CN114861211B - A data privacy protection method, system, and storage medium for metaverse scenarios - Google Patents

Abstract

The invention discloses a data privacy protection method, a system and a storage medium for a meta-universe scene; the method comprises the following steps: uploading non-private data to a data storage module in the virtual world as a first local private data set to train a first local model; while private data with sensitive information is stored locally as a second local private data set training a second local model; constructing a metauniverse cross-chain federal machine learning framework, wherein the metauniverse cross-chain federal machine learning framework comprises a task publisher, a first client for training a first local model through a first local private data set, and a second client for training a second local model through a second local private data set; the task publisher and the first client are arranged in the virtual world, and the second client is arranged in the real world; the first client and the second client update and aggregate the first local model parameters and the second local model parameters based on a cross-chain aggregation method to obtain a new global model.

Description

A data privacy protection method, system, and storage medium for metaverse scenarios

Technical Field

The present invention relates to the field of communication signal demodulation technology, and more specifically, to a data privacy protection method, system, and storage medium for a metaverse scenario.

Background technique

Federated learning is an emerging artificial intelligence technology. Its design goal is to ensure information security when sharing big data, protect terminal data and personal data privacy; and carry out efficient machine learning model training between multiple participants or multiple computing nodes under the premise of ensuring legality and compliance. Federated learning achieves "data is available but invisible" and "data does not leave the house" through iterative training operations. In the process of model training, interactive data is encrypted, which to a certain extent achieves efficient model training and protects the privacy of participants' data to the greatest extent, thereby solving the data privacy problem caused by the need for traditional machine learning models to access data. At the same time, it can introduce data from more organizations or institutions to improve the overall quality of the model.

Blockchain, with its anonymity, immutability, and distribution, provides a secure and reliable solution between multiple untrusted parties. The essence of blockchain is a distributed ledger. Its biggest feature is that it has changed from a traditional centralized solution to a distributed network structure, ensuring the security of on-chain data through cryptographic technologies such as asymmetric encryption; at the same time, through consensus mechanisms, smart contracts, etc., the reliability of chain data is guaranteed between multiple untrusted distributed participants. Through the authorization mechanism and identity management of blockchain, mutually untrusted users can be integrated as participants to establish a secure and reliable cooperation mechanism; and the model parameters of federated learning can be stored in the blockchain, ensuring the security and reliability of the model parameters.

Currently, there are many and diverse blockchain networks in operation, but each blockchain network is independent of each other, which greatly limits the interoperability between blockchains. In order to connect different blockchain networks to realize the value Internet, blockchain cross-chain technology has become the key. The overall architecture of cross-chain technology can be divided into information collection, information verification, system connection and information feedback. By establishing a connection party or expanding a smart contract, information interaction between different parallel chains is realized. After verification by two or more parties, it is recorded in the total blockchain ledger. The relay chain is one of the current mainstream cross-chain technologies. The relay chain does not completely rely on the verification judgment of a trusted third party. It only collects the data status of the two blockchains through an intermediary for self-verification. Its verification method varies significantly depending on its own structure. Although the relay chain is relatively difficult to implement, it is the most complete in terms of functions and characteristics, and has unique and important application value.

The Metaverse is the next generation of the Internet after the web and mobile Internet. People can freely socialize and entertain themselves in a virtual world. Through technologies such as VR and AR, people can immersively access the Metaverse. In order to enhance the immersion of the Metaverse, the Metaverse needs to synchronize data continuously and obtain fresh data from the real world. The amount and variety of personal data collected in this process will be unprecedented. For example, in VR games, in order to maintain the authenticity of the rendered scene, it is necessary to read the user's actions in real time through sensors. At the same time, it is also possible to read the user's physiological reactions and EEG patterns and other private data without the user's consent. If the entire virtual world is uploaded, there will be a risk of privacy data leakage and abuse. Therefore, it is particularly important to protect privacy data in the virtual world.

The shortcomings of the current existing technology are as follows: the AI model in the metaverse needs to be trained with the help of data generated in the real world and the virtual world to obtain better model performance. However, if all local private data is uploaded to the virtual world for training, on the one hand, uploading the real world data to the virtual world storage module will incur large communication and storage overheads. On the other hand, it will also lead to the risk of privacy leakage and data abuse.

Summary of the invention

In order to solve the problems of deficiencies and shortcomings in the prior art, the present invention provides a data privacy protection method, system, and storage medium for metaverse scenarios. The method realizes local privacy data protection while better training AI models in the virtual world.

In order to achieve the above-mentioned purpose of the present invention, the technical scheme adopted is as follows:

A data privacy protection method for a metaverse scenario, the method comprising the following steps:

The data is divided into two parts according to the data type and privacy protection requirements. The non-privacy data is directly uploaded to the data storage module in the virtual world and used as the first local private data set to train the first local model. The privacy data with sensitive information is stored locally and used as the second local private data set to train the second local model.

Constructing a Metaverse cross-chain federated machine learning framework with privacy protection, the Metaverse cross-chain federated machine learning framework comprising a task publisher for storing a global model, a first client for training a first local model through a first local private data set, and a second client for training a second local model through a second local private data set; the task publisher and the first client are set in a virtual world, and the second client is set in a real world;

The first client and the second client update the first local model parameters and the second local model parameters based on the cross-chain aggregation method and aggregate them to obtain a new global model.

Preferably, a decentralized metaverse cross-chain federated machine learning framework is built using blockchain technology, wherein one blockchain is used as the main chain, and the main chain is used as the storage management module of the global model parameters; one blockchain is used as the first slave chain, and the first slave chain is used as the first storage module of the first local model parameters; one blockchain is used as the second slave chain, and the second slave chain is used as the second storage module of the second local model parameters; a relay chain is also provided, and the relay chain is used as the cross-chain management platform;

The task publisher sends the initialized global model through the main chain, and receives the updated first local model and second local model parameters through the main chain;

The first client receives the initialized global model through the first slave chain, and uploads the updated first local model parameters to the first slave chain;

The second client receives the initialized global model through the second slave link, and uploads the updated second local model parameters to the second slave link;

The relay chain realizes the cross-connection between the first slave chain, the second slave chain and the main chain.

Furthermore, before the Metaverse cross-chain federated machine learning framework is trained, it is initialized as follows:

The first slave chain and the second slave chain register with the relay chain, and the task publisher determines the federated machine learning task and searches for the first client and the second client that participate in the federated machine learning training;

The first client and the second client determine whether to participate in the federated machine learning task according to their needs; after the two parties reach an agreement, they enter the federated machine learning process;

The task publisher uploads the initialized global model _wt to the main chain, and the first client in the virtual world and the second client in the real world obtain the initialized global model _wt through the corresponding first slave chain and second slave chain respectively.

Going further, the federated machine learning process is as follows:

The second client and the first client receive the initialized global model w _t issued by the task publisher, and initialize the global model w _t respectively to obtain the second local model First local model/> The initialization process is as follows:

in, Represents the second client set in the real world,/> represents the first client set in the virtual world; then, the first client and the second client use the first local private data set D _j and the second local private data set D _i for training respectively, where D _i represents the second local private data set held by the second client i in the real world, and D _j represents the first local private data set held by the first client j in the virtual world; the loss functions to be optimized for the training of the second local model and the first local model are defined as follows:

real world:

virtual reality:

Where w is the global model parameter, |D _i | is the size of the data set of the second client i, |D _j | is the size of the data set of the first client j, f _k (w) is the local loss function, and each client performs stochastic gradient descent for a given number of iterations to train the second local model and the first local model as follows:

Where η is the learning rate, is the gradient.

Furthermore, after the first client in the virtual world and the second client in the real world train the first local model and the second local model, the trained first local model parameters and the second local model parameters are uploaded to the first slave chain and the second slave chain respectively, and cross-chain requests are initiated to the main chain in the first slave chain and the second slave chain respectively. The first local model parameters and the second local model parameters are updated and uploaded to the main chain through the blockchain cross-chain method. The main chain verifies the received first local model parameters and the second local model parameters, and performs aggregation operations to obtain a new global model w _t+1 , which is defined as follows:

The federated machine learning task ends after a certain number of iterations until the global model converges; the learned global model will be retained in the main chain for subsequent prediction or classification work.

Furthermore, there are four participants in the cross-chain aggregation method, and their functions are as follows:

Validator: responsible for the block generation of the relay chain network, running a relay chain client, and verifying the blocks generated by the blockchain nominated by it;

Collator: Maintains a full node of the blockchain, helps validators collect and verify the correctness of transactions, and submits candidate blocks to validators;

Nominator: A party that owns tokens and is responsible for maintaining and securing the validator.

Supervisor: Earn income by reporting illegal transactions or illegal blocks.

Furthermore, the cross-chain process between the second slave chain and the main chain is as follows:

S101: The second client creates an account on the second slave chain and initializes the information stored on the second slave chain;

S102: After the second client completes training the second local model, the second client creates a transaction on the second slave chain and sends the second local model parameters and identity information to the main chain; the transaction is a second local model parameter upload request;

S103: The second client signs and broadcasts the transaction;

S104: The organizer of the second slave chain collects transaction information, verifies the validity of the transaction, organizes the transaction data, and packages it into a candidate block;

S105: The collation staff shows the candidate block and the proof of state transfer to the validator of the second slave chain;

S106: The validator verifies the received candidate block. Only if the verification is passed, ensuring that the candidate block contains only valid transactions, the validator will pledge its tokens to prepare for the candidate block to be put on the chain;

S107: When enough nominators stake their tokens and nominate validators, the broadcast of their candidate blocks to the blockchain will be authorized;

S108: When all validators reach a consensus on the relay chain block, the validator moves the transaction on the second slave chain from the exit of the second slave chain to the entrance of the main chain to complete the message transmission;

S109: The main chain executes the transaction in the entry queue and modifies its own ledger;

S110: The task issuing server obtains and temporarily stores the updated parameters of the second local model, and waits for several second clients to return results before performing an aggregation operation.

Furthermore, the cross-chain process between the first slave chain and the master chain is as follows:

S201: The first client creates an account on the first slave chain and initializes the information stored on the first slave chain;

S202: After the first client completes training the first local model, the first client creates a transaction on the first slave chain and sends the first local model parameters and identity information to the master chain; the transaction is a first local model parameter upload request;

S203: The first client signs and broadcasts the transaction;

S204: First, the chain organizer collects transaction information, verifies the validity of the transaction, organizes the transaction data, and packages it into a candidate block;

S205: The collation staff shows the candidate block and the proof of state transfer to the validator of the first slave chain;

S206: The validator verifies the received candidate block. Only if the verification is passed, ensuring that the candidate block contains only valid transactions, the validator will pledge its tokens to prepare for the candidate block to be put on the chain;

S207: When enough nominators stake their tokens and nominate validators, the broadcast of their candidate blocks to the blockchain will be authorized;

S208: When all validators reach a consensus on the relay chain block, the validator moves the transaction on the first slave chain from the exit of the first slave chain to the entrance of the main chain to complete the message transmission;

S209: The main chain executes the transaction in the entry queue and modifies its own ledger;

S210: The task issuing server obtains and temporarily stores the updated parameters of the first local model, and waits for a number of first clients to return results before performing an aggregation operation.

A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above method are implemented.

A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

The beneficial effects of the present invention are as follows:

The present invention does not simply upload all physical data collected by sensors directly to the virtual world, but divides the data into two parts according to data type and privacy protection requirements. Non-privacy data can be directly uploaded to the virtual world data storage module (such as cloud servers, edge servers, etc.), and privacy data with sensitive information is stored locally, thereby realizing privacy protection of artificial intelligence applications in the metaverse scenario.

The present invention uses blockchain technology to build a decentralized Metaverse cross-chain federated machine learning framework. This framework is different from the traditional federated machine learning framework. One blockchain in this framework is called the main chain, which serves as a storage management module for global model parameters, and multiple slave chains serve as storage modules for local model parameters. Local training model parameters are uploaded to the slave chains. The Metaverse cross-chain federated machine learning framework can record the training process of the local model, achieve privacy protection and the auditability and traceability of the local training process.

The present invention also proposes to adopt a cross-chain aggregation method, in which the interaction of local model parameters between the master and slave chains is carried out in a cross-chain form, ensuring traceability while also ensuring the security of local model parameter interaction to a certain extent.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic diagram of the data privacy protection method according to Embodiment 1.

FIG. 2 is a schematic diagram of the cross-chain polymerization method described in Example 1.

Detailed ways

The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

Example 1

As shown in FIG1 , a data privacy protection method for a metaverse scenario is provided, wherein the method comprises the following steps:

The data is divided into two parts according to the data type and privacy protection requirements. The non-privacy data is directly uploaded to the data storage module in the virtual world and used as the first local private data set to train the first local model. The privacy data with sensitive information is stored locally and used as the second local private data set to train the second local model.

Constructing a Metaverse cross-chain federated machine learning framework with privacy protection, the Metaverse cross-chain federated machine learning framework comprising a task publisher for storing a global model, a first client for training a first local model through a first local private data set, and a second client for training a second local model through a second local private data set; the task publisher and the first client are set in a virtual world, and the second client is set in a real world;

The first client and the second client update the first local model parameters and the second local model parameters based on the cross-chain aggregation method and aggregate them to obtain a new global model.

In a specific embodiment, a decentralized metaverse cross-chain federated machine learning framework is built using blockchain technology, wherein one blockchain is used as the main chain, and the main chain is used as the storage management module of the global model parameters; one blockchain is used as the first slave chain, and the first slave chain is used as the first storage module of the first local model parameters; one blockchain is used as the second slave chain, and the second slave chain is used as the second storage module of the second local model parameters; a relay chain is also provided, and the relay chain is used as the cross-chain management platform;

The task publisher sends the initialized global model through the main chain, and receives the updated first local model and second local model parameters through the main chain;

The first client receives the initialized global model through the first slave chain, and uploads the updated first local model parameters to the first slave chain;

The second client receives the initialized global model through the second slave link, and uploads the updated second local model parameters to the second slave link;

The relay chain realizes the cross-connection between the first slave chain, the second slave chain and the main chain.

In a specific embodiment, before the Metaverse cross-chain federated machine learning framework is trained, it is initialized as follows:

The first slave chain and the second slave chain register with the relay chain, and the task publisher determines the federated machine learning task and searches for the first client and the second client that participate in the federated machine learning training;

The first client and the second client determine whether to participate in the federated machine learning task according to their needs; after the two parties reach an agreement, they enter the federated machine learning process;

The task publisher uploads the initialized global model _wt to the main chain, and the first client in the virtual world and the second client in the real world obtain the initialized global model _wt through the corresponding first slave chain and second slave chain respectively.

In a specific embodiment, the federated machine learning process is as follows:

The second client and the first client receive the initialized global model w _t issued by the task publisher, and initialize the global model w _t respectively to obtain the second local model First local model/> The initialization process is as follows:

in, Represents the second client set in the real world,/> represents the first client set in the virtual world; then, the first client and the second client use the first local private data set D _j and the second local private data set D _i for training respectively, where D _i represents the second local private data set held by the second client i in the real world, and D _j represents the first local private data set held by the first client j in the virtual world; the loss functions to be optimized for the training of the second local model and the first local model are defined as follows:

real world:

virtual reality:

Where w is the global model parameter, |D _i | is the size of the data set of the second client i, |D _j | is the size of the data set of the first client j, f _k (w) is the local loss function, and each client performs stochastic gradient descent for a given number of iterations to train the second local model and the first local model as follows:

Where η is the learning rate, is the gradient.

In a specific embodiment, after the first client in the virtual world and the second client in the real world train the first local model and the second local model, the trained first local model parameters and the second local model parameters are uploaded to the first slave chain and the second slave chain respectively, and cross-chain requests are initiated to the main chain in the first slave chain and the second slave chain respectively. The first local model parameters and the second local model parameters are updated and uploaded to the main chain through the blockchain cross-chain method. The main chain verifies the received first local model parameters and the second local model parameters, and performs aggregation operations to obtain a new global model w _t+1 , which is defined as follows:

The federated machine learning task ends after a certain number of iterations until the global model converges; the learned global model will be retained in the main chain for subsequent prediction or classification work.

In a specific embodiment, there are four participants in the cross-chain aggregation method, and their functions are as follows:

Validator: responsible for the block generation of the relay chain network, running a relay chain client, and verifying the blocks generated by the blockchain nominated by it;

Collator: Maintains a full node of the blockchain, helps validators collect and verify the correctness of transactions, and submits candidate blocks to validators;

Nominator: A party that owns tokens and is responsible for maintaining and securing the validator.

Supervisor: Earn income by reporting illegal transactions or illegal blocks.

In a specific embodiment, the cross-chain process between the second slave chain and the master chain is as follows:

S101: The second client creates an account on the second slave chain and initializes the information stored on the second slave chain;

S102: After the second client completes training the second local model, the second client creates a transaction on the second slave chain and sends the second local model parameters and identity information to the main chain; the transaction is a second local model parameter upload request;

S103: The second client signs and broadcasts the transaction;

S104: The organizer of the second slave chain collects transaction information, verifies the validity of the transaction, organizes the transaction data, and packages it into a candidate block;

S105: The collation staff shows the candidate block and the proof of state transfer to the validator of the second slave chain;

S106: The validator verifies the received candidate block. Only if the verification is passed, ensuring that the candidate block contains only valid transactions, the validator will pledge its tokens to prepare for the candidate block to be put on the chain;

S107: When enough nominators stake their tokens and nominate validators, the broadcast of their candidate blocks to the blockchain will be authorized;

S108: When all validators reach a consensus on the relay chain block, the validator moves the transaction on the second slave chain from the exit of the second slave chain to the entrance of the main chain to complete the message transmission;

S109: The main chain executes the transaction in the entry queue and modifies its own ledger;

S110: The task issuing server obtains and temporarily stores the updated parameters of the second local model, and waits for several second clients to return results before performing an aggregation operation.

In a specific embodiment, the cross-chain process between the first slave chain and the master chain is as follows:

S201: The first client creates an account on the first slave chain and initializes the information stored on the first slave chain;

S202: After the first client completes training the first local model, the first client creates a transaction on the first slave chain and sends the first local model parameters and identity information to the master chain; the transaction is a first local model parameter upload request;

S203: The first client signs and broadcasts the transaction;

S204: First, the chain organizer collects transaction information, verifies the validity of the transaction, organizes the transaction data, and packages it into a candidate block;

S205: The collation staff shows the candidate block and the proof of state transfer to the validator of the first slave chain;

S206: The validator verifies the received candidate block. Only if the verification is passed, ensuring that the candidate block contains only valid transactions, the validator will pledge its tokens to prepare for the candidate block to be put on the chain;

S207: When enough nominators stake their tokens and nominate validators, the broadcast of their candidate blocks to the blockchain will be authorized;

S208: When all validators reach a consensus on the relay chain block, the validator moves the transaction on the first slave chain from the exit of the first slave chain to the entrance of the main chain to complete the message transmission;

S209: The main chain executes the transaction in the entry queue and modifies its own ledger;

S210: The task issuing server obtains and temporarily stores the updated parameters of the first local model, and waits for a number of first clients to return results before performing an aggregation operation.

The present invention does not simply upload all physical data collected by sensors directly to the virtual world, but divides the data into two parts according to data type and privacy protection requirements. Non-privacy data can be directly uploaded to the virtual world data storage module (such as cloud servers, edge servers, etc.), and privacy data with sensitive information is stored locally, thereby realizing privacy protection of artificial intelligence applications in the metaverse scenario.

The present invention uses blockchain technology to build a decentralized Metaverse cross-chain federated machine learning framework. This framework is different from the traditional federated machine learning framework. One blockchain in this framework is called the main chain, which serves as a storage management module for global model parameters, and multiple slave chains serve as storage modules for local model parameters. Local training model parameters are uploaded to the slave chains. The Metaverse cross-chain federated machine learning framework can record the training process of the local model, achieve privacy protection and the auditability and traceability of the local training process.

The present invention also proposes to adopt a cross-chain aggregation method, in which the interaction of local model parameters between the master and slave chains is carried out in a cross-chain form, ensuring traceability while also ensuring the security of local model parameter interaction to a certain extent.

Example 2

A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a data privacy protection method for a metaverse scenario are implemented:

The data is divided into two parts according to the data type and privacy protection requirements. The non-privacy data is directly uploaded to the data storage module in the virtual world and used as the first local private data set to train the first local model. The privacy data with sensitive information is stored locally and used as the second local private data set to train the second local model.

Constructing a Metaverse cross-chain federated machine learning framework with privacy protection, the Metaverse cross-chain federated machine learning framework comprising a task publisher for storing a global model, a first client for training a first local model through a first local private data set, and a second client for training a second local model through a second local private data set; the task publisher and the first client are set in a virtual world, and the second client is set in a real world;

The first client and the second client update the first local model parameters and the second local model parameters based on the cross-chain aggregation method and aggregate them to obtain a new global model.

Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripherals, voltage regulators, and power management circuits, which are all well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium via an antenna, and further, the antenna also receives data and transmits the data to the processor.

Example 3

A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of implementing a data privacy protection method for a metaverse scenario are as follows:

The data is divided into two parts according to the data type and privacy protection requirements. The non-privacy data is directly uploaded to the data storage module in the virtual world and used as the first local private data set to train the first local model. The privacy data with sensitive information is stored locally and used as the second local private data set to train the second local model.

Constructing a Metaverse cross-chain federated machine learning framework with privacy protection, the Metaverse cross-chain federated machine learning framework comprising a task publisher for storing a global model, a first client for training a first local model through a first local private data set, and a second client for training a second local model through a second local private data set; the task publisher and the first client are set in a virtual world, and the second client is set in a real world;

The first client and the second client update the first local model parameters and the second local model parameters based on the cross-chain aggregation method and aggregate them to obtain a new global model.

Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

Obviously, the above embodiments of the present invention are only examples for clearly explaining the present invention, and are not intended to limit the implementation methods of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the claims of the present invention.

Claims (10)

1. A data privacy protection method for a meta-universe scene is characterized by comprising the following steps of: the method comprises the following steps:

Dividing data into two parts according to data types and privacy protection requirements, and directly uploading non-private data to a data storage module in a virtual world to be used as a first local private data set to train a first local model; while private data with sensitive information is stored locally as a second local private data set training a second local model;

Constructing a metauniverse cross-chain federation machine learning framework with privacy protection, wherein the metauniverse cross-chain federation machine learning framework comprises a task publisher for storing a global model, a first client for training a first local model through a first local private data set and a second client for training a second local model through a second local private data set; the task publisher and the first client are arranged in the virtual world, and the second client is arranged in the real world;

The first client and the second client update and aggregate the first local model parameters and the second local model parameters based on a cross-chain aggregation method to obtain a new global model;

Constructing a decentralised meta universe cross-chain federation machine learning framework by using a blockchain technology, wherein one blockchain is used as a main chain, and the main chain is used as a storage management module of global model parameters; a first storage module taking one blockchain as a first slave chain and taking the first slave chain as a first local model parameter; a second storage module for taking one blockchain as a second slave chain and taking the second slave chain as a second local model parameter; and a relay chain is also arranged, and the relay chain is used as a cross-chain management platform.

2. The meta-universe scene-oriented data privacy protection method of claim 1, characterized by:

The task publisher transmits the initialized global model through the main chain, and receives updated parameters of the first local model and the second local model through the main chain;

The first client receives the initialized global model through a first slave chain and uploads the updated first local model parameters to the first slave chain;

the second client receives the initialized global model through a second slave chain and uploads the updated second local model parameters to the second slave chain;

the relay chain realizes bridging of the first slave chain, the second slave chain and the main chain.

3. The meta-universe scene-oriented data privacy protection method of claim 2, characterized by: before training the meta-universe cross-chain federal machine learning framework, initializing is carried out, and the method specifically comprises the following steps:

the first slave chain and the second slave chain register to the relay chain, a task publisher determines a federal machine learning task and searches a first client and a second client which participate in federal machine learning training;

the first client and the second client determine whether to participate in a federal machine learning task according to requirements; after the two parties reach the agreement, entering a federal machine learning process;

The task publisher uploads the initialized global model w _t to the main chain, and the first client of the virtual world and the second client of the real world acquire the initialized global model w _t through the corresponding first slave chain and the second slave chain respectively.

4. The meta-universe scene-oriented data privacy protection method of claim 3, characterized by: the federal machine learning process is specifically as follows:

The second client and the first client receive the initialized global model w _t issued by the task issuing party and initialize the global model w _t respectively to obtain a second local model First local model/>The initialization process is as follows:

Where P represents a second set of real-world clients, A first set of clients representing a virtual world; then, the first client and the second client respectively use the first local private data set D _j and the second local private data set D _i for training, D _i represents the second local private data set held by the real-world second client i, and D _j represents the first local private data set held by the virtual world first client j; the second local model and the first local model are respectively defined as follows:

Real world:

Virtual world:

Wherein w is a global model parameter, |d _i | is the size of the dataset of the second client i, |d _j | is the size of the dataset of the first client j, f _k (w) is a local loss function, each client performs random gradient descent for a given number of iterations, and the training results in a second local model, the first local model being as follows:

wherein, eta is the learning rate, Is a gradient.

5. The meta-universe scene-oriented data privacy protection method of claim 4, characterized by: after the first client in the virtual world and the second client in the real world train the first local model and the second local model, the trained first local model parameters and second local model parameters are respectively uploaded to a first slave chain and a second slave chain, a cross-link request is respectively initiated on the main chain by the first slave chain and the second slave chain, the first local model parameters and the second local model parameters are updated and uploaded to the main chain in a block chain cross-link mode, and the main chain verifies the received first local model parameters and second local model parameters and performs aggregation operation to obtain a new global model w _t+1, which is defined as follows:

Iterating for a certain number of times until the global model converges, and ending the federal machine learning task at the moment; the learned global model will be retained in the backbone for subsequent predictive or classification work services.

6. The meta-universe scene-oriented data privacy protection method of claim 5, characterized by: four participants are in total in the cross-chain polymerization method, and the functions of the four participants are as follows:

the verifier: the method comprises the steps of taking charge of the block out of a relay chain network, running a client of one relay chain, and verifying a block generated by a nominated blockchain;

and (3) finishing: maintaining all nodes of a blockchain, helping a verifier to collect and verify the correctness of a transaction and submitting candidate blocks to the verifier;

the nominator: the party concerned who owns the token maintains and is responsible for verifying the security of the person;

the supervisor: the benefit is obtained by the detection of illegal transactions or illegal blocks.

7. The meta-universe scene-oriented data privacy protection method of claim 6, characterized by: the process of crossing the chain between the second slave chain and the main chain is as follows:

s101: the second client creates an account on the second slave chain and initializes information stored on the second slave chain;

s102: after the second client finishes training the second local model, the second client creates a transaction on the second slave chain and sends the transaction including the second local model parameters and the identity information to the main chain; the transaction is a second local model parameter uploading request;

s103: the second client signs and broadcasts the transaction;

S104: collecting transaction information from a collator of the chain, verifying the validity of the transaction, collating transaction data, and packaging into candidate blocks;

s105: the collator presents the candidate block and the transition proof of state to a verifier of the second slave chain;

s106: the verifier verifies the received candidate block, only the verification passes, only effective transactions are ensured to be contained in the candidate block, and the verifier mortises the tokens of the candidate block to prepare for the uplink of the candidate block;

s107: broadcasting the candidate block to the blockchain will be authorized when there are enough nominators to mortgage their tokens and nominating the verifier;

s108: when all authenticators agree on the relay chain block, the authenticators move the transaction on the second slave chain from the exit of the second slave chain to the entrance of the master chain to complete the transmission of the message;

S109: the main chain executes the transaction in the entrance queue and modifies the account book of the main chain;

s110: the task issuing side server obtains the updated parameters of the second local model, temporarily stores the updated parameters, and performs aggregation operation after waiting for the return results of a plurality of second clients.

8. The meta-universe scene-oriented data privacy protection method of claim 6, characterized by: the first slave chain and the main chain have the following cross-chain flow path:

s201: the first client creates an account on the first slave chain and initializes information stored on the first slave chain;

S202: after the first client finishes training the first local model, the first client creates a transaction on the first slave chain and sends the first local model parameters and identity information to the main chain; the transaction is a first local model parameter uploading request;

s203: the first client signs and broadcasts the transaction;

S204: collecting transaction information by a collator of a chain, verifying the validity of a transaction, collating transaction data, and packaging into candidate blocks;

s205: the collator presents the candidate block and the transition proof of state to the verifier of the first slave chain;

S206: the verifier verifies the received candidate block, only the verification passes, only effective transactions are ensured to be contained in the candidate block, and the verifier mortises the tokens of the candidate block to prepare for the uplink of the candidate block;

S207: broadcasting the candidate block to the blockchain will be authorized when there are enough nominators to mortgage their tokens and nominating the verifier;

S208: when all authenticators agree on the relay chain block, the authenticators move the transaction on the first slave chain from the exit of the first slave chain to the entrance of the master chain to complete the transmission of the message;

s209: the main chain executes the transaction in the entrance queue and modifies the account book of the main chain;

S210: the task issuing side server obtains the updated parameters of the first local model, temporarily stores the parameters, and performs aggregation operation after waiting for the return results of a plurality of first clients.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized by: the processor, when executing the computer program, performs the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, characterized by: the computer program, when executed by a processor, performs the steps of the method according to any one of claims 1 to 8.