WO2022193173A1

WO2022193173A1 - Blockchain-based financial data information federated transfer learning system and method

Info

Publication number: WO2022193173A1
Application number: PCT/CN2021/081308
Authority: WO
Inventors: 王化; 赵建
Original assignee: 深圳技术大学
Priority date: 2021-03-17
Filing date: 2021-03-17
Publication date: 2022-09-22

Abstract

The present invention provides a blockchain-based financial data information federated transfer learning system and method. The system comprises: a task publishing module, which is used to publish on a blockchain a smart contract of a model training task, wherein the smart contract of the model training task comprises a model training graph; a task management module, which is connected to the task publishing module and used to record and manage the smart contract of the model training task and generate an address list; and a running node, which is connected to the task management module and used to read the address list, download the model training graph from the smart contract of the model training task, train the model training graph by using local data, and upload gradient data of training to a specified storage location of the smart contract of the model training task, so as to perform data sharing. According to the present invention, data privacy leakage of financial data information during machine learning is avoided, and training accuracy is improved.

Description

A blockchain-based financial data information federated transfer learning system and method

technical field

The invention relates to the technical field of blockchain, in particular to a blockchain-based financial data information federated transfer learning system and method.

Background technique

Machine learning techniques have achieved significant success in many fields, but machine learning methods only work well under the assumption that the training and test data are in the same feature space or have the same distribution. When the distribution changes, most statistical models require rebuilding the model using newly collected training data. In many real-world applications, re-collecting the required training data and rebuilding the model is prohibitively expensive. Machine learning generates predictive models from data, so high-quality data is often needed to help regulate statistical models.

At present, financial institutions have various data sources, non-standard data formats, and unstable data update cycles. Therefore, when applying machine learning to actual financial scenarios, there are data islands resulting in less data in application scenarios, and in practical applications, it is often necessary to use multiple User privacy and data security issues caused by datasets.

technical problem

Aiming at the deficiencies of the prior art, the present invention provides a blockchain-based financial data information federated transfer learning system and method, aiming to solve the problem that when traditional machine learning is applied to actual financial scenarios, there is less application scenario data and users Privacy and data security concerns.

technical solutions

In order to solve the above-mentioned technical problems, the present invention adopts the following technical solutions:

In the first aspect, a blockchain-based financial data information federated transfer learning system is provided, including:

A task publishing module, used to publish the smart contract of the model training task on the blockchain, where the smart contract of the model training task includes the model training graph;

a task management module, which is connected with the task release module, and is used for recording and managing the smart contracts of the model training tasks and generating an address list;

A running node, which is connected to the task management module, is used to read the address list, download the model training graph from the smart contract of the model training task, and use local data to train the model training graph. The gradient data is uploaded to the designated storage location of the smart contract of the model training task for data sharing.

In a second aspect, a blockchain-based financial data information federated transfer learning method is provided, which is applied to a blockchain-based financial data information federated transfer learning system. The system includes a task publishing module, a task management module, and an operating node. , the method includes:

The task publishing module publishes the smart contract of the model training task on the blockchain, and the smart contract of the model training task includes the model training graph;

The task management module records and manages the smart contract of the model training task and generates an address list;

The operating node reads the address list, downloads the model training graph from the smart contract of the model training task, uses local data to train the model training graph, and uploads the trained gradient data to the model training task's Designated storage location for smart contracts for data sharing.

beneficial effect

The beneficial effects of the present invention are:

The task management module of the present invention is respectively connected with the task release module and the running node, and the smart contract of the model training task is released on the blockchain through the task release module. The task management module records and manages the released smart contract and generates its address list, and runs the The node reads the address list, downloads the model training graph from the corresponding smart contract, uses local data to train its own model, and uploads the trained gradient data to the designated storage location of the smart contract for data sharing, avoiding the risk of uploading original data. At the same time, sharing gradient data can solve the problem of less data in application scenarios and improve training accuracy.

Description of drawings

The specific structure of the present invention will be described in detail below in conjunction with the accompanying drawings

1 is a block connection diagram of a blockchain-based financial data information federated transfer learning system provided by an embodiment of the present invention;

FIG. 2 is a flowchart of a blockchain-based method for federated transfer learning of financial data information provided by an embodiment of the present invention.

Embodiments of the present invention

In order to describe the technical content, structural features, achieved objects and effects of the present invention in detail, the following detailed description is given in conjunction with the embodiments and the accompanying drawings.

The following describes in detail the embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary, and are intended to explain the present invention and should not be construed as limiting the present invention.

Please refer to FIG. 1. FIG. 1 is a module connection diagram of a blockchain-based financial data information federated transfer learning system provided by an embodiment of the present invention.

In a first aspect, the present invention provides a blockchain-based financial data information federated transfer learning system, including:

The task publishing module 10 is used to publish the smart contract of the model training task on the blockchain, and the smart contract of the model training task includes the model training graph;

Specifically, the blockchain is a consortium chain, which is between the private chain and the public chain. It is a blockchain that requires registration and permission, and is limited to members of the consortium who have permission to participate in the reading and writing of the ledger. The roles and functional division of nodes need to be preset, and the consensus, operation and maintenance and access in the network are controlled by the preset nodes. Generally speaking, the alliance chain is suitable for cross-institutional transactions, settlement, collaborative office and certificate deposit. After the transaction is reached, each participant on the blockchain first verifies the transaction. Once all participants reach a consensus, the transaction information is stamped with a timestamp indicating the order in which the transaction occurred. The timestamp function guarantees the traceability of transactions. The application of blockchain technology solves the pain point of high credit risk in traditional transactions and improves the security of transactions. At the same time, each participant in the blockchain has a complete set of books, which has unique advantages in reconciliation, which reduces the It reduces the cost of reconciliation and improves the efficiency of settlement. The product has the characteristics of decentralization, de-trust, time stamp, etc., as the underlying technology of the platform architecture, making all transaction information open, transparent and tamper-proof, greatly reducing the occurrence of operational risks and credit risks, enabling transactions safer.

Further, the blockchain includes a data layer, a network layer, a consensus layer, an incentive layer, a contract layer and an application layer.

The task management module 20, which is connected with the task release module, is used for recording and managing the smart contracts of the model training tasks and generating an address list.

Specifically, the task management module 20 is a factory-mode smart contract, which is encapsulated in the contract layer of the blockchain.

Running node 30, which is connected to the task management module 20, is used to read the address list, download the model training graph from the smart contract of the model training task, and use local data to train the model training graph, The trained gradient data is uploaded to the designated storage location of the smart contract of the model training task for data sharing.

The beneficial effects of the present invention are:

The task management module 20 of the present invention is respectively connected with the task release module 10 and the operation node 30, and the smart contract of the model training task is released on the blockchain through the task release module 10. The task management module 20 records and manages the released smart contract and generates For its address list, the operating node 30 reads the address list, downloads the model training graph from the corresponding smart contract, uses local data to train its own model, and uploads the trained gradient data to the designated storage location of the smart contract for data sharing, The privacy problem caused by uploading the original data is avoided, and the sharing of gradient data can solve the problem of less data in application scenarios and improve the training accuracy.

Wherein, the blockchain-based financial data information federated transfer learning system further includes a file storage module for storing data files, which is connected to the task management module 20, and the hash values and acquisition paths of the data files are stored in the data file. in the smart contract that describes the model training task.

Wherein, the task management module 20 has an API interface, so that the running node 30 can read the address list.

Wherein, the smart contract of the model training task further includes a training data set storage path, a test data set storage path and accuracy requirements.

Please refer to FIG. 2. FIG. 2 is a flowchart of a method for federated transfer learning of financial data information based on blockchain provided by an embodiment of the present invention.

In the second aspect, the present invention also provides a blockchain-based financial data information federated transfer learning method, which is applied to a blockchain-based financial data information federated transfer learning system. The system includes a task issuing module 10, a task management Module 20 and operating node 30, the method includes:

Step S101, the task publishing module 10 publishes the smart contract of the model training task on the blockchain, and the smart contract of the model training task includes the model training graph;

Step S102, the task management module 20 records and manages the smart contract of the model training task and generates an address list;

Step S103, the operating node 30 reads the address list, downloads the model training graph from the smart contract of the model training task, uses local data to train the model training graph, and uploads the trained gradient data to the model training graph. Designated storage location for smart contracts for model training tasks for data sharing.

Further, in a possible implementation manner, the running node 30 includes a first running node and a second running node, then the model training graph is trained by using local data, and the trained gradient data is uploaded to The designated storage location of the smart contract of the model training task includes:

The first running node and the second running node respectively initialize their respective models;

The second running node calculates part of the estimated value and part of the loss based on its own characteristics, and encrypts it and sends it to the first running node;

The first running node calculates part of the estimated value based on its own features, and combines the estimated value of the second running node to calculate the final loss function and gradient data, and then sends the gradient data and loss function required by the second running node back to the first running node. Two running nodes;

After completing the gradient data calculation, the first running node and the second running node encrypt and mask the gradient data respectively, and send them to the off-chain storage server;

Off-chain storage servers decrypt and aggregate gradient data;

The first running node and the second running node download the aggregated gradient data from the off-chain storage server, remove their own masks and update their own models;

The above steps are repeated until convergence, and the models of the first running node and the second running node are obtained respectively.

In another possible implementation manner, the operating nodes 30 include multiple, and the model training graph is trained by using local data, and the trained gradient data is uploaded to the smart contract of the model training task. Designated storage locations include:

The off-chain storage server initializes and encrypts the item profile;

The off-chain storage server sends the encrypted item profile to each running node;

Each running node decrypts the item profile and calculates the loss based on the local data, updates its own user profile, and encrypts and transmits the gradient data of the item profile to the off-chain storage server;

Off-chain storage server summary item profile gradient data and update item profiles;

Each running node uses a matrix to decompose its user profile and item profile and recommend it.

The above descriptions are only the embodiments of the present invention, and are not intended to limit the scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied to other related technologies Fields are similarly included in the scope of patent protection of the present invention.

Claims

A blockchain-based financial data information federated transfer learning system, characterized in that it includes:

A task publishing module, used to publish the smart contract of the model training task on the blockchain, where the smart contract of the model training task includes the model training graph;

a task management module, which is connected with the task release module, and is used for recording and managing the smart contracts of the model training tasks and generating an address list;

A running node, which is connected to the task management module, is used to read the address list, download the model training graph from the smart contract of the model training task, and use local data to train the model training graph. The gradient data is uploaded to the designated storage location of the smart contract of the model training task for data sharing.
The blockchain-based financial data information federated transfer learning system according to claim 1, further comprising a file storage module for storing data files, the hash values and acquisition paths of the data files are stored in the in the smart contract that describes the model training task.
The blockchain-based financial data information federated transfer learning system according to claim 1, wherein the task management module has an API interface, so that the running node can read the address list.
The blockchain-based financial data information federated transfer learning system according to claim 1, wherein the smart contract for the model training task further includes a training data set storage path, a test data set storage path, and accuracy requirements.
A blockchain-based financial data information federated transfer learning method is applied to a blockchain-based financial data information federated transfer learning system. The system includes a task issuing module, a task management module and an operation node, and is characterized in that it includes: :

The task publishing module publishes the smart contract of the model training task on the blockchain, and the smart contract of the model training task includes the model training graph;

The task management module records and manages the smart contract of the model training task and generates an address list;

The operating node reads the address list, downloads the model training graph from the smart contract of the model training task, uses local data to train the model training graph, and uploads the trained gradient data to the model training task's Designated storage location for smart contracts for data sharing.
The method for federated transfer learning of financial data information based on blockchain according to claim 5, wherein the operation node includes a first operation node and a second operation node, and the model is trained by using local data The specified storage location of the smart contract for uploading the trained gradient data to the model training task includes:

The first running node and the second running node respectively initialize their respective models;

The second running node calculates part of the estimated value and part of the loss based on its own characteristics, and encrypts it and sends it to the first running node;

The first running node calculates part of the estimated value based on its own features, and combines the estimated value of the second running node to calculate the final loss function and gradient data, and then sends the gradient data and loss function required by the second running node back to the first running node. Two running nodes;

After completing the gradient data calculation, the first running node and the second running node encrypt and mask the gradient data respectively, and send them to the off-chain storage server;

Off-chain storage servers decrypt and aggregate gradient data;

The first running node and the second running node download the aggregated gradient data from the off-chain storage server, remove their own masks and update their own models;

The above steps are repeated until convergence, and the models of the first running node and the second running node are obtained respectively.
The method for federated transfer learning of financial data information based on blockchain according to claim 5, characterized in that, if the operating nodes include a plurality of nodes, the model training graph is trained by using local data, and the trained The designated storage location of the gradient data uploaded to the smart contract of the model training task includes:

The off-chain storage server initializes and encrypts the item profile;

The off-chain storage server sends the encrypted item profile to each running node;

Each running node decrypts the item profile and calculates the loss based on the local data, updates its own user profile, and encrypts and transmits the gradient data of the item profile to the off-chain storage server;

Off-chain storage server summary item profile gradient data and update item profiles;

Each running node uses a matrix to decompose its user profile and item profile and recommend it.