Method, apparatus, device, medium and product for calculating quality information of edge device

Abstract

The present application relates to a method, apparatus, device, medium and program product for calculating quality information of edge devices. The method is applied to edge devices in a federated learning system, and the federated learning system includes multiple edge devices and an edge server that has established communication connections with each edge device. First, data state information for characterizing the importance of the local model update information deployed in the edge device is calculated. Then, channel state information for characterizing the channel conditions of the edge device is calculated. Next, transmission energy information for characterizing the transmission energy of the edge device is calculated. Finally, quality information is determined based on the data state information, channel state information, and transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server. Using this method can improve the accuracy of calculating the quality information of edge devices.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular, to a method, device, equipment, medium, and product for calculating quality information of edge devices. Background Art

[0002] Federated learning is a new machine learning paradigm. A federated learning system generally includes multiple edge devices and an edge server. The edge server can perform distributed model training among multiple edge devices with local data. In this way, during the training process of the global model on the edge server, it is not necessary to exchange local individuals or sample data on the edge devices, but only to exchange model parameters or gradients of the global model and local models, thereby enabling data security and privacy protection.

[0003] In order to improve the efficiency of global model training, federated learning can select appropriate edge devices for global model training according to the quality of edge devices during the training of the global model. However, the current method for calculating the quality of edge devices is not accurate enough. Summary of the Invention

[0004] Based on this, it is necessary to provide a method, device, equipment, medium, and product for calculating quality information of edge devices in view of the above technical problems.

[0005] In a first aspect, this application provides a method for calculating quality information of edge devices, which is applied to edge devices in a federated learning system. The federated learning system includes multiple edge devices and an edge server that has established a communication connection with each edge device. The method includes: calculating data status information, where the data status information is used to characterize the importance of the local model update information deployed in the edge device; calculating channel status information, where the channel status information is used to characterize the channel conditions of the edge device; calculating transmission energy information, where the transmission energy information is used to characterize the transmission energy of the edge device; and determining quality information according to the data status information, channel status information, and transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server.

[0006] In one embodiment, calculating the data status information includes: obtaining the gradient information of the local model deployed in the edge device; and calculating the data status information according to the gradient information.

[0007] In one embodiment, calculating the channel status information includes: obtaining the quasi-static channel gain between the edge device and the edge server; and calculating the channel status information according to the quasi-static channel gain.

[0008] In one embodiment, calculating transmission energy information includes: obtaining the quasi-static channel gain between the edge device and the edge server and a preset signal-to-noise ratio threshold at the receiving end; calculating the transmission energy information according to the quasi-static channel gain and the preset signal-to-noise ratio threshold at the receiving end.

[0009] In one embodiment, obtaining the gradient information of the local model deployed in the edge device includes: determining whether the edge device received an instruction sent by the edge server during the previous round of global model training; if it is determined that the edge device received an instruction sent by the edge server during the previous round of global model training, using the gradient of the local model deployed in the edge device calculated in this round as the gradient information; if it is determined that the edge device did not receive an instruction sent by the edge server during the previous round of global model training, using the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradients not sent as the gradient information.

[0010] In one embodiment, the method further includes: receiving the current global model parameters of the global model deployed in the edge server; performing this round of training on the local model deployed in the edge device according to the current global model parameters to obtain the gradient information of the local model deployed in the edge device.

[0011] In a second aspect, the present application further provides a quality information calculation device for an edge device, which is applied to an edge device in a federated learning system. The federated learning system includes multiple edge devices and an edge server that has established a communication connection with each edge device. The device includes:

[0012] A first calculation module, configured to calculate data status information, where the data status information is used to characterize the importance of the local model update information deployed in the edge device;

[0013] A second calculation module, configured to calculate channel status information, where the channel status information is used to characterize the channel condition of the edge device;

[0014] A third calculation module, configured to calculate transmission energy information, where the transmission energy information is used to characterize the transmission energy of the edge device;

[0015] A determination module, configured to determine quality information according to the data status information, the channel status information, and the transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the global model training deployed in the edge server.

[0016] In one embodiment, the first calculation module is specifically configured to obtain the gradient information of the local model deployed in the edge device; calculate the data status information according to the gradient information.

[0017] In one embodiment, the second calculation module is specifically configured to obtain the quasi-static channel gain between the edge device and the edge server; and calculate the channel state information according to the quasi-static channel gain.

[0018] In one embodiment, the third calculation module is specifically configured to obtain the quasi-static channel gain between the edge device and the edge server and the preset receiver signal-to-noise ratio threshold; and calculate the transmission energy information according to the quasi-static channel gain and the preset receiver signal-to-noise ratio threshold.

[0019] In one embodiment, the first calculation module is specifically configured to determine whether the edge device received an instruction sent by the edge server during the previous round of global model training; if it is determined that the edge device received an instruction sent by the edge server during the previous round of global model training, then use the gradient of the local model deployed in the edge device calculated in this round as the gradient information; if it is determined that the edge device did not receive an instruction sent by the edge server during the previous round of global model training, then use the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradient that has not been sent as the gradient information.

[0020] In one embodiment, it further includes a receiving module, which is configured to receive the current global model parameters of the global model deployed in the edge server; and perform training on the local model deployed in the edge device in this round according to the current global model parameters to obtain the gradient information of the local model deployed in the edge device.

[0021] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the method for calculating the quality information of the edge device according to any one of the above first aspects.

[0022] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the method for calculating the quality information of the edge device according to any one of the above first aspects or second aspects.

[0023] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, it implements the method for calculating the quality information of the edge device according to any one of the above first aspects or second aspects.

[0024] The above method, device, equipment, medium and product for calculating the quality information of edge devices are applied to edge devices in a federated learning system. The federated learning system includes multiple edge devices and an edge server that has established communication connections with each edge device. First, data state information is calculated to characterize the importance of the local model update information deployed in the edge device. Then, channel state information is calculated to characterize the channel conditions of the edge device. Next, transmission energy information is calculated to characterize the transmission energy of the edge device. Finally, quality information is determined based on the data state information, channel state information, and transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server. In this way, by comprehensively considering the data state information, channel state information, and transmission energy information of the edge device to determine the quality information of the edge device, it is possible to select edge devices with high communication efficiency, high precision, and high energy efficiency. At the same time, the quality information of the edge device is quantified, and the calculation of the quality information of the edge device is more accurate, thus facilitating the selection of appropriate numbers and device qualities of edge devices for training during the training process of the global model in the edge server. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a diagram of the application environment of the method for calculating the quality information of edge devices in an embodiment;

[0026] Figure 2 It is a schematic flowchart of the method for calculating the quality information of edge devices in an embodiment;

[0027] Figure 3 It is a schematic flowchart of the method for calculating the quality information of edge devices in another embodiment;

[0028] Figure 4 It is a schematic flowchart of the method for calculating the quality information of edge devices in another embodiment;

[0029] Figure 5 It is a schematic flowchart of the method for calculating the quality information of edge devices in another embodiment;

[0030] Figure 6 It is a schematic flowchart of the method for calculating the quality information of edge devices in another embodiment;

[0031] Figure 7 It is a schematic flowchart of the method for calculating the quality information of edge devices in another embodiment;

[0032] Figure 8 It is a block diagram of the structure of the model training device in another embodiment;

[0033] Figure 9Internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0034] To make the objectives, technical solutions and advantages of this application clearer, the following further elaborates on this application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely used to explain this application and are not used to limit this application.

[0035] The method for calculating the quality information of an edge device provided in an embodiment of this application can be applied to, for example Figure 1 the federated learning system shown as follows. The federated learning system may include an edge server 101 and multiple edge devices 102, and a communication connection is established between the edge server 101 and the edge devices 102. Among them, the edge server 101 can be implemented by an independent server or a server cluster composed of multiple servers. The edge devices 102 can be, but are not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc.

[0036] A global model is deployed on the edge server 101, and a local model and a local data set are deployed on the edge devices 102.

[0037] In federated learning, during the training process of the global model deployed in the edge server, the gradient information of the local model deployed in the edge device is required. To reduce the communication delay and resource overhead during the transmission process, only a certain number of edge devices are selected to participate in the training of the global model in each round of training of the global model. At the same time, since the contribution degrees of the gradient information of the local models in different edge devices to improving the accuracy and convergence of the global model in the edge server are different, therefore, the edge server needs to receive the quality information sent by multiple edge devices, and select the edge devices used for training the global model of the edge server through the quality information.

[0038] In one embodiment, as Figure 2 shown, a method for calculating the quality information of an edge device is provided. Taking the edge device in Figure 1 as an example for illustration, taking the federated learning system shown as Figure 1 in this application as an example, the federated learning system consists of an edge server and N distributed single-antenna edge devices. where the data set collected or generated locally by each edge device n is D n . One update of the global model on the edge server can be defined as a communication round, that is, one training process. Taking as the index, where T is the total number of communication rounds for global model training. Among them, in one training of the global model, the calculation method of the quality information of each edge device includes the following steps:

[0039] Step 201, the edge device calculates the data state information.

[0040] Among them, the data state information is used to characterize the importance of the update information of the local model deployed in the edge device. The data state information (Data State Information, DSI) is represented by and is measured by the l2-norm of the update information of the local model in the edge device. The edge device with a larger DSI value makes a greater contribution to the global model training, that is, the importance of the update information of the local model is higher.

[0041] Optionally, the data state information can be calculated based on the gradient information of the local model deployed in the edge device. The gradient information of the local model is also the update information of the local model.

[0042] Step 202, the edge device calculates the channel state information.

[0043] The channel state information is used to characterize the channel condition of the edge device. The channel state information (Channel State Information, CSI) is represented by and is measured by |h n,t |, where is the quasi-static channel gain between the edge device n and the edge server in the t-th round, represents the set of complex numbers. The larger the CSI value, the better the channel condition.

[0044] Optionally, the channel state information is calculated based on the quasi-static channel gain between the edge device and the edge server.

[0045] Step 203, the edge device calculates the transmission energy information.

[0046] The transmission energy information is used to characterize the transmission energy of the edge device. In order to pursue high energy efficiency, the edge device selection mechanism also considers the transmission energy consumption levels of different devices.

[0047] Optionally, the transmission energy information is calculated based on the quasi-static channel gain between the edge device and the edge server and the preset receiver signal-to-noise ratio threshold.

[0048] Step 204, the edge device determines the quality information according to the data state information, the channel state information, and the transmission energy information.

[0049] Among them, the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server. By determining the quality information, an appropriate number of edge devices with high quality information, that is, edge devices with a high contribution degree of the gradient information of the local model in the edge device to the global model training, can be selected to participate in the training of the global model of the edge server. The quality information comprehensively considers the importance of the update information of the local model deployed on the edge device, the channel condition, and the energy consumption level. Optionally, different weight parameters can be used to balance the proportions of the data state information, the channel state information, and the transmission energy information in the calculation of the quality information.

[0050] In the above embodiment, it is applied to an edge device in a federated learning system. The federated learning system includes multiple edge devices and an edge server that has established communication connections with each edge device. First, calculate the data state information used to characterize the importance of the update information of the local model deployed in the edge device. Then, calculate the channel state information used to characterize the channel condition of the edge device. Next, calculate the transmission energy information used to characterize the transmission energy of the edge device. Finally, determine the quality information according to the data state information, the channel state information, and the transmission energy information. Among them, the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server. In this way, by comprehensively considering the data state information, the channel state information, and the transmission energy information of the edge device to determine the quality information of the edge device, edge devices with high communication efficiency, high precision, and high energy efficiency are selected, and at the same time, the quality information of the edge device is quantified, and the calculation of the quality information of the edge device is more accurate, so as to facilitate selecting appropriate numbers and device qualities of edge devices for training according to the quality information sent by the edge device during the training process of the global model of the edge server.

[0051] In one embodiment, as Figure 3 shown, when the edge device performs local model update, it needs the current global model parameters of the edge server, which specifically includes the following steps.

[0052] Step 301, the edge device receives the current global model parameters of the global model deployed in the edge server.

[0053] At the beginning of each round of training of the global model, the edge server sends the current global model parameter w t-1 to all edge devices by broadcasting, and the edge device performs local model training according to the global model parameter.

[0054] Step 302, the edge device performs this round of training on the local model deployed in the edge device according to the current global model parameter, and obtains the gradient information of the local model deployed in the edge device.

[0055] Among them, define l(w; s i , q i ) as the loss function of the data sample (s i , q i ). Then, for the edge device n on the local dataset D n , the local loss function is as follows:

[0056]

[0057] Among them, |D n | represents the size of the dataset D n .

[0058] After receiving the global model parameters w t-1 , each edge device calculates the gradient of the local model by running the mini-batch stochastic gradient descent algorithm on the local mini-batch , that is:

[0059]

[0060] In one embodiment, a dynamic cumulative residual feedback mechanism is proposed. When an edge device is not selected during this round of training, the edge device saves the gradient information of the local model of this round of training locally, and then transmits the accumulated local gradients that have not been sent in the past saved locally when the edge device is selected next time. The steps of sending the gradient information of the local model deployed in the edge device to the edge server for training the global model deployed in the edge server include:

[0061] First, the edge device determines whether the edge device received an instruction sent by the edge server during the global model training in the previous round.

[0062] When the edge device determines the gradient information of the local model in the current round, it is necessary to first determine whether the edge device received an instruction sent by the edge server during the global model training in the previous round. Receiving the instruction means that the edge device was selected to send the gradient information of the local model during the global model training in the previous round, and not receiving the instruction means that the edge device was not selected to send the gradient information of the local model during the global model training in the previous round.

[0063] Then, based on the determination result, the edge device obtains the gradient information of the local model deployed in the edge device and sends the gradient information to the edge server.

[0064] Optionally, the calculation of the gradient information of the edge device includes two cases:

[0065] In the first case, if it is determined that the edge device received the instruction sent by the edge server during the previous round of global model training, the gradient of the local model deployed in the edge device calculated in this round is used as the gradient information.

[0066] In the second case, if it is determined that the edge device did not receive the instruction sent by the edge server during the previous round of global model training, the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradient that has not been sent is used as the gradient information.

[0067] For example, define the gradient information of the local model of the selected edge device n in the t-th round as the gradient g of the current communication round n，t and the accumulated residual r n，t combination. Specifically, r n，t is given by the following equations respectively:

[0068]

[0069]

[0070] where 0 ≤ ξ ≤ 1 reflects the importance of the accumulated residual for the current training. S t-1 represents the set of selected edge devices in the (t - 1)-th round. That is, if the edge device n is selected in the (t - 1)-th round, the accumulated residual is 0, and the gradient of the local model deployed in the edge device calculated in this round is used as the gradient information. If the edge device n was not selected in the previous round, the accumulated residual is ξg n，t-1 The sum of the gradient of the local model calculated in this round and the accumulated residual is used as the gradient information.

[0071] In the above embodiments, through this dynamic accumulated residual feedback mechanism, the edge device considers the accumulated local gradients that have not been sent in the past when sending the gradient information of the local model, so as to be able to more comprehensively utilize the dataset of the edge device to improve the accuracy of quality information calculation.

[0072] In the embodiments of the present application, the quality information is determined according to the data status information, the channel status information, and the transmission energy information. Among them, the data status information is used to characterize the importance of the local model update information deployed in the edge device. The data status information is denoted by, measured by the l2-norm of the local model update information. Edge devices with larger DSI values contribute more to the global model training. As Figure 4 shown, the calculation steps include:

[0073] Step 401, obtain the gradient information of the local model deployed in the edge device The calculation of the gradient information is as described in the above embodiments.

[0074] Step 402: Calculate the data status information according to the gradient information.

[0075] To unify the DSI output value range, the present application performs a normalization operation on the metric DSI, and the specific definition is as follows:

[0076]

[0077] where g max is the maximum value of

[0078] Optionally, the channel state information is used to characterize the channel condition of the edge device, and is measured by |h n，t |. The larger the CSI value, the better the channel condition. As Figure 5 shown, the calculation steps include:

[0079] Step 501: Obtain the quasi-static channel gain h n，t .

[0080] Step 502: Calculate the channel state information according to the quasi-static channel gain, and perform a normalization operation on the CSI, as shown in the following formula:

[0081]

[0082] where h max is the maximum value of |h n，t |.

[0083] In one embodiment, as Figure 6 shown, the calculation steps of the transmission energy information include:

[0084] Step 601: Obtain the quasi-static channel gain between the edge device and the edge server and the preset receiver signal-to-noise ratio threshold.

[0085] To effectively resist channel fading, the present invention designs a power control mechanism based on the channel inversion method. In the t-th round, the transmission power P n，t of the edge device n is:

[0086]

[0087] where σ tDenote the power expansion factor, i.e., the denoising factor, to determine the signal-to-noise ratio (SNR) at the server receiver. For the convenience of energy constraint, according to the designed power control mechanism, the SNR at the server receiver in the t-th round can be expressed as:

[0088]

[0089] where x n，t is the gradient information sent by the edge device n in the t-th round, Φ(·) represents the preprocessing and normalization operations to ensure that x n，t has zero mean and variance P n，t ; is the local model gradient information of the edge device n in the t-th round training calculated above. is additive white Gaussian noise with mean 0 and variance , and E[·] is to find the expected value.

[0090] Let γ thr represent the preset SNR threshold, and it is required to satisfy γ s ≥γ thr . Therefore, σ t 2 can be expressed as:

[0091]

[0092] Step 602, calculate the transmission energy information according to the quasi-static channel gain and the preset receiver SNR threshold. As shown in the following formula:

[0093]

[0094] Optionally, the edge device determines the quality information according to the data status information, channel status information, and transmission energy information.

[0095] Considering the importance of the update information of the edge device n and the channel condition, the importance of the edge device can be defined by the following formula:

[0096]

[0097] where the weight parameters ρ1 and ρ2 jointly balance the proportion of DSI and CSI, and ρ1 + ρ2 = 1. Based on the above definition, considering the importance of the update information, channel condition, and energy consumption level, the quality information of the edge device is defined as:

[0098] I n，t = λ V V n，t - λ E E n，t(11)

[0099] Among them, the weight parameter λ E and λ V are used to balance the energy consumption loss and the gain brought by the device importance, and λ E +λ V = 1.

[0100] In the above embodiments, due to the diversity of different edge devices in terms of local datasets, channel conditions, energy states, etc., the contributions of different edge devices to the global model are different. Therefore, the quality information of the edge devices is determined by comprehensively considering the importance of the update information, channel conditions, and energy consumption levels, and the selection of the edge devices is determined through the quality information, so as to achieve high communication efficiency, high precision, and high energy efficiency.

[0101] In the embodiments of the present application, please refer to Figure 7 , which shows a flowchart of a method for calculating the quality information of an edge device provided by the embodiments of the present application. The method for calculating the quality information of the edge device includes the following steps:

[0102] Step 701, the edge device receives the current global model parameters of the global model deployed in the edge server.

[0103] Step 702, the edge device performs this round of training on the local model deployed in the edge device according to the current global model parameters, and obtains the gradient information of the local model deployed in the edge device.

[0104] Step 703, obtain the gradient information of the local model deployed in the edge device.

[0105] Step 704, calculate the data state information according to the gradient information.

[0106] Step 705, obtain the quasi-static channel gain between the edge device and the edge server.

[0107] Step 706, calculate the channel state information according to the quasi-static channel gain.

[0108] Step 707, obtain the quasi-static channel gain and the preset receiver signal-to-noise ratio threshold between the edge device and the edge server.

[0109] Step 708, calculate the transmission energy information according to the quasi-static channel gain and the preset receiver signal-to-noise ratio threshold.

[0110] Step 709, the edge device determines the quality information according to the data state information, channel state information, and transmission energy information.

[0111] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0112] Based on the same inventive concept, an embodiment of the present application further provides a model training device for implementing the above-mentioned edge device quality information calculation method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the following model training devices can refer to the limitations on the edge device quality information calculation method in the above text, and will not be repeated here.

[0113] In one embodiment, as Figure 8 shown, a quality information calculation device 800 for an edge device is provided, which is applied to an edge device in a federated learning system. The federated learning system includes multiple edge devices and an edge server that has established a communication connection with each edge device. The device includes: a first calculation module 801, a second calculation module 802, a third calculation module 803, and a determination module 804, where:

[0114] The first calculation module 801 is used to calculate data status information, and the data status information is used to characterize the importance of the local model update information deployed in the edge device;

[0115] The second calculation module 802 is used to calculate channel status information, and the channel status information is used to characterize the channel conditions of the edge device;

[0116] The third calculation module 803 is used to calculate transmission energy information, and the transmission energy information is used to characterize the transmission energy of the edge device;

[0117] The determination module 804 is used to determine quality information according to the data status information, channel status information, and transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the global model training deployed in the edge server.

[0118] In an alternative embodiment of the present application, the first calculation module 801 is specifically configured to obtain the gradient information of the local model deployed in the edge device; and calculate the data state information according to the gradient information.

[0119] In an alternative embodiment of the present application, the second calculation module 802 is specifically configured to obtain the quasi-static channel gain between the edge device and the edge server; and calculate the channel state information according to the quasi-static channel gain.

[0120] In an alternative embodiment of the present application, the third calculation module 803 is specifically configured to obtain the quasi-static channel gain between the edge device and the edge server and the preset receiver signal-to-noise ratio threshold; and calculate the transmission energy information according to the quasi-static channel gain and the preset receiver signal-to-noise ratio threshold.

[0121] In an alternative embodiment of the present application, the first calculation module 801 is specifically configured to determine whether the edge device received an instruction sent by the edge server during the previous round of global model training; if it is determined that the edge device received an instruction sent by the edge server during the previous round of global model training, then use the gradient of the local model deployed in the edge device calculated in this round as the gradient information; if it is determined that the edge device did not receive an instruction sent by the edge server during the previous round of global model training, then use the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradients that have not been sent as the gradient information.

[0122] In an alternative embodiment of the present application, it further includes a receiving module, which is configured to receive the current global model parameters of the global model deployed in the edge server; and perform training on the local model deployed in the edge device in this round according to the current global model parameters to obtain the gradient information of the local model deployed in the edge device.

[0123] Each module in the above model training device can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor in the computer device in the form of hardware or be independent of it, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0124] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 9As shown in the figure. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for calculating quality information of an edge device. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0125] Those skilled in the art can understand that Figure 9 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0126] In one embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the following steps are implemented: calculating data status information, where the data status information is used to characterize the importance of the local model update information deployed in the edge device; calculating channel status information, where the channel status information is used to characterize the channel conditions of the edge device; calculating transmission energy information, where the transmission energy information is used to characterize the transmission energy of the edge device; determining quality information according to the data status information, the channel status information, and the transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server.

[0127] In one embodiment, when the processor executes the computer program, the following steps are further implemented: obtaining the gradient information of the local model deployed in the edge device; calculating the data status information according to the gradient information.

[0128] In one embodiment, when the processor executes the computer program, the following steps are further implemented: obtaining the quasi-static channel gain between the edge device and the edge server; calculating the channel status information according to the quasi-static channel gain.

[0129] In one embodiment, when the processor executes the computer program, the following steps are further implemented: obtaining the quasi-static channel gain between the edge device and the edge server and the preset signal-to-noise ratio threshold at the receiving end; calculating the transmission energy information according to the quasi-static channel gain and the preset signal-to-noise ratio threshold at the receiving end.

[0130] In one embodiment, when the processor executes the computer program, the following steps are further implemented: determining whether the edge device received an instruction sent by the edge server during the previous round of global model training; if it is determined that the edge device received an instruction sent by the edge server during the previous round of global model training, using the gradient of the local model deployed in the edge device calculated in this round as the gradient information; if it is determined that the edge device did not receive an instruction sent by the edge server during the previous round of global model training, using the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradient that has not been sent as the gradient information.

[0131] In one embodiment, when the processor executes the computer program, the following steps are further implemented: receiving the current global model parameters of the global model deployed in the edge server; training the local model deployed in the edge device in this round according to the current global model parameters to obtain the gradient information of the local model deployed in the edge device.

[0132] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: calculating data status information, which is used to characterize the importance of the local model update information deployed in the edge device; calculating channel status information, which is used to characterize the channel condition of the edge device; calculating transmission energy information, which is used to characterize the transmission energy of the edge device; determining quality information according to the data status information, the channel status information, and the transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the global model training deployed in the edge server.

[0133] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining the gradient information of the local model deployed in the edge device; calculating data status information according to the gradient information.

[0134] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining the quasi-static channel gain between the edge device and the edge server; calculating channel status information according to the quasi-static channel gain.

[0135] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining the quasi-static channel gain between the edge device and the edge server and a preset received signal-to-noise ratio threshold at the receiving end; calculating transmission energy information according to the quasi-static channel gain and the preset received signal-to-noise ratio threshold at the receiving end.

[0136] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: determining whether the edge device received an instruction sent by the edge server during the previous round of global model training; if it is determined that the edge device received an instruction sent by the edge server during the previous round of global model training, using the gradient of the local model deployed in the edge device calculated in this round as gradient information; if it is determined that the edge device did not receive an instruction sent by the edge server during the previous round of global model training, using the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradients not sent as gradient information.

[0137] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: receiving the current global model parameters of the global model deployed in the edge server; training the local model deployed in the edge device in this round according to the current global model parameters to obtain the gradient information of the local model deployed in the edge device.

[0138] In one embodiment, a computer program product is provided, including a computer program, which when executed by a processor implements the following steps: calculating data status information, where the data status information is used to characterize the importance of the local model update information deployed in the edge device; calculating channel status information, where the channel status information is used to characterize the channel condition of the edge device; calculating transmission energy information, where the transmission energy information is used to characterize the transmission energy of the edge device; determining quality information according to the data status information, the channel status information, and the transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the global model training deployed in the edge server.

[0139] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining the gradient information of the local model deployed in the edge device; calculating data status information according to the gradient information.

[0140] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining the quasi-static channel gain between the edge device and the edge server; calculating channel status information according to the quasi-static channel gain.

[0141] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining the quasi-static channel gain between the edge device and the edge server and a preset signal-to-noise ratio threshold at the receiving end; calculating transmission energy information according to the quasi-static channel gain and the preset signal-to-noise ratio threshold at the receiving end.

[0142] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: determining whether the edge device received an instruction sent by the edge server during the previous round of global model training; if it is determined that the edge device received an instruction sent by the edge server during the previous round of global model training, using the gradient of the local model deployed in the edge device calculated in this round as gradient information; if it is determined that the edge device did not receive an instruction sent by the edge server during the previous round of global model training, using the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradients not sent as gradient information.

[0143] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: receiving the current global model parameters of the global model deployed in the edge server; training the local model deployed in the edge device in this round according to the current global model parameters to obtain the gradient information of the local model deployed in the edge device.

[0144] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties.

[0145] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0146] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0147] The above-described embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for calculating quality information of an edge device, characterized in that, Applied to an edge device in a federated learning system, the federated learning system includes a plurality of the edge devices and an edge server that has established a communication connection with each of the edge devices. The method includes: Calculating data status information, where the data status information is used to characterize the importance of the local model update information deployed in the edge device; Calculating channel status information, where the channel status information is used to characterize the channel condition of the edge device; Calculating transmission energy information, where the transmission energy information is used to characterize the transmission energy of the edge device; Determining quality information according to the data status information, the channel status information, and the transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server; The calculating the transmission energy information includes: Obtaining the quasi-static channel gain between the edge device and the edge server and a preset receiver signal-to-noise ratio threshold; Calculating the transmission energy information according to the quasi-static channel gain and the preset receiver signal-to-noise ratio threshold.

2. The method according to claim 1, wherein The calculating the data status information includes: Obtaining the gradient information of the local model deployed in the edge device; Calculating the data status information according to the gradient information.

3. The method according to claim 1, wherein The calculating the channel status information includes: Obtaining the quasi-static channel gain between the edge device and the edge server; Calculating the channel status information according to the quasi-static channel gain.

4. The method according to claim 2, wherein The obtaining the gradient information of the local model deployed in the edge device includes: Determining whether the edge device received an instruction sent by the edge server during the previous round of global model training; If it is determined that the edge device received an instruction sent by the edge server during the previous round of global model training, then taking the gradient of the local model deployed in the edge device calculated in this round as the gradient information; If it is determined that the edge device did not receive an instruction sent by the edge server during the previous round of global model training, then taking the sum of the gradient of the local model deployed in the edge device calculated in this round and the accumulated gradient not sent as the gradient information.

5. The method according to claim 4, wherein The method further includes: Receiving the current global model parameters of the global model deployed in the edge server; Performing training on the local model deployed in the edge device according to the current global model parameters to obtain the gradient information of the local model deployed in the edge device.

6. A quality information calculation device for an edge device, characterized in that, Applied to an edge device in a federated learning system, the federated learning system includes a plurality of the edge devices and an edge server that has established a communication connection with each of the edge devices. The device includes: A first calculation module, configured to calculate data status information, where the data status information is used to characterize the importance of the local model update information deployed in the edge device; A second calculation module, configured to calculate channel status information, where the channel status information is used to characterize the channel condition of the edge device; A third calculation module, configured to calculate transmission energy information, where the transmission energy information is used to characterize the transmission energy of the edge device; A determination module, configured to determine quality information according to the data status information, the channel status information, and the transmission energy information, where the quality information is used to characterize the contribution degree of the gradient information of the local model deployed in the edge device to the training of the global model deployed in the edge server; The third calculation module is specifically configured to obtain the quasi-static channel gain between the edge device and the edge server and a preset receiver signal-to-noise ratio threshold; calculate the transmission energy information according to the quasi-static channel gain and the preset receiver signal-to-noise ratio threshold.

7. The device according to claim 6, characterized in that, The first calculation module is specifically configured to obtain the gradient information of the local model deployed in the edge device; calculate the data status information according to the gradient information.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images