WO2020210979A1 - Blockchain-network-based joint learning method and joint learning device - Google Patents

Abstract

Provided are a blockchain-network-based joint learning method and joint learning device. The method comprises: acquiring an initial model parameter, wherein the initial model parameter is a parameter, used for building an initial model, of a device participating in joint learning; training the initial model parameter, and generating a training result used to update the initial model parameter; and sending the training result to an application server node by means of a state channel, wherein the state channel is an off-chain channel between the device and the application server node, and the off-chain channel is located outside a blockchain network. According to the technical solution provided in the present application, the device participating in joint learning feeds back a training result of the joint learning by means of the state channel, and can meet the requirement for high transactions per second of a combination of the joint learning and blockchain technology.

Description

Joint learning method and joint learning equipment based on blockchain network Technical field

This application relates to the field of information technology, and more specifically, to a joint learning method and a joint learning device based on a blockchain network.

Background technique

The proposed joint learning scheme makes it possible to share models on mobile devices. Joint learning refers to storing all training data on the mobile terminal, decoupling the model training of machine learning and the cloud storage process. This makes it possible to perform model training and evolution on the mobile terminal, and solves the problem that the previous model could only deliver the trained model in the cloud instead of training locally. In modern society, the privacy of personal data is becoming more and more important, and in supervised learning, training data is a necessary element.

In order to realize the process automation of the joint learning method and provide an incentive mechanism for the equipment participating in the joint learning, the industry has considered the scheme of combining the joint learning technology and the blockchain technology, but the method of combining the joint learning technology and the blockchain technology , That is, the training results of joint learning need to be shared with any node in the blockchain. If multiple mobile devices participate in joint learning, each training result needs to be shared by each node in the blockchain, which will cause the existing blockchain architecture to fail to meet the transaction throughput (transactions per second, TPS) Needs.

Summary of the invention

This application provides a joint learning method and a joint learning device based on a blockchain network. The method feeds back the training results of joint learning through a state channel, which can meet the high TPS requirements of the combination of joint learning and blockchain technology.

In the first aspect, a joint learning method based on a blockchain network is provided, including: obtaining initial model parameters, where the initial model parameters are parameters used by devices participating in the joint learning to establish an initial model; and training the initial model parameters , Generating a training result for updating the initial model parameters; sending the training result to an application server node through a state channel, where the state channel is an off-chain channel between the device and the application server node, and the off-chain The channel is outside the blockchain network.

In the joint learning method provided by this application, the device can send the training results of participating in the joint learning through the state channel, that is, a state channel can be established between the application server node and the device participating in the joint learning, so that the device does not need to participate in the joint learning for local training The training results obtained later are shared at multiple nodes of the blockchain network, which can meet the high TPS requirements of the combination of joint learning and blockchain technology. In addition, sending the training results through the state channel can avoid the problem of sharing the training results at each node of the blockchain network and ensure the data privacy of the training results.

It should be understood that the State Channel is an "off-chain" technology used to execute transactions and other state updates. That is, the state channel can refer to the off-chain channel of the blockchain, through the off-chain channel. The data or information sent does not need to be shared among the nodes of the blockchain, and the off-chain channel can be an off-chain channel established between two nodes.

For example, the state channel can be a one-way off-chain payment channel. For example, a one-way off-chain payment channel can be established between the application server node and the device participating in the joint learning. The one-way off-chain payment channel is used by the application server node to receive the device After sending the training result of participating in the joint learning, the terminal device is paid with digital currency.

For example, the state channel can be a two-way off-chain channel, such as a two-way off-chain channel established between an application server node and a device participating in joint learning, and the device can participate in the local training result of the joint learning to the application server node through the two-way off-chain channel. The application server node can pay the device with digital currency after receiving the training result sent by the device.

It should be noted that the initial model can be a machine learning model or a computer algorithm model. For example, the initial model can be a neural network model used in machine learning. When the initial model is a neural network model, the initial parameters can be used for Establish the weight value of the neural network model. When the initial model is a computer algorithm model, the initial parameters may be the parameters required to establish the algorithm model. The above are examples and do not limit this application.

With reference to the first aspect, in some implementations of the first aspect, the method further includes: receiving channel address information sent by the application server node, where the channel address information is used to identify the state channel.

With reference to the first aspect, in some implementations of the first aspect, the method further includes: obtaining a joint learning smart contract, where the joint learning smart contract includes training logic instructions for the initial model parameters;

The training of the initial model parameters and generating a training result for updating the initial model parameters includes: running the joint learning smart contract to train the initial model parameters to generate the training result.

In a possible implementation, the joint learning smart contract may include data information required for joint learning, content requirements for training results fed back by the device to the application server, and cost information corresponding to the training results fed back by the device.

With reference to the first aspect, in some implementations of the first aspect, the obtaining a joint learning smart contract includes: obtaining the joint learning smart contract through the blockchain network, wherein the joint learning smart contract is An off-chain smart contract, the execution environment of the off-chain smart contract does not belong to the smart contract of the blockchain network.

In this application, the joint learning smart contract can be a smart contract that deploys an off-chain smart contract on the blockchain network, where the off-chain smart contract means that the execution environment of the smart contract does not belong to the blockchain network Smart contract. Deploying the off-chain smart contract on the blockchain network can ensure the credibility of the joint smart contract.

With reference to the first aspect, in some implementations of the first aspect, the running the joint learning smart contract includes: running the joint learning smart contract in a joint learning application.

For example, the equipment participating in the joint learning can install the joint learning application, the application server node can publish the address of the joint learning smart contract to the joint learning application, and the equipment participating in the joint learning can download and run the joint learning application in the joint learning application. Learn smart contracts.

With reference to the first aspect, in some implementations of the first aspect, the running the joint learning smart contract in a joint learning application to train the initial model parameters includes: running under a trusted execution environment TEE The joint learning smart contract is used to train the initial model parameters.

In this application, the device can run the joint learning smart contract in a trusted execution environment, thereby ensuring the security of running the joint smart contract.

For example, equipment based on system-on-chip (SoC) hardware can provide a three-layer hardware security system, including rich execution environment (REE), trusted execution environment (TEE) and safe operation Environment (secure execution environment, SEE), in which TEE can run security-sensitive programs and store security-sensitive data.

For example, the joint learning smart contract may be run in a joint learning application under a trusted execution environment TEE to train the initial model parameters.

With reference to the first aspect, in some implementations of the first aspect, after the sending the training result to the application server node through the state channel, the method further includes: receiving the application server node to pay through the state channel Digital currency.

With reference to the first aspect, in some implementations of the first aspect, the method further includes: sending a first message to the application server node, where the first message is used to instruct the device to participate in joint learning, the The first message includes the digital currency wallet address of the device.

In this application, the device can send a message to the application server node indicating that it participates in joint learning and sends the information of the digital currency wallet address, so as to obtain the corresponding digital currency after sending the training result to the application server node.

It should be understood that the digital currency wallet address can be used to identify the device. When multiple devices participate in joint learning, the application server node distinguishes different devices according to the digital currency wallet addresses of multiple devices, so that after receiving the training results of joint learning sent by multiple devices, they pay digital currency to multiple devices respectively .

With reference to the first aspect, in some implementations of the first aspect, the receiving the digital currency paid by the application server node through the state channel includes: receiving the application server node under a trusted execution environment TEE Digital currency paid through the state channel.

With reference to the first aspect, in some implementations of the first aspect, the method further includes: determining the end of the joint learning; sending a first transaction to the blockchain network, the first transaction being used to instruct to close the Status channel.

It should be understood that the device can receive many transactions sent by the application server node. Therefore, the digital currency of the device in the state channel is continuously increasing, and only the latest transactions need to be maintained. When the device decides to close the state channel, it can sign the latest transaction, send it to the blockchain network, and take out the digital currency belonging to the device.

In a second aspect, a joint learning method based on a blockchain network is provided, including: sending an initial model parameter to a device, where the initial model parameter is a parameter used by the device to establish an initial model participating in joint learning; The channel receives the training result of updating the initial model parameters, wherein the state channel is an off-chain channel between the device and the application server node, and the off-chain channel is located outside the blockchain network.

In the joint learning method based on the blockchain network provided in this application, the application server node can receive the training result sent by the device participating in the joint learning through the state channel, that is, the state channel can be established between the application server node and the device participating in the joint learning, Therefore, it is not necessary to share the training results obtained after the device participates in joint learning for local training at multiple nodes of the blockchain network, which can meet the high TPS requirements of the combination of joint learning and blockchain technology.

It should be understood that the State Channel is an "off-chain" technology used to execute transactions and other state updates. That is, the state channel can refer to the off-chain channel of the blockchain, through the off-chain channel. The data or information sent does not need to be shared among the nodes of the blockchain, and the off-chain channel can be an off-chain channel established between two nodes.

For example, the state channel can be a one-way off-chain payment channel. For example, a one-way off-chain payment channel can be established between the application server node and the device participating in the joint learning. The one-way off-chain payment channel is used by the application server node to receive the terminal After the training result of participating in the joint learning sent by the device, the digital currency is paid to the participating device in the joint learning.

For example, the state channel can be a two-way off-chain channel, such as a two-way off-chain channel established between an application server node and a device participating in joint learning, and the device can participate in the local training result of the joint learning to the application server node through the two-way off-chain channel. The application server node can pay the device with digital currency after receiving the training result sent by the device.

With reference to the second aspect, in some implementations of the second aspect, the method further includes: receiving channel address information sent by a blockchain network, where the channel address information is used to identify the state channel; Send the channel address information.

With reference to the second aspect, in some implementations of the second aspect, before the receiving the channel address information sent by the blockchain network, the method further includes: sending a second transaction to the blockchain network, the second The transaction is used to deploy the state channel.

In this application, the application server node may send transactions to be deployed in the off-chain channel between the devices participating in the joint learning, so as to receive the training result sent by the device to update the initial model parameters through the off-chain channel. It can not only meet the needs of TPS, but also protect the data privacy of training results.

With reference to the second aspect, in some implementations of the second aspect, the method further includes: sending a third transaction to the blockchain network, where the third transaction is used to deploy a joint learning smart contract on the Blockchain network, the joint learning smart contract is an off-chain smart contract, the execution environment of the off-chain smart contract does not belong to the blockchain network, and the joint learning smart contract includes the training of the initial model parameters Logic instructions.

In this application, the application server node can deploy the off-chain smart contract in the smart contract of the blockchain network, where the off-chain smart contract refers to the smart contract whose execution environment does not belong to the blockchain network . Deploying the off-chain smart contract on the blockchain network can ensure the credibility of the joint smart contract.

With reference to the second aspect, in some implementations of the second aspect, after receiving the training result through the state channel, the method further includes: paying digital currency to the device through the state channel.

With reference to the second aspect, in some implementations of the second aspect, before the payment of digital currency to the device through the state channel, the method further includes: receiving a first message sent by the device, and The first message is used to instruct the device to participate in joint learning, and the first message includes the digital currency wallet address of the device.

It should be understood that the digital currency wallet address can be used to identify the device. When multiple devices participate in joint learning, the application server node distinguishes different devices according to the digital currency wallet addresses of multiple devices, so that after receiving the training results of joint learning sent by multiple devices, they pay digital currency to multiple devices respectively .

In a third aspect, a joint learning device is provided, including: a receiving unit for receiving initial parameters sent by an application server node, where the initial model parameters are parameters used by devices participating in the joint learning to establish an initial model; a processing unit, Used to train the initial parameters and generate training results for updating the initial parameters; a sending unit, used to send the training results to the application server node through a state channel, where the state channel is the device and the application server node The off-chain channel is located outside the blockchain network.

With reference to the third aspect, in some implementations of the third aspect, the receiving unit is further configured to: receive channel address information sent by the application server node, where the channel address information is used to identify the state channel.

With reference to the third aspect, in some implementations of the third aspect, the receiving unit is further configured to: obtain a joint learning smart contract, and the joint learning smart contract includes training logic instructions for the initial model parameters; The processing unit is specifically configured to: run the joint learning smart contract to train the initial model parameters, and generate training results for updating the initial model parameters.

With reference to the third aspect, in some implementations of the third aspect, the receiving unit is specifically configured to: obtain the joint learning smart contract through the blockchain network, wherein the joint learning smart contract is off-chain Smart contract, the execution environment of the off-chain smart contract does not belong to the blockchain network.

With reference to the third aspect, in some implementations of the third aspect, the processing unit is specifically configured to: the device runs the joint learning smart contract in a joint learning application.

With reference to the third aspect, in some implementations of the third aspect, the processing unit is specifically configured to run the joint learning smart contract to train the initial model parameters in a trusted execution environment TEE.

With reference to the third aspect, in some implementation manners of the third aspect, the receiving unit is further configured to: receive the digital currency paid by the application server node through the state channel.

With reference to the third aspect, in some implementations of the third aspect, the sending unit is further configured to send a first message to the application server node, where the first message is used to instruct the device to participate in joint learning, The first message includes the digital currency wallet address of the device.

With reference to the third aspect, in some implementation manners of the third aspect, the receiving unit is specifically configured to receive the digital currency paid by the application server node through the state channel under the trusted execution environment TEE.

With reference to the third aspect, in some implementations of the third aspect, the processing unit is further configured to: determine the end of the training task of the joint learning; the sending unit is further configured to: send the first one to the blockchain network Transaction, the first transaction is used to instruct to close the state channel.

In a fourth aspect, a joint learning device is provided, including: a sending unit, configured to send initial model parameters to the device, where the initial model parameters are parameters used by the device to establish an initial model participating in joint learning; and a receiving unit, It is used to receive the training result for updating the initial model parameters through a state channel, wherein the state channel is an off-chain channel between the device and the application server node, and the off-chain channel is located in the blockchain network outer.

With reference to the fourth aspect, in some implementations of the fourth aspect, the receiving unit is further configured to: receive channel address information sent by the blockchain network, where the channel address information is used to identify the state channel; The sending unit is further configured to send the channel address information to the device.

With reference to the fourth aspect, in some implementations of the fourth aspect, the sending unit is further configured to send a second transaction to the blockchain network, and the second transaction is used to deploy the state channel.

With reference to the fourth aspect, in some implementations of the fourth aspect, the sending unit is further configured to: send a third transaction to the blockchain network, and the third transaction is used to sign the joint learning smart contract on In the blockchain network, the joint learning smart contract is an off-chain smart contract, the execution environment of the off-chain smart contract does not belong to the blockchain network, and the joint learning smart contract includes the initial model parameters Training logic instructions.

With reference to the fourth aspect, in some implementations of the fourth aspect, the sending unit is further configured to: pay digital currency to the device through the state channel.

With reference to the fourth aspect, in some implementations of the fourth aspect, the receiving unit is further configured to: receive a first message sent by the device, where the first message is used to instruct the terminal device to participate in joint learning, The first message includes information of the digital currency wallet address of the terminal device.

In a fifth aspect, a joint learning device is provided, including: a processor, a memory, the memory is used to store a computer program, the processor is used to call and run the computer program from the memory, so that the joint learning device executes the first aspect And its various possible implementation methods of joint learning.

For example, the joint learning device may be a terminal device.

Optionally, there are one or more processors and one or more memories.

Optionally, the memory may be integrated with the processor, or the memory and the processor may be provided separately.

In a sixth aspect, a joint learning device is provided, including: a processor, a memory, the memory is used to store a computer program, the processor is used to call and run the computer program from the memory, so that the joint learning device executes the second aspect And its various possible implementation methods of joint learning.

For example, the joint learning device may be an application server node.

Optionally, there are one or more processors and one or more memories.

Optionally, the memory may be integrated with the processor, or the memory and the processor may be provided separately.

In a seventh aspect, a computer program product is provided. The computer program product includes: a computer program (also referred to as code, or instruction) that, when the computer program is executed, causes a computer or any one of at least one processing The device executes the methods in the first aspect and its various implementations.

In an eighth aspect, a computer program product is provided. The computer program product includes: a computer program (also called code, or instruction) that, when the computer program is executed, causes a computer or any one of at least one processing The device executes the methods in the second aspect and its various implementations.

In a ninth aspect, a computer-readable medium is provided, and the computer-readable medium stores a computer program (also called code, or instruction) when it runs on a computer or any at least one processor, so that The computer or the processor executes the methods in the first aspect and various implementations thereof.

In a tenth aspect, a computer-readable medium is provided, and the computer-readable medium stores a computer program (also called code, or instruction) when it runs on a computer or any at least one processor, so that The computer or the processor executes the methods in the second aspect and various implementations thereof.

In an eleventh aspect, a chip system is provided. The chip system includes a processor for supporting a server in a computer to implement the functions involved in the first aspect and various implementations thereof.

In a twelfth aspect, a chip system is provided, and the chip system includes a processor for supporting a server in a computer to implement the functions involved in the second aspect and various implementation manners thereof.

Description of the drawings

Figure 1 is an exemplary flow chart of a joint learning technology in the prior art;

Fig. 2 is a schematic flowchart of a joint learning method provided according to an embodiment of the present application;

FIG. 3 is a schematic flowchart of a joint learning method provided according to another embodiment of the present application;

Fig. 4 is a schematic structural diagram of a joint learning device provided according to an embodiment of the present application;

Fig. 5 is a schematic structural diagram of a joint learning device provided according to another embodiment of the present application;

Fig. 6 is a schematic structural diagram of a joint learning device provided according to an embodiment of the present application;

Fig. 7 is a schematic structural diagram of a joint learning device provided according to another embodiment of the present application;

Fig. 8 is a schematic structural diagram of a joint learning device provided according to another embodiment of the present application.

detailed description

The technical solution in this application will be described below in conjunction with the drawings.

First, introduce the concepts of blockchain technology, smart contracts, and joint learning involved in the embodiments of this application for a brief explanation.

1. Smart contract

A smart contract is a collection of code and data, and can also be called a "programmable contract". Generally speaking, smart contracts are defined by program coding and preset operating conditions; when the operating conditions are triggered, the behavior is executed. The "intelligence" is the execution intelligence, that is, if a certain preset condition is reached, the contract will run automatically.

2. Joint learning

As shown in Figure 1, it is a schematic flowchart of joint learning. Including step 110 to step 140.

110. The model request sends initial parameters to one or more devices participating in the joint learning, where the initial parameters are used to provide the one or more devices with an initial model participating in the joint learning.

It should be understood that the initial model can be a machine learning model or a computer algorithm model. For example, the initial model can be a neural network model used in machine learning. When the initial model is a neural network model, the initial parameters can be used to establish neural networks. The weight value of the network model. When the initial model is a computer algorithm model, the initial parameters may be the parameters required to establish the algorithm model. The above are examples and do not limit this application.

It should be understood that the model requester can be an application server node or an application APP in an application server node. The device may be a terminal device, for example, it may be a user equipment, a mobile device, a user terminal, a terminal, a wireless communication device, or a user device. The above are examples and do not limit this application.

120. The equipment participating in the joint learning performs local training according to the acquired initial model parameters, and obtains a training result for updating the initial model parameters.

130. The equipment participating in the joint learning feeds back the training results to the model demander.

It should be noted that the training result may be to improve the initial model by learning based on the private data of the user using the device, and then compress the changed part of the initial model into a small update package for feedback.

For example, when the initial model is a neural network model, the changed part of the initial model may be the changed part of the network weight value of the initial model.

140. The model demander adjusts the model parameters according to the training results fed back by one or more devices participating in the joint learning.

In other words, in the joint learning scenario, the devices participating in the joint learning can download the current latest initial model, perform local training on the latest initial model to improve the initial model, and then compress the changed part of the initial model into a small update package. The updated part of the model is sent to the model demander using an encrypted communication method, and the model demander averages the model update parts received from the feedback of the equipment participating in the joint learning to improve the initial model. Therefore, all the training data is on the equipment participating in the joint learning, and the personal privacy data of the user used for local training in the equipment is not sent to the model requester, only the changed part of the model is sent.

3. Off-chain payment channel technology

The off-chain payment channel of the blockchain is one of the technical routes for off-chain expansion, that is, a general technical solution for updating some transaction-related states in the blockchain outside the blockchain. The core idea is that a party involved in a transaction establishes an off-chain communication channel, so that both parties can perform interactive behaviors in the off-chain channel. During the period, state updates are not submitted to the main chain miners. When the state channel needs to be closed, the final state is passed. "Close Channel Transaction" is submitted to the main chain and synchronized to the main chain ledger. Since the intermediate process does not interact with the main chain, the state channel has a very high execution efficiency.

4. Blockchain technology

Blockchain technology realizes a chained data structure in which data and information blocks are sequentially connected in chronological order, and it is cryptographically guaranteed that cannot be tampered with and cannot be forged distributed storage. In general, the data and information in the blockchain are called "transactions".

Blockchain technology is not a single technology, but as a system integrating point-to-point transmission, consensus mechanism, distributed data storage, and cryptographic principles. The system has the technical characteristics of full disclosure and tamper-proof.

First, point-to-point transmission: the nodes participating in the blockchain are independent and peer-to-peer, and the synchronization of data and information between nodes is achieved through point-to-point transmission technology. The nodes can be different physical machines or different instances of the cloud.

Second, the consensus mechanism: The consensus mechanism of the blockchain refers to the process by which nodes participating in multiple parties achieve agreement on specific data and information through the interaction between nodes under preset logical rules. Consensus mechanisms need to rely on well-designed algorithms, so there are certain differences in the performance of different consensus mechanisms (such as TPS), the delay in reaching a consensus, the cost of computing resources, and the cost of transmission resources.

Third, distributed data storage: Distributed storage in the blockchain is that each node participating in the blockchain stores independent and complete data, which ensures that the data stored in the node is fully open. Different from traditional distributed data storage, traditional distributed data storage divides the data into multiple copies for backup or synchronous storage according to certain rules, while blockchain distributed data storage relies on the equivalent, location-based data storage in the blockchain. Consensus between independent nodes to achieve highly consistent data storage.

Fourth, the principle of cryptography: Blockchain is usually based on asymmetric encryption technology to achieve credible information dissemination and verification.

The concept of "block" is to organize one or more data records in the form of "blocks". The size of "blocks" can be customized according to actual application scenarios; and "chain" is a data structure, which The "blocks" storing data records are connected in chronological order and connected by HASH technology. In the blockchain, each "block" contains two parts: a "block header" and a "block body". The "block body" contains the transaction records packaged into the "block"; the "block header" contains " The root HASH of all transactions in the block and the HASH of the previous "block". The data structure of the blockchain ensures that the data stored on the blockchain is non-tamperable.

In the existing technology block chain and joint learning combination scheme, the training results of each training of the device are sent to the nodes in the block chain network for processing. That is, the training results after each device participates in joint learning for local training need to be shared among multiple nodes in the blockchain. Every training is on the chain. When the device is tens of millions of mobile phones, there is a high demand for transaction throughput (transactions per second, TPS). The existing blockchain architecture cannot meet the requirements of high TPS.

In view of this, this application proposes a method that combines blockchain and joint learning. The training result of the device participating in the joint learning can be sent to the model requester through the state channel, that is, the application server node can be connected with the device participating in the joint learning State channels are established between them, so there is no need to share the training results obtained after the device participates in joint learning for local training at multiple nodes in the blockchain network, which can meet the high requirements of combining joint learning and blockchain technology. TPS requirements.

In this application, the device may be a terminal device, for example, it may be a user equipment, a mobile device, a user terminal, a terminal, a wireless communication device, or a user device. The terminal device can also be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), and a wireless communication Functional handheld devices, computing devices, or other processing devices connected to wireless modems, in-vehicle devices, wearable devices, terminal devices in the future 5G network or future evolution of the public land mobile network (PLMN) Terminal equipment, etc., this embodiment of the present application does not limit this.

The following describes the method 200 for combining blockchain and joint learning provided by an embodiment of the present application with reference to FIG. 2. FIG. 2 shows a schematic flowchart of the method 200 for combining blockchain and joint learning provided by an embodiment of the present application. The method includes step 210 to step 230.

Step 210: The device receives an initial model parameter sent by the application server node, where the initial model parameter is a parameter used by the device participating in the joint learning to establish an initial model.

It should be noted that the initial model can be a machine learning model or a computer algorithm model. For example, the initial model can be a neural network model used in machine learning. When the initial model is a neural network model, the initial parameters can be used for Establish the weight value of the neural network model. When the initial model is a computer algorithm model, the initial parameters may be the parameters required to establish the algorithm model. The above are examples and do not limit this application.

It should be understood that, in this application, the application server node may be a model demander as shown in FIG. 1, for example, the application server node may be a server, a device used to provide computing services. Alternatively, the application server node may also be an application program APP in the application server node.

It should be understood that, in the embodiments of the present application, the device and application server nodes may be light nodes, that is, nodes that have transaction data related to themselves. For the blockchain network, if the full node, that is, the node that has all the transaction data of the entire network, is regarded as a node in the blockchain network, the equipment and application server nodes do not belong to the blockchain network; if the full node is Both and light nodes are regarded as nodes in the blockchain network, and the equipment and application server nodes belong to the blockchain network.

Step 220: The device trains the initial model parameters, and generates a training result for updating the initial model parameters.

Among them, the training result may be data information generated by the device learning according to the private data to improve the initial model, and then compressing the changed part of the initial model.

For example, when the initial model is a neural network model, the training result can be the changed part of the neural network weight value, that is, assuming that the initial parameter is the weight value W1 of the neural network model, the equipment participating in the joint learning can train the neural network locally. The weight value of the neural network model obtained by training is W2, and the training result may be the difference between W1 and W2.

In one example, the device participating in the joint learning can obtain the joint learning smart contract, and the joint learning smart contract may include the training logic instructions of the initial model parameters; the device may train the initial model parameters according to the obtained joint learning smart contract, Generate training results. Among them, the structure design of the joint learning smart contract can be shown in Table 1.

Table 1

Joint smart contract structure
1. Demand for user data, data processing flow
2. Model architecture (DNN super parameter)
3. Model training logic
4. Device return content
5. Single training cost

As shown in Table 1, the joint learning smart contract can include the data information required for joint learning, the content requirements of the training results fed back to the application server node by the equipment participating in the joint learning, and the corresponding cost of the equipment participating in the joint learning feedback training result. information.

It should be understood that the initial model can be a machine learning model. For example, the initial model can be a neural network model used in machine learning. When the initial model is a neural network model, the initial parameters can be the weight values used to establish the neural network model. Smart contract The training logic instruction of the initial model included in the DLL may be to execute a gradient descent algorithm.

It should be understood that the initial model may be a computer algorithm model. For example, when the initial model is a computer algorithm model, the initial parameters may be the parameters required to establish the algorithm model, and the training logic instructions for the initial model included in the smart contract may be performed by the computer once. Algorithm required for model calculation. The above are examples and do not limit the application. In the embodiment of this application, the joint learning smart contract may be a smart contract deployed by the off-chain smart contract on the blockchain network, where the off-chain smart contract means that the execution environment of the smart contract does not belong to the zone. The smart contract of the blockchain network.

For example, the application server node can obtain a smart contract that redesigns the structure of the off-chain smart contract. The redesigned off-chain smart contract can meet the needs of joint learning, that is, the smart contract can be used as the carrier of the joint learning program, such as , Joint learning can be a piece of code encapsulated in a smart contract, and the redesigned off-chain smart contract can be regarded as a joint learning smart contract. The application server node may send a transaction (for example, a third transaction) to the blockchain network, and the transaction may be used to deploy an off-chain smart contract on the blockchain network.

Further, the application server node can also publish the address of the joint learning smart contract to accept the third-party organization to review the joint learning program, thereby ensuring the credibility of the joint learning program. It should be understood that the address of the joint smart contract may be the hash value of the joint smart contract, and the third-party organization may find the joint learning smart contract in the blockchain network according to the hash value of the joint smart contract. The information published by the application server node can be shown in Table 2.

Table 2

Transaction ID (Transaction ID)
From: Application server node address (demand-side Address)
Hash String: The hash value encoded by the joint learning program
Data: Joint learning of smart contract procedures

In one example, the devices participating in the joint learning can obtain the joint learning smart contract through the blockchain network, where the joint learning smart contract is a smart contract that the application server node deploys the off-chain smart contract on the blockchain network.

For example, the devices participating in the joint learning can install the joint learning application, the application server node will publish the address of the joint learning smart contract to the joint learning application, and the devices participating in the joint learning can download the joint learning intelligence in the joint learning application contract.

It should be understood that the joint learning smart contract deployed on the blockchain network is shared among all nodes of the blockchain network. Therefore, to a certain extent, the joint learning smart contract can be prevented from being maliciously tampered with, and the joint smart contract can be ensured Credibility.

In an example, the device participating in the joint learning may also obtain the joint learning smart contract from the application server node. This application does not make any restrictions on the specific source of the joint learning smart contract acquired by the device while ensuring the authenticity of the joint learning smart contract.

In the embodiment of this application, after the device obtains the joint learning smart contract, it can send a first message to the application server node. The first message is used to instruct the device to participate in the joint learning, that is, the device can indicate by sending the first message to the application server node Participate in joint learning, where the first message also includes the digital currency wallet address of the device. After the device sends the local training result of the joint learning to the application server, the application server node can pay digital currency according to the digital currency wallet address of the device.

It should be understood that the digital currency wallet address can be used to identify devices participating in joint learning. When multiple devices participate in joint learning, the application server node distinguishes different devices according to the digital currency wallet addresses of multiple devices, so that after receiving the training results of joint learning sent by multiple devices, they pay digital currency to multiple devices respectively .

In one example, after receiving the digital currency wallet address of the device, the application server node can publish a transaction (for example, the second transaction) to the blockchain network, deploy an off-chain channel contract, and establish a connection between the application server node and the device The off-chain channel in between, where the off-chain channel is located outside the blockchain network.

Further, the blockchain network sends the address information of the state channel to the application server, and the address information of the state channel is used to identify the state channel.

It should be noted that the channel address information may be an index value, and the application server node can find the established state channel according to the index value. For example, the channel address information may be the hash value of the state channel.

It should be understood that the State Channel is an "off-chain" technology used to execute transactions and other state updates. That is, the state channel can refer to the off-chain channel of the blockchain network. The data or information sent by the channel does not need to be shared among the nodes of the blockchain, and the off-chain channel can be an off-chain channel established between two nodes. For example, the state channel can be a one-way off-chain payment channel. For example, a one-way off-chain payment channel can be established between the application server node and the device, and the one-way off-chain payment channel is used for the application server node to participate in the alliance after receiving the device sent After learning the training results, the digital currency is paid to the device.

For example, the state channel can be a two-way off-chain channel, such as a two-way off-chain channel established between an application server node and a device. The device can participate in the local training results of joint learning from the application server node through the two-way off-chain channel, and the application server node can After receiving the training result sent by the device, the digital currency is paid to the device.

In the embodiment of the present application, the device is based on system-on-chip (SoC) hardware, and can provide a three-layer hardware security system, including rich execution environment (REE) and trusted execution environment (REE) , TEE) and secure execution environment (secure execution environment, SEE). Among them, REE runs security-insensitive programs to save security-insensitive data, TEE runs security-sensitive programs and saves security-sensitive data, and SEE runs financial payment high-security programs and saves High security data for financial payment.

As an example and not limitation, the SoC may provide at least one operating environment such as Trustzone, Bowmore, eSE, or inSE. The operating environment of Trustzone or Bowmore may be referred to as TEE, and the operating environment of eSE or inSE may be referred to as SEE.

For example, when the device is a smart phone, an SoC can be configured in the smart phone, and the smart phone provides operating environments such as Trustzone, Bowmore, eSE, and inSE through the SoC.

As an example and not a limitation, the foregoing SoC may support running an instruction set based on an advanced RISC machine (ARM) architecture, and an SoC that supports running an instruction set based on the ARM architecture is called an ARM-based SoC. For example, an ARM-based SoC can be configured on the device to provide a three-layer hardware security system for the device.

In one example, the joint learning smart contract is run under the trusted execution environment TEE to train the initial model parameters.

For example, the device may run a joint learning smart contract in a joint learning application under a trusted execution environment TEE to train the initial model parameters and generate training results for updating the initial model parameters.

Step 230: Send the training result to the application server node through a state channel. The state channel refers to an off-chain channel between the device and the application server node, and the off-chain channel is located outside the blockchain network.

For example, the state channel can be a one-way off-chain payment channel. For example, a one-way off-chain payment channel can be established between the application server node and the device, and the one-way off-chain payment channel is used for the application server node to participate in the alliance after receiving the device sent After learning the training results, the digital currency is paid to the device.

It should be understood that in the embodiments of the present application, digital currency refers to electronic currency that is created, issued, and circulated by relying on checksum encryption technology. Its characteristic is to use P2P peer-to-peer network technology to issue, manage and circulate currency.

For example, the state channel can be a two-way off-chain channel. For example, the two-way off-chain channel established between the application server node and the device of the joint learning device. The device can send the local training results of the joint learning to the device through the two-way off-chain channel. The server node can pay digital currency to the device after receiving the training result sent by the device.

The application server node can adjust the initial model parameters according to the received training result sent by the device. Among them, the device may refer to one or more devices participating in the joint learning.

In the embodiment of the present application, after the device sends the training result to the application server node through the state channel, the application server node can pay the device with digital currency according to the acquired digital currency wallet address of the device.

For example, an application server node can send a transaction to the device through a state channel (for example, an off-chain payment channel), where the total digital currency in the channel at this time can be S(A+B=S), the digital currency of the application server node Is A, the digital currency of the device is B, and the cost of a single training is t. After this transaction, the digital currency of the application server node is At, and the digital currency of the device is B+t. In this transaction, the signature of the application server node is attached. After the device signs the transaction, it can send the transaction to the blockchain network.

In one example, in order to ensure the security and credibility of the payment environment for digital currency, the device may receive the digital currency paid by the application server node through the state channel under the trusted execution environment TEE.

In the embodiment of the present application, the device determines that the training task of joint learning ends; the device may send a transaction (for example, the first transaction) to the blockchain network, and the first transaction may be used to instruct to close the state channel.

It should be understood that the device can receive many transactions sent by the application server node. Therefore, the digital currency of the device in the state channel is continuously increasing, and only the latest transactions need to be maintained. When the device decides to close the state channel, it can sign the latest transaction, send it to the blockchain network, and take out the digital currency belonging to the device.

In the embodiment of the present application, the training parameters generated after the device participates in the joint learning for local training can be fed back to the application server node through the state channel, and the training result can be sent to the application server node through the state channel to meet the high TPS in the joint learning and blockchain combination scenario At the same time, the training results of the device also have a certain degree of privacy. The training results are sent through the state channel to avoid the sharing of each node in the blockchain network and protect the privacy of the device data.

The following describes the specific process of the joint learning method in the embodiment of the present application with reference to FIG. 3.

FIG. 3 is a schematic flowchart of a joint learning method provided by an embodiment of the application. The method shown in FIG. 3 includes steps 301 to 314, and steps 301 to 314 are respectively described in detail below.

It should be understood that the model requester in FIG. 3 may be the application server node shown in FIG. 2, or may also be the application APP in the application server node. The device may be a terminal device, for example, it may be a user equipment, a mobile device, a user terminal, a terminal, a wireless communication device, or a user device. The above are examples and do not limit this application.

Step 301: The device installs a joint learning APP locally.

Among them, the joint learning APP can collect different personal privacy data for each different joint learning smart contract. Do data preprocessing. At the same time, the initialization model data w provided by the model requester will be received.

In an example, as shown in Figure 4, an ARM-based SoC can provide a three-layer hardware security system for the device. The joint learning App can ensure the safety of the data processing process in the TEE trusted environment. After a reliable joint learning App is installed on the device, multiple joint learning smart contracts can be loaded at the same time to meet the needs of multiple model demanders.

For example, different personal privacy data collected by the United Learning APP can be stored in a local personal privacy data database. Among them, personal privacy data is stored in the trusted environment of TEE, which can protect the security of personal privacy data from other malicious software; on the other hand, it can ensure that the personal privacy data provided to users who use the device are trusted Yes, it is really valuable to the demand side of the model.

Step 302: The model requester deploys the off-chain smart contract to the chain, and accepts audit to ensure credibility and traceability.

For example, the model demander deploys off-chain smart contracts on the chain through transactions. At the same time, the address of the off-chain smart contract can be published, for example, the address of the off-chain smart contract can be published to the joint learning APP. Thereby accepting third-party institutions to review the joint learning program to ensure credibility.

It should be understood that the address of the off-chain smart contract may refer to the hash value of the off-chain smart contract, and the off-chain smart contract can be found according to the address of the off-chain smart contract. It should be noted that, unlike ordinary smart contracts in the blockchain, the execution environment for smart contracts under the chain is not on the chain, but on the off-chain mobile phone. The joint learning can be combined with blockchain and off-chain smart contracts to ensure the credibility of the joint learning program. Among them, the structure design of the off-chain smart contract can be shown in Table 1.

Step 303: The device may receive a recommendation of a program for joint learning of smart contracts, and download the off-chain smart contract program deployed on the blockchain network.

Step 304: The device sends the digital currency wallet address to the model demander, and indicates to the model demander to participate in the joint learning process.

Step 305: The model demander publishes a transaction (for example, a second transaction) to the blockchain network for deploying an off-chain channel contract and establishing a state channel with the device (for example, an off-chain payment channel).

Among them, the state channel established by the application server node and the device can be a one-way off-chain payment channel for the model demander to pay digital currency to the device; or it can be a two-way off-chain channel for the device to train the model demander The result and the demand direction of the model pay digital currency to the device after receiving the training result sent by the device.

For example, the model demander deposits S in the one-way off-chain payment channel. The off-chain channel contract requires a transaction between the signature SigA of the demander and the signature SigB of the user to close the payment channel. The total amount of this transaction is S, the amount output to the demander's address is A, and the amount of equipment address is B, where A+B=S.

Step 306: The model requester receives the channel address information sent by the blockchain network, and the channel address information is used to identify the state channel.

Step 307: The model request sends channel address information to the device.

It should be noted that in step 305, the model demand sends a transaction to the blockchain network, requesting the establishment of a state channel with the device; when the blockchain network receives the transaction, it deploys the model between the demander and the device The state channel is an off-chain channel located outside the blockchain network and sends the address information of the state channel to the model demander. After receiving the address information of the state channel, the model demander sends the address information of the state channel to the device for the device to identify the established state channel.

Step 308: The model requester sends initial model parameters to the device.

For example, the model request sends the initial model parameters of this joint learning to the device, and the initial parameters may be the value of the weight w of the neural network.

Step 309: The device executes the smart contract and performs single-step training locally, where the local single-step training may be a stochastic gradient descent optimization algorithm, thereby updating the value of the parameter w.

In an example, as shown in Figure 4, the execution environment of the smart contract under the joint learning chain is in a virtual machine (VM). The joint learning App uses the user data of the device and the initial model parameters for a certain joint learning contract to run in the off-chain smart contract VM. The execution environment of the off-chain smart contract can not only support the joint learning program, but also other off-chain smart contract programs.

Step 310: The model requester receives the training result sent by the device.

For example, it may be that the device feeds back the training result to the model consumer through an off-chain channel established with the model consumer.

Step 311: The model demander pays digital currency to the device through a state channel (for example, an off-chain payment channel).

For example, through the state channel, the model demander sends a transaction to the device. In this case, the total asset in the channel is S, where the demander has A and the device has B. The cost of this single training is t. Then in the output of this transaction, the digital currency of the model demander is A-t, and the digital currency of the device is B+t, where A+B=S. In this transaction, the signature of the demander of the model is attached. After the device signs the transaction again, it can send the transaction to the blockchain network, close the channel, and withdraw the digital currency belonging to the device.

In one example, the state channel can run in a TEE environment, and the device can be in a safe environment when receiving digital currency from the model demander and sending model data updates to the model demander. To spend digital currency, you need to use the private key in the digital currency wallet to sign in the SEE environment.

Step 312: The model requester receives the feedback training results of the equipment participating in the joint learning and performs averaging processing to obtain the latest model parameters, returning to step 306, and then delivering to the equipment participating in the joint learning.

Step 313: The device can view the on-chain digital currency and off-chain digital currency belonging to the device at any time.

The device can receive many transactions sent by the demand side of the model. Therefore, the digital currency of the device in the state channel is increasing, and only the latest transactions need to be maintained.

Step 314: When the joint learning end device decides to close the channel, it signs the latest transaction, sends it to the blockchain network, and takes out the digital currency belonging to the device.

It should be noted that the example in FIG. 3 is only to help those skilled in the art understand the embodiments of the present application, and is not intended to limit the embodiments of the present application to the specific scenarios illustrated. Those skilled in the art can obviously make various equivalent modifications or changes based on the example of FIG. 3 given, and such modifications or changes also fall within the scope of the embodiments of the present application.

It should be understood that, in the various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, rather than corresponding to the embodiments of the present application. The implementation process constitutes any limitation.

The above describes the joint learning method according to the embodiment of the application in detail. In the embodiment of the application, the training parameters generated after the device participating in the joint learning participates in the local training can be fed back to the application server node through the state channel. The state channel sending training results can meet the requirements of high TPS in the combined learning and block chain scenario. At the same time, the training results of the device also have a certain degree of privacy. The training results can be sent through the state channel to avoid each node in the blockchain network. Sharing can protect the privacy of device data. It should be understood that the joint learning device of the embodiment of the present application can execute the various methods of the foregoing embodiment of the present application, that is, for the specific working processes of the following various products, refer to the corresponding process in the foregoing method embodiment.

FIG. 5 is a schematic block diagram of a joint learning device 500 provided by an embodiment of the present application. It should be understood that the joint learning device 500 can execute each step performed by the device in the method of FIG. 2 or FIG. 3, and in order to avoid repetition, it will not be described in detail here. The joint learning device 500 includes: a receiving unit 510, a processing unit 520, and a sending unit 530.

Wherein, the receiving unit 510 is configured to receive the initial model parameters sent by the application server node, where the initial model parameters are parameters used by the equipment participating in the joint learning to establish the initial model; the processing unit 520 is configured to train the initial model parameters, Generate training results for updating the initial model parameters; sending unit 530, configured to send the training results to the application server node through a state channel, where the state channel is an off-chain connection between the device and the application server node Channel, the off-chain channel is located outside the blockchain network.

Optionally, as an embodiment, the receiving unit 510 is further configured to: receive channel address information sent by the application server node, where the channel address information is used to identify the state channel.

Optionally, as an embodiment, the receiving unit 510 is further configured to: obtain a joint learning smart contract, where the joint learning smart contract includes training logic instructions for the initial model parameters; the processing unit 520 is specifically configured to: run The joint learning smart contract is used to train the initial model parameters, and generate training results for updating the initial model parameters.

Optionally, as an embodiment, the receiving unit 510 is specifically further configured to: obtain the joint learning smart contract through a blockchain network, where the joint learning smart contract is an off-chain smart contract department, and the chain The execution environment of the next smart contract does not belong to the smart contract of the blockchain network.

Optionally, as an embodiment, the processing unit 520 is specifically configured to: run the joint learning smart contract in a joint learning application to train the initial model parameters.

Optionally, as an embodiment, the processing unit 520 is specifically configured to: in a trusted execution environment TEE, run the joint learning smart contract to train the initial model parameters.

Optionally, as an embodiment, the receiving unit 510 is specifically configured to: receive the digital currency paid by the application server node through the state channel.

Optionally, as an embodiment, the receiving unit 510 is specifically configured to: in a trusted execution environment TEE, receive the digital currency paid by the application server node through the state channel.

Optionally, as an embodiment, the sending unit 530 is further configured to send a first message to the application server node, where the first message is used to instruct the device to participate in joint learning, and the first message is Including the digital currency wallet address of the device.

Optionally, as an embodiment, the processing unit 520 is further configured to: determine the end of the training task of joint learning; the sending unit 530 is further configured to: send the first transaction to the blockchain network, and the first transaction A transaction is used to instruct to close the state channel.

It should be understood that the joint learning device 500 here is embodied in the form of a functional unit. The term "unit" herein can be implemented in the form of software and/or hardware, which is not specifically limited. For example, a "unit" may be a software program, a hardware circuit, or a combination of the two that realize the above-mentioned functions. The hardware circuit may include an application specific integrated circuit (ASIC), an electronic circuit, a processor for executing one or more software or firmware programs (such as a shared processor, a dedicated processor, or a group processor). Etc.) and memory, merged logic circuits and/or other suitable components that support the described functions. Therefore, the units of the examples described in the embodiments of the present application can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software depends on the specific application and design constraint conditions of the technical solution. Professionals and technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of this application.

FIG. 6 is a schematic block diagram of a joint learning device 600 provided by an embodiment of the present application. It should be understood that the joint learning device 600 may be an application server node, which can execute each step performed by the application server node in the method of FIG. 2 or FIG. 3, and in order to avoid repetition, it will not be detailed here. The joint learning device 600 includes a sending unit 610 and a receiving unit 620.

Wherein, the sending unit 610 is configured to send initial model parameters to the device, and the initial model parameters are parameters used by the device to establish an initial model participating in joint learning; the receiving unit 620 is configured to receive through a state channel The training result used to update the initial model parameters, wherein the state channel is an off-chain channel between the device and the application server node, and the off-chain channel is located outside the blockchain network.

It should be understood that the application server node 600 may further include a processing unit, and the processing unit may be used to control the receiving unit 620 and the sending unit 610 to perform related steps.

Optionally, as an embodiment, the receiving unit 620 is further configured to: receive channel address information sent by the blockchain network, where the channel address information is used to identify the status channel; the sending unit 610, It is also used to send the channel address information to the device.

Optionally, as an embodiment, the sending unit 610 is further configured to send a second transaction to the blockchain network, where the second transaction is used to deploy the state channel.

Optionally, as an embodiment, the sending unit 610 is further configured to send a third transaction to the blockchain network, and the third transaction is used to sign the joint learning smart contract on the blockchain network, so The joint learning smart contract is an off-chain smart contract, the execution environment of the off-chain smart contract does not belong to the blockchain network, and the joint learning smart contract includes training logic instructions for the initial model parameters.

Optionally, as an embodiment, the sending unit 610 is further configured to: pay digital currency to the device through the state channel.

Optionally, as an embodiment, the receiving unit 620 is further configured to: receive a first message sent by the device, where the first message is used to instruct the device to participate in joint learning, and the first message includes The digital currency wallet address of the device.

It should be understood that the joint learning device 600 here is embodied in the form of a functional unit. The term "unit" herein can be implemented in the form of software and/or hardware, which is not specifically limited. For example, a "unit" may be a software program, a hardware circuit, or a combination of the two that realize the above-mentioned functions. The hardware circuit may include an application specific integrated circuit (ASIC), an electronic circuit, a processor for executing one or more software or firmware programs (such as a shared processor, a dedicated processor, or a group processor). Etc.) and memory, merged logic circuits and/or other suitable components that support the described functions. Therefore, the units of the examples described in the embodiments of the present application can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software depends on the specific application and design constraint conditions of the technical solution. Professionals and technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of this application.

FIG. 7 shows a schematic block diagram of a joint learning device 700 according to another embodiment of the present application. The joint learning device 700 may be a terminal device. As shown in FIG. 7, the joint learning device 700 includes a processor 720, a memory 760, a communication interface 740, and a bus 750. Among them, the processor 720, the memory 760, and the communication interface 740 communicate through the bus 750, and may also communicate through other means such as wireless transmission. The memory 760 is used to store instructions, and the processor 720 is used to execute instructions stored in the memory 760. The memory 760 stores program codes 711, and the processor 720 can call the program codes 711 stored in the memory 760 to execute the joint learning method shown in FIG. 2 or FIG. 3.

For example, the processor 720 may be used to execute the training model parameters in 220 in FIG. 2 to generate training results; or 307 in FIG. 3 to execute a smart contract to perform local training.

The memory 760 may include a read-only memory and a random access memory, and provides instructions and data to the processor 720. The memory 760 may also include a non-volatile random access memory. The memory 760 may be volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), and electronic Erase programmable read-only memory (electrically EPROM, EEPROM) or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of exemplary but not restrictive description, many forms of RAM are available, such as static random access memory (static RAM, SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), Double data rate synchronous dynamic random access memory (double data date SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous connection dynamic random access memory (synchlink DRAM, SLDRAM) and direct Memory bus random access memory (direct rambus RAM, DR RAM).

In addition to the data bus, the bus 750 may also include a power bus, a control bus, and a status signal bus. However, for clarity of description, various buses are marked as the bus 750 in FIG. 7.

It should be understood that the joint learning device 700 shown in FIG. 7 can implement various processes performed by the device in the method embodiments shown in FIG. 2 and FIG. 3. The operations and/or functions of the various modules in the joint learning device 700 are to implement the corresponding procedures of the devices in the foregoing method embodiments. For details, please refer to the descriptions in the foregoing method embodiments. To avoid repetition, detailed descriptions are appropriately omitted here.

FIG. 8 shows a schematic block diagram of a joint learning device 800 according to another embodiment of the present application. The joint learning device 800 may be an application server node. As shown in FIG. 8, the joint learning device 800 includes a processor 820, a memory 860, a communication interface 840, and a bus 850. Among them, the processor 820, the memory 860, and the communication interface 840 communicate through the bus 850, and may also communicate through other means such as wireless transmission. The memory 860 is used to store instructions, and the processor 820 is used to execute instructions stored in the memory 860. The memory 860 stores the program code 811, and the processor 820 can call the program code 811 stored in the memory 860 to execute the joint learning method shown in FIG. 2 or FIG. 3.

Wherein, the communication interface 840 shown in FIG. 8 may correspond to the receiving unit 620 and the sending unit 610 in the terminal device shown in FIG. 6.

The memory 820 may include a read-only memory and a random access memory, and provides instructions and data to the processor 820. The memory 860 may also include a non-volatile random access memory. The memory 860 may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), and electronic Erase programmable read-only memory (electrically EPROM, EEPROM) or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of exemplary but not restrictive description, many forms of RAM are available, such as static random access memory (static RAM, SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), Double data rate synchronous dynamic random access memory (double data date SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous connection dynamic random access memory (synchlink DRAM, SLDRAM) and direct Memory bus random access memory (direct rambus RAM, DR RAM).

In addition to the data bus, the bus 850 may also include a power bus, a control bus, and a status signal bus. However, for clear description, various buses are marked as bus 850 in FIG. 8.

It should be understood that the joint learning device 800 shown in FIG. 8 can implement various processes performed by the application server node in the method embodiments shown in FIG. 2 and FIG. 3. The operations and/or functions of each module in the joint learning device 800 are used to implement the corresponding procedures of the application server nodes in the foregoing method embodiments. For details, please refer to the descriptions in the foregoing method embodiments. To avoid repetition, detailed descriptions are appropriately omitted here.

This application also provides a computer-readable storage medium that stores instructions in the computer-readable storage medium. When the instructions run on a computer, the computer executes the joint learning methods shown in FIG. 2 and FIG. 3. The various steps.

The present application also provides a computer program product containing instructions. When the computer program product runs on a computer or any one of at least one processor, the computer executes the joint learning methods shown in FIG. 2 and FIG. 3 Various steps.

This application also provides a chip including a processor. The processor is used to read and run the computer program stored in the memory to execute the corresponding operation and/or process performed by the method for waking up the screen provided in this application.

Optionally, the chip further includes a memory, the memory and the processor are connected to the memory through a circuit or a wire, and the processor is used to read and execute the computer program in the memory. Further optionally, the chip further includes a communication interface, and the processor is connected to the communication interface. The communication interface is used to receive data and/or information that needs to be processed, and the processor obtains the data and/or information from the communication interface, and processes the data and/or information. The communication interface can be an input and output interface.

In the above embodiments, the processors involved may include, for example, a central processing unit (CPU), a microprocessor, a microcontroller, or a digital signal processor, and may also include GPU, NPU, and ISP. The processor may also include It may include necessary hardware accelerators or logic processing hardware circuits, such as application-specific integrated circuits (ASICs), or one or more integrated circuits used to control the execution of the technical solutions of this application. In addition, the processor may have a function of operating one or more software programs, and the software programs may be stored in the memory.

The memory can be read-only memory (ROM), other types of static storage devices that can store static information and instructions, random access memory (RAM), or other types that can store information and instructions Dynamic storage devices can also be electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), or other optical disk storage, optical disc storage ( Including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or can be used to carry or store desired program codes in the form of instructions or data structures and can Any other medium accessed by the computer, etc.

In the embodiments of the present application, "at least one" refers to one or more, and "multiple" refers to two or more. "And/or" describes the association relationship of the associated objects, indicating that there can be three types of relationships, for example, A and/or B, which can mean that A exists alone, A and B exist at the same time, and B exists alone. Among them, A and B can be singular or plural. The character "/" generally indicates that the associated objects are in an "or" relationship. "The following at least one item" and similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, and c can represent: a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

A person of ordinary skill in the art may be aware that the units and algorithm steps described in the embodiments disclosed herein can be implemented by a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are executed by hardware or software depends on the specific application and design constraint conditions of the technical solution. Professionals and technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of this application.

Those skilled in the art can clearly understand that, for the convenience and conciseness of description, the specific working process of the above-described system, device, and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

In the several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disk or optical disk and other media that can store program code .

The above are only specific implementations of this application. Any person skilled in the art can easily conceive of changes or substitutions within the technical scope disclosed in this application, which should be covered by the protection scope of this application. The protection scope of this application shall be subject to the protection scope of the claims.

Claims (17)

1. A joint learning method based on blockchain network, which is characterized in that it includes:

   Receiving an initial model parameter sent by an application server node, where the initial model parameter is a parameter used by a device participating in the joint learning to establish an initial model;

   Training the initial model parameters, and generating training results for updating the initial model parameters;

   The training result is sent to the application server node through a state channel, which is an off-chain channel between the device and the application server node, and the off-chain channel is located outside the blockchain network.
2. The method of claim 1, wherein the method further comprises:

   Receiving channel address information sent by the application server node, where the channel address information is used to identify the state channel.
3. The method according to claim 1 or 2, wherein the method further comprises:

   Acquiring a joint learning smart contract, where the joint learning smart contract includes training logic instructions for the initial model parameters;

   The training of the initial model parameters and generating a training result for updating the initial model parameters includes:

   Run the joint learning smart contract to train the initial model parameters, and generate training results for updating the initial model parameters.
4. The method of claim 3, wherein said acquiring a joint learning smart contract comprises:

   The joint learning smart contract is acquired through the blockchain network, wherein the joint learning smart contract is an off-chain smart contract, and the execution environment of the off-chain smart contract does not belong to the blockchain network.
5. The method according to claim 3 or 4, wherein the running the joint learning smart contract to train the initial model parameters comprises:

   Under the trusted execution environment TEE, run the joint learning smart contract to train the initial model parameters.
6. The method according to any one of claims 1 to 5, wherein after the sending the training result to an application server node through a state channel, the method further comprises:

   Receiving the digital currency paid by the application server node through the state channel.
7. The method according to any one of claims 1 to 6, wherein before the receiving the digital currency paid by the application server node through the state channel, the method further comprises:

   Send a first message to the application server node, where the first message is used to instruct the device to participate in joint learning, and the first message includes the digital currency wallet address of the device.
8. The method according to claim 6 or 7, wherein the receiving digital currency paid by the application server node through the state channel comprises:

   Under the trusted execution environment TEE, the digital currency paid by the application server node through the state channel is received.
9. 8. The method according to any one of claims 1 to 8, wherein the method further comprises:

   Confirm the end of the training task of joint learning;

   Send a first transaction to the blockchain network, where the first transaction is used to instruct to close the state channel.
10. A joint learning method based on blockchain network, which is characterized in that it includes:

    Sending initial model parameters to the device, where the initial model parameters are parameters used by the device to establish an initial model participating in joint learning;

    The training result for updating the initial model parameters is received through a state channel, where the state channel is an off-chain channel between the device and the application server node, and the off-chain channel is located outside the blockchain network.
11. The method of claim 10, wherein the method further comprises:

    Receiving channel address information sent by the blockchain network, where the channel address information is used to identify the state channel;

    Sending the channel address information to the device.
12. The method according to claim 11, wherein before said receiving the channel address information sent by the blockchain network, the method further comprises:

    A second transaction is sent to the blockchain network, and the second transaction is used to deploy the state channel.
13. The method according to any one of claims 10 to 12, wherein the method further comprises:

    Send a third transaction to the blockchain network, the third transaction is used to sign a joint learning smart contract on the blockchain network, the joint learning smart contract is an off-chain smart contract, and the off-chain smart The execution environment of the contract does not belong to the blockchain network, and the joint learning smart contract includes training logic instructions for the initial model parameters.
14. The method according to any one of claims 10 to 13, wherein after the receiving the training result for updating the initial model parameters through the state channel, the method further comprises:

    The digital currency is paid to the device through the state channel.
15. The method according to claim 14, characterized in that, before the payment of digital currency to the device through the state channel, the method further comprises:

    A first message sent by the device is received, the first message is used to instruct the device to participate in joint learning, and the first message includes the digital currency wallet address of the device.
16. A joint learning device, characterized by comprising a memory and a processor, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory to execute claims 1 to 9. The method of any one of 9.
17. A joint learning device, characterized by comprising a memory and a processor, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory to execute claims 10 to The method of any one of 15.

Images