WO2021051610A1

WO2021051610A1 - Data training method, apparatus and system

Info

Publication number: WO2021051610A1
Application number: PCT/CN2019/118407
Authority: WO
Inventors: 何安珣; 王健宗
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-09-20
Filing date: 2019-11-14
Publication date: 2021-03-25
Also published as: CN110795477A

Abstract

A data training method, apparatus and system. The method comprises: sending an initial training model to a plurality of clients, wherein the plurality of clients all independently communicate with a server (S202); receiving a plurality of sets of first model parameters sent by the plurality of clients, wherein the first model parameters are obtained by training, according to first medical data of a local database, the initial training model by the clients (S204); performing weighted averaging on the plurality of sets of first model parameters to obtain second model parameters (S206); and sending the second model parameters to the plurality of clients, wherein the second model parameters are used for respectively constructing the same second training model on the plurality of clients (S208). The method solves the technical problem, in the related technology, of an algorithm model for processing medical data being relatively complicated and being unable to process large-scale medical data that has a relatively high security and is inconvenient to circulate.

Description

Data training method, device and system

Technical field

This application relates to the computer field, specifically, to a data training method, device, and system.

Background technique

Among related technologies, medical imaging assisted recognition is a relatively mature application of artificial intelligence image recognition technology in the medical field. Many institutions at home and abroad have established standardized regional medical imaging data center cloud platform services based on this technology. Integrated auxiliary diagnosis, data centralized storage management, regional major disease analysis, and regional population health portraits and other functions. At present, the widely used regional cloud platform, as the name suggests, is just a regional health information sharing system. In essence, this is a private cloud with clinics, hospitals as units, or several hospitals as units.

Due to the privacy of medical and health data, the scale effect cannot be produced. The problem of data islands still exists. The training of medical and health models is still restricted by limited data. Some medical institutions need to spend a higher cost to purchase the training that has been trained by third-party institutions. Model, the industry as a whole has a low degree of information sharing and low economic efficiency. It is difficult for the healthcare ecosystem to develop further on this basis.

Traditional data structure and machine learning integrate data and then train based on the integrated data set. This type of method requires data to be transmitted between the distributed data set and the central server. Since the central server integrates massive data, the computing power required for training the model is high, the calculation cost is correspondingly high, and the response time is relatively long. At the same time, for some data that is relatively safe and inconvenient to flow, such as medical health data, this method cannot be used for model training on a large scale.

In view of the above-mentioned problems existing in related technologies, no effective solutions have been found so far.

Application content

The embodiments of the present application provide a data training method, device, and system, so as to at least solve the technical problems that the algorithm model for processing medical data in related technologies is relatively complicated and cannot handle large-scale medical data that is inconvenient to flow.

According to an embodiment of the present application, a data training method is provided, including: sending an initial training model to multiple clients, wherein each of the multiple clients communicates with a server separately; and receiving the multiple clients Sent multiple sets of first model parameters, where the first model parameters are obtained by the client training the initial training model according to the first medical data in the local database; and the multiple sets of first model parameters Perform a weighted average to obtain the second model parameters; send the second model parameters to the multiple clients, where the second model parameters are used to construct the same second training on the multiple clients respectively model.

According to another embodiment of the present application, a data training method is provided, including: receiving an initial training model sent by a server; training the initial training model according to first medical data in a local database to obtain the first training model Send the first model parameters of the first training model to the server, where the server is used to perform a weighted average on multiple sets of first model parameters of multiple clients to obtain a second model of the second training model Parameters, and feed back the second model parameters to the multiple clients; construct a second training model according to the second model parameters, and use the second training model to train the second medical data of the local database .

According to an embodiment of the present application, there is provided a data training device, including: a first sending module, configured to send an initial training model to multiple clients, wherein each of the multiple clients communicates with the server separately; The receiving module is configured to receive multiple sets of first model parameters sent by the multiple clients, where the first model parameters are the clients training the initial training model according to the first medical data in the local database A calculation module for weighted average of the multiple sets of first model parameters to obtain a second model parameter; a second sending module for sending the second model parameters to the multiple clients, Wherein, the second model parameters are used to construct the same second training model on the multiple clients respectively.

According to another embodiment of the present application, there is provided a data training device, which includes: a receiving module for receiving an initial training model sent by a server; a first training module for performing data training based on first medical data in a local database; The initial training model is trained to obtain a first training model; a sending module is used to send the first model parameters of the first training model to the server, where the server is used to provide multiple sets of The first model parameters are weighted and averaged to obtain the second model parameters of the second training model, and the second model parameters are fed back to the multiple clients; the second training module is used to obtain the second model parameters according to the second model parameters. Constructing a second training model, and using the second training model to train the second medical data of the local database.

According to another embodiment of the present application, there is also provided a data training system, including: a server and a plurality of clients, wherein the server includes: a first sending module, configured to send initial training to the plurality of clients Model; receiving module for receiving multiple sets of first model parameters sent by said multiple clients; calculation module for weighted average of said multiple sets of first model parameters to obtain second model parameters; second sending The module is used to send the second model parameters to the multiple clients; the multiple clients all communicate with the server separately, including: a receiving module, used to receive the initial training model; first training Module, used to train the initial training model according to the first medical data in the local database to obtain the first training model; sending module, used to send the first model parameters of the first training model to the Server; a second training module for constructing a second training model according to the second model parameters, and using the second training model to train the second medical data of the local database.

According to another embodiment of the present application, there is also provided a storage medium in which a computer program is stored, wherein the computer program is configured to execute the steps in any one of the above-mentioned device embodiments when running.

According to another embodiment of the present application, there is also provided a computer device, including a memory and a processor, the memory is stored with a computer program, and the processor is configured to run the computer program to execute any of the above Steps in the method embodiment.

Through this application, the server sends the initial training model to multiple clients, so that the client trains local medical data locally to obtain the updated first training model, and only needs to send the first model parameters of the first training model to the server , There is no need to integrate local medical data to the server, which ensures the security of local data, reduces the workload and storage resources of the server; the server weights the model parameters obtained and returns them to multiple clients for training, so Multiple clients share the same training model, which solves the technical problems of complicated medical data processing algorithm models in related technologies, and large-scale medical data that cannot be processed and inconvenient to flow.

Description of the drawings

The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application, and do not constitute an improper limitation of the application. In the attached picture:

FIG. 1 is a block diagram of the hardware structure of a data training method applied to a computer terminal according to an embodiment of the present application;

Fig. 2 is a flowchart of a data training method provided according to the present application;

Fig. 3 is a structural block diagram of another data training method according to an embodiment of the present application;

Figure 4 is a flow chart based on federated learning medical data provided according to an embodiment of the present application;

Fig. 5 is a structural block diagram of a data training device according to an embodiment of the present application;

Fig. 6 is a structural block diagram of another data training device according to an embodiment of the present application;

Fig. 7 is a structural block diagram of a data training system according to an embodiment of the present application.

detailed description

Hereinafter, the present application will be described in detail with reference to the drawings and in conjunction with the embodiments. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other if there is no conflict.

It should be noted that the terms "first" and "second" in the specification and claims of the application and the above-mentioned drawings are used to distinguish similar objects, and not necessarily used to describe a specific sequence or sequence.

Example 1

The method embodiment provided in Embodiment 1 of the present application may be executed in a mobile terminal, a server, a computer terminal, or a similar computing device. Taking running on a computer terminal as an example, FIG. 1 is a hardware structural block diagram of a data training method applied to a computer terminal in an embodiment of the present application. As shown in FIG. 1, the computer terminal may include one or more (only one is shown in FIG. 1) processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) And the memory 104 for storing data. Optionally, the above-mentioned computer terminal may also include a transmission device 106 and an input/output device 108 for communication functions. A person of ordinary skill in the art can understand that the structure shown in FIG. 1 is only for illustration, and does not limit the structure of the foregoing computer terminal. For example, the computer terminal may also include more or fewer components than shown in FIG. 1, or have a different configuration from that shown in FIG.

The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer programs corresponding to the data training method in the embodiment of the present application. The processor 102 executes the computer programs stored in the memory 104 by running the computer programs stored in the memory 104. This kind of functional application and data processing realizes the above-mentioned method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely provided with respect to the processor 102, and these remote memories may be connected to a computer terminal through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC for short), which can be connected to other network devices through a base station to communicate with the Internet. In an example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet in a wireless manner.

In this embodiment, a data training method is provided, and FIG. 2 is a flowchart of a data training method provided in this application. As shown in Figure 2, the process includes the following steps:

Step S202, sending the initial training model to multiple clients, where each of the multiple clients communicates with the server individually;

Step S204, receiving multiple sets of first model parameters sent by multiple clients, where the first model parameters are obtained by the client training the initial training model according to the first medical data in the local database;

Among them, the first medical data in the client's local database may include the patient's attribute information, the patient's diagnosis and treatment information, etc., such as: the patient's age, gender and other personal identification information, past medical history, prescription effects and other diagnosis and treatment records.

Step S206: Perform weighted average on multiple sets of first model parameters to obtain second model parameters;

In this embodiment, the server performs a weighted calculation on the model parameters sent by multiple clients according to the federated average algorithm, where the weight is determined according to the training effect of each client.

Step S208: Send the second model parameters to multiple clients, where the second model parameters are used to construct the same second training model on the multiple clients respectively.

Through this application, the server sends the initial training model to multiple clients, so that the client trains local medical data locally to obtain the updated first training model, and only needs to return the first model parameters of the first training model to the server , There is no need to integrate local medical data to the server, which guarantees the security of local data and reduces the workload and storage resources of the server; the server weights the model parameters obtained and returns them to multiple clients for training, making more Each client shares the same training model, which solves the technical problems that the algorithm model for processing medical data in related technologies is relatively complex and cannot handle large-scale medical data with high security and inconvenient flow.

Optionally, before the weighted average of multiple sets of first model parameters is performed to obtain the second model parameters, the method further includes: decrypting the first model parameters according to a preset private key, where the private key corresponds to multiple clients The public key forms a set of key pairs, and the public key is used to encrypt the first model parameters.

In this embodiment, in order to ensure the security of the information between each client and the server, the parameters transmitted between the two are encrypted and trained. The encryption method is as follows: (1) The server sends the public key to each client for Encrypt the parameters that need to be interacted (that is, the above-mentioned first model parameters), where the server is also provided with a private key corresponding to the public key, that is, the public key and the private key are a set of key pairs; (2) the server receives After the encrypted parameters, the parameters are decrypted according to the private key.

In an optional example, the weighted average of multiple sets of first model parameters to obtain the second model parameters includes: selecting among M sets of first model parameters

The first model parameters of the second time, where N sets of first model parameters are selected each time, and the N sets of first model parameters selected each time are weighted and averaged to obtain a first-level model parameter, where N is an integer less than M;

The first-level model parameters are weighted and averaged to obtain the second model parameters.

In an optional embodiment, the server uses a federated average algorithm to averagely weight the first model parameters after each client's local database training. Take 3 clients as an example, (namely client 1, client 2, client 3), according to the selection ratio of the clients in each round of calculation, assuming that 2 of the 3 clients are selected in each round, there are a total of

A selection method (ie a group of client 1 and client 2, a group of client 1 and client 3, a group of client 2 and client 3); the first model parameter sent by client 1 and client 2 Perform weighting to obtain parameter 1 (that is, the above-mentioned first-level model parameters), weight the first model parameters sent by client 1 and client 3 to obtain parameter 2, and weight the first model parameters sent by client 2 and client 3 to obtain Parameter 3; Finally, parameter 1, parameter 2, and parameter 3 are averagely weighted to obtain the second model parameter (equivalent to the second-level model parameter).

In this embodiment, another data training method is provided, which is applied to the client. FIG. 3 is a structural block diagram of another data training method according to an embodiment of the present application. As shown in Figure 3, the process includes the following steps:

Step S302, receiving the initial training model sent by the server;

Step S304, training the initial training model according to the first medical data in the local database to obtain the first training model;

Step S306: Send the first model parameters of the first training model to the server, where the server is used to perform a weighted average on multiple sets of first model parameters of multiple clients to obtain the second model parameters of the second training model, and Feedback of the second model parameters to multiple clients;

In step S308, a second training model is constructed according to the second model parameters, and the second training model is used to train the second medical data of the local database.

Through the embodiments of this application, multiple clients train based on the initial training model provided by the server and their respective local medical data to obtain the updated first training model, and only need to send the first model parameters of the first training model to the server , There is no need to integrate and aggregate local medical data to the server, which ensures the security of local data; multiple clients are averagely weighted according to the second model parameters returned by the server, so that multiple clients continue to train local medical data, thus achieving The purpose of multiple clients sharing the same training model solves the technical problems that the algorithm model for processing medical data in related technologies is relatively complicated and cannot handle large-scale medical data with high security and inconvenient flow.

In an optional embodiment, training the initial training model according to the first medical data of the local database to obtain the first training model includes: using the first medical data of the local database to perform batch gradient calculation on the initial training model, Obtain multiple gradient values; calculate the average gradient of the multiple gradient values; use the average gradient to update the initial weight value of the initial training model to obtain the first model parameter.

In this embodiment, since the local medical data of each client is constantly updated, in order to make the model trained by federated learning adaptive to each client, and the loss of local medical data is minimized, the initial training model is Perform batch gradient calculation (SGD algorithm, full name Stochastic Gradient Descent, stochastic gradient descent), according to the proportion of client devices that perform calculations in each round, calculate the loss gradient of multiple client local medical data, which is equivalent to multiple parallel data The channel calculates the average value of the gradient of the client in a randomly selected subset, and updates the weight of the initial training model according to the average value of the gradient; then each client uses the local medical data according to the average value of the gradient in the current model (that is, the initial training model mentioned above) ) Performs a step gradient descent, and the server performs a weighted average on the obtained model (that is, the aforementioned second model parameter). By averaging the gradients of multiple models of multiple clients, the loss of local medical data is minimized, which is better than the model obtained by training on two clients separately.

The following further describes the embodiments of the application in combination with a specific embodiment:

Figure 4 is a flow chart based on federated learning medical data provided according to an embodiment of the application. As shown in Figure 4, it is assumed that the distributed data center has 3 clients, namely data set No. 1 and data set No. 2 in Fig. 4 , Data Set No. 3, and the central server. The central server provides an initial model for the distributed data center (that is, the above-mentioned initial training model), and data set 1 performs model training on the initial model based on its own data recorded in its local database (that is, the first medical data in the above-mentioned local database) , Get model update 1 and the first model parameter 1 of the model; at the same time, data set 2 trains the initial model according to its own data recorded in its local data to obtain model update 2 and first model parameter 2; the same goes for , For data set 3, get model update 3 and first model parameter 3.

Sending the three model parameters to the central server eliminates the need to integrate the data sets on each distributed data center side into the central server, thereby reducing the workload of the central server and increasing the processing speed of the central server. The central server performs weighting calculation on the three received parameters according to the federated average algorithm to obtain the second model parameter, and the central server returns the second model parameter to the data set 1, 2, 3.

Through the above steps, the distributed data center does not have to send the locally stored medical data to the central server, but sends the model parameters of the training models of each data center to the central server after encryption processing, thereby ensuring the data security on the data set side and Personal privacy of the user; the central server does not need to integrate the data sets of each terminal side, but averagely weights the model parameters from each data set to obtain the second model parameter, which achieves the purpose of uniformly updating the model of each data set. The calculation cost and calculation time of the central server are reduced, thereby improving the processing efficiency of the server. It solves the technical problems such as the inability to train large-scale user medical data with high security and inconvenient flow in related technologies, resulting in a low degree of medical data sharing.

Example 2

In this embodiment, a data training device is also provided, which is used to implement the above-mentioned embodiments and preferred implementations, and those that have been explained will not be repeated. As used below, the term "module" can implement a combination of software and/or hardware with predetermined functions. Although the devices described in the following embodiments are preferably implemented by software, implementation by hardware or a combination of software and hardware is also possible and conceived.

FIG. 5 is a structural block diagram of a data training device according to an embodiment of the present application. As shown in FIG. 5, the device includes: a first sending module 502, configured to send initial training models to multiple clients, where multiple Each client communicates with the server separately; the receiving module 504, connected to the first sending module 502, is used to receive multiple sets of first model parameters sent by multiple clients, where the first model parameter is the client according to the local database The calculation module 506, connected to the receiving module 504, is used to perform a weighted average of multiple sets of first model parameters to obtain the second model parameters; the second sending module 508, It is connected to the aforementioned calculation module 506 and is used to send the second model parameters to multiple clients, where the second model parameters are used to construct the same second training model on the multiple clients respectively.

Optionally, the above-mentioned device further includes: a decryption module, which is used to decrypt the first model parameters according to a preset private key before performing a weighted average of multiple sets of first model parameters to obtain the second model parameters, wherein the private The key and the public keys corresponding to multiple clients form a set of key pairs, and the public key is used to encrypt the first model parameter.

Optionally, the calculation module includes: a selection unit for selecting among M sets of first model parameters

The second first model parameter, where N sets of first model parameters are selected each time, and the N sets of first model parameters selected each time are weighted and averaged to obtain a first-level model parameter, where N is an integer less than M; calculation Unit for

Fig. 6 is a structural block diagram of another data training device according to an embodiment of the present application. As shown in Fig. 6, the device includes: a receiving module 602 for receiving an initial training model sent by a server; and a first training module 604, Connected to the receiving module 602, used to train the initial training model according to the first medical data in the local database, to obtain the first training model; the sending module 606, connected to the first training module 604, used to train the first training model The first model parameters are sent to the server, where the server is used to perform a weighted average of multiple sets of first model parameters of multiple clients to obtain the second model parameters of the second training model, and feed back the second model parameters to the multiple clients The second training module 608, connected to the above-mentioned sending module 606, is used to construct a second training model according to the second model parameters, and use the second training model to train the second medical data of the local database.

Optionally, the first training module includes: a first calculation unit for performing batch gradient calculations on the initial training model using the first medical data in a local database to obtain multiple gradient values; and a second calculation unit for calculating multiple gradients. The average gradient of each gradient value; the third calculation unit is used to update the initial weight value of the initial training model using the average gradient to obtain the first model parameter.

According to another embodiment of the present application, a data training system is also provided. FIG. 7 is a structural block diagram of a data training system according to an embodiment of the present application, including: a server and multiple clients, where the server Including: a first sending module, used to send initial training models to multiple clients; a receiving module, used to receive multiple sets of first model parameters sent by multiple clients; a calculation module, used to compare multiple sets of first model parameters Perform a weighted average to obtain the second model parameters; the second sending module is used to send the second model parameters to multiple clients; multiple clients, all communicating with the server separately, including: a receiving module, used to receive the initial training model The first training module is used to train the initial training model according to the first medical data in the local database to obtain the first training model; the sending module is used to send the first model parameters of the first training model to the server; second The training module is used to construct a second training model according to the second model parameters, and use the second training model to train the second medical data of the local database.

It should be noted that each of the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following manner, but not limited to this: the above modules are all located in the same processor; or, the above modules can be combined in any combination. The forms are located in different processors.

Example 3

The embodiment of the present application also provides a storage medium in which a computer program is stored, wherein the computer program is configured to execute the steps in any one of the foregoing method embodiments when running.

Optionally, in this embodiment, the aforementioned storage medium may be configured to store a computer program for executing the following steps:

S1. Send an initial training model to multiple clients, where each of the multiple clients communicates with the server individually;

S2, receiving multiple sets of first model parameters sent by the multiple clients, where the first model parameters are obtained by the client training the initial training model according to the first medical data in a local database;

S3: Perform a weighted average on the multiple sets of first model parameters to obtain second model parameters;

S4. Send the second model parameters to the multiple clients, where the second model parameters are used to construct the same second training model on the multiple clients respectively.

S1, receiving the initial training model sent by the server;

S2, training the initial training model according to the first medical data in the local database to obtain the first training model;

S3. Send the first model parameters of the first training model to the server, where the server is used to perform a weighted average on multiple sets of first model parameters of multiple clients to obtain the second model parameters of the second training model. Model parameters, and feedback the second model parameters to the multiple clients;

S4: Construct a second training model according to the second model parameters, and use the second training model to train second medical data of the local database.

Optionally, in this embodiment, the foregoing storage medium may include, but is not limited to: U disk, Read-Only Memory (Read-Only Memory, ROM for short), Random Access Memory (Random Access Memory, RAM for short), Various media that can store computer programs, such as mobile hard disks, magnetic disks, or optical disks.

An embodiment of the present application also provides an electronic device, including a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the foregoing method embodiments.

Optionally, the aforementioned electronic device may further include a transmission device and an input-output device, wherein the transmission device is connected to the aforementioned processor, and the input-output device is connected to the aforementioned processor.

Optionally, in this embodiment, the foregoing processor may be configured to execute the following steps through a computer program:

S1, receiving the initial training model sent by the server;

Optionally, for specific examples in this embodiment, reference may be made to the examples described in the above-mentioned embodiments and optional implementation manners, and details are not described herein again in this embodiment.

Obviously, those skilled in the art should understand that the above-mentioned modules or steps of this application can be implemented by a general computing device, and they can be concentrated on a single computing device or distributed in a network composed of multiple computing devices. Above, alternatively, they can be implemented with program codes executable by a computing device, so that they can be stored in a storage device for execution by the computing device, and in some cases, can be executed in a different order than here. Perform the steps shown or described, or fabricate them into individual integrated circuit modules respectively, or fabricate multiple modules or steps of them into a single integrated circuit module for implementation. In this way, this application is not limited to any specific combination of hardware and software.

The above descriptions are only preferred embodiments of the application, and are not intended to limit the application. For those skilled in the art, the application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the principles of this application shall be included in the protection scope of this application.

Claims

A data training method, which includes:

Sending the initial training model to multiple clients, where each of the multiple clients communicates with the server separately;

Receiving multiple sets of first model parameters sent by the multiple clients, where the first model parameters are obtained by the client training the initial training model according to the first medical data in the local database;

Performing a weighted average on the multiple sets of first model parameters to obtain a second model parameter;

The second model parameter is sent to the multiple clients, where the second model parameter is used to construct the same second training model on the multiple clients respectively.
The method according to claim 1, before the multiple sets of first model parameters are weighted and averaged to obtain the second model parameters, the method further comprises:

The first model parameter is decrypted according to the preset private key, wherein the private key and the public keys corresponding to the multiple clients form a set of key pairs, and the public key is used to decrypt the first model parameter. Model parameters are encrypted.
The method according to claim 1, wherein the weighted average of the multiple sets of first model parameters to obtain the second model parameters includes:

Choose from M sets of first model parameters
The first model parameter of the second time, where N sets of first model parameters are selected each time, and the N sets of the first model parameters selected each time are weighted and averaged to obtain a first-level model parameter, where N is an integer less than M ；

Correct
Performing a weighted average of the first-level model parameters to obtain the second model parameter.
A data training method, which includes:

Receive the initial training model sent by the server;

Training the initial training model according to the first medical data in the local database to obtain the first training model;

The first model parameters of the first training model are sent to the server, where the server is used to perform a weighted average of multiple sets of first model parameters of multiple clients to obtain the second model parameters of the second training model , And feed back the second model parameters to the multiple clients;

A second training model is constructed according to the second model parameters, and the second training model is used to train the second medical data of the local database.
The method according to claim 4, training the initial training model according to the first medical data in the local database to obtain the first training model, comprising:

Using the first medical data of the local database to perform batch gradient calculation on the initial training model to obtain multiple gradient values;

Calculating the average gradient of the multiple gradient values;

The average gradient is used to update the initial weight value of the initial training model to obtain the first model parameter.
A data processing and training device, which includes:

The first sending module is configured to send the initial training model to multiple clients, where each of the multiple clients communicates with the server separately;

The receiving module is configured to receive multiple sets of first model parameters sent by the multiple clients, where the first model parameters are the clients training the initial training model according to the first medical data in the local database owned;

A calculation module, configured to perform a weighted average of the multiple sets of first model parameters to obtain a second model parameter;

The second sending module is configured to send the second model parameters to the multiple clients, where the second model parameters are used to construct the same second training model on the multiple clients respectively.
The device according to claim 6, the device further comprising:

The decryption module is used to decrypt the first model parameters according to a preset private key before the multiple sets of first model parameters are weighted and averaged to obtain the second model parameters. The public key sent by the target terminal is a set of key pairs, and the public key is used to encrypt the first model parameter.
The device according to claim 6, wherein the calculation module comprises:

The selection unit is used to select among the M sets of first model parameters
The first model parameter of the second time, where N sets of first model parameters are selected each time, and the N sets of the first model parameters selected each time are weighted and averaged to obtain a first-level model parameter, where N is an integer less than M ；Computer unit, used to
Performing a weighted average of the first-level model parameters to obtain the second model parameter.
A data training device, which includes:

The receiving module is used to receive the initial training model sent by the server;

The first training module is configured to train the initial training model according to the first medical data in the local database to obtain the first training model;

The sending module is configured to send the first model parameters of the first training model to the server, where the server is configured to perform a weighted average on multiple sets of first model parameters of multiple clients to obtain a second training model And feedback the second model parameters to the multiple clients;

The second training module is configured to construct a second training model according to the second model parameters, and use the second training model to train the second medical data of the local database.
The device according to claim 9, wherein the first training module comprises:

The first calculation unit is configured to use the first medical data of the local database to perform batch gradient calculation on the initial training model to obtain multiple gradient values;

The second calculation unit is used to calculate the average gradient of the multiple gradient values;

The third calculation unit is configured to use the average gradient to update the initial weight value of the initial training model to obtain the first model parameter.
A data training system, including: a server and multiple clients, among which,

The server includes: a first sending module, used to send initial training models to multiple clients; a receiving module, used to receive multiple sets of first model parameters sent by the multiple clients; Performing a weighted average of the multiple sets of first model parameters to obtain a second model parameter; a second sending module, configured to send the second model parameter to the multiple clients;

The multiple clients all communicate with the server separately, and include: a receiving module for receiving the initial training model; a first training module for training the initial training model according to the first medical data in the local database , Obtain the first training model; a sending module, for sending the first model parameters of the first training model to the server; a second training module, for constructing a second training according to the second model parameters Model, and use the second training model to train the second medical data of the local database.
The system according to claim 11, the server further comprising:

The decryption module is used to decrypt the first model parameters according to a preset private key before the multiple sets of first model parameters are weighted and averaged to obtain the second model parameters. The public key sent by the target terminal is a set of key pairs, and the public key is used to encrypt the first model parameter.
The system according to claim 11, said calculation module comprising:

The selection unit is used to select among the M sets of first model parameters
The first model parameter of the second time, where N sets of first model parameters are selected each time, and the N sets of the first model parameters selected each time are weighted and averaged to obtain a first-level model parameter, where N is an integer less than M ；Computer unit, used to
Performing a weighted average of the first-level model parameters to obtain the second model parameter.
The system according to claim 11, the first training module comprises:

The first calculation unit is configured to use the first medical data of the local database to perform batch gradient calculation on the initial training model to obtain multiple gradient values;

The second calculation unit is used to calculate the average gradient of the multiple gradient values;

The third calculation unit is configured to use the average gradient to update the initial weight value of the initial training model to obtain the first model parameter.
A computer device includes a memory and a processor, the memory stores a computer program, and the steps of implementing a data training method when the processor executes the computer program include:

Sending the initial training model to multiple clients, where each of the multiple clients communicates with the server separately;

Receiving multiple sets of first model parameters sent by the multiple clients, where the first model parameters are obtained by the client training the initial training model according to the first medical data in the local database;

Performing a weighted average on the multiple sets of first model parameters to obtain a second model parameter;

The second model parameter is sent to the multiple clients, where the second model parameter is used to construct the same second training model on the multiple clients respectively.
The computer device according to claim 15, before the weighted average of the multiple sets of first model parameters is performed to obtain the second model parameters, the method further comprises:

The first model parameter is decrypted according to the preset private key, wherein the private key and the public keys corresponding to the multiple clients form a set of key pairs, and the public key is used to decrypt the first model parameter. Model parameters are encrypted.
A computer device includes a memory and a processor, the memory stores a computer program, and the steps of implementing a data training method when the processor executes the computer program include:

Receive the initial training model sent by the server;

Training the initial training model according to the first medical data in the local database to obtain the first training model;

The first model parameters of the first training model are sent to the server, where the server is used to perform a weighted average of multiple sets of first model parameters of multiple clients to obtain the second model parameters of the second training model , And feed back the second model parameters to the multiple clients;

A second training model is constructed according to the second model parameters, and the second training model is used to train the second medical data of the local database.
The computer device according to claim 17, training the initial training model according to the first medical data in the local database to obtain the first training model, comprising:

Using the first medical data of the local database to perform batch gradient calculation on the initial training model to obtain multiple gradient values;

Calculating the average gradient of the multiple gradient values;

The average gradient is used to update the initial weight value of the initial training model to obtain the first model parameter.
A computer storage medium having a computer program stored thereon, and the steps of implementing a data training method when the computer program is executed by a processor include:

Sending the initial training model to multiple clients, where each of the multiple clients communicates with the server separately;

Receiving multiple sets of first model parameters sent by the multiple clients, where the first model parameters are obtained by the client training the initial training model according to the first medical data in the local database;

Performing a weighted average on the multiple sets of first model parameters to obtain a second model parameter;

The second model parameter is sent to the multiple clients, where the second model parameter is used to construct the same second training model on the multiple clients respectively.
A computer storage medium having a computer program stored thereon, and the steps of implementing a data training method when the computer program is executed by a processor include:

Receive the initial training model sent by the server;

Training the initial training model according to the first medical data in the local database to obtain the first training model;

The first model parameters of the first training model are sent to the server, where the server is used to perform a weighted average of multiple sets of first model parameters of multiple clients to obtain the second model parameters of the second training model , And feed back the second model parameters to the multiple clients;

A second training model is constructed according to the second model parameters, and the second training model is used to train the second medical data of the local database.