CN116910524A - Federal learning method, apparatus, device and storage medium - Google Patents

Abstract

The application belongs to the technical field of computers, and discloses a federal learning method, a federal learning device, federal learning equipment and a federal learning storage medium. The method comprises the steps of initializing a federation model and constructing a model training task corresponding to the federation model; distributing the model training task to a target client so that the target client trains the federal model carried in the model training task; receiving local model parameters fed back by a target client, and carrying out safety verification on the local model parameters based on a preset block chain; and when the safety verification is passed, updating the federal model according to the local model parameters. Because the blockchain is introduced in the federation learning process, the local model parameters fed back by each target client can be safely checked through the data stored in the blockchain, and the federation model can be updated after the safety check passes, so that the reliability and the safety of the federation learning method are improved.

Description

Federated learning methods, devices, equipment and storage media

Technical field

The present invention relates to the field of computer technology, and in particular to a federated learning method, device, equipment and storage medium.

Background technique

With the development of smart devices and the Internet, the application of technologies such as artificial intelligence in data processing has been promoted. However, most companies are unwilling to disclose their own data, and data privacy protection in various places has led to the phenomenon of data islands. Federated learning stipulates that model training be completed in a distributed collaborative training method. Each client trains the model locally and transmits the training parameters to the central server without leaking its own data, which solves the problem of data privacy to a large extent.

However, the current credibility of the federated learning system is not guaranteed, resulting in the illegal behavior of malicious terminals being unable to be traced, and the fairness and reliability of the training process unable to be guaranteed.

The above content is only used to assist in understanding the technical solution of the present invention, and does not represent an admission that the above content is prior art.

Contents of the invention

The main purpose of the present invention is to provide a federated learning method, device, equipment and storage medium, aiming to solve the technical problem that the security and reliability of federated learning in the existing technology cannot be guaranteed.

In order to achieve the above objectives, the present invention provides a federated learning method, which includes the following steps:

Initialize the federated model and construct a model training task corresponding to the federated model;

Distribute the model training task to the target client so that the target client trains the federated model carried in the model training task;

Receive the local model parameters fed back by the target client, and perform security verification on the local model parameters based on the preset blockchain;

When the security check passes, the federated model is updated according to the local model parameters.

Optionally, the step of distributing the model training task to the target client so that the target client trains the model carried in the model training task includes:

Distribute the model training task to the target client, so that the target client extracts the federated model from the model training task, trains the federated model according to the local data set, and after the training is completed, feeds back the local model parameters.

Optionally, the model training task is distributed to the target client, so that the target client extracts the federated model from the model training task, trains the federated model according to the local data set, and After training, the steps for feeding back local model parameters include:

Distribute the model training task to the target client, so that the target client trains the federated model in the model training task according to the local data set, and after the training is completed, obtains the trained federation The model parameters of the model are stored in the preset blockchain and fed back as local model parameters.

Optionally, the step of receiving local model parameters fed back by the target client and performing security verification on the local model parameters based on a preset blockchain includes:

Receive local model parameters fed back by the target client, and extract model identification information from the local model parameters;

Search parameter verification data in the preset blockchain according to the model identification information;

If the local model parameters are consistent with the parameter verification data, it is determined that the security verification has passed.

Optionally, if the local model parameters are consistent with the parameter verification data, the step of determining that the security verification has passed includes:

Perform a hash operation on the local model parameters to obtain parameter hash values;

If the parameter hash value is consistent with the parameter verification data, it is determined that the security verification has passed.

Optionally, after the step of updating the federated model according to the local model parameters when the security verification is passed, the step includes:

Use the updated federated model as the model to be tested;

Detect whether the model to be detected satisfies preset global convergence conditions;

If not, construct a new model training task based on the model to be detected, and return to the step of distributing the model training task to the target client.

Optionally, the step of detecting whether the updated federation model meets the preset global convergence conditions includes:

If satisfied, the model to be detected is sent to the target client, so that the target client performs multiple rounds of training on the model to be detected based on the local data set according to the preset rounds, and after training When the model to be detected meets the local convergence condition, training of the model to be detected is stopped.

In addition, to achieve the above objectives, the present invention also proposes a federated learning device, which includes the following modules:

The task construction module is used to initialize the federated model and construct the model training task corresponding to the federated model;

A task distribution module, configured to distribute the model training task to the target client so that the target client can train the federated model carried in the model training task;

A parameter verification module, configured to receive local model parameters fed back by the target client, and perform security verification on the local model parameters based on the preset blockchain;

The model update module is used to update the federated model according to the local model parameters when the security verification is passed.

In addition, to achieve the above object, the present invention also proposes a federated learning device. The federated learning device includes: a processor, a memory, and a federated learning program stored on the memory and capable of running on the processor. When the federated learning program is executed by the processor, the steps of implementing the federated learning method as described above are implemented.

In addition, in order to achieve the above object, the present invention also proposes a computer-readable storage medium. The computer-readable storage medium stores a federated learning program. When the federated learning program is executed, the steps of the federated learning method as described above are implemented. .

The present invention initializes the federated model and constructs a model training task corresponding to the federated model; distributes the model training task to the target client so that the target client can train the federated model carried in the model training task; and receives feedback from the target client. Local model parameters, and perform security verification on the local model parameters based on the preset blockchain; when the security verification passes, the federated model is updated based on the local model parameters. Since the blockchain is introduced in the federated learning process, the local model parameters fed back by each target client can be safely verified through the data stored in the blockchain, and the federated model will be updated only after the security verification is passed. , which improves the credibility and security of the federated learning method of the present invention.

Description of the drawings

Figure 1 is a schematic structural diagram of an electronic device with a hardware operating environment involved in an embodiment of the present invention;

Figure 2 is a schematic flow chart of the first embodiment of the federated learning method of the present invention;

Figure 3 is a schematic flow chart of the second embodiment of the federated learning method of the present invention;

Figure 4 is a schematic diagram of a hierarchical clustering process according to an embodiment of the present invention;

Figure 5 is a schematic diagram of an interaction process according to an embodiment of the present invention;

Figure 6 is a structural block diagram of the first embodiment of the federated learning device of the present invention.

The realization of the purpose, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention.

Referring to Figure 1, Figure 1 is a schematic structural diagram of the federated learning device of the hardware operating environment involved in the embodiment of the present invention.

As shown in Figure 1, the electronic device may include: a processor 1001, such as a central processing unit (Central Processing Unit, CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize connection communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). The optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001.

Those skilled in the art can understand that the structure shown in Figure 1 does not constitute a limitation of the electronic device, and may include more or fewer components than shown, or combine certain components, or arrange different components.

As shown in Figure 1, memory 1005, which is a storage medium, may include an operating system, a network communication module, a user interface module, and a federated learning program.

In the electronic device shown in Figure 1, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be provided in In the federated learning device, the electronic device calls the federated learning program stored in the memory 1005 through the processor 1001, and executes the federated learning method provided by the embodiment of the present invention.

An embodiment of the present invention provides a federated learning method. Refer to Figure 2. Figure 2 is a schematic flow chart of a federated learning method according to the first embodiment of the present invention.

In this embodiment, the federated learning method includes the following steps:

Step S10: Initialize the federated model and construct a model training task corresponding to the federated model.

It should be noted that the execution subject of this embodiment may be the federated learning device. The federated learning device may be an electronic device such as a server, a cloud server, or a personal computer. Of course, it may also be other devices that can implement the same or similar functions. Equipment, this embodiment is not limited to this. In this embodiment and the following embodiments, the federated learning method of the present invention is explained by taking the federated learning equipment as an example.

It should be noted that the federated model can be a global model that needs to be trained through federated learning, and the federated model can be preset by the manager of the federated learning device. Initializing the federated model can be to initialize the model parameters and/or hyperparameters of the federated model. The model training task corresponding to building the federated model can be to use the model structure, model parameters and number of training iterations (epoch) of the federated model as task parameters. Build a model training task. The number of training iterations can be preset by the administrator of the federated learning device, for example, the number of training iterations is set to 5.

Step S20: Distribute the model training task to the target client, so that the target client trains the federated model carried in the model training task.

It should be noted that the target client can be a client selected for distributed training of the federated model. The target client can be specified in advance by the administrator of the federated learning device, or it can be a client of the federated learning device. Select a preset number of clients as target clients.

In the specific implementation, the federated learning device distributes the model training task to the target client, and the target client can parse the model training task, extract the federated model from the model training task, and then compare the federated model based on the local data of the target client. When training is performed, step S20 in this embodiment may include:

Distribute the model training task to the target client, so that the target client extracts the federated model from the model training task, trains the federated model according to the local data set, and after the training is completed, feeds back the local model parameters.

It should be noted that the local data set can be a data set stored locally on the target client. The local data sets stored by different clients are different, and their data structures can also be different, that is, the data stored in each client is heterogeneous data.

In actual use, the target client can also train the number of iterations from the model training task, and then perform multiple rounds of training on the federated model based on the number of training iterations and the local data set. Among them, in order to ensure that the problem can be traced, the target client can also store the local data set used for training or the hash value of the local data set in the preset blockchain.

Furthermore, in order to facilitate the federated learning device to verify the local model parameters fed back by the target client, in this embodiment, the model training task is distributed to the target client, so that the target client extracts all the parameters from the model training task. The steps of training the federated model based on the local data set and feeding back the local model parameters after the training is completed may include:

Distribute the model training task to the target client, so that the target client extracts the federated model from the model training task, trains the federated model in the model training task according to the local data set, and After the training is completed, the model parameters of the trained federated model are obtained, the model parameters are stored in the preset blockchain, and the model parameters are fed back as local model parameters.

It should be noted that since federated learning may perform multiple rounds of model training, the target client will also feedback local model parameters to the federated learning device multiple times, and the local model parameters fed back each time may be different. For convenience, The federated learning device verifies whether the local model parameters fed back by the target client are forged by malicious attackers. When storing the model parameters in the preset blockchain, the model identification information of the federated model in the current target client can be extracted, and then based on The model identification information stores the model parameters into the preset blockchain.

Among them, the model identification information can be the unique identification assigned by the target client to the federated model. Different models are set with different model identification information for the same federated model in different clients, and different training stages of the federated model in the same client can also correspond to different Model identification information, for example: set the model version number for the federated model as the model identification information. If necessary, you can also combine the device identification (device ID) of the target client with the model version number to generate model identification information.

Step S30: Receive the local model parameters fed back by the target client, and perform security verification on the local model parameters based on the preset blockchain.

It should be noted that in order to ensure data security, both the client and the federated learning device must install at least one blockchain node with a preset blockchain. During the training process, model parameters, training data and other data can be uploaded to the chain area. Blockchain, then the federated learning device can verify whether the received data is forged data through the data stored in the preset blockchain.

In specific implementation, after the target client completes training of the federated model based on the model training task, it will feed back the parameters in the trained federated model to the federated learning device. At this time, in order to ensure the security in the federated learning process, the federated learning device will perform a security check on the local model parameters through the data stored in the preset blockchain to determine whether the local model parameters have been forged by a malicious attacker. At this time, the steps of security verification of the local model parameters based on the preset blockchain described in this embodiment may include:

Receive local model parameters fed back by the target client, and extract model identification information from the local model parameters;

Search parameter verification data in the preset blockchain according to the model identification information;

If the local model parameters are consistent with the parameter verification data, it is determined that the security verification has passed.

It should be noted that when the target client feeds back the local model parameters, it can add the model identification information to the local model parameters and feed them back to the federated learning device. After receiving the local model parameters, the federated learning device quickly presets If the parameter verification data for secure verification of local model parameters is found in the blockchain, the model identification information can be extracted from the local model parameters, and then the corresponding parameter verification data can be found in the preset blockchain based on the model identification information. .

It can be understood that since the target client can store the local model parameters as parameter verification data in the preset blockchain, after obtaining the parameter verification data, the federated learning device can store the received local model parameters. Compare it with the parameter verification data. If the two are consistent, the federated learning device can determine that the local model parameters are not forged by a malicious attacker. Therefore, it can be determined that the security verification has passed.

In specific implementation, since the model parameters may have many parameter items, if the local model parameters are directly stored on the preset blockchain, it may take up a large space and even affect the performance of the preset blockchain. In order to To avoid this situation, it is necessary to reduce the amount of data stored in the preset blockchain. At this time, the target client can hash the local model parameters, obtain the hash value corresponding to the local model parameter, and then use the hash value as When the parameter verification data is stored in the preset blockchain, the verification method of the federated learning device will also change accordingly. Therefore, as described in this embodiment, if the local model parameters are consistent with the parameter verification data, Then the steps to determine whether the security verification is passed may include:

Perform a hash operation on the local model parameters to obtain parameter hash values;

If the parameter hash value is consistent with the parameter verification data, it is determined that the security verification has passed.

It is understandable that in order to reduce the amount of data stored in the preset blockchain, the target client will store the hash value of the local model parameters as parameter verification data in the preset blockchain. At this time, the federated learning device The local model parameters can be hashed in the same way, and then the parameter hash value obtained by the hash operation is compared with the parameter verification data. If the parameter hash value is consistent with the parameter verification data, the local model can be determined The parameters were not forged by malicious attackers, so it can be determined that the security check passed.

In the specific implementation, if only the hash value is used for verification, the verification method may still be cracked by malicious attackers. In order to improve the difficulty of cracking, the target client can first rearrange the local model parameters in a preset order, and then rearrange the parameters. The ranked local model parameters are hashed, and the hash value obtained at this time is stored in the preset blockchain as parameter verification data. Among them, the preset sequence can be preset by the administrator of the federated learning device, and added as task parameters to the model training task and distributed to each target client.

Of course, in order to ensure security, the default blockchain can store encrypted data when storing data. For example, when the target client stores data A to the default blockchain, it can use the block connection point in the target client. The stored encryption key encrypts data A, and then stores it on the preset blockchain after encryption. Then the federated learning device can decrypt the obtained blockchain data based on the decryption key corresponding to the encryption key, thereby obtaining the parameters. Check data. The encryption key may be an encryption key of a symmetric encryption algorithm or an encryption key of a fee-symmetric encryption algorithm, which is not limited in this embodiment.

In the same way, when the federated learning device distributes model training tasks to each target client, it can also store the verification data related to the model training task in the preset blockchain. Then the target client can use the preset blockchain The stored data performs security verification on the model training task. When the security verification passes, the model training is performed based on the model training task.

Step S40: When the security check passes, update the federated model according to the local model parameters.

It can be understood that if the security check is passed, it means that the local model parameters are not forged by malicious attackers. Therefore, the federated learning device can update the federated model based on the local model parameters, for example: after receiving the information from each target customer After receiving the local model parameters fed back by the client, the average value of the local model parameters can be calculated, and then the model parameters in the federated model are updated based on the average value.

This embodiment initializes the federated model and constructs a model training task corresponding to the federated model; distributes the model training task to the target client so that the target client can train the federated model carried in the model training task; and receives feedback from the target client. local model parameters, and perform security verification on the local model parameters based on the preset blockchain; when the security verification passes, the federated model is updated based on the local model parameters. Since the blockchain is introduced in the federated learning process, the local model parameters fed back by each target client can be safely verified through the data stored in the blockchain, and the federated model will be updated only after the security verification is passed. , which improves the credibility and security of the federated learning method of the present invention.

Refer to Figure 3, which is a schematic flow chart of a second embodiment of a federated learning method of the present invention.

Based on the above first embodiment, after the step S40 of the federated learning method in this embodiment, it also includes:

Step S50: Use the updated federated model as the model to be detected.

It should be noted that after updating the federated model based on the local model parameters fed back by each target client, the updated federated model needs to be further verified to determine whether federated training needs to continue. Therefore, the updated federated model can be The federated model serves as the model to be tested.

Step S60: Detect whether the model to be detected satisfies the preset global convergence condition.

It should be noted that the preset global convergence conditions can be set in advance by the administrator of the federated learning device.

For example: managers of federated learning equipment can set a model verification set in advance, and use the model verification set to verify the accuracy of the model to be detected. When the model accuracy of the model to be detected reaches the preset accuracy threshold, it is determined that the model to be detected satisfies the preset global convergence conditions.

Of course, when verifying whether the model to be tested satisfies the preset global convergence conditions, not only the accuracy can be combined, but also the generalization, learning rate, etc. can be combined for comprehensive determination. This embodiment is not limited to this.

It can be understood that if the model to be tested does not meet the preset global convergence conditions, it means that during the federated learning process, the trained federated model has not reached the initially set training goal, and a new model will be constructed based on the model to be tested at this time. Create a new model training task, and then return to the step of distributing the model training task to the target client, and train the model to be tested again.

Step S70: If satisfied, send the model to be detected to the target client, so that the target client performs multiple rounds of training on the model to be detected based on the local data set according to preset rounds, and When the trained model to be detected meets the local convergence condition, the training of the model to be detected is stopped.

It should be noted that if the model to be detected meets the preset global convergence conditions, it means that during the federated learning process, the trained federated model has reached the initially set training goal, and the federated learning of the federated model can be ended at this time.

However, since the data stored in each target client is heterogeneous data, the federated model may not be effective in the application scenario of the target client. In order to avoid this situation, the model to be detected satisfies the preset global convergence conditions. At the same time, the model to be detected can also be sent to the target client, so that the target client can perform multiple rounds of training on the model to be detected based on the local data set according to the preset rounds, and continuously detect whether the model to be detected satisfies the target customer during the training process. local convergence conditions stored in the terminal, and then stop training the model to be detected when the trained model to be detected meets the local convergence conditions.

Among them, the preset round-level local convergence conditions can be set in advance by the administrator of the target client according to actual needs. The setting method of the local convergence conditions is similar to the setting method of the global convergence conditions, and will not be described again here.

It can be understood that by sending the model to be detected that meets the preset global convergence conditions to each target client for re-training, a personalized model that meets the target client can be built. Personalization will allow each client to learn from more similarities. Get more benefits from your clients.

In order to facilitate understanding, the description will now be made in conjunction with Figure 4 and Figure 5. This solution will not be limited. Figure 4 is a schematic diagram of the hierarchical clustering process of this embodiment. Figure 5 shows the client and server of this embodiment (that is, as mentioned above). Schematic diagram of the interaction process between federated learning devices).

First, the federated learning problem can be defined, assuming that D represents the global data distribution, s ₁ ,...s _N =S, S follows the D distribution, the data is distributed among M client terminals, and each client can only access and utilize Local data s ₁ ,...s _Ni =S _i , local data belong to a subset of global data. Di represents the local data distribution, θ _g represents the parameters of the global parameters, and the global training task can be expressed as:

Di is different for each client, but clustering operations can be performed based on their similarities, resulting in a specific model θ _c (c∈C, C is the cluster set, θ _c represents all clusters) for each clustering result A collection of result-like models). The loss function of the global model defined by the clustering results is expressed as follows:

The loss function of each clustering result is expressed as follows

Extend this model to a personalized version where each client has its own model:

Both equations (2) and (4) introduce additional degrees of freedom, allowing clients to better learn sampled training data points from their respective distributions. The results are compared as follows:

Second, the federated learning training process can be defined

Assume that there are M clients, N _i =|D _i | represents the number of samples owned by client i, and the global training loss function is expressed as follows:

Among them/>

where F _i (θ) is the local objective function of each client, and f (θ) is the average global objective of all clients. The training process includes two stages. First, the client uses the initial distribution to train a predetermined round on its local data. number, client and server communication. The server then calculates a weighted average based on all locally updated model parameters, and the results are returned to the client to update its model parameters.

Third, set up a personalized federated learning model

Global model parameters are initialized. Before joint training starts, the parameters are transmitted to randomly selected n clients, local training is performed, and updated local model parameters are generated in each client, using adaptive elements based on the current cycle number. The learning rate updates parameters, and the adaptive meta-learning rate decreases linearly throughout the training process. After the last round of joint training, the local models of all clients are personalized by performing a fixed number of gradient optimization epochs on the local training data. Personalization enables each client to learn from more similar clients. Get more benefits.

Fourth, hierarchical clustering operation of federated learning process

Hierarchical clustering is performed after an initial stage of global federated learning. After performing federated learning for a specified number of federation rounds, clients are clustered by hierarchical clustering based on model updates in the current round. Then, each generated client cluster is treated as an isolated federated learning problem, and the process is shown in Figure 4.

The above steps only introduce the principles related to federated learning. The following steps introduce the blockchain and communication related processes in detail.

Fifth, client and central server settings based on blockchain scenarios

This step involves the job creation process of federated learning and the interaction process with the blockchain. The behavior of the server-side processor and client is shown in Figure 2. The specific process is as follows.

Server behavior: The server's job creator creates a training job by initializing the global model and hyperparameters, and then broadcasts the job to the client model trainer. The job creator transfers the initial global model to the global model database. The server waits for local model parameters from qualified clients. The model aggregator receives local model parameters from the client, performs model aggregation, and calculates the average of all model parameters to update global model parameters;

The updated global model is stored in the global model database. A hash of the global model version is uploaded to the blockchain. The model deployer broadcasts the updated global model to the client decider for use in the next round of training. The model deployer is the component that selects client devices to receive global model updates.

Client behavior: The client preprocesses the local data and stores it in the local database. The output in the database is transmitted to the model trainer for local training. The hash value of the training data version used in each epoch is uploaded to the blockchain. After training, the local model is transferred to the local model evaluator for evaluation, and the hash value of the current model version is also transferred to the blockchain;

Local model parameters are stored in the local model database and sent to the server's model aggregator. After that, multiple rounds of communication are carried out. If the decision-maker checks that the results of the model meet the requirements, the training process is terminated.

Sixth, data model registration smart contract construction

In each local training epoch, the client uploads its local model parameters and corresponding hashes of the data version to the data model registration smart contract. After each global aggregation, the server sends the hash of the global model parameters to the smart contract.

In smart contracts, the hash values of local and global model parameters are recorded in the model structure. Once uploaded, the information cannot be modified. Set up another structure to calculate the number of uploads for each client. Through on-chain hash mapping, these two structures are connected to the client through their on-chain addresses and are used to retrieve on-chain data versions and model parameter information. You may encounter problems during the upload process. The following lists possible problems and gives solutions.

Seventh, define the system blockchain functions and processing procedures

In the design of a blockchain-based trusted federated learning architecture, each client and server installs at least one blockchain node to form a blockchain network. Each node keeps a local copy of the complete transaction data in the form of a blockchain. Modifying any historical data state in any block requires updating all subsequent blocks stored in all participating nodes. Furthermore, blockchain primarily handles data models using smart contracts, where all participants are identified by their blockchain addresses. In order to maintain the feasibility and efficiency of the joint learning process while tracing the model, the architecture applies an off-chain data storage design pattern, using a database system to store the actual model, while the blockchain only records the hashed version of the model. The details of the off-chain part and the up-chain part in the blockchain system are as follows.

Off-chain part: All model parameters (i.e. local and global models) are stored in the off-chain database of the local model and the database of the global model respectively.

On-chain part: The data version consists of timestamp and data size. A hash value is generated based on the local data version and recorded in the registered smart contract of the on-chain data model to achieve source and version control of data and models.

For data privacy reasons, the present invention tracks the data used to train each local model using smart contracts without recording the actual local data on the blockchain. Verify its origin by comparing the off-chain model’s hash with the corresponding on-chain record.

In this embodiment, the updated federated model is used as the model to be detected; it is detected whether the model to be detected satisfies the preset global convergence condition; if it is satisfied, the model to be detected is sent to the target client, so that all The target client performs multiple rounds of training on the model to be detected based on the local data set according to preset rounds, and stops training the model to be detected when the trained model to be detected meets local convergence conditions. Since the model to be detected that meets the preset global convergence conditions will be distributed to each target client, each target client will perform multiple rounds of training on the model to be detected based on the local data set and preset rounds until the local convergence conditions are met, ensuring A personalized model that meets the target client can be constructed to ensure that the personalized model obtained by each target client can meet the actual needs of the local usage scenario of each target client.

In addition, embodiments of the present invention also provide a storage medium on which a federated learning program is stored. When the federated learning program is executed by a processor, the steps of the federated learning method described above are implemented.

Referring to Figure 6, Figure 6 is a structural block diagram of the first embodiment of the federated learning device of the present invention.

As shown in Figure 6, the federated learning device proposed by this embodiment of the present invention includes:

The task construction module 10 is used to initialize the federated model and construct the model training task corresponding to the federated model;

Task distribution module 20 is used to distribute the model training task to the target client, so that the target client trains the federated model carried in the model training task;

The parameter verification module 30 is used to receive the local model parameters fed back by the target client, and perform security verification on the local model parameters based on the preset blockchain;

The model update module 40 is configured to update the federated model according to the local model parameters when the security verification is passed.

This embodiment initializes the federated model and constructs a model training task corresponding to the federated model; distributes the model training task to the target client so that the target client can train the federated model carried in the model training task; and receives feedback from the target client. local model parameters, and perform security verification on the local model parameters based on the preset blockchain; when the security verification passes, the federated model is updated based on the local model parameters. Since the blockchain is introduced in the federated learning process, the local model parameters fed back by each target client can be safely verified through the data stored in the blockchain, and the federated model will be updated only after the security verification is passed. , which improves the credibility and security of the federated learning method of the present invention.

Further, the task distribution module 20 is also used to distribute the model training task to the target client, so that the target client extracts the federated model from the model training task, and analyzes the federated model according to the local data set. The federated model is trained, and after training, local model parameters are fed back.

Further, the task distribution module 20 is also used to distribute the model training task to the target client, so that the target client extracts the federated model from the model training task, and analyzes the federated model according to the local data set. The federated model in the model training task is trained, and after the training is completed, the model parameters of the trained federated model are obtained, the model parameters are stored in the preset blockchain, and the model parameters are used as local Model parameters are fed back.

Further, the parameter verification module 30 is also used to receive the local model parameters fed back by the target client, extract model identification information from the local model parameters; and perform the preset blockchain according to the model identification information. Search parameter verification data in; if the local model parameters are consistent with the parameter verification data, it is determined that the security verification has passed.

Further, the parameter verification module 30 is also used to perform a hash operation on the local model parameters to obtain a parameter hash value; if the parameter hash value is consistent with the parameter verification data, it is determined to be safe. Verification passed.

Further, the model update module 40 is also used to use the updated federated model as a model to be detected; detect whether the model to be detected satisfies the preset global convergence conditions; if not, construct a model based on the model to be detected. Create a new model training task, and return to the step of distributing the model training task to the target client.

Further, the model update module 40 is also configured to, if satisfied, send the model to be detected to the target client, so that the target client can update the model according to the local data set according to the preset round. The model to be detected is trained for multiple rounds, and when the trained model to be detected meets the local convergence condition, training of the model to be detected is stopped.

It should be understood that the above are only examples and do not constitute any limitation on the technical solution of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any limitations on this.

It should be noted that the workflow described above is only illustrative and does not limit the scope of the present invention. In practical applications, those skilled in the art can select some or all of them for implementation according to actual needs. The purpose of this embodiment is not limited here.

In addition, for technical details that are not described in detail in this embodiment, please refer to the federated learning method provided by any embodiment of the present invention, and will not be described again here.

Furthermore, it should be noted that, as used herein, the terms "include", "comprises" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or system that includes a list of elements includes not only those elements, but also other elements not expressly listed or elements inherent to the process, method, article or system. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

The above serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product in essence or that contributes to the existing technology. The computer software product is stored in a storage medium (such as a read-only memory). , ROM)/RAM, magnetic disk, optical disk), including several instructions to cause a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the method described in various embodiments of the present invention.

The above are only preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the description and drawings of the present invention may be directly or indirectly used in other related technical fields. , are all similarly included in the scope of patent protection of the present invention.

Claims (10)

1. A federal learning method, comprising the steps of:

Initializing a federation model and constructing a model training task corresponding to the federation model;

distributing the model training task to a target client so that the target client trains the federation model carried in the model training task;

receiving local model parameters fed back by the target client, and carrying out safety verification on the local model parameters based on a preset blockchain;

and updating the federation model according to the local model parameters when the safety verification passes.

2. The federal learning method according to claim 1, wherein the step of distributing the model training task to a target client to cause the target client to train the model carried in the model training task comprises:

and distributing the model training task to a target client so that the target client extracts the federation model from the model training task, trains the federation model according to a local data set, and feeds back local model parameters after training is finished.

3. The federal learning method according to claim 2, wherein the step of distributing the model training task to a target client to cause the target client to extract the federal model from the model training task, train the federal model according to a local data set, and feedback local model parameters after training is completed, comprises:

Distributing the model training task to a target client so that the target client extracts the federation model from the model training task, training the federation model in the model training task according to a local data set, acquiring model parameters of the federation model after training, storing the model parameters into a preset blockchain, and feeding back the model parameters as local model parameters.

4. The federal learning method according to claim 3, wherein the step of receiving the local model parameters fed back by the target client and performing security check on the local model parameters based on a preset blockchain comprises:

receiving local model parameters fed back by the target client, and extracting model identification information from the local model parameters;

searching parameter verification data in a preset blockchain according to the model identification information;

and if the local model parameters are consistent with the parameter verification data, judging that the safety verification is passed.

5. The federal learning method according to claim 4, wherein the step of determining that the security check passes if the local model parameter is consistent with the parameter check data comprises:

Carrying out hash operation on the local model parameters to obtain parameter hash values;

and if the parameter hash value is consistent with the parameter verification data, judging that the safety verification is passed.

6. The federal learning method according to claim 1, wherein the step of updating the federal model based on the local model parameters when the security check passes comprises:

taking the updated federal model as a model to be detected;

detecting whether the model to be detected meets a preset global convergence condition or not;

if not, a new model training task is constructed according to the model to be detected, and the step of distributing the model training task to the target client is returned.

7. The federal learning method according to claim 6, wherein the step of detecting whether the updated federal model satisfies a preset global convergence condition comprises:

if yes, the model to be detected is sent to the target client, so that the target client carries out multi-round training on the model to be detected according to a local data set according to preset rounds, and training on the model to be detected is stopped when the trained model to be detected meets a local convergence condition.

8. A federal learning apparatus, comprising:

the task construction module is used for initializing a federation model and constructing a model training task corresponding to the federation model;

the task distribution module is used for distributing the model training task to a target client so that the target client trains the federal model carried in the model training task;

the parameter verification module is used for receiving local model parameters fed back by the target client and carrying out safety verification on the local model parameters based on a preset block chain;

and the model updating module is used for updating the federal model according to the local model parameters when the safety verification passes.

9. A federal learning apparatus, the federal learning apparatus comprising: a processor, a memory and a federal learning program stored on the memory and executable on the processor, which federal learning program when executed by the processor implements the steps of the federal learning method according to any one of claims 1-7.

10. A computer readable storage medium, wherein a federal learning program is stored on the computer readable storage medium, which when executed implements the steps of the federal learning method according to any one of claims 1-7.