Federal learning method and device, communication equipment, storage medium and computer program product

Abstract

The invention relates to a federal learning method and device, communication equipment, a storage medium and a computer program product. The method comprises the following steps: in each round of model cooperative training, receiving local model data obtained by a plurality of data holders in the previous round of model cooperative training; obtaining the contribution degree of each data holder; obtaining first basic model data according to the local model data of each data holder and the contribution degree of each data holder; and according to the contribution degree of each data holder, determining the data holder participating in the current round of model cooperative training, and sending first basic model data to the data holder participating in the current round of model cooperative training, the first basic model data being used for the data holder participating in the current round of model cooperative training to perform the current round of model cooperative training. By adopting the method, each data holder can be promoted to actively participate in each round of model cooperative training, so that a high-quality model can be obtained.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a federated learning method, apparatus, communication device, storage medium, and computer program product. Background Technology

[0002] Federated learning, as a novel distributed machine learning approach, allows data holders to collaboratively train machine learning models without exposing their raw data. This avoids uploading massive amounts of raw data to the cloud, significantly reduces data transmission latency, saves valuable communication resources, and effectively protects users' privacy data.

[0003] However, existing federated learning methods need further improvement to enhance the motivation of data holders in the collaborative training process, thereby obtaining high-quality models. Summary of the Invention

[0004] This application provides a federated learning method, apparatus, communication device, storage medium, and computer program product.

[0005] In a first aspect, this application provides a federated learning method, the method comprising: in each round of model co-training, receiving local model data obtained by multiple data holders in the previous round of model co-training; obtaining the contribution of each data holder, the contribution being used to characterize the value of the local model data provided by the data holder in the previous round of model co-training; obtaining first basic model data based on the local model data of each data holder and the contribution of each data holder; determining the data holders participating in the current round of model co-training based on the contribution of each data holder, and sending the first basic model data to the data holders participating in the current round of model co-training, the first basic model data being used by the data holders participating in the current round of model co-training for the current round of model co-training.

[0006] In one embodiment, obtaining the contribution of each data holder includes: evaluating the contribution of each data holder and determining the contribution of each data holder based on the results of the contribution evaluation.

[0007] In one embodiment, the contribution assessment includes at least one of the following: calculating the similarity between the local model data corresponding to each data holder and the basic model data sent to each data holder in the previous round of model co-training; determining the semantic information quality of each data holder based on the data feature values ​​sent by each data holder; wherein the data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set; and determining the proportion of the number of samples in the local training sample set of each data holder to the total number of samples.

[0008] In one embodiment, the contribution of each data holder is determined based on the contribution assessment results, including: for each data holder, a weighted summation of the sub-contributions obtained from the various contribution assessments of the data holder is performed to obtain the contribution of the data holder.

[0009] In one embodiment, the data holders participating in this round of model co-training are determined based on their contribution, including: data holders whose contribution is greater than or equal to a preset contribution threshold are selected as data holders participating in this round of model co-training.

[0010] In one embodiment, determining the data holders participating in this round of model collaborative training based on the contribution of each data holder includes: summing the contributions of each data holder to obtain the total contribution; determining the contribution percentage of each data holder based on the ratio of the data holder's contribution to the total contribution; and selecting data holders whose contribution percentage is greater than or equal to a preset percentage threshold as data holders participating in this round of model collaborative training.

[0011] In one embodiment, obtaining first basic model data based on the local model data of each data holder and the contribution of each data holder includes: determining the weight of each data holder based on the contribution of each data holder; and aggregating the local model data of each data holder based on the weight of each data holder to obtain the first basic model data.

[0012] In one embodiment, the method further includes: sending training counts to data holders participating in this round of model co-training, the training counts being used to indicate the number of iterations the data holders participating in this round of model co-training perform based on the first base model data.

[0013] In one embodiment, the method further includes: determining whether the training termination condition is met during the current round of model co-training; if the training termination condition is met, sending the second basic model data obtained from the current round of model co-training to the data holder participating in the current round of model co-training, so that the data holder participating in the current round of model co-training can adjust the second basic model data based on the local training sample set to obtain fine-tuned model data.

[0014] In one embodiment, the method further includes: establishing and publishing a federated learning model co-training task; receiving participation requests from data holders who wish to participate in the model co-training task; and sending initial basic model data to the data holders corresponding to each participation request, so that the data holders corresponding to each participation request can perform model co-training based on the local training sample set and the initial basic model data to obtain local model data.

[0015] In one embodiment, the coordinator and each data holder are semantic agents; the first basic model data and the second basic model data are both semantic knowledge bases common to all semantic agents; the fine-tuned model data are personalized semantic knowledge bases formed after fine-tuning with local training sample sets.

[0016] Secondly, this application provides a federated learning method for data holders, the method comprising: in each round of model co-training, sending local model data obtained in the previous round of model co-training to a coordinator; so that the coordinator can obtain first basic model data based on the local model data sent by each data holder and the contribution of each data holder; receiving the first basic model data sent by the coordinator, and performing model co-training in the current round based on the first basic model data; wherein, the data holders are the data holders participating in the current round of model co-training determined by the coordinator based on the contribution of each data holder, and the contribution is used to characterize the value of the local model data provided by the data holder in the previous round of model co-training.

[0017] In one embodiment, the method further includes: sending at least one of data feature values ​​and the number of samples in the local training sample set to a coordinator, so that the coordinator can evaluate the contribution of the data holder based on at least one of the data feature values ​​and the number of samples, and determine the contribution of the data holder; wherein the data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set.

[0018] In one embodiment, the method further includes: extracting features from each sample data in the local training sample set to obtain a feature vector corresponding to each sample data; aligning the feature vectors corresponding to each sample data, and obtaining a feature matrix of the local training sample set based on the aligned feature vectors corresponding to each sample data; and using the product of the feature matrix and the transpose of the feature matrix as a variable of the sqrt function to calculate the data feature value through the sqrt function.

[0019] In one embodiment, the method further includes: receiving the number of training iterations sent by the coordinator; correspondingly, performing this round of model co-training based on the first basic model data includes: performing this round of model co-training based on the first basic model data and the number of training iterations.

[0020] In one embodiment, the method further includes: receiving second basic model data sent by a coordinator, and adjusting the second basic model data based on a local training sample set to obtain fine-tuned model data.

[0021] In one embodiment, the second basic model data is sent by the coordinator when the training termination condition is met.

[0022] In one embodiment, the method further includes: performing a performance evaluation on the fine-tuned model data; if the fine-tuned model data meets the performance requirements, then encapsulating the fine-tuned model data to obtain encapsulated fine-tuned model data.

[0023] In one embodiment, the method further includes: sending the fine-tuned model data to an affiliate of the data holder; and / or sending the fine-tuned model data to a trusted party trusted by the data holder.

[0024] In one embodiment, the method further includes: sending a participation request to the coordinator after the coordinator publishes the federated learning model co-training task; receiving initial basic model data sent by the coordinator; and performing model co-training based on the local training sample set and the initial basic model data to obtain local model data.

[0025] Thirdly, this application provides a federated learning device, disposed at a coordinator, the device comprising: a receiving module, configured to receive local model data obtained by multiple data holders in the previous round of model co-training in each round of model co-training; a first acquisition module, configured to acquire the contribution of each data holder, the contribution being used to characterize the value of the local model data provided by the data holder in the previous round of model co-training; a second acquisition module, configured to acquire first basic model data based on the local model data of each data holder and the contribution of each data holder; and a sending module, configured to determine the data holders participating in the current round of model co-training based on the contribution of each data holder, and send the first basic model data to the data holders participating in the current round of model co-training, the first basic model data being used by the data holders participating in the current round of model co-training for model co-training.

[0026] Fourthly, this application provides a federated learning device, disposed at a data holder, the device comprising: a sending module, configured to send local model data obtained in the previous round of model co-training to a coordinator in each round of model co-training, so that the coordinator can obtain first basic model data based on the local model data sent by each data holder and the contribution of each data holder; and a receiving module, configured to receive the first basic model data sent by the coordinator and perform model co-training in the current round based on the first basic model data; wherein, the data holder is the data holder participating in the current round of model co-training determined by the coordinator based on the contribution of each data holder, and the contribution is used to characterize the value of the local model data provided by the data holder in the previous round of model co-training.

[0027] Fifthly, this application provides a communication device including a transceiver, a memory, and a processor, wherein the memory stores a computer program, and the processor executes the computer program to control the transceiver to implement the steps of the method described in any one of the first aspects above.

[0028] In a sixth aspect, this application provides a communication device including a transceiver, a memory, and a processor, wherein the memory stores a computer program, and the processor executes the computer program to control the transceiver to implement the steps of the method described in any one of the second aspects above.

[0029] In a seventh aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the first or second aspects above.

[0030] Eighthly, this application also provides a chip including programmable logic circuitry and / or program instructions, which, when the chip is running, can perform the steps of the method described in any one of the first or second aspects above.

[0031] Ninthly, this application also provides a computer program product, including a computer program, characterized in that, when executed by a processor, the computer program implements the steps of the method described in any one of the first or second aspects above.

[0032] The aforementioned federated learning methods, devices, communication equipment, storage media, and computer program products involve a coordinator receiving local model data from multiple data holders in each round of model co-training. This local model data, obtained from the previous round, is used to assess the value of the data provided by each data holder in the previous round. Based on this value, a first basic model data is generated. Then, based on the data holders' contributions, the data holders participating in the current round of model co-training are determined, and the first basic model data is sent to them. This data is used for the participating data holders in the current round of model co-training. This method of determining participants based on their contributions encourages active participation from all data holders in each round of model co-training, leading to higher-quality models. Attached Figure Description

[0033] Figure 1 This is a diagram illustrating the application environment of the federated learning method in one embodiment;

[0034] Figure 2This is a flowchart illustrating the federated learning method in one embodiment;

[0035] Figure 3 This is a schematic diagram of the signaling interaction process during the initiation of federated learning in one embodiment.

[0036] Figure 4 This is a flowchart illustrating a federated learning method in another embodiment;

[0037] Figure 5 This is a schematic diagram of the signaling interaction process of a federated learning method in one embodiment;

[0038] Figure 6 This is a structural block diagram of a federated learning device in one embodiment;

[0039] Figure 7 This is a structural block diagram of a federated learning device in another embodiment;

[0040] Figure 8 This is an internal structure diagram of a communication device in one embodiment;

[0041] Figure 9 This is a schematic structure of a chip in one embodiment. Detailed Implementation

[0042] 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.

[0043] Before introducing the specific embodiments of this application, the technical terms involved in this application will be explained:

[0044] The coordinator, data owner, and model user can all be semantic information agents, which are three roles in federated learning training.

[0045] The coordinator coordinates the model training for the data holders and obtains the semantic agent of the final trained model.

[0046] The data holder is a semantic intelligent agent that owns the local data.

[0047] Affiliated parties are entities that have some kind of association or cooperative relationship with the data holder, which may be based on business cooperation, technology sharing, common interests, or legal agreements.

[0048] Trusted parties (with similar abilities) refer to entities that are considered highly trustworthy by data holders, and these entities have demonstrated excellent performance in areas such as data security, privacy protection, technical capabilities, and cooperation history.

[0049] Model users are semantic agents who use the basic model or fine-tuned models, such as coordinators, data holders, affiliates, and trusted parties.

[0050] Data holders, leveraging pre-trained personalized models, can accurately capture and analyze deep semantic features extracted from multimodal information, thereby achieving efficient and precise end-to-end semantic communication. Federated learning, as a novel distributed machine learning approach, allows data holders to collaboratively train machine learning models without exposing their raw data. This avoids uploading massive amounts of raw data to the cloud, significantly reducing data transmission latency, saving valuable communication resources, and effectively protecting user privacy.

[0051] However, while federated learning is a promising machine learning method, the participation of data holders in the process consumes significant private resources, such as computing power, bandwidth, and data. Due to a lack of incentive mechanisms, data holders are unwilling to sacrifice their various equipment resources to assist in federated learning. This selfishness of data holders limits the application of federated learning in scenarios such as mobile networks and mobile edge computing. Therefore, further improvements are needed, such as designing reasonable incentive mechanisms to enhance the enthusiasm of data holders in the collaborative training process, thereby obtaining high-quality models.

[0052] Current incentive mechanisms for federated learning primarily target data holders with similar computing power. However, in existing technologies, some data holders with low computing power also want to participate in semantic communication systems, but they cannot complete effective federated learning model training, thus failing to obtain personalized models. This results in low accuracy or even no semantic information extraction at all. Therefore, how to utilize incentive mechanisms to enable these data holders with low computing power to obtain high-quality personalized models is a key issue that needs to be considered.

[0053] Therefore, it is necessary to propose effective technical means to solve the above problems. The technical solution of this application and how it solves the above technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0054] It should be noted that the beneficial effects or technical problems solved by the embodiments of this application are not limited to this one, but may also be other implicit or related problems. For details, please refer to the description of the embodiments below.

[0055] Figure 1 This is a schematic diagram illustrating an application scenario of a federated learning method provided in an embodiment of this application. For example... Figure 1 As shown, this scenario involves multiple data holders, a coordinator, and affiliated and trusted parties connected to each data holder. Data transmission occurs between the data holders and the coordinator via a network, as do data transmission between the data holders and their affiliated and trusted parties.

[0056] In one embodiment, such as Figure 2 As shown, a federated learning method is provided, which can be applied to... Figure 1 Taking the coordinator in the example, the steps include:

[0057] Step 201: In each round of model co-training, receive local model data obtained by multiple data holders in the previous round of model co-training.

[0058] Optionally, in each round of model co-training, multiple data holders will send the local model data obtained in the previous round of model co-training to the coordinator, so that the coordinator will receive the local model data obtained by multiple data holders in the previous round of model co-training.

[0059] Step 202: Obtain the contribution of each data holder. The contribution is used to characterize the value of the local model data provided by the data holder in the previous round of model co-training.

[0060] Optionally, each data holder's contribution can be sent directly to the coordinator, thus the coordinator receives the contribution information from each data holder. Alternatively, the coordinator can evaluate the contribution of each data holder and determine their contribution based on the evaluation results.

[0061] Step 203: Obtain the first basic model data based on the local model data of each data holder and the contribution of each data holder.

[0062] The first basic model data is the global model data obtained by the coordinator after aggregating the local model data of each data holder.

[0063] Optionally, the weight of each data holder is determined based on their contribution; and the local model data of each data holder is aggregated based on their weight to obtain the first basic model data.

[0064] Step 204: Based on the contribution of each data holder, determine the data holders participating in this round of model co-training, and send the first basic model data to the data holders participating in this round of model co-training. The first basic model data is used by the data holders participating in this round of model co-training to carry out this round of model co-training.

[0065] Optionally, the data holders participating in this round of model collaborative training can be determined based on the size of their contribution or the percentage of their contribution.

[0066] In one possible implementation, data holders whose contribution is greater than or equal to a preset contribution threshold are designated as data holders participating in this round of model collaborative training.

[0067] In another possible implementation, the contributions of each data holder are summed to obtain the total contribution; for each data holder, the contribution percentage of the data holder is determined based on the ratio of the data holder's contribution to the total contribution; data holders whose contribution percentage is greater than or equal to a preset percentage threshold are selected as data holders to participate in this round of model collaborative training.

[0068] After identifying the data holders participating in this round of model co-training, the coordinator sends the first set of basic model data to them. In other words, data holders whose contribution is less than a preset contribution threshold and whose contribution percentage is less than a preset percentage threshold are not allowed to participate in this round of model co-training.

[0069] After receiving the first base model data, the data holders participating in this round of model co-training conduct this round of model co-training. Specifically, the data holders participating in this round of model co-training use their local training sample sets to train the first base model data, calculate the local loss function, determine the learning rate used for training, and can use the mini-batch stochastic gradient descent method. By minimizing the local loss function of the data holder on its local training sample set, the trained local model data is obtained and uploaded to the coordinator.

[0070] The local loss function of data holder i participating in this round of model co-training for:

[0071]

[0072] in, This represents the local model data of data holder i. This represents the local training sample set of data holder i. This represents the j-th sample data of data holder i. This represents the true label of the j-th sample data of data holder i. Let represent the loss function of data holder i on the j-th sample data.

[0073] The process by which data holder i, participating in this round of model co-training, trains the first basic model data using the local training sample set is as follows:

[0074]

[0075] Where t represents the current iteration round during training. This indicates the learning rate used during training. This indicates that during training, data holder i in the current iteration is on the local training sample set. The average gradient over.

[0076] In the aforementioned federated learning method, the coordinator receives local model data from multiple data holders in each round of model co-training, based on the data from the previous round. The coordinator then calculates the contribution of each data holder, representing the value of their local model data provided in the previous round. Based on the local model data and contribution of each data holder, the coordinator obtains the first basic model data. Finally, based on the contribution of each data holder, the coordinator determines the data holders participating in the current round of model co-training and sends the first basic model data to them. This first basic model data is used by the data holders participating in the current round of model co-training. This method of determining the data holders based on their contribution encourages active participation from each data holder in each round of model co-training, leading to a higher quality model.

[0077] In one embodiment, step 202 above, which obtains the contribution of each data holder, includes: evaluating the contribution of each data holder and determining the contribution of each data holder based on the evaluation results.

[0078] Contribution assessment includes at least one of the following:

[0079] Calculate the similarity between the local model data corresponding to each data holder and the basic model data sent to each data holder in the previous round of model co-training;

[0080] The semantic information quality of each data holder is determined based on the data feature values ​​sent by each data holder; whereby the data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set.

[0081] Determine the proportion of samples in each data holder's local training sample set to the total number of samples.

[0082] This involves calculating the similarity between the local model data corresponding to each data holder and the basic model data sent to each data holder in the previous round of model co-training. This yields the first sub-contribution of each data holder, which can also be referred to as the update quality of the model provided by each data holder. Specifically, the similarity can be cosine similarity.

[0083] Specifically, the first sub-contribution of data holder i for:

[0084]

[0085] in, This represents the local model data of data holder i. This represents the basic model data sent to each data holder in the previous round of model co-training.

[0086] Based on the data feature values ​​sent by each data holder, the semantic information quality of each data holder is determined, and the second sub-contribution of each data holder can be obtained. Specifically, the data feature values ​​sent by the data holders are obtained by extracting features from each sample data in the local training sample set, obtaining the corresponding feature vectors for each sample data, aligning the feature vectors, and obtaining the feature matrix of the local training sample set based on the aligned feature vectors. The product of the feature matrix and its transpose is then used as the variable for the sqrt function calculation.

[0087] The coordinator can directly determine the semantic information quality of each data holder by using the data feature values ​​sent by each data holder, or the coordinator can process the data feature values ​​sent by each data holder to obtain the semantic information quality of each data holder.

[0088] Taking the example of the coordinator directly determining the semantic information quality of each data holder by using the data feature values ​​sent by each data holder, the second sub-contribution of data holder i... for:

[0089]

[0090] in, This represents the aligned feature vector. Each sample data has d features. This is the Gram determinant value measurement function.

[0091] Determining the proportion of samples in each data holder's local training sample set to the total number of samples yields each data holder's third sub-contribution, also known as the amount of semantic information provided by each data holder from their local training sample set. The number of samples in each data holder's local training sample set is the amount sent by each data holder to the coordinator.

[0092] Specifically, the third sub-contribution of data holder i for:

[0093]

[0094] in, This represents the number of samples in the local training sample set of data holder i. This represents the total number of samples from all data holders.

[0095] Optionally, the contribution of each data holder can be determined based on the results of the contribution assessment, including: for each data holder, the contribution can be the data holder's first sub-contribution, second sub-contribution, or third sub-contribution, or the contribution can be the sum of any two or all three of the data holder's first, second, and third sub-contributions.

[0096] In one embodiment, the contribution of each data holder is determined based on the contribution assessment results, including: for each data holder, a weighted summation of the sub-contributions obtained from the various contribution assessments of the data holder is performed to obtain the contribution of the data holder.

[0097] Specifically, the contribution of data holder i for:

[0098]

[0099] in, , , This is a weighting parameter that can be dynamically adjusted.

[0100] In this embodiment, the sub-contributions obtained from the evaluation of various contributions of the data holder are weighted and summed to realize the contribution of the data holder from multiple perspectives, thus making the contribution of the data holder more accurate.

[0101] In one embodiment, such as Figure 2 As shown, a method for obtaining first basic model data is provided. Step 103 above obtains the first basic model data based on the local model data of each data holder and the contribution of each data holder, including:

[0102] Step 201: Determine the weight of each data holder based on their contribution.

[0103] Optionally, for each data holder, the ratio of the data holder's contribution to the sum of the contributions of all data holders can be used as the data holder's weight.

[0104] Step 202: Aggregate the local model data of each data holder according to the weight of each data holder to obtain the first basic model data.

[0105] The objective function of the first basic model for:

[0106]

[0107] Where N represents the number of data holders; This represents the total contribution of all data holders; This represents the weight of data holder i, which measures the contribution of data holder i to the first basic model. The first basic model is obtained by aggregating the local model data from different data holders by minimizing the weighted average of the local loss function. .

[0108] In this embodiment, by determining the weight of each data holder based on their contribution, and then aggregating the local model data of each data holder based on their weight, the quality of the first basic model data can be improved.

[0109] In one embodiment, the method further includes: sending training counts to data holders participating in this round of model co-training, the training counts being used to indicate the number of iterations the data holders participating in this round of model co-training perform based on the first base model data.

[0110] In other words, in addition to sending the first basic model data, the coordinator also sends the number of training iterations to the data holders participating in this round of model co-training. After receiving the first basic model data and the number of training iterations, the data holders participating in this round of model co-training will conduct this round of model co-training based on the first basic model data and the number of training iterations, that is, to perform the same iterative training on the first basic model data as the number of training iterations.

[0111] It should be noted that the number of training sessions sent to different data holders participating in this round of model collaborative training may be the same or different, such as determining the number of training sessions based on the data holder's contribution.

[0112] In this embodiment, by sending training times to the data holders participating in this round of model co-training, the number of times the data holders train the first basic model data is specified, thereby achieving flexible control over the model training of the data holders.

[0113] In one embodiment, the method further includes: determining whether the training termination condition is met during the current round of model co-training; if the training termination condition is met, sending the second basic model data obtained from the current round of model co-training to the data holder participating in the current round of model co-training, so that the data holder participating in the current round of model co-training can adjust the second basic model data based on the local training sample set to obtain fine-tuned model data.

[0114] The training termination condition can be reaching the number of co-training iterations, the loss value of the second basic model data reaching the loss threshold, or other training termination conditions. Here, no limit is placed on the training termination condition.

[0115] When the training termination condition is met, the second basic model data obtained from this round of model co-training is sent to the data holders who participated in this round of model co-training. Upon receiving the second basic model data, the data holders will adjust it based on their local training sample set. For example, they may freeze the bottom-level weights of the second basic model data, unfreeze and train the top-level weights, and set appropriate hyperparameters such as learning rate, batch size, and training epochs according to the characteristics and size of their local training sample set. They will then update the weights of the unfrozen layers using the backpropagation algorithm, ultimately obtaining the fine-tuned model data. The bottom-level weights include at least the weights of the last layer, and the top-level weights include at least the weights of the first layer.

[0116] In this embodiment, by allowing data holders to adjust the trained second basic model data using local training sample sets, i.e. model fine-tuning, this method ensures that the model can better adapt to the specific needs and environment of users, thereby providing higher performance in specific application scenarios and further improving the adaptability and personalization level of the model.

[0117] After receiving the fine-tuned model data, the data holders participating in this round of model co-training will evaluate the performance of the fine-tuned model data. If the fine-tuned model data meets the performance requirements, it will be encapsulated to obtain the encapsulated fine-tuned model data.

[0118] Specifically, the data holders participating in this round of model co-training determine whether the performance of the fine-tuned model data is greater than or equal to a preset performance threshold; if so, the performance requirements are met. Then, the fine-tuned model data is packaged into a deployable format, such as an API (Application Programming Interface) or a Docker container.

[0119] After receiving the packaged fine-tuned model data, the data holders participating in this round of model co-training will send the fine-tuned model data to their affiliates; and / or, to a trusted party trusted by the data holder. Specifically, by using an incentive mechanism to send the fine-tuned model data to the data holder's affiliates; and / or, to a trusted party trusted by the data holder, this not only solves the problem of resource-constrained parties being unable to directly participate in model training, but also indirectly obtains training results by sharing the fine-tuned model data, achieving efficient model utilization, optimizing overall resource allocation, and improving model training efficiency.

[0120] In this context, an affiliate refers to an entity that has some kind of association or cooperative relationship with the data holder. This relationship may be based on business cooperation, technology sharing, common interests, or legal agreements. Affiliates typically also possess certain data processing capabilities, technical resources, or datasets, which may be similar to or complementary to the data holder's resources. Within the framework of model fine-tuning and data sharing, affiliates receive fine-tuned model data provided by the data holder through incentive mechanisms and may utilize this data for further training, validation, or optimization to improve their own model performance or develop new applications.

[0121] Trusted parties (with similar capabilities) refer to entities considered highly trustworthy by data holders. These entities demonstrate excellence in data security, privacy protection, technical capabilities, and a history of cooperation, thus being granted authority to process sensitive or advanced model data. Unlike affiliates, the definition of a trusted party emphasizes a high level of trust, rather than merely the existence of a cooperative relationship. During model fine-tuning and data sharing, trusted parties may be responsible for performing specific data processing tasks, such as model evaluation, security audits, or data anonymization, to ensure the legitimacy and security of data flow. Simultaneously, trusted parties may also leverage their expertise and technical resources to provide valuable feedback and suggestions to data holders, helping to optimize model performance or improve data sharing mechanisms.

[0122] It should be noted that the coordinator and each data holder are semantic agents; the first basic model data and the second basic model data are semantic knowledge bases common to all semantic agents; the fine-tuned model data are personalized semantic knowledge bases formed after fine-tuning with local training sample sets.

[0123] In one embodiment, such as Figure 3 As shown, this application also provides a schematic diagram of the signaling interaction process at the start of federated learning, including the following steps:

[0124] Step 301: Establish and publish the federated learning model collaborative training task.

[0125] Step 302: Receive participation requests from data holders who are willing to participate in the model co-training task.

[0126] Step 303: Send the initial basic model data to the data holders corresponding to each participating request, so that each data holder can perform model co-training based on the local training sample set and the initial basic model data to obtain local model data.

[0127] Optionally, the coordinator establishes and publishes a federated learning model co-training task. Data holders who wish to participate in the model co-training task send participation requests to the coordinator. The coordinator receives the participation requests from each data holder who wishes to participate in the model co-training task and sends the initial basic model data to the data holder corresponding to each participation request. After receiving the initial basic model data, the data holder performs model co-training based on the local training sample set and the initial basic model data to obtain local model data, and sends the local model data to the coordinator. This process can be called the first round of model co-training.

[0128] In this embodiment, by allowing all willing data holders to participate in the first round of training, not only is the diversity of participants greatly increased, but also, through this broad participation, data features from different sources can be quickly integrated, thereby improving the generalization and accuracy of the model.

[0129] In one embodiment, such as Figure 4 As shown, a federated learning method is provided, which can be applied to... Figure 1 Taking data holders as an example, the explanation includes the following steps:

[0130] Step 401: In each round of model co-training, send the local model data obtained in the previous round of model co-training to the coordinator so that the coordinator can obtain the first basic model data based on the local model data sent by each data holder and the contribution of each data holder.

[0131] Step 402: Receive the first basic model data sent by the coordinator, and perform this round of model collaborative training based on the first basic model data; wherein, the data holder is the data holder participating in this round of model collaborative training determined by the coordinator according to the contribution of each data holder, and the contribution is used to characterize the value of the local model data provided by the data holder in the previous round of model collaborative training.

[0132] In one embodiment, the method further includes: sending at least one of data feature values ​​and the number of samples in the local training sample set to a coordinator, so that the coordinator can evaluate the contribution of the data holder based on at least one of the data feature values ​​and the number of samples, and determine the contribution of the data holder; wherein the data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set.

[0133] In one embodiment, the method further includes: extracting features from each sample data in the local training sample set to obtain a feature vector corresponding to each sample data; aligning the feature vectors corresponding to each sample data, and obtaining a feature matrix of the local training sample set based on the aligned feature vectors corresponding to each sample data; and using the product of the feature matrix and the transpose of the feature matrix as a variable of the sqrt function to calculate the data feature value through the sqrt function.

[0134] In one embodiment, the method further includes: receiving the number of training iterations sent by the coordinator; correspondingly, performing this round of model co-training based on the first basic model data includes: performing this round of model co-training based on the first basic model data and the number of training iterations.

[0135] In one embodiment, the method further includes: receiving second basic model data sent by a coordinator, and adjusting the second basic model data based on a local training sample set to obtain fine-tuned model data.

[0136] In one embodiment, the second basic model data is sent by the coordinator when the training termination condition is met.

[0137] In one embodiment, the method further includes: performing a performance evaluation on the fine-tuned model data; if the fine-tuned model data meets the performance requirements, then encapsulating the fine-tuned model data to obtain encapsulated fine-tuned model data.

[0138] In one embodiment, the method further includes: sending the fine-tuned model data to an affiliate of the data holder; and / or sending the fine-tuned model data to a trusted party trusted by the data holder.

[0139] In one embodiment, the method further includes: sending a participation request to the coordinator after the coordinator publishes the federated learning model co-training task; receiving initial basic model data sent by the coordinator; and performing model co-training based on the local training sample set and the initial basic model data to obtain local model data.

[0140] The specific implementation of the federated learning method embodiment on the data holder side is described in detail in the federated learning method embodiment on the coordinator side, and will not be repeated here.

[0141] In summary, as Figure 5 As shown, this application also provides a schematic diagram of the signaling interaction process for a federated learning method, including the following steps:

[0142] Step 501: In each round of model co-training, the data holder sends the local model data obtained in the previous round of model co-training to the coordinator.

[0143] Step 502: The coordinator obtains the contribution of each data holder and determines the data holders participating in this round of model collaborative training based on the contribution of each data holder.

[0144] Step 503: Obtain the first basic model data based on the local model data of each data holder and the contribution of each data holder.

[0145] Step 504: The coordinator sends the first basic model data to the data holders participating in this round of model co-training.

[0146] Step 505: The data holder performs this round of model co-training based on the first basic model data.

[0147] Step 506: Repeat steps 502-505. After the training termination condition is met, both the coordinator and the data holder obtain the second basic model data.

[0148] Step 507: The data holder adjusts the second basic model data to obtain the fine-tuned data model.

[0149] Step 508: The data holder sends the fine-tuned data model to the affiliate / trusted party.

[0150] It should be understood that, although Figure 2-5 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2-5At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0151] In one embodiment, such as Figure 6 As shown, a federated learning device 600 is provided, disposed in a coordinator. The federated learning device 600 includes: a receiving module 601, a first acquisition module 602, a second acquisition module 603, and a sending module 604, wherein:

[0152] The receiving module 601 is used to receive local model data obtained by multiple data holders in the previous round of model co-training in each round of model co-training.

[0153] The first acquisition module 602 is used to acquire the contribution of each data holder. The contribution is used to characterize the value of the local model data provided by the data holder in the previous round of model co-training.

[0154] The second acquisition module 603 is used to acquire the first basic model data based on the local model data of each data holder and the contribution of each data holder;

[0155] The sending module 604 is used to determine the data holders participating in this round of model collaborative training based on the contribution of each data holder, and to send the first basic model data to the data holders participating in this round of model collaborative training. The first basic model data is used by the data holders participating in this round of model collaborative training to carry out this round of model collaborative training.

[0156] In one embodiment, the first acquisition module 602 is specifically used to evaluate the contribution of each data holder and determine the contribution of each data holder based on the results of the contribution evaluation.

[0157] In one embodiment, the contribution assessment includes at least one of the following: calculating the similarity between the local model data corresponding to each data holder and the basic model data sent to each data holder in the previous round of model co-training; determining the semantic information quality of each data holder based on the data feature values ​​sent by each data holder; wherein the data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set; and determining the proportion of the number of samples in the local training sample set of each data holder to the total number of samples.

[0158] In one embodiment, the first acquisition module 602 is specifically used to perform a weighted summation of the sub-contributions obtained from the various contribution assessments of each data holder to obtain the contribution of each data holder.

[0159] In one embodiment, the data holders participating in this round of model co-training are determined based on their contribution, including: data holders whose contribution is greater than or equal to a preset contribution threshold are selected as data holders participating in this round of model co-training.

[0160] In one embodiment, the sending module 604 includes a determining unit, which is specifically used to sum the contributions of each data holder to obtain a total contribution; for each data holder, determine the contribution ratio of the data holder based on the ratio of the data holder's contribution to the total contribution; and select data holders whose contribution ratio is greater than or equal to a preset ratio threshold as data holders participating in this round of model collaborative training.

[0161] In one embodiment, the determining unit is further configured to determine the weight of each data holder based on the contribution of each data holder; and to aggregate the local model data of each data holder based on the weight of each data holder to obtain the first basic model data.

[0162] In one embodiment, the sending module 604 is further configured to send the training count to the data holder participating in the current round of model co-training, the training count being used to indicate the number of iterations the data holder participating in the current round of model co-training performs based on the first basic model data.

[0163] In one embodiment, the sending module 604 is further configured to determine whether the training termination condition is met in the current round of model co-training; if the training termination condition is met, the second basic model data obtained in the current round of model co-training is sent to the data holder participating in the current round of model co-training, so that the data holder participating in the current round of model co-training can adjust the second basic model data based on the local training sample set to obtain fine-tuned model data.

[0164] In one embodiment, the federated learning apparatus further includes a publishing module for establishing and publishing a model co-training task for federated learning; a receiving module 601 for receiving participation requests sent by data holders who intend to participate in the model co-training task; and a sending module 604 for sending initial basic model data to the data holders corresponding to each participation request, so that the data holders corresponding to each participation request can perform model co-training based on their local training sample set and the initial basic model data to obtain local model data.

[0165] In one embodiment, the coordinator and each data holder are semantic agents; the first basic model data and the second basic model data are both semantic knowledge bases common to all semantic agents; the fine-tuned model data are personalized semantic knowledge bases formed after fine-tuning with local training sample sets.

[0166] In one embodiment, such as Figure 7 As shown, a federated learning device 700 is provided, disposed at a data holder. The federated learning device 700 includes: a sending module 701 and a receiving module 702, wherein:

[0167] The sending module 701 is used to send the local model data obtained in the previous round of model co-training to the coordinator in each round of model co-training, so that the coordinator can obtain the first basic model data based on the local model data sent by each data holder and the contribution of each data holder.

[0168] The receiving module 702 is used to receive the first basic model data sent by the coordinator and to perform the current round of model collaborative training based on the first basic model data; wherein, the data holder is the data holder who participates in the current round of model collaborative training, determined by the coordinator according to the contribution of each data holder, and the contribution is used to characterize the value of the local model data provided by the data holder in the previous round of model collaborative training.

[0169] In one embodiment, the sending module 701 is further configured to send at least one of data feature values ​​and the number of samples in the local training sample set to the coordinator, so that the coordinator can evaluate the contribution of the data holder based on at least one of the data feature values ​​and the number of samples, and determine the contribution of the data holder; wherein, the data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set.

[0170] In one embodiment, the federated learning device 700 further includes a first processing module, configured to extract features from each sample data in the local training sample set to obtain a feature vector corresponding to each sample data; align the feature vectors corresponding to each sample data, and obtain a feature matrix of the local training sample set based on the aligned feature vectors corresponding to each sample data; and use the product of the feature matrix and the transpose of the feature matrix as a variable of the sqrt function to calculate the data feature value through the sqrt function.

[0171] In one embodiment, the receiving module 702 is further configured to receive the number of training iterations sent by the coordinator; correspondingly, the receiving module 702 is specifically configured to perform this round of model collaborative training based on the first basic model data and the number of training iterations.

[0172] In one embodiment, the receiving module 702 is further configured to receive second basic model data sent by the coordinator, and adjust the second basic model data based on the local training sample set to obtain fine-tuned model data.

[0173] In one embodiment, the second basic model data is sent by the coordinator when the training termination condition is met.

[0174] In one embodiment, the federated learning device 700 further includes a second processing module for performance evaluation of the fine-tuned model data; if the fine-tuned model data meets the performance requirements, the fine-tuned model data is encapsulated to obtain encapsulated fine-tuned model data.

[0175] In one embodiment, the sending module 701 is further configured to send the fine-tuning model data to an affiliate of the data holder; and / or to send the fine-tuning model data to a trusted party trusted by the data holder.

[0176] In one embodiment, the sending module 701 is further configured to send a participation request to the coordinator after the coordinator publishes the model co-training task of federated learning; the receiving module 702 is further configured to receive the initial basic model data sent by the coordinator; and to perform model co-training based on the local training sample set and the initial basic model data to obtain local model data.

[0177] Specific limitations regarding the federated learning apparatus can be found in the limitations of the federated learning method described above, and will not be repeated here. Each module in the aforementioned federated learning apparatus 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 the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0178] Figure 8 This is a schematic diagram of the structure of a communication device provided in an embodiment of this application. The communication device may include a transceiver 801, a memory 802, a processor 803, and at least one communication bus 804. The communication bus 804 is used to implement communication connections between components. The memory 802 may include a high-speed RAM memory, and may also include non-volatile memory (NVM), such as at least one disk storage device. The memory 802 can store various programs for performing various processing functions and implementing the method steps of this embodiment. In this embodiment, the transceiver 801 can be a radio frequency processing module or a baseband processing module in the communication device. The transceiver 801 can be coupled to the processor 803, and can perform receiving or transmitting actions under the instruction or control of the processor 803.

[0179] In one embodiment, a communication device is provided in which the processor 803, when executing a computer program, controls the transceiver 801 to implement the steps of any one of the methods described above on the coordinator side or data holder side.

[0180] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described above on the coordinator side or the data holder side.

[0181] In one embodiment, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the steps of any one of the methods described above on the coordinator side or the data holder side.

[0182] In one embodiment, a chip is provided. Figure 9 This is a schematic structural diagram of the chip according to an embodiment of this application. Figure 9 The chip 900 shown includes a processor 901, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0183] Optionally, such as Figure 9 As shown, chip 900 may further include memory 902. Processor 901 can call and run computer programs from memory 902 to implement the methods in the embodiments of this application. Memory 902 may be a separate device independent of processor 901, or it may be integrated into processor 901.

[0184] Optionally, the chip 900 may further include an input interface 903. The processor 901 can control the input interface 903 to communicate with other devices or chips; specifically, it can acquire information or data sent by other devices or chips. Optionally, the chip 900 may further include an output interface 904. The processor 901 can control the output interface 904 to communicate with other devices or chips; specifically, it can output information or data to other devices or chips.

[0185] Optionally, the chip 900 can be applied to the communication device in the embodiments of this application, and the chip 900 can implement the corresponding processes implemented by the communication device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0186] It should be understood that the chip 900 mentioned in the embodiments of this application can also be called a system-on-a-chip, system-on-a-chip, chip system, or system-on-a-chip, etc. It should be understood that the processor in the embodiments of this application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiments can be completed by the integrated logic circuit in the processor's hardware or by instructions in software form. The processor mentioned above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0187] 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 methods described above. Any references to memory, storage, 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, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0188] 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.

[0189] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. 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 patent application should be determined by the appended claims.

Claims

1. A federated learning method, characterized in that, For use by a coordinator, the method includes: In each round of model co-training, local model data obtained from multiple data holders in the previous round of model co-training is received; Obtain the contribution score of each data holder, wherein the contribution score is used to characterize the value of the local model data provided by the data holder in the previous round of model co-training; Based on the local model data of each data holder and the contribution of each data holder, the first basic model data is obtained; Based on the contribution of each data holder, the data holders participating in this round of model co-training are determined, and the first basic model data is sent to the data holders participating in this round of model co-training. The first basic model data is used by the data holders participating in this round of model co-training to conduct this round of model co-training.

2. The method according to claim 1, characterized in that, The process of obtaining the contribution of each data holder includes: The contribution of each data holder is evaluated, and the contribution of each data holder is determined based on the evaluation results.

3. The method according to claim 2, characterized in that, The contribution assessment includes at least one of the following: Similarity calculation is performed on the local model data corresponding to each data holder and the basic model data sent to each data holder in the previous round of model co-training. The semantic information quality of each data holder is determined based on the data feature values ​​sent by each data holder; wherein, the data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set; Determine the proportion of the number of samples in each data holder's local training sample set to the total number of samples.

4. The method according to claim 2, characterized in that, The determination of the contribution of each data holder based on the contribution assessment results includes: For each data holder, the sub-contributions obtained from the various contribution assessments of the data holder are weighted and summed to obtain the contribution of the data holder.

5. The method according to claim 1, characterized in that, The process of determining the data holders participating in this round of model collaborative training based on their respective contributions includes: Data holders whose contribution is greater than or equal to a preset contribution threshold will be selected as data holders to participate in this round of model collaborative training.

6. The method according to claim 1, characterized in that, The process of determining the data holders participating in this round of model collaborative training based on their respective contributions includes: The total contribution is obtained by summing the contributions of each data holder. For each data holder, the contribution percentage of each data holder is determined based on the ratio of the data holder's contribution to the total contribution. Data holders whose contribution percentage is greater than or equal to a preset threshold will be considered as data holders participating in this round of model collaborative training.

7. The method according to claim 1, characterized in that, The step of obtaining the first basic model data based on the local model data of each data holder and the contribution of each data holder includes: The weight of each data holder is determined based on their contribution. Based on the weights of each data holder, the local model data of each data holder is aggregated to obtain the first basic model data.

8. The method according to claim 1, characterized in that, The method further includes: The training count is sent to the data holders participating in this round of model co-training. The training count is used to indicate the number of iterations the data holders participate in when performing this round of model co-training based on the first basic model data.

9. The method according to claim 1, characterized in that, The method further includes: In this round of model co-training, determine whether the training termination condition is met; If the training termination condition is met, the second basic model data obtained from this round of model co-training will be sent to the data holders participating in this round of model co-training, so that the data holders participating in this round of model co-training can adjust the second basic model data based on their local training sample set to obtain fine-tuned model data.

10. The method according to claim 1, characterized in that, The method further includes: Establish and publish collaborative training tasks for federated learning models; Receive participation requests from data holders who wish to participate in the model co-training task; The initial basic model data is sent to the data holders corresponding to each of the participation requests, so that the data holders corresponding to each of the participation requests can perform model co-training based on the local training sample set and the initial basic model data to obtain local model data.

11. The method according to claim 9, characterized in that, The coordinator and each of the data holders are semantic agents; the first basic model data and the second basic model data are semantic knowledge bases common to all semantic agents; the fine-tuned model data are personalized semantic knowledge bases formed after fine-tuning with local training sample sets.

12. A federated learning method, characterized in that, For data holders, the method includes: In each round of model co-training, the local model data obtained in the previous round of model co-training is sent to the coordinator so that the coordinator can obtain the first basic model data based on the local model data sent by each data holder and the contribution of each data holder. Receive the first basic model data sent by the coordinator, and perform this round of model collaborative training based on the first basic model data; The data holders are those who participate in this round of model collaborative training, as determined by the coordinator based on their respective contributions. The contribution level is used to characterize the value of the local model data provided by the data holders in the previous round of model collaborative training.

13. The method according to claim 12, characterized in that, The method further includes: Send at least one of the data feature value and the number of samples in the local training sample set to the coordinator so that the coordinator can evaluate the contribution of the data holder based on at least one of the data feature value and the number of samples, and determine the contribution of the data holder. The data feature values ​​are obtained by the data holder after extracting features from the sample data in the local training sample set.

14. The method according to claim 13, characterized in that, The method further includes: Feature extraction is performed on each sample data in the local training sample set to obtain the feature vector corresponding to each sample data. The feature vectors corresponding to each sample data are aligned, and the feature matrix of the local training sample set is obtained based on the aligned feature vectors corresponding to each sample data. The product of the feature matrix and its transpose is used as the variable of the sqrt function to calculate the data feature value.

15. The method according to claim 12, characterized in that, The method further includes: Receive the number of training sessions sent by the coordinator; Correspondingly, the step of performing this round of model co-training based on the first basic model data includes: This round of model co-training is performed based on the first basic model data and the number of training iterations.

16. The method according to claim 12, characterized in that, The method further includes: The system receives the second basic model data sent by the coordinator and adjusts the second basic model data based on the local training sample set to obtain fine-tuned model data.

17. The method according to claim 16, characterized in that, The second basic model data is sent by the coordinator when the training termination condition is met.

18. The method according to claim 16, characterized in that, The method further includes: The performance of the fine-tuned model data is evaluated. If the fine-tuned model data meets the performance requirements, then the fine-tuned model data is encapsulated to obtain encapsulated fine-tuned model data.

19. The method according to claim 16 or 18, characterized in that, The method further includes: Send the fine-tuned model data to an affiliate of the data holder; and / or, The fine-tuned model data is sent to a trusted party trusted by the data holder.

20. The method according to claim 12, characterized in that, The method further includes: After the coordinator publishes the model collaborative training task of federated learning, a participation request is sent to the coordinator. Receive the initial basic model data sent by the coordinator; Based on the local training sample set and the initial basic model data, local model data is obtained through model co-training.

21. The method according to claim 16, characterized in that, The coordinator and the data holder are both semantic agents; the first basic model data and the second basic model data are both semantic knowledge bases common to all semantic agents; the fine-tuned model data are personalized semantic knowledge bases formed after fine-tuning with local training sample sets.

22. A federated learning device, characterized in that, Located in the coordinator, the device includes: The receiving module is used to receive local model data obtained by multiple data holders in the previous round of model co-training in each round of model co-training; The first acquisition module is used to acquire the contribution of each data holder, wherein the contribution is used to characterize the value of the local model data provided by the data holder in the previous round of model co-training. The second acquisition module is used to acquire the first basic model data based on the local model data of each data holder and the contribution of each data holder; The sending module is used to determine the data holders participating in this round of model collaborative training based on the contribution of each data holder, and to send the first basic model data to the data holders participating in this round of model collaborative training. The first basic model data is used by the data holders participating in this round of model collaborative training to carry out this round of model collaborative training.

23. A federated learning device, characterized in that, The device, located by the data holder, includes: The sending module is used to send the local model data obtained in the previous round of model co-training to the coordinator in each round of model co-training, so that the coordinator can obtain the first basic model data according to the local model data sent by each data holder and the contribution of each data holder. The receiving module is used to receive the first basic model data sent by the coordinator and perform this round of model collaborative training based on the first basic model data. The data holders are those who participate in this round of model collaborative training, as determined by the coordinator based on their respective contributions. The contribution level is used to characterize the value of the local model data provided by the data holders in the previous round of model collaborative training.

24. A communication device comprising a transceiver, 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 11 by controlling the transceiver.

25. A communication device comprising a transceiver, 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 12 to 21 by controlling the transceiver.

26. 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 11 or 12 to 21.

27. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 11 or 12 to 21.

Images