WO2021232754A1 - Federated learning modeling method and device, and computer-readable storage medium - Google Patents

Abstract

A federated learning modeling method and device, and a computer-readable storage medium. The federated learning modeling method comprises: receiving encryption model parameters and verification parameters corresponding to the encryption model parameters that are sent by second devices (S10); on the basis of the verification parameters, respectively performing zero knowledge verification on the encryption model parameters so as to determine false encryption model parameters within the encryption model parameters and obtain a zero knowledge verification result (S20); and on the basis of the zero knowledge verification result and the encryption model parameters, coordinating the second devices to perform federated learning modeling (S30).

Description

Federated learning modeling method, equipment and computer readable storage medium

Cross-references to related applications

This application requires: the priority of the Chinese patent application filed on May 22, 2020, with application number 202010445868.8, and titled "Federal Learning Modeling Method, Equipment, and Readable Storage Medium", which is hereby incorporated by reference.

Technical field

This application relates to the field of artificial intelligence of financial technology (Fintech), in particular to a federated learning modeling method, equipment and computer-readable storage medium.

Background technique

With the continuous development of financial technology, especially Internet technology and finance, more and more technologies (such as distributed, blockchain, artificial intelligence, etc.) are applied in the financial field, but the financial industry has also proposed higher technology Requirements, such as the distribution of to-do items in the financial industry, also have higher requirements.

With the continuous development of computer software and artificial intelligence, the development of federated learning modeling has become more and more mature. At present, each participant in federated learning modeling usually feeds back his own encryption model parameters to the coordinator of federated learning modeling. Furthermore, it coordinates the aggregation of the encryption model parameters of the parties and feeds back the aggregated parameters to each participant for federated learning modeling. However, if a malicious party provides false encryption model parameters during the training process, it will It directly affects the overall model quality of the federated model obtained by federated learning modeling, and even leads to the failure of the entire federated learning modeling process, which in turn makes the efficiency and accuracy of federated learning modeling low.

The above content is only used to assist the understanding of the technical solution of the application, and does not mean that the above content is recognized as prior art.

Summary of the invention

The main purpose of this application is to provide a federated learning modeling method, device, and computer-readable storage medium, aiming to solve the technical problem of low efficiency and poor accuracy of federated learning modeling in the prior art.

To achieve the above objective, the present application provides a federated learning modeling method, the federated learning modeling method is applied to a first device, and the federated learning modeling method includes:

Receiving encrypted model parameters and verification parameters corresponding to the encrypted model parameters sent by each second device;

Based on each of the verification parameters, perform zero-knowledge verification on each of the encryption model parameters to determine a false encryption model parameter in each of the encryption model parameters to obtain a zero-knowledge verification result; and

Based on the zero-knowledge verification result and each of the encryption model parameters, coordinate each of the second devices to perform federated learning modeling.

In an embodiment, the step of coordinating each of the second devices to perform federated learning modeling based on the zero-knowledge verification result and each of the encryption model parameters includes:

Based on the zero-knowledge verification result, remove the false encryption model parameters from each of the encryption model parameters to obtain each trusted model parameter; and

Perform aggregation processing on each of the trusted model parameters to obtain aggregation parameters, and feed back the aggregation parameters to each of the second devices, so that each of the second devices can update their respective local training models until the The local training model reaches the preset training end condition.

In an embodiment, the step of performing zero-knowledge verification on each of the encryption model parameters based on each of the verification parameters to determine a false encryption model parameter in each of the encryption model parameters, and obtaining a zero-knowledge verification result include:

Respectively calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the verification parameters; and

Based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results, it is verified whether each encryption model parameter is a false encryption model parameter, and a zero-knowledge verification result is obtained.

In an embodiment, the verification parameters include verification model parameters and verification random parameters,

The step of separately calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the verification parameters includes:

Based on preset verification challenge parameters, perform legality verification on each of the encryption model parameters to obtain each of the first zero-knowledge verification results; and

Encryption processing is performed on each verification model parameter based on the preset coordinator public key and each of the verification random parameters to obtain each of the second zero-knowledge verification results.

In an embodiment, the preset verification challenge parameters include a first verification challenge parameter and a second verification challenge parameter;

The step of performing legality verification on each of the encryption model parameters based on preset verification challenge parameters to obtain each of the first zero-knowledge verification results includes:

Performing an exponentiation operation on the first verification challenge parameter and each of the encryption model parameters to obtain a first exponentiation operation result corresponding to each of the encryption model parameters;

Performing an exponentiation operation on the second verification challenge parameter and each of the encryption model parameters to obtain a second exponentiation operation result corresponding to each of the encryption model parameters; and

Based on each of the first power operation results and each of the second power operation results, each of the first zero-knowledge verification results is generated.

In an embodiment, the step of separately verifying whether each encryption model parameter is a false encryption model parameter based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results includes:

Respectively comparing the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the encryption model parameters;

If the first zero-knowledge proof result corresponding to the encryption model parameter is inconsistent with the second zero-knowledge proof result, it is determined that the encryption model parameter is the fake encryption model parameter; and if the encryption model parameter corresponds to If the first zero-knowledge proof result is consistent with the second zero-knowledge proof result, it is determined that the encryption model parameter is not the false encryption model parameter.

In an embodiment, the first device and the second device both perform encryption based on a homomorphic encryption algorithm, wherein,

The homomorphic encryption algorithm satisfies:

C = Enc (PK, m, r), for C ₁ = Enc (PK, m ₁ , r ₁ )) and C ₂ = Enc (PK, m ₂ , r ₂ ), satisfying:

Among them, C, C ₁ and C ₂ are the parameters to be encrypted after encryption, PK is the encryption key, m, m ₁ and m ₂ are the parameters to be encrypted, and r, r ₁ and r ₂ are the random parameters required for encryption. number.

In an embodiment, the aggregation processing is performed on each of the trusted model parameters to obtain the aggregation parameters, and the aggregation parameters are fed back to each of the second devices, so that each of the second devices can update their respective The steps for the local training model of the local training model until the local training model reaches the preset training end condition include:

Decrypt the aggregation parameters to obtain decrypted aggregation parameters;

And based on the decryption aggregation parameters, update the local training model held by oneself to obtain the updated local training model;

Determine whether the updated local training model meets the preset training end conditions;

If the updated local training model reaches the preset training end condition, it is determined that the task of the federated learning modeling is completed;

If the updated local training model does not reach the preset training end condition, the local training model is re-trained iteratively until the local model reaches the preset number of iterations threshold, then the model of the local training model is reacquired Training parameters;

Re-encrypting the model training parameters and sending them to the coordinator for re-federation until the local training model reaches a preset training end condition; and

Wherein, the training end condition includes reaching a preset maximum number of iterations, and the loss function corresponding to the local training model converges. To achieve the above objective, the present application also provides a federated learning modeling method, the federated learning modeling method is applied to a second device, and the federated learning modeling method includes:

Acquiring model training parameters and first verification random parameters, and performing encryption processing on the model training parameters based on the first verification random parameters and the preset public key to obtain encrypted model parameters;

Generating a verification model parameter and a second verification random parameter based on the first verification random parameter, the model training parameter, and the preset verification challenge parameter;

Sending the encryption model parameters, the verification model parameters, and the second verification random parameters to a first device for the first device to perform zero-knowledge verification and obtain a zero-knowledge verification result; and

Receive the aggregation parameter fed back by the first device based on the zero-knowledge verification result and the encryption model parameter, and based on the aggregation parameter, update the local training model corresponding to the model training parameter until the local training The model reaches the preset training end condition.

In an embodiment, the model training parameters include current model parameters and auxiliary model parameters,

The step of obtaining model training parameters includes:

Performing iterative training on the local training model corresponding to the model training parameters until the local training model reaches a preset iteration number threshold, and obtaining the current model parameters of the local training model; and

Obtain the previous model parameters of the local training model, and generate the auxiliary model parameters based on the previous model parameters.

The present application also provides a federated learning modeling device. The federated learning modeling device is a physical device. The federated learning modeling device includes a memory, a processor, and a device stored on the memory and available on the processor. When the program of the federated learning modeling method is executed by the processor, the steps of the federated learning modeling method can be realized.

The present application also provides a computer-readable storage medium, the computer-readable storage medium stores a program for implementing the federated learning modeling method, and the program of the federated learning modeling method is executed by a processor to realize the federation as described above. Learn the steps of modeling methods.

This application receives the encryption model parameters sent by each second device and the verification parameters corresponding to the encryption model parameters, and then performs zero-knowledge verification on each of the encryption model parameters based on each of the verification parameters, so as to perform zero-knowledge verification on each of the encryption model parameters. A false encryption model parameter is determined in the encryption model parameters to obtain a zero-knowledge verification result, and then based on the zero-knowledge verification result, each of the second devices is coordinated to perform federated learning modeling. That is, after receiving the encrypted model parameters sent by each second device and the verification parameters corresponding to the encrypted model parameters, the present application performs zero-knowledge verification on each encrypted model parameter based on each of the verification parameters, so as to A false encryption model parameter is determined in each encryption model parameter to obtain a zero-knowledge verification result. In one embodiment, based on the zero-knowledge verification result, the false encryption model parameter can be eliminated from each encryption model parameter, The federated learning modeling is performed by coordinating each of the second devices. That is, this application provides a method for determining false encryption model parameters in each of the local models based on zero-knowledge proof, and furthermore, when a malicious participant provides false encryption model parameters during the training process, it can accurately identify and The false encryption model parameters are removed, thereby avoiding the occurrence of federated learning modeling based on the encryption model parameters mixed with false encryption model parameters, thereby improving the overall model quality of the federated model obtained through federated learning modeling, and thus improving The efficiency and accuracy of federated learning modeling further solve the technical problems of low efficiency and poor accuracy of federated learning modeling.

Description of the drawings

The drawings herein are incorporated into the specification and constitute a part of the specification, show embodiments that conform to the application, and are used together with the specification to explain the principle of the application.

In order to more clearly describe the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, those of ordinary skill in the art are In other words, other drawings can be obtained based on these drawings without creative labor.

Fig. 1 is a schematic flow chart of the first embodiment of the federated learning modeling method according to the application;

FIG. 2 is a schematic flowchart of a second embodiment of a federal learning modeling method according to the application;

FIG. 3 is a schematic flowchart of a third embodiment of a federal learning modeling method according to the application;

FIG. 4 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the application.

The realization of the objectives, functional characteristics, and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application.

The embodiment of the present application provides a federated learning modeling method. In the first embodiment of the federated learning modeling method of the present application, the federated learning modeling method is applied to a first device. Referring to FIG. 1, the federated learning modeling method is Modal methods include:

Step S10, receiving encrypted model parameters sent by each second device and verification parameters corresponding to the encrypted model parameters;

In this embodiment, it should be noted that before performing federated learning modeling, the first device negotiates and interacts with each of the second devices to determine standard verification challenge parameters, where the standard verification challenge parameters The quantity can be determined during the negotiation interaction. The first device is a coordinator performing federated learning modeling, and is used to coordinate each of the second devices to perform federated learning modeling, the second device is a participant performing federated learning modeling, and the encryption model parameters It is a model training parameter that is encrypted based on a homomorphic encryption algorithm. For example, suppose that the public key of the participant held by the second device is P, and the first verification random parameter used for homomorphic encryption is r ₁ , based on the homomorphic encryption algorithm _{The encrypted model parameter h m} after encrypting the model training parameter m = Enc(P, m, r ₁ ), where Enc is a homomorphic encryption symbol. In an embodiment, the model training parameter is the second The model parameters of the local training model of the device, for example, assuming that the local training model is a linear model, and the expression is Y=β ₀ +β ₁ x ₁ +β ₂ x ₂ +...+β _n x _n , then the model The parameter is a vector (β ₀ +β ₁ +β ₂ +...+β _n ).

In addition, it should be noted that the verification parameter is a parameter for zero-knowledge proof, where the verification parameter includes a second verification random parameter and a verification model parameter, wherein the verification model parameter is the first The second device is generated based on model training parameters and verification challenge parameters. For example, assuming that the verification challenge parameters include parameters x ₁ and x ₂ , and the model training parameter is m, then the verification model parameter is n=m*x ₁ +m*x ₂ , in one embodiment, the second verification random parameter is generated by the second device based on the first verification random parameter and the verification challenge parameter. For example, suppose the first verification The random parameter is r ₁ , and the verification challenge parameter includes a parameter x ₁ and a parameter x ₂ , then the second verification random parameter

In addition, it should be noted that malicious participants in each of the participating parties can change the homomorphic encryption encryption parameters of the model training parameters to achieve the purpose of generating false encryption model parameters. When modeling, usually directly aggregate the encrypted model parameters sent by each of the second devices to obtain the aggregated parameters. If there are false encryption model parameters in each of the encrypted model parameters, it will affect the efficiency and accuracy of federated model training. For example, assuming that the encryption model parameter sent by the second device A is 5a, when the second device B is a malicious participant, the false encryption model parameter sent is 100b, and the aggregation process is a weighted average, then after the aggregation process The obtained first aggregation parameter is (5a+100b)/2. When the second device B is not a malicious participant, the sent encryption model parameter is 5b, and the second aggregation parameter obtained after the aggregation process is ( 5a+5b)/2. It can be seen that if there is a malicious participant in each second device, the The gap between the first set of parameters obtained by the time is extremely large, and if there are malicious participants in each of the second devices, the efficiency and accuracy of the federation model training will be greatly affected.

In addition, it should be noted that the first device and the second device in this embodiment both perform encryption based on a homomorphic encryption algorithm, where, in an implementable solution, the homomorphic encryption algorithm should satisfy the following nature:

C = Enc (PK, m, r), and for C ₁ = Enc (PK, m ₁ , r ₁ )) and C ₂ = Enc (PK, m ₂ , r ₂ ), satisfy:

Among them, C, C ₁ and C ₂ are the parameters to be encrypted after encryption, PK is the encryption key, m, m ₁ and m ₂ are the parameters to be encrypted, and r, r ₁ and r ₂ are the random parameters required for encryption. number.

Step S20, performing zero-knowledge verification on each of the encryption model parameters based on each of the verification parameters, so as to determine a false encryption model parameter in each of the encryption model parameters, and obtain a zero-knowledge verification result;

In this embodiment, based on each of the verification parameters, zero-knowledge verification is performed on each of the encryption model parameters to determine a false encryption model parameter in each of the encryption model parameters to obtain a zero-knowledge verification result, specifically, Based on each of the verification parameters, the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the encryption model parameters are respectively calculated, and then based on the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the encryption model parameters. Zero-knowledge proof result, respectively verifying whether each encryption model parameter is a false encryption model parameter, so as to determine the false encryption model parameter in each encryption model parameter to obtain a zero-knowledge verification result, wherein the zero-knowledge verification result is Whether each encryption model parameter is a verification result of a false encryption model parameter, the false encryption model parameter is a model training parameter that a malicious participant maliciously encrypts the model training parameter, for example, the encryption when homomorphic encryption is performed by changing Parameters to achieve malicious encryption.

Wherein, the step of performing zero-knowledge verification on each of the encryption model parameters based on each of the verification parameters to determine a false encryption model parameter in each of the encryption model parameters, and obtaining a zero-knowledge verification result includes:

Step S21, respectively calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each verification parameter;

In this embodiment, it should be noted that the verification parameters include verification model parameters and second verification random parameters.

The first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each verification parameter are respectively calculated. Specifically, the following steps are performed for each verification parameter:

Based on the preset verification challenge parameters, homomorphic addition is performed on the encryption model parameters to obtain the first zero-knowledge proof result, and based on the preset coordinator public key and the second verification random parameters, the verification model parameters are performed The homomorphic encryption operation obtains the second zero-knowledge proof result. For example, suppose that the preset verification challenge parameters are x ₁ and x ₂ , and the encryption model parameters are h _m =Enc(P ₁ , m, r ₁ ), where P ₁ Is the public key of the participant, r ₁ is the first verification random parameter, m is the model training parameter, and the verification model parameter n=m*x ₁ +m*x ₂ , and

Then the first zero-knowledge verification result is

The second zero-knowledge verification result is

Where P ₂ is the public key of the coordinator, and if each of the second devices is maliciously modified, the public key of the participant and the public key of the coordinator should be consistent.

Step S22, based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results, respectively verify whether each encryption model parameter is a false encryption model parameter, and obtain a zero-knowledge verification result.

In this embodiment, based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results, it is verified whether each encryption model parameter is a false encryption model parameter, and the zero-knowledge verification result is obtained, specifically For the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each verification parameter, the following steps are performed:

The first zero-knowledge proof result is compared with the second zero-knowledge proof result to determine whether the first zero-knowledge proof result is consistent with the second zero-knowledge proof result, if the first zero-knowledge proof result If the proof result is consistent with the second zero-knowledge proof result, it is determined that when the second device homomorphically encrypts the model training parameters, it has not been maliciously modified, that is, the encrypted model parameters are not false encrypted model parameters If the first zero-knowledge proof result is inconsistent with the second zero-knowledge proof result, it is determined that the second device performed malicious modification when homomorphic encryption of the model training parameters, that is, the encryption The model parameters are fake encrypted model parameters.

Wherein, the step of separately verifying whether each encryption model parameter is a false encryption model parameter based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results includes:

Step S221, comparing the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each encryption model parameter respectively;

In this embodiment, the first zero-knowledge proof result corresponding to each encryption model parameter is compared with the second zero-knowledge proof result, specifically, the encryption model parameter corresponding to the The first zero-knowledge proof result and the second zero-knowledge proof result are respectively compared to calculate the difference between the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each encryption model parameter. The difference.

Step S222: If the first zero-knowledge proof result corresponding to the encryption model parameter is inconsistent with the second zero-knowledge proof result, determine that the encryption model parameter is the fake encryption model parameter;

In this embodiment, if the first zero-knowledge proof result corresponding to the encryption model parameter is inconsistent with the second zero-knowledge proof result, it is determined that the encryption model parameter is the fake encryption model parameter, specifically If the difference is not 0, it is determined that the second device corresponding to the encrypted model parameter maliciously encrypts the model training parameter, and then it is determined that the encrypted model parameter is a false encrypted model parameter, and then the encrypted model The parameter is given a false identification to identify the encryption model parameter as a false encryption model parameter.

Step S223: If the first zero-knowledge proof result corresponding to the encryption model parameter is consistent with the second zero-knowledge proof result, it is determined that the encryption model parameter is not the false encryption model parameter.

In this embodiment, if the first zero-knowledge proof result corresponding to the encryption model parameter is consistent with the second zero-knowledge proof result, it is determined that the encryption model parameter is not the false encryption model parameter, specifically Ground, if the difference is 0, it is determined that the second device corresponding to the encrypted model parameter has not maliciously encrypted the model training parameter, and then it is determined that the encrypted model parameter is not a false encrypted model parameter, and then the encrypted model The parameter is given a credible identifier to identify the encrypted model parameter as a credible model parameter.

Step S30, based on the zero-knowledge verification result and each of the encryption model parameters, coordinate each of the second devices to perform federated learning modeling.

In this embodiment, it should be noted that the zero-knowledge verification result includes a calibration identifier corresponding to each of the encryption model parameters, where the calibration identifier is an identifier that identifies whether the encryption model parameter is a false encryption model parameter. .

Based on the zero-knowledge verification result and each of the encryption model parameters, coordinate each of the second devices to perform federated learning modeling, specifically, based on each of the calibration identifiers, remove false encryption models from each of the encryption model parameters Parameters, each trusted model parameter is obtained, and then each of the trusted model parameters is aggregated to obtain the aggregated parameter, and then based on each of the aggregated model parameters, each of the second devices is coordinated to perform federated learning modeling.

Wherein, the step of coordinating each of the second devices to perform federated learning modeling based on the zero-knowledge verification result and each of the encryption model parameters includes:

Step S31, based on the zero-knowledge verification result, remove the false encryption model parameters from each encryption model parameter to obtain each trusted model parameter;

In this embodiment, it should be noted that after discovering the false encryption model parameters, the coordinator can punish the malicious participant corresponding to the false encryption model parameter according to the preset incentive mechanism or cancel the malicious participant's subsequent participation in the federated learning modeling Qualifications.

Step S32, performing aggregation processing on each of the trusted model parameters to obtain aggregation parameters, and feeding back the aggregation parameters to each of the second devices, so that each of the second devices can update their respective local training models, Until the local training model reaches the preset training end condition.

In this embodiment, each of the trusted model parameters is aggregated to obtain the aggregated parameters, and the aggregated parameters are fed back to each of the second devices, so that each of the second devices can update their local Train the model until the local training model reaches a preset training end condition, specifically, based on preset aggregation processing rules, perform aggregation processing on each of the trusted model parameters to obtain aggregation parameters, wherein the preset aggregation processing The rules include weighted average, summation, etc. In one embodiment, each of the aggregation parameters is sent to each of the second devices, so that each of the second devices can participate based on the public key of the participant. The party’s private key decrypts the aggregation parameters to obtain the decryption aggregation parameters, and based on the decryption aggregation parameters, updates the local training model held by the party, obtains the updated local training model, and judges the updated local training model Whether the preset training end condition is reached, if the updated local training model meets the preset training end condition, it is determined that the task of federated learning modeling is completed, and if the updated local training model does not meet the preset training end condition, then Re-train the local training model iteratively until the local model reaches the threshold of the preset number of iterations, then re-acquire the model training parameters of the local training model, and re-encrypt and send the model training parameters to the The coordinator re-feds until the local training model reaches a preset training end condition, where the training end condition includes reaching a preset maximum number of iterations, and the loss function corresponding to the local training model converges.

In one embodiment, the local training model includes a risk control model, where the risk control model is a machine learning model used to assess the risk of a user’s loan, and when there is a malicious participant in each of the participants, then Perform aggregation based on the false encryption model parameters sent by the malicious participants and the encryption model parameters sent by the normal participants in each of the participants, and the error aggregation parameters that differ greatly from the accurate aggregation parameters will be obtained, and then each The second device updates the risk control model based on the error aggregation parameters, which will reduce the accuracy of the risk control model for user loan risk assessment. Furthermore, if it is based on the federated learning modeling method in this application, the encrypted model can be sent from each participant The false encryption model parameters are screened and eliminated from the parameters to obtain the trusted encryption model parameters. Then, in the entire federated learning modeling process, the risk control model is always based on the aggregate parameters obtained by the aggregate processing of the trusted encryption model parameters. The update makes the risk control model's assessment of user loan risk more accurate, that is, improves the accuracy of the risk control model's loan risk assessment.

In this embodiment, by receiving the encryption model parameters sent by each second device and the verification parameters corresponding to the encryption model parameters, and then based on each of the verification parameters, perform zero-knowledge verification on each of the encryption model parameters, so as to perform zero-knowledge verification on each of the encryption model parameters. In the encryption model parameters, a false encryption model parameter is determined to obtain a zero-knowledge verification result, and then based on the zero-knowledge verification result, each of the second devices is coordinated to perform federated learning modeling. That is, in this embodiment, after receiving the encryption model parameters sent by each second device and the verification parameters corresponding to the encryption model parameters, based on each of the verification parameters, zero-knowledge verification is performed on each of the encryption model parameters respectively, so as to Determine a false encryption model parameter in each encryption model parameter to obtain a zero-knowledge verification result. In one embodiment, based on the zero-knowledge verification result, the false encryption model parameter can be eliminated from each encryption model parameter , To coordinate each of the second devices to perform federated learning modeling. That is, this embodiment provides a method for determining false encryption model parameters in each of the local models based on zero-knowledge proof, and furthermore, when a malicious participant provides false encryption model parameters during the training process, it can accurately identify And remove the false encryption model parameters, thereby avoiding the occurrence of federated learning modeling based on the encryption model parameters mixed with false encryption model parameters, thereby improving the overall model quality of the federated model obtained through federated learning modeling, and thereby improving It improves the efficiency and accuracy of federated learning modeling, and then solves the technical problems of low efficiency and poor accuracy of federated learning modeling.

In an embodiment, referring to FIG. 2, based on the first embodiment of the present application, in another embodiment of the present application, the verification parameters include verification model parameters and verification random parameters,

The step of separately calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the verification parameters includes:

Step S211: Perform legality verification on each of the encryption model parameters based on preset verification challenge parameters to obtain each of the first zero-knowledge verification results;

In this embodiment, it should be noted that the preset verification challenge parameters include a first verification challenge parameter and a second verification challenge parameter, and the encryption model parameters include a first encryption model parameter and a second encryption model parameter, where , The first encryption model parameter is the encryption parameter after the second device homomorphically encrypts the current model parameter, and the second encryption model parameter is after the second device homomorphically encrypts the previous model parameter Where the current model parameter is the model parameter extracted when the local training model reaches the threshold of the preset number of training iterations during the current round of federation, and the previous model parameter is based on the previous round of federation The model parameters of the previous round of federation are, for example, taking the historical model parameters corresponding to the previous three rounds of federation, and performing a weighted average of each of the historical model parameters to obtain the previous model parameters, for example, assuming that The current model parameter is m, the previous model parameter is m ₀ , then the first encrypted model parameter h ₀ =Enc(P, m ₀ , r ₁ ), where P is the public key of the participant, and r ₁ is the first encryption model parameter. One verification random parameter, the second encryption model parameter h _m =Enc(P, m, r ₂ ), where r ₂ is the second verification random parameter.

Based on the preset verification challenge parameters, the legality verification is performed on each encryption model parameter, and each of the first zero-knowledge verification results is obtained. Specifically, the following steps are performed for each encryption model parameter:

Based on the first verification challenge parameter and the second verification challenge parameter, the first encryption model parameter and the second encryption model parameter are respectively exponentiated and summed to obtain the first zero-knowledge verification result, for example, suppose The first encryption model parameter h ₀ =Enc(P, m ₀ , r ₁ ), the second encryption model parameter h _m = Enc(P, m, r ₂ ), the first verification challenge parameter is x ₁ , and the second verification challenge The parameter is x ₂ , then the first zero-knowledge proof result

Based on the nature of the homomorphic encryption algorithm, we can get

Where P is the public key of the participant, x ₁ is the first verification challenge parameter, x ₂ is the second verification challenge parameter, r ₁ is the first verification random parameter, r ₂ is the second verification random parameter, the current model parameter Is m, and the previous model parameter is m ₀ .

Wherein, the preset verification challenge parameters include a first verification challenge parameter and a second verification challenge parameter;

The step of performing legality verification on each of the encryption model parameters based on preset verification challenge parameters to obtain each of the first zero-knowledge verification results includes:

Step A10, performing exponentiation operations on the first verification challenge parameter and each encryption model parameter, respectively, to obtain a first exponentiation operation result corresponding to each encryption model parameter;

In this embodiment, the first verification challenge parameter and each encryption model parameter are respectively subjected to exponentiation operation to obtain the first exponentiation operation result corresponding to each encryption model parameter. Specifically, for each encryption model parameter, For model parameters, the following steps are performed: based on the first verification challenge parameter, the first encryption model parameter is exponentiated to obtain the first power operation result. For example, assuming that the first verification challenge parameter is x, If the first encryption model parameter is h, the result of the first power operation is h ^x .

Step A20, performing exponentiation operations on the second verification challenge parameter and each encryption model parameter, respectively, to obtain a second exponentiation operation result corresponding to each encryption model parameter;

In this embodiment, the second verification challenge parameter and each encryption model parameter are respectively subjected to exponentiation operation to obtain the second exponentiation operation result corresponding to each encryption model parameter. Specifically, for each encryption model parameter For the model parameters, the following steps are performed: based on the second verification challenge parameter, an exponentiation operation is performed on the second encryption model parameter to obtain a second exponentiation operation result.

Step A30, generating each of the first zero-knowledge verification results based on each of the first power operation results and each of the second power operation results.

In this embodiment, based on each of the first power operation results and each of the second power operation results, each of the first zero-knowledge verification results is generated, and specifically, the first power operation result and the result of the second power operation are obtained. The product of the second power operation result, and use the product as the first zero-knowledge verification result.

Step S212: Perform encryption processing on each verification model parameter based on the preset coordinator public key and each of the verification random parameters to obtain each of the second zero-knowledge verification results.

In this embodiment, it should be noted that each of the participants is a legal participant, and the public key of the participant is consistent with the public key of the preset coordinator, and the verification random parameter includes a third verification random Parameters, and the third verification random parameter is calculated based on the first verification random parameter, the second verification random parameter, the first verification challenge parameter, and the second verification challenge parameter. For example, assuming that the first verification random parameter is r1, The second verification random parameter is r2, the first verification challenge parameter is x1, the second verification challenge parameter is x2, then the third verification challenge parameter

Additionally, the verification model parameters are calculated based on the first verification challenge parameters, the second verification challenge parameters, the current model parameters, and the previous model parameters. For example, assuming that the first verification challenge parameter is x1, the second verification challenge parameter is x1. The challenge parameter is x2, the current model parameter is m, and the previous model parameter is m ₀ , then the verification model parameter is n=m ₀ x ₁ +mx ₂ .

Based on the preset public key of the coordinator and each of the verification random parameters, each verification model parameter is encrypted to obtain each of the second zero-knowledge verification results. Specifically, each verification model parameter is executed The following steps:

Based on the preset coordinator public key and the third verification challenge parameter, homomorphic encryption is performed on the verification model parameter to obtain the second zero-knowledge verification result, for example, suppose the third verification challenge parameter

Wherein, the first verification random parameter is r1, the second verification random parameter is r2, the first verification challenge parameter is x1, the second verification challenge parameter is x2, and the verification model parameter is n=m ₀ x ₁ +mx ₂ , The default coordinator’s public key is P, then

In one embodiment, if the participant does not maliciously encrypt the encryption model parameters, for example, maliciously tampering with the encryption algorithm, maliciously tampering with the encrypted parameters, etc., the first zero-knowledge proof result is the same as the second zero-knowledge proof result. That is, the encryption model parameter provided by the participant to the coordinator is a trusted model parameter. If the participant maliciously encrypts the encryption model parameter, the first zero-knowledge proof result is different from the second zero-knowledge proof result That is, the encryption model parameter provided by the participant to the coordinator is a false encryption model parameter.

In this embodiment, by performing legality verification on each of the encryption model parameters based on preset verification challenge parameters, each of the first zero-knowledge verification results is obtained, and then based on the preset coordinator public key and each of the verification random parameters, Encryption processing is performed on each of the verification model parameters to obtain each of the second zero-knowledge verification results. That is, this embodiment provides a method for calculating the first zero-knowledge proof result and the second zero-knowledge proof result, and then after the first zero-knowledge proof result and the second zero-knowledge proof result are obtained by calculation, the first zero-knowledge proof result and the second zero-knowledge proof result are calculated. The knowledge proof result and the second zero-knowledge proof result can only be compared to determine whether the encryption model parameters are false encryption model parameters, which lays a foundation for determining the false encryption model parameters in each of the encryption model parameters, and then solves the problem of federation. It laid the foundation for learning the technical problems of low modeling efficiency and poor accuracy.

In an embodiment, referring to FIG. 3, based on the first and second embodiments of the present application, in another embodiment of the present application, the federated learning modeling method is applied to a second device, and the federated Learning modeling methods include:

Step B10: Obtain model training parameters and first verification random parameters, and perform encryption processing on the model training parameters based on the first verification random parameters and the preset public key to obtain encrypted model parameters;

In this embodiment, the federated learning modeling includes at least one round of federation, and in each round of federation, the second device performs iterative training on the local training model until the threshold of the preset number of iterations is reached. The model parameters of the local training model are sent to the first device, and the aggregation parameters fed back by the first device based on the model parameters are connected, and based on the aggregation parameters, the local training model is updated, and all the parameters are updated. The local training model serves as the initial model of the next round of federation until the local training model reaches a preset training end condition, where the preset training end condition includes reaching the maximum number of iterations, convergence of the loss function, and the like.

Obtain model training parameters and first verification random parameters, and perform encryption processing on the model training parameters based on the first verification random parameters and the preset public key to obtain encrypted model parameters, specifically, when the local training model When the threshold of the preset number of iterations is reached, the model parameters of the local training model are extracted as the model training parameters, and the first verification random parameters are obtained, and then based on the first verification random parameters and the preset public key, the model training parameters are Perform homomorphic encryption processing to obtain encryption model parameters. For example, assuming that the model training parameter is m, the first verification random parameter is r ₁ , and the preset public key is P, then the encryption model parameter is h _m = Enc(P, m, r ₁ ), where Enc is a homomorphic encryption symbol.

Wherein, the model training parameters include current model parameters and auxiliary model parameters,

The step of obtaining model training parameters includes:

Step B11: Perform iterative training on the local training model corresponding to the model training parameters until the local training model reaches a preset iteration number threshold, and obtain the current model parameters of the local training model;

In this embodiment, it should be noted that the current model parameter is the current iteration model parameter when the local training model reaches the preset iteration number threshold in the current round of federation.

Step B12: Obtain the previous model parameters of the local training model, and generate the auxiliary model parameters based on the previous model parameters.

In this embodiment, the previous model parameters of the local training model are acquired, and the auxiliary model parameters are generated based on the previous model parameters, specifically, the previous-round federation corresponding to the current round of federation is obtained Each of the previous iteration model parameters of each of the previous iteration model parameters is weighted and averaged to obtain the auxiliary model parameters. For example, assuming that each of the previous iteration model parameters is a, b, c, a corresponding to The weight of is 20%, the weight of b is 30%, and the weight of c is 50%, then the auxiliary model parameter m ₀ =a*20%+b*30%+c*50%.

Step B20, generating a verification model parameter and a second verification random parameter based on the first verification random parameter, the model training parameter, and the preset verification challenge parameter;

In this embodiment, it should be noted that, in a feasible implementation, the preset verification challenge parameters can be encrypted by the coordinator according to the previous encryption sent by each participant in the previous round of the federation. The model parameters and the preset hash function are calculated. For example, if there are 10 participants, then the corresponding 10 previously encrypted model parameters are freely combined, and then the n results of the free combination are input to the preset In the hash function, the verification challenge parameters x ₁ and x _{2 are obtained} . x _n , and specific preset verification challenge parameters x ₁ , x ₂ . The generation method and number of x _{n are not limited.}

Based on the first verification random parameter, the model training parameter, and the preset verification challenge parameter, a verification model parameter and a second verification random parameter are generated, and specifically, the first verification random parameter and the preset verification challenge The parameters perform the exponentiation operation to obtain the second verification random parameter, and based on the model training parameters and preset verification challenge parameters, the verification model parameters are generated, for example, assuming that the first verification random parameter is r ₁ , the model The training parameter is m, the preset verification challenge parameters are x ₁ and x ₂ , the verification model parameter n=m*x ₁ +m*x ₂ , the second verification random parameter

Step B30, sending the encryption model parameters, the verification model parameters, and the second verification random parameters to a first device for the first device to perform zero-knowledge verification and obtain a zero-knowledge verification result;

In this embodiment, the encryption model parameters, the verification model parameters, and the second verification random parameters are sent to a first device for the first device to perform zero-knowledge verification and obtain a zero-knowledge verification result, Specifically, the encryption model parameters, the verification model parameters, and the second verification random parameters are sent to a first device associated with the second device, so that the first device can base on the encryption model parameters , The verification model parameter and the second verification random parameter, calculate the first zero-knowledge proof result and the second zero-knowledge proof result, and determine the result based on the first zero-knowledge proof result and the second zero-knowledge proof result Whether the encryption model parameter is a false encryption model parameter, a determination result is obtained, and the determination result is recorded in the zero-knowledge verification result, where the zero-knowledge verification result includes the determination result corresponding to each of the second devices.

Step B40: Receive the aggregation parameter fed back by the first device based on the zero-knowledge verification result and the encryption model parameter, and update the local training model corresponding to the model training parameter based on the aggregation parameter, until all The local training model reaches the preset training end condition.

In this embodiment, the aggregation parameter fed back by the first device based on the zero-knowledge verification result and the encryption model parameter is received, and based on the aggregation parameter, the local training model corresponding to the model training parameter is updated , Until the local training model reaches the preset training end condition, specifically, after the first device obtains the zero-knowledge proof result, the encrypted model sent by each second device will be based on the zero-knowledge proof result The false encryption model parameters are removed from the parameters, and each trusted model parameter is obtained, and each trusted model parameter is aggregated. The aggregation processing includes summation, weighted averaging, etc., to obtain aggregated parameters, and aggregate The parameters are fed back to each of the second devices respectively, and then after the second device ends the aggregation parameters, it decrypts the aggregation parameters based on the preset private key corresponding to the preset public key to obtain the decrypted aggregation parameters, And based on the decryption aggregation parameters, the local training model is updated, and the updated local training model is used as the initial model of the next round of federation until the local training model reaches the preset training end condition. The training end conditions include reaching the maximum number of iterations and the convergence of the loss function.

In this embodiment, the model training parameters and the first verification random parameters are obtained, and based on the first verification random parameters and the preset public key, the model training parameters are encrypted to obtain encrypted model parameters, and then based on the first verification random parameters. 1. Verify random parameters, the model training parameters, and preset verification challenge parameters, generate verification model parameters and second verification random parameters, and then send the encryption model parameters, the verification model parameters, and the second verification random parameters To the first device for the first device to perform zero-knowledge verification to obtain a zero-knowledge verification result, and then to receive the aggregate parameter fed back by the first device based on the zero-knowledge verification result and the encryption model parameter, and based on The aggregation parameter updates the local training model corresponding to the model training parameter until the local training model reaches a preset training end condition. That is, this embodiment provides a federated learning modeling method based on zero-knowledge proof, that is, when the model training parameters are encrypted into encrypted model parameters, the verification model parameters and the second verification dependent parameters are generated at the same time, and then the The encryption model parameters, the verification model parameters, and the second verification random parameters are sent to the first device for the first device to perform zero-knowledge verification to obtain a zero-knowledge verification result, and then the first device The false encryption model parameters are determined and eliminated in each of the second devices, and then the aggregate parameters received by the second device are obtained by aggregation based on the trusted encryption model parameters by the first device, and further based on the aggregate parameters, The local training model is updated to complete the federated learning modeling, thereby avoiding the update of the local training model based on the aggregated parameters of each encryption model parameter aggregation mixed with false encryption model parameters, making it difficult for the local training model to reach the preset training end conditions and local The low determination of the training model occurs, which in turn improves the efficiency and accuracy of federated learning modeling, thereby solving the technical problem of low efficiency and poor accuracy of federated learning modeling.

Referring to FIG. 4, FIG. 4 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application.

As shown in FIG. 4, the federated learning modeling device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to implement connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as a magnetic disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

In an embodiment, the federated learning modeling 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, and so on. The rectangular user interface may include a display screen (Display) and an input sub-module 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 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

Those skilled in the art can understand that the structure of the federated learning modeling device shown in FIG. 4 does not constitute a limitation on the federated learning modeling device, and may include more or fewer components than shown in the figure, or a combination of certain components, Or different component arrangements.

As shown in FIG. 4, the memory 1005 as a computer storage medium may include an operating system, a network communication module, and a federated learning modeling method program. The operating system is a program that manages and controls the hardware and software resources of the federated learning modeling equipment, and supports the running of the federated learning modeling method program and other software and/or programs. The network communication module is used to realize the communication between the various components in the memory 1005 and the communication with other hardware and software in the federated learning modeling method system.

In the federated learning modeling device shown in FIG. 4, the processor 1001 is used to execute the federated learning modeling method program stored in the memory 1005 to implement the steps of the federated learning modeling method described in any one of the above.

The specific implementation of the federated learning modeling device of the present application is basically the same as each embodiment of the above-mentioned federated learning modeling method, and will not be repeated here.

An embodiment of the present application also provides a federated learning modeling device, the federated learning modeling device is applied to a first device, and the federated learning modeling device includes:

A receiving module, configured to receive encrypted model parameters sent by each second device and verification parameters corresponding to the encrypted model parameters;

The zero-knowledge verification module is configured to perform zero-knowledge verification on each of the encryption model parameters based on each of the verification parameters, so as to determine a false encryption model parameter in each of the encryption model parameters, and obtain a zero-knowledge verification result;

The coordination module is configured to coordinate each of the second devices to perform federated learning modeling based on the zero-knowledge verification result and each of the encryption model parameters.

In an embodiment, the coordination module includes:

The elimination sub-module is used to eliminate the false encryption model parameters from the encryption model parameters based on the zero-knowledge verification result to obtain the trusted model parameters;

The aggregation sub-module is configured to perform aggregation processing on each of the trusted model parameters to obtain aggregation parameters, and feed back the aggregation parameters to each of the second devices, so that each of the second devices can update their local The model is trained until the local training model reaches a preset training end condition.

In an embodiment, the zero-knowledge verification module includes:

The calculation sub-module is used to calculate the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each verification parameter;

The zero-knowledge verification sub-module is configured to verify whether each encryption model parameter is a false encryption model parameter based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results, and obtain a zero-knowledge verification result.

In an embodiment, the calculation sub-module includes:

A legitimacy verification unit, configured to verify the legitimacy of each encryption model parameter based on preset verification challenge parameters, and obtain each of the first zero-knowledge verification results;

The encryption unit is configured to perform encryption processing on each verification model parameter based on the preset coordinator public key and each of the verification random parameters to obtain each of the second zero-knowledge verification results.

In an embodiment, the legality verification unit includes:

The first power operation subunit is configured to perform power operations on the first verification challenge parameter and each of the encryption model parameters to obtain the first power operation result corresponding to each of the encryption model parameters;

The second power operation subunit is configured to perform power operations on the second verification challenge parameter and each of the encryption model parameters to obtain the second power operation result corresponding to each of the encryption model parameters;

A generating subunit is configured to generate each of the first zero-knowledge verification results based on each of the first power operation results and each of the second power operation results.

In an embodiment, the zero-knowledge verification sub-module includes:

A comparison unit, configured to compare the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each encryption model parameter respectively;

A first determining unit, configured to determine that the encryption model parameter is the false encryption model parameter if the first zero-knowledge proof result corresponding to the encryption model parameter is inconsistent with the second zero-knowledge proof result;

The second determination unit is configured to determine that the encryption model parameter is not the false encryption model parameter if the first zero-knowledge proof result corresponding to the encryption model parameter is consistent with the second zero-knowledge proof result.

The specific implementation of the federated learning modeling device of the present application is basically the same as each embodiment of the above-mentioned federated learning modeling method, and will not be repeated here.

To achieve the objective, this embodiment also provides a federated learning modeling device, the federated learning modeling device is applied to a second device, and the federated learning modeling device includes:

An encryption module, configured to obtain model training parameters and first verification random parameters, and based on the first verification random parameters and a preset public key, perform encryption processing on the model training parameters to obtain encrypted model parameters;

A generating module, configured to generate a verification model parameter and a second verification random parameter based on the first verification random parameter, the model training parameter, and the preset verification challenge parameter;

A sending module, configured to send the encryption model parameters, the verification model parameters, and the second verification random parameters to a first device for the first device to perform zero-knowledge verification and obtain a zero-knowledge verification result;

The model update module is configured to receive the aggregate parameter fed back by the first device based on the zero-knowledge verification result and the encryption model parameter, and update the local training model corresponding to the model training parameter based on the aggregate parameter , Until the local training model reaches the preset training end condition.

In an embodiment, the encryption module includes:

An obtaining sub-module, configured to perform iterative training on the local training model corresponding to the model training parameters until the local training model reaches a preset iteration number threshold, and obtain the current model parameters of the local training model;

A generating sub-module is used to obtain the previous model parameters of the local training model, and generate the auxiliary model parameters based on the previous model parameters.

The specific implementation of the federated learning modeling device of the present application is basically the same as each embodiment of the above-mentioned federated learning modeling method, and will not be repeated here.

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent processing of this application.

Claims (20)

1. A federated learning modeling method, wherein the federated learning modeling method is applied to a first device, and the federated learning modeling method includes:

   Receiving encrypted model parameters and verification parameters corresponding to the encrypted model parameters sent by each second device;

   Based on each of the verification parameters, perform zero-knowledge verification on each of the encryption model parameters to determine a false encryption model parameter in each of the encryption model parameters to obtain a zero-knowledge verification result; and

   Based on the zero-knowledge verification result and each of the encryption model parameters, coordinate each of the second devices to perform federated learning modeling.
2. 5. The federated learning modeling method according to claim 1, wherein the step of coordinating each of the second devices to perform federated learning modeling based on the zero-knowledge verification result and each of the encryption model parameters comprises:

   Based on the zero-knowledge verification result, remove the false encryption model parameters from each of the encryption model parameters to obtain each trusted model parameter; and

   Perform aggregation processing on each of the trusted model parameters to obtain aggregation parameters, and feed back the aggregation parameters to each of the second devices, so that each of the second devices can update their respective local training models until the The local training model reaches the preset training end condition.
3. 5. The federated learning modeling method according to claim 1, wherein said zero-knowledge verification is performed on each of said encryption model parameters based on each of said verification parameters, so as to determine a false encryption model parameter in each of said encryption model parameters , The steps to obtain zero-knowledge verification results include:

   Respectively calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the verification parameters; and

   Based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results, it is verified whether each encryption model parameter is a false encryption model parameter, and a zero-knowledge verification result is obtained.
4. The federated learning modeling method according to claim 3, wherein the verification parameters include verification model parameters and verification random parameters,

   The step of separately calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the verification parameters includes:

   Based on preset verification challenge parameters, perform legality verification on each of the encryption model parameters to obtain each of the first zero-knowledge verification results; and

   Encryption processing is performed on each verification model parameter based on the preset coordinator public key and each of the verification random parameters to obtain each of the second zero-knowledge verification results.
5. 5. The federated learning modeling method according to claim 4, wherein the preset verification challenge parameters include a first verification challenge parameter and a second verification challenge parameter;

   The step of performing legality verification on each of the encryption model parameters based on preset verification challenge parameters to obtain each of the first zero-knowledge verification results includes:

   Performing exponentiation operations on the first verification challenge parameter and each encryption model parameter to obtain a first exponentiation operation result corresponding to each encryption model parameter;

   Performing an exponentiation operation on the second verification challenge parameter and each of the encryption model parameters to obtain a second exponentiation operation result corresponding to each of the encryption model parameters; and

   Based on each of the first power operation results and each of the second power operation results, each of the first zero-knowledge verification results is generated.
6. 5. The federated learning modeling method according to claim 3, wherein the verification is based on each of the first zero-knowledge proof results and each of the second zero-knowledge proof results to verify whether each of the encryption model parameters is a false encryption model The parameters of the steps include:

   Respectively comparing the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the encryption model parameters;

   If the first zero-knowledge proof result corresponding to the encryption model parameter is inconsistent with the second zero-knowledge proof result, determining that the encryption model parameter is the fake encryption model parameter; and

   If the first zero-knowledge proof result corresponding to the encryption model parameter is consistent with the second zero-knowledge proof result, it is determined that the encryption model parameter is not the false encryption model parameter.
7. The federated learning modeling method according to claim 1, wherein the first device and the second device both perform encryption based on a homomorphic encryption algorithm, wherein,

   The homomorphic encryption algorithm satisfies:

   C = Enc (PK, m, r), for C ₁ = Enc (PK, m ₁ , r ₁ )) and C ₂ = Enc (PK, m ₂ , r ₂ ), satisfying:

   

   Among them, C, C ₁ and C ₂ are the parameters to be encrypted after encryption, PK is the encryption key, m, m ₁ and m ₂ are the parameters to be encrypted, and r, r ₁ and r ₂ are the random parameters required for encryption. number.
8. 5. The federated learning modeling method according to claim 3, wherein said aggregation processing is performed on each of said trusted model parameters to obtain aggregation parameters, and said aggregation parameters are fed back to each of said second devices respectively for Each of the second devices updates their local training model until the local training model reaches a preset training end condition. The steps include:

   Decrypt the aggregation parameters to obtain decrypted aggregation parameters;

   And based on the decryption aggregation parameters, update the local training model held by oneself to obtain the updated local training model;

   Determine whether the updated local training model meets the preset training end conditions;

   If the updated local training model reaches the preset training end condition, it is determined that the task of the federated learning modeling is completed;

   If the updated local training model does not reach the preset training end condition, the local training model is re-trained iteratively until the local model reaches the preset number of iterations threshold, then the model of the local training model is reacquired Training parameters; and

   Re-encrypting the model training parameters and sending them to the coordinator for re-federation until the local training model reaches a preset training end condition;

   Wherein, the training end condition includes reaching a preset maximum number of iterations, and the loss function corresponding to the local training model converges.
9. A federated learning modeling method, wherein the federated learning modeling method is applied to a second device, and the federated learning modeling method includes:

   Acquiring model training parameters and first verification random parameters, and performing encryption processing on the model training parameters based on the first verification random parameters and the preset public key to obtain encrypted model parameters;

   Generating a verification model parameter and a second verification random parameter based on the first verification random parameter, the model training parameter, and the preset verification challenge parameter;

   Sending the encryption model parameters, the verification model parameters, and the second verification random parameters to a first device for the first device to perform zero-knowledge verification and obtain a zero-knowledge verification result; and

   Receive the aggregation parameter fed back by the first device based on the zero-knowledge verification result and the encryption model parameter, and based on the aggregation parameter, update the local training model corresponding to the model training parameter until the local training The model reaches the preset training end condition.
10. 9. The federated learning modeling method according to claim 9, wherein the model training parameters include current model parameters and auxiliary model parameters,

    The step of obtaining model training parameters includes:

    Performing iterative training on the local training model corresponding to the model training parameters until the local training model reaches a preset iteration number threshold, and obtaining the current model parameters of the local training model; and

    Obtain the previous model parameters of the local training model, and generate the auxiliary model parameters based on the previous model parameters.
11. A federated learning modeling device, wherein the federated learning modeling device includes a memory, a processor, and a program stored on the memory for implementing the federated learning modeling method,

    The memory is used to store a program for implementing the federated learning modeling method;

    The processor is used to execute a program for implementing the federated learning modeling method to achieve:

    Receiving encrypted model parameters and verification parameters corresponding to the encrypted model parameters sent by each second device;

    Based on each of the verification parameters, perform zero-knowledge verification on each of the encryption model parameters to determine a false encryption model parameter in each of the encryption model parameters to obtain a zero-knowledge verification result; and

    Based on the zero-knowledge verification result and each of the encryption model parameters, coordinate each of the second devices to perform federated learning modeling.
12. The federated learning modeling device according to claim 11, wherein, in the processor for executing the program for realizing the federated learning modeling method, the verification result based on the zero-knowledge and each of the encryption model parameters , The steps of coordinating each of the second devices to perform federated learning modeling include:

    Based on the zero-knowledge verification result, remove the false encryption model parameters from each of the encryption model parameters to obtain each trusted model parameter; and

    Perform aggregation processing on each of the trusted model parameters to obtain aggregation parameters, and feed back the aggregation parameters to each of the second devices, so that each of the second devices can update their respective local training models until the The local training model reaches the preset training end condition.
13. The federated learning modeling device according to claim 11, wherein, in the processor for executing the program for realizing the federated learning modeling method, based on each of the verification parameters, each of the encrypted model parameters is performed separately Zero-knowledge verification, to determine the false encryption model parameters in each of the encryption model parameters, and the steps of obtaining the zero-knowledge verification result include:

    Respectively calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the verification parameters; and

    Based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results, it is verified whether each encryption model parameter is a false encryption model parameter, and a zero-knowledge verification result is obtained.
14. The federated learning modeling device according to claim 13, wherein, in the processor for executing the program for realizing the federated learning modeling method, the verification parameters include verification model parameters and verification random parameters, and the respective The step of calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each verification parameter includes:

    Based on preset verification challenge parameters, perform legality verification on each of the encryption model parameters to obtain each of the first zero-knowledge verification results; and

    Encryption processing is performed on each verification model parameter based on the preset coordinator public key and each of the verification random parameters to obtain each of the second zero-knowledge verification results.
15. The federated learning modeling device according to claim 14, wherein, in the processor for executing the program for realizing the federated learning modeling method, the preset verification challenge parameters include a first verification challenge parameter and a second verification challenge parameter. Verification challenge parameters; the step of performing legality verification on each of the encryption model parameters based on preset verification challenge parameters, and obtaining each of the first zero-knowledge verification results includes:

    Performing an exponentiation operation on the first verification challenge parameter and each of the encryption model parameters to obtain a first exponentiation operation result corresponding to each of the encryption model parameters;

    Performing an exponentiation operation on the second verification challenge parameter and each of the encryption model parameters to obtain a second exponentiation operation result corresponding to each of the encryption model parameters; and

    Based on each of the first power operation results and each of the second power operation results, each of the first zero-knowledge verification results is generated.
16. A computer-readable storage medium, wherein a program for implementing the federated learning modeling method is stored on the computer-readable storage medium, and the program for implementing the federated learning modeling method is executed by a processor to realize:

    Receiving encrypted model parameters and verification parameters corresponding to the encrypted model parameters sent by each second device;

    Based on each of the verification parameters, perform zero-knowledge verification on each of the encryption model parameters to determine a false encryption model parameter in each of the encryption model parameters to obtain a zero-knowledge verification result; and

    Based on the zero-knowledge verification result and each of the encryption model parameters, coordinate each of the second devices to perform federated learning modeling.
17. The federated learning modeling device according to claim 16, wherein, in the processor for executing the program for realizing the federated learning modeling method, the verification result based on the zero-knowledge and each of the encryption model parameters , The steps of coordinating each of the second devices to perform federated learning modeling include:

    Based on the zero-knowledge verification result, remove the false encryption model parameters from each of the encryption model parameters to obtain each trusted model parameter; and

    Perform aggregation processing on each of the trusted model parameters to obtain aggregation parameters, and feed back the aggregation parameters to each of the second devices, so that each of the second devices can update their respective local training models until the The local training model reaches the preset training end condition.
18. The federated learning modeling device according to claim 16, wherein, in the processor for executing the program for realizing the federated learning modeling method, based on each of the verification parameters, each of the encryption model parameters is performed separately Zero-knowledge verification, to determine the false encryption model parameters in each of the encryption model parameters, and the steps of obtaining the zero-knowledge verification result include:

    Respectively calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each of the verification parameters; and

    Based on each of the first zero-knowledge certification results and each of the second zero-knowledge certification results, it is verified whether each encryption model parameter is a false encryption model parameter, and a zero-knowledge verification result is obtained.
19. The federated learning modeling device according to claim 18, wherein, in the processor for executing the program for realizing the federated learning modeling method, the verification parameters include verification model parameters and verification random parameters, and the respective The step of calculating the first zero-knowledge proof result and the second zero-knowledge proof result corresponding to each verification parameter includes:

    Based on preset verification challenge parameters, perform legality verification on each of the encryption model parameters to obtain each of the first zero-knowledge verification results; and

    Encryption processing is performed on each verification model parameter based on the preset coordinator public key and each of the verification random parameters to obtain each of the second zero-knowledge verification results.
20. The federated learning modeling device according to claim 19, wherein, in the processor for executing the program for realizing the federated learning modeling method, the preset verification challenge parameters include a first verification challenge parameter and a second verification challenge parameter. Verification challenge parameters; the step of performing legality verification on each of the encryption model parameters based on preset verification challenge parameters, and obtaining each of the first zero-knowledge verification results includes:

    Performing an exponentiation operation on the first verification challenge parameter and each of the encryption model parameters to obtain a first exponentiation operation result corresponding to each of the encryption model parameters;

    Performing an exponentiation operation on the second verification challenge parameter and each of the encryption model parameters to obtain a second exponentiation operation result corresponding to each of the encryption model parameters; and

    Based on each of the first power operation results and each of the second power operation results, each of the first zero-knowledge verification results is generated.

Images