Model parameter updating method and device, computer device and storage medium
By using Fourier series in federated learning to determine specific and shared parameters, the heterogeneity of client data is addressed, improving the accuracy and adaptability of the data distribution prediction model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG LAB
- Filing Date
- 2023-07-20
- Publication Date
- 2026-05-19
AI Technical Summary
Existing federated learning methods cannot effectively address the heterogeneity of data across clients, resulting in low accuracy in predicting data distribution using neural network models.
By passing shared and specific parameters of the data distribution prediction model between the federated learning client and server, using Fourier series to perform finite summation to determine the specific parameters, fitting the data distribution of each client, and performing iterative training on the client, the shared parameters are aggregated to improve the accuracy of the model's data distribution prediction.
It improves the accuracy of data distribution prediction models, enabling them to better adapt to the data characteristics of various clients and enhance the performance and accuracy of models on unknown data.
Smart Images
Figure CN117057439B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, computer device, and storage medium for updating model parameters. Background Technology
[0002] With the development of artificial intelligence and deep learning, neural network models are increasingly permeating our daily lives and work. For example, in the field of autonomous driving, neural network models can be used for path planning, improving driving safety; in the medical field, they can classify and recognize medical images, improving the accuracy of disease diagnosis. Therefore, the performance of neural network models is extremely important.
[0003] To ensure the performance of neural network models, a large amount of data is often required during training. However, the training data is distributed across different clients. Currently, due to data security concerns, much useful data can only be stored on its respective client. Therefore, to ensure the performance of neural network models, a federated learning approach is used to share useful data from each client to train the neural network model.
[0004] However, current federated learning methods can only utilize useful data from each client in independent and identically distributed scenarios to train neural network models, failing to address the data heterogeneity problem among clients, which in turn leads to low accuracy in predicting data distribution in neural network models. Summary of the Invention
[0005] Therefore, it is necessary to provide a model parameter update method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of data distribution prediction in response to the above-mentioned technical problems.
[0006] Firstly, this application provides a method for updating model parameters. Applied to a federated learning client, the method includes:
[0007] The system receives a parameter update instruction from the federated learning server for a data distribution prediction model shared by all federated learning clients. The parameter update instruction carries shared parameters and specific parameters of the data distribution prediction model. The specific parameters are determined by a finite summation of Fourier series and are used to fit the data distribution of each federated learning client. The shared parameters are used to extract common data features from each federated learning client. The data distribution prediction model is used to predict the data type of the data to be processed.
[0008] In response to the parameter update instruction, the shared parameters and specific parameters are updated to obtain the updated shared parameters and updated specific parameters;
[0009] The updated shared parameters are sent to the federated learning server, while the updated specific parameters are retained locally.
[0010] Receive the aggregated shared parameters sent by the federated learning server after aggregating the updated shared parameters;
[0011] Based on the aggregated shared parameters and the specific parameters, the corresponding target shared parameters and target specific parameters are determined.
[0012] In one embodiment, the specific parameter is determined by performing a finite summation of the Fourier series to determine the activation function of the data distribution prediction model common to all federated learning clients, and by initializing the coefficients of the Fourier series of the activation function.
[0013] In one embodiment, updating the shared parameters and specific parameters in response to the parameter update instruction to obtain the updated shared parameters and updated specific parameters includes:
[0014] The locally stored sample dataset is divided into a training dataset and a validation dataset.
[0015] In response to the parameter update instruction, determine the parameter update order of the shared parameter and the specific parameter;
[0016] According to the parameter update order, the data distribution prediction model is trained iteratively multiple times using the training dataset and the validation dataset, and the shared parameters and specific parameters are updated to obtain the updated shared parameters and updated specific parameters.
[0017] In one embodiment, the parameter update order is as follows: first, fix the specific parameter and update the shared parameter; then, fix the updated shared parameter and update the specific parameter. Each iteration of the multi-iteration training includes:
[0018] With the specific parameters remaining unchanged, the data distribution prediction model is trained using the training dataset, and the shared parameters are updated to obtain the updated shared parameters.
[0019] With the updated shared parameters unchanged, the data distribution prediction model is trained using the validation dataset, and the specific parameters are updated to obtain the updated specific parameters.
[0020] Secondly, this application also provides a model parameter updating device. The device includes:
[0021] The instruction receiving module is used to receive parameter update instructions sent by the federated learning server for a data distribution prediction model common to all federated learning clients. The parameter update instructions carry shared parameters and specific parameters of the data distribution prediction model. The specific parameters are determined by a finite summation of Fourier series and are used to fit the data distribution of each federated learning client. The shared parameters are used to extract common data features from each federated learning client. The data distribution prediction model is used to predict the data type of the data to be processed.
[0022] An update module is used to update the shared parameters and specific parameters in response to the parameter update instruction, so as to obtain the updated shared parameters and updated specific parameters;
[0023] The data transceiver module is used to send the updated shared parameters to the federated learning server and retain the updated specific parameters locally.
[0024] Receive the aggregated shared parameters sent by the federated learning server after aggregating the updated shared parameters;
[0025] The update module is used to determine the corresponding target shared parameters and target specific parameters based on the aggregated shared parameters and the specific parameters.
[0026] Thirdly, this application also provides a method for updating model parameters. Applied to a federated learning server, the method includes:
[0027] By performing finite summation on the Fourier series, specific parameters of the data distribution prediction model common to all federated learning clients are determined; these specific parameters are used to fit the data distribution of each federated learning client; the data distribution prediction model is used to predict the data type of the data to be processed.
[0028] The parameter update instructions for the data distribution prediction model are sent to each of the federated learning clients. The parameter update instructions carry the shared parameters and the specific parameters of the data distribution prediction model. The shared parameters are used to extract the common features of the data from each of the federated learning clients.
[0029] The system receives updated shared parameters sent by each of the federated learning clients in response to the parameter update instruction, aggregates the updated shared parameters sent by each of the federated learning clients, and obtains aggregated shared parameters; wherein, each of the federated learning clients updates the specific parameter in response to the parameter update instruction to determine the updated specific parameter, and retains the updated specific parameter locally;
[0030] The aggregated shared parameters are sent to each of the federated learning clients, so that each of the federated learning clients updates the target model parameters and target-specific parameters based on the aggregated shared parameters and the updated specific parameters.
[0031] In one embodiment, the step of performing a finite summation of the Fourier series to determine specific parameters of the data distribution prediction model common to all federated learning clients includes:
[0032] The activation function of the data distribution prediction model shared by all federated learning clients is determined by finite summation of Fourier series.
[0033] The coefficients of the Fourier series of the activation function are initialized to determine the specific parameters of the data distribution prediction model shared by all federated learning clients.
[0034] Fourthly, this application also provides a model parameter updating device. The device includes:
[0035] The parameter determination module is used to perform finite summation on the Fourier series to determine specific parameters of the data distribution prediction model common to all federated learning clients; the specific parameters are used to fit the data distribution of each federated learning client; the data distribution prediction model is used to predict the data type of the data to be processed.
[0036] The instruction sending module is used to send parameter update instructions for the data distribution prediction model to each of the federated learning clients. The parameter update instructions carry the shared parameters of the data distribution prediction model and the specific parameters. The shared parameters are used to extract the common features of the data from each of the federated learning clients.
[0037] An aggregation processing module is used to receive updated shared parameters sent by each of the federated learning clients in response to the parameter update instruction, and to perform aggregation processing on the updated shared parameters sent by each of the federated learning clients to obtain aggregated shared parameters; wherein, each of the federated learning clients updates the specific parameter in response to the parameter update instruction to determine the updated specific parameter, and retains the updated specific parameter locally;
[0038] The parameter sharing module is used to send the aggregated shared parameters to each of the federated learning clients, so that each of the federated learning clients updates the target model parameters and target-specific parameters based on the aggregated shared parameters and the updated specific parameters.
[0039] Fifthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described model parameter update method.
[0040] Sixthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the above-described model parameter update method.
[0041] Seventhly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the aforementioned model parameter update method.
[0042] The aforementioned model parameter update method, apparatus, computer equipment, storage medium, and computer program product receive parameter update instructions for the data distribution prediction model sent by the federated learning server from each federated learning client. Based on the specific parameters determined by finite summation of Fourier series and shared parameters carried in the parameter update instructions, each federated learning client updates the shared parameters and specific parameters. The updated shared parameters are then sent to the federated learning server for aggregation processing to obtain aggregated shared parameters. The updated specific parameters are stored in their respective federated learning clients for fitting their respective data distributions without being shared. Based on the aggregated shared parameters and the updated specific parameters, the corresponding target model parameters and target specific parameters are determined. By extracting the common data features of each federated learning client from the shared parameters in the data distribution prediction model shared by all federated learning clients, aggregating and sharing them, and fitting the specific parameters of each federated learning client's shared data distribution prediction model to their respective data distributions, the target shared parameters and target specific parameters of the data distribution prediction model on each federated learning client are determined. This solves the data heterogeneity problem among clients and thus improves the accuracy of data distribution prediction. Attached Figure Description
[0043] Figure 1 This is a diagram illustrating the application environment of a model parameter update method in one embodiment.
[0044] Figure 2 This is a flowchart illustrating a model parameter update method in one embodiment;
[0045] Figure 3 This is a schematic diagram of using a Fourier series as an activation function in one embodiment;
[0046] Figure 4 This is a flowchart illustrating the steps for updating shared parameters and specific parameters in one embodiment.
[0047] Figure 5 This is a flowchart illustrating the model parameter update method in another embodiment;
[0048] Figure 6 This is a schematic diagram illustrating parameter updates for model parameter updates in one embodiment.
[0049] Figure 7 This is a flowchart illustrating the model parameter update method in another embodiment;
[0050] Figure 8 This is a structural block diagram of a model parameter update device in one embodiment;
[0051] Figure 9 This is a structural block diagram of a model parameter update device in one embodiment;
[0052] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0054] Federated learning is essentially a distributed training framework. Based on this framework, while protecting the data security of each client, it can train a common data distribution prediction model using data from each client. However, in real-world applications, the data on each client is not independently and identically distributed, i.e., data heterogeneity. The data distribution prediction model trained using federated learning cannot fit the individual characteristics of each client's data, resulting in low prediction accuracy. Data heterogeneity can be understood as class imbalance in the data across clients; that is, the data types of each client are consistent (p(x) are the same), but their class ratios are inconsistent (p(y|x) are different). For example, suppose there are two clients with two data categories, labeled 0 and 1 respectively. The first client has 400 data points labeled 0 and 100 data points labeled 1. The second client has 100 data points labeled 0 and 400 data points labeled 1. To address this technical problem, a model parameter update method is proposed.
[0055] The model parameter update method provided in this application can be applied to, for example, Figure 1The application environment shown includes at least one federated learning client 102 and a federated learning server 104. The federated learning client 102 communicates with the federated learning server 104 via a network. A data storage system can store the data that the federated learning server 104 needs to process. The data storage system can be integrated into the federated learning server 104 or located in the cloud or on another network server. The federated learning server performs finite summation on the Fourier series to determine the specific parameters of the data distribution prediction model shared by all federated learning clients. It then sends parameter update instructions for the data distribution prediction model to each federated learning client. These instructions carry the shared parameters and specific parameters of the data distribution prediction model. Each federated learning client receives the parameter update instructions from the federated learning server and updates its shared and specific parameters accordingly. The updated shared parameters are then sent to the federated learning server, while the specific parameters are stored locally. The federated learning server aggregates the updated shared parameters from each client to obtain aggregated shared parameters, which are then sent to each client. This allows each client to update the target model parameters and target specific parameters based on the aggregated shared parameters and the updated specific parameters. The federated learning client 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, etc. The federated learning server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.
[0056] In one embodiment, such as Figure 2 As shown, a model parameter update method is provided, which is applied to... Figure 1 Taking the federated learning client in [the context of the project] as an example, the steps are as follows:
[0057] Step 202: Receive parameter update instructions from the federated learning server for the data distribution prediction model common to all federated learning clients; wherein, the parameter update instructions carry the shared parameters and specific parameters of the data distribution prediction model.
[0058] The specific parameters are determined by a finite summation of Fourier series and are used to fit the data distribution of each federated learning client. In other words, when updating the parameters of the data distribution prediction model, both the specific parameters and the shared parameters need to be initialized. Initializing the specific parameters involves initializing the coefficients of the finite summation Fourier series, which are personalized parameters. The shared parameters are used to extract common features of the data from each federated learning client; these shared parameters can be understood as the initial model parameters of the data distribution prediction model.
[0059] Data distribution prediction models are used to predict the data to be processed, determining the probability value of the data belonging to various configured data types. Based on the obtained probability values, the data type of the data to be processed is determined. For example, in a medical application scenario, the acquired data to be processed is a medical image. The image is input into the data distribution prediction model, and by predicting the image, the probability value of the image belonging to each pre-configured medical image is determined. Based on the obtained probability value, the image type of the image is determined, and further, based on the image type, the target object's physical condition can be predicted and diagnosed. Similarly, in an interactive behavior scenario, the acquired data to be processed is the target object's behavioral data. The behavioral data is input into the data distribution prediction model, and by predicting the behavioral data, the probability value of the behavioral data belonging to each pre-configured behavior is determined. Based on the obtained probability value, the target object's behavior type can be determined.
[0060] Specifically, before training all data distribution prediction models for each federated learning client, the federated learning server initializes the shared parameters and specific parameters of the data distribution prediction models common to all federated learning clients, obtains the corresponding initialized shared parameters and specific parameters, triggers the generation of parameter update instructions carrying the shared parameters and specific parameters, and each federated learning client receives the parameter update instructions sent by the federated learning server.
[0061] Step 204: In response to the parameter update instruction, update the shared parameters and specific parameters to obtain the updated shared parameters and updated specific parameters.
[0062] The optimization of shared parameters and specific parameters can be performed in a single-level or two-level manner. Single-level optimization involves updating both shared and specific parameters simultaneously. Two-level optimization can involve updating the shared parameters first and then the specific parameters, or vice versa.
[0063] Specifically, since the data of each federated learning client is not independent and identically distributed, to improve the training effect and accuracy of the data distribution prediction model, the data distribution prediction model on each federated learning client is iteratively trained using the sample datasets corresponding to each client, based on the acquired shared parameters and specific parameters. The shared parameters and specific parameters are then updated to obtain the updated shared parameters and updated specific parameters corresponding to the specific parameters. The number of iterations can be determined according to actual needs, such as 5 or 10 times. It is worth noting that the iterative training here refers to obtaining the updated shared parameters and updated specific parameters through a predetermined number of iterations, not that the loss function of the data distribution prediction model must converge to obtain the shared parameters and specific parameters.
[0064] Step 206: Send the updated shared parameters to the federated learning server and retain the updated specific parameters locally.
[0065] Step 208: Receive the aggregated shared parameters sent by the federated learning server after aggregating the updated shared parameters.
[0066] The aggregation process includes summing and averaging the updated shared parameters to obtain aggregated shared parameters, which are then sent to each federated learning client for sharing. The aggregated shared parameters can extract common features from all federated learning clients.
[0067] Step 210: Determine the corresponding target shared parameters and target specific parameters based on the aggregated shared parameters and the updated specific parameters.
[0068] Specifically, upon obtaining the aggregated shared parameters and the updated specific parameters, if the data distribution prediction model algorithm converges, the aggregated shared parameters are determined as the target shared parameters, and the updated specific parameters are determined as the target specific parameters, resulting in a trained data distribution prediction model. If the algorithm does not converge, the steps of responding to parameter update instructions to update the shared parameters and specific parameters are continued until the algorithm converges, resulting in a trained data distribution prediction model. Algorithm convergence can be determined based on the magnitude of the gradient update of the shared parameters. When the L2 norm of the gradient of the shared parameters is less than or equal to a preset value, the data distribution prediction model can be considered converged.
[0069] Taking medical data as an example, the datasets of each hospital's clients have certain characteristics and distributions, and these characteristics and distributions affect the performance and accuracy of the model. Based on the trained data distribution prediction model, it can better adapt to the data characteristics of each client, that is, it can better adapt to the local data of each client, thereby effectively improving the performance and accuracy of the model on unknown data, that is, improving the accuracy of the data distribution prediction model.
[0070] In the aforementioned model parameter update method, each federated learning client receives a parameter update instruction for the data distribution prediction model sent by the federated learning server. Based on the specific parameters determined by finite summation of Fourier series carried in the parameter update instruction, as well as shared parameters, each federated learning client updates the shared parameters and specific parameters. The updated shared parameters are then sent to the federated learning server for aggregation processing to obtain aggregated shared parameters. The updated specific parameters are stored in their respective federated learning clients for fitting their respective data distributions without being shared. Based on the aggregated shared parameters and the updated specific parameters, the corresponding target model parameters and target specific parameters are determined. By extracting the common data features of each federated learning client from the shared parameters in the data distribution prediction model shared by all federated learning clients, aggregating them, and then sharing them, and by fitting the specific parameters of each federated learning client's shared data distribution prediction model to their respective data distributions, the target shared parameters and target specific parameters of the data distribution prediction model on each federated learning client are determined. This solves the data heterogeneity problem among clients, thereby improving the accuracy of data distribution prediction. Understandably, activation functions in neural networks can introduce nonlinear relationships and increase the expressive power of neural network functions, thereby enabling neural networks to handle various complex problems and better adapt to different tasks and data distributions. However, current activation functions in neural networks are pre-set based on human experience, which cannot adapt to all application scenarios or accurately determine data distributions. Therefore, we use finite summation Fourier series to replace the activation functions in the data distribution prediction model. For each activation function in the data distribution prediction model, we use a different Fourier series to represent it.
[0071] like Figure 3 The diagram shown illustrates a Fourier series as an activation function in one embodiment. The activation function can be applied to the network layer of a data distribution prediction model according to actual needs. Figure 3 The activation functions shown are applied to the outputs of the convolutional and fully connected layers. The functions of the convolutional, max-pooling, and fully connected layers are the same as those in the existing ones, and will not be elaborated here.
[0072] The Fourier series can be expressed as:
[0073]
[0074] Where f(x) is a finite-sum Fourier series, where finite summation means that summing over variables k from 0 to n yields a finite Fourier series summation, and n is a hyperparameter; a is a hyperparameter used to determine the period of the trigonometric function; b k c k The Fourier coefficients are parameters that need to be optimized.
[0075] Optionally, in one embodiment, the activation function of the data distribution prediction model common to all federated learning clients is determined by performing a finite summation of the Fourier series, and specific parameters are initialized by initializing the coefficients of the Fourier series of the activation function. Before updating the model parameters of the data distribution prediction model, the coefficients of the Fourier series are used as personalized parameters to replace the activation function, without pre-setting them, to determine the activation function that matches the corresponding scenario of each federated client.
[0076] In one embodiment, such as Figure 4 As shown, the steps for updating shared parameters and specific parameters include:
[0077] Step 402: Divide the locally stored sample dataset to obtain the training dataset and the validation dataset.
[0078] The method for partitioning the locally stored sample dataset into training and validation datasets can be determined using existing partitioning methods. It should be noted that while the number of sample datasets used for partitioning is the same across all federated learning clients, the data distribution differs, and the partitioning method for the sample datasets can be the same.
[0079] Step 404: Determine the parameter update order for shared parameters and specific parameters.
[0080] The parameter update order for shared parameters and specific parameters includes any one of the following: updating both shared parameters and specific parameters simultaneously, updating shared parameters first and then specific parameters, or updating specific parameters first and then shared parameters. The specific parameter update order can be determined in response to an update instruction carrying a target parameter update order, based on the target parameter update order.
[0081] Step 406: Following the parameter update order, the data distribution prediction model is trained iteratively multiple times using the training dataset and validation dataset, and the shared parameters and specific parameters are updated to obtain the updated shared parameters and updated specific parameters.
[0082] The training dataset is used to update the shared parameters of the data distribution prediction model, while the validation dataset is used to update specific parameters of the model, namely the coefficients of the Fourier series. Updating the shared parameters of the data distribution prediction model using the training dataset incurs a corresponding training loss, while updating the specific parameters using the validation dataset incurs a corresponding validation loss.
[0083] Training loss It can be represented as:
[0084] Where α is the coefficient of the Fourier series, and w is the shared parameter.
[0085] Verify loss It can be represented as:
[0086] Where α is the coefficient of the Fourier series, and w is the shared parameter.
[0087] Specifically, following the parameter update order, the data distribution prediction model is trained iteratively multiple times using the training dataset and the validation dataset. The shared parameters are updated using the training dataset, and the specific parameters are updated using the validation dataset, resulting in the corresponding updated shared parameters and updated specific parameters.
[0088] Optionally, in one embodiment, when the parameter update order is to first fix specific parameters and update shared parameters, and then fix the updated shared parameters and update specific parameters, each iteration step in the multiple iterations of training the data distribution prediction model using the training dataset and validation dataset includes the following:
[0089] With specific parameters remaining constant, a data distribution prediction model is trained using the training dataset, and the shared parameters are updated to obtain the updated shared parameters. Then, with the updated shared parameters unchanged, the data distribution prediction model is trained using the validation dataset, and the specific parameters are updated to obtain the updated specific parameters. This training method can specifically meet the needs of different scenarios and improve the adaptability of the data distribution prediction model. It's understandable that when the model parameters are first updated, the initial parameters with fixed Fourier coefficients are used. If the update at round t ends and the update at round t+1 begins, fixing the Fourier coefficients means fixing the Fourier coefficients obtained in round t of training. The same applies to updating the shared parameters.
[0090] In the above embodiments, by dividing the locally stored sample dataset into a training dataset and a validation dataset, the shared parameters of the data distribution prediction model are updated using the training dataset, and the specific parameters of the data distribution prediction model are updated using the validation dataset. This allows the data distribution prediction model to better adapt to new data distributions, thereby improving the model's generalization ability and performance, and increasing the prediction accuracy of the data distribution prediction model in practical applications.
[0091] In another embodiment, such as Figure 5 As shown, a model parameter update method is provided, which is applied to... Figure 1 Taking the federated learning client in [the context of the project] as an example, the steps are as follows:
[0092] Step 502: Receive parameter update instructions from the federated learning server for the data distribution prediction model common to all federated learning clients; wherein the parameter update instructions carry the shared parameters and specific parameters of the data distribution prediction model.
[0093] The specific parameters are determined by finite summation of Fourier series and are used to fit the data distribution of each federated learning client; the shared parameters are used to extract common features of the data from each federated learning client.
[0094] Step 504: Divide the locally stored sample dataset to obtain the training dataset and the validation dataset.
[0095] Step 506: Respond to the parameter update command and determine the parameter update order for shared parameters and specific parameters.
[0096] Step 508: Following the parameter update order, the data distribution prediction model is trained iteratively multiple times using the training dataset and validation dataset, and the shared parameters and specific parameters are updated to obtain the updated shared parameters and updated specific parameters.
[0097] Step 510: Send the updated shared parameters to the federated learning server and retain the updated specific parameters locally.
[0098] Step 512: Receive the aggregated shared parameters sent by the federated learning server after aggregating the updated shared parameters.
[0099] Step 514: If the data distribution prediction model does not converge based on the aggregated shared parameters and the updated specific parameters, then proceed to step 506; if the data distribution prediction model converges, then proceed to step 516.
[0100] Step 516: Determine the target shared parameters and target specific parameters of the data distribution prediction model to obtain the trained data distribution prediction model.
[0101] Given a learning rate, the method for updating the parameters can be specifically specified. For example, if the learning rate of the shared parameters is ηw, then the method for updating the shared parameters w is as follows:
[0102]
[0103] Let the learning rate of a specific parameter be ηα, then the update method for the specific parameter α is as follows:
[0104]
[0105] In one embodiment, such as Figure 6 The diagram illustrates the parameter update process for the model, including client 1, client 2... client n, and the server. The data distribution prediction model shared by each federated learning client includes its own shared parameters and specific parameters. For example, client 1's specific parameter is the Fourier coefficient a1, and its corresponding shared parameter is w1; client 2's specific parameter is the Fourier coefficient a2, and its corresponding shared parameter is w2; and client n's specific parameter is the Fourier coefficient a... n The corresponding shared parameter is w n Upon receiving a parameter update command, the system first fixes specific parameters, trains a data distribution prediction model using the validation dataset, and updates the shared parameters. Then, it fixes the updated shared parameters, trains the data distribution prediction model again using the validation dataset, and updates the specific parameters, obtaining the updated specific parameters. The updated shared parameters are sent to the federated learning server, while the updated specific parameters are stored locally. The system then receives the aggregated shared parameters from the federated learning server after aggregating the updated shared parameters. Based on the aggregated shared parameters and the specific parameters, the system determines the corresponding target shared parameters and target specific parameters.
[0106] Optionally, in one embodiment, medical image data to be processed is acquired, the medical image to be processed is input into a trained data distribution prediction model, the target object in the medical image to be processed is identified by the trained data distribution prediction model, and the extracted features of the target object are extracted. The extracted features are compared with the features of each pre-configured medical image type to obtain the probability value belonging to each pre-configured medical image, and the image type of the medical image to be processed is determined according to the probability value.
[0107] Optionally, in one embodiment, medical image data to be processed is acquired, the medical image to be processed is input into a trained data distribution prediction model, the distribution of the medical image to be processed is predicted by the trained data distribution prediction model, image enhancement processing is performed on the image region corresponding to the target object in the medical image to be processed, and the enhanced medical image is obtained. Image recognition is performed on the enhanced medical image to obtain the probability value belonging to each pre-configured medical image, and the image type of the enhanced medical image is determined according to the probability value, thereby improving the accuracy of medical image classification.
[0108] In the above embodiments, by using Fourier series as the activation function to replace the original activation function, specific parameters of the data distribution prediction model are obtained as personalized parameters to fit the data distribution of each client; and the model parameters of the data distribution prediction model other than the activation function are used as shared parameters to learn the common features of the data of each client; using the locally stored sample dataset, the shared parameters and Fourier series coefficients of the data distribution prediction model are alternately updated, and the updated network model parameters are sent to the server; the server sums and averages the network model parameters of all clients, and then sends them to the federated learning client. Through iterative training, until the final data distribution prediction model converges, the target shared parameters and target specific parameters are determined, and a trained data distribution prediction model is obtained, thereby improving the accuracy of the data distribution prediction model.
[0109] Based on the same inventive concept, this application also provides an application of the aforementioned model parameter update method in a federated learning server. The provided solution is similar to the implementation described in the above method; therefore, the specific limitations in one or more federated learning server embodiments provided below can be found in the limitations of the model parameter update method applied to the federated learning client described above, and will not be repeated here.
[0110] In another embodiment, such as Figure 7 As shown, a model parameter update method is provided, which is applied to... Figure 1 Taking the federated learning server in China as an example, the following steps are included:
[0111] Step 702: Perform a finite summation of the Fourier series to determine specific parameters of the data distribution prediction model common to all federated learning clients; the specific parameters are used to fit the data distribution of each federated learning client.
[0112] Step 704: The parameter update instructions for the data distribution prediction model are sent to each federated learning client. The parameter update instructions carry the shared parameters and specific parameters of the data distribution prediction model. The shared parameters are used to extract the common features of the data from each federated learning client.
[0113] Step 706: Receive the updated shared parameters sent by each federated learning client in response to the parameter update instruction, aggregate the updated shared parameters sent by each federated learning client to obtain aggregated shared parameters; wherein, each federated learning client updates specific parameters in response to the parameter update instruction to determine the updated specific parameters, and retains the updated specific parameters locally.
[0114] Step 708: The aggregated shared parameters are sent to each federated learning client so that each federated learning client can update the target model parameters and target-specific parameters based on the aggregated shared parameters and the updated specific parameters.
[0115] The aforementioned model parameter update method involves each federated learning client receiving parameter update instructions for the data distribution prediction model sent by the federated learning server. Based on the specific parameters determined by finite summation of Fourier series carried in the parameter update instructions, as well as shared parameters, each federated learning client updates the shared and specific parameters. The updated shared parameters are then sent to the federated learning server for aggregation processing to obtain aggregated shared parameters. Each specific parameter is stored in its respective federated learning client for fitting its own data distribution without being shared. Based on the aggregated shared and specific parameters, the corresponding target model parameters and target specific parameters are determined. By extracting the common data features from the shared parameters in the data distribution prediction model shared by all federated learning clients, aggregating them, and then sharing them, and by fitting the specific parameters from the shared data distribution prediction model of each federated learning client to its own data distribution, the target shared parameters and target specific parameters of the data distribution prediction model on each federated learning client are determined. This solves the data heterogeneity problem among clients and thus improves the accuracy of data distribution prediction.
[0116] Optionally, in one embodiment, the specific parameters of the data distribution prediction model shared by all federated learning clients are determined by performing a finite summation of the Fourier series, including: determining the activation function of the data distribution prediction model shared by all federated learning clients by performing a finite summation of the Fourier series; and determining the specific parameters of the data distribution prediction model shared by all federated learning clients by using the coefficients of the Fourier series of the activation function.
[0117] Optionally, in one embodiment, the updated shared parameters are received, aggregated to obtain aggregated shared parameters, and then sent to each federated learning client. The shared parameters can be parameters from convolutional layers. Aggregating the shared parameters ensures that the parameters in the convolutional layers of the data distribution prediction model on each federated learning client are consistent, allowing the learning of common features from the datasets of each client.
[0118] Optionally, the following is an application scenario for the model parameter update method, using the big data healthcare field as an example. Each healthcare institution's client stores patients' electronic medical records, CT images, and other medical data. In practical applications, the medical data stored on each institution's client is not independently and identically distributed. To enable each institution's client to obtain a well-trained personalized neural network model that better fits the local medical data distribution, the model parameters of the data distribution prediction model are updated using a personalized federated learning approach based on Fourier series.
[0119] The activation function of the data distribution prediction model shared by all federated learning clients is determined by finite summation of the Fourier series; the coefficients of the Fourier series of the activation function are used to determine the specific parameters of the data distribution prediction model shared by all federated learning clients, and the specific parameters and shared parameters of the data distribution prediction model are initialized.
[0120] The federated learning client receives parameter update instructions from the federated learning server for the data distribution prediction model shared by all federated learning clients. These instructions carry shared parameters and specific parameters of the data distribution prediction model. The specific parameters are determined by a finite summation of Fourier series and are used to fit the data distribution of each federated learning client. The shared parameters are used to extract common features of the data from each federated learning client. The federated learning client divides the locally stored sample dataset into training and validation datasets. Responding to parameter update commands, it determines the update order of shared and specific parameters. Following this order, it iteratively trains the data distribution prediction model using the training and validation datasets, updating the shared and specific parameters. The updated shared parameters are then sent to the federated learning server, while the updated specific parameters are stored locally. The client receives the aggregated shared parameters from the federated learning server after aggregating the updated shared parameters. If the data distribution prediction model fails to converge, it continues to update the shared and specific parameters. If the model converges, it determines the target shared and target specific parameters, resulting in a well-trained data distribution prediction model.
[0121] The process involves acquiring medical image data to be processed, inputting the medical images to be processed into a trained data distribution prediction model, identifying target objects in the medical images to be processed through the trained data distribution prediction model, extracting the features of the target objects, comparing the extracted features with the features of each pre-configured medical image type to obtain the probability value of belonging to each pre-configured medical image, and determining the image type of the medical image to be processed based on the probability value.
[0122] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed 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 performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0123] Based on the same inventive concept, this application also provides a model parameter updating device for implementing the model parameter updating method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more model parameter updating device embodiments provided below can be found in the limitations of the model parameter updating method described above, and will not be repeated here.
[0124] In one embodiment, such as Figure 8 As shown, a model parameter update device is provided, including: an instruction receiving module 802, an update module 804, and a data transceiver module 806, wherein:
[0125] The instruction receiving module 802 is used to receive parameter update instructions sent by the federated learning server for the data distribution prediction model common to all federated learning clients. The parameter update instructions carry shared parameters and specific parameters of the data distribution prediction model. The specific parameters are determined by finite summation of Fourier series and are used to fit the data distribution of each federated learning client. The shared parameters are used to extract the common features of the data of each federated learning client.
[0126] The update module 804 is used to update the shared parameters and specific parameters in response to the parameter update command, so as to obtain the updated shared parameters and updated specific parameters.
[0127] The data transceiver module 806 is used to send the updated shared parameters to the federated learning server and retain the updated specific parameters locally.
[0128] Receive the aggregated shared parameters sent by the federated learning server after processing the updated shared parameters.
[0129] The update module 804 is used to update the aggregated shared parameters and the updated specific parameters, and to determine the corresponding target model parameters and target-specific parameters.
[0130] In the aforementioned model parameter update device, each federated learning client receives parameter update instructions for the data distribution prediction model sent by the federated learning server. Based on the specific parameters determined by finite summation of Fourier series carried in the parameter update instructions, as well as shared parameters, each federated learning client updates the shared parameters and specific parameters. The updated shared parameters are then sent to the federated learning server for aggregation processing to obtain aggregated shared parameters. Each specific parameter is stored in its corresponding federated learning client for fitting its respective data distribution without being shared. Based on the aggregated shared parameters and specific parameters, the corresponding target model parameters and target specific parameters are determined. By extracting the common data features of each federated learning client from the shared parameters in the data distribution prediction model shared by all federated learning clients, aggregating them, and then sharing them, and by fitting the specific parameters of each federated learning client's shared data distribution prediction model to its respective data distribution, the target shared parameters and target specific parameters of the data distribution prediction model on each federated learning client are determined. This solves the data heterogeneity problem among clients, thereby improving the accuracy of data distribution prediction.
[0131] In another embodiment, a model parameter update apparatus is provided, which, in addition to including an instruction receiving module 802, an update module 804, and a data transceiver module 806, further includes: a dataset partitioning module, wherein:
[0132] The specific parameters are determined by summing the Fourier series to find the activation function of the data distribution prediction model common to all federated learning clients, based on the coefficients of the Fourier series of the activation function.
[0133] The dataset partitioning module is used to partition the locally stored sample dataset to obtain the training dataset and the validation dataset;
[0134] The update module 804 is also used to respond to parameter update instructions and determine the parameter update order of shared parameters and specific parameters;
[0135] Following the parameter update order, the data distribution prediction model is trained iteratively multiple times using the training and validation datasets, updating the shared parameters and specific parameters to obtain the updated shared parameters and updated specific parameters.
[0136] The update module 804 is also used to update the shared parameters by first fixing the specific parameters and then fixing the updated shared parameters and updating the specific parameters, and then using the training dataset to train a data distribution prediction model to update the shared parameters while keeping the specific parameters unchanged, so as to obtain the updated shared parameters.
[0137] With the updated shared parameters unchanged, a data distribution prediction model is trained using the validation dataset, and specific parameters are updated to obtain the updated specific parameters.
[0138] In another embodiment, such as Figure 9 As shown, a model parameter update device is provided, including: a parameter determination module 902, an instruction sending module 904, an aggregation processing module 906, and a parameter sharing module 908, wherein:
[0139] The parameter determination module 902 is used to perform finite summation on the Fourier series to determine specific parameters of the data distribution prediction model common to all federated learning clients; the specific parameters are used to fit the data distribution of each federated learning client.
[0140] The instruction sending module 904 is used to send parameter update instructions for the data distribution prediction model to each federated learning client. The parameter update instructions carry the shared parameters and specific parameters of the data distribution prediction model. The shared parameters are used to extract the common features of the data from each federated learning client.
[0141] The aggregation processing module 906 is used to receive the updated shared parameters sent by each federated learning client in response to the parameter update instruction, and to aggregate the updated shared parameters sent by each federated learning client to obtain aggregated shared parameters; wherein, each federated learning client updates specific parameters in response to the parameter update instruction to determine the updated specific parameters, and retains the updated specific parameters locally.
[0142] The parameter sharing module 908 is used to send the aggregated shared parameters to each federated learning client, so that each federated learning client can update the target model parameters and target-specific parameters based on the aggregated shared parameters and the updated specific parameters.
[0143] In the above embodiments, each federated learning client receives a parameter update instruction for the data distribution prediction model sent by the federated learning server. Based on the specific parameters determined by finite summation of Fourier series carried in the parameter update instruction, as well as shared parameters, each federated learning client updates the shared parameters and specific parameters. The updated shared parameters are then sent to the federated learning server for aggregation processing to obtain aggregated shared parameters. Each specific parameter is stored in its corresponding federated learning client for fitting its respective data distribution without being shared. Based on the aggregated shared parameters and specific parameters, the corresponding target model parameters and target specific parameters are determined. By extracting the common data features of each federated learning client from the shared parameters in the data distribution prediction model shared by all federated learning clients, aggregating them, and then sharing them, and by fitting the specific parameters of each federated learning client's shared data distribution prediction model to its respective data distribution, the target shared parameters and target specific parameters of the data distribution prediction model on each federated learning client are determined. This solves the data heterogeneity problem among clients and thus improves the accuracy of data distribution prediction.
[0144] Each module in the aforementioned model parameter update device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.
[0145] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a model parameter update method. The display unit of the computer device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.
[0146] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0147] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.
[0148] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.
[0149] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments. 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 used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.
[0150] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. 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), magnetic 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 take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0151] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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, they should be considered to be within the scope of this specification.
[0152] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for updating model parameters, characterized in that, When applied to a federated learning client, the method includes: The system receives a parameter update instruction from the federated learning server for a data distribution prediction model shared by all federated learning clients. The parameter update instruction carries shared parameters and specific parameters of the data distribution prediction model. The specific parameters are determined by a finite summation of Fourier series and are used to fit the data distribution of each federated learning client. The shared parameters are used to extract common data features from each federated learning client. The data distribution prediction model is used to predict the data type of the data to be processed. In response to the parameter update instruction, the shared parameters and specific parameters are updated to obtain the updated shared parameters and updated specific parameters; The updated shared parameters are sent to the federated learning server, while the updated specific parameters are retained locally. Receive the aggregated shared parameters sent by the federated learning server after aggregating the updated shared parameters; Based on the aggregated shared parameters and the specific parameters, determine the corresponding target shared parameters and target specific parameters; The method further includes: acquiring medical image data to be processed, inputting the medical image to be processed into a trained data distribution prediction model, identifying target objects in the medical image to be processed through the trained data distribution prediction model, extracting the extraction features of the target objects, comparing the extracted features with the features of each pre-configured medical image type to obtain the probability value belonging to each pre-configured medical image, and determining the image type of the medical image to be processed based on the probability value. The specific parameters are determined by summing the Fourier series to find the activation function of the data distribution prediction model common to all federated learning clients, based on the coefficients of the Fourier series of the activation function.
2. The method according to claim 1, characterized in that, The step of updating the shared parameters and specific parameters in response to the parameter update instruction to obtain the updated shared parameters and updated specific parameters includes: The locally stored sample dataset is divided into a training dataset and a validation dataset. In response to the parameter update instruction, determine the parameter update order of the shared parameter and the specific parameter; According to the parameter update order, the data distribution prediction model is trained iteratively multiple times using the training dataset and the validation dataset, and the shared parameters and specific parameters are updated to obtain the updated shared parameters and updated specific parameters.
3. The method according to claim 2, characterized in that, The parameter update order is as follows: first, fix the specific parameter and update the shared parameter; then, fix the updated shared parameter and update the specific parameter. Each iteration of the multi-iteration training includes: With the specific parameters remaining unchanged, the data distribution prediction model is trained using the training dataset, and the shared parameters are updated to obtain the updated shared parameters. With the updated shared parameters unchanged, the data distribution prediction model is trained using the validation dataset, and the specific parameters are updated to obtain the updated specific parameters.
4. A method for updating model parameters, characterized in that, When applied to a federated learning server, the method includes: By performing finite summation on the Fourier series, specific parameters of the data distribution prediction model common to all federated learning clients are determined; these specific parameters are used to fit the data distribution of each federated learning client; the data distribution prediction model is used to predict the data type of the data to be processed. The parameter update instructions for the data distribution prediction model are sent to each of the federated learning clients. The parameter update instructions carry the shared parameters and the specific parameters of the data distribution prediction model. The shared parameters are used to extract the common features of the data from each of the federated learning clients. The system receives updated shared parameters sent by each of the federated learning clients in response to the parameter update instruction, aggregates the updated shared parameters sent by each of the federated learning clients, and obtains aggregated shared parameters; wherein, each of the federated learning clients updates the specific parameter in response to the parameter update instruction to determine the updated specific parameter, and retains the updated specific parameter locally; The aggregated shared parameters are sent to each of the federated learning clients, so that each federated learning client updates the target model parameters and target-specific parameters based on the aggregated shared parameters and the updated specific parameters; and the medical image data to be processed is acquired, the medical image to be processed is input into the trained data distribution prediction model, the target object in the medical image to be processed is identified by the trained data distribution prediction model, and the extracted features of the target object are extracted. The extracted features are compared with the features of each pre-configured medical image type to obtain the probability value belonging to each pre-configured medical image, and the image type of the medical image to be processed is determined according to the probability value. The specific parameters are determined by summing the Fourier series to find the activation function of the data distribution prediction model common to all federated learning clients, based on the coefficients of the Fourier series of the activation function.
5. The method according to claim 4, characterized in that, The process of performing finite summation on the Fourier series to determine specific parameters of the data distribution prediction model shared by all federated learning clients includes: The activation function of the data distribution prediction model shared by all federated learning clients is determined by finite summation of Fourier series. The coefficients of the Fourier series of the activation function are used to determine specific parameters of the data distribution prediction model shared by all federated learning clients.
6. A model parameter update device, characterized in that, The device includes: The instruction receiving module is used to receive parameter update instructions sent by the federated learning server for a data distribution prediction model common to all federated learning clients; wherein, the parameter update instructions carry shared parameters and specific parameters of the data distribution prediction model; the specific parameters are determined by finite summation of Fourier series, and the specific parameters are used to fit the data distribution of each federated learning client; the shared parameters are used to extract common data features of each federated learning client; the data distribution prediction model is used to predict the data type of the data to be processed; An update module is used to update the shared parameters and specific parameters in response to the parameter update instruction, so as to obtain the updated shared parameters and updated specific parameters; The data transceiver module is used to send the updated shared parameters to the federated learning server and retain the updated specific parameters locally. Receive the aggregated shared parameters sent by the federated learning server after aggregating the updated shared parameters; The update module is used to determine the corresponding target shared parameters and target specific parameters based on the aggregated shared parameters and the specific parameters; The update module is also used to acquire medical image data to be processed, input the medical image to be processed into a trained data distribution prediction model, identify target objects in the medical image to be processed through the trained data distribution prediction model, extract the extraction features of the target objects, compare the extracted features with the features of each pre-configured medical image type, obtain the probability value belonging to each pre-configured medical image, and determine the image type of the medical image to be processed based on the probability value. The specific parameters are determined by summing the Fourier series to find the activation function of the data distribution prediction model common to all federated learning clients, based on the coefficients of the Fourier series of the activation function.
7. A model parameter update device, characterized in that, The device includes: The parameter determination module is used to perform finite summation on the Fourier series to determine specific parameters of the data distribution prediction model common to all federated learning clients; the specific parameters are used to fit the data distribution of each federated learning client; the data distribution prediction model is used to predict the data type of the data to be processed. The instruction sending module is used to send parameter update instructions for the data distribution prediction model to each of the federated learning clients. The parameter update instructions carry the shared parameters of the data distribution prediction model and the specific parameters. The shared parameters are used to extract the common features of the data from each of the federated learning clients. An aggregation processing module is used to receive updated shared parameters sent by each of the federated learning clients in response to the parameter update instruction, and to perform aggregation processing on the updated shared parameters sent by each of the federated learning clients to obtain aggregated shared parameters; wherein, each of the federated learning clients updates the specific parameter in response to the parameter update instruction to determine the updated specific parameter, and retains the updated specific parameter locally; The parameter sharing module is used to send the aggregated shared parameters to each of the federated learning clients, so that each federated learning client updates the target model parameters and target-specific parameters based on the aggregated shared parameters and the updated specific parameters; and to acquire medical image data to be processed, input the medical image to be processed into a trained data distribution prediction model, identify target objects in the medical image to be processed through the trained data distribution prediction model, extract the extraction features of the target objects, compare the extracted features with the features of each pre-configured medical image type to obtain the probability value belonging to each pre-configured medical image, and determine the image type of the medical image to be processed based on the probability value; The specific parameters are determined by summing the Fourier series to find the activation function of the data distribution prediction model common to all federated learning clients, based on the coefficients of the Fourier series of the activation function.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.