CN114329544A

CN114329544A - Model application method, block chain client, computing platform, block chain and equipment

Info

Publication number: CN114329544A
Application number: CN202111620281.7A
Authority: CN
Inventors: 程龙; 李浩然; 李艳鹏; 陆旭明
Original assignee: Ant Blockchain Technology Shanghai Co Ltd
Current assignee: Ant Blockchain Technology Shanghai Co Ltd
Priority date: 2021-12-27
Filing date: 2021-12-27
Publication date: 2022-04-12

Abstract

The embodiment of the specification provides a model application method, a block chain client, a computing platform, a block chain and equipment. The method comprises the following steps: the blockchain client sends a participation request of the target federated learning plan to the blockchain. And calling a recruitment intelligent contract deployed according to the target federal learning plan by the blockchain, adding the blockchain client to the target federal learning plan, recording the identification and the public key of the blockchain client, and sending the public key of the blockchain client to a computing platform of the target federal learning plan. And the block chain client trains the local learning model based on the private data, and sends the training result carrying the private key signature model to the computing platform. And the computing platform verifies the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan, and integrates at least two model parameters in the model training results verified by the private key signature to obtain the target learning model.

Description

Model application method, block chain client, computing platform, block chain and equipment

Technical Field

The present document belongs to the technical field of information processing, and in particular, relates to a model application method, a block chain client, a computing platform, a block chain, and a device.

Background

Learning models based on artificial intelligence technology are gaining more and more attention from mechanisms by virtue of the capability of mechanical information processing. Data is the basis of model construction, and for most industries, data used for training a learning model belongs to customer information in the industry, so when the need of jointly developing the model exists among organizations, federal learning is usually selected to complete modeling. Federal learning is a mode of joint training model proposed on the premise of ensuring data privacy among organizations.

The federal study is not completed all at once, and many factors need to be considered in practical application, such as: how to achieve recruitment of member subjects, such as proving credible data, how to ensure that model training is fair, and the like. Obviously, it is a technical problem that needs to be solved at present to specially design an operation architecture aiming at the mechanism of federal learning.

Disclosure of Invention

Embodiments of the present description aim to provide a model application method, a blockchain client, a computing platform, a blockchain, and a device, where a blockchain is introduced into a framework of federal learning, so that the blockchain client can be recruited quickly to participate in a federal learning plan by using advantages of the blockchain, and an identity certificate of a member object of the federal learning plan is configured for the blockchain client, so that a local model training result can be provided by using the blockchain client on the basis of the identity certificate of the member object when the computing platform models. Obviously, the development efficiency of the federal learning can be greatly improved by applying the framework of the federal learning, and the opportunity of participating in the joint modeling is provided for more organizations.

In order to achieve the above object, the embodiments of the present specification are implemented as follows:

in a first aspect, a model application method is provided, including:

a block chain client sends a participation request of a target federal learning plan to a block chain, wherein the participation request carries an identifier and a public key of the block chain client;

the block chain responds to the participation request, invokes a recruitment intelligent contract deployed aiming at the target federal learning plan, adds the block chain client to the target federal learning plan, records the identification and the public key of the block chain client, and sends the public key of the block chain client to a computing platform of the target federal learning plan;

the block chain client trains a local learning model based on private data and sends a model training result to the computing platform, wherein the model training result of the block chain client carries model parameters after the learning model training and a private key signature of the block chain client; the computing platform verifies the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan;

and the computing platform integrates model parameters in at least two model training results which pass through the private key signature verification to obtain a target learning model.

In a second aspect, a model application method is provided, including:

the block chain client trains a learning model of the local target federal learning plan based on private data, and sends a model training result to the computing platform, wherein the model training result of the block chain client carries model parameters after the learning model training and a private key signature of the block chain client, and the computing platform checks the private key signature in each obtained model training result based on a public key of a member object of the target federal learning plan, and integrates the model parameters in at least two model training results which pass through the private key signature check to obtain the target learning model.

In a third aspect, a model application method is provided, including:

a computing platform of a target federal learning plan receives a public key of a blockchain client sent by a blockchain, wherein the public key of the blockchain client is sent by the blockchain to add the blockchain client to the target federal learning plan and is used for proving that the blockchain client belongs to a member object in the target federal learning plan;

the method comprises the steps that a computing platform of a target federal learning plan receives a model training result sent by a member object of the target federal learning plan, wherein the model training result sent by the member object of the target federal learning plan carries model parameters after the member object sends a learning model which is trained on the local target federal learning plan based on private data and a private key signature sent by the member object, and the member object of the target federal learning plan comprises a block chain client;

the computing platform verifies the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan;

In a fourth aspect, a model application method is provided, including:

receiving a participation request aiming at a target federal learning plan, which is sent by a blockchain client, by the blockchain, wherein the participation request carries an identifier and a public key of the blockchain client;

and the block chain responds to the participation request, calls a recruitment intelligent contract deployed aiming at the target federal learning plan, adds the block chain client to the target federal learning plan, records the identification and the public key of the block chain client, and sends the public key of the block chain client to a computing platform of the target federal learning plan.

In a fifth aspect, a blockchain client is provided, which comprises

The participation request module is used for sending a participation request of a target federal learning plan to the block chain, and the participation request carries the identification and the public key of the block chain client;

the federated learning execution module trains a local learning model of the target federated learning plan based on private data and sends a model training result to the computing platform, wherein the model training result of the block chain client carries model parameters after learning model training and the private key signature of the block chain client, and the computing platform verifies the private key signature in each obtained model training result based on the public key of the member object of the target federated learning plan and integrates the model parameters in at least two model training results verified by the private key signature to obtain the target learning model.

In a sixth aspect, there is provided an electronic device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:

sending a participation request of a target federal learning plan to a block chain, wherein the participation request carries an identifier and a public key of a block chain client;

based on private data, training a learning model of the local target federal learning plan, and sending a model training result to the computing platform, wherein the model training result of the blockchain client carries model parameters after the learning model training and a private key signature of the blockchain client, and the computing platform checks the private key signature in each obtained model training result based on a public key of a member object of the target federal learning plan, and integrates model parameters in at least two model training results which pass through the private key signature check to obtain the target learning model.

In a seventh aspect, a computer-readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the steps of:

In an eighth aspect, a computing platform is provided, comprising

The certification information receiving module is used for receiving a public key of a block chain client side sent by a block chain, wherein the public key of the block chain client side is sent by the block chain through adding the block chain client side to a target federal learning plan and is used for certifying that the block chain client side belongs to a member object in the target federal learning plan;

the model training result receiving module is used for receiving a model training result sent by a member object of a target federal learning plan, wherein the model training result sent by the member object of the target federal learning plan carries model parameters after the member object sends a learning model which is trained on the local target federal learning plan based on private data and a private key signature sent by the member object, and the member object of the target federal learning plan comprises the block chain client;

the core body detection module is used for verifying the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan;

and the model parameter integration module is used for integrating the model parameters in at least two model training results which pass through the private key signature verification to obtain the target learning model.

In a ninth aspect, there is provided an electronic device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:

receiving a public key of a blockchain client sent by a blockchain, wherein the public key of the blockchain client is sent by the blockchain when the blockchain adds the blockchain client to a target federal learning plan and is used for proving that the blockchain client belongs to a member object in the target federal learning plan;

receiving a model training result sent by a member object of a target federal learning plan, wherein the model training result sent by the member object of the target federal learning plan carries model parameters sent by the member object and a private key signature sent by the member object after the member object sends private data to train a learning model of the local target federal learning plan, and the member object of the target federal learning plan comprises the block chain client;

verifying the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan;

and integrating the model parameters in at least two model training results which pass through the private key signature verification to obtain a target learning model.

In a tenth aspect, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:

In an eleventh aspect, there is provided a blockchain, comprising

The participation request receiving module is used for receiving a participation request aiming at a target federal learning plan, which is sent by a blockchain client, wherein the participation request carries an identifier and a public key of the blockchain client;

and the federal learning plan filing module is used for responding to the participation request, calling a recruitment intelligent contract deployed aiming at the target federal learning plan, adding the block chain client to the target federal learning plan, filing the identification and the public key of the block chain client, and sending the public key of the block chain client to a computing platform of the target federal learning plan.

In a twelfth aspect, there is provided an electronic device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:

receiving a participation request aiming at a target federal learning plan, which is sent by a blockchain client, wherein the participation request carries an identifier and a public key of the blockchain client;

and responding to the participation request, calling a recruitment intelligent contract deployed aiming at the target federal learning plan, adding the blockchain client to the target federal learning plan, recording the identification and the public key of the blockchain client, and sending the public key of the blockchain client to a computing platform of the target federal learning plan.

A thirteenth aspect provides a computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a processor, implementing the steps of:

According to the scheme of the embodiment of the specification, a federal learning plan is issued through a block chain, and an organization can apply for participation to the block chain after logging in a block chain client, so that member objects can be recruited quickly by means of the propagation advantages of the block chain. Meanwhile, the block chain is in butt joint with the computing platform, a public key of the block chain client is sent to the computing platform to serve as a proof that the block chain client participates in the federal learning plan, and after a subsequent computing platform receives a training result of a local federal learning model which is sent by the block chain client and carries a private key signature, the private key signature in the training result can be verified through the public key of the block chain client, so that whether the block chain client belongs to a member object of the federal learning plan or not is verified, and whether the provided training result is credible or not is proved. For a computing platform, after the credible model training results are accumulated to a certain degree, a modeling task can be executed, and finally a target learning model with high reliability is obtained. It can be seen that under the federal learning architecture in the embodiments of the present specification, an organization can participate in a joint learning plan more conveniently by means of a blockchain, thereby improving the usability of federal learning.

Drawings

In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only some embodiments described in the embodiments of the present specification, and for a person of ordinary skill in the relevant art, other drawings can be obtained according to the drawings without inventive labor.

Fig. 1 is a first flowchart of a model application method provided in an embodiment of the present disclosure.

Fig. 2 is a second flowchart of a model application method provided in an embodiment of the present disclosure.

Fig. 3 is a schematic structural diagram of a blockchain client according to an embodiment of the present disclosure.

Fig. 4 is a schematic structural diagram of a computing platform provided in an embodiment of the present specification.

Fig. 5 is a schematic structural diagram of a block chain provided in an embodiment of the present disclosure.

Fig. 6 is a schematic structural diagram of an electronic device provided in an embodiment of this specification.

Detailed Description

In order to make those skilled in the art better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only a part of the embodiments of the present specification, and not all of the embodiments. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in the present specification without any creative efforts shall fall within the protection scope of the present specification.

As mentioned above, currently, when there is a need for developing models together among organizations, federal learning is usually selected to complete modeling. The federal study is not performed on kick, and in practical application, many factors need to be considered to meet the mechanism requirement of the federal study, such as: how to implement recruitment of member subjects, how to prove trustworthiness of data, how to ensure that the model is just trained, etc. In order to reduce the threshold of the federal learning application, the document specially designs an architecture for realizing the federal learning based on the mechanism of the federal learning, and can quickly develop a federal learning plan based on the architecture, thereby facilitating the participation of organizations in the joint modeling.

FIG. 1 is a flow diagram of a model application method of one embodiment of the present description, including:

s102, the blockchain client sends a participation request of the target federal learning plan to the blockchain, and the participation request carries the identification and the public key of the blockchain client.

In this embodiment, the blockchain may be used as a leading party of the federal learning plan and is responsible for target federal learning plan creation and release. The entity that needs to collaborate in the modeling sends a request for participation in the target federated learning plan to the blockchain in the identity of the blockchain client.

Alternatively, the federal learning plan may be dominated by a third party, that is, the target federal learning in this step is created by the third party and distributed in a blockchain, which may be regarded as a channel for registering participation in the target federal learning plan.

And S104, responding to the participation request by the blockchain, calling a recruitment intelligent contract deployed aiming at the target federal learning plan, adding the blockchain client to the target federal learning plan, recording the identification and the public key of the blockchain client, and sending the public key of the blockchain client to a computing platform of the target federal learning plan.

In the embodiment of the specification, the computing platform is responsible for integrating the training results of the member objects to determine a final target learning model of the target federal learning plan. If the target federated learning plan is dominated by the computing platform side, i.e., the computing platform publishes the target federated learning plan through the blockchain, the computing platform need not prove trustworthiness to the blockchain and member objects. If the target federated learning plan is dominated by the blockchain party, the computing platform may be specified by the blockchain party, i.e., the trustworthiness of the computing platform is demonstrated by the dominating party specifying means.

It should be understood that a smart contract is a type of computing program executing code. The recruitment intelligence contract of the embodiment of the specification is obtained by compiling execution logic related to a blockchain recruiting member objects according to a target federal learning plan into codes through a mechanical language. Specifically, the blockchain call recruitment intelligence contract may perform the following flow:

and auditing the qualification of the blockchain client participating in the target federal learning plan based on the blockchain data corresponding to the blockchain client, such as: and checking the credit information, the risk information, the client level information and the like of the block chain client. If the qualification of the block chain client passes the audit, adding the block chain client to the target federal learning plan; otherwise, the block chain client is denied addition to the target federated learning plan.

After the blockchain client is added to the target federal learning plan, the targeted target federal learning plan is recorded on the blockchain client, and the identification and the public key of the blockchain client are recorded on the record. After being put on record, the blockchain client can become a member object of a legal target federal learning plan.

In addition, after completing the record of the blockchain client, the public key of the blockchain client needs to be sent to the computing platform. For the computing platform, the public key of the blockchain client sent by the blockchain may prove that the blockchain client belongs to the member object of the target federal learning plan, that is, the public key of the blockchain client is also regarded as the public key of one member object of the target federal learning plan by the computing platform.

S106, the block chain client trains the local learning model based on the private data, and sends a model training result to the computing platform, wherein the model training result of the block chain client carries the model parameters after the learning model training and the private key signature of the block chain client.

In the embodiment of the specification, the target federated learning plan requires a plurality of member objects to collaboratively participate in building the target learning model. The model construction process is that the computing platform provides an initial learning model of the target federal learning plan for each member object, and the member objects use respective private data to train the local initial learning model. Because the private data of each member object are different, the model parameters obtained by training different member objects are different. And the computing platform is responsible for collecting model parameters finished by training of each member object, and integrating the differentiated model parameters to obtain a final target learning model.

Specifically, after participating in the target federal learning plan, the blockchain client may send a model acquisition request for the target federal learning plan to the computing platform, where the model acquisition request carries a private key signature of the blockchain client. Then, the computing platform verifies the private key signature of the blockchain client in the model acquisition request by using the public key of the blockchain client acquired in S104. If the private key signature of the blockchain client in the model acquisition request passes verification, the computing platform determines that the blockchain client belongs to the member object of the target federal learning plan initial, and sends the learning model of the target federal learning plan initial to the blockchain client.

Similarly, the blockchain client sends its own model training result to the computing platform, and needs to carry its own private key signature of the blockchain client, so that the computing platform checks the private key signature in the model training result by using the public key of the blockchain client obtained in S104. If the private key signature in the model training result passes verification, the computing platform determines that the block chain client providing the model training result belongs to the member object of the target Federal learning plan initiation, and judges that the model training result provided by the computing platform has credibility.

And S108, the computing platform verifies the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan.

Based on the verification design of the asymmetric encryption algorithm of public and private keys, the computing platform can identify whether the provider of each model training result belongs to a member object of a target federal learning plan. In the embodiment of the specification, the computing platform only uses the model training result provided by the member object of the target federal learning plan to carry out modeling, so that the condition that other people pretend to provide unrealistic model parameters by the member object to interfere with the modeling result is avoided. It should be understood that the model parameters in the model training result refer to the result data obtained after the learning model is trained, such as the weight values of the feature vectors in the learning model. The result data is not private data used by local training learning models of organizations, is relatively low in sensitivity and cannot damage data safety.

It should also be noted that if the target federal learning plan is created by a computing platform and is distributed to a blockchain, the blockchain may serve as a conduit for the computing platform to recruit member objects. That is, the member object of the target federated learning plan may also include non-blockchain clients that are enrolled in other channels to participate in the target federated learning plan. The member objects of the non-blockchain clients can also provide model training results to the computing platform to participate in modeling. Here, the description is not repeated by way of example. In the target federal learning plan S110, the computing platform integrates model parameters in at least two model training results verified by private key signatures, so as to obtain a target learning model.

It should be understood that the integration of model parameters falls within the category of mathematical quantification, and is not specifically limited herein since the implementation is not unique. By way of exemplary introduction, the computing platform may integrate the model parameters in a weighted summation manner to obtain target model parameters, and configure the learning model based on the target model parameters to obtain a final target learning model.

Such as: the larger the business scale is, the more comprehensive the collected private data samples of the training learning model are; the higher the model training capability of the member object, the higher the contribution degree of the member object to the alliance, so that the influence degree of the member objects belonging to the two classes on the target science model construction can be improved. That is, in the weighted summation process, the weighting factor corresponding to the model parameter provided by any target member object should be determined based on the business scale index (which can be quantified by information such as transaction amount and customer volume) of the target member object and the contribution degree index for target federal learning.

According to the method, a federal learning plan is released through a block chain, and an organization can apply for participation to the block chain after logging in a block chain client, so that member objects can be recruited quickly by means of the propagation advantages of the block chain. Meanwhile, the block chain is in butt joint with the computing platform, a public key of the block chain client is sent to the computing platform to serve as a proof that the block chain client participates in the federal learning plan, and after a subsequent computing platform receives a training result of a local federal learning model which is sent by the block chain client and carries a private key signature, the private key signature in the training result can be verified through the public key of the block chain client, so that whether the block chain client belongs to a member object of the federal learning plan or not is verified, and whether the provided training result is credible or not is proved. For a computing platform, after the credible model training results are accumulated to a certain degree, a modeling task can be executed, and finally a target learning model with high reliability is obtained. It can be seen that under the federal learning architecture in the embodiments of the present specification, an organization can participate in a joint learning plan more conveniently by means of a blockchain, thereby improving the usability of federal learning. In addition, after the target learning model is constructed and obtained based on the target federal learning, the advantages of block chain decentralization and incapability of tampering can be utilized, and member objects and other clients participating in the target federal learning can be assisted to apply the correct target learning model to actual production.

The application of the block chain-based target learning model is described below with reference to two implementations.

Implementation mode one

After the target learning model is generated, the computing platform encrypts model parameters of the final target learning model based on an irreversible encryption algorithm or an asymmetric encryption algorithm to obtain an electronic deposit certificate corresponding to the target learning model, and uploads the electronic deposit certificate to a block chain through block chain transaction.

The target learning model can be released by a network and used offline after being downloaded by member objects in the target federal learning plan or other authorized members.

Here, taking the member object in the target federal learning plan as an example, the member object may obtain the electronic deposit certificate of the target learning model from the blockchain after the member object is saved and downloaded to the target learning model. And then, the member object in the target federal learning plan checks the model parameters of the downloaded target learning model based on the electronic deposit certificate acquired from the blockchain so as to judge whether the target learning model downloaded from the network is tampered.

If the target learning model passes the verification, the member objects in the target federated learning plan can perform offline application on the target learning model. Conversely, if the target learning model fails to verify, the member object in the target federated learning plan may choose to obtain the target federated learning in other ways, for example, obtain the target learning model from the computing platform of the target federated learning.

Implementation mode two

After the computing platform generates the target learning model, compiling execution logic predicted by using the target learning model into a predicted intelligent contract according to the mechanical language, and deploying the intelligent contract into a block chain to realize online application.

When a block chain client (which may also be an authorized non-member object block chain client) in a target federal learning plan needs to use a target learning model to predict an object to be predicted, a prediction request carrying feature information of the object to be predicted can be submitted to a block chain to call a block chain trigger prediction intelligent contract, feature data of the object to be predicted is predicted, and a prediction result is sent to the block chain client.

The method of the present embodiment is described in detail below with reference to an actual application scenario.

The application scene is based on an artificial intelligence technology, and mechanical prediction of insurance premium is provided for the block chain client. The insurance premium is thousands of people and thousands of prices, the premium for buying insurance of the insurance application target by different insurance applicants is different due to various factors, customized pricing is carried out through a premium inquiry model, and the insurance department, the insurance applicants and the platform win more. Traditional insurance uses different centralized actuarial tables according to different risk types, such as: "Life tables" and "incidence tables of major diseases" to price premium pricing. Because different insurance companies are based on the unchanged actuarial model similar to the above when pricing the insurance fee, the homogenization of the insurance product is increasingly aggravated; by using a model construction technology, a risk judgment model for individuals can be trained, risk levers of different subjects are estimated through the model, so that differentiated pricing of insurance rates of different groups is realized, and the problem that the accuracy of determining the risk levers is lost on the basis of the total occurrence probability of a traditional insurance product is solved; however, the traditional model construction concentrates data to a server side, and relies on a centralized data set to train a corresponding model by running a model construction algorithm. At present, with the requirements and emphasis of national laws on user privacy data protection, the centralized model construction algorithm faces a huge challenge of privacy protection.

Therefore, the application scene provides an insurance premium intelligent enquiry scheme based on block chain and federal learning, and the safety problem of user privacy data in the process of modeling the price model by using a machine learning technology under the cross-trust domain scene is solved through the processes of data collection, model training, model integration and model storage; during model training, a training task is transferred to a user side, only model parameters obtained through training are sent to credit granting identification for fusion in the model integration process, and privacy data of the user still remain in the local, so that the privacy of the data is guaranteed in a physical isolation mode. In the process of storing and using the model, the block chain provides a credible mechanism for each participant (user) of federal learning, and users which are not credible can be integrated together as the participants through the authorization mechanism, identity management and the like of the block chain, particularly a alliance chain, so that a safe and credible data sharing and using mechanism is established; finally, in the aspect of mode usage, the federally learned model parameters can be stored in the world state of the block chain in the mode of original survivor anchor or intelligent contract, the former provides credible anchor for the offline downloading model of the user, and the latter provides safety and reliability of data access for the real-time price inquiry of the user chain through the intelligent contract.

Referring to fig. 2, the whole process is divided into four stages of data collection, model training, model integration and model application, wherein:

first, data collection phase

In the data collection phase, the blockchain issues the target federal learning plan externally. If the blockchain client desires to participate in the target federated learning plan, a participation request for the target federated learning plan may be sent to the blockchain, where the participation request requires the identification and public key of the blockchain client to be provided to the blockchain. The blockchain client which successfully participates in the target federal learning plan is responsible for collecting data required by the target federal learning task and is filed as a member object (trust domain) of the federal learning by the blockchain, and the filing information comprises: a blockchain client identification (trust domain identification) and a public key of the blockchain client.

Second, model training phase

In the model training stage, each trust domain (block chain client) of the federal learning is allowed to train each learning model by using different private data or even algorithm, and model parameters of the trained learning model are coordinated with a public key of the model parameters to be sent to a computing platform serving as credit granting equipment.

The computing platform can use a stacking method (stacking) to fuse model parameters obtained by training different trust domains to obtain a final target learning model, and upload a hash certificate of the model parameters of the target learning model to a block chain for different trust domains to perform tamper-proof verification on a subsequently obtained local target learning model. On the other hand, the computing platform can compile the operation process of the target learning model into an execution code of a mechanical language through a code programming rule in the m2cgen library, and deploy a prediction intelligent contract carrying the execution code into a block chain, so that a block chain client can call the target learning model to perform related prediction in a mode of initiating a block chain transaction on line. The method comprises the following specific steps:

training local models of the trust domains according to local data sources, signing the locally trained model files (carrying model parameters) by using private keys corresponding to the recorded public keys, and generating signature model files carrying the private keys. Then, the blockchain client of each trust domain transmits the model file and the signed digital signature to a computing platform of a trusted environment, where the trusted environment refers to a running environment that is implemented by any technical means and is not interfered by other factors, such as a Trusted Execution Environment (TEE), an on-chain private computing environment, an off-chain private computing environment, and the like, and is not specifically limited herein.

And after receiving the model file, the computing platform of the trusted environment verifies the validity of the digital signature.

Third, model fusion phase

The model fusion stage is executed by a computing platform of a trusted environment, and the specific process is as follows:

(1) a stacking ensemble learning algorithm is used to integrate the model parameters for the various trust domains.

The Stacking algorithm example procedure is as follows:

step 1, dividing a training set D into k parts for a base model 1 (a base model of a certain member object which is not trained locally), training the model by using a residual data set for each part, and then predicting the result of the part.

And 2, repeating the steps until each part is predicted to obtain a training set of the secondary model.

And 3, obtaining k test sets, and averaging to obtain the test set of the secondary model.

And 4, repeating the steps for the base model 2 and the base model 3 … to obtain m-dimensional model parameters.

And 5, integrating the m-dimensional model parameter data to determine a final model. Here, the application scenario does not limit the specific implementation of the model, but a logistic regression model may be preferably used in consideration of the difficulty and stability of the model training.

(2) And converting the integrated model into a file in a pkl format.

(3) Using the m2cgen library, the operational logic of the model file is written as execution code to act as a chain of blocks to apply a predictive intelligence contract to the target learning model.

Fourth, model application phase

The model application phase is divided into two modes of using under the chain and using on the chain.

Used under chain:

(1) and the computing platform of the trusted environment computes the Hash value of the fused model file and uploads the Hash value to the block chain for storage.

(2) And when each trust domain downloads and uses the fused model file, calculating the Hash value of each trust domain, and verifying and comparing the Hash values provided by the block chains.

(3) If the comparison is consistent, the downloaded model file is not tampered, and corresponding prediction is completed through the model file.

On-chain use:

(1) and the computing platform of the trusted environment writes the application process of the model file into an intelligent contract to be deployed on the block chain.

(2) The block chain informs the block chain client of each trust domain that the intelligent contract of the target learning model is deployed, and when the block chain client of the trust domain has the demand of premium inquiry, the block chain transaction (the transaction carries the characteristic information of model prediction premium) is submitted to the intelligent contract so that the block chain triggers the intelligent contract, and the premium prediction is completed based on the characteristic information of the premium in the block chain transaction.

It should be understood that the application scenario of premium inquiry is an exemplary introduction of a mode of only operating a federated learning and block chain implementation model, and is not limited to the protection scope of this document, and common artificial intelligence recognition scenarios such as risk recognition and identity recognition may be applied to the solution of the embodiment of the present description. Suitable changes may be made in the steps of the method of embodiments without departing from the principles described herein, and such changes are to be considered within the scope of the embodiments of the present disclosure.

In addition, the embodiment of the present specification further provides a blockchain client corresponding to the method shown in fig. 3. Fig. 3 is a schematic structural diagram of a blockchain client 300 according to an embodiment of the present disclosure, including:

the participation request module 310 sends a participation request of the target federal learning plan to the blockchain, where the participation request carries the identifier and the public key of the blockchain client.

The federal learning execution module 320 trains a local learning model of the target federal learning plan based on private data, and sends a model training result to the computing platform, wherein the model training result of the block chain client carries model parameters after learning model training and private key signatures of the block chain client, and the computing platform checks the private key signatures in the obtained model training results based on the public keys of member objects of the target federal learning plan, and integrates the model parameters in at least two model training results which pass through private key signature checking to obtain the target learning model.

Optionally, the blockchain client further includes:

and the model acquisition module is used for sending a model acquisition request to the computing platform, wherein the model acquisition request carries the private key signature of the block chain client, and receiving the initial learning model of the target federal learning plan fed back by the computing platform. After the private key signature of the block chain client in the model acquisition request passes verification, the computing platform sends the initial learning model of the target federal learning plan to the block chain client; the model training result of the blockchain client is obtained by training the initial learning model of the target federal learning plan based on private data.

Optionally, the compiling, by the computing station, the execution logic predicted by using the target learning model into a predicted intelligent contract and deploying the intelligent contract to the blockchain client in the blockchain further includes:

and the prediction request module is used for sending a prediction request of the target learning model to the block chain, wherein the prediction request carries the characteristic information of the object to be predicted so as to call the block chain to trigger the intelligent prediction contract, predict the characteristic data of the object to be predicted and send a prediction result to the block chain client.

Obviously, the blockchain client in the embodiment of the present specification may be used as an execution main body of the step of the corresponding blockchain client in the method shown in fig. 1, and the principle is the same, and therefore, the details are not described herein again.

In addition, corresponding to the method shown in fig. 1, the embodiment of the present specification further provides a computing platform for federated learning. Fig. 4 is a schematic structural diagram of a computing platform 400 according to an embodiment of the present specification, including:

the certification information receiving module 410 is configured to receive a public key of a blockchain client sent by a blockchain, where the public key of the blockchain client is sent by the blockchain to add the blockchain client to a target federal learning plan, and is used to certify that the blockchain client belongs to a member object in the target federal learning plan.

The model training result receiving module 420 is configured to receive a model training result sent by a member object of a target federal learning plan, where the model training result sent by the member object of the target federal learning plan carries model parameters after the member object sends a learning model trained on the local target federal learning plan based on private data and a private key signature sent by the member object, and the member object of the target federal learning plan includes the blockchain client.

And the core body detection module 430 is used for verifying the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan.

The model parameter integration module 430 integrates model parameters in at least two model training results passing through the private key signature verification to obtain a target learning model.

Optionally, the computing platform further comprises:

and the evidence storage module is used for generating an electronic evidence corresponding to the target learning model after the model parameter integration module 430 obtains the target learning model, and uploading the electronic evidence to the block chain through block chain transaction.

Optionally, the computing platform further comprises:

and the intelligent contract deployment module is used for compiling the execution logic predicted by using the target learning model into a predicted intelligent contract according to the mechanical language after the model parameter integration module 430 obtains the target learning model, and deploying the intelligent contract into the block chain.

Obviously, the computing platform in fig. 4 in the embodiment of this specification may be used as an execution subject of the steps corresponding to the computing platform in the method shown in fig. 1, and the principle is the same, and therefore, the details are not described herein again.

In addition, the embodiment of the present specification further provides a block chain corresponding to the method shown in fig. 1. Fig. 5 is a schematic structural diagram of a block chain 500 according to an embodiment of the present specification, including:

the participation request receiving module 510 receives a participation request for the target federal learning plan sent by the blockchain client, where the participation request carries the identifier and the public key of the blockchain client.

And a federal learning plan filing module 520, which is used for responding to the participation request, calling a recruitment intelligent contract deployed for the target federal learning plan, adding the blockchain client to the target federal learning plan, filing the identification and the public key of the blockchain client, and sending the public key of the blockchain client to a computing platform of the target federal learning plan.

Optionally, the computing platform compiles, in a mechanical language, execution logic predicted using a target learning model into a predicted intelligent contract and deploys the intelligent contract into the blockchain. The block chain of the embodiment of the present specification further includes: the model application module is used for receiving a prediction request which is submitted by the block chain client and carries the characteristic information of the object to be predicted, wherein the prediction request carries the characteristic information of the object to be predicted; and responding to the prediction request, calling the prediction intelligent contract, predicting the characteristic data of the object to be predicted, and sending a prediction result to the block chain client.

Optionally, the federal learning plan docket module 520, in response to the participation request, invokes a recruitment intelligence contract for the target federal learning plan deployment that further performs: and before adding the block chain client to the target federal learning plan, auditing the qualification of the block chain client participating in the target federal learning plan based on the block chain data corresponding to the block chain client, wherein if the qualification of the block chain client participating in the target federal learning plan passes the audit, the block chain client is added to the target federal learning plan and is put on record.

Obviously, the block chain in fig. 5 in the embodiment of this specification may be used as an execution main body of the step of the corresponding block chain in the method shown in fig. 1, and the principle is the same, so that the details are not described herein again.

Fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present specification. Referring to fig. 6, at a hardware level, the electronic device includes a processor, and optionally further includes an internal bus, a network interface, and a memory. The Memory may include a Memory, such as a high-speed Random-Access Memory (RAM), and may further include a non-volatile Memory, such as at least 1 disk Memory. Of course, the electronic device may also include hardware required for other services.

The processor, the network interface, and the memory may be connected to each other via an internal bus, which may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one double-headed arrow is shown in FIG. 6, but that does not indicate only one bus or one type of bus.

And the memory is used for storing programs. In particular, the program may include program code comprising computer operating instructions. The memory may include both memory and non-volatile storage and provides instructions and data to the processor.

The processor reads a corresponding computer program from the nonvolatile memory into the memory and then runs the computer program, and the blockchain client shown in fig. 1 is formed on a logic level. The processor is used for executing the program stored in the memory and is specifically used for executing the following operations:

and sending a participation request of a target federated learning plan to the blockchain, wherein the participation request carries the identification and the public key of the blockchain client.

In the target federal learning plan, or the processor reads the corresponding computer program from the nonvolatile memory into the memory and runs the computer program, so that the computing platform shown in fig. 4 is formed on a logic level. The processor is used for executing the program stored in the memory and is specifically used for executing the following operations:

and receiving a public key of a blockchain client sent by a blockchain, wherein the public key of the blockchain client is sent by the blockchain client when the blockchain client is added to a target federal learning plan and is used for proving that the blockchain client belongs to a member object in the target federal learning plan.

Receiving a model training result sent by a member object of a target federated learning plan, wherein the model training result sent by the member object of the target federated learning plan carries model parameters after the member object sends a learning model for the local target federated learning plan based on private data and a private key signature sent by the member object, and the member object of the target federated learning plan comprises the block chain client.

And verifying the private key signature in each obtained model training result based on the public key of the member object of the target federal learning plan.

Alternatively, the processor reads a corresponding computer program from the non-volatile memory into the memory and then runs the computer program, forming the block chain shown in fig. 5 on a logical level. The processor is used for executing the program stored in the memory and is specifically used for executing the following operations:

receiving a participation request aiming at a target federated learning plan sent by a blockchain client, wherein the participation request carries an identifier and a public key of the blockchain client.

The steps performed by the blockchain client, the steps performed by the computing platform, or the steps performed by the blockchain client, in the target federated learning scheme as disclosed in the embodiment shown in fig. 1 of this specification may be implemented in and implemented by a processor. The processor may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in a processor or instructions in the form of software. The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components. The various methods, steps and logic blocks disclosed in the embodiments of the present specification may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the embodiments of the present specification may be embodied directly in a hardware decoding processor, or in a combination of hardware and software modules in the decoding processor. The software module may be located in ram, flash memory, rom, prom, or eprom, registers, etc. storage media as is well known in the art. The storage medium is located in a memory, and a processor reads information in the memory and completes the steps of the method in combination with hardware of the processor.

It should be understood that the electronic device of the embodiment of the present specification can implement the functions of the embodiment of the forensics method shown in fig. 1. Since the principle is the same, the detailed description is omitted here.

Of course, besides the software implementation, the electronic device in this specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may also be hardware or logic devices.

Furthermore, the present specification embodiments also propose a computer-readable storage medium storing one or more programs, the one or more programs including instructions.

Wherein the instructions, when executed by a portable electronic device comprising a plurality of applications, enable the portable electronic device to perform the steps of the embodiment shown in fig. 1, which are performed with respect to the blockchain client, include:

Alternatively, the instructions, when executed by a portable electronic device comprising a plurality of application programs, can cause the portable electronic device to perform the steps of the embodiment shown in fig. 1, which are performed with respect to a computing platform, including:

Still alternatively, the above instructions, when executed by a portable electronic device including a plurality of application programs, can cause the portable electronic device to perform the steps performed with respect to the block chain in the embodiment shown in fig. 1, including:

As will be appreciated by one of ordinary skill in the art, the embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and changes may occur to those skilled in the art to which it pertains. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification. Moreover, all other embodiments obtained by persons of ordinary skill in the art without making any inventive step shall fall within the scope of protection of this document.

Claims

1. A model application method, comprising:

the block chain client trains a local learning model based on private data and sends a model training result to the computing platform, wherein the model training result of the block chain client carries model parameters after the learning model training and a private key signature of the block chain client;

2. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,

before the blockchain client trains a learning model of the local target federal learning plan based on private data, the method further includes:

the block chain client sends a model obtaining request to the computing platform, wherein the model obtaining request carries a private key signature of the block chain client;

the computing platform verifies the private key signature of the blockchain client in the model acquisition request based on the public key of the blockchain client acquired from the blockchain; and the number of the first and second groups,

and after the private key signature of the block chain client in the model acquisition request passes verification, the computing platform sends the initial learning model of the target federal learning plan to the block chain client, wherein the model training result of the block chain client is obtained by training the initial learning model of the target federal learning plan based on private data.

3. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,

after the computing platform device integrates model parameters in at least two model training results which pass through private key signature verification to obtain a target learning model, the method further comprises the following steps:

and the computing platform generates an electronic deposit certificate corresponding to the target learning model and uploads the electronic deposit certificate to a block chain.

4. The method of claim 3, wherein the first and second light sources are selected from the group consisting of,

after the computing platform generates an electronic deposit corresponding to the target learning model and uploads the electronic deposit to a block chain, the method further includes:

the block chain client acquires an electronic deposit certificate of the target learning model from the block chain;

the block chain client checks the target learning model based on the acquired electronic deposit certificate; and the number of the first and second groups,

and after the target learning model passes the verification, the block chain client completes corresponding prediction based on the target learning model.

5. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,

the computing platform compiles execution logic predicted by using the target learning model into a predicted intelligent contract according to a mechanical language and deploys the intelligent contract into the block chain;

the block chain client sends a prediction request of the target learning model to the block chain, wherein the prediction request carries characteristic information of an object to be predicted;

and the block chain responds to the prediction request, calls the prediction intelligent contract, predicts the characteristic data of the object to be predicted and sends a prediction result to the block chain client.

6. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,

the blockchain, in response to the participation request, invoking a recruitment intelligence contract deployed in advance for the target federated learning plan further performs: and before adding the block chain client to the target federal learning plan, auditing the qualification of the block chain client participating in the target federal learning plan based on the block chain data corresponding to the block chain client, wherein if the qualification of the block chain client participating in the target federal learning plan passes the audit, the block chain client is added to the target federal learning plan and is put on record.

7. The method of claim 6, wherein the first and second light sources are selected from the group consisting of,

auditing the qualification of the blockchain client participating in the target federal learning plan includes: and auditing at least one of credit information, risk information and client level information of the blockchain client.

8. A model application method, comprising:

9. A model application method, comprising:

10. A model application method, comprising:

11. A block chain client comprises

12. An electronic device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:

13. A computer-readable storage medium having a computer program stored thereon, which when executed by a processor, performs the steps of:

14. A computing platform comprising

15. An electronic device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:

16. A computer-readable storage medium having a computer program stored thereon, which when executed by a processor, performs the steps of:

17. A block chain comprises

18. An electronic device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:

19. A computer-readable storage medium having a computer program stored thereon, which when executed by a processor, performs the steps of: