Multi-task learning method based on federated learning and related devices
By clustering nodes and masking features, the problems of decreased model accuracy and excessively long training time in federated learning in heterogeneous network nodes are solved, and efficient multi-task learning is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2023-02-02
- Publication Date
- 2026-07-24
AI Technical Summary
Existing federated learning suffers from decreased model accuracy and excessively long training time in heterogeneous transmission networks, and existing solutions cannot solve both problems simultaneously, resulting in limited application scenarios.
By clustering the participating nodes, several clusters are determined. The global cluster model and key feature set of the cluster are calculated using federated learning and the SHAP framework. Combined with feature masking technology, multi-task learning is achieved.
It alleviates the problems caused by device heterogeneity and data heterogeneity, improves model accuracy and reduces training time, and enables efficient multi-task learning.
Smart Images
Figure CN116306987B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a multi-task learning method and related equipment based on federated learning. Background Technology
[0002] In heterogeneous transmission networks, existing heterogeneity issues lead to decreased model accuracy and excessively long training times in traditional federated learning. Furthermore, existing solutions to heterogeneity problems suffer from limited scope, addressing only individual issues like decreased accuracy or excessive training time, rather than simultaneously resolving both. This limits their application scenarios. Summary of the Invention
[0003] In view of this, the purpose of this application is to propose a multi-task learning method, apparatus, electronic device and storage medium based on federated learning.
[0004] To achieve the above objectives, this application provides a multi-task learning method based on federated learning.
[0005] Identify participating nodes;
[0006] The participating nodes are clustered to determine several clusters;
[0007] Based on several clusters, a global cluster model is determined through federated learning.
[0008] Based on the global cluster model, the key cluster feature set of any of the clusters is determined by calculation using the SHAP framework.
[0009] Based on the cluster key feature set, determine the global model;
[0010] The global model is trained on any of the clusters to determine the cluster global model of any of the clusters; wherein, multiple clusters are used to implement multi-task learning.
[0011] Optionally, before clustering the participating nodes to determine several clusters, the following steps are included:
[0012] Determine a node training model, and train the participating nodes using the node training model;
[0013] In response to determining that a preset number of training iterations has been reached, the training time and model weights are determined;
[0014] The process of clustering the participating nodes to determine several clusters includes:
[0015] Based on the training time and the model weights, the participating nodes are clustered to determine several clusters.
[0016] Optionally, the method includes:
[0017] The participating nodes are clustered using the KMeans algorithm.
[0018] Optionally, the step of determining a global cluster model through federated learning based on several clusters includes:
[0019] Determine the cluster center based on the clusters;
[0020] The clusters are trained using the federated learning method.
[0021] In response to the determination that the number of training iterations has reached a preset threshold, the computational model is determined to be a cluster global model.
[0022] Optionally, determining the cluster key feature set of any of the clusters based on the cluster global model using the SHAP framework includes:
[0023] The SHAP framework is used to analyze the global cluster model to determine the data characteristics and data characteristic values of any participating node in the cluster.
[0024] Based on the data feature values, determine the key feature set of the node;
[0025] The cluster key feature set is determined by taking the union of several node key feature sets.
[0026] Optionally, determining the node key feature set based on the data feature values includes:
[0027] In response to determining that the value of the data feature is greater than a preset threshold, the data feature corresponding to the value of the data feature is determined as a key feature;
[0028] Based on the key features, determine the set of key features for the node.
[0029] Optionally, determining the global model based on the cluster key feature set includes:
[0030] The global key feature set is determined by taking the intersection of several cluster key feature sets.
[0031] Based on the global key feature set, feature masking is performed on the data of the participating nodes to determine the global model.
[0032] Based on the same inventive concept, embodiments of this application also provide a multi-task learning device based on federated learning, including:
[0033] The first determining module is configured to determine the participating nodes;
[0034] The clustering module is configured to cluster the participating nodes to determine several clusters.
[0035] The first computation module is configured to perform computation through federated learning based on several said clusters to determine a global cluster model.
[0036] The second calculation module is configured to calculate, based on the cluster global model and using the SHAP framework, the cluster key feature set of any of the clusters.
[0037] The second determining module is configured to determine a global model based on the cluster key feature set;
[0038] The training module is configured to train the global model based on any of the said clusters to determine the cluster global model of any of the said clusters; wherein, multiple said clusters are used to implement multi-task learning.
[0039] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the multi-task learning method based on federated learning as described in any of the above.
[0040] Based on the same inventive concept, embodiments of this application also provide a non-transitory computer-readable storage medium storing computer instructions, characterized in that the computer instructions are used to cause a computer to execute any of the above-described multi-task learning methods based on federated learning.
[0041] As can be seen from the above, this application provides a multi-task learning method, apparatus, electronic device, and storage medium based on federated learning. The multi-task learning method based on federated learning includes: determining participating nodes; clustering the participating nodes to determine several clusters; calculating a global cluster model using federated learning based on the several clusters; calculating a cluster key feature set for any cluster using the SHAP framework based on the global cluster model; determining a global model based on the cluster key feature set; and training the global model in any cluster to determine a global cluster model for any cluster. Multiple clusters are used to implement multi-task learning. This application aggregates participating nodes with similar device performance and data distribution into the same class through clustering, allowing participating nodes within the same cluster to be trained together, avoiding the influence of participating nodes with different device types or data distributions, and mitigating problems caused by device heterogeneity. Through feature masking, only the relevant parameters of features considered key by all nodes are trained, mitigating problems caused by data heterogeneity. Attached Figure Description
[0042] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0043] Figure 1 This is a schematic diagram of the multi-task learning method based on federated learning in an embodiment of this application;
[0044] Figure 2 This is a schematic diagram illustrating the implementation of the multi-task learning method based on federated learning in this application.
[0045] Figure 3 This is a schematic diagram of the structure of a multi-task learning device based on federated learning according to an embodiment of this application;
[0046] Figure 4 This is a schematic diagram of the electronic device structure according to an embodiment of this application. Detailed Implementation
[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.
[0048] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0049] As described in the background section, in heterogeneous transmission networks, existing heterogeneity issues lead to decreased model accuracy and excessively long training times in traditional federated learning. Furthermore, existing solutions for heterogeneity problems suffer from limited scope, addressing only individual issues like decreased accuracy or excessive training time, rather than simultaneously resolving both. This limits their application scenarios.
[0050] Specifically, in heterogeneous transmission networks, federated learning currently faces two major challenges: First, the model accuracy problem caused by data heterogeneity: data from different devices is not independently and identically distributed. For example, people from different regions have different facial features and accents. From a data protection perspective, federated learning cannot achieve data sharing between devices, increasing the difficulty of perceiving data heterogeneity. Second, the problem of excessively long training times caused by device heterogeneity: tasks on mobile or edge computing devices are only executed when the device is idle, charging, or connected to an uncharged network. Furthermore, the connection between the device and the remote server may often be unavailable or slow. Many researchers have mitigated data heterogeneity by using federated multi-task learning to study the potential relationships between tasks; in addition, many researchers have used asynchronous updates to alleviate the problem of excessively long waiting times caused by device differences. However, currently, no research has effectively solved both of these problems simultaneously, failing to meet the needs of user scenarios and computational requirements.
[0051] In view of this, embodiments of this application provide a multi-task learning method, apparatus, electronic device, and storage medium based on federated learning to solve the problems of decreased model accuracy and excessively long training time in traditional federated learning caused by heterogeneity issues in the prior art.
[0052] like Figure 1 As shown in the embodiment of this application, the multi-task learning method based on federated learning includes:
[0053] Step 102: Determine the participating nodes;
[0054] Step 104: Cluster the participating nodes to determine several clusters;
[0055] Step 106: Based on the aforementioned clusters, calculate and determine the global cluster model through federated learning;
[0056] Step 108: Based on the global cluster model, calculate using the SHAP framework to determine the cluster key feature set of any of the clusters.
[0057] Step 110: Determine the global model based on the cluster key feature set;
[0058] Step 112: Train the global model based on any of the clusters to determine the cluster global model of any of the clusters; wherein, multiple clusters are used to implement multi-task learning.
[0059] This application includes a central server and several participating nodes. In step 102, the central server first determines the number and type of participating nodes, and determines the node data calculation model for the participating nodes based on the number and type of participating nodes, and sends the parameters of the node data calculation model to the participating nodes.
[0060] Furthermore, in step 102, participating nodes can be understood as terminal devices with computing or data processing capabilities. The data processed by the devices can be of various types, such as image data or formatted data for analysis, or for predicting health events such as hypoglycemia or heart disease risk caused by wearable devices; or for detecting theft in smart homes. Therefore, the types of participating nodes are also diverse, and the types of participating nodes vary depending on the application scenario of the terminal device.
[0061] In step 104, before clustering the participating nodes to determine several clusters, a node training model for each participating node is first determined. This model is then trained using the local data of the participating nodes, and the training time and model weights are determined through this training. Further, based on the training time and model weights, the participating nodes are clustered using the KMeans algorithm to determine several clusters. Furthermore, the amount of data and the data processing capabilities of the terminal devices both affect the training time during the training process.
[0062] In some alternative implementations, training time is equivalent to device performance, that is, the shorter the training time, the better the device performance, i.e., the better the device's training capability; model weights are equivalent to data distribution, that is, the closer the model weight values are to the data distribution, the more similar they are; the closer the training time and weight values of any two, three or more participating nodes are, the more similar the participating nodes are, and similar participating nodes are clustered to determine several clusters.
[0063] It should be noted that when training the model via node training, a threshold for the number of training iterations is first set. Training stops when the number of training iterations exceeds the preset threshold, and the training time and the weights of the most recently trained model are recorded. The preset number of training iterations can be set according to actual conditions, for example, it can be set to 50. When it exceeds 50, for example, when the number of training iterations reaches 51, training stops, and the current training time and the weights of the most recently trained model are recorded. Of course, in actual use, the threshold for the number of training iterations may not necessarily be set to 50 due to differences in data volume and / or device data processing capabilities; it can be adjusted according to actual conditions, and this application does not impose any limitations on this.
[0064] In some alternative implementations, the model weights do not necessarily have to be the model weights from the last training session as the clustering basis. The largest or smallest model weights can also be selected as the clustering basis, or the average of the weights from multiple training sessions can be chosen.
[0065] In some alternative implementations, the K-means algorithm is called the K_means algorithm and is used for clustering. Furthermore, clustering is a form of unsupervised learning. Clustering refers to discovering relationships between data objects without prior "labels," grouping the data into groups, each group being called a "cluster." The greater the similarity within a group and the greater the difference between groups, the better the clustering effect. In other words, the higher the similarity of objects within a cluster and the lower the similarity between objects in different clusters, the better the clustering effect.
[0066] In some optional implementations, after the participating nodes determine the training time and model weights, they upload the training time and model weights to the central server. The central server receives the training time and model weights uploaded by all participating nodes and clusters the participating nodes into multiple clusters based on the training time and model weights. At this time, heterogeneous nodes are divided into different clusters, and the global model and training time of each cluster are different; however, the training time and model weights of each participating node within a cluster are the same, and the training time and model weights correspond to the device performance and data distribution, respectively. Therefore, they can be regarded as having similar data distribution and device performance.
[0067] In some optional implementations, after determining the clusters, any participating node in the cluster is selected as the cluster center. Federated learning is performed within each cluster for iterative training. Once the preset number of iterations is reached, the global cluster model is determined, denoted as w. k Where w is the global model and k is the cluster index, k = 1, 2, 3, ..., n. It should be noted that after clustering the participating nodes, multiple clusters can be determined and ordered. Before iterative training, an iteration threshold is first set. When the number of iterations reaches the preset threshold, the global cluster model w is output. k .
[0068] In some optional implementations, participating nodes determine their feature values based on the iteratively determined cluster global model. Specifically, this includes: analyzing the cluster global model using the SHAP framework to determine the data features of participating nodes; calculating the feature values of participating nodes based on the data features and the cluster global model; further comparing the feature values of participating nodes with a preset threshold; if a feature value is greater than the threshold, the data feature of the participating node corresponding to that feature value is determined as a key feature. Simultaneously, the data features of participating nodes in each cluster whose feature values are greater than the preset threshold are grouped into one set, thus obtaining multiple sets of key node features. The threshold can be set according to actual calculation conditions; different computational loads or calculation methods may result in different thresholds, which are not specifically limited here. Further, the union of several sets of key node features is taken to further determine the cluster key feature set. Figure 2 The diagram shown illustrates the application process of the multi-task learning method based on federated learning in this application. Figure 2 It can be seen that in the initial stage, the participating nodes are in an unordered state and their arrangement is also very random. The K-means algorithm in this application is used to cluster the participating nodes, determining several clusters, as follows: Figure 2 The diagram shows clusters 1, 2, and 3. However, in some other implementations, the number of clusters can be infinite, not necessarily three. The number of clusters depends on the types of nodes. It is understood that when there are three types of nodes, all nodes can be divided into three clusters; when there are ten or twenty types of nodes, all nodes can be divided into ten or twenty clusters; and when there are more types of nodes, there can be more clusters.
[0069] Furthermore, after determining the key features of the clusters, calculations can be performed based on these key features. The central server performs an intersection operation on the key features of all clusters to obtain the global key features.
[0070] In some optional implementations, after determining the global key feature set, feature masking techniques are used to mask features in each participating node that are not included in the global key feature set. Training is then performed using data other than the masked data, thereby preventing each participating node from being affected by features unrelated to it and mitigating the problems caused by data heterogeneity. In the final clustering training stage, each cluster trains independently without waiting for participating nodes in other slower clusters, reducing the waiting time for each participating node.
[0071] In some optional implementations, the global model of any cluster is determined by training the global model within any of the clusters. Specifically, this includes: each node in each cluster training the global model based on local data within the cluster, and aggregating the models within their respective clusters until the global model converges, thus determining the cluster global model. Once the cluster global model is determined, each cluster processes different types of data for different application scenarios according to its corresponding cluster global model. When different types of data are received, the cluster global models in different clusters can simultaneously train or process different types of data separately. As can be seen from the above, each cluster trains independently, without waiting for participating nodes in other, slower-computing clusters, reducing the waiting time for each participating node and mitigating the problems caused by device heterogeneity.
[0072] As can be seen from the above, this application provides a multi-task learning method, apparatus, electronic device, and storage medium based on federated learning. The multi-task learning method based on federated learning includes: determining participating nodes; clustering the participating nodes to determine several clusters; calculating a global cluster model using federated learning based on the several clusters; calculating a cluster key feature set for any cluster using the SHAP framework based on the global cluster model; determining a global model based on the cluster key feature set; and training the global model in any cluster to determine a cluster global model for any cluster. Multiple clusters are used to implement multi-task learning. This application aggregates participating nodes with similar device performance and data distribution into the same class through clustering, allowing participating nodes within the same cluster to be trained together, avoiding the influence of participating nodes with different device types or data distributions, and mitigating problems caused by device heterogeneity and data heterogeneity. Furthermore, through feature masking, only the relevant parameters of features considered key by all nodes are trained, mitigating problems caused by data heterogeneity.
[0073] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.
[0074] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0075] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides a multi-task learning device based on federated learning.
[0076] refer to Figure 3 The federated learning-based multi-task learning device includes:
[0077] The first determining module 302 is configured to determine the participating nodes;
[0078] Clustering module 304 is configured to cluster the participating nodes to determine several clusters;
[0079] The first computing module 306 is configured to perform calculations based on several clusters through federated learning to determine a global cluster model.
[0080] The second calculation module 308 is configured to perform calculations based on the cluster global model using the SHAP framework to determine the cluster key feature set of any of the clusters.
[0081] The second determining module 310 is configured to determine a global model based on the cluster key feature set;
[0082] Training module 310 is configured to train the global model based on any of the said clusters to determine the cluster global model of any of the said clusters; wherein, multiple said clusters are used to implement multi-task learning.
[0083] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.
[0084] The apparatus of the above embodiments is used to implement the corresponding multi-task learning method based on federated learning in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0085] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the multi-task learning method based on federated learning described in any of the above embodiments.
[0086] Figure 4 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.
[0087] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.
[0088] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.
[0089] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.
[0090] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0091] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.
[0092] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.
[0093] The electronic devices described above are used to implement the corresponding federated learning-based multi-task learning methods in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0094] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the multi-task learning method based on federated learning as described in any of the above embodiments.
[0095] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0096] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the multi-task learning method based on federated learning as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0097] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.
[0098] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.
[0099] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.
[0100] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.
Claims
1. A multi-task learning method based on federated learning, characterized in that, include: Determine the participating nodes; wherein the data processed by the participating nodes is image data; The participating nodes are clustered to determine several clusters, including: Determine a node training model, and train the participating nodes using the node training model; In response to determining that a preset number of training iterations has been reached, the training time and model weights are determined; Based on the training time and the model weights, the participating nodes are clustered to determine several clusters; Based on several clusters, a global cluster model is determined through federated learning. Based on the cluster global model, the SHAP framework is used to calculate and determine the set of key cluster features for any of the clusters, including: The SHAP framework is used to analyze the global cluster model to determine the data characteristics and data characteristic values of any participating node in the cluster. Based on the data feature values, a set of key node features is determined, including: In response to determining that the value of the data feature is greater than a preset threshold, the data feature corresponding to the value of the data feature is determined as a key feature; Based on the key features, determine the set of key features for the node; The cluster key feature set is determined by taking the union of several node key feature sets. Based on the cluster key feature set, determine the global model; The global model is trained on any of the clusters to determine the cluster global model of any of the clusters; wherein, multiple clusters are used to implement multi-task learning.
2. The method according to claim 1, characterized in that, The method includes: The participating nodes are clustered using the KMeans algorithm.
3. The method according to claim 1, characterized in that, The step of determining a global cluster model through federated learning based on several clusters includes: Determine the cluster center based on the clusters; The clusters are trained using the federated learning method. In response to determining that the number of training iterations has reached a preset threshold, the computational model is determined as the cluster global model.
4. The method according to claim 1, characterized in that, The step of determining the global model based on the cluster key feature set includes: taking the intersection of several cluster key feature sets to determine the global key feature set; Based on the global key feature set, feature masking is performed on the data of the participating nodes to determine the global model.
5. A multi-task learning method apparatus based on federated learning, characterized in that, include: The first determining module is configured to determine participating nodes; wherein the processing data of the participating nodes is image data; The clustering module is configured to cluster the participating nodes to determine several clusters, including: Determine a node training model, and train the participating nodes using the node training model; In response to determining that a preset number of training iterations has been reached, the training time and model weights are determined; Based on the training time and the model weights, the participating nodes are clustered to determine several clusters; The first computation module is configured to perform computation through federated learning based on several said clusters to determine a global cluster model. The second calculation module is configured to determine the set of key cluster features for any of the clusters based on the global cluster model using the SHAP framework, including: The SHAP framework is used to analyze the global cluster model to determine the data characteristics and data characteristic values of any participating node in the cluster. Based on the data feature values, a set of key node features is determined, including: In response to determining that the value of the data feature is greater than a preset threshold, the data feature corresponding to the value of the data feature is determined as a key feature; Based on the key features, determine the set of key features for the node; The cluster key feature set is determined by taking the union of several node key feature sets. The second determining module is configured to determine a global model based on the cluster key feature set; The training module is configured to train the global model based on any of the said clusters to determine the cluster global model of any of the said clusters; wherein, multiple said clusters are used to implement multi-task learning.
6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor, when executing the computer program, implements the method according to any one of claims 1 to 4.
7. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method according to any one of claims 1 to 4.