WO2025246373A1 - Information transmission method and related apparatus - Google Patents

Abstract

Provided in the embodiments of the present application are an information transmission method and a related apparatus, which are used for allocating corresponding model training information to a first data collection apparatus, facilitating the completion of model training on the basis of data provided by the first data collection apparatus, thus solving the problem of it being impossible to perform model training when the first data collection apparatus has weaker computing capability or has no computing capability at all. The method provided in the present application comprises: a first data collection apparatus sending a first request to a model training management apparatus, wherein the first request is used for requesting that model training information is allocated to the first data collection apparatus; and the first data collection apparatus receiving first indication information from the model training management apparatus, wherein the first indication information is used for indicating first model training information allocated to the first data collection apparatus.

Description

Information transmission methods and related devices

This application claims priority to Chinese Patent Application No. 202410705939.1, filed with the State Intellectual Property Office of China on May 31, 2024, entitled "Information Transmission Method and Related Apparatus", the entire contents of which are incorporated herein by reference.

Technical Field

This application relates to the field of artificial intelligence technology, and more particularly to an information transmission method and related apparatus.

Background Technology

Distributed learning can complete the learning task of AI models while fully ensuring user data privacy and security. Distributed learning mainly includes federated learning, partitioned learning, and decentralized learning. Assuming that the distributed nodes (e.g., terminal devices) have sufficient computing power, the distributed nodes train the model based on locally collected data to obtain a local model, and then transmit the local model to other nodes (e.g., the central node or other distributed nodes).

However, if the computing power of a distributed node is limited and it is unable to train the model, then how to achieve distributed learning to complete the model training task is a question worth considering.

Summary of the Invention

This application provides an information transmission method and related apparatus for allocating first model training information to a first data acquisition device. This facilitates the first data acquisition device in determining how to complete model training based on the first model training information, thereby enabling model training to be completed based on the local data of the first data acquisition device.

This application provides an information transmission method, which can be executed by a first data acquisition device. The first data acquisition device can be a terminal device, a component within the terminal device (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the terminal device's functions. The method includes: the first data acquisition device sending a first request to a model training management device, the first request requesting the allocation of model training information for the first data acquisition device; and the first data acquisition device receiving first instruction information from the model training management device, the first instruction information indicating the allocation of first model training information for the first data acquisition device. This enables the allocation of first model training information to the first data acquisition device. It facilitates the first data acquisition device in determining how to complete model training based on the first model training information, thereby enabling model training based on the local data of the first data acquisition device. For example, when the first data acquisition device has weak computing power or lacks computing power, the above technical solution helps solve the problem of the first data acquisition device being unable to perform model training. For example, the first data acquisition device can determine a first model training device based on the first model training information and send its local data to the first model training device. This facilitates the first model training device in completing model training based on the local data. For example, the first data acquisition device is a terminal device, and the first model training device is the terminal manufacturer's server. Given the limited computing power of the terminal device and its data privacy protection requirements, the terminal manufacturer's server can complete model training based on the terminal device's local data. This ensures both the data privacy protection requirements of the first data acquisition device and the ability to train the model.

Based on the first aspect, in one possible implementation, the first model training information includes information about the first model training device and/or the first model training resources. This allows for the allocation of a corresponding model training device and training resources to the first data acquisition device, facilitating the completion of the model training task based on the data provided by the first data acquisition device.

Based on the first aspect, in one possible implementation, the first indication information includes a first identifier, which indicates that the first model training device is a model training device for the first data acquisition device. This allows the first data acquisition device to be paired or matched with the first model training device via the first identifier. This facilitates the first model training device in subsequently determining whether to perform model training based on data from the first data acquisition device using the first identifier.

Based on the first aspect, in one possible implementation, the first identifier is either the identifier of the first data acquisition device, the identifier of the first model training device, or a pairing identifier. The pairing identifier is used to indicate that the first data acquisition device is paired with the first model training device. The first identifier can be similar to the destination address of a packet in network routing, which facilitates the first model training device in determining whether to perform model training based on the data from the first data acquisition device.

Based on the first aspect, in one possible implementation, the method further includes: a first data acquisition device sending first data to a first model training device, the first data including a first identifier, the first data being used for model training. This facilitates the first model training device to perform model training based on the first identifier and the first data.

Based on the first aspect, in one possible implementation, the method further includes: a first data acquisition device receiving a first model from a first model training device, wherein the first model is trained based on the first data; or, the first data acquisition device receiving a second model from a model fusion device, wherein the second model is obtained by fusing the first model reported by the first model training device, and the first model is trained based on the first data. In this implementation, in decentralized learning, the first data acquisition device can receive the first model from the first model training device. In federated learning, the first data acquisition device can receive the second model from the model fusion device. This enables the first data acquisition device to obtain the trained model, facilitating model inference operations based on the first model or the second model, thereby improving model inference performance.

Based on the first aspect, in one possible implementation, the method further includes: a first data acquisition device receiving a third model from a second model training device or a second data acquisition device, the first data acquisition device sending the third model to the first model training device, and the third model being used for model fusion.

Based on the first aspect, in one possible implementation, the method further includes: a first data acquisition device receiving a fourth model from a first model training device, the fourth model being obtained by fusing the first model and the third model.

Based on the first aspect, in one possible implementation, before the first data acquisition device sends the first data to the first model training device, the method further includes: the first data acquisition device receiving scheduling information from the control device, the scheduling information being used to schedule the first data acquisition device to send the first data. This achieves scheduling of the first data acquisition device.

Based on the first aspect, in one possible implementation, before the first data acquisition device receives scheduling information from the control device, the method further includes: the first data acquisition device sending second indication information to the control device, the second indication information being used to indicate the data status of the first data. This facilitates the control device in determining whether to schedule the first data acquisition device based on the data status of the first data. It also facilitates the control device in scheduling data acquisition devices with higher data quality to provide data, thereby improving the performance of model training.

Based on the first aspect, in one possible implementation, the second indication information is further used to indicate the first identifier. This facilitates the control device in determining the first model training device that matches the first data acquisition device based on the first identifier, so that the control device can determine whether to schedule the first data acquisition device.

A second aspect of this application provides an information transmission method, which can be executed by a model training management device. The model training management device can be a model training management server, a component within a model training management server (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of a model training management server. The method includes: the model training management device receiving a first request from a first data acquisition device, the first request requesting the allocation of model training information to the first data acquisition device; and the model training management device sending first instruction information to the first data acquisition device, the first instruction information indicating the allocation of first model training information to the first data acquisition device. This method allocates first model training information to the first data acquisition device, facilitating the first data acquisition device to determine how to complete model training based on the first model training information, thereby enabling model training based on the local data of the first data acquisition device. For example, when the first data acquisition device has weak computing power or no computing power at all, the above technical solution helps to solve the problem that the first data acquisition device cannot perform model training. For example, the first data acquisition device can determine the first model training device based on the first model training information and send its local data to the first model training device. This facilitates the first model training device to complete model training based on the local data. For example, if the first data acquisition device is a terminal device and the first model training device is a server from the terminal manufacturer, and the terminal device has limited computing power and data privacy protection requirements, the terminal manufacturer's server can complete model training based on the terminal device's local data. This ensures both the data privacy protection requirements of the first data acquisition device and the ability to train the model.

Based on the second aspect, in one possible implementation, the method further includes: the model training management device sending first instruction information to the first model training device. This facilitates the first model training device in determining the first data acquisition device it is paired with.

Based on the second aspect, one possible implementation further includes: the model training management device sending first instruction information to the control device. This facilitates the control device in determining the pairing between the first data acquisition device and the first model training device. It also allows the control device to better schedule the data acquisition device, thereby improving the performance of model training.

Based on the second aspect, in one possible implementation, the first model training information includes information about the first model training device and/or the first model training resources. This allows for the allocation of a corresponding model training device and training resources to the first data acquisition device, facilitating the provision of model training functionality to the first data acquisition device and thus completing the model training task.

Based on the second aspect, in one possible implementation, the first indication information includes a first identifier, which indicates that the first model training device is a model training device for the first data acquisition device. This allows the first data acquisition device to be paired or matched with the first model training device via the first identifier. This facilitates the first model training device in subsequently determining whether to perform model training based on data from the first data acquisition device using the first identifier.

Based on the second aspect, in one possible implementation, the first identifier is either the identifier of the first data acquisition device, the identifier of the first model training device, or a pairing identifier. The pairing identifier is used to indicate that the first data acquisition device is paired with the first model training device. The first identifier can be similar to the destination address of a packet in network routing, which facilitates the first model training device to determine whether to perform model training based on the data from the first data acquisition device.

A third aspect of this application provides an information transmission method, which can be executed by a first model training device. The first model training device can be an AI server, or a component within an AI server (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of an AI server. The method includes: the first model training device receiving first instruction information from a model training management device, the first instruction information indicating first model training information allocated to a first data acquisition device; and the first model training device determining the first model training information based on the first instruction information. This method allocates first model training information to the first data acquisition device, facilitating the first model training device to determine, based on local data from the first data acquisition device, to perform model training, thereby completing the model training. For example, when the first data acquisition device has weak computing power or lacks computing power, the above technical solution helps solve the problem of the first data acquisition device being unable to perform model training. For example, the first model training device determines itself as the model training device for the first data acquisition device based on the first model training information. The first model training device can receive local data from the first data acquisition device and complete model training based on that local data. This solves the problem of model training being impossible when the first data acquisition device has limited or no computing power. For example, if the first data acquisition device is a terminal device and the first model training device is the terminal manufacturer's server, and the terminal device has limited computing power and data privacy protection requirements, the terminal manufacturer's server can complete model training based on the terminal device's local data. This ensures both the data privacy protection requirements of the first data acquisition device and the ability to train the model.

Based on the third aspect, in one possible implementation, the first model training information includes information about the first model training device and/or the first model training resources. This allows for the allocation of a corresponding model training device and training resources to the first data acquisition device. This facilitates the first model training device providing model training functionality to the first data acquisition device, thereby completing the model training task.

Based on the third aspect, in one possible implementation, the first indication information includes a first identifier, which identifies the first model training device as a model training device that is also a first data acquisition device. This allows the first data acquisition device to be paired or matched with the first model training device via the first identifier. This facilitates the first model training device in subsequently determining whether to perform model training based on data from the first data acquisition device using the first identifier.

Based on the third aspect, in one possible implementation, the first identifier is either the identifier of the first data acquisition device, the identifier of the first model training device, or a pairing identifier. The pairing identifier is used to identify the pairing of the first data acquisition device and the first model training device. The first identifier can be similar to the destination address of a packet in network routing, which facilitates the first model training device in determining whether to perform model training based on the data from the first data acquisition device.

Based on the third aspect, in one possible implementation, the method further includes: a first model training device receiving first data from a first data acquisition device, the first data including a first identifier; the first model training device performing model training based on the first data to obtain a first model. This enables the first model training device to provide model training functionality to the first data acquisition device, completing the model training task.

Based on the third aspect, one possible implementation further includes: the first model training device sending the first model to the first data acquisition device or the model fusion device. This facilitates model inference by the first data acquisition device, improving model inference performance. Alternatively, it facilitates model fusion by the model fusion device, improving model training performance.

Based on the third aspect, in one possible implementation, the method further includes: a first model training device receiving a second model from a model fusion device, the second model being obtained by fusing the first model by the model fusion device. This facilitates the first model training device performing model inference and/or model training based on the second model.

Based on the third aspect, one possible implementation further includes: a first model training device receiving a third model from a first data acquisition device, a second data acquisition device, or a second model training device, the third model being used for model fusion. In decentralized learning, this allows models from different data acquisition devices or different model training devices to be transferred to each other, facilitating model fusion.

Based on the third aspect, in one possible implementation, the first model training device fuses the first model and the third model to obtain a fourth model; the first model training device then sends the fourth model to the first data acquisition device. In decentralized learning, models from different data acquisition devices or different model training devices can be transferred to each other and fused, thereby improving model training performance.

Based on the third aspect, in one possible implementation, before the first model training device receives the first data from the first data acquisition device, the method further includes: the first model training device sending third instruction information to the control device, the third instruction information being used to indicate training-related information of the first model training device. This facilitates better scheduling of the corresponding data acquisition devices by the control device, improving the performance of model training.

Based on the third aspect, in one possible implementation, the third indication information is also used to indicate the first identifier. This facilitates the control device in determining the first model training device that matches the first data acquisition device based on the first identifier, so that the control device can determine whether to schedule the first data acquisition device.

A fourth aspect of this application provides an information transmission method, which can be executed by a control device. The control device can be a network device, a component within the network device (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of the network device. The method includes: the control device receiving first instruction information from a model training management device, the first instruction information indicating first model training information allocated to a first data acquisition device; and the control device determining the first model training information based on the first instruction information. This facilitates the control device scheduling the first data acquisition device based on the first model training information, enabling model training to be completed based on the local data of the first data acquisition device. For example, if the first data acquisition device has weak computing power or no computing power at all, the above technical solution can solve the problem that the first data acquisition device cannot perform model training. For example, if the first data acquisition device is a terminal device and the first model training device is a server of the terminal manufacturer, and the terminal device has limited computing power and data privacy protection requirements, the control device can schedule the terminal device, and the terminal device can send local data to the terminal manufacturer's server. Then, the terminal manufacturer's server can complete model training based on the local data of the terminal device. It can both ensure the data privacy protection needs of the primary data acquisition device and enable model training.

Based on the fourth aspect, in one possible implementation, the first model training information includes: information about the first model training device, and/or, the first model training resources. The control device determines the model training device and/or model training resources allocated to the first data acquisition device.

Based on the fourth aspect, in one possible implementation, the first indication information includes a first identifier, which indicates that the first model training device is used as the model training device for the first data acquisition device. This allows the first identifier to indicate pairing or matching between the first data acquisition device and the first model training device. This facilitates the control device in determining whether to schedule the first data acquisition device based on its data status and the training capability of the first model training device. Consequently, it allows the control device to schedule data acquisition devices with higher data quality, thereby improving model training performance.

Based on the fourth aspect, in one possible implementation, the first identifier is either the identifier of the first data acquisition device, the identifier of the first model training device, or a pairing identifier, which is used to indicate that the first data acquisition device and the first model training device are paired. The control device determines the pairing of the first model training device and the first data acquisition device based on the first identifier.

Based on the fourth aspect, in one possible implementation, the method further includes: the control device sending scheduling information to the first data acquisition device, the scheduling information being used to schedule the first data acquisition device to send the first data.

Based on the fourth aspect, in one possible implementation, before the control device sends scheduling information to the first data acquisition device, the method further includes: the control device receiving second indication information from the first data acquisition device, the second indication information indicating the data status of the first data; and the control device determining the scheduling of the first data acquisition device based on the second indication information. This is beneficial for the control device to schedule data acquisition devices with higher data quality, thereby improving the performance of model training.

Based on the fourth aspect, in one possible implementation, the method further includes: the control device receiving third instruction information from the first model training device; the control device determining the scheduling of the first data acquisition device based on the second instruction information, including: the control device determining the scheduling of the first data acquisition device based on the second instruction information and the third instruction information. This enables the control device to better schedule the data acquisition device, ensuring that the model training device matched with the scheduled data acquisition device can provide model training functionality.

This application provides a fifth aspect of an information transmission method, which can be executed by a control device. The control device can be a network device, a component within the network device (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of the network device. The method includes: the control device receiving second indication information from a first data acquisition device, the second indication information indicating the data status and a first identifier of first data, the first identifier indicating a first model training device as the model training device for the first data acquisition device; the control device receiving third indication information from the first model training device, the third indication information indicating training-related information and a first identifier for the first model training device; and the control device determining the first model training device as the model training device for the first data acquisition device based on the second and third indication information. This facilitates the control device scheduling the first data acquisition device based on the second and third indication information to achieve model training based on the local data of the first data acquisition device. For example, if the first data acquisition device has weak computing power or no computing power at all, the above technical solution can solve the problem that the first data acquisition device cannot perform model training. For example, if the first data acquisition device is a terminal device and the first model training device is the terminal manufacturer's server, and the terminal device has limited computing power and data privacy protection requirements, the control device can schedule the terminal device based on the second and third instruction information. The terminal device then sends local data to the terminal manufacturer's server. The terminal manufacturer's server can then complete model training based on the terminal device's local data. This approach ensures both the data privacy protection requirements of the first data acquisition device and the successful training of the model.

Based on the fifth aspect, one possible implementation further includes: the control device determining the scheduling of the first data acquisition device according to the second and third instruction information. This enables the control device to better schedule data acquisition devices, ensuring that the model training device matched with the scheduled data acquisition device can provide model training functionality. Furthermore, it facilitates the control device scheduling data acquisition devices with higher data quality, thereby improving model training performance.

Based on the fifth aspect, in one possible implementation, the method further includes: the control device sending scheduling information to the first data acquisition device, the scheduling information being used to schedule the first data acquisition device.

A sixth aspect of this application provides an information transmission method, which can be executed by a model fusion device. The model fusion device can be a network device, a component within a network device (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of a network device. The method includes: the model fusion device receiving a first model from a first model training device, the first model being obtained based on first data from a first data acquisition device; and the model fusion device fusing the first model to obtain a second model. This completes the training and fusion of the models.

Based on the sixth aspect, one possible implementation further includes: the model fusion device sending the second model to the first data acquisition device. This facilitates the first data acquisition device in performing model inference based on the second model, improving model inference performance.

Based on the sixth aspect, in one possible implementation, the method further includes: the model fusion device sending the second model to the first model training device. This facilitates the first model training device to perform model training and/or model inference based on the second model, thereby improving model trainability and/or model inference performance.

A seventh aspect of this application provides a first data acquisition device, the first data acquisition device comprising:

The transceiver module is used to send a first request to the model training management device, the first request being used to request the allocation of model training information to the first data acquisition device; and to receive first instruction information from the model training management device, the first instruction information being used to indicate the allocation of first model training information to the first data acquisition device.

Based on the seventh aspect, in one possible implementation, the first model training information includes: information about the first model training device, and/or, the first model training resources.

Based on the seventh aspect, in one possible implementation, the first indication information includes a first identifier, which is used to indicate that the first model training device is a model training device for the first data acquisition device.

Based on the seventh aspect, in one possible implementation, the first identifier is the identifier of the first data acquisition device, or the identifier of the first model training device, or a pairing identifier, which is used to indicate that the first data acquisition device is paired with the first model training device.

Based on the seventh aspect, in one possible implementation, the transceiver module is further configured to: receive a first model from a first model training device, wherein the first model is trained based on first data, or receive a second model from a model fusion device, wherein the second model is obtained by fusing the first model reported by the first model training device, and the first model is trained based on the first data.

Based on the seventh aspect, in one possible implementation, the transceiver module is further configured to: receive a third model from a second model training device or a second data acquisition device; and send the third model to a first model training device, wherein the third model is used for model fusion.

Based on the seventh aspect, in one possible implementation, the transceiver module is further configured to: receive a fourth model from the first model training device, the fourth model being obtained by fusing the first model and the third model.

Based on the seventh aspect, in one possible implementation, the transceiver module is further configured to: receive scheduling information from the control device, the scheduling information being used to schedule the first data acquisition device to send the first data.

Based on the seventh aspect, in one possible implementation, the transceiver module is further configured to: send a second indication message to the control device, the second indication message being used to indicate the data status of the first data.

Based on the seventh aspect, in one possible implementation, the second indication information is also used to indicate the first identifier.

The eighth aspect of this application provides a model training management device, which includes:

The transceiver module is used to receive a first request from a first data acquisition device, the first request being used to request the allocation of model training information to the first data acquisition device; and to send first instruction information to the first data acquisition device, the first instruction information being used to indicate the allocation of first model training information to the first data acquisition device.

Based on the eighth aspect, in one possible implementation, the transceiver module is further configured to: send first instruction information to the first model training device.

Based on the eighth aspect, in one possible implementation, the transceiver module is also used to: send first instruction information to the control device.

Based on the eighth aspect, in one possible implementation, the first model training information includes information about the first model training device and/or, the first model training resources.

Based on the eighth aspect, in one possible implementation, the first indication information includes a first identifier, which is used to indicate that the first model training device is a model training device for the first data acquisition device.

Based on the eighth aspect, in one possible implementation, the first identifier is the identifier of the first data acquisition device, or the identifier of the first model training device, or a pairing identifier, which is used to indicate that the first data acquisition device is paired with the first model training device.

A ninth aspect of this application provides a first model training apparatus, the first model training apparatus comprising:

The transceiver module is used to receive first instruction information from the model training management device, the first instruction information being used to indicate the first model training information allocated to the first data acquisition device;

The processing module is used to determine the first model training information based on the first instruction information.

Based on the ninth aspect, in one possible implementation, the first model training information includes information about the first model training device and/or, the first model training resources.

Based on the ninth aspect, in one possible implementation, the first indication information includes a first identifier, which is used to identify the first model training device as a model training device for the first data acquisition device.

Based on the ninth aspect, in one possible implementation, the first identifier is the identifier of the first data acquisition device, or the identifier of the first model training device, or a pairing identifier, which is used to identify the pairing of the first data acquisition device and the first model training device.

Based on the ninth aspect, in one possible implementation, the transceiver module is further configured to: receive first data from the first data acquisition device, the first data including a first identifier; the processing module is further configured to: perform model training based on the first data to obtain a first model.

Based on the ninth aspect, in one possible implementation, the transceiver module is further configured to: send the first model to the first data acquisition device or the model fusion device.

Based on the ninth aspect, in one possible implementation, the transceiver module is further configured to: receive a second model from the model fusion device, the second model being obtained by the model fusion device fusing the first model.

Based on the ninth aspect, in one possible implementation, the transceiver module is further configured to: receive a third model from the first data acquisition device, the second data acquisition device, or the second model training device, wherein the third model is used for model fusion.

Based on the ninth aspect, in one possible implementation, the processing module is further configured to: fuse the first model and the third model to obtain a fourth model; the transceiver module is further configured to: send the fourth model to the first data acquisition device.

Based on the ninth aspect, in one possible implementation, the transceiver module is further configured to: send third instruction information to the control device, the third instruction information being used to indicate training-related information of the first model training device.

Based on the ninth aspect, in one possible implementation, the third indication information is also used to indicate the first identifier.

The tenth aspect of this application provides a control device, the control device comprising:

The transceiver module is used to receive first instruction information from the model training management device, the first instruction information being used to indicate the first model training information allocated to the first data acquisition device;

The processing module is used to determine the first model training information based on the first instruction information.

Based on the tenth aspect, in one possible implementation, the first model training information includes: information about the first model training device, and/or, the first model training resources.

Based on the tenth aspect, in one possible implementation, the first indication information includes a first identifier, which is used to indicate that the first model training device is a model training device for the first data acquisition device.

Based on the tenth aspect, in one possible implementation, the first identifier is the identifier of the first data acquisition device, or the identifier of the first model training device, or a pairing identifier, which is used to indicate that the first data acquisition device is paired with the first model training device.

Based on the tenth aspect, in one possible implementation, the transceiver module is further configured to: send scheduling information to the first data acquisition device, wherein the scheduling information is used to schedule the first data acquisition device to send the first data.

Based on the tenth aspect, in one possible implementation, the transceiver module is further configured to: receive second indication information from the first data acquisition device, the second indication information being used to indicate the data status of the first data; the processing module is further configured to: determine the scheduling of the first data acquisition device based on the second indication information.

Based on the tenth aspect, in one possible implementation, the transceiver module is further configured to: receive third instruction information from the first model training device; the processing module is specifically configured to: determine the scheduling of the first data acquisition device based on the second instruction information and the third instruction information.

The eleventh aspect of this application provides a control device, the control device comprising:

The transceiver module is used to receive second instruction information from the first data acquisition device, the second instruction information being used to indicate the data status and first identifier of the first data, the first identifier being used to indicate that the first model training device is the model training device of the first data acquisition device; and to receive third instruction information from the first model training device, the third instruction information being used to indicate the training-related information and first identifier of the first model training device.

The processing module is used to determine the first model training device as the model training device of the first data acquisition device based on the second instruction information and the third instruction information.

Based on the eleventh aspect, in one possible implementation, the processing module is further configured to: determine the scheduling of the first data acquisition device based on the second instruction information and the third instruction information.

Based on the eleventh aspect, in one possible implementation, the transceiver module is further configured to: send scheduling information to the first data acquisition device, the scheduling information being used to schedule the first data acquisition device.

The twelfth aspect of this application provides a model fusion apparatus, the model fusion apparatus comprising:

The transceiver module is used to receive a first model from a first model training device, wherein the first model is obtained based on first data from a first data acquisition device.

The processing module is used to fuse the first model to obtain the second model.

Based on the twelfth aspect, in one possible implementation, the transceiver module is also used to: send the second model to the first data acquisition device.

Based on the twelfth aspect, in one possible implementation, the transceiver module is also used to: send the second model to the first model training device.

Regarding the seventh aspect mentioned above, the first data acquisition device may be a terminal device, or a component of a terminal device (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of the terminal device. The transceiver module may be a transceiver, or an input/output interface; the processing module may be a processor.

In one implementation, the first data acquisition device is a chip, chip system, or circuit configured in the terminal device. When the first data acquisition device is a chip, chip system, or circuit configured in the terminal device, the transceiver module may be an input/output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; the processing module may be a processor, processing circuit, or logic circuit.

Regarding the eighth aspect above, the model training management device can be a server, or a component of a server (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the server's functions. The transceiver module can be a transceiver, or an input/output interface; the processing module can be a processor.

In one implementation, the model training management device is a chip, chip system, or circuit configured in a server. When the model training management device is a chip, chip system, or circuit configured in a server, the transceiver module may be an input/output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; the processing module may be a processor, processing circuit, or logic circuit.

Regarding the ninth aspect mentioned above, the first model training device may be an AI server, or a component of an AI server (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of an AI server. The transceiver module may be a transceiver, or an input/output interface; the processing module may be a processor.

In one implementation, the first model training device is a chip, chip system, or circuit configured in an AI server. When the first model training device is a chip, chip system, or circuit configured in an AI server, the transceiver module may be an input/output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; the processing module may be a processor, processing circuit, or logic circuit.

Regarding the tenth or eleventh aspect above, the control device may be a network device, or a component of a network device (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of the network device. The transceiver module may be a transceiver, or an input/output interface; the processing module may be a processor.

In one implementation, the control device is a chip, chip system, or circuit configured in the network device. When the control device is a chip, chip system, or circuit configured in the network device, the transceiver module may be an input/output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; the processing module may be a processor, processing circuit, or logic circuit.

Regarding the twelfth aspect above, the model fusion apparatus may be a network device, or a component of a network device (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of a network device. The transceiver module may be a transceiver, or an input/output interface; the processing module may be a processor.

In one implementation, the model fusion device is a chip, chip system, or circuit configured in a network device. When the model fusion device is a chip, chip system, or circuit configured in a network device, the transceiver module may be an input/output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; the processing module may be a processor, processing circuit, or logic circuit.

The thirteenth aspect of this application provides an apparatus comprising a processor and a memory. The memory stores a computer program or computer instructions, and the processor is configured to invoke and execute the computer program or computer instructions stored in the memory, such that the processor implements any one of the implementations of the first to sixth aspects.

Optionally, the device may also include a transceiver, the processor of which controls the transceiver to transmit and receive signals.

The fourteenth aspect of this application provides an apparatus including a processor and an interface circuit, the processor being configured to communicate with other devices via the interface circuit and to perform the method described in any one of the first to sixth aspects. The processor may include one or more devices.

The fifteenth aspect of this application provides an apparatus including a processor connected to a memory for invoking a program stored in the memory to perform the method described in any one of the first to sixth aspects. The memory may be located within or outside the apparatus. The processor may include one or more processors.

In one implementation, the first data acquisition device shown in the first and seventh aspects above can be a chip or a chip system. The model training management device shown in the second and eighth aspects above can be a chip or a chip system. The first model training device shown in the third and ninth aspects above can be a chip or a chip system. The control device shown in the fourth, fifth, tenth, and eleventh aspects above can be a chip or a chip system. The model fusion device shown in the sixth and twelfth aspects above can be a chip or a chip system.

The sixteenth aspect of this application provides a computer program product including computer instructions, characterized in that, when run on a computer, it causes the computer to perform any of the implementations of any one of the first to sixth aspects.

The seventeenth aspect of this application provides a computer-readable storage medium including computer instructions that, when executed on a computer, cause the computer to perform any of the implementations of any one of the first to sixth aspects.

The eighteenth aspect of this application provides a chip device including a processor for calling a computer program or computer instructions in memory to cause the processor to execute any one of the implementations of the first to sixth aspects described above.

Optionally, the processor is coupled to the memory via an interface.

The nineteenth aspect of this application provides a communication system comprising a first data acquisition device as shown in the first aspect and a model training management device as shown in the second aspect. Optionally, the communication system further comprises a first model training device as described in the third aspect. Optionally, the communication system further comprises a control device as shown in the fourth or fifth aspect and/or a model fusion device as shown in the sixth aspect.

As described in the above technical solution, the first data acquisition device sends a first request to the model training management device. The first request requests the allocation of model training information for the first data acquisition device. Then, the first data acquisition device receives first instruction information from the model training management device. This first instruction information indicates the allocation of first model training information to the first data acquisition device. This enables the first data acquisition device to allocate first model training information, facilitating its determination of how to complete model training based on the first model training information, thus enabling model training based on the device's local data. For example, if the first data acquisition device has weak computing power or lacks computing power, the above technical solution helps solve the problem of the first data acquisition device being unable to perform model training. For example, the first data acquisition device can determine the first model training device based on the first model training information and send its local data to the first model training device. This facilitates the first model training device completing model training based on this local data. For example, if the first data acquisition device is a terminal device and the first model training device is a server of the terminal manufacturer, and the terminal device has limited computing power and data privacy protection requirements, the terminal manufacturer's server can complete model training based on the terminal device's local data. It can both ensure the data privacy protection needs of the primary data acquisition device and enable model training.

Attached Figure Description

Figure 1 is a schematic diagram of a communication system according to an embodiment of this application;

Figure 2 is a schematic diagram of federated learning in an embodiment of this application;

Figure 3 is a schematic diagram of segmentation learning in an embodiment of this application;

Figure 4 is a schematic diagram of decentralized learning in an embodiment of this application;

Figure 5 is a schematic diagram of an embodiment of the information transmission method of this application;

Figure 6A is a schematic diagram of a scenario of the information transmission method according to an embodiment of this application;

Figure 6B is a schematic diagram of another scenario of the information transmission method according to an embodiment of this application;

Figure 6C is a schematic diagram of another scenario of the information transmission method according to an embodiment of this application;

Figure 7 is a schematic diagram of another embodiment of the information transmission method of this application;

Figure 8 is a schematic diagram of another embodiment of the information transmission method of this application;

Figure 9 is a schematic diagram of another embodiment of the information transmission method of this application;

Figure 10 is a structural schematic diagram of a first data acquisition device according to an embodiment of this application;

Figure 11 is a schematic diagram of a model training management device according to an embodiment of this application;

Figure 12 is a schematic diagram of the structure of a first model training device according to an embodiment of this application;

Figure 13 is a schematic diagram of a control device according to an embodiment of this application;

Figure 14 is a schematic diagram of a model fusion device according to an embodiment of this application;

Figure 15 is a schematic diagram of a device according to an embodiment of this application;

Figure 16 is a structural schematic diagram of a terminal device according to an embodiment of this application;

Figure 17 is a schematic diagram of a network device according to an embodiment of this application.

Detailed Implementation

This application provides an information transmission method and related apparatus for allocating first model training information to a first data acquisition device. This facilitates the first data acquisition device in determining how to complete model training based on the first model training information, thereby enabling model training to be completed based on the local data of the first data acquisition device.

The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

References to "one embodiment" or "some embodiments" as described in this application mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

In the description of this application, unless otherwise stated, "/" means "or". For example, A/B can mean A or B. "And/or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and/or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent: a, b, c; a and b; a and c; b and c; or a and b and c. Where a, b, and c can be single or multiple.

It is understood that in this application, "instruction" can include direct instruction, indirect instruction, explicit instruction, and implicit instruction. When describing a certain instruction information to indicate A, it can be understood that the instruction information carries A, directly indicates A, or indirectly indicates A.

In this application, the information indicated by the instruction information is called the information to be instructed. In specific implementations, there are many ways to instruct the information to be instructed, such as, but not limited to, directly instructing the information to be instructed, such as the information to be instructed itself or its index; indirectly instructing the information to be instructed by instructing other information, where there is a relationship between the other information and the information to be instructed; or instructing only a part of the information to be instructed, while the other parts are known or pre-agreed upon. For example, the instruction of specific information can be achieved by using a pre-agreed (e.g., protocol-defined) arrangement of various pieces of information, thereby reducing instruction overhead to some extent.

The information to be instructed can be sent as a whole or divided into multiple sub-information messages, and the sending period and/or timing of these sub-information messages can be the same or different. This application does not limit the specific sending method. The sending period and/or timing of these sub-information messages can be predefined, for example, according to a protocol, or configured by the transmitting device by sending configuration information to the receiving device.

It is understood that "send" and "receive" in this application refer to the direction of signal transmission. For example, "send information to XX" can be understood as the destination of the information being XX, which can include direct transmission via the air interface or indirect transmission via the air interface from other units or modules. "Receive information from YY" can be understood as the source of the information being YY, which can include direct reception from YY via the air interface or indirect reception from YY via the air interface from other units or modules. "Send" can also be understood as the "output" of the chip interface, and "receive" can also be understood as the "input" of the chip interface.

In other words, sending and receiving can occur between devices, such as between network devices and terminal devices, or within a device, such as between components, modules, chips, software modules, or hardware modules within the device via buses, wiring, or interfaces.

It is understandable that information may undergo necessary processing, such as encoding and modulation, between the source and destination, but the destination can understand the valid information from the source. Similar statements in this application can be interpreted in a similar way and will not be elaborated further.

The technical solution of this application can be applied to cellular communication systems related to the 3rd Generation Partnership Project (3GPP). For example, 4th generation (4G) communication systems, 5th generation (5G) communication systems, and communication systems beyond the 5th generation, such as 6th generation (6G) communication systems. For example, 4th generation communication systems may include Long Term Evolution (LTE) communication systems. 5th generation communication systems may include New Radio (NR) communication systems. The technical solution of this application can also be applied to Wireless Fidelity (WiFi) systems, communication systems supporting the convergence of multiple wireless technologies, device-to-device (D2D) systems, or vehicle-to-everything (V2X) communication systems, etc.

The communication system to which the technical solution of this application applies includes a first data acquisition device, a first model training device, and a model training management device.

The first data acquisition device has a data acquisition function. For example, the first data acquisition device acquires local data. Then, the first data acquisition device can send the local data to the first model training device. This local data is used for model training. Optionally, the first data acquisition device can be a terminal device, or a component in the terminal device (e.g., a processor, chip, or chip system), or a logic module or software that can implement all or part of the functions of the terminal device.

The first model training device has a model training function. It receives local data from the first data acquisition device and trains the model based on the local data. Optionally, the first model training device can send the trained model to the first data acquisition device. Optionally, the first model training device can be an AI server, or a component within an AI server (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of an AI server.

Optionally, the first data acquisition device and the first model training device can jointly form a distributed node in distributed learning, but they are different physical devices. For example, as shown in Figure 1, the first data acquisition device can be a terminal device 101, and the first model training device can be an AI server 103. The first model training device can be understood as the terminal manufacturer's server. In other words, in distributed learning, the data collection (DC) module of the distributed node is deployed on the terminal device, and the model training (MT) module of the distributed node is deployed on the terminal manufacturer's server. This enables the terminal device to collect data through the data acquisition module and provide the data to the model training module. The terminal manufacturer's server then uses the model training module to train the model based on the data.

The model training management device is used to manage model training devices and allocate model training devices to data acquisition devices. For example, the model training management device can allocate a first model training device to a first data acquisition device. This enables the first model training device to perform model training using local data provided by the first data acquisition device. It also enables the first model training device to provide model training functionality to the first data acquisition device. Optionally, the model training management device can be a model training management server or a core network element, or a component within a model training management server or core network element (e.g., a processor, chip, or chip system), or a logic module or software capable of implementing all or part of the functions of a model training management server or core network element.

Optionally, the communication system also includes a model fusion device and/or a control device. In federated learning, the model fusion device can be a central node, which has the following functions: fusing the models sent by the model training device and sending the fused model to the data acquisition device and/or the model training device. The control device can be used to schedule the data acquisition device. Optionally, both the model fusion device and the control device can be network devices, or components within network devices (e.g., processors, chips, or chip systems), or logical modules or software capable of implementing all or part of the functions of network devices. For example, as shown in Figure 1, both the model fusion device and the control device are network devices 102. Optionally, the model fusion device and the control device can jointly form a central node in distributed learning; or, the model fusion device can be the central node in distributed learning, while the control device can be the control node in distributed learning. Optionally, the model fusion device and the control device can be two different physical devices.

Optionally, the communication system also includes a second data acquisition device and a second model training device. The function of the second data acquisition device is similar to that of the first data acquisition device; for details, please refer to the aforementioned description of the first data acquisition device, which will not be repeated here. The function of the second model training device is similar to that of the first model training device; for details, please refer to the aforementioned description of the first model training device, which will not be repeated here. For example, as shown in Figure 1, the second data acquisition device is a terminal device 104, and the second model training device is an AI server 105.

It should be noted that the above communication system is merely an example. In practical applications, the communication system includes a model training management device, at least one data acquisition device, and at least one model training device; the specific details are not limited in this application. That is, the communication system shown in Figure 1 is just an example. The communication system includes at least one terminal device and at least one AI server. Optionally, the communication system includes at least one network device.

In this application, the number of data acquisition devices and the number of model training devices in the communication system may be the same or different. Typically, the number of data acquisition devices exceeds the number of model training devices. For example, the data acquisition devices may be terminal equipment, and the model training devices may be server provided by the terminal manufacturer. The number of terminal equipment in the communication system may exceed the number of server provided by the terminal manufacturer.

In this application, the data acquisition device has a data acquisition function. The data acquisition device can also be called a data acquisition device, data collection device, data acquisition node, data collection device, or data acquisition node, etc., and this application does not limit the specific name of the data acquisition device. The model training device has a model training function. The model training device can also be called a local training device, training device, local training node, or training node, etc., and this application does not limit the specific name of the model training device. The control device can be used to schedule the data acquisition device. The model fusion device has the following functions: fusing the model sent by the model training device and sending the fused model to the data acquisition device and/or the model training device. The control device and the model fusion device can also be called a central device, central control node, central control device, or control center, etc., and this application does not limit the specific name of the control device.

The following describes the terminal equipment, network equipment, and AI server involved in this application.

The terminal device can be a wireless terminal device capable of receiving scheduling and instruction information from network devices. The wireless terminal device can be a device that provides voice and/or data connectivity to the user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem.

Terminal devices can communicate with one or more core networks or the Internet via an access network. Terminal devices can be mobile terminal devices, such as mobile phones (or "cellular" phones), computers, and data cards. For example, they can be portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile devices that exchange voice and/or data with the radio access network. Examples include personal communication service (PCS) phones, cordless phones, session initiation protocol phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), tablets, and computers with wireless transceiver capabilities. Wireless terminal equipment can also be referred to as a system, subscriber unit, subscriber station, mobile station, mobile station (MS), remote station, access point (AP), remote terminal, access terminal, user terminal, user agent, subscriber station (SS), customer premises equipment (CPE), terminal, user equipment (UE), mobile terminal (MT), etc.

By way of example and not limitation, in this embodiment, the terminal device can also be a wearable device. Wearable devices, also known as wearable smart devices or smart wearable devices, are a general term for devices that utilize wearable technology to intelligently design and develop everyday wearables. Examples include glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices that are worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not merely hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are feature-rich, large in size, and can achieve complete or partial functions without relying on a smartphone, such as smartwatches or smart glasses, as well as those that focus on a specific type of application function and require the use of other devices such as smartphones, such as various smart bracelets, smart helmets, and smart jewelry for vital sign monitoring.

Terminal devices can also be drones, robots, device-to-device (D2D) communication devices, vehicle-to-everything (V2X) devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless devices in industrial control, wireless devices in self-driving, wireless devices in remote medical care, wireless devices in smart grids, wireless devices in transportation safety, wireless devices in smart cities, and wireless devices in smart homes, etc.

Furthermore, terminal devices can also be terminal devices in communication systems evolved from fifth-generation (5G) communication systems (such as sixth-generation (6G) communication systems) or in future public land mobile networks (PLMNs). For example, 6G communication systems can further expand the form and function of 5G communication terminals; 6G terminals include, but are not limited to, vehicles, cellular network terminals (integrating satellite terminal functions), drones, or Internet of Things (IoT) devices.

In this embodiment, the terminal device has artificial intelligence (AI) capabilities. For example, the terminal device can obtain AI services provided by network devices or servers. The terminal device also has AI processing capabilities.

It should be noted that the terminal device may be a device or apparatus with a chip, or a device or apparatus with integrated circuitry, or a chip, module or control unit in the device or apparatus shown above. This application does not limit the specific device.

Network devices can be devices within a wireless network. For example, a network device can be an access network node (or access network equipment) that connects terminal devices to a wireless network, also known as a base station. Currently, some examples of access network equipment include: base stations (gNodeB, gNB), transmission reception points (TRP), evolved Node B (eNB), radio network controllers (RNC), Node Bs (NB), home base stations (e.g., home evolved Node B, or home Node B, HNB), base band units (BBU), or Wi-Fi access points (APs) in 5G communication systems. In another network architecture, network devices may include centralized unit (CU) nodes, distributed unit (DU) nodes, CU-control plane (CP), CU-user plane (UP), or radio unit (RU), or RAN equipment including CU and DU nodes. CU and DU may be separate entities or included in the same network element, such as a baseband unit (BBU). RU may be included in radio equipment or radio units, such as remote radio units (RRU), active antenna units (AAU), or remote radio heads (RRH). In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but their meanings will be understood by those skilled in the art. For example, in an Open RAN (ORAN) system, a CU can also be called an Open CU (O-CU), a DU can also be called an Open DU (O-DU), a CU-CP can also be called an Open CU-CP (O-CU-CP), a CU-UP can also be called an Open CU-UP (O-CU-UP), and a RU can also be called an Open RU (O-RU). Any of the units among the CU (or CU-CP, CU-UP), DU, and RU can be implemented through software modules, hardware modules, or a combination of software and hardware modules.

Network devices can be other devices that provide wireless communication functions for terminal devices. The embodiments of this application do not limit the specific technology or form of the network device. For ease of description, the embodiments of this application are not limited.

Network equipment may also include core network equipment, such as the Mobility Management Entity (MME), Home Subscriber Server (HSS), Serving Gateway (S-GW), Policy and Charging Rules Function (PCRF), and Public Data Network Gateway (PDN Gateway) in 4G networks; and access and mobility management function (AMF), user plane function (UPF), or session management function (SMF) in 5G networks. Furthermore, the core network equipment may also include other core network equipment in 5G networks and next-generation networks (e.g., 6G networks).

In this embodiment, the network device can also be a network node with AI capabilities, which can provide AI services to terminal devices or other network devices. For example, the network device can be an AI node, computing power node, access network node with AI capabilities, or core network element with AI capabilities on the network side (access network or core network).

It should be noted that the network device can be the device or apparatus shown above, or a component (e.g., a chip), module, or unit in the device or apparatus shown above; this application does not limit the specifics.

An AI server is a device equipped with AI capabilities. For example, an AI server can train models based on data and manage models.

Distributed learning can complete the learning tasks of AI models while fully ensuring user data privacy and security. Distributed learning mainly includes federated learning, partitioning learning, and decentralized learning. These three methods will be introduced below.

I. Federal Learning.

Federated learning is a typical distributed learning method that efficiently completes model learning tasks by facilitating collaboration between distributed nodes and a central node, while fully protecting user data privacy and security. As shown in Figure 2, the communication system includes a central node and one or more distributed nodes. For example, as shown in Figure 2, the communication system includes distributed node n, distributed node k, and distributed node m. Each distributed node collects its local dataset and performs local training on the model to obtain local parameters. Then, the distributed node sends its local parameters to the central node. The central node itself does not have a dataset; it collects the local parameters reported by multiple distributed nodes, fuses these parameters to obtain global parameters, and then distributes them to the distributed nodes. This enables the training and learning of the model.

II. Segmented learning.

In segmentation learning, the complete neural network model is divided into multiple parts. For example, taking a two-part model, the neural network model is divided into two sub-networks. One part is deployed on distributed nodes, and the other part is deployed on a central node. The part where the complete neural network model is segmented can be called a segmentation layer.

As shown in Figure 3, during forward inference of the model, distributed node 1 inputs local data into its local sub-network and infers it to the segmentation layer to obtain the segmentation layer's output result F1. Distributed node 1 sends this F1 to the central node via the communication link. The central node inputs the received F1 into another sub-network it has deployed and continues forward inference to obtain the final inference result. In gradient backpropagation during model training, the central node performs backpropagation to the segmentation layer through another sub-network it has deployed, obtaining the backpropagation result G1. Then, the central node sends G1 to distributed node 1. Distributed node 1 continues gradient backpropagation based on G1 through one of its deployed sub-networks. The forward inference and gradient backpropagation between other distributed nodes and the central node are similar.

As can be seen, the forward inference and backward gradient propagation processes of segmentation learning involve only one distributed node and one central node. The subnetworks trained on the distributed nodes can be stored locally on the distributed nodes or on a specific model storage server. When a new distributed node joins the segmentation learning process, it can download the trained subnetwork and then use its local data to further train that subnetwork.

3. Decentralized learning.

Unlike federated learning, decentralized learning is a learning method without a central node. As shown in Figure 4, the design objective f(x) of decentralized learning is generally the average of the local objective _fi (x) obtained by each node, i.e. Here, n is the number of distributed nodes, and x is the parameter to be optimized. In machine learning, x is the parameter of the machine learning model (such as a neural network). Each node uses its local data and local objective f _{_i} (x) to calculate its local gradient. and the local gradient The gradient is sent to its neighboring nodes. Upon receiving the local gradient from its neighboring nodes, any node can update the parameters x of its local model according to Formula 1 below. Thus, the model learning task is completed through information exchange between nodes.

Where N _{_i} is the set of neighboring nodes of node i, |N_{_i} | represents the number of elements in the set of neighboring nodes of node i, i.e. the number of neighboring nodes of node i, and α _{_k} is the fusion weight of the k-th training round. These are the model parameters obtained by node i in the (k+1)th round of training. These are the model parameters obtained by node i in the kth round of training. These are the model parameters obtained from neighbor node j in the k-th round of training. k is an integer greater than or equal to 1.

Currently, distributed nodes (e.g., terminal devices) possess sufficient computing power. Distributed nodes train models based on locally collected data to obtain local models, which are then transmitted to other nodes (e.g., central nodes or other distributed nodes). However, if a distributed node has limited computing power but requires data privacy protection, it cannot directly send local data to a central node or other distributed nodes with computing power, thus preventing model training. Therefore, how to achieve distributed learning to complete the model training task is a problem worth considering. This application provides a corresponding technical solution for allocating first model training information to a first data acquisition device. This facilitates the first data acquisition device in determining how to complete model training based on the first model training information, enabling model training based on the local data of the first data acquisition device. This solves the problem of being unable to perform model training when the first data acquisition device has weak or no computing power. Please refer to the following description of the embodiments for details.

The technical solution of this application is described below with reference to specific embodiments.

Figure 5 is a schematic diagram of an embodiment of the information transmission method of this application. Referring to Figure 5, the method includes:

501. The first data acquisition device sends a first request to the model training management device. Correspondingly, the model training management device receives the first request from the first data acquisition device.

The first request is used to request the allocation of model training information for the first data acquisition device. Optionally, the first request includes task information for the first data acquisition device. For example, the task information includes at least one of the following: task type, computing power required to perform the task, computing accuracy, or computing energy efficiency. Specifically, the first request can be used to request the allocation of model training equipment and/or model training resources for the first data acquisition device.

502. The model training management device sends a first instruction message to the first data acquisition device. Correspondingly, the first data acquisition device receives the first instruction message from the model training management device.

The first instruction information is used to indicate the first model training information allocated to the first data acquisition device. Optionally, the first model training information includes information about the first model training device and/or, first model training resources. The first model training resources include at least one of the following: first computing resources or first communication resources. Specifically, the model training management device allocates the first model training device and/or the first model training resources to the first data acquisition device according to the first request.

Optionally, the first indication information includes a first identifier. The first identifier is used to indicate that the first model training device is a model training device that serves as the first data acquisition device. Some possible forms of the first identifier are described below.

I. The first identifier is the identifier of the first data acquisition device.

In this implementation, the identifier of the first data acquisition device is assigned by the model training management device. The model training management device determines that the first data acquisition device and the first model training device are paired, and then sends the identifier of the first data acquisition device to both the first data acquisition device and the first model training device, so that the first model training device can know that it is paired with the first data acquisition device. When the first data acquisition device sends data, the identifier of the first data acquisition device can be carried in the data of the first data acquisition device. Forwarding nodes can forward the data carrying the identifier of the first data acquisition device to the first model training device according to the known pairing relationship (the method of knowing is not limited, it can be known through the model training management device); or, if the first data acquisition device sends its data in a broadcast manner, each model training device, after receiving the data, determines whether it is its paired data acquisition device based on the identifier of the first data acquisition device carried in it, and thus determines whether to perform model training based on the data. Therefore, the first model training device can perform model training based on the data of the first data acquisition device.

II. The first identifier is the identifier of the first model training device.

In this implementation, the identifier of the first model training device is assigned by the model training management device. The model training management device determines that the first data acquisition device and the first model training device are paired, and then sends the identifier of the first model training device to both the first data acquisition device and the first model training device. Thus, the first data acquisition device knows that it is paired with the first model training device. When the first data acquisition device sends data, the identifier of the first model training device can be carried within the data. The first model training device determines that the data comes from the paired first data acquisition device based on the identifier of the first model training device carried in the data (i.e., its own identifier), thus determining that model training needs to be performed based on this data. Therefore, the first model training device can perform model training based on the data from the first data acquisition device.

Third, the first identifier is a pairing identifier. This pairing identifier is used to indicate that the first data acquisition device is paired with the first model training device. Optionally, the pairing identifier is generated based on the identifier of the first data acquisition device and/or the identifier of the first model training device.

In this implementation, the model training management device determines that the first data acquisition device and the first model training device are paired, and then sends a pairing identifier to both the first data acquisition device and the first model training device. When the first data acquisition device sends data, the pairing identifier can be carried in the data of the first data acquisition device. The first model training device determines that the pairing identifier carried in the data is the pairing identifier between the first data acquisition device and the first model training device. Therefore, the first model training device can perform model training based on this data.

It should be noted that step 502 above describes the technical solution of this application by using the example of the model training management device allocating the first model training information to the first data acquisition device. In practical applications, the first model training information can also be pre-configured in the first data acquisition device, and this application does not limit this. For example, the terminal manufacturer may pre-configure the first model training information offline in the first data acquisition device.

Optionally, the embodiment shown in FIG5 further includes step 501a. Step 501a may be performed before step 502.

501a. The first model training device sends a registration request to the model training management device. Correspondingly, the model training management device receives the registration request from the first model training device.

The registration request is used to request the registration of the first model training device. Optionally, the registration request includes registration information of the first model training device. For example, the registration information includes at least one of the following: the computing power, computing power type, computing power margin, computing accuracy, or computing energy efficiency of the first model training device. Optionally, the computing power type includes at least one of the following: central processing unit (CPU), graphics processing unit (GPU), or neural network processing unit (NPU).

It should be noted that there is no fixed execution order between steps 501a and 501. Step 501a can be executed first, followed by step 501; or step 501 can be executed first, followed by step 501a; or, depending on the circumstances, steps 501 and 501a can be executed simultaneously. This application does not impose any specific restrictions on this.

Optionally, the embodiment shown in FIG5 further includes step 503. Step 503 may be performed after step 501.

503. The model training management device sends a first instruction message to the first model training device. Correspondingly, the first model training device receives the first instruction message from the model training management device.

For information on the first instruction, please refer to the relevant description in step 502 above, which will not be repeated here.

It should be noted that there is no fixed execution order between steps 502 and 503. Step 502 can be executed first, followed by step 503; or step 503 can be executed first, followed by step 502; or, depending on the circumstances, steps 502 and 503 can be executed simultaneously. This application does not impose any specific restrictions on this.

Optionally, the embodiment shown in FIG5 further includes step 504. Step 504 may be performed after step 501.

504. The model training management device sends a first instruction message to the control device. Correspondingly, the control device receives the first instruction message from the model training management device.

For information on the first instruction, please refer to the relevant description in step 502 above, which will not be repeated here.

It should be noted that, optionally, in federated learning, the control device can be a central node.

It should be noted that there is no fixed execution order between step 504 and steps 502 and 503. For example, step 504 can be executed first, then step 502, and finally step 503; or step 504 can be executed first, then step 502, and finally step 503; or step 502 can be executed first, then step 503, and finally step 504; or step 503 can be executed first, then step 502, and finally step 504. This application does not limit the specific execution order.

Optionally, the embodiment shown in FIG5 further includes steps 505 to 506. Steps 505 to 506 may be performed after step 503.

505. The first data acquisition device sends first data to the first model training device. The first data includes a first identifier. Correspondingly, the first model training device receives the first data from the first data acquisition device.

The first data is used for model training. In other words, the first data is model training data. The first data includes a first identifier, which instructs the first model training device to determine whether to train the model based on the first data according to the first identifier. For example, as shown in Figure 6A, the first data acquisition device is terminal device 1, and the first model training device is AI server 2. Terminal device 1 sends the first data to AI server 2.

506. The first model training device trains the model based on the first data to obtain the first model.

For example, the first identifier may be the identifier of a first data acquisition device. The first model training device determines that model training can be performed based on the first data based on the identifier of the first data acquisition device carried in the first data. As another example, the first identifier may be the identifier of a first model training device. The first model training device determines that model training can be performed based on the first data based on the identifier of the first model training device carried in the first data. As yet another example, the first identifier may be a pairing identifier. The first model training device determines that the pairing identifier carried in the first data is a pairing identifier between the first data acquisition device and the first model training device; therefore, the first model training device can perform model training based on the first data.

It should be noted that there is no fixed execution order between steps 505 to 506 and the aforementioned step 504. Steps 505 to 506 can be executed first, followed by step 504; or, step 504 can be executed first, followed by steps 506 to 506; or, depending on the circumstances, steps 504 and steps 505 to 506 can be executed simultaneously. This application does not impose any specific restrictions on this.

Optionally, the embodiment shown in FIG5 further includes step 505a. Step 505a may be performed before step 505.

505a. The control device sends scheduling information to the first data acquisition device. Correspondingly, the first data acquisition device receives the scheduling information from the control device.

The scheduling information is used to schedule the first data acquisition device to send the first data.

Optionally, the embodiment shown in FIG5 further includes steps 505b to 505c. Steps 505b to 505c may be performed before step 505a.

505b. The first data acquisition device sends a second instruction to the control device. Correspondingly, the control device receives the second instruction from the first data acquisition device.

The second indication information is used to indicate the data status of the first data.

Optionally, the data state of the first data includes at least one of the following: the data type, data distribution, number of samples included in the first data, or data attribute. Optionally, the data type includes at least one of the following: image data, voice data, text data, channel data, signal data, or radar data, etc. Optionally, the data distribution of the first data conforms to any one of the following: Gaussian distribution, exponential distribution, uniform distribution, or Poisson distribution, etc. Optionally, the data attribute includes at least one of the following: the time, location, or conditions of the first data acquisition.

505c. The control device determines the scheduling of the first data acquisition device based on the second instruction information.

For example, if the control device determines that the first data is helpful in improving model performance based on its type and distribution, it can choose to schedule the first data acquisition device. Therefore, the control device determines whether to schedule the first data acquisition device based on the second instruction information, which helps it schedule data acquisition devices with higher data quality to improve model training performance.

Optionally, the embodiment shown in Figure 5 further includes step 505d. Step 505d may be performed before step 505c.

505d. The first model training device sends a third instruction message to the control device. Correspondingly, the control device receives the third instruction message from the first model training device.

The third indication information is used to indicate training-related information of the first model training device. For example, the third indication information is used to indicate the training resources of the first model training device, such as computing power type, computing power margin, computing accuracy, and/or computing energy efficiency.

Optionally, step 505c specifically includes: the control device determining whether to schedule the first data acquisition device based on the second and third indication information. In this implementation, the control device further combines the third indication information to determine whether to schedule the first data acquisition device. This achieves better scheduling of the data acquisition device and improves the performance of model training.

It should be noted that the above example illustrates the technical solution of this application, where the control device determines the scheduling of the first data acquisition device based on the second instruction information. Optionally, the control device may further determine the scheduling of the first data acquisition device in conjunction with the third instruction information. In practice, the control device can determine the scheduling of the first data acquisition device based on the third instruction information, or optionally, the control device may further determine the scheduling of the first data acquisition device in conjunction with the second instruction information. This application does not impose any specific limitations on this.

Optionally, the embodiment shown in Figure 5 further includes step 507. Step 507 may be performed after step 506.

507. The first model training device sends the first model to the model fusion device. Correspondingly, the model fusion device receives the first model from the first model training device.

For example, as shown in Figure 6B, the first model training device is AI server 1, the model fusion device is network device, and AI server 1 sends the first model to network device.

Optionally, when the first model converges or the training epochs corresponding to the first model reach a preset threshold, the first model training device sends the first model to the model fusion device. The training epochs corresponding to the first model refer to the training epochs in which the first model training device trains the model based on the first data.

Optionally, the embodiment shown in FIG5 further includes step 508, which can be performed after step 507.

508. The model fusion device fuses the first model to obtain the second model.

For example, as shown in Figures 6B and 6C, the model fusion device is a network device that receives a first model from AI server 1. The network device also receives a fifth model from AI server 2. Then, the network device fuses the first and fifth models to obtain a second model.

Optionally, the embodiment shown in Figure 5 further includes step 509. Step 509 may be performed after step 508.

509. The model fusion device sends the second model to the first data acquisition device. Correspondingly, the first data acquisition device receives the second model from the model fusion device.

The second model is used for model inference. For example, as shown in Figure 6C, the model fusion device is a network device, and the first data acquisition device is terminal device 1. The network device sends the second model to terminal device 1. Terminal device 1 can perform model inference using the second model.

Optionally, the embodiment shown in Figure 5 further includes step 510. Step 510 may be performed after step 508.

510. The model fusion device sends the second model to the first model training device. Correspondingly, the first model training device receives the second model from the model fusion device.

The second model is used for model inference and/or model training.

It should be noted that there is no fixed execution order between steps 509 and 510. Step 509 can be executed first, followed by step 510; or step 510 can be executed first, followed by step 509; or, depending on the circumstances, steps 509 and 510 can be executed simultaneously. This application does not impose any specific restrictions on this.

In the embodiment shown in Figure 5 above, the first data acquisition device sends a first request to the model training management device. The first request requests the allocation of model training information for the first data acquisition device. Then, the first data acquisition device receives first instruction information from the model training management device. The first instruction information indicates the allocation of first model training information to the first data acquisition device. This enables the allocation of first model training information to the first data acquisition device. This facilitates the first data acquisition device in determining how to complete model training based on the first model training information, thereby enabling model training based on the local data of the first data acquisition device. For example, if the first data acquisition device has weak computing power or no computing power at all, the above technical solution helps solve the problem of the first data acquisition device being unable to perform model training. For example, the first data acquisition device can determine the first model training device based on the first model training information and send its local data to the first model training device. This facilitates the first model training device in completing model training based on the local data. For example, if the first data acquisition device is a terminal device and the first model training device is a server of the terminal manufacturer, and the terminal device has limited computing power and data privacy protection requirements, the terminal manufacturer's server can complete model training based on the local data of the terminal device. It can both ensure the data privacy protection needs of the primary data acquisition device and enable model training.

Figure 7 is a schematic diagram of another embodiment of the information transmission method of this application. Referring to Figure 7, the method includes:

701. The first data acquisition device sends a first request to the model training management device. Correspondingly, the model training management device receives the first request from the first data acquisition device.

702. The model training management device sends a first instruction message to the first data acquisition device. Correspondingly, the first data acquisition device receives the first instruction message from the model training management device.

Steps 701 to 702 are similar to steps 501 to 502 in the embodiment shown in Figure 5 above. For details, please refer to the relevant descriptions of steps 501 to 502 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG7 further includes step 701a. Step 701a may be performed before step 702.

701a. The first model training device sends a registration request to the model training management device. Correspondingly, the model training management device receives the registration request from the first model training device.

Step 701a is similar to step 501a in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 501a in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG7 further includes step 703. Step 703 may be performed after step 701.

703. The model training management device sends a first instruction message to the first model training device. Correspondingly, the first model training device receives the first instruction message from the model training management device.

Step 703 is similar to step 503 in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 503 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG7 further includes steps 704a to 704c. Steps 704a to 704c may be performed after step 703.

704a. The first data acquisition device sends a second indication message to the control device. The second indication message is used to indicate the first identifier. Correspondingly, the control device receives the second indication message from the first data acquisition device.

The second indication information is used to indicate the data status and the first identifier of the first data. The first identifier is used to indicate that the first model training device is a model training device that serves as the first data acquisition device. For details regarding the data status and the first identifier of the first data, please refer to the relevant description in the embodiment shown in Figure 5 above; it will not be repeated here.

704b. The first model training device sends a third instruction message to the control device. The third instruction message is used to indicate the first identifier. Correspondingly, the control device receives the third instruction message from the first model training device.

The third instruction information is used to indicate the training-related information and the first identifier of the first model training device. For details regarding the training-related information and the first identifier, please refer to the relevant descriptions in the embodiment shown in Figure 5 above; they will not be repeated here.

704c. The control device determines the first model training device as the model training device of the first data acquisition device based on the second instruction information and the third instruction information.

Specifically, the control device can determine that the first model training device is the model training device for the first data acquisition device based on the first identifier carried in the second instruction information and the first identifier carried in the third instruction information. Alternatively, the control device can determine that the first data acquisition device is paired with the first model training device based on the first identifier carried in the second instruction information and the first identifier carried in the third instruction information.

Optionally, the embodiment shown in FIG7 further includes steps 704 to 705. Steps 704 to 705 may be performed after step 703.

704. The first data acquisition device sends first data to the first model training device. The first data includes a first identifier. Correspondingly, the first model training device receives the first data from the first data acquisition device.

705. The first model training device trains the model based on the first data to obtain the first model.

Steps 704 to 705 are similar to steps 505 to 506 in the embodiment shown in Figure 5 above. For details, please refer to the relevant descriptions of steps 505 to 506 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG7 further includes steps 704d to 704e. Steps 704d to 704e may be performed after step 704c and before step 704.

704d. The control device determines the scheduling of the first data acquisition device based on the second and third instruction information.

Step 704d is similar to step 505c in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 505c in the embodiment shown in Figure 5 above, which will not be repeated here.

704e. The control device sends scheduling information to the first data acquisition device. Correspondingly, the first data acquisition device receives the scheduling information from the control device.

The scheduling information is used to schedule the first data acquisition device to send the first data.

Optionally, the embodiment shown in FIG7 further includes step 706. Step 706 may be performed after step 705.

706. The first model training device sends the first model to the model fusion device. Correspondingly, the model fusion device receives the first model from the first model training device.

Step 706 is similar to step 507 in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 507 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG7 further includes step 707. Step 707 may be performed after step 706.

707. The model fusion device fuses the first model to obtain the second model.

Step 707 is similar to step 508 in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 508 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG7 further includes step 708. Step 708 may be performed after step 707.

708. The model fusion device sends the second model to the first data acquisition device. Correspondingly, the first data acquisition device receives the second model from the model fusion device.

Step 708 is similar to step 509 in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 509 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG7 further includes step 709. Step 709 may be performed after step 707.

709. The model fusion device sends the second model to the first model training device. Correspondingly, the first model training device receives the second model from the model fusion device.

Step 709 is similar to step 510 in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 510 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, there is no fixed execution order between steps 708 and 709. For example, step 708 can be executed first, followed by step 709; or step 709 can be executed first, followed by step 708; or, depending on the circumstances, steps 708 and 709 can be executed simultaneously. This application does not impose any specific restrictions on this.

As shown in the embodiment of Figure 7 above, the first data acquisition device sends a first request to the model training management device. The first request is used to request the allocation of model training information for the first data acquisition device. Then, the first data acquisition device receives first instruction information from the model training management device. The first instruction information is used to indicate the allocation of first model training information for the first data acquisition device. This realizes the allocation of first model training information to the first data acquisition device. This facilitates the first data acquisition device to determine how to complete model training based on the first model training information, so as to realize the completion of model training based on the local data of the first data acquisition device. For example, when the computing power of the first data acquisition device is weak or the first data acquisition device does not have computing power, the above technical solution helps to solve the problem that the first data acquisition device cannot perform model training. For example, the first data acquisition device can determine the first model training device based on the first model training information and send the local data of the first data acquisition device to the first model training device. This facilitates the first model training device to complete model training based on the local data. For example, if the first data acquisition device is a terminal device and the first model training device is a server of the terminal manufacturer, and the terminal device has limited computing power and data privacy protection requirements, the server of the terminal manufacturer can complete model training based on the local data of the terminal device. It can both ensure the data privacy protection needs of the primary data acquisition device and enable model training.

It should be noted that the embodiments shown in Figures 5 and 7 above illustrate the technical solution of this application by taking the model fusion device as the central node in federated learning as an example. In practical applications, in segmentation learning, the model fusion device can be replaced by a model processing device, which can be the central node in segmentation learning. Step 506 in the embodiment shown in Figure 5 above or step 705 in the embodiment shown in Figure 7 above can be replaced by describing the following: The first model training device inputs the first data into the first model and pushes it to the segmentation layer to obtain the intermediate inference result. The first model is a sub-network of the complete neural network model. For information on the segmentation layer, please refer to the relevant introduction in the aforementioned segmentation learning. Step 507 in the embodiment shown in Figure 5 above or step 706 in the embodiment shown in Figure 7 above can be replaced by: The model processing device receives the intermediate inference result from the first model training device. Step 508 in the embodiment shown in Figure 5 above or step 707 in the embodiment shown in Figure 7 above can be replaced by: The model processing device performs inference, gradient calculation, and/or parameter update of another part of the model based on the intermediate inference result. The other part of the model is another sub-network of the neural network model, and the first model and the other part of the model constitute the neural network model. Step 509 in the embodiment shown in Figure 5 or step 708 in the embodiment shown in Figure 7 may be omitted, while step 510 in the embodiment shown in Figure 5 or step 709 in the embodiment shown in Figure 7 may be replaced by: the model processing device sending the gradient backpropagation result to the first model training device. The first model training device updates the first model based on the gradient backpropagation result. Optionally, the first model training device sends the updated first model to the first data acquisition device.

Figure 8 is a schematic diagram of another embodiment of the information transmission method of this application. Referring to Figure 8, the method includes:

801. The first data acquisition device sends a first request to the model training management device. Correspondingly, the model training management device receives the first request from the first data acquisition device.

802. The model training management device sends a first instruction message to the first data acquisition device. Correspondingly, the first data acquisition device receives the first instruction message from the model training management device.

Steps 801 to 802 are similar to steps 501 to 502 in the embodiment shown in Figure 5 above. For details, please refer to the relevant descriptions of steps 501 to 502 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG8 further includes step 801a. Step 801a may be performed before step 802.

801a. The first model training device sends a registration request to the model training management device. Correspondingly, the model training management device receives the registration request from the first model training device.

Step 801a is similar to step 501a in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 501a in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG8 further includes step 803. Step 803 may be performed after step 801.

803. The model training management device sends a first instruction message to the first model training device. Correspondingly, the first model training device receives the first instruction message from the model training management device.

Step 803 is similar to step 503 in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 503 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG8 further includes step 804. Step 804 may be performed after step 801.

804. The model training management device sends a first instruction message to the control device. Correspondingly, the control device receives the first instruction message from the model training management device.

Step 804 is similar to step 504 in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 504 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG8 further includes steps 805 to 806. Steps 805 to 806 may be performed after step 803.

805. The first data acquisition device sends first data to the first model training device. The first data includes a first identifier. Correspondingly, the first model training device receives the first data from the first data acquisition device.

806. The first model training device trains the model based on the first data to obtain the first model.

Steps 805 to 806 are similar to steps 505 to 506 in the embodiment shown in Figure 5 above. For details, please refer to the relevant descriptions of steps 505 to 506 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG8 further includes step 805a. Step 805a may be performed after step 804 and before step 805.

805a. The control device sends scheduling information to the first data acquisition device. Correspondingly, the first data acquisition device receives the scheduling information from the control device.

The scheduling information is used to schedule the first data acquisition device to send the first data.

Optionally, the embodiment shown in FIG8 further includes steps 805b to 805c. Steps 805b to 805c may be performed before step 805a.

805b. The first data acquisition device sends a second instruction to the control device. Correspondingly, the control device receives the second instruction from the first data acquisition device.

805c. The control device determines the scheduling of the first data acquisition device based on the second instruction information.

Steps 805b to 805c are similar to steps 505b to 505c in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of steps 505b to 505c in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in Figure 8 further includes step 805d. Step 805d may be performed before step 805c.

805d. The first model training device sends a third instruction message to the control device. Correspondingly, the control device receives the third instruction message from the first model training device.

Step 805d is similar to step 505d in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 505d in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG8 further includes step 807. Step 807 may be performed after step 806.

807. The first model training device sends the first model to the first data acquisition device. Correspondingly, the first data acquisition device receives the first model from the first model training device.

Specifically, for the first data acquisition device, the first model can be used for model inference. For the first model training device, the first model is used for model inference and/or model training. In decentralized learning, the first model training device trains the model based on the first data to obtain the first model and sends the first model to the first data acquisition device.

Optionally, the first model training device sends the first model to the first data acquisition device when the first model converges or when the training epochs corresponding to the first model reach a preset threshold. Instead of the first model training device sending the trained model to the first data acquisition device after each training iteration, the first model training device sends the trained model to the first data acquisition device.

Optionally, the embodiment shown in FIG8 further includes step 808. Step 808 may be performed after step 806.

808. The first model training device sends the first model to the second data acquisition device or the second model training device. Correspondingly, the second data acquisition device or the second model training device receives the first model from the first model training device.

In decentralized learning, the first model training device sends the first model to the second data acquisition device or the second model training device. The second data acquisition device can also send the first model to the second model training device, facilitating the fusion of the first model with its local model. Optionally, the second model training device sends the fused model to the second data acquisition device. The second model training device is the model training unit of the second data acquisition device, used for model training based on data provided by the second data acquisition device. For the second data acquisition device, the fused model can be used for model inference. For the second model training device, the fused model can be used for model inference and/or model training.

It should be noted that there is no fixed execution order between steps 807 and 808. For example, step 807 can be executed first, followed by step 808; or step 808 can be executed first, followed by step 807; or, depending on the circumstances, steps 807 and 808 can be executed simultaneously. This application does not impose any specific restrictions on this.

The above step 808 illustrates the technical solution of this application by taking the implementation of the first model training device sending the first model to the second data acquisition device or the second model training device as an example. In practical applications, the following solution is also possible: the first data acquisition device sends the first model to the second data acquisition device. In one possible implementation, the first data acquisition device sends the first model to the second model training device. In another possible implementation, the second data acquisition device sends the first model to the second model training device.

Optionally, the embodiment shown in FIG8 further includes steps 809 to 810. Steps 809 to 810 may be performed after step 806.

809. The second data acquisition device or the second model training device sends the third model to the first model training device. Correspondingly, the first model training device receives the third model from the second data acquisition device or the second model training device.

The above step 809 illustrates the technical solution of this application by using the implementation method of the first model training device acquiring the third model from the second data acquisition device or the second model training device as an example. In practical applications, the first data acquisition device may also acquire the third model from the second data acquisition device or the second model training device. Then, the first data acquisition device sends the third model to the first model training device; this application does not limit the specific implementation.

810. The first model training device merges the first model and the third model to obtain the fourth model.

In decentralized learning, the first model training device can receive a third model from a second data acquisition device or a second model training device. The first model training device merges the first model and the third model to obtain a fourth model. This is beneficial for improving model performance. For the first model training device, the fourth model can be used for model inference and/or model training.

Optionally, if the embodiment shown in FIG8 further includes step 807, steps 809 to 810 may be performed after step 807.

Optionally, if the embodiment shown in FIG8 further includes step 808, there is no fixed execution order between step 808 and steps 809 to 810. Step 808 can be executed first, followed by steps 809 to 810; or, steps 809 to 810 can be executed first, followed by step 808; or, depending on the situation, steps 808 and steps 809 to 810 can be executed simultaneously. This application does not limit the specific execution order.

Optionally, if the embodiment shown in FIG8 includes steps 807 and 808, steps 809 to 810 can be executed after step 807. There is no fixed execution order between steps 808 and steps 809 to 810.

Optionally, the embodiment shown in FIG8 further includes step 811. Step 811 may be performed after step 810.

811. The first model training device sends the fourth model to the first data acquisition device. Correspondingly, the first data acquisition device receives the fourth model from the first model training device.

For the first data acquisition device, the fourth model can be used for model inference.

As shown in Figure 8 above, in the embodiment, the first data acquisition device sends a first request to the model training management device. The first request requests the allocation of model training information for the first data acquisition device. Then, the first data acquisition device receives first instruction information from the model training management device. The first instruction information indicates the allocation of first model training information to the first data acquisition device. This enables the allocation of first model training information to the first data acquisition device. This facilitates the first data acquisition device in determining how to complete model training based on the first model training information, thereby enabling model training based on the local data of the first data acquisition device. For example, if the first data acquisition device has weak computing power or no computing power at all, the above technical solution helps solve the problem of the first data acquisition device being unable to perform model training. For example, the first data acquisition device can determine the first model training device based on the first model training information and send its local data to the first model training device. This facilitates the first model training device in completing model training based on the local data. For example, if the first data acquisition device is a terminal device and the first model training device is a server of the terminal manufacturer, and the terminal device has limited computing power and data privacy protection requirements, the terminal manufacturer's server can complete model training based on the local data of the terminal device. It can both ensure the data privacy protection needs of the primary data acquisition device and enable model training.

Figure 9 is a schematic diagram of another embodiment of the information transmission method of this application. Referring to Figure 9, the method includes:

901. The first data acquisition device sends a first request to the model training management device. Correspondingly, the model training management device receives the first request from the first data acquisition device.

902. The model training management device sends a first instruction message to the first data acquisition device. Correspondingly, the first data acquisition device receives the first instruction message from the model training management device.

Steps 901 to 902 are similar to steps 501 to 502 in the embodiment shown in Figure 5 above. For details, please refer to the relevant descriptions of steps 501 to 502 in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes step 901a. Step 901a may be performed before step 902.

901a. The first model training device sends a registration request to the model training management device. Correspondingly, the model training management device receives the registration request from the first model training device.

Step 901a is similar to step 501a in the embodiment shown in Figure 5 above. For details, please refer to the relevant description of step 501a in the embodiment shown in Figure 5 above, which will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes step 903. Step 903 may be performed after step 901.

903. The model training management device sends a first instruction message to the first model training device. Correspondingly, the first model training device receives the first instruction message from the model training management device.

Step 903 is similar to step 703 in the embodiment shown in Figure 7 above. For details, please refer to the relevant description of step 703 in the embodiment shown in Figure 7 above, which will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes steps 904a to 904c. Steps 904a to 904c may be performed after step 903.

904a. The first data acquisition device sends a second indication message to the control device. The second indication message is used to indicate the first identifier. Correspondingly, the control device receives the second indication message from the first data acquisition device.

904b. The first model training device sends a third instruction message to the control device. The third instruction message is used to indicate the first identifier. Correspondingly, the control device receives the third instruction message from the first model training device.

904c. The control device determines the first model training device as the model training device of the first data acquisition device based on the second instruction information and the third instruction information.

Steps 904a to 904c are similar to steps 704a to 704c in the embodiment shown in FIG7 above. For details, please refer to the relevant description of steps 704a to 704c in the embodiment shown in FIG7 above, which will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes steps 904 to 905. Steps 904 to 905 may be performed after step 903.

904. The first data acquisition device sends first data to the first model training device. The first data includes a first identifier. Correspondingly, the first model training device receives the first data from the first data acquisition device.

905. The first model training device trains the model based on the first data to obtain the first model.

Steps 904 to 905 are similar to steps 704 to 705 in the embodiment shown in Figure 7 above. For details, please refer to the relevant descriptions of steps 704 to 705 in the embodiment shown in Figure 7 above, which will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes step 904d. Step 904d may be performed after step 904c and before step 904.

904d. The control device sends scheduling information to the first data acquisition device. Correspondingly, the first data acquisition device receives the scheduling information from the control device.

The scheduling information is used to schedule the first data acquisition device to send the first data.

Optionally, the embodiment shown in FIG9 further includes step 906. Step 906 may be performed after step 905.

906. The first model training device sends the first model to the first data acquisition device. Correspondingly, the first data acquisition device receives the first model from the first model training device.

Step 906 is similar to step 807 in the embodiment shown in Figure 8 above. For details, please refer to the relevant description of step 807 in the embodiment shown in Figure 8. It will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes step 907. Step 907 may be performed after step 905.

907. The first model training device sends the first model to the second data acquisition device or the second model training device. Correspondingly, the second data acquisition device or the second model training device receives the first model from the first model training device.

Step 907 is similar to step 808 in the embodiment shown in Figure 8 above. For details, please refer to the relevant description of step 808 in the embodiment shown in Figure 8. It will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes steps 908 to 909. Steps 908 to 909 may be performed after step 905.

908. The second data acquisition device or the second model training device sends the third model to the first model training device. Correspondingly, the first model training device receives the third model from the second data acquisition device or the second model training device.

909. The first model training device merges the first model and the third model to obtain the fourth model.

Steps 908 to 909 are similar to steps 809 to 810 in the embodiment shown in Figure 8 above. For details, please refer to the relevant descriptions of steps 809 to 810 in the embodiment shown in Figure 8 above, which will not be repeated here.

Optionally, the embodiment shown in FIG9 further includes step 910. Step 910 may be performed after step 909.

910. The first model training device sends the fourth model to the first data acquisition device. Correspondingly, the first data acquisition device receives the fourth model from the first model training device.

For the first data acquisition device, the fourth model can be used for model inference.

As shown in Figure 9 above, in the embodiment, the first data acquisition device sends a first request to the model training management device. The first request requests the allocation of model training information for the first data acquisition device. Then, the first data acquisition device receives first instruction information from the model training management device. The first instruction information indicates the allocation of first model training information to the first data acquisition device. This enables the allocation of first model training information to the first data acquisition device. This facilitates the first data acquisition device in determining how to complete model training based on the first model training information, thereby enabling model training based on the local data of the first data acquisition device. For example, if the first data acquisition device has weak computing power or no computing power at all, the above technical solution helps solve the problem of the first data acquisition device being unable to perform model training. For example, the first data acquisition device can determine the first model training device based on the first model training information and send its local data to the first model training device. This facilitates the first model training device in completing model training based on the local data. For example, if the first data acquisition device is a terminal device and the first model training device is a server of the terminal manufacturer, and the terminal device has limited computing power and data privacy protection requirements, the terminal manufacturer's server can complete model training based on the local data of the terminal device. It can both ensure the data privacy protection needs of the primary data acquisition device and enable model training.

The first data acquisition device provided in the embodiments of this application is described below. Please refer to FIG10, which is a structural schematic diagram of the first data acquisition device in the embodiments of this application. The first data acquisition device 1000 can be used to execute the steps performed by the first data acquisition device in the embodiments shown in FIG5, FIG7 to FIG9. For details, please refer to the relevant description of the above method embodiments. The first data acquisition device 1000 includes a transceiver module 1001. Optionally, the first data acquisition device 1000 further includes a processing module 1002.

The processing module 1002 is used for data processing. The transceiver module 1001 can implement the corresponding communication functions. The transceiver module 1001 can also be called a communication interface or a communication module.

Optionally, the first data acquisition device 1000 may further include a storage module, which can be used to store program code, program instructions and/or data. The processing module 1002 can read the instructions and/or data in the storage module so that the first data acquisition device 1000 can implement the aforementioned method embodiment.

The first data acquisition device 1000 can be used to perform the actions performed by the first data acquisition device in the above method embodiment. The first data acquisition device 1000 can be a terminal device or a component configurable on a terminal device. The processing module 1002 is used to perform processing-related operations on the first data acquisition device side in the above method embodiment. The transceiver module 1001 is used to perform receiving-related operations on the first data acquisition device side in the above method embodiment.

Optionally, the transceiver module 1001 may include a sending module and a receiving module. The sending module is used to perform the sending operation in the above method embodiments. The receiving module is used to perform the receiving operation in the above method embodiments.

It should be noted that the first data acquisition device 1000 may include a transmitting module but not a receiving module. Alternatively, the first data acquisition device 1000 may include a receiving module but not a transmitting module. Specifically, it depends on whether the above-described scheme executed by the first data acquisition device 1000 includes both transmitting and receiving actions. For example, the first data acquisition device 1000 is used to execute the actions performed by the first data acquisition device in the embodiments shown in Figures 5, 7 to 9. For details, please refer to the relevant descriptions in the embodiments shown in Figures 5, 7 to 9, which will not be elaborated here. For example, the first data acquisition device 1000 is used to execute the following scheme:

The transceiver module 1001 is used to send a first request to the model training management device, the first request being used to request the allocation of model training information to the first data acquisition device 1000; and to receive first instruction information from the model training management device, the first instruction information being used to indicate the allocation of first model training information to the first data acquisition device 1000.

In one possible implementation, the first model training information includes: information about the first model training device, and/or, the first model training resources.

In another possible implementation, the first indication information includes a first identifier, which indicates that the first model training device is a model training device of the first data acquisition device 1000.

In another possible implementation, the first identifier is the identifier of the first data acquisition device 1000, or the identifier of the first model training device, or a pairing identifier, which is used to indicate that the first data acquisition device 1000 is paired with the first model training device.

In another possible implementation, the transceiver module 1001 is further configured to: receive a first model from a first model training device, wherein the first model is trained based on the first data; or receive a second model from a model fusion device, wherein the second model is obtained by fusing the first model reported by the first model training device, and the first model is trained based on the first data.

In another possible implementation, the transceiver module 1001 is further configured to: receive a third model from the second model training device or the second data acquisition device; and send the third model to the first model training device, wherein the third model is used for model fusion.

In another possible implementation, the transceiver module 1001 is also used to receive a fourth model from the first model training device, the fourth model being obtained by fusing the first model and the third model.

In another possible implementation, the transceiver module 1001 is also used to: receive scheduling information from the control device, the scheduling information being used to schedule the first data acquisition device 1000 to send the first data.

In another possible implementation, the transceiver module 1001 is further configured to: send a second indication message to the control device, the second indication message being used to indicate the data status of the first data.

In another possible implementation, the second indication information is also used to indicate the first identifier.

It should be understood that the specific procedures for each module to perform the above-mentioned corresponding processes have been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

The processing module 1002 in the above embodiments can be implemented by at least one processor or processor-related circuitry. The transceiver module 1001 can be implemented by a transceiver or transceiver-related circuitry. The transceiver module 1001 can also be referred to as a communication module or communication interface. The storage module can be implemented by at least one memory.

The model training management device provided in the embodiments of this application is described below. Please refer to FIG11, which is a structural schematic diagram of the model training management device according to an embodiment of this application. The model training management device 1100 can be used to execute the steps performed by the model training management device in the embodiments shown in FIG5, FIG7 to FIG9. For details, please refer to the relevant description of the above method embodiments. The model training management device 1100 includes a transceiver module 1101. Optionally, the model training management device 1100 also includes a processing module 1102.

The processing module 1102 is used for data processing. The transceiver module 1101 can implement the corresponding communication functions. The transceiver module 1101 can also be called a communication interface or a communication module.

Optionally, the model training management device 1100 may further include a storage module, which can be used to store program code, program instructions and/or data. The processing module 1102 can read the instructions and/or data in the storage module so that the model training management device 1100 can implement the aforementioned method embodiment.

The model training management device 1100 can be used to execute the actions performed by the model training management device in the above method embodiment. The model training management device 1100 can be a server or a component configurable on a server. The processing module 1102 is used to execute processing-related operations on the model training management device side in the above method embodiment. The transceiver module 1101 is used to execute receiving-related operations on the model training management device side in the above method embodiment.

Optionally, the transceiver module 1101 may include a sending module and a receiving module. The sending module is used to perform the sending operation in the above method embodiments. The receiving module is used to perform the receiving operation in the above method embodiments.

It should be noted that the model training management device 1100 may include a sending module but not a receiving module. Alternatively, the model training management device 1100 may include a receiving module but not a sending module. Specifically, it depends on whether the above-described scheme executed by the model training management device 1100 includes both sending and receiving actions. For example, the model training management device 1100 is used to execute the actions performed by the model training management device in the embodiments shown in Figures 5 and 7 to 9. For details, please refer to the relevant descriptions in the embodiments shown in Figures 5, 7 to 9, which will not be elaborated here. For example, the model training management device 1100 is used to execute the following scheme:

The transceiver module 1101 is used to receive a first request from the first data acquisition device, the first request being used to request the allocation of model training information to the first data acquisition device; and to send first instruction information to the first data acquisition device, the first instruction information being used to indicate the allocation of first model training information to the first data acquisition device.

In one possible implementation, the transceiver module 1101 is further configured to: send first instruction information to the first model training device.

In another possible implementation, the transceiver module 1101 is also used to: send first instruction information to the control device.

In another possible implementation, the first model training information includes information about the first model training device and/or the first model training resources.

In another possible implementation, the first indication information includes a first identifier, which is used to indicate that the first model training device is a model training device for the first data acquisition device.

In another possible implementation, the first identifier is the identifier of the first data acquisition device, or the identifier of the first model training device, or a pairing identifier, which is used to indicate that the first data acquisition device is paired with the first model training device.

It should be understood that the specific procedures for each module to perform the above-mentioned corresponding processes have been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

The processing module 1102 in the above embodiments can be implemented by at least one processor or processor-related circuitry. The transceiver module 1101 can be implemented by a transceiver or transceiver-related circuitry. The transceiver module 1101 can also be referred to as a communication module or communication interface. The storage module can be implemented by at least one memory.

The first model training apparatus provided in the embodiments of this application is described below. Please refer to FIG12, which is a structural schematic diagram of the first model training apparatus in the embodiments of this application. The first model training apparatus 1200 can be used to execute the steps performed by the first model training apparatus in the embodiments shown in FIG5, FIG7 to FIG9. For details, please refer to the relevant description of the above method embodiments. The first model training apparatus 1200 includes a transceiver module 1201 and a processing module 1202.

The processing module 1202 is used for data processing. The transceiver module 1201 can implement the corresponding communication functions. The transceiver module 1201 can also be called a communication interface or a communication module.

Optionally, the first model training device 1200 may further include a storage module, which can be used to store program code, program instructions and/or data. The processing module 1202 can read the instructions and/or data in the storage module so that the first model training device 1200 can implement the aforementioned method embodiment.

The first model training device 1200 can be used to execute the actions performed by the first model training device in the above method embodiment. The first model training device 1200 can be an AI server or a component configurable on an AI server. The processing module 1202 is used to execute processing-related operations on the first model training device side in the above method embodiment. The transceiver module 1201 is used to execute receiving-related operations on the first model training device side in the above method embodiment.

Optionally, the transceiver module 1201 may include a sending module and a receiving module. The sending module is used to perform the sending operation in the above method embodiments. The receiving module is used to perform the receiving operation in the above method embodiments.

It should be noted that the first model training device 1200 may include a sending module but not a receiving module. Alternatively, the first model training device 1200 may include a receiving module but not a sending module. Specifically, it depends on whether the above-described scheme executed by the first model training device 1200 includes both sending and receiving actions. For example, the first model training device 1200 is used to execute the actions performed by the first model training device in the embodiments shown in Figures 5 and 7 to 9. For details, please refer to the relevant descriptions in the embodiments shown in Figures 5, 7 to 9, which will not be elaborated here. For example, the first model training device 1200 is used to execute the following scheme:

The transceiver module 1201 is used to receive first instruction information from the model training management device, the first instruction information being used to indicate the first model training information allocated to the first data acquisition device;

The processing module 1202 is used to determine the first model training information based on the first instruction information.

In one possible implementation, the first model training information includes information about the first model training device 1200 and/or the first model training resources.

In another possible implementation, the first indication information includes a first identifier, which is used to identify the first model training device 1200 as a model training device for the first data acquisition device.

In another possible implementation, the first identifier is the identifier of the first data acquisition device, or the identifier of the first model training device 1200, or a pairing identifier, which is used to identify the pairing of the first data acquisition device and the first model training device 1200.

In another possible implementation, the transceiver module 1201 is further configured to: receive first data from the first data acquisition device, the first data including a first identifier; and the processing module 1202 is further configured to: perform model training based on the first identifier and the first data to obtain a first model.

In another possible implementation, the transceiver module 1201 is also used to: send the first model to the first data acquisition device or the model fusion device.

In another possible implementation, the transceiver module 1201 is also used to: receive a second model from the model fusion device, the second model being obtained by the model fusion device fusing the first model.

In another possible implementation, the transceiver module 1201 is also used to: receive a third model from the first data acquisition device, the second data acquisition device, or the second model training device, the third model being used for model fusion.

In another possible implementation, the processing module 1202 is further configured to: fuse the first model and the third model to obtain a fourth model; the transceiver module 1201 is further configured to: send the fourth model to the first data acquisition device.

In another possible implementation, the transceiver module 1201 is also used to: send third instruction information to the control device, the third instruction information being used to indicate training-related information of the first model training device 1200.

In another possible implementation, the third indication information is also used to indicate the first identifier.

It should be understood that the specific procedures for each module to perform the above-mentioned corresponding processes have been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

The processing module 1202 in the above embodiments can be implemented by at least one processor or processor-related circuitry. The transceiver module 1201 can be implemented by a transceiver or transceiver-related circuitry. The transceiver module 1201 can also be referred to as a communication module or communication interface. The storage module can be implemented by at least one memory.

The control device provided in the embodiments of this application is described below. Please refer to FIG13, which is a structural schematic diagram of the control device in the embodiments of this application. The control device 1300 can be used to execute the steps performed by the control device in the embodiments shown in FIG5, FIG7 to FIG9. For details, please refer to the relevant description of the above method embodiments. The control device 1300 includes a transceiver module 1301 and a processing module 1302.

The processing module 1302 is used for data processing. The transceiver module 1301 can implement the corresponding communication functions. The transceiver module 1301 can also be called a communication interface or a communication module.

Optionally, the control device 1300 may further include a storage module, which can be used to store program code, program instructions and/or data. The processing module 1302 can read the instructions and/or data in the storage module so that the control device 1300 can implement the aforementioned method embodiments.

The control device 1300 can be used to execute the actions performed by the control device in the above method embodiments. The control device 1300 can be a network device or a component configurable on a network device. The processing module 1302 is used to execute processing-related operations on the control device side in the above method embodiments. The transceiver module 1301 is used to execute receiving-related operations on the control device side in the above method embodiments.

Optionally, the transceiver module 1301 may include a sending module and a receiving module. The sending module is used to perform the sending operation in the above method embodiments. The receiving module is used to perform the receiving operation in the above method embodiments.

It should be noted that the control device 1300 may include a transmitting module but not a receiving module. Alternatively, the control device 1300 may include a receiving module but not a transmitting module. Specifically, it depends on whether the above-described scheme executed by the control device 1300 includes both transmitting and receiving actions. For example, the control device 1300 is used to execute the actions performed by the control device in the embodiments shown in Figures 5 and 7 to 9. For details, please refer to the relevant descriptions in the embodiments shown in Figures 5, 7 to 9; these will not be elaborated upon here.

For example, the control device 1300 is used to execute the following scheme:

The transceiver module 1301 is used to receive first instruction information from the model training management device, the first instruction information being used to indicate the first model training information allocated to the first data acquisition device;

The processing module 1302 is used to determine the first model training information based on the first instruction information.

For example, control device 1300 is used to execute the following scheme:

The transceiver module 1301 is used to receive second instruction information from the first data acquisition device, the second instruction information being used to indicate the data status and first identifier of the first data, the first identifier being used to indicate the first model training device as the model training device of the first data acquisition device; and to receive third instruction information from the first model training device, the third instruction information being used to indicate the training-related information and first identifier of the first model training device.

The processing module 1302 is used to determine the first model training device as the model training device of the first data acquisition device based on the second instruction information and the third instruction information.

For other implementation methods, please refer to the relevant descriptions in the foregoing method embodiments.

It should be understood that the specific procedures for each module to perform the above-mentioned corresponding processes have been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

The processing module 1302 in the above embodiments can be implemented by at least one processor or processor-related circuitry. The transceiver module 1301 can be implemented by a transceiver or transceiver-related circuitry. The transceiver module 1301 can also be referred to as a communication module or communication interface. The storage module can be implemented by at least one memory.

The model fusion apparatus provided in the embodiments of this application is described below. Please refer to FIG14, which is a schematic diagram of the structure of the model fusion apparatus in the embodiments of this application. The model fusion apparatus 1400 can be used to perform the steps performed by the model fusion apparatus in the embodiments shown in FIG5 and FIG7. For details, please refer to the relevant description of the above method embodiments. The model fusion apparatus 1400 includes a transceiver module 1401 and a processing module 1402.

The processing module 1402 is used for data processing. The transceiver module 1401 can implement the corresponding communication functions. The transceiver module 1401 can also be called a communication interface or a communication module.

Optionally, the model fusion apparatus 1400 may further include a storage module, which can be used to store program code, program instructions and/or data. The processing module 1402 can read the instructions and/or data in the storage module so that the model fusion apparatus 1400 can implement the aforementioned method embodiments.

The model fusion device 1400 can be used to perform the actions performed by the model fusion device in the above method embodiments. The control device 1400 can be a network device or a component configurable on a network device. The processing module 1402 is used to perform processing-related operations on the model fusion device side in the above method embodiments. The transceiver module 1401 is used to perform receiving-related operations on the model fusion device side in the above method embodiments.

Optionally, the transceiver module 1401 may include a sending module and a receiving module. The sending module is used to perform the sending operation in the above method embodiments. The receiving module is used to perform the receiving operation in the above method embodiments.

It should be noted that the model fusion device 1400 may include a transmitting module but not a receiving module. Alternatively, the model fusion device 1400 may include a receiving module but not a transmitting module. Specifically, it depends on whether the above-described scheme executed by the model fusion device 1400 includes both transmitting and receiving actions. For example, the model fusion device 1400 is used to execute the actions performed by the model fusion device in the embodiments shown in Figures 5 and 7. For details, please refer to the relevant descriptions in the embodiments shown in Figures 5 and 7, which will not be elaborated here. For example, the model fusion device 1400 can be used to execute the following scheme:

The transceiver module 1401 is used to receive a first model from the first model training device, the first model being obtained based on the first data from the first data acquisition device; the processing module 1402 is used to fuse the first model to obtain a second model.

In one possible implementation, the transceiver module 1401 is also used to: send the second model to the first data acquisition device.

In another possible implementation, the transceiver module 1401 is also used to send the second model to the first model training device.

It should be understood that the specific procedures for each module to perform the above-mentioned corresponding processes have been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

The processing module 1402 in the above embodiments can be implemented by at least one processor or processor-related circuitry. The transceiver module 1401 can be implemented by a transceiver or transceiver-related circuitry. The transceiver module 1401 can also be referred to as a communication module or communication interface. The storage module can be implemented by at least one memory.

This application embodiment also provides an apparatus 1500. Referring to FIG15, the apparatus 1500 includes a processor 1510, which is coupled to a memory 1520. The memory 1520 is used to store computer programs or instructions and/or data. The processor 1510 is used to execute the computer programs or instructions and/or data stored in the memory 1520, causing the methods in the above method embodiments to be executed. The apparatus 1500 is used to implement the operations performed by the first data acquisition device, the first model training device, the model training management device, the control device, or the model fusion device in the above method embodiments.

Optionally, the device 1500 may include one or more processors 1510.

Optionally, as shown in Figure 15, the device 1500 may also include a memory 1520.

Optionally, the device 1500 may include one or more memory 1520.

Optionally, the memory 1520 can be integrated with the processor 1510 or set separately.

Optionally, as shown in FIG15, the device 1500 may further include a transceiver 1530 for receiving and/or transmitting signals. For example, the processor 1510 is used to control the transceiver 1530 to receive and/or transmit signals.

This application also provides an apparatus 1600, which may be a terminal device, a processor in the terminal device, or a chip. The apparatus 1600 can be used to perform the operations performed by the first data acquisition device in the above method embodiments.

When device 1600 is a terminal device, Figure 16 shows a simplified schematic diagram of the terminal device. As shown in Figure 16, the terminal device includes a processor, a memory, and a transceiver. The memory can store computer program code, and the transceiver includes a transmitter 1631, a receiver 1632, radio frequency circuitry (not shown), an antenna 1633, and input/output devices (not shown).

The processor is mainly used to process communication protocols and communication data; control terminal devices; execute software programs; and process data from software programs.

Memory is mainly used to store software programs and data.

Radio frequency (RF) circuits are mainly used for the conversion between baseband signals and RF signals, as well as for the processing of RF signals.

Antennas are primarily used for transmitting and receiving radio frequency signals in the form of electromagnetic waves.

Input/output devices can include touchscreens, displays, or keyboards. They are primarily used to receive user input and output data to the user. It should be noted that some types of terminal devices may not have input/output devices.

When data needs to be transmitted, the processor performs baseband processing on the data to be transmitted and outputs a baseband signal to the radio frequency (RF) circuit. The RF circuit then processes the baseband signal and transmits it outwards via an antenna as electromagnetic waves. When data is sent to the terminal device, the RF circuit receives the RF signal through the antenna. The RF circuit converts the RF signal back into a baseband signal and outputs it to the processor. The processor converts the baseband signal back into data and processes the data. For ease of explanation, Figure 16 only shows one memory, processor, and transceiver. In actual terminal device products, there may be one or more processors and one or more memories. Memory can also be called storage medium or storage device, etc. Memory can be independent of the processor or integrated with the processor; this embodiment does not limit this.

In this embodiment, the antenna and radio frequency circuit with transceiver function can be regarded as the transceiver module of the terminal device, and the processor with processing function can be regarded as the processing module of the terminal device.

As shown in Figure 16, the terminal device includes a processor 1610, a memory 1620, and a transceiver 1630. The processor 1610 can also be referred to as a processing unit, processing board, processing module, or processing device, etc. The transceiver 1630 can also be referred to as a transceiver unit, transceiver, or transceiver device, etc.

Optionally, the device in transceiver 1630 used to implement the receiving function can be considered a receiving module, and the device in transceiver 1630 used to implement the transmitting function can be considered a transmitting module. That is, transceiver 1630 includes a receiver and a transmitter. A transceiver may also be called a transceiver unit, transceiver module, or transceiver circuit, etc. A receiver may also be called a receiver unit, receiving module, or receiving circuit, etc. A transmitter may also be called a transmitter, transmitting module, or transmitting circuit, etc.

The processor 1610 is used to execute the processing operations on the first data acquisition device side of the embodiments shown in Figures 5, 7 to 9. The transceiver 1630 is used to execute the transmission and reception operations on the first data acquisition device side of the embodiments shown in Figures 5, 7 to 9.

It should be understood that Figure 16 is merely an example and not a limitation, and the terminal device described above, including the transceiver module and the processing module, may not depend on the structure shown in Figure 10 or Figure 16.

When device 1600 is a chip, the chip includes a processor, a memory, and a transceiver. The transceiver can be an input/output circuit or a communication interface. The processor can be a processing module integrated on the chip, a microprocessor, or an integrated circuit. In the above method embodiments, the transmitting operation of the first data acquisition device can be understood as the chip's output, and the receiving operation of the first data acquisition device in the above method embodiments can be understood as the chip's input.

This application also provides a device 1700, which can be a network device or a chip. The device 1700 can be used to perform the operations performed by the control device or model fusion device in the embodiments shown in Figures 5, 7 to 9 above.

When device 1700 is a network device, such as a base station, Figure 17 shows a simplified schematic diagram of a base station structure. The base station includes parts 1710, 1720, and 1730.

The 1710 section is mainly used for baseband processing and controlling the base station; the 1710 section is usually the control center of the base station, which can be called the processor, and is used to control the base station to perform the processing operations on the control device or model fusion device side in the above method embodiments.

Part 1720 is primarily used to store computer program code and data.

Section 1730 is primarily used for transmitting and receiving radio frequency (RF) signals, as well as converting RF signals to baseband signals. Section 1730 is commonly referred to as a transceiver module, transceiver, transceiver circuit, or transceiver unit. The transceiver module of section 1730, also known as a transceiver or transceiver unit, includes antenna 1733 and RF circuitry (not shown in the figure), where the RF circuitry is mainly used for RF processing. Optionally, the device in section 1730 used for receiving can be considered a receiver, and the device used for transmitting can be considered a transmitter; that is, section 1730 includes receiver 1732 and transmitter 1731. The receiver can also be called a receiving module, receiver circuit, or receiving circuit, and the transmitter can be called a transmitting module, transmitter, or transmitting circuit.

Sections 1710 and 1720 may include one or more circuit boards, each of which may include one or more processors and one or more memories. The processors are used to read and execute programs from the memories to implement baseband processing functions and control the base station. If multiple circuit boards exist, they can be interconnected to enhance processing capabilities. As an alternative implementation, multiple circuit boards may share one or more processors, multiple circuit boards may share one or more memories, or multiple circuit boards may simultaneously share one or more processors.

For example, in one implementation, the transceiver module of section 1730 is used to execute the transceiver-related processes performed by the control device or model fusion device in the embodiments shown in Figures 5, 7 to 9. The processor of section 1710 is used to execute the processing-related processes performed by the control device or model fusion device in the embodiments shown in Figures 5, 7 to 9.

It should be understood that Figure 17 is merely an example and not a limitation, and the network device described above, including the processor, memory, and transceiver, may not depend on the structure shown in Figure 13 or Figure 17.

When device 1700 is a chip, the chip includes a transceiver, a memory, and a processor. The transceiver can be an input/output circuit or a communication interface; the processor can be an integrated processor, a microprocessor, or an integrated circuit on the chip. In the above method embodiments, the transmitting operation of the control device or model fusion device can be understood as the chip's output, and the receiving operation of the control device or model fusion device in the above method embodiments can be understood as the chip's input.

This application also provides a computer-readable storage medium storing computer instructions for implementing the methods executed by the first data acquisition device, the first model training device, the model training management device, the control device, or the model fusion device in the above-described method embodiments.

For example, when the computer program is executed by a computer, the computer can implement the method performed by the first data acquisition device, the first model training device, the model training management device, the control device, or the model fusion device in the above method embodiments.

This application also provides a computer program product containing instructions that, when executed by a computer, cause the computer to implement the method described above, which is performed by the first data acquisition device, the first model training device, the model training management device, the control device, or the model fusion device.

This application also provides a communication system, which includes a model training management device and a first data acquisition device. The model training management device is used to perform some or all of the operations performed by the model training management device in the embodiments shown in Figures 5, 7 to 9. The first data acquisition device is used to perform some or all of the operations performed by the first data acquisition device in the embodiments shown in Figures 5, 7 to 9.

Optionally, the communication system further includes a first model training device, which is used to perform some or all of the operations performed by the first model training device in the embodiments shown in Figures 5, 7 to 9.

Optionally, the communication system also includes a control device for performing some or all of the operations performed by the control device in the embodiments shown in Figures 5, 7 to 9.

Optionally, the communication system also includes a model fusion device, which performs some or all of the operations performed by the model fusion device in the embodiments shown in Figures 5 and 7.

This application also provides a chip device, including a processor, for calling computer programs or computer instructions stored in the memory, so that the processor executes the method provided in the embodiments shown in Figures 5, 7 to 9 above.

In one possible implementation, the input of the chip device corresponds to the receiving operation in any one of the embodiments shown in Figures 5, 7 to 9, and the output of the chip device corresponds to the sending operation in any one of the embodiments shown in Figures 5, 7 to 9.

Optionally, the processor is coupled to the memory via an interface.

Optionally, the chip device may also include a memory that stores computer programs or computer instructions.

The processor mentioned above can be a general-purpose central processing unit, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of a program for controlling the method provided in any of the embodiments shown in Figures 5, 7 to 9. The memory mentioned above can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, such as random access memory (RAM).

Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the explanations and beneficial effects of the relevant contents in any of the above-mentioned devices can be referred to the corresponding method embodiments provided above, and will not be repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

If the integrated unit is implemented as 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 essential contribution of the technical solution of this application, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims (30)

An information transmission method, characterized in that the method includes: The first data acquisition device sends a first request to the model training management device, the first request being used to request the allocation of model training information for the first data acquisition device; The first data acquisition device receives first instruction information from the model training management device, the first instruction information being used to indicate first model training information allocated to the first data acquisition device. The method according to claim 1, wherein the first model training information includes information about the first model training device and/or the first model training resources. The method according to claim 2, wherein the first indication information includes a first identifier, the first identifier being used to indicate that the first model training device is a model training device for the first data acquisition device. According to the method of claim 3, the first identifier is the identifier of the first data acquisition device, or the identifier of the first model training device, or a pairing identifier, wherein the pairing identifier is used to indicate that the first data acquisition device is paired with the first model training device. The method according to claim 3 or 4, characterized in that the method further comprises: The first data acquisition device sends first data to the first model training device. The first data includes the first identifier and is used for model training. The method according to claim 5, characterized in that the method further comprises: The first data acquisition device receives a first model from the first model training device, wherein the first model is trained based on the first data; or, the first data acquisition device receives a second model from the model fusion device, wherein the second model is obtained by fusing the first model reported by the first model training device, and the first model is trained based on the first data. The method according to claim 5 or 6, characterized in that, before the first data acquisition device sends the first data to the first model training device, the method further includes: The first data acquisition device receives scheduling information from the control device, the scheduling information being used to schedule the first data acquisition device to send the first data. The method according to claim 7, characterized in that, before the first data acquisition device receives scheduling information from the control device, the method further includes: The first data acquisition device sends a second instruction to the control device, the second instruction being used to indicate the data status of the first data. The method according to claim 8, wherein the second indication information is further used to indicate the first identifier. An information transmission method, characterized in that the method includes: The model training management device receives a first request from the first data acquisition device, the first request being used to request the allocation of model training information to the first data acquisition device; The model training management device sends a first instruction message to the first data acquisition device, the first instruction message being used to indicate the first model training information allocated to the first data acquisition device. The method according to claim 10, characterized in that the method further comprises: The model training management device sends the first instruction information to the first model training device. The method according to claim 10 or 11, characterized in that the method further comprises: The model training management device sends the first instruction information to the control device. The method according to any one of claims 10 to 12 is characterized in that the first model training information includes information about the first model training device and/or the first model training resources. An information transmission method, characterized in that the method includes: The first model training device receives first instruction information from the model training management device, the first instruction information being used to indicate first model training information allocated to the first data acquisition device; The first model training device determines the first model training information based on the first instruction information. The method according to claim 14, wherein the first model training information includes information about the first model training device and/or the first model training resources. The method according to claim 14 or 15 is characterized in that the first indication information includes a first identifier, the first identifier being used to identify the first model training device as a model training device of the first data acquisition device. According to the method of claim 16, the first identifier is an identifier of the first data acquisition device, or an identifier of the first model training device, or a pairing identifier, wherein the pairing identifier is used to identify that the first data acquisition device is paired with the first model training device. The method according to claim 16 or 17, characterized in that the method further comprises: The first model training device receives first data from the first data acquisition device, the first data including the first identifier; The first model training device trains the model based on the first data to obtain the first model. The method according to claim 18, characterized in that the method further comprises: The first model training device sends the first model to the first data acquisition device or the model fusion device. The method according to claim 19, characterized in that the method further comprises: The first model training device receives a second model from the model fusion device, the second model being obtained by fusing the first model with the model fusion device. The method according to any one of claims 18 to 20, characterized in that, before the first model training device receives the first data from the first data acquisition device, the method further includes: The first model training device sends a third instruction to the control device, the third instruction being used to indicate training-related information of the first model training device. The method according to claim 21, wherein the third indication information is further used to indicate the first identifier. An apparatus, characterized in that the apparatus includes a transceiver module, the transceiver module being configured to perform a transceiver operation of the method as described in any one of claims 1 to 9; or, the transceiver module being configured to perform a transceiver operation of the method as described in any one of claims 10 to 13. The apparatus according to claim 23, wherein the apparatus further comprises a processing module; If the transceiver module is used to perform the transceiver operation of the method as described in any one of claims 1 to 9, the processing module is used to perform the processing operation of the method as described in any one of claims 1 to 9; or, If the transceiver module is used to perform the transceiver operation of the method as described in any one of claims 10 to 13, the processing module is used to perform the processing operation of the method as described in any one of claims 10 to 13. An apparatus, characterized in that the apparatus comprises a transceiver module and a processing module; the transceiver module is used to perform a transceiver operation of the method as described in any one of claims 14 to 22, and the processing module is used to perform a processing operation of the method as described in any one of claims 14 to 22. An apparatus, characterized in that the apparatus includes a processor for executing a computer program or computer instructions in a memory to perform the method as described in any one of claims 1 to 22. The apparatus according to claim 26, wherein the apparatus further comprises the memory. The device according to claim 26 or 27 is characterized in that the device is a chip or a chip system. A computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed by a device, causes the device to perform the method as described in any one of claims 1 to 22. A computer program product, characterized in that the computer program product includes computer instructions that, when the computer program product is run on a computer, cause the computer to perform the method as described in any one of claims 1 to 22.