that there are relatively few devices in the device set that have passed the request, and the electronic devices in the device set can continue to be used for training. In this case, a pass instruction message for passing the target training request is issued. The instruction message is used to instruct the target electronic device to train the current federated global model.

[0065] If the total number of requests exceeds the preset number, it means that there are too many devices in the device set that have made the requests, and the electronic devices in the device set cannot continue to be used for training. In this case, a rejection instruction message is issued to reject the target training request. The rejection instruction message is used to prohibit the target electronic device from training the current federated global model.

[0066] Furthermore, after issuing an instruction to approve the target training request, the total number of approved training requests is incremented and updated. After issuing a rejection instruction to reject the target training request, the total number of approved training requests remains unchanged.

[0067] Furthermore, the training method for the federated global model also includes: obtaining the client model parameters returned by each electronic device after training the current federated global model, aggregating at least two client model parameters, and obtaining the updated federated global model. Specifically, after the target electronic device trains the current federated global model and obtains the client model parameters, it returns the client model parameters to the server. The server then obtains the client model parameters returned by each electronic device after training the current federated global model, aggregating at least two client model parameters to obtain the updated federated global model.

[0068] In one specific embodiment, the process involves obtaining client model parameters returned by each electronic device after training the current federated global model, and aggregating at least two client model parameters to obtain an updated federated global model. This includes: obtaining client model parameters returned by each electronic device after training the current federated global model according to a preset aggregation period; and aggregating at least two client model parameters obtained within the preset aggregation period to obtain the updated federated global model. The preset aggregation period can be 2 hours, 3 hours, etc., and can be set according to specific circumstances.

[0069] In another specific embodiment, the process involves obtaining client model parameters returned by each electronic device after training the current federated global model, and aggregating at least two client model parameters to obtain an updated federated global model. This includes: obtaining client model parameters returned by each electronic device after training the current federated global model; and when the obtained client model parameters reach a preset number, aggregating the preset number of client model parameters to obtain the updated federated global model. The preset number of models can be 10, 15, etc., and can be set according to specific circumstances.

[0070] In this embodiment, aggregating at least two client model parameters to obtain an updated federated global model may include: obtaining the aggregation weight coefficients of each client model parameter; aggregating at least two client model parameters based on the aggregation weight coefficients of each client model parameter to obtain the updated federated global model. The aggregation weight coefficients of each client model parameter can be preset. Specifically, aggregating at least two client model parameters based on the aggregation weight coefficients of each client model parameter to obtain updated model parameters, and then updating the current federated global model using the updated model parameters to obtain the updated federated global model.

[0071] To improve the accuracy of the aggregated weight parameters, in a specific embodiment, obtaining the aggregated weight coefficients of each client model parameter may include:

[0072] (1) Obtain the initial return time parameters of each client model parameter in the current training cycle.

[0073] The initial return time parameter indicates how fast the client model parameters are returned. The larger the initial return time parameter is, the slower the client model parameters are returned, and the smaller the initial return time parameter is, the faster the client model parameters are returned.

[0074] In one specific embodiment, the model training cycle in which the client model parameters were first iterated and the number of model aggregations h that the server has currently completed are obtained. The difference between the number of model aggregations h that the server has currently completed and the model training cycle in which the client model parameters started training is determined as the initial return time parameter. Each time the server aggregates the model, the number of model aggregations increases by one. For example, if the number of model aggregations is h, it is 0 at the start of the first round of training. For instance, if the current training cycle is the kth training cycle, and the target electronic device iterates the client model parameters for the first time in the (k-1)th training cycle and returns converged client model parameters in the current training cycle, then the number of model aggregations the server has currently completed is h, and the initial return time parameter d is h+1-k.

[0075] (2) Determine the aggregate weight coefficient of each client model parameter based on the initial return time parameter of each client model parameter in the current training cycle.

[0076] In this context, a larger initial return time parameter d indicates a slower return, and a smaller aggregate weight coefficient for the corresponding client-side model parameters. For example, the aggregate weight coefficient could be (1-1 / (1+e^(-d))), where e is the natural constant and d is the initial return time parameter. Since a larger initial return time parameter indicates a slower model return, it should be discarded, and its weight should be set smaller.

[0077] Furthermore, when the training period is short, the return time requirement for the returned model is high. In this case, the slower-returning model needs to be discarded. When the training period is long, the return time requirement for the returned model is low. In this case, all models need to be utilized to improve the convergence speed.

[0078] In another specific embodiment, the aggregate weight coefficients of each client-side model parameter are determined based on the initial return time parameters of each client-side model parameter in the current training cycle, including:

[0079] (1) Obtain the speed parameter influence factor of each client model parameter in the current training cycle.

[0080] Among them, the velocity parameter influence factor 'a' decreases as the training cycle increases. Specifically, during the first training cycle on the server, the velocity parameter influence factor 'a' is set to the initial value A that influences the shadow.

[0081] In a specific embodiment, a preset annealing step size and a preset number of training iterations are obtained. The product of the preset annealing step size and the preset number of training iterations is determined as the initial value A of the influence shadow. The initial value A of the influence shadow is decreased round by round according to the preset annealing step size s to obtain the velocity parameter influence factor a of each client model parameter in the current training cycle. Here, the preset annealing step size s is a positive number and can be 1, 2, etc. The preset number of training iterations can be 1000, 100, etc. The preset annealing step size and the preset number of training iterations can be set according to specific circumstances. a = A - ks.

[0082] (2) Based on the influence factor of the velocity parameter, the initial return time parameters of each client model parameter are proportionally corrected to obtain the corrected return time parameters of each client model parameter.

[0083] Specifically, the velocity parameter influence factor is multiplied by the velocity parameter influence factors of each client-side model parameter to obtain the corrected velocity parameter influence factor for each client-side model parameter. For example, if the velocity parameter influence factor is 'a' and the initial return time parameter is 'd', the corrected velocity parameter influence factor is a*d.

[0084] (3) Determine the aggregate weight coefficient of each client model parameter based on the corrected return time parameter of each client model parameter in the current training cycle.

[0085] In this context, a larger adjustment rate factor indicates a slower return time, and consequently, a smaller aggregate weight coefficient for the corresponding client-side model parameters. For example, the aggregate weight coefficient might be (1-1 / (1+e^(-a*d))), where e is the natural constant, a*d is the adjusted return time parameter, a is the adjustment rate factor, and d is the initial return time parameter. When the training period is short, the adjustment rate factor a is larger. In this case, the difference in the initial return time parameter d between different client-side models is amplified by the adjustment rate factor a, resulting in a larger difference in the aggregate weight coefficients of each client-side model parameter. This allows models with slower returns to be discarded. Conversely, when the training period is long, the adjustment rate factor a is smaller. In this case, the difference in the initial return time parameter d between different client-side models is reduced by the adjustment rate factor a, resulting in a smaller difference in the aggregate weight coefficients of each client-side model parameter. This allows all models to be utilized, improving the convergence speed.

[0086] Furthermore, the server stores at least two different current federated global models, and the server issues different training tasks for each different current federated global model. The electronic device initiates a target training request for one of the at least two different current federated global models.

[0087] To further ensure uniform sampling, please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the steps before determining whether the total number of training requests passed by electronic devices within the target device set during the current training period exceeds the preset number of requests corresponding to the target device set, as shown in one embodiment of the training method for the federated global model provided in this application. Figure 3 As shown in this embodiment, before determining whether the total number of training requests that electronic devices within the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set, the process includes:

[0088] 301. Obtain the preset device allocation ratio for each device set.

[0089] In one specific embodiment, the preset device allocation ratio of each device set can be set manually.

[0090] In another specific embodiment, to ensure reasonable allocation, if the current training period is the first round, the number of training requests initiated by electronic devices in each device set within a preset historical time period is obtained, and the proportion of training requests initiated by electronic devices in each device set is determined as the preset device allocation ratio for each device set. If the current training period is not the first round, the number of devices in each device set that have returned client models in the previous training period is obtained, and the proportion of devices in each device set that have returned client models in the previous training period is determined as the preset device allocation ratio for each device set. The preset historical time period can be 24 hours, 12 hours, etc., and can be set according to specific circumstances. For example, if the current training period is k, the preset device allocation ratios for each device set are recorded as H_{k-1}.

[0091] 302. Obtain the required number of devices for the current training cycle.

[0092] The number of devices required for the current training cycle is set manually. For example, if the required number of devices for the current training cycle is 100, then 100 electronic devices are needed for training in the current training cycle. Specifically, let n_k be the number of devices required for the current training cycle k.

[0093] 303. Determine the preset number of requests for each equipment set based on the required number of equipment, the preset redundancy ratio, and the preset equipment allocation ratio for each equipment set.

[0094] The preset redundancy ratio p can be 0.1 or 0.2, etc., and can be set manually according to the specific situation. The required number of devices n_k is allocated according to the preset device allocation ratio H_{k-1} of each device set to obtain the allocated number of devices n^i_k for each device set. The allocated number of devices n^i_k for each device set is increased by the preset redundancy ratio p to obtain the preset number of requests m^i_k = (1+p)*n^i_k for each device set.

[0095] 304. Obtain the preset number of requests corresponding to the target device set.

[0096] After obtaining the preset number of requests corresponding to the target device set, determine whether the total number of training requests that the electronic devices in the target device set have passed in the current training period exceeds the preset number of requests corresponding to the target device set.

[0097] Furthermore, if the total number of requests does not exceed the preset number of requests corresponding to the target device set, it indicates that there are relatively few devices in the device set that have passed the request, and training can continue using the electronic devices in the device set. In this case, an approval instruction message for the target training request is issued, which instructs the target electronic device to train the current federated global model. If the total number of requests exceeds the preset number, it indicates that there are relatively many devices in the device set that have passed the request, and training cannot continue using the electronic devices in the device set. In this case, a rejection instruction message for the target training request is issued, which prohibits the target electronic device from training the current federated global model.

[0098] Furthermore, at the end of the current training period, the percentage of devices that have returned the client model in each device set during the current training period is determined as the preset device allocation ratio for each device set in the next training period.

[0099] For further details, please refer to Figure 4 , Figure 4 This is a flowchart illustrating another embodiment of the training method for the federated global model provided in this application, as shown below. Figure 4 As shown, the training method of the federated global model provided in this application is applied to a server. The process of the training method of the federated global model provided in this application is as follows:

[0100] 401. When a target training request initiated by a target electronic device is obtained, the target device set to which the target electronic device belongs is determined according to the preset device set mapping relationship.

[0101] The device set mapping relationship includes the mapping relationship between each electronic device and each device set.

[0102] In this embodiment, the mapping relationship between each electronic device and each device set can be preset. For example, electronic devices located in the same time zone can be placed in the same device set. For example, there are 24 device sets according to the time zone, namely device set 1, device set 2, ..., device set 24. Each device set includes at least one electronic device.

[0103] In one specific embodiment, when a target training request initiated by a target electronic device is obtained, the location of the target electronic device is obtained, the time zone of the target electronic device is determined based on the location of the target electronic device, and the target device set to which the target electronic device belongs is determined based on the time zone of the target electronic device.

[0104] In this embodiment, at the start of the current training cycle, each electronic device sends a training request to the server. The server then identifies each electronic device as the target electronic device based on the order in which the training requests are received. A training cycle is one iterative training iteration, and the duration of a training cycle can be set according to specific circumstances.

[0105] 402. Obtain the preset device allocation ratio for each device set.

[0106] To ensure reasonable allocation, if the current training period is the first round, the number of training requests initiated by electronic devices in each device set within a preset historical time period is obtained, and the percentage of training requests initiated by electronic devices in each device set is determined as the preset device allocation ratio for each device set. If the current training period is not the first round, the number of devices in each device set that have returned client models in the previous training period is obtained, and the percentage of devices in each device set that have returned client models in the previous training period is determined as the preset device allocation ratio for each device set. The preset historical time period can be 24 hours, 12 hours, etc., and can be set according to specific circumstances. For example, if the current training period is k, the preset device allocation ratios for each device set are recorded as H_{k-1}.

[0107] 403. Obtain the required number of devices for the current training cycle.

[0108] The number of devices required for the current training cycle is set manually. For example, if the required number of devices for the current training cycle is 100, then 100 electronic devices are needed for training in the current training cycle. Specifically, let n_k be the number of devices required for the current training cycle k.

[0109] 404. Determine the preset number of requests for each equipment set based on the required number of equipment, the preset redundancy ratio, and the preset equipment allocation ratio for each equipment set.

[0110] The preset redundancy ratio p can be 0.1 or 0.2, etc., and can be set manually according to the specific situation. The required number of devices n_k is allocated according to the preset device allocation ratio H_{k-1} of each device set to obtain the allocated number of devices n^i_k for each device set. The allocated number of devices n^i_k for each device set is increased by the preset redundancy ratio p to obtain the preset number of requests m^i_k = (1+p)*n^i_k for each device set.

[0111] 405. Obtain the preset number of requests corresponding to the target device set.

[0112] After obtaining the preset number of requests corresponding to the target device set, determine whether the total number of training requests that the electronic devices in the target device set have passed in the current training period exceeds the preset number of requests corresponding to the target device set.

[0113] Furthermore, if the total number of requests does not exceed the preset number of requests corresponding to the target device set, it indicates that there are relatively few devices in the device set that have passed the request, and training can continue using the electronic devices in the device set. In this case, an approval instruction message for the target training request is issued, which instructs the target electronic device to train the current federated global model. If the total number of requests exceeds the preset number, it indicates that there are relatively many devices in the device set that have passed the request, and training cannot continue using the electronic devices in the device set. In this case, a rejection instruction message for the target training request is issued, which prohibits the target electronic device from training the current federated global model.

[0114] Furthermore, at the end of the current training period, the percentage of devices that have returned the client model in each device set during the current training period is determined as the preset device allocation ratio for each device set in the next training period.

[0115] 406. Determine whether the total number of training requests that electronic devices in the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set.

[0116] In this embodiment of the application, at the start of the current training cycle, the server initializes the total number of training requests that have passed to a preset initial value, for example, the preset initial value is zero.

[0117] Specifically, when a target training request initiated by a target electronic device is received, it is determined whether the target electronic device is the first electronic device to initiate a training request in the current training cycle. If the target electronic device is the first electronic device to initiate a training request in the current training cycle, the total number of training requests that have been passed is determined by incrementing a preset initial value by one. If the target electronic device is not the first electronic device to initiate a training request in the current training cycle, the total number of training requests that have been passed is obtained, and it is determined whether the total number of training requests that have been passed by electronic devices in the target device set in the current training cycle exceeds the preset number of requests corresponding to the target device set.

[0118] The preset number of requests corresponding to the target device set can be set according to specific circumstances. For example, the preset number of requests corresponding to the target device set can be 100 or other values. The preset number of requests corresponding to different target device sets can be different or the same.

[0119] 407. Issue a pass instruction message to request training to the target.

[0120] The instruction information is used to instruct the target electronic device to train the current federal global model.

[0121] If the total number of requests does not exceed the preset number of requests corresponding to the target device set, it means that there are relatively few devices in the device set that have passed the request, and the electronic devices in the device set can continue to be used for training. In this case, an approval instruction message is issued to allow the target training request, which instructs the target electronic device to train the current federated global model. If the total number of requests exceeds the preset number, it means that there are relatively many devices in the device set that have passed the request, and the electronic devices in the device set cannot continue to be used for training. In this case, a rejection instruction message is issued to reject the target training request, which prohibits the target electronic device from training the current federated global model.

[0122] 408. Obtain the client model parameters returned by each electronic device after training the current federal global model.

[0123] Specifically, after issuing the instruction message to request training through the target, the current federated global model is sent to the target electronic device. After the target electronic device trains the current federated global model, it obtains the client model parameters and returns the client model parameters to the server. The server then obtains the client model parameters returned by each electronic device after training the current federated global model.

[0124] 409. Obtain the initial return time parameters of each client model parameter in the current training cycle.

[0125] The initial return time parameter indicates how fast the client model parameters are returned. The larger the initial return time parameter is, the slower the client model parameters are returned, and the smaller the initial return time parameter is, the faster the client model parameters are returned.

[0126] In one specific embodiment, the model training cycle in which the client model parameters begin training and the number of model aggregations h that the server has currently completed are obtained. The difference between the number of model aggregations h that the server has currently completed and the number of model aggregations h that the client model parameters began training in is determined as the initial return time parameter. Each time the server aggregates a model, the model aggregation count increases by one. For example, if the number of model aggregations is h, at the start of the first round of training, the number of model aggregations h is 0. If the current training cycle is the k-th training cycle, and the target electronic device starts training in the (k-1)-th training cycle and returns the client model parameters in the current training cycle, then the number of model aggregations the server has currently completed is h. Therefore, the initial return time parameter d is h+1-k.

[0127] 410. Obtain the velocity parameter influence factor of each client model parameter in the current training cycle.

[0128] Among them, the velocity parameter influence factor 'a' decreases as the training cycle increases. Specifically, during the first training cycle on the server, the velocity parameter influence factor 'a' is set to the initial value A that influences the shadow.

[0129] In a specific embodiment, a preset annealing step size *s* and a preset number of training iterations are obtained. The product of the preset annealing step size *s* and the preset number of training iterations is determined as the initial value *A* of the influence shadow. The initial value *A* of the influence shadow is decreased round by round according to the preset annealing step size *s* to obtain the velocity parameter influence factor *a* of each client model parameter in the current training cycle. Here, the preset annealing step size *s* is a positive number and can be 1, 2, etc., and the preset number of training iterations can be 1000, 100, etc. The preset annealing step size and the preset number of training iterations can be set according to specific circumstances. *a* = *A* - *ks*.

[0130] 411. Based on the velocity parameter influence factor, the initial return time parameters of each client model parameter are proportionally corrected to obtain the corrected return time parameters of each client model parameter.

[0131] Specifically, the velocity parameter influence factor is multiplied by the velocity parameter influence factors of each client-side model parameter to obtain the corrected velocity parameter influence factor for each client-side model parameter. For example, if the velocity parameter influence factor is 'a' and the initial return time parameter is 'd', the corrected velocity parameter influence factor is a*d.

[0132] 412. Determine the aggregate weight coefficient of each client model parameter based on the correction return time parameter of each client model parameter in the current training cycle.

[0133] In this context, a larger adjustment rate factor indicates a slower return time, and consequently, a smaller aggregate weight coefficient for the corresponding client-side model parameters. For example, the aggregate weight coefficient might be (1-1 / (1+e^(-a*d))), where e is the natural constant, a*d is the adjusted return time parameter, a is the adjustment rate factor, and d is the initial return time parameter. When the training period is short, the adjustment rate factor a is larger. In this case, the difference in the initial return time parameter d between different client-side model parameters is amplified by the adjustment rate factor a, resulting in a larger difference in the aggregate weight coefficients between the various client-side model parameters. This allows model parameters with slower returns to be discarded. Conversely, when the training period is long, the adjustment rate factor a is smaller. In this case, the difference in the initial return time parameter d between different client-side model parameters is reduced by the adjustment rate factor a, resulting in a smaller difference in the aggregate weight coefficients between the various client-side model parameters. This allows all models to be utilized, improving the convergence speed.

[0134] 413. Aggregate the parameters of at least two client models based on the aggregated weight coefficients of each client model to obtain the updated federated global model.

[0135] For further details, please refer to Figure 5 , Figure 5 This is a flowchart illustrating yet another embodiment of the training method for the federated global model provided in this application, as shown below. Figure 5 As shown, the training method of the federated global model provided in this application is applied to electronic devices. The process of the training method of the federated global model provided in this application is as follows:

[0136] 501. Send a target training request to the server.

[0137] In one specific embodiment, a target training request is sent to the server according to a preset request period. This preset request period can be 10 seconds, 15 seconds, etc., and can be set according to specific circumstances.

[0138] In another specific embodiment, the device determines whether it meets preset training conditions according to a preset request period; if it meets the preset training conditions, it initiates a target training request. Specifically, the preset training conditions may include at least one of the following: the electronic device is charging; the electronic device is connected to a network; and the local time of the device is within a preset time interval. For example, the preset time interval is from 0:00 to 2:00.

[0139] Furthermore, the system determines whether it meets the preset training conditions according to a preset request cycle. If it meets the preset training conditions, it obtains the training task list on the server and sends a target training request to the server for the highest priority training task on the training task list. The server stores at least two different current federated global models. The server publishes different training tasks for each different current federated global model, forming a training task list. The electronic device initiates a target training request for one of the at least two different current federated global models.

[0140] 502. When the server returns a pass indication, retrieve the current federated global model on the server.

[0141] Specifically, when the server receives a target training request initiated by the electronic device, it determines the target device set to which the electronic device belongs based on a preset device set mapping relationship. It then issues the request when the total number of training requests that each electronic device in the target device set has passed in the current training cycle does not exceed the preset number of requests corresponding to the target device set. The server then performs local training on the current federated global model to obtain client model parameters.

[0142] 503. Train the current federated global model locally to obtain the client model parameters.

[0143] Specifically, the current federated global model is trained locally using local data on the electronic device to obtain the client model parameters.

[0144] 504. Return the client model parameters to the server.

[0145] For further details, please refer to Figure 6 , Figure 6 This is a flowchart illustrating yet another embodiment of the training method for the federated global model provided in this application, as shown below. Figure 6 As shown, the training process of the federated global model provided in this application is as follows:

[0146] 601. Send a target training request to the server.

[0147] 602. When a target training request initiated by the target electronic device is obtained, the target device set to which the target electronic device belongs is determined according to the preset device set mapping relationship.

[0148] 603. Determine whether the total number of requests exceeds the preset number of requests corresponding to the target device set.

[0149] Specifically, it determines whether the total number of training requests that electronic devices within the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set.

[0150] 604. Issue a pass instruction message to request training to the target.

[0151] Specifically, if the total number of requests does not exceed the preset number of requests corresponding to the target device set, is the total number of requests exceeded the preset number of requests corresponding to the target device set?

[0152] 605. Obtain the current federated global model on the server.

[0153] When the server returns a pass indication, retrieve the current federated global model on the server.

[0154] 606. Train the current global federation model locally to obtain the client model parameters.

[0155] 607. Return the client model parameters to the server.

[0156] 608. Obtain the client model parameters returned by each electronic device after training the current federal global model.

[0157] 609. Aggregate at least two client model parameters to obtain the updated federated global model.

[0158] To facilitate better implementation of the training method for the federated global model provided in this application, this application also provides a training apparatus for the federated global model based on the above-described training method. The meanings of the terms used are the same as in the above-described training method for the federated global model; for specific implementation details, please refer to the descriptions in the above method embodiments.

[0159] Please refer to Figure 7 , Figure 7 This is a schematic diagram of an embodiment of a training device for a federated global model provided in this application. The training device for the federated global model may include a determining module 701, a judging module 702, a sending module 703, and an aggregation module 704, wherein...

[0160] The determination module is used to determine the target device set to which the target electronic device belongs based on a preset device set mapping relationship when a target training request initiated by the target electronic device is received. The device set mapping relationship includes the mapping relationship between each electronic device and each device set.

[0161] The judgment module is used to determine whether the total number of training requests that electronic devices in the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set.

[0162] The sending module is used to send a pass instruction message for the target training request if the total number of requests does not exceed the preset number of requests corresponding to the target device set. The pass instruction message is used to instruct the target electronic device to train the current federated global model.

[0163] To facilitate better implementation of the training method for the federated global model provided in this application, this application also provides a training apparatus for the federated global model based on the above-described training method. The meanings of the terms used are the same as in the above-described training method for the federated global model; for specific implementation details, please refer to the descriptions in the above method embodiments.

[0164] Please refer to Figure 8 , Figure 8 This is a schematic diagram of another embodiment of the training device for the federated global model provided in this application. The training device for the federated global model may include a sending module 801, an acquisition module 802, a training module 803, and a return module 804, wherein...

[0165] The sending module is used to send target training requests to the server;

[0166] The acquisition module is used to acquire the current federated global model on the server when the server returns the pass indication information. The pass indication information is obtained by the server when it receives the target training request initiated by the electronic device, determines the target device set to which the electronic device belongs according to the preset device set mapping relationship, and determines that the total number of training requests that each electronic device in the target device set has passed in the current training period does not exceed the preset number of requests corresponding to the target device set.

[0167] The training module is used to train the current federated global model locally to obtain client model parameters;

[0168] The return module is used to return the client model parameters to the server.

[0169] This application also provides an electronic device, including a memory and a processor, wherein the processor executes steps in the training method of the federated global model provided in this embodiment by calling a computer program stored in the memory.

[0170] This application also provides a server, including a memory and a processor, wherein the processor executes steps in the training method of the federated global model provided in this embodiment by calling a computer program stored in the memory.

[0171] Please refer to Figure 9 , Figure 9This is a schematic diagram of the structure of the electronic device / server provided in the embodiments of this application. Figure 9 The electronic device mentioned can be either the electronic device of this application or the server of this application.

[0172] The electronic device may include components such as a processor 101 with one or more processing cores, a memory 102 with one or more computer-readable storage media, a power supply 103, and an input unit 104. Those skilled in the art will understand that the electronic device structure shown in the figures does not constitute a limitation on the electronic device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0173] The processor 101 is the control center of the electronic device, connecting various parts of the device via various interfaces and lines. It executes software programs and / or modules stored in the memory 102, and calls data stored in the memory 102, to perform various functions and process data. Optionally, the processor 101 may include one or more processing cores; alternatively, the processor 101 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 101.

[0174] The memory 102 can be used to store software programs and modules. The processor 101 executes various functional applications and data processing by running the software programs and modules stored in the memory 102. The memory 102 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 102 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 102 may also include a memory controller to provide the processor 101 with access to the memory 102.

[0175] The electronic device also includes a power supply 103 that supplies power to the various components. Optionally, the power supply 103 can be logically connected to the processor 101 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 103 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0176] The electronic device may also include an input unit 104, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0177] Although not shown, the electronic device may also include a display unit, an image acquisition component, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 101 in the electronic device loads the executable code corresponding to one or more computer programs into the memory 102 according to the following instructions, and the processor 101 executes the steps in the training method of the federated global model provided in this application, such as:

[0178] When a target training request initiated by a target electronic device is received, the target device set to which the target electronic device belongs is determined according to a preset device set mapping relationship, wherein the device set mapping relationship includes the mapping relationship between each electronic device and each device set;

[0179] Determine whether the total number of training requests that electronic devices in the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set;

[0180] If the total number of requests does not exceed the preset number of requests corresponding to the target device set, then a pass instruction information is issued to pass the target training request. The pass instruction information is used to instruct the target electronic device to train the current federated global model.

[0181] or,

[0182] Send a target training request to the server;

[0183] When the server returns the pass indication information, the current federated global model on the server is obtained. The pass indication information is obtained by the server when it receives the target training request initiated by the electronic device, and the server determines the target device set to which the electronic device belongs according to the preset device set mapping relationship. The server issues the pass indication information when it determines that the total number of training requests that each electronic device in the target device set has passed in the current training cycle does not exceed the preset number of requests corresponding to the target device set.

[0184] The current federated global model is trained locally to obtain the client model parameters;

[0185] The client model parameters are returned to the server.

[0186] It should be noted that the electronic device provided in this application embodiment and the training method of the federated global model in the above embodiment belong to the same concept. The specific implementation process can be found in the above related embodiments, and will not be repeated here.

[0187] This application also provides a computer-readable storage medium storing a computer program thereon. When the computer program stored thereon is executed on the processor of the electronic device provided in the embodiments of this application, the processor of the electronic device performs the steps in the training method of the federated global model provided in this application. The storage medium may be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0188] This application also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform various optional implementations of the training method for the aforementioned federated global model.

[0189] The above provides a detailed description of the training method and apparatus for a federated global model provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0190] It should be noted that when the above embodiments of this application are applied to specific products or technologies, and user-related data is involved, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

Claims

1. A training method for a federated global model, characterized in that, Applied to servers, including: When a target training request initiated by a target electronic device is received, the target device set to which the target electronic device belongs is determined according to a preset device set mapping relationship. The device set mapping relationship includes the mapping relationship between each electronic device and each device set. If the current training cycle is not the first round, the proportion of the number of devices that have returned the client model in each device set in the previous training cycle is determined as the preset device allocation ratio of each device set. The required number of devices in the current training cycle is obtained. Based on the required number of devices, the preset redundancy ratio, and the preset device allocation ratio of each device set, the preset number of requests corresponding to each device set is determined. One training cycle is one iteration of training. Determine whether the total number of training requests that electronic devices in the target device set have passed during the current training period exceeds the preset number of requests corresponding to the target device set; If the total number of requests does not exceed the preset number of requests corresponding to the target device set, a pass indication message for the target training request is issued. The pass indication message is used to instruct the target electronic device to train the current federated global model, obtain the client model parameters returned by each electronic device after training the current federated global model, obtain the aggregate weight coefficient of each client model parameter in the current training cycle, wherein the aggregate weight coefficient is (1-1 / (1+e^(-a*d))), where e is a natural constant, a is a speed parameter influence factor, and d is an initial return time parameter, wherein the speed parameter influence factor decreases as the training cycle increases, and at least two client model parameters are aggregated based on the aggregate weight coefficient of each client model parameter to obtain the updated federated global model.

2. A training method for a federated global model, characterized in that, Applied to electronic devices, including: Send a target training request to the server; When the server returns the pass indication information, the current federated global model on the server is obtained. The pass indication information is obtained by the server when it receives the target training request initiated by the electronic device, and the server determines the target device set to which the electronic device belongs according to the preset device set mapping relationship. The server issues the pass indication information when it determines that the total number of training requests that each electronic device in the target device set has passed in the current training cycle does not exceed the preset number of requests corresponding to the target device set. The current federated global model is trained locally to obtain the client model parameters; The client model parameters are returned to the server. The server obtains the client model parameters returned by each electronic device after training the current federated global model. The server obtains the aggregated weight coefficient of each client model parameter in the current training cycle, where the aggregated weight coefficient is (1-1 / (1+e^(-a*d))), where e is a natural constant, a is a velocity parameter influence factor, and d is an initial return time parameter. The velocity parameter influence factor decreases as the training cycle increases. At least two client model parameters are aggregated based on the aggregated weight coefficient of each client model parameter to obtain the updated federated global model.

3. The training method for the federated global model according to claim 2, characterized in that, Sending the target training request to the server includes: Determine whether it meets the preset training conditions according to the preset request cycle; If it meets the preset training conditions, it will initiate a target training request.

4. A training device for a federated global model, characterized in that, The training apparatus for the federated global model includes: The determination module is used to determine the target device set to which the target electronic device belongs based on a preset device set mapping relationship when a target training request initiated by the target electronic device is obtained. The device set mapping relationship includes the mapping relationship between each electronic device and each device set. The judgment module is used to determine whether the total number of training requests that electronic devices in the target device set have passed in the current training period exceeds the preset number of requests corresponding to the target device set. The sending module is configured to, if the total number of requests does not exceed the preset number of requests corresponding to the target device set, send a pass indication information for the target training request. The pass indication information is used to instruct the target electronic device to train the current federated global model, obtain the client model parameters returned by each electronic device after training the current federated global model, obtain the aggregated weight coefficient of each client model parameter in the current training cycle, wherein the aggregated weight coefficient is (1-1 / (1+e^(-a*d))), where e is a natural constant, a is a speed parameter influence factor, and d is an initial return time parameter, wherein the speed parameter influence factor decreases as the training cycle increases, and aggregate at least two client model parameters based on the aggregated weight coefficient of each client model parameter to obtain an updated federated global model.

5. A training device for a federated global model, characterized in that, include: The sending module is used to send target training requests to the server; The acquisition module is used to acquire the current federated global model on the server when it receives the pass indication information returned by the server. The pass indication information is obtained by the server when it receives the target training request initiated by the electronic device, based on a preset device set mapping relationship to determine the target device set to which the electronic device belongs. If the current training cycle is not the first round, the percentage of devices that have returned client models in each device set in the previous training cycle is determined as the preset device allocation ratio of each device set. The module also acquires the number of devices required in the current training cycle and determines the preset number of requests corresponding to each device set based on the number of devices required, the preset redundancy ratio, and the preset device allocation ratio of each device set. One training cycle is one iteration of training. The module issues a request when the total number of training requests that have been passed by each electronic device in the target device set in the current training cycle does not exceed the preset number of requests corresponding to the target device set. The training module is used to train the current federated global model locally to obtain client model parameters; The return module is used to return the client model parameters to the server. The server obtains the client model parameters returned by each electronic device after training the current federated global model. It obtains the aggregated weight coefficient of each client model parameter in the current training cycle, where the aggregated weight coefficient is (1-1 / (1+e^(-a*d))), where e is a natural constant, a is a velocity parameter influence factor, and d is an initial return time parameter. The velocity parameter influence factor decreases as the training cycle increases. Based on the aggregated weight coefficient of each client model parameter, at least two client model parameters are aggregated to obtain the updated federated global model.

6. A server, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor running the computer program in the memory to perform the steps in the training method of the federated global model of claim 1.

7. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor running the computer program in the memory to perform the steps in the training method of the federated global model according to any one of claims 2 to 3.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform steps in the training method of the federated global model of claim 1 or any one of claims 2 to 3.

9. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the steps in the training method of the federated global model according to claim 1 or the steps in the training method of the federated global model according to any one of claims 2 to 3.

Images