Model training method, central node device, and computer program product

Patent metadata

Field	Value
Publication number	US-12519786-B2
Application number	US-202418589547-A
Country	US
Kind code	B2
Filing date	Feb 28, 2024
Priority date	Jan 26, 2024
Publication date	Jan 6, 2026
Grant date	Jan 6, 2026

Abstract

A method in one embodiment comprises: communicating, in response to receiving configuration information, with a remote attestation device to attest a trusted execution environment (TEE) of a central node device, the configuration information including a division manner for training sample data for model training; establishing a secure channel with one or more edge node devices in response to passing the attestation for the TEE, wherein the one or more edge node devices each have a TEE; and performing the following operations iteratively until a predetermined condition is satisfied: selecting, from the one or more edge node devices, at least one edge node device for training a model; updating a global model parameter based on a training result received over the secure channel from the selected at least one edge node device; and sending, to the one or more edge node devices, the updated global model parameter over the secure channel.

First claim

What is claimed is:

1 . A federated learning-based model training method, which is executed by a central node device having a hardware-based Trusted Execution Environment (TEE), the method comprising:
communicating, in response to receiving configuration information, with a remote attestation device to attest the hardware-based TEE of the central node device, the configuration information comprising a division manner for training sample data for model training; establishing a secure channel with one or more edge node devices in response to passing the attestation for the hardware-based TEE, wherein the one or more edge node devices each have a hardware-based TEE; and performing the following operations iteratively until a predetermined condition is satisfied: selecting, from the one or more edge node devices, at least one edge node device for training a model; controlling deployment of at least a portion of the model in the at least one edge node device; initiating training of the model in the at least one edge node device to produce a training result; receiving the training result from the selected at least one edge node device over the secure channel; updating a global model parameter based on the training result received over the secure channel from the selected at least one edge node device, wherein the selected at least one edge node device trains corresponding models separately according to the division manner, and the training result corresponds to the division manner; and sending, to the one or more edge node devices, the updated global model parameter over the secure channel.

2 . The method according to claim 1 , wherein establishing a secure channel with the one or more edge node devices comprises:
sending edge node configuration information to the one or more edge node devices, the edge node configuration information comprising information about a remote attestation device that performs TEE attestation with the one or more edge node devices; sending, in response to the one or more edge node devices each passing the TEE attestation, to the one or more edge node devices, a request to establish a secure channel; and establishing the secure channel between the central node device and the one or more edge node devices in response to an acknowledgment from the one or more edge node devices.

3 . The method according to claim 1 , wherein the configuration information further comprises structure information of a model for training that corresponds to the division manner and is deployed in each of the one or more edge node devices.

4 . The method according to claim 1 , wherein in response to the division manner comprising a horizontal federated learning manner, the training result comprises model parameters obtained from model training for a model in each of the selected at least one edge node device.

5 . The method according to claim 4 , further comprising:
processing the obtained model parameters according to a model parameter update manner in the configuration information to obtain the updated global model parameter; and sending the updated global model parameter to each of the one or more edge node devices via the secure channel.

6 . The method according to claim 1 , wherein in response to the division manner comprising a vertical federated learning manner, the training result comprises an intermediate result of model training for a model in each of the selected at least one edge node device.

7 . The method according to claim 6 , wherein the intermediate result comprises an element associated with a sequence number of the training sample data.

8 . The method according to claim 6 , further comprising:
constructing an operator used during the model training; and storing the constructed operator in the hardware-based TEE of the central node device.

9 . The method according to claim 7 , further comprising:
invoking a corresponding operator stored in the hardware-based TEE based on the sequence number and a label of the training sample data and according to a model parameter update approach in the configuration information to process the intermediate result to obtain the updated global model parameter; and sending the updated global model parameter to each of the one or more edge node devices via the secure channel.

10 . The method according to claim 1 , wherein a model corresponding to each of the one or more edge node devices is stored in the hardware-based TEE of the corresponding edge node device.

11 . A central node device having a hardware-based Trusted Execution Environment (TEE), the central node device comprising:
at least one processor, and memory coupled to the at least one processor and having instructions stored therein, wherein the instructions, when executed by the at least one processor, cause the central node device to perform actions comprising: communicating, in response to receiving configuration information, with a remote attestation device to attest the hardware-based TEE of the central node device, the configuration information comprising a division manner for training sample data for model training; establishing a secure channel with one or more edge node devices in response to passing the attestation for the hardware-based TEE, wherein the one or more edge node devices each have a hardware-based TEE; and performing the following operations iteratively until a predetermined condition is satisfied: selecting, from the one or more edge node devices, at least one edge node device for training a model; controlling deployment of at least a portion of the model in the at least one edge node device; initiating training of the model in the at least one edge node device to produce a training result; receiving the training result from the selected at least one edge node device over the secure channel; updating a global model parameter based on the training result received over the secure channel from the selected at least one edge node device, wherein the selected at least one edge node device trains corresponding models separately according to the division manner, and the training result corresponds to the division manner; and sending, to the one or more edge node devices, the updated global model parameter over the secure channel.

12 . The central node device according to claim 11 , wherein establishing a secure channel with the one or more edge node devices comprises:
sending edge node configuration information to the one or more edge node devices, the edge node configuration information comprising information about a remote attestation device that performs TEE attestation with the one or more edge node devices; sending, in response to the one or more edge node devices each passing the TEE attestation, to the one or more edge node devices, a request to establish a secure channel; and establishing the secure channel between the central node device and the one or more edge node devices in response to an acknowledgment from the one or more edge node devices.

13 . The central node device according to claim 11 , wherein the configuration information further comprises structure information of a model for training that corresponds to the division manner and is deployed in each of the one or more edge node devices.

14 . The central node device according to claim 11 , wherein in response to the division manner comprising a horizontal federated learning manner, the training result comprises model parameters obtained from model training for a model in each of the selected at least one edge node device.

15 . The central node device according to

Assignees

• Dell Products Lp

Inventors

• Chen Tianxiang
• Liu Jinpeng
• Wang Zijia
• Ni Jiacheng
• Jia Zhen

Classifications

• H04L63/0435 Primary

  wherein the sending and receiving network entities apply symmetric encryption, i.e. same key used for encryption and decryption (cryptographic mechanisms or cryptographic arrangements for symmetric key encryption H04L9/06) · CPC title

• H04L63/0876 Primary

  based on the identity of the terminal or configuration, e.g. MAC address, hardware or software configuration or device fingerprint · CPC title

• G06F21/57

  Certifying or maintaining trusted computer platforms, e.g. secure boots or power-downs, version controls, system software checks, secure updates or assessing vulnerabilities · CPC title

• G06N20/20 Primary

  Ensemble learning · CPC title