Machine learning training method, controller, device, server, terminal and medium

1. A machine learning training method, applied to a machine learning controller in a server, comprising:

acquiring training data uploaded by a terminal, wherein the training data is encrypted training data encrypted with an encryption key at the terminal, and comprises training data from at least two data providers;

creating a trusted execution environment in response to a machine learning training request from the terminal;

establishing a trusted communication link between the terminal and the trusted execution environment, wherein the trusted communication link is configured to transmit the encryption key of the terminal to a key manager in the trusted execution environment, the key manager being configured to manage the encryption key;

submitting the encrypted training data of each data provider and preset training parameters to a data fusion manager in the trusted execution environment, decrypting the encrypted training data of each data provider by the data fusion manager according to an encryption key of each data provider, and fusing the decrypted training data according to a preset fused data format to obtain fused training data;

triggering a target algorithm in a machine learning algorithm library in the executed execution environment, and training the fused training data according to the training parameters in the trusted execution environment; and

acquiring a machine learning model obtained after training.

2. He method according to claim 1 , before establishing the trusted communication link between the terminal and the trusted execution environment, further comprising:

sending information to be authenticated of the server and the trusted execution environment to the terminal, and enabling the terminal to perform remote authentication on the trusted execution environment through a remote authentication server of the trusted execution environment based on the information to be authenticated; and

executing an operation of establishing the trusted communication link when the authentication is successful.

3. The method according to claim 1 , wherein an operation for fusing the decrypted training data comprises: splitting data column by column and splitting data row by row.

4. The method according to claim 1 , wherein the acquired machine learning model is a model encrypted in the machine learning algorithm library in the trusted execution environment.

5. The method according to claim 3 , wherein the acquired machine learning model is a model encrypted in the machine learning algorithm library in the trusted execution environment.

6. A server, comprising:

one or more processors;

a storage device, configured to store one or more programs,

wherein, when the one or more programs are executed by the one or more processors, the one or more processors are configured to implement a machine learning training method, the method comprising:

acquiring training data uploaded by a terminal, wherein the training data is encrypted training data encrypted with an encryption key at the terminal, and comprises training data from at least two data providers;

creating a trusted execution environment in response to a machine learning training request from the terminal;

establishing a trusted communication link between the terminal and the trusted execution environment, wherein the trusted communication link is configured to transmit the encryption key of the terminal to a key manager in the trusted execution environment, the key manager being configured to manage the encryption key;

submitting the encrypted training data of each data provider and preset training parameters to a data fusion manager in the trusted execution environment, decrypting the encrypted training data of each data provider by the data fusion manager according to an encryption key of each data provider, and fusing the decrypted training data according to a preset fused data format to obtain fused training data;

triggering a target algorithm in a machine learning algorithm library in the executed execution environment, and training the fused training data according to the training parameters in the trusted execution environment; and

acquiring a machine learning model obtained after training.

7. The server according to claim 6 , wherein, before establishing the trusted communication link between the terminal and the trusted execution environment, the method further comprises:

sending information to be authenticated of the server and the trusted execution environment to the terminal, and enabling the terminal to perform remote authentication on the trusted execution environment through a remote authentication server of the trusted execution environment based on the information to be authenticated; and

executing an operation of establishing the trusted communication link when the authentication is successful.

8. The server according to claim 6 , wherein an operation for fusing the decrypted training data comprises: splitting data column by column and splitting data row by row.

9. The server according to claim 6 , wherein the acquired machine learning model is a model encrypted in the machine learning algorithm library in the trusted execution environment.