Federated learning modeling optimization method, device, medium and computer program product

Abstract

The present application discloses a method, device, medium and computer program product for optimizing federated learning modeling, wherein the method comprises: receiving encrypted model parameters and social information sent by each second device, wherein the encrypted model parameters are generated by the second device through iterative training of a local model, and according to each social information, selectively densely aggregate each encrypted model parameter based on social association, obtain encrypted aggregate model parameters corresponding to each social information, and feed back each encrypted aggregate model parameter to the corresponding second device, so that the second device can iteratively optimize the local model according to the received encrypted aggregate model parameters to obtain a federated learning model. The present application solves the technical problem of low efficiency of federated learning modeling.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a federated learning modeling optimization method, device, medium and computer program product. Background Art

[0002] With the development of computer technology, the application of federated learning is becoming more and more extensive. Currently, in federated learning, all participants need to send the updated model parameters to the server after completing the local model update, and then complete the model aggregation process through the server. At present, the server usually collects the model parameters uploaded by all participants, and randomly selects some uploaded model parameters for weighted average as the updated global model parameters. However, randomly selecting some uploaded model parameters will make it difficult for the updated global model to iterate and converge, which will lead to a large amount of calculation in the model aggregation process, resulting in low efficiency of federated learning modeling. Summary of the invention

[0003] The main purpose of this application is to provide a federated learning modeling optimization method, device, medium and computer program product, aiming to solve the technical problem of low efficiency of federated learning modeling in the prior art.

[0004] To achieve the above-mentioned object, the present application provides a federated learning modeling optimization method, which is applied to a first device and includes:

[0005] Receiving encryption model parameters and social information sent by each second device, wherein the encryption model parameters are generated by the second device through iterative training of a local model;

[0006] According to each of the social information, selective encryption aggregation based on social relevance is performed on each of the encryption model parameters to obtain encryption aggregation model parameters corresponding to each of the social information;

[0007] The encrypted aggregate model parameters are fed back to the corresponding second devices respectively, so that the second devices can iteratively optimize the local model according to the received encrypted aggregate model parameters to obtain the federated learning model.

[0008] To achieve the above object, the present application provides a federated learning modeling optimization method, which is applied to a second device and includes:

[0009] Iteratively train the local model according to the training sample set to obtain encryption model parameters corresponding to the local model;

[0010] The encryption model parameters and the social information corresponding to the second device are sent to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encryption aggregation model parameters corresponding to each second device;

[0011] Receive the encrypted aggregation model parameters fed back by the first device, and iteratively optimize the local model according to the encrypted aggregation model parameters to obtain a federated learning model.

[0012] The present application also provides a federated recommendation optimization method, the federated recommendation optimization method applies a second device, and the federated recommendation optimization method includes:

[0013] Acquire user data to be recommended, and input the user data to be recommended into a federated recommendation model to obtain a user feature representation variable, wherein the federated recommendation model is obtained by the second device iteratively optimizing a local model based on encrypted aggregated model parameters sent by the first device, and the encrypted aggregated model parameters are obtained by the first device selectively aggregating encrypted model parameters of each second device based on social information corresponding to each second device;

[0014] Calculating user similarity results corresponding to the user feature representation variables, and generating a similar user candidate set based on the similarity results;

[0015] Based on the similar user candidate set, a preset item recommendation process is executed.

[0016] The present application also provides a federated learning modeling optimization device, which is a virtual device and is applied to a first device. The federated learning modeling optimization device includes:

[0017] A receiving module, configured to receive encryption model parameters and social information sent by each second device, wherein the encryption model parameters are generated by the second device through iterative training of a local model;

[0018] An aggregation module, configured to perform selective encryption aggregation based on social relevance on each of the encryption model parameters according to each of the social information, to obtain encryption aggregation model parameters corresponding to each of the social information;

[0019] A feedback module is used to feed back each of the encrypted aggregation model parameters to the corresponding second device, so that the second device can iteratively optimize the local model according to the received encrypted aggregation model parameters to obtain a federated learning model.

[0020] The present application also provides a federated learning modeling optimization device, which is a virtual device and is applied to a second device. The federated learning modeling optimization device includes:

[0021] A training module, used for iteratively training the local model according to the training sample set to obtain encryption model parameters corresponding to the local model;

[0022] A sending module, configured to send the encryption model parameters and the social information corresponding to the second device to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encryption aggregation model parameters corresponding to each second device;

[0023] The optimization module is used to receive the encrypted aggregation model parameters fed back by the first device, and iteratively optimize the local model according to the encrypted aggregation model parameters to obtain a federated learning model.

[0024] The present application also provides a federal recommendation optimization device, which is a virtual device and is applied to a second device. The federal recommendation optimization device includes:

[0025] an acquisition module, configured to acquire user data to be recommended, and input the user data to be recommended into a federated recommendation model to obtain a user feature representation variable, wherein the federated recommendation model is obtained by the second device iteratively optimizing a local model based on encrypted aggregated model parameters sent by the first device, and the encrypted aggregated model parameters are obtained by the first device selectively aggregating encrypted model parameters of each second device based on social information corresponding to each second device;

[0026] A generation module, used to calculate the user similarity results corresponding to the user feature representation variables, and generate a similar user candidate set based on the similarity results;

[0027] The recommendation module is used to execute a preset item recommendation process based on the similar user candidate set.

[0028] The present application also provides a federated learning modeling optimization device, which is a physical device, and includes: a memory, a processor, and a program of the federated learning modeling optimization method stored in the memory and executable on the processor. When the program of the federated learning modeling optimization method is executed by the processor, the steps of the federated learning modeling optimization method as described above can be implemented.

[0029] The present application also provides a federal recommendation optimization device, which is a physical device. The federal recommendation optimization device includes: a memory, a processor, and a program of the federal recommendation optimization method stored in the memory and executable on the processor. When the program of the federal recommendation optimization method is executed by the processor, the steps of the federal recommendation optimization method as described above can be implemented.

[0030] The present application also provides a medium, which is a readable storage medium, on which is stored a program for implementing a federated learning modeling optimization method, and when the program of the federated learning modeling optimization method is executed by a processor, the steps of the federated learning modeling optimization method as described above are implemented.

[0031] The present application also provides a medium, which is a readable storage medium, on which is stored a program for implementing the federal recommendation optimization method, and when the program of the federal recommendation optimization method is executed by a processor, the steps of the federal recommendation optimization method as described above are implemented.

[0032] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned federated learning modeling optimization method when executed by a processor.

[0033] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned federal recommendation optimization method when executed by a processor.

[0034] The present application provides a method, device, medium and computer program product for optimizing federated learning modeling. Compared with the technical means adopted in the prior art for performing federated learning modeling by collecting model parameters uploaded by all participants based on a server, and randomly selecting some of the uploaded model parameters for weighted averaging as updated global model parameters, it should be noted that when randomly selecting some of the uploaded model parameters for model aggregation in the prior art, since the correlation between the model parameters uploaded by the federated learning participants is different, if model parameters with low correlation are selected for aggregation, it will make it difficult for the updated global model to iteratively converge. The present application first receives the encrypted model parameters and social information sent by each second device, wherein the encrypted model parameters are generated by the second device through iterative training of a local model, and then, according to each of the social information, selective encryption aggregation based on social correlation is performed on each of the encrypted model parameters to obtain the corresponding social information. The encrypted aggregation model parameters can be selected based on social information to perform encrypted aggregation on the encrypted model parameters corresponding to the second device with high social correlation, that is, the encrypted model parameters with low social correlation can be quickly screened and eliminated, so that the encrypted aggregation model parameters after encryption aggregation are not negatively affected by the encrypted model parameters with low social correlation. Furthermore, each of the encrypted aggregation model parameters is fed back to the corresponding second device respectively, so that the second device can iteratively optimize the local model according to the received encrypted aggregation model parameters, thereby improving the iterative convergence efficiency of the iterative optimization of the local model by each second device, that is, improving the iterative convergence efficiency of the global model, and thus obtaining the federated learning model. Therefore, the technical defect that the randomly selected partially uploaded model parameters in the prior art will make it difficult for the updated global model to iteratively converge, thereby resulting in a large amount of calculation in the model aggregation process, which leads to a low efficiency of federated learning modeling is overcome, thereby improving the efficiency of federated learning modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0037] Figure 1 This is a flowchart of the first embodiment of the federated learning modeling optimization method of the present application;

[0038] Figure 2This is a flow chart of the second embodiment of the federated learning modeling optimization method of the present application;

[0039] Figure 3 This is a flowchart of the third embodiment of the federated learning modeling optimization method of the present application;

[0040] Figure 4 A schematic diagram of the device structure of the hardware operating environment involved in the federated learning modeling optimization method in the embodiment of the present application;

[0041] Figure 5 A schematic diagram of the device structure of the hardware operating environment involved in the federal recommendation optimization method in the embodiment of the present application;

[0042] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0043] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0044] The present application embodiment provides a federated learning modeling optimization method. In the first embodiment of the federated learning modeling optimization method of the present application, refer to Figure 1 , the federated learning modeling optimization method is applied to a first device, and the federated learning modeling optimization method includes:

[0045] Step S10, receiving encryption model parameters and social information sent by each second device, wherein the encryption model parameters are generated by the second device through iterative training of a local model;

[0046] In this embodiment, it should be noted that the encryption model parameters are model parameters corresponding to the homomorphically encrypted local model, and the social information is information recording the social association between each second device. In one embodiment, the social information can be represented by a social information vector. For example, the social association between device A and device B is 90%, and the social association between device A and device C is 80%. The social information of device A can be recorded as A(0.9, 0.8), wherein the social association is the degree of social association between each second device, which can be determined by the social behavior data between each second device, such as chat frequency, chat record keywords, and Moments like records.

[0047] Receive encrypted model parameters and social information sent by each second device, wherein the encrypted model parameters are generated by the second device through iterative training of a local model. Specifically, each second device iteratively trains its own local model to a preset number of iterations based on its own training sample set to respectively obtain local model parameters corresponding to its own local model, and respectively homomorphically encrypts its own model parameters to obtain its own encrypted model parameters, and then each second device sends its own encrypted model parameters and its own social information to the first device, and then the first device receives the encrypted model parameters and social information sent by each second device.

[0048] Step S20, according to each of the social information, selectively encrypt and aggregate each of the encryption model parameters based on the social association degree, to obtain the encryption aggregation model parameters corresponding to each of the social information;

[0049] In this embodiment, it should be noted that the encryption aggregation method includes weighted average and summation, and the social information at least includes a social association degree.

[0050] According to each piece of social information, selective encryption aggregation based on social relevance is performed on each encryption model parameter to obtain an encryption aggregation model parameter corresponding to each piece of social information. Specifically, the following steps are performed for each piece of social information:

[0051] In the social information, each social association degree greater than a preset association degree threshold is selected as the target social association degree, and then the encryption model parameters corresponding to the second device sending the social information and the encryption model parameters sent by other second devices corresponding to each target social association degree are extracted as each social association model parameter, wherein each of the other second devices has a target social association degree with each of the second devices, and then the social association model parameters are encrypted and aggregated to obtain the encrypted aggregation model parameters corresponding to the social information, and then the encrypted aggregation model parameters corresponding to each of the social information are obtained.

[0052] The step of selectively encrypting and aggregating the encrypted model parameters based on social relevance according to the social information to obtain the encrypted aggregation model parameters corresponding to the social information includes:

[0053] Step S21, based on the social association degree corresponding to each of the social information, selecting each social association model parameter corresponding to each of the second devices from each of the encryption model parameters;

[0054] In this embodiment, it should be noted that the social association model parameters are encrypted model parameters whose social association degree is greater than a preset association degree threshold. The social association model parameters include the encrypted model parameters of the second device itself and the encrypted model parameters corresponding to each other second device that has a target social association degree with the second device. For example, assuming that there are three second devices, namely, the target device, device a and device b, the social association degree between the target device and device a is 90%, and the social association degree between the target device and device b is 50%. If the preset association degree threshold is 60%, the encrypted model parameters corresponding to the target device and the encrypted model parameters corresponding to device a are both used as social association model parameters.

[0055] Based on the social association degree corresponding to each of the social information, each of the social association model parameters corresponding to each of the second devices is selected from each of the encryption model parameters. Specifically, the following steps are performed for each of the social information sent by the second device:

[0056] According to the social association degrees in the social information, other second devices whose social association degrees are greater than a preset association degree threshold are selected from other second devices as social association devices, and then the encrypted model parameters corresponding to the second device and each of the social association devices are used as the social association model parameters, so as to obtain the social association model parameters corresponding to each of the second devices.

[0057] The step of selecting the social association model parameters corresponding to the second devices in the encryption model parameters based on the social associations corresponding to the social information includes:

[0058] Step S211, comparing the social relevance in each of the social information with a preset relevance threshold, to obtain a relevance comparison result corresponding to each of the social information;

[0059] In this embodiment, the social relevance in each of the social information is compared with a preset relevance threshold to obtain a relevance comparison result corresponding to each of the social information. Specifically, the following steps are performed for each of the social information sent by the second device:

[0060] Compare each social correlation in the social information with a preset correlation threshold, and if the social correlation is greater than the preset correlation threshold, generate a first correlation comparison result; if the social correlation is not greater than the preset correlation threshold, generate a second correlation comparison result, and then use each first correlation comparison result and each second correlation comparison result corresponding to the social information as the correlation comparison result, wherein the first correlation comparison result and the second correlation comparison result can be labels or identifiers corresponding to the social correlation, which are used to mark the result of the correlation comparison between the social correlation and the preset correlation threshold, for example, set 1 to mark that the social correlation is greater than the preset correlation threshold, and set 0 to mark that the social correlation is not greater than the preset correlation threshold.

[0061] Step S212: According to the comparison results of the respective correlations, the encryption model parameters whose social correlations satisfy a preset correlation threshold condition are selected as the social correlation model parameters corresponding to the respective second devices.

[0062] In this embodiment, it should be noted that the preset relevance threshold condition can be set as the social relevance being greater than the preset relevance threshold.

[0063] According to each of the correlation comparison results, each of the encrypted model parameters whose social correlation meets the preset correlation threshold condition is selected as each of the social correlation model parameters corresponding to each of the second devices. Specifically, for each of the correlation comparison results corresponding to the second device, the following steps are performed:

[0064] According to the correlation comparison result, the encryption model parameters of other second devices corresponding to each social correlation degree not greater than the preset correlation threshold are eliminated, and the encryption model parameters corresponding to the second device and the encryption model parameters corresponding to other second devices corresponding to each social correlation degree greater than the preset correlation threshold are selected as social association model parameters, and then the social association model parameters corresponding to each second device are obtained. For example, M(i, j) represents the association strength between the target device i and the device j. If M(i, j) is greater than the preset correlation threshold, the encryption model parameters of the device j are used as the social association model parameters. If M(i, j) is not greater than the preset correlation threshold, the encryption model parameters of the user j are eliminated.

[0065] Step S22: encrypt and aggregate the social association model parameters corresponding to the second devices respectively to obtain encrypted aggregate model parameters corresponding to the social information.

[0066] In this embodiment, the social association model parameters corresponding to the second devices are encrypted and aggregated respectively to obtain the encrypted aggregate model parameters corresponding to the social information. Specifically, the following steps are performed for each second device:

[0067] According to a preset encryption aggregation rule, the social association model parameters corresponding to the second device are encrypted and aggregated to obtain the encrypted aggregation model parameters corresponding to the social information sent by the second device, wherein the preset encryption aggregation rule includes weighted average and summation, etc.

[0068] Step S30, feeding back each of the encrypted aggregate model parameters to the corresponding second device respectively, so that the second device can iteratively optimize the local model according to the received encrypted aggregate model parameters to obtain a federated learning model.

[0069] In this embodiment, each of the encrypted aggregate model parameters is fed back to the corresponding second device respectively, so that the second device can iteratively optimize the local model according to the received encrypted aggregate model parameters to obtain a federated learning model. Specifically, each of the encrypted aggregate model parameters is sent to the corresponding second device respectively, so that the second device can replace and update the model parameters of the local model with the aggregate model parameters according to the received encrypted aggregate model parameters to optimize the local model, and determine whether the optimized local model meets the preset training end conditions, wherein the preset training end conditions include conditions such as loss function convergence and reaching the maximum number of iterations threshold. If satisfied, the optimized local model is used as the federated learning model. If not satisfied, the execution step is returned: each of the second devices iteratively trains its own local model to a preset number of iterations based on its own training sample set to respectively obtain the local model parameters corresponding to its own local model.

[0070] The embodiment of the present application provides a method for optimizing federated learning modeling. Compared with the technical means adopted in the prior art for performing federated learning modeling by collecting model parameters uploaded by all participants based on a server, and randomly selecting some of the uploaded model parameters for weighted averaging as updated global model parameters, it should be noted that when randomly selecting some of the uploaded model parameters for model aggregation in the prior art, since the correlation between the model parameters uploaded by the participants of the federated learning is different, if model parameters with low correlation are selected for aggregation, it will make it difficult for the updated global model to iteratively converge. The present application first receives the encrypted model parameters and social information sent by each second device, wherein the encrypted model parameters are generated by the second device through iterative training of a local model, and then, according to each of the social information, selectively encrypts and aggregates the encrypted model parameters based on the social correlation to obtain the encrypted aggregation model parameters corresponding to each of the social information. type parameters, and then based on social information, the encryption model parameters corresponding to the second device with high social correlation can be selected for encryption aggregation, that is, the encryption model parameters with low social correlation can be quickly screened and eliminated, so that the encrypted aggregation model parameters after encryption aggregation are not negatively affected by the encryption model parameters with low social correlation. Furthermore, each of the encrypted aggregation model parameters is fed back to the corresponding second device respectively, so that the second device can iteratively optimize the local model according to the received encrypted aggregation model parameters, so as to improve the iterative convergence efficiency of the iterative optimization of the local model by each second device, that is, improve the iterative convergence efficiency of the global model, and then obtain the federated learning model. Therefore, it overcomes the technical defect in the prior art that randomly selecting part of the uploaded model parameters will make it difficult for the updated global model to iteratively converge, which will lead to a large amount of calculation in the model aggregation process, resulting in low efficiency of federated learning modeling, thereby improving the efficiency of federated learning modeling.

[0071] Further, refer to Figure 2 In another embodiment of the present application, the federated learning modeling optimization method is applied to a second device, and the federated learning modeling optimization method further includes:

[0072] Step A10, iteratively training the local model according to the training sample set to obtain encryption model parameters corresponding to the local model;

[0073] In this embodiment, it should be noted that the encrypted model parameters are model parameters corresponding to the local model after iterative training of the homomorphic encryption model.

[0074] The local model is iteratively trained according to the training sample set to obtain encrypted model parameters corresponding to the local model. Specifically, the local model is iteratively trained according to the training sample set until the number of iterations of the local model reaches a preset number of iterations, the model parameters of the local model are obtained, and the model parameters are homomorphically encrypted to obtain the encrypted model parameters.

[0075] Step A20, sending the encryption model parameters and the social information corresponding to the second device to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encryption aggregation model parameters corresponding to each second device;

[0076] In this embodiment, the encryption model parameters and the social information corresponding to the second device are sent to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encrypted aggregation model parameters corresponding to each second device. Specifically, the encryption model parameters and the social information corresponding to the second device are sent to the first device, so that the first device can select the encryption model parameters corresponding to each second device whose social association is greater than a preset association threshold based on the social association corresponding to each social information, and then encrypt and aggregate the encryption model parameters corresponding to each second device, and obtain the encrypted aggregation model parameters corresponding to each second device. The specific content of the first device generating the encrypted aggregation model parameters corresponding to each second device can refer to the specific content in step S20, which will not be repeated here.

[0077] Step A30: Receive the encrypted aggregation model parameters fed back by the first device, and iteratively optimize the local model according to the encrypted aggregation model parameters to obtain a federated learning model.

[0078] In this embodiment, the encrypted aggregate model parameters fed back by the first device are received, and according to the encrypted aggregate model parameters, the local model is iteratively optimized to obtain a federated learning model. Specifically, the encrypted aggregate model parameters fed back by the first device are received, and the encrypted aggregate model parameters are decrypted to obtain aggregate model parameters, and then the model parameters of the local model are replaced and updated with the aggregate model parameters to optimize the local model, and it is determined whether the optimized local model meets the preset training end conditions, wherein the preset training end conditions include conditions such as loss function convergence and reaching the maximum number of iterations threshold. If the conditions are met, the optimized local model is used as the federated learning model. If not, the execution step is returned to: iteratively train the local model according to the training sample set to obtain the encrypted model parameters corresponding to the local model.

[0079] Wherein, the federated learning model includes a federated recommendation model,

[0080] After the step of iteratively optimizing the local model according to the encrypted aggregation model parameters to obtain the federated learning model, the federated learning modeling optimization includes:

[0081] Step B10, obtaining user data, and inputting the user data into the federated learning model to obtain user feature representation variables;

[0082] In this example, it should be noted that the user data is data recording the target user about items, for example, the number of times the target user purchases a certain product and the number of visits to a certain web page by the target user, etc. The user feature representation corresponding to the user feature representation variable can be represented by a string of real numbers, and the user feature representation can uniquely represent a certain thing. For example, if the number of clicks on a certain web page by a user is 5, the user feature representation variable can be set to the real number 5.

[0083] Acquire user data, and input the user data into the federated learning model to obtain user feature representation variables. Specifically, extract the user data from a preset local database, input the user data into the federated learning model to obtain the user feature representation variables.

[0084] Step B20, calculating the user similarity results corresponding to the user feature representation variables, and generating a similar user candidate set based on the similarity results;

[0085] In this embodiment, it should be noted that the user similarity is the degree of similarity between users, and the federated learning model includes a scoring model, wherein the method for calculating the user similarity result includes Euclidean distance, cosine (cosine function), etc.

[0086] Calculate the user similarity result corresponding to the user feature representation variable, and generate a similar user candidate set based on the similarity result. Specifically, based on the user feature representation variable, score each to-be-selected user corresponding to the target user through the scoring model to obtain a to-be-selected user score corresponding to each to-be-selected user, wherein the higher the to-be-selected user score is, the higher the user similarity between the target user and the to-be-selected user is. Then, based on the scores of each to-be-selected user, sort each to-be-selected user to obtain a to-be-selected user list, that is, obtain a user similarity result, and select the similar user candidate set from the to-be-selected user list. For example, select the top three to-be-selected users with the highest scores from the to-be-selected user list to form the similar user candidate set.

[0087] Step B30: executing a preset item recommendation process based on the similar user candidate set.

[0088] In this embodiment, it should be noted that the purpose of executing the preset item recommendation process is to make personalized recommendations to the target user corresponding to the preset item recommendation process, so as to recommend the items that the user to be recommended is interested in to the target user.

[0089] Based on the similar user candidate set, a preset item recommendation process is executed. Specifically, a target item set corresponding to the similar user candidate set is obtained, and the target item set includes one or more items associated with each of the similar users, such as movies watched, goods purchased, web pages clicked, etc. by the similar users. Then, each item in the target item set is screened to obtain a set of items to be recommended. Further, items to be recommended are selected from the set of items to be recommended, and the items to be recommended are recommended to the target user. For example, assuming that the similar user candidate set includes similar user A, similar user B and similar user C, similar user A has watched movie a, movie b and movie c, similar user B has watched movie a and movie b, and similar user C has watched movie a. The number of item appearances of movie a is 3, that is, the number of clicks on movie a is 3. Similarly, the number of clicks on movie b is 2, and the number of clicks on movie c is 1. If the preset click threshold is 1, movie a and movie b are selected to form the set of items to be recommended, and the items to be recommended are recommended to the users to be recommended.

[0090] The embodiment of the present application provides a method for optimizing federated learning modeling, that is, compared with the technical means adopted in the prior art to collect model parameters uploaded by all participants based on a server, and randomly select some of the uploaded model parameters for weighted averaging as updated global model parameters to perform federated learning modeling, it should be noted that when randomly selecting some of the uploaded model parameters for model aggregation in the prior art, since the correlation between the model parameters uploaded by the participants of the federated learning is different, if model parameters with low correlation are selected for aggregation, it will make it difficult for the updated global model to iteratively converge. The embodiment of the present application iteratively trains the local model according to the training sample set to obtain the encrypted model parameters corresponding to the local model, and then sends the encrypted model parameters and the social information corresponding to the second device to the first device, so that the first device can perform social-based optimization on the encrypted model parameters sent by each second device based on the social information sent by each second device. Selective encryption aggregation of the correlation degree can obtain the encrypted aggregation model parameters corresponding to each of the second devices, so that the encrypted model parameters corresponding to the second devices with high social correlation can be selected for encrypted aggregation based on social information, that is, the encrypted model parameters with low social correlation can be quickly screened and eliminated, so that the encrypted aggregation model parameters after encryption aggregation are not negatively affected by the encrypted model parameters with low social correlation. Furthermore, the encrypted aggregation model parameters fed back by the first device are received, and the local model is iteratively optimized according to the encrypted aggregation model parameters, so that the iterative convergence efficiency of the iterative optimization of the local model by each second device can be improved, that is, the iterative convergence efficiency of the global model is improved, and then the federated learning model is obtained. Therefore, the technical defect that the random selection of part of the uploaded model parameters in the prior art will make it difficult for the updated global model to iteratively converge, which will lead to a large amount of calculation in the model aggregation process, resulting in low efficiency of federated learning modeling is overcome, thereby improving the efficiency of federated learning modeling.

[0091] Further, refer to Figure 3 In another embodiment of the present application, the federal recommendation optimization method is applied to a second device, and the federal recommendation optimization method includes:

[0092] Step D10: obtaining user data to be recommended, and inputting the user data to be recommended into a federated recommendation model to obtain a user feature representation variable, wherein the federated recommendation model is obtained by the second device iteratively optimizing a local model based on encrypted aggregated model parameters sent by the first device, and the encrypted aggregated model parameters are obtained by the first device selectively aggregating encrypted model parameters of each second device based on social information corresponding to each second device;

[0093] In this embodiment, it should be noted that the user data to be recommended is data recording the items of the user to be recommended, for example, the number of times the user to be recommended has purchased a certain product and the number of visits to a certain web page by the user to be recommended, etc. The federal recommendation optimization method is applied to the second device, and the second device periodically updates the federal recommendation model, that is, iterative training is performed based on the existing users and newly added users in the second device based on the federal learning modeling optimization method.

[0094] Obtain the user data to be recommended, and input the user data to be recommended into the federated recommendation model to obtain user feature representation variables, wherein the federated recommendation model is obtained by the second device iteratively optimizing the local model based on the encrypted aggregation model parameters sent by the first device, and the encrypted aggregation model parameters are obtained by the first device selectively aggregating the encrypted model parameters of each second device based on the social information corresponding to each second device. Specifically, extract the user data to be recommended from a preset local database, input the user data to be recommended into the federated recommendation model, and then obtain the user feature representation variables, wherein the federated recommendation model is obtained by the second device iteratively optimizing the local model based on the encrypted aggregation model parameters sent by the first device, and the encrypted aggregation model parameters are obtained by the first device selectively aggregating the encrypted model parameters of each second device based on the social information corresponding to each second device. The construction process of the federated recommendation model can be specifically referred to the contents of steps S10 to S30, which will not be repeated here.

[0095] Step D20, calculating the user similarity results corresponding to the user feature representation variables, and generating a similar user candidate set based on the similarity results;

[0096] In this embodiment, the user similarity results corresponding to the user feature representation variables are calculated, and a similar user candidate set is generated based on the similarity results. Specifically, based on the user feature representation variables, each to-be-selected user corresponding to the to-be-recommended user is scored by the scoring model to obtain a to-be-selected user score corresponding to each to-be-selected user, and then based on each to-be-selected user score, each to-be-selected user is sorted to obtain a user similarity result, and the similar user candidate set is selected from the to-be-selected user list.

[0097] Step D30: executing a preset item recommendation process based on the similar user candidate set.

[0098] In this embodiment, based on the similar user candidate set, a preset item recommendation process is executed. Specifically, a target item set corresponding to the similar user candidate set is obtained, and then each item in the target item set is screened to obtain a to-be-recommended item set. Further, items to be recommended are selected from the to-be-recommended item set, and the items to be recommended are recommended to the to-be-recommended user. For example, assuming that the purpose of executing the preset item recommendation process is to recommend movies to the to-be-recommended user, then all the movies watched by the similar users are the to-be-recommended items, and then the movies that are ranked top in the number of times watched by the similar users and have not been watched by the to-be-recommended user can be selected from all the movies and recommended to the to-be-recommended user. The specific implementation methods of steps D10 to D30 can refer to the contents of steps B10 to B30 in the second embodiment of the present application, and will not be repeated here.

[0099] Before the step of obtaining the user data to be recommended and inputting the user data to be recommended into the federated recommendation model to obtain the user feature representation variable, wherein the federated recommendation model is obtained by the second device iteratively optimizing the local model based on the encrypted aggregated model parameters sent by the first device, and the encrypted aggregated model parameters are obtained by the first device selectively aggregating the encrypted model parameters of each second device based on the social information corresponding to each second device, the federated recommendation optimization method further includes:

[0100] Step E10, iteratively training the local model according to the training sample set to obtain encryption model parameters corresponding to the local model;

[0101] In this embodiment, the local model is iteratively trained according to the training sample set to obtain encrypted model parameters corresponding to the local model. Specifically, the local model is iteratively trained according to the training sample set until the number of iterations of the local model reaches a preset number of iterations, the model parameters of the local model are obtained, and the model parameters are homomorphically encrypted to obtain the encrypted model parameters.

[0102] Step E20, sending the encryption model parameters and the social information corresponding to the second device to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encryption aggregation model parameters corresponding to each second device;

[0103] In this embodiment, the encryption model parameters and the social information corresponding to the second device are sent to the first device, so that the first device can perform selective encryption aggregation based on social correlation on the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encrypted aggregation model parameters corresponding to each second device. Specifically, each encryption model parameter and each social information are sent to the first device, so that the first device can perform encryption aggregation on each encryption model parameter whose social correlation meets a preset correlation threshold, and obtain the encrypted aggregation model parameter.

[0104] Step E30: receiving the encrypted aggregation model parameters fed back by the first device, and iteratively optimizing the local model according to the encrypted aggregation model parameters to obtain a federated recommendation model.

[0105] In this embodiment, the encrypted aggregate model parameters fed back by the first device are received, and based on the encrypted aggregate model parameters, the local model is iteratively optimized to obtain a federal recommendation model. Specifically, the encrypted aggregate model parameters fed back by the first device are received, and the model parameters of the local model are replaced and updated with the aggregate model parameters to optimize the local model until the optimized local model meets the preset training end conditions to obtain the federal recommendation model. The specific implementation content of steps E10 to E30 can refer to the specific content of steps A10 to A30, which will not be repeated here.

[0106] The embodiment of the present application provides an item recommendation method, that is, obtaining user data to be recommended, and inputting the user data to be recommended into a federated recommendation model to obtain a user feature representation variable, wherein the federated recommendation model is obtained by iteratively optimizing a local model by the second device based on the encrypted aggregate model parameters sent by the first device, and the encrypted aggregate model parameters are obtained by the first device selectively aggregating the encrypted model parameters of each second device based on the social information corresponding to each second device, and then calculating the user similarity result corresponding to the user feature representation variable, and generating a similar user candidate set based on the similarity result, and further, based on the similar user candidate set, executing a preset item recommendation process to achieve execution through the federated recommendation model An item recommendation process is provided to achieve the purpose of personalized recommendation for users, and the federated recommendation model is obtained by the second device iteratively optimizing the local model based on the encrypted aggregate model parameters sent by the first device, and the encrypted aggregate model parameters are obtained by the first device selectively aggregating the encrypted model parameters of each second device based on the social information corresponding to each second device, that is, based on the social information, the encrypted model parameters corresponding to the second device with high social association are selected for federated learning modeling, so that the accuracy of the federated recommendation model constructed based on federated learning is not negatively affected by the encrypted model parameters with low social association, thereby improving the accuracy of the federated recommendation model, thereby making the accuracy of item recommendation higher, and thereby improving the recommendation effect of item recommendation.

[0107] Reference Figure 4 , Figure 4 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0108] like Figure 4 As shown, the federated learning modeling optimization device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0109] Optionally, the federated learning modeling optimization device may also include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen (Display), an input submodule such as a keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0110] Those skilled in the art will understand that Figure 4 The structure of the federated learning modeling optimization device shown in the figure does not constitute a limitation of the federated learning modeling optimization device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0111] like Figure 4 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, and a federated learning modeling optimization program. The operating system is a program that manages and controls the hardware and software resources of the federated learning modeling optimization device, and supports the operation of the federated learning modeling optimization program and other software and / or programs. The network communication module is used to realize communication between the components inside the memory 1005, and communication with other hardware and software in the federated learning modeling optimization system.

[0112] exist Figure 4 In the federated learning modeling optimization device shown, the processor 1001 is used to execute the federated learning modeling optimization program stored in the memory 1005 to implement the steps of any of the above-mentioned federated learning modeling optimization methods.

[0113] The specific implementation methods of the federated learning modeling optimization device of the present application are basically the same as the embodiments of the above-mentioned federated learning modeling optimization method, and will not be repeated here.

[0114] Reference Figure 5 , Figure 5 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0115] like Figure 5 As shown, the federated recommendation optimization device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0116] Optionally, the federated recommendation optimization device may also include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen (Display), an input submodule such as a keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0117] Those skilled in the art will understand that Figure 5 The structure of the federal recommendation optimization device shown in the figure does not constitute a limitation of the federal recommendation optimization device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0118] like Figure 5 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module and a federal recommendation optimization program. The operating system is a program that manages and controls the hardware and software resources of the federal recommendation optimization device, and supports the operation of the federal recommendation optimization program and other software and / or programs. The network communication module is used to realize the communication between the components inside the memory 1005, and the communication with other hardware and software in the federal recommendation optimization system.

[0119] exist Figure 5 In the federal recommendation optimization device shown, the processor 1001 is used to execute the federal recommendation optimization program stored in the memory 1005 to implement the steps of any of the above-mentioned federal recommendation optimization methods.

[0120] The specific implementation of the federal recommendation optimization device of the present application is basically the same as the embodiments of the above-mentioned federal recommendation optimization method, and will not be repeated here.

[0121] The present application also provides a federated learning modeling optimization device, which is applied to a first device and includes:

[0122] A receiving module, configured to receive encryption model parameters and social information sent by each second device, wherein the encryption model parameters are generated by the second device through iterative training of a local model;

[0123] An aggregation module, configured to perform selective encryption aggregation based on social relevance on each of the encryption model parameters according to each of the social information, to obtain encryption aggregation model parameters corresponding to each of the social information;

[0124] A feedback module is used to feed back each of the encrypted aggregation model parameters to the corresponding second device, so that the second device can iteratively optimize the local model according to the received encrypted aggregation model parameters to obtain a federated learning model.

[0125] Optionally, the aggregation module is further used for:

[0126] Based on the social association degree corresponding to each of the social information, selecting each social association model parameter corresponding to each of the second devices from each of the encryption model parameters;

[0127] The social association model parameters corresponding to the second devices are encrypted and aggregated respectively to obtain encrypted aggregate model parameters corresponding to the social information.

[0128] Optionally, the federated learning modeling optimization device is also used for:

[0129] Compare the social relevance in each social information with a preset relevance threshold to obtain a relevance comparison result corresponding to each social information;

[0130] According to the comparison results of the respective correlations, the encryption model parameters whose social correlations satisfy a preset correlation threshold condition are selected as the social correlation model parameters corresponding to the respective second devices.

[0131] The specific implementation of the federated learning modeling optimization device of the present application is basically the same as the various embodiments of the above-mentioned federated learning modeling optimization method, and will not be repeated here.

[0132] The present application also provides a federated learning modeling optimization device, which is applied to a second device and includes:

[0133] A training module, used for iteratively training the local model according to the training sample set to obtain encryption model parameters corresponding to the local model;

[0134] A sending module, configured to send the encryption model parameters and the social information corresponding to the second device to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encryption aggregation model parameters corresponding to each second device;

[0135] The optimization module is used to receive the encrypted aggregation model parameters fed back by the first device, and iteratively optimize the local model according to the encrypted aggregation model parameters to obtain a federated learning model.

[0136] Optionally, the federated learning modeling optimization device is also used for:

[0137] Acquire user data, and input the user data into the federated learning model to obtain user feature representation variables;

[0138] Calculating user similarity results corresponding to the user feature representation variables, and generating a similar user candidate set based on the similarity results;

[0139] Based on the similar user candidate set, a preset item recommendation process is executed.

[0140] The specific implementation of the federated learning modeling optimization device of the present application is basically the same as the various embodiments of the above-mentioned federated learning modeling optimization method, and will not be repeated here.

[0141] The present application also provides a federal recommendation optimization device, which is applied to a second device and includes:

[0142] an acquisition module, configured to acquire user data to be recommended, and input the user data to be recommended into a federated recommendation model to obtain a user feature representation variable, wherein the federated recommendation model is obtained by the second device iteratively optimizing a local model based on encrypted aggregated model parameters sent by the first device, and the encrypted aggregated model parameters are obtained by the first device selectively aggregating encrypted model parameters of each second device based on social information corresponding to each second device;

[0143] A generation module, used to calculate the user similarity results corresponding to the user feature representation variables, and generate a similar user candidate set based on the similarity results;

[0144] The recommendation module is used to execute a preset item recommendation process based on the similar user candidate set.

[0145] Optionally, the federal recommendation optimization device is further used for:

[0146] Iteratively train the local model according to the training sample set to obtain encryption model parameters corresponding to the local model;

[0147] The encryption model parameters and the social information corresponding to the second device are sent to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encryption aggregation model parameters corresponding to each second device;

[0148] Receive the encrypted aggregation model parameters fed back by the first device, and iteratively optimize the local model according to the encrypted aggregation model parameters to obtain the federated recommendation model.

[0149] The specific implementation of the federal recommendation optimization device of the present application is basically the same as the embodiments of the above-mentioned federal recommendation optimization method, and will not be repeated here.

[0150] An embodiment of the present application provides a medium, which is a readable storage medium, and the readable storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement the steps of the federated learning modeling optimization method described in any one of the above items.

[0151] The specific implementation methods of the readable storage medium of the present application are basically the same as the embodiments of the above-mentioned federated learning modeling optimization method, and will not be repeated here.

[0152] An embodiment of the present application provides a medium, which is a readable storage medium, and the readable storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement the steps of any of the above-mentioned federal recommendation optimization methods.

[0153] The specific implementation of the readable storage medium of the present application is basically the same as the embodiments of the above-mentioned federal recommendation optimization method, and will not be repeated here.

[0154] An embodiment of the present application provides a computer program product, and the computer program product includes one or more computer programs, and the one or more computer programs can also be executed by one or more processors to implement the steps of any of the above-mentioned federated learning modeling optimization methods.

[0155] The specific implementation methods of the computer program product of the present application are basically the same as the embodiments of the above-mentioned federated learning modeling optimization method, and will not be repeated here.

[0156] An embodiment of the present application provides a computer program product, and the computer program product includes one or more computer programs, and the one or more computer programs can also be executed by one or more processors to implement the steps of any of the above-mentioned federal recommendation optimization methods.

[0157] The specific implementation methods of the computer program product of the present application are basically the same as the embodiments of the above-mentioned federal recommendation optimization method, and will not be repeated here.

[0158] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. A federated learning modeling optimization method, characterized in that: The federated learning modeling optimization method is applied to a first device, and the federated learning modeling optimization method includes: Receiving encryption model parameters and social information sent by each second device, wherein the encryption model parameters are generated by the second device through iterative training of a local model; According to each of the social information, selective encryption aggregation based on social relevance is performed on each of the encryption model parameters to obtain encryption aggregation model parameters corresponding to each of the social information; Feeding back each of the encrypted aggregate model parameters to the corresponding second device respectively, so that the second device can iteratively optimize the local model according to the received encrypted aggregate model parameters to obtain a federated learning model; The step of selectively encrypting and aggregating the encrypted model parameters based on social relevance according to the social information to obtain the encrypted aggregation model parameters corresponding to the social information includes: Based on the social association degree corresponding to each of the social information, selecting each social association model parameter corresponding to each of the second devices from each of the encryption model parameters; The social association model parameters corresponding to the second devices are encrypted and aggregated respectively to obtain encrypted aggregate model parameters corresponding to the social information.

2. The federated learning modeling optimization method according to claim 1, characterized in that: The step of selecting the social association model parameters corresponding to the second devices in the encryption model parameters based on the social association degree corresponding to each social information includes: Compare the social relevance in each social information with a preset relevance threshold to obtain a relevance comparison result corresponding to each social information; According to the comparison results of the respective correlations, the encryption model parameters whose social correlations satisfy a preset correlation threshold condition are selected as the social correlation model parameters corresponding to the respective second devices.

3. A federated learning modeling optimization method, characterized in that: The federated learning modeling optimization method is applied to a second device, and the federated learning modeling optimization method includes: Iteratively train the local model according to the training sample set to obtain encryption model parameters corresponding to the local model; Sending the encrypted model parameters and the social information corresponding to the second device to the first device, so that the first device selects the social association model parameters corresponding to the second devices respectively from the encrypted model parameters based on the social association degree corresponding to the social information sent by each second device, and encrypts and aggregates the social model parameters corresponding to each second device respectively, to obtain the encrypted aggregated model parameters corresponding to each second device; Receive the encrypted aggregation model parameters fed back by the first device, and iteratively optimize the local model according to the encrypted aggregation model parameters to obtain a federated learning model.

4. The federated learning modeling optimization method according to claim 3, characterized in that: The federated learning model includes a federated recommendation model. After the step of iteratively optimizing the local model according to the encrypted aggregation model parameters to obtain the federated learning model, the federated learning modeling optimization includes: Acquire user data, and input the user data into the federated learning model to obtain user feature representation variables; Calculating user similarity results corresponding to the user feature representation variables, and generating a similar user candidate set based on the similarity results; Based on the similar user candidate set, a preset item recommendation process is executed.

5. A federated recommendation optimization method, characterized in that: The federated recommendation optimization method is applied to a second device, and the federated recommendation optimization method includes: Acquire the user data to be recommended, and input the user data to be recommended into the federated recommendation model to obtain the user feature representation variable, wherein the federated recommendation model is obtained by the second device iteratively optimizing the local model based on the encrypted aggregated model parameters sent by the first device, and the encrypted aggregated model parameters are obtained by the first device based on the social association degree corresponding to the social information corresponding to each of the second devices, and the social association model parameters corresponding to each of the second devices are selected from each encrypted model parameter, and the social model parameters corresponding to each of the second devices are encrypted and aggregated respectively; Calculating user similarity results corresponding to the user feature representation variables, and generating a similar user candidate set based on the similarity results; Based on the similar user candidate set, a preset item recommendation process is executed.

6. The federated recommendation optimization method according to claim 5, characterized in that: Before the step of obtaining the user data to be recommended and inputting the user data to be recommended into the federated recommendation model to obtain the user feature representation variable, wherein the federated recommendation model is obtained by the second device iteratively optimizing the local model based on the encrypted aggregated model parameters sent by the first device, and the encrypted aggregated model parameters are obtained by the first device selectively aggregating the encrypted model parameters of each second device based on the social information corresponding to each second device, the federated recommendation optimization method further includes: Iteratively train the local model according to the training sample set to obtain encryption model parameters corresponding to the local model; Sending the encryption model parameters and the social information corresponding to the second device to the first device, so that the first device can selectively encrypt and aggregate the encryption model parameters sent by each second device based on the social information sent by each second device, and obtain the encryption aggregation model parameters corresponding to each second device; Receive the encrypted aggregation model parameters fed back by the first device, and iteratively optimize the local model according to the encrypted aggregation model parameters to obtain the federated recommendation model.

7. A federated learning modeling optimization device, characterized in that: The federated learning modeling optimization device includes: a memory, a processor, and a program stored in the memory for implementing the federated learning modeling optimization method. The memory is used to store a program for implementing the federated learning modeling optimization method; The processor is used to execute a program for implementing the federated learning modeling optimization method to implement the steps of the federated learning modeling optimization method as described in any one of claims 1 to 2 or 3 to 4.

8. A federated recommendation optimization method device, characterized in that: The federated recommendation optimization method device includes: a memory, a processor, and a program stored in the memory for implementing the federated recommendation optimization method. The memory is used to store a program for implementing the federal recommendation optimization method; The processor is used to execute a program for implementing the federated recommendation optimization method to implement the steps of the federated learning modeling optimization method as described in any one of claims 5 to 6.

9. A medium, the medium being a readable storage medium, characterized in that: The readable storage medium stores a program for implementing the federated learning modeling optimization method, and the program for implementing the federated learning modeling optimization method is executed by a processor to implement the steps of the federated learning modeling optimization method as described in any one of claims 1 to 2 or 3 to 4.

10. A medium, the medium being a readable storage medium, characterized in that: The readable storage medium stores a program for implementing the federated recommendation optimization method, and the program for implementing the federated recommendation optimization method is executed by a processor to implement the steps of the federated recommendation optimization method as described in any one of claims 5 to 6.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the steps of the federated learning modeling optimization method as described in any one of claims 1 to 2 or 3 to 4, or implements the steps of the federated recommendation optimization method as described in any one of claims 5 to 6.

Images