WO2025237119A1 - Communication method and related apparatus - Google Patents

Abstract

Embodiments of the present application disclose a communication method and a related apparatus. The method comprises: a transmitting terminal transmits first information to a receiving terminal, wherein the first information comprises one or more pieces of first data and one or more pieces of first training label data, the first data includes data outputted by the middle layers of a first model, and the first training label data is used for updating the first model; and the transmitting terminal receives second information transmitted by the receiving terminal, wherein the second information indicates model updating data reversely outputted by the middle layers of the first model. By means of the method, the transmitting terminal and the receiving terminal cooperatively train a neural network model, the method can be flexibly suitable for various different scenes and tasks, and the compatibility is improved. In addition, the first information transmitted by the transmitting terminal to the receiving terminal comprises one or more pieces of first data and one or more pieces of first training label data, and rich training-related data is transmitted to the receiving terminal, so that the generalization of a model is improved.

Description

A communication method and related apparatus

This application claims priority to Chinese Patent Application No. 202410585456.2, filed on May 11, 2024, entitled "A Communication Method and Related Device", the entire contents of which are incorporated herein by reference.

Technical Field

This application relates to the field of communication technology, and in particular to a communication method and related apparatus.

Background Technology

Wireless communication can be a transmission communication between two or more communication devices that does not propagate through conductors or cables. Generally, the two or more communication devices include network devices and terminal devices, or the two or more communication devices include different terminal devices.

With the development of artificial intelligence (AI) and communication technologies, AI models (which can also be simply referred to as models in this application embodiment) suitable for various scenarios or tasks have emerged. Currently, AI models are usually trained on one side of the network device or the terminal device, and then the trained model is sent to the other end for deployment.

However, due to limitations in terminal device capabilities or data privacy issues, one-sided training may not be feasible. Furthermore, models trained on one side may suffer from poor generalization. Therefore, how to achieve two-sided training of models has become a pressing problem to be solved.

Summary of the Invention

This application provides a communication method for collaboratively training a neural network model between a sending end and a receiving end. This method is flexible and applicable to various scenarios and tasks, thus improving compatibility.

In a first aspect, embodiments of this application propose a communication method applied to a sending end. The sending end can be a communication device (such as a terminal device or network device), or it can be a component within the communication device (such as a processor, chip, or chip system), or it can be a logic module or software capable of implementing all or part of the functions of the communication device. The method includes: first, the sending end sending first information to a receiving end, the first information including one or more first data and one or more first training label data, the first data including data output from the intermediate layer of a first model, the first training label data being used to update the first model; second, the sending end receiving second information sent by the receiving end, the second information indicating model update data output in reverse by the intermediate layer of the first model.

In this application, "model" may be replaced with other terms, such as neural network model, artificial intelligence (AI) model, AI network, AI neural network model, neural network, machine learning model, or AI processing model, etc.

It should be understood that the processing of updating the first model in the embodiments of this application includes, but is not limited to, one or more of the following: training, iteration, optimization, fine-tuning, tuning, or improvement.

It should be noted that the first model in the embodiments of this application may include one or more models. The intermediate layers of the first model may also be referred to as the hidden layers of the first model, which refers to a series of processing layers connecting the input layer and the output layer in the first model.

In this embodiment, the sending end sends first information to the receiving end. The first information includes one or more first data points and one or more first training label data points. The first data points include data output from the intermediate layers of the first model, and the first training label data is used to update the first model. Then, the sending end receives second information sent by the receiving end. The second information instructs the intermediate layers of the first model to output model update data in reverse order. This method enables collaborative training of the neural network model between the sending end and the receiving end, allowing for flexible application to various scenarios and tasks and improving compatibility. Furthermore, the first information sent by the sending end to the receiving end includes one or more first data points and one or more first training label data points. By sending rich training-related data to the receiving end, the generalization ability of the model is improved.

In one possible implementation, if one or more first data are obtained from a first training label, then the first information includes a first training label.

In another possible implementation, if one or more first data are obtained from multiple first training labels, then the first information includes multiple first training labels.

Optionally, the first information sent by the sending end to the receiving end may be first information obtained after data compression processing. In other words, the sending end compresses and sends the first information to the receiving end.

Optionally, the second information sent from the receiving end to the sending end may be second information obtained through data compression. In other words, the receiving end compresses and sends the second information to the sending end.

Optionally, the sending end can also directly send the first information, and the receiving end can also directly send the second information.

In one possible implementation of the first aspect, the first training label data includes: index information of the first training label in the training label set, or the first training label, and the training label set includes one or more training labels.

In the above technical solution, the first training label can be implemented in multiple ways, which improves the flexibility of the solution.

In one possible implementation of the first aspect, the first information further includes: grouping indication information, which indicates the grouping method of one or more first data included in the first information; and/or, group number information, which indicates the number of one or more first data included in the first information.

In the above technical solution, the first information can be realized through various grouping methods, thus effectively improving the generalization of the model.

In one possible implementation of the first aspect, the second information includes: a plurality of second data, wherein each second data corresponds to a first data; or, the second information includes: a weighted value of the plurality of second data, and the first information further includes the weight of the first data.

In the above technical solution, the second information can carry the second data in a variety of ways, thus effectively improving the generalization of the model.

In one possible implementation of the first aspect, the method further includes: the sending end acquiring a first training sample, the training label corresponding to the first training sample being a first training label; the sending end using the first training sample to process the first model multiple times, outputting one or more first data, wherein the processing method and/or processing parameters are different each time.

In one possible implementation of the first aspect, the method further includes:

The sending end acquires multiple first training samples, and the training label corresponding to the first training sample is the first training label.

The sending end uses multiple first training samples to process the first model once or multiple times, and outputs one or more first data, wherein the first training samples used in each processing are different.

In one possible implementation of the first aspect, the sending end acquires multiple first training samples, including:

The sending end obtains public training samples;

The sending end processes the common training samples to generate multiple first training samples, and the multiple first training samples are different from each other.

In the above technical solution, the sending end can obtain multiple first training samples in various ways, which improves the flexibility of the solution implementation.

In one possible implementation of the first aspect, the method further includes: the sending end sending third information to the receiving end, the third information including any one or more of the following:

The third data, the input location information of the third data in the first model, the dimension information of the third data, the identification information of the first model, or the first type indication information, the first type indication information is used to indicate the data type of the third data, wherein the third data includes any one or more of the following information: first channel information, first environment information, or multimodal information.

Optionally, the third information sent by the sending end to the receiving end may be third information obtained through data compression. In other words, the sending end compresses and sends the third information to the receiving end.

Alternatively, the sending end can also directly send this third information.

In one example, the third information includes: third data, input location information of the third data in the first model, dimension information of the third data, identification information of the first model, and first type indication information, wherein the third data includes: first channel information, first environment information, and multimodal information.

In the above technical solution, in addition to sending the first information to the receiver, the sending end can also send other information, which improves the generalization ability of the model and enhances the flexibility of implementation. Furthermore, by defining a unified interaction format for the data exchanged between the sending and receiving ends, the two ends can be flexibly trained for different training scenarios and tasks, improving compatibility.

In one possible implementation of the first aspect, the method further includes:

The sending end receives a fourth message sent by the receiving end, which includes any one or more of the following:

The fourth data, the second type indication information, the termination interaction instruction of the first model, the dimension information of the fourth data, or the identification information of the first model, the second type indication information is used to indicate the data type of the fourth data, and the termination interaction instruction of the first model is used to instruct the sender to stop sending information related to the first model. The fourth data includes any one or more of the following information: second channel information, second environment information, multimodal information, training residual information generated by the receiver during the training of the first model, or the performance information of the updated first model.

Optionally, the fourth information sent from the receiving end to the sending end may be fourth information obtained through data compression. In other words, the receiving end compresses and sends the fourth information to the sending end.

Alternatively, the receiving end can also directly send this fourth piece of information.

In one example, the fourth information includes: fourth data, second type indication information, termination interaction instruction of the first model, dimension information of the fourth data, and identification information of the first model, wherein the fourth data includes: second channel information, second environment information, multimodal information, training residual information generated by the receiver during the training of the first model, and performance information of the updated first model.

In the above technical solution, the receiving end can send other information besides the second information to the sending end, improving the model's generalization ability and implementation flexibility. Furthermore, by defining a unified interaction format for the data exchanged between the sending and receiving ends, the two ends can be flexibly trained for different training scenarios and tasks, improving compatibility.

In one possible implementation of the first aspect, the first model includes one or more splitting positions; the first data includes: data output by intermediate layers corresponding to one or more splitting positions in the first model; and/or, the second data includes: model update data output in reverse from one or more splitting positions in the first model.

In the above technical solution, since the first model can output feature data or update data at multiple segmentation positions, it enriches the amount of data exchanged between the sender and receiver and improves the generalization of the model.

In one possible implementation of the first aspect, the method further includes: the sending end sending configuration information to the receiving end, the configuration information being used to configure the training task for training the first model, the configuration information including any one or more of the following:

Model information, wherein the model information includes any one or more of the following: the identifier of the first model, the structural information of the first model, the training task information of the first model, the initialization parameters of the first model, or the parameters of the first model to be updated;

Segmentation information, wherein the segmentation information includes: the number of segmentation positions in the first model, and/or, the position information of the segmentation positions in the first model;

Alternatively, training information, which includes any one or more of the following: training label set information, validation set information, number of training iterations, learning rate, or loss function, wherein the training label set includes one or more training labels;

And/or, the sending end receives configuration information sent by the receiving end.

In the above technical solution, the sending end and the receiving end can support the adjustment of the trained model within a training cycle, thereby improving training efficiency.

In one possible implementation of the first aspect, the configuration information further includes: first compression configuration information, and/or, second compression configuration information, wherein: the first compression configuration information is used to indicate the compression method and compression parameters of the first information and/or the third information, and the second compression configuration information is used to indicate the compression method and compression parameters of the second information and/or the fourth information.

In the above technical solution, the data exchanged between the sending end and the receiving end supports multiple compression methods, which can effectively reduce the amount of data transmitted and improve communication efficiency.

In one possible implementation of the first aspect, the first information is carried in a radio resource control (RRC) message or a media access control (MAC) message.

And/or, the second information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

In one possible implementation of the first aspect, the third information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message;

And/or, the fourth information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

Secondly, embodiments of this application propose a communication method applied to a receiving end. The receiving end can be a communication device (such as a terminal device or network device), or it can be a component within the communication device (such as a processor, chip, or chip system), or it can be a logic module or software capable of implementing all or part of the functions of the communication device. The method includes: the receiving end receiving first information sent by a sending end, the first information including one or more first data items and one or more first training label data items, the first data including data output from the intermediate layer of a first model, the first training label data being used to update the first model; the receiving end sending second information to the sending end, the second information indicating model update data output in reverse from the intermediate layer of the first model.

In this application, "model" may be replaced with other terms, such as neural network model, artificial intelligence (AI) model, AI network, AI neural network model, neural network, machine learning model, or AI processing model, etc.

It should be understood that the processing of updating the first model in the embodiments of this application includes, but is not limited to, one or more of the following: training, iteration, optimization, fine-tuning, tuning, or improvement.

It should be noted that the first model in the embodiments of this application may include one or more models.

In this embodiment, the sending end sends first information to the receiving end. The first information includes one or more first data points and one or more first training label data points. The first data points include data output from the intermediate layers of the first model, and the first training label data is used to update the first model. Then, the sending end receives second information sent by the receiving end. The second information instructs the intermediate layers of the first model to output model update data in reverse order. This method enables collaborative training of the neural network model between the sending end and the receiving end, allowing for flexible application to various scenarios and tasks and improving compatibility. Furthermore, the first information sent by the sending end to the receiving end includes one or more first data points and one or more first training label data points. By sending rich training-related data to the receiving end, the generalization ability of the model is improved.

In one possible implementation of the second aspect, the first training label data includes: index information of the first training label in the training label set, or the first training label, and the training label set includes one or more training labels.

In the above technical solution, the first training label can be implemented in multiple ways, which improves the flexibility of the solution.

In one possible implementation of the second aspect, the first information further includes: grouping indication information, which indicates the grouping method of one or more first data included in the first information; and/or, group number information, which indicates the number of one or more first data included in the first information.

In the above technical solution, the first information can be realized through various grouping methods, thus effectively improving the generalization of the model.

In one possible implementation of the second aspect, the second information includes: a plurality of second data, wherein each second data corresponds to a first data; or, the second information includes: a weighted value of a plurality of second data, and the first information further includes the weight of the first data.

In the above technical solution, the second information can carry the second data in a variety of ways, thus effectively improving the generalization of the model.

In one possible implementation of the second aspect, the method further includes: the receiving end receiving third information sent by the sending end, the third information including any one or more of the following:

The third data, the input location information of the third data in the first model, the dimension information of the third data, the identification information of the first model, or the first type indication information, wherein the first type indication information is used to indicate the data type of the third data, wherein the third data includes any one or more of the following information: first channel information, first environment information, or multimodal information.

In the above technical solution, in addition to sending the first information to the receiver, the sending end can also send other information, which improves the generalization ability of the model and enhances the flexibility of implementation. Furthermore, by defining a unified interaction format for the data exchanged between the sending and receiving ends, the two ends can be flexibly trained for different training scenarios and tasks, improving compatibility.

In one possible implementation of the second aspect, the method further includes:

The receiving end sends a fourth piece of information to the sending end, which includes any one or more of the following:

The fourth data, the second type indication information, the termination interaction instruction of the first model, the dimension information of the fourth data, or the identification information of the first model, the second type indication information is used to indicate the data type of the fourth data, and the termination interaction instruction of the first model is used to instruct the sending end to stop sending information related to the first model, wherein the fourth data includes any one or more of the following information: second channel information, second environment information, multimodal information, training residual information generated by the receiving end during the training of the first model, or the performance information of the updated first model.

In the above technical solution, the receiving end can send other information besides the second information to the sending end, improving the model's generalization ability and implementation flexibility. Furthermore, by defining a unified interaction format for the data exchanged between the sending and receiving ends, the two ends can be flexibly trained for different training scenarios and tasks, improving compatibility.

In one possible implementation of the second aspect, the first model includes one or more segmentation positions; the first data includes: data output from intermediate layers corresponding to one or more segmentation positions in the first model;

And/or, the second data includes: model update data output in reverse from one or more split positions in the first model.

In the above technical solution, since the first model can output feature data or update data at multiple segmentation positions, it enriches the amount of data exchanged between the sender and receiver and improves the generalization of the model.

In one possible implementation of the second aspect, the method further includes: receiving configuration information from the receiver, the configuration information being used to configure the training task for training the first model, the configuration information including any one or more of the following:

Model information, wherein the model information includes any one or more of the following: the identifier of the first model, the structural information of the first model, the training task information of the first model, the initialization parameters of the first model, or the parameters of the first model to be updated;

Segmentation information, wherein the segmentation information includes: the number of segmentation positions in the first model, and/or, the position information of the segmentation positions in the first model;

Alternatively, training information, which includes any one or more of the following: training label set information, validation set information, number of training iterations, learning rate, or loss function, wherein the training label set includes one or more training labels;

And/or, the receiving end sends configuration information to the sending end.

In the above technical solution, the sending end and the receiving end can support the adjustment of the trained model within a training cycle, thereby improving training efficiency.

In one possible implementation of the second aspect, the configuration information also includes:

First compression configuration information, and/or, second compression configuration information, wherein:

The first compression configuration information is used to indicate the compression method and compression parameters of the first information and/or the third information.

The second compression configuration information is used to indicate the compression method and compression parameters of the second and/or fourth information.

In the above technical solution, the data exchanged between the sending end and the receiving end supports multiple compression methods, which can effectively reduce the amount of data transmitted and improve communication efficiency.

In one possible implementation of the second aspect, the first information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message.

And/or, the second information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

In one possible implementation of the second aspect, the third information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message.

And/or, the fourth information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

The third aspect of this application provides a communication device, which is a transmitting end or a receiving end. The device includes a transceiver unit and a processing unit. The constituent modules of the communication device can also be used to perform the steps performed in various possible implementations of the first aspect and achieve the corresponding technical effects. For details, please refer to the first aspect, which will not be repeated here.

The fourth aspect of this application provides a communication device, which is a transmitting end or a receiving end. The device includes a transceiver unit and a processing unit. The constituent modules of the communication device can also be used to perform the steps performed in various possible implementations of the second aspect and achieve the corresponding technical effects. For details, please refer to the second aspect, which will not be repeated here.

The fifth aspect of this application provides a communication device including at least one processor coupled to a memory; the memory is used to store a program or instructions; the at least one processor is used to execute the program or instructions to enable the device to implement the method described in any possible implementation of the first or second aspect.

The sixth aspect of this application provides a communication device including at least one logic circuit and an input/output interface; the logic circuit is used to perform the method described in any of the possible implementations of the first or second aspect described above.

The seventh aspect of this application provides a communication system that includes the aforementioned transmitting end or receiving end.

Optionally, the communication system may also include other communication devices that communicate with the transmitting end or the receiving end.

An eighth aspect of this application provides a computer-readable storage medium for storing one or more computer-executable instructions that, when executed by a processor, perform the method as described in any possible implementation of the first or second aspect above.

The ninth aspect of this application provides a computer program product (or computer program) that, when executed by a processor, performs the method described in any possible implementation of the first or second aspect described above.

The tenth aspect of this application provides a chip or chip system including at least one processor for supporting a communication device in implementing the method described in any possible implementation of the first or second aspect above.

In one possible design, the chip or chip system may further include a memory for storing program instructions and data necessary for the communication device. The chip system may be composed of chips or may include chips and other discrete devices. Optionally, the chip system may also include interface circuitry that provides program instructions and/or data to the at least one processor.

The technical effects of any of the design methods in aspects three through ten can be found in the technical effects of different design methods in aspects one or two above, and will not be repeated here.

Attached Figure Description

Figures 1a to 1c are schematic diagrams of the communication system provided in this application;

Figures 2a to 2e are schematic diagrams of the AI processing involved in this application;

Figure 3 is a schematic diagram of a communication scenario in an embodiment of this application;

Figure 4 is a schematic diagram of another communication scenario in the embodiments of this application;

Figure 5 is a flowchart illustrating one embodiment of the communication method in this application.

Figure 6 is a schematic diagram of the structure of the first information in an embodiment of this application;

Figure 7 is a schematic diagram of another structure of the first information in an embodiment of this application;

Figure 8 is a schematic diagram of a structure of the second information in an embodiment of this application;

Figure 9 is a schematic diagram of another structure of the second information in an embodiment of this application;

Figure 10 is a schematic diagram of the segmentation position of a first model in an embodiment of this application;

Figure 11 is a schematic diagram of the segmentation position of another first model in an embodiment of this application;

Figure 12 is a schematic diagram of another structure of the first model in the embodiments of this application;

Figure 13 is a schematic diagram of a forward interaction data structure in an embodiment of this application;

Figure 14 is a schematic diagram of another structure of forward interaction data in an embodiment of this application;

Figure 15 is a schematic diagram of a structure of the second information in an embodiment of this application;

Figure 16 is a schematic diagram of the structure of MAC message carrying forward interaction data in an embodiment of this application;

Figure 17 is a schematic diagram of the structure of MAC messages carrying reverse interaction data in an embodiment of this application;

Figure 18 is a structural schematic diagram of a communication device 1800 according to an embodiment of this application;

Figure 19 is another schematic structural diagram of the communication device 1900 provided in this application;

Figure 20 is a schematic diagram of the structure of the communication device 2000 provided in an embodiment of this application;

Figure 21 is a schematic diagram of the structure of the communication device 2100 provided in an embodiment of this application.

Detailed Implementation

First, some terms used in the embodiments of this application will be explained to facilitate understanding by those skilled in the art.

(1) Terminal device: can be a wireless terminal device that can receive network device scheduling and instruction information. The wireless terminal device can be a device that provides voice and/or data connectivity to the user, or a handheld device with wireless connection function, or other processing device connected to a wireless modem.

Terminal devices can communicate with one or more core networks or the Internet via a radio access network (RAN). 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 RAN. Examples include personal communication service (PCS) phones, cordless phones, session initiation protocol (SIP) 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, such as 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.

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

Furthermore, terminal devices can also be terminal devices in communication systems evolved from fifth-generation (5G) communication systems (such as 5G Advanced or sixth-generation (6G) communication systems), or terminal devices in future public land mobile networks (PLMNs). For example, 5G Advanced or 6G networks 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, and Internet of Things (IoT) devices.

In this embodiment, the terminal device can also obtain artificial intelligence (AI) services provided by the network device. Optionally, the terminal device can also have AI processing capabilities.

(2) Network equipment: This can be equipment in a wireless network. For example, network equipment can be a RAN node (or device) that connects terminal devices to the wireless network, and can also be called a base station. Currently, some examples of RAN equipment include: base station, evolved NodeB (eNodeB), gNB (gNodeB) in 5G communication systems, transmission reception point (TRP), evolved Node B (eNB), radio network controller (RNC), Node B (NB), home base station (e.g., home evolved Node B, or home Node B, HNB), base band unit (BBU), or wireless fidelity (Wi-Fi) access point (AP), etc. In addition, in a network architecture, network devices may include centralized unit (CU) nodes, distributed unit (DU) nodes, or RAN devices that include both CU and DU nodes.

Optionally, RAN nodes can also be macro base stations, micro base stations or indoor stations, relay nodes or donor nodes, or radio controllers in cloud radio access network (CRAN) scenarios. RAN nodes can also be servers, wearable devices, vehicles, or in-vehicle equipment. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU).

In another possible scenario, multiple RAN nodes collaborate to assist the terminal in achieving wireless access, with different RAN nodes each implementing some of the base station's functions. For example, RAN nodes can be CUs, DUs, CUs (control plane, CP), CUs (user plane, UP), or radio units (RUs). CUs and DUs can be set up separately or included in the same network element, such as a baseband unit (BBU). RUs can be included in radio frequency equipment or radio frequency units, such as remote radio units (RRUs), active antenna units (AAUs), radio heads (RHs), or remote radio heads (RRHs).

In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an open access network (open RAN, O-RAN, or ORAN) system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. For ease of description, this application uses CU, CU-CP, CU-UP, DU, and RU as examples. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software modules and hardware modules.

Communication between access network devices and terminal devices follows a specific protocol layer structure. This protocol layer may include a control plane protocol layer and a user plane protocol layer. The control plane protocol layer may include at least one of the following: radio resource control (RRC) layer, packet data convergence protocol (PDCP) layer, radio link control (RLC) layer, media access control (MAC) layer, or physical (PHY) layer, etc. The user plane protocol layer may include at least one of the following: service data adaptation protocol (SDAP) layer, PDCP layer, RLC layer, MAC layer, or physical layer, etc.

The correspondence between network elements and their achievable protocol layer functions in the ORAN system can be found in Table 1 below.

Table 1

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, this core network equipment may also include other core network equipment in 5G networks and next-generation networks of 5G networks.

In this embodiment of the application, the network device may also have network nodes with AI capabilities, which can provide AI services to terminals or other network devices. For example, it may be an AI node, computing node, RAN node with AI capabilities, or core network element with AI capabilities on the network side (access network or core network).

In this application embodiment, the device for implementing the function of the network device can be the network device itself, or it can be a device capable of supporting the network device in implementing that function, such as a chip system, which can be installed in the network device. In the technical solutions provided in this application embodiment, the example of a network device being used to implement the function of the network device is used to describe the technical solutions provided in this application embodiment.

(3) Configuration and Pre-configuration: In this application, both configuration and pre-configuration are used. Configuration refers to the network device/server sending configuration information or parameter values to the terminal via messages or signaling, so that the terminal can determine communication parameters or resources for transmission based on these values or information. Pre-configuration is similar to configuration; it can be parameter information or parameter values pre-negotiated between the network device/server and the terminal device, parameter information or parameter values specified by standard protocols for use by the base station/network device or terminal device, or parameter information or parameter values pre-stored in the base station/server or terminal device. This application does not limit this.

Furthermore, these values and parameters can be changed or updated.

(4) The terms "system" and "network" in the embodiments of this application can be used interchangeably. "Multiple" refers to two or more. "And/or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and/or B can mean: A exists alone, A and B exist simultaneously, or B exists alone, where A and B can be singular or plural. The character "/" generally indicates that the related objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, "at least one of A, B and C" includes A, B, C, AB, AC, BC or ABC. And, unless otherwise specified, the ordinal numbers such as "first" and "second" mentioned in the embodiments of this application are used to distinguish multiple objects and are not used to limit the order, sequence, priority or importance of multiple objects.

(5) In the embodiments of this application, "send" and "receive" indicate the direction of signal transmission. For example, "send information to XX" can be understood as the destination of the information being XX, which may include sending directly through the air interface or sending indirectly through the air interface by other units or modules. "Receive information from YY" can be understood as the source of the information being YY, which may include receiving directly from YY through the air interface or receiving indirectly from YY through the air interface by 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.

(6) In the embodiments of this application, "instruction" may include direct instruction and indirect instruction, as well as explicit instruction and implicit instruction. The information indicated by a certain piece of information (hereinafter referred to as instruction information) is called the information to be instructed. In the specific implementation process, there are many ways to indicate the information to be instructed, such as, but not limited to, directly indicating the information to be instructed, such as the information to be instructed itself or its index. It can also indirectly indicate the information to be instructed by indicating other information, where there is an association between the other information and the information to be instructed; or it can only indicate a part of the information to be instructed, while the other parts of the information to be instructed are known or pre-agreed upon. For example, the instruction of specific information can be achieved by using a pre-agreed (e.g., protocol predefined) arrangement order of various information, thereby reducing the instruction overhead to a certain extent. This application does not limit the specific method of instruction. It is understood that for the sender of the instruction information, the instruction information can be used to indicate the information to be instructed, and for the receiver of the instruction information, the instruction information can be used to determine the information to be instructed.

In this application, unless otherwise specified, the same or similar parts between the various embodiments can be referred to each other. In the various embodiments of this application, and the various methods/designs/implementations within each embodiment, unless otherwise specified or logically conflicting, the terminology and/or descriptions between different embodiments and between the various methods/designs/implementations within each embodiment are consistent and can be mutually referenced. The technical features in different embodiments and the various methods/designs/implementations within each embodiment can be combined to form new embodiments, methods, or implementations based on their inherent logical relationships. The following embodiments of this application do not constitute a limitation on the scope of protection of this application.

This application can be applied to long-term evolution (LTE) systems, new radio (NR) systems, or communication systems evolving after 5G (such as 6G). The communication system includes at least one network device and/or at least one terminal device.

Please refer to Figure 1a, which is a schematic diagram of a communication system according to this application. Figure 1a exemplarily shows one network device and six terminal devices, namely terminal device 1, terminal device 2, terminal device 3, terminal device 4, terminal device 5, and terminal device 6. In the example shown in Figure 1a, terminal device 1 is a smart teacup, terminal device 2 is a smart air conditioner, terminal device 3 is a smart gas pump, terminal device 4 is a vehicle, terminal device 5 is a mobile phone, and terminal device 6 is a printer.

As shown in Figure 1a, the entity sending the AI configuration information can be a network device. The entity receiving the AI configuration information can be terminal devices 1-6. In this case, the network device and terminal devices 1-6 form a communication system. In this communication system, terminal devices 1-6 can send data to the network device, and the network device needs to receive the data sent by terminal devices 1-6. At the same time, the network device can send configuration information to terminal devices 1-6.

For example, in Figure 1a, terminal devices 4 and 6 can also form a communication system. Terminal device 5 acts as a network device, i.e., the entity sending AI configuration information; terminal devices 4 and 6 act as terminal devices, i.e., the entities receiving AI configuration information. For instance, in a vehicle-to-everything (V2X) system, terminal device 5 sends AI configuration information to terminal devices 4 and 6 respectively, and receives data sent by terminal devices 4 and 6; correspondingly, terminal devices 4 and 6 receive the AI configuration information sent by terminal device 5 and send data back to terminal device 5.

Taking the communication system shown in Figure 1a as an example, in addition to performing communication-related services, different devices (including network devices and network devices, network devices and terminal devices, and/or terminal devices and terminal devices) may also perform AI-related services.

As shown in Figure 1b, taking a network device as a base station as an example, the base station can perform communication-related services and AI-related services with one or more terminal devices, and different terminal devices can also perform communication-related services and AI-related services.

As shown in Figure 1c, taking terminal devices including televisions and mobile phones as an example, communication-related services and AI-related services can also be performed between televisions and mobile phones.

The technical solutions provided in this application can be applied to wireless communication systems (such as the systems shown in Figures 1a, 1b, or 1c). For example, AI network elements can be introduced into the communication system provided in this application to implement some or all AI-related operations. AI network elements can also be called AI nodes, AI devices, AI entities, AI modules, AI models, or AI units, etc. AI network elements can be built into network elements within the communication system. For example, an AI network element can be an AI module built into: access network equipment, core network equipment, cloud servers, or operation, administration, and maintenance (OAM) systems to implement AI-related functions. OAM can act as the network management system for core network equipment and/or access network equipment. Alternatively, AI network elements can also be independently configured network elements within the communication system. Optionally, the terminal or its built-in chip can also include an AI entity to implement AI-related functions.

The following is a brief introduction to the AI that may be involved in this application.

AI can endow machines with human-like intelligence, for example, allowing them to use computer hardware and software to simulate certain intelligent human behaviors. To achieve artificial intelligence, machine learning methods can be employed. In machine learning, machines learn (or train) a model using training data. This model represents the mapping between inputs and outputs. The learned model can be used for reasoning (or prediction), that is, it can be used to predict the output corresponding to a given input. This output can also be called the reasoning result (or prediction result).

Machine learning can include supervised learning, unsupervised learning, and reinforcement learning. Unsupervised learning can also be called learning without supervision.

Supervised learning, based on collected sample values and labels, uses machine learning algorithms to learn the mapping relationship between sample values and labels, and then expresses this learned mapping relationship using an AI model. The process of training the machine learning model is the process of learning this mapping relationship. During training, sample values are input into the model to obtain the model's predicted values, and the model parameters are optimized by calculating the error between the model's predicted values and the sample labels (ideal values). After the mapping relationship is learned, it can be used to predict new sample labels. The mapping relationship learned in supervised learning can include linear or non-linear mappings. Based on the type of label, the learning task can be divided into classification tasks and regression tasks.

Unsupervised learning relies on collected sample values to discover inherent patterns within the samples themselves. One type of unsupervised learning algorithm uses the samples themselves as supervisory signals, meaning the model learns the mapping relationship from sample to sample; this is called self-supervised learning. During training, model parameters are optimized by calculating the error between the model's predictions and the samples themselves. Self-supervised learning can be used for signal compression and decompression recovery applications; common algorithms include autoencoders and generative adversarial networks.

Reinforcement learning, unlike supervised learning, is a type of algorithm that learns problem-solving strategies through interaction with the environment. Unlike supervised and unsupervised learning, reinforcement learning problems do not have explicit "correct" action labels. The algorithm needs to interact with the environment to obtain reward signals from the environment, and then adjust its decision actions to obtain a larger reward signal value. For example, in downlink power control, the reinforcement learning model adjusts the downlink transmission power of each user based on the total system throughput feedback from the wireless network, aiming to achieve a higher system throughput. The goal of reinforcement learning is also to learn the mapping relationship between the environment state and a better (e.g., optimal) decision action. However, because the label of the "correct action" cannot be obtained in advance, the network cannot be optimized by calculating the error between the action and the "correct action." Reinforcement learning training is achieved through iterative interaction with the environment.

Neural networks (NNs) are a specific model in machine learning techniques. According to the general approximation theorem, neural networks can theoretically approximate any continuous function, thus enabling them to learn arbitrary mappings. Traditional communication systems rely on extensive expert knowledge to design communication modules, while deep learning communication systems based on neural networks can automatically discover hidden pattern structures from large datasets, establish mapping relationships between data, and achieve performance superior to traditional modeling methods.

The idea behind neural networks comes from the neuronal structure of the brain. For example, each neuron performs a weighted summation of its input values and outputs the result through an activation function.

Figure 2a shows a schematic diagram of a neuron structure. Assume the neuron's input is x = [ _x0 , _x1 , ..., _xn ], and the corresponding weights for each input are w = [w, _w1 , ..., _wn ], where n is a positive integer, and w_{_i} and _xi can be decimals, integers (e.g., 0, positive integers, or negative integers), or complex numbers, etc. _{w_i} is used as the weight for _xi , and is used to weight _xi . The bias for the weighted summation of the input values based on the weights is, for example, b. The activation function can take many forms. Assuming a neuron's activation function is y = f(z) = max(0, z), then the neuron's output is: For example, if the activation function of a neuron is y = f(z) = z, then the output of that neuron is: Here, b can be any possible type, such as a decimal, an integer (e.g., 0, a positive integer, or a negative integer), or a complex number. The activation functions of different neurons in a neural network can be the same or different.

Furthermore, neural networks generally consist of multiple layers, each of which may include one or more neurons. Increasing the depth and/or width of a neural network can improve its expressive power, providing more powerful information extraction and abstract modeling capabilities for complex systems. The depth of a neural network can refer to the number of layers it includes, and the number of neurons in each layer can be called the width of that layer. In one implementation, a neural network includes an input layer and an output layer. The input layer processes the received input information through neurons and passes the processing result to the output layer, which then obtains the output of the neural network. In another implementation, a neural network includes an input layer, hidden layers, and an output layer. The input layer processes the received input information through neurons and passes the processing result to the hidden layer. The hidden layer calculates the received processing result and passes the calculation result to the output layer or the next adjacent hidden layer, ultimately obtaining the output of the neural network. A neural network may include one hidden layer or multiple sequentially connected hidden layers, without limitation.

Neural networks, for example, are deep neural networks (DNNs). Depending on how the network is constructed, DNNs can include feedforward neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs).

Figure 2b is a schematic diagram of an FNN network. A characteristic of FNN networks is that neurons in adjacent layers are completely connected pairwise. This characteristic makes FNNs typically require a large amount of storage space, leading to high computational complexity.

CNNs are neural networks specifically designed to process data with a grid-like structure. For example, time-series data (e.g., discrete sampling along a time axis) and image data (e.g., two-dimensional discrete sampling) can both be considered grid-like data. CNNs do not use all the input information at once for computation; instead, they use a fixed-size window to extract a portion of the information for convolution operations, which significantly reduces the computational cost of model parameters. Furthermore, depending on the type of information extracted by the window (e.g., people and objects in an image represent different types of information), each window can use different convolution kernels, allowing CNNs to better extract features from the input data.

Recurrent Neural Networks (RNNs) are a type of distributed neural network (DNN) that utilizes feedback time-series information. The input to an RNN includes the current input value and its own output value from the previous time step. RNNs are well-suited for acquiring temporally correlated sequence features, and are particularly applicable to applications such as speech recognition and channel coding/decoding.

In the model training process described above for machine learning, a loss function can be defined. The loss function describes the difference or discrepancy between the model's output value and the ideal target value. The loss function can be expressed in various forms, and there are no restrictions on its specific form. The model training process can be viewed as follows: by adjusting some or all of the model's parameters, the value of the loss function is made to be less than a threshold value or to meet the target requirement.

A model can also be called an AI model, a rule, or other names. An AI model can be considered a specific method for implementing AI functions. An AI model represents the mapping relationship or function between the model's input and output. AI functions can include one or more of the following: data collection, model training (or model learning), model information dissemination, model inference (or model reasoning, inference, or prediction, etc.), model monitoring or model validation, or inference result publication, etc. AI functions can also be called AI (related) operations or AI-related functions.

The implementation process of a fully connected neural network will be described below with reference to the accompanying drawings. A fully connected neural network is also called a multilayer perceptron (MLP).

As shown in Figure 2c, an MLP consists of an input layer (left side), an output layer (right side), and multiple hidden layers (middle). Each layer of an MLP contains several nodes, called neurons. Neurons in adjacent layers are connected pairwise.

Optionally, considering neurons in two adjacent layers, the output h of a neuron in the next layer is the weighted sum of all neurons x in the previous layer connected to it, after passing through an activation function, and can be expressed as:
h = f(wx + b).

Where w is the weight matrix, b is the bias vector, and f is the activation function.

Alternatively, the output of the neural network can be recursively expressed as:
y=f _n (w _n f _n-1 (…)+b _n ).

Where n is the index of the neural network layer, n is greater than or equal to 1 and less than or equal to N, where N is the total number of layers in the neural network.

In other words, a neural network can be understood as a mapping from an input data set to an output data set. Neural networks are typically initialized randomly; the process of obtaining this mapping from random values w and b using existing data is called training the neural network.

Optionally, the training process can involve using a loss function to evaluate the output of the neural network.

As shown in Figure 2d, the error can be backpropagated, and the neural network parameters (including w and b) can be iteratively optimized using gradient descent until the loss function reaches its minimum, which is the "better point (e.g., the optimal point)" in Figure 2d. It can be understood that the neural network parameters corresponding to the "better point (e.g., the optimal point)" in Figure 2d can be used as the neural network parameters in the trained AI model information.

Alternatively, the gradient descent process can be represented as:

Where θ represents the parameters to be optimized (including w and b), L is the loss function, and η is the learning rate, controlling the step size of gradient descent. This represents the differentiation operation. This indicates taking the derivative of θ with respect to L.

Alternatively, the backpropagation process can utilize the chain rule for partial derivatives.

As shown in Figure 2e, the gradient of the parameters in the previous layer can be recursively calculated from the gradient of the parameters in the next layer, and can be expressed as:

Where w_{_ij} is the weight connecting node j to node i, and s _{_i} is the weighted sum of the inputs at node i.

In current communication systems, AI models are typically trained on one side of the network device or the terminal device, and then the trained model is sent to the other end for deployment. However, due to limitations in the capabilities of the terminal device or data privacy issues, one-sided training may not be feasible. Furthermore, models trained on one side may suffer from poor generalization. Therefore, how to achieve two-sided training of models has become an urgent problem to be solved.

Based on this, this application proposes a communication method and related apparatus. First, the sending end sends first information to the receiving end. The first information includes one or more first data and one or more first training label data, wherein the first data includes data output from the intermediate layer of the first model, and the first training label data is used to update the first model. Then, the sending end receives second information sent by the receiving end, the second information indicating model update data output in reverse from the intermediate layer of the first model. Through the above method, the sending end and receiving end can collaboratively train a neural network model, which can be flexibly applied to various different scenarios and tasks, improving compatibility. Furthermore, the first information sent by the sending end to the receiving end includes one or more first data and one or more first training label data; by sending rich training-related data to the receiving end, the generalization ability of the model is improved.

Next, the communication scenarios applicable to the embodiments of this application are described. Please refer to Figure 3, which is a schematic diagram of a communication scenario according to an embodiment of this application. The communication scenario involved in this embodiment includes: a sending end and a receiving end, wherein the direction from the sending end to the receiving end is defined as forward, and the direction from the receiving end to the sending end is defined as reverse (or backward). The sending end sends forward interaction data to the receiving end, the forward interaction data including first information; the receiving end sends reverse interaction data (or backward interaction data) to the sending end, the reverse interaction data including second information.

In one possible implementation, the sending end can be a terminal device, and the receiving end can be a network device, as shown in Figure 4, which is a schematic diagram of another communication scenario in an embodiment of this application. In the communication scenario illustrated in Figure 4, the terminal device, as the sending end, sends forward interaction data to the network device, which is the receiving end, and the network device sends reverse interaction data to the terminal device.

In another possible implementation, the sending end can be a network device, and the receiving end can be a terminal device.

In another possible implementation, the sender can be a network device, and the receiver can be another network device.

In another possible implementation, the sending end can be a terminal device, and the receiving end can be another terminal device.

Based on the above communication scenarios, the communication method proposed in this application is described below. Please refer to Figure 5, which is a schematic flowchart of an embodiment of the communication method in this application. The communication method proposed in this application includes:

S1. The sending end sends configuration information to the receiving end, and/or the receiving end sends configuration information to the sending end.

In step S1, when the sending end and the receiving end need to collaboratively train the neural network model, the sending end can send configuration information to the receiving end. In this embodiment, the neural network model to be trained is referred to as the first model, which may include one or more neural network models.

Furthermore, when the sending and receiving ends need to collaboratively train the first model, the sending end can send configuration information to the receiving end, or the receiving end can send the configuration information to the sending end. Alternatively, both the sending and receiving ends can send configuration information to each other, ensuring that the configuration information at both ends remains consistent. Through these methods, it is ensured that the sending and receiving ends are training the same neural network model (i.e., the first model).

Optionally, during the training of the first model at the sending and receiving ends, the configuration information related to the first model may also change. In this case, either the sending or receiving end can send the changed configuration information to the other end to ensure consistency between the configuration information at both ends. This method ensures that the sending and receiving ends are training the same neural network model (i.e., the first model). For example, if the segmentation information (segmentation position and/or the number of segmentation positions) of the first model changes during training, the sending and receiving ends need to exchange configuration information (including the updated segmentation information) to ensure consistency between their first models. In other words, the execution order of step S1 and subsequent steps S2-S3 is not restricted here.

Optionally, the first model may include two parts: a transmitter model and a receiver model, wherein the transmitter trains the transmitter model in the first model, and the receiver trains the receiver model in the first model.

The following describes this configuration information. Specifically, this configuration information is used to configure the training task for training the first model, and it includes one or more of the following:

Model information, wherein the model information includes any one or more of the following: the identifier of the first model, the structural information of the first model, the training task information of the first model, the initialization parameters of the first model, or the parameters of the first model to be updated;

Segmentation information, wherein the segmentation information includes: the number of segmentation positions in the first model, and/or, the position information of the segmentation positions in the first model;

Alternatively, training information, which includes any one or more of the following: training label set information, validation set information, number of training iterations, learning rate, or loss function, wherein the training label set includes one or more training labels.

For example, when multiple first models are trained in parallel between the sending end and the receiving end, the identification information of the first model may include multiple identification information, each of which indicates a first model. When the first model trained in parallel between the sending end and the receiving end changes, the identification information of the first model included in the configuration information needs to be updated. For example, before the update, if the sending end and the receiving end train first model #1 and first model #2, then the identification information of the first model in the configuration information before the update includes: the identifier (ID) of first model #1 and the ID of first model #2. After the update, if the sending end and the receiving end train first model #3 and first model #4, then the identification information of the first model in the updated configuration information includes: the ID of first model #3 and the ID of first model #4.

For example, the training task information of the first model indicates the scenario (or environment) in which the first model is applicable, such as: reconstruction scenario, detection scenario, outdoor scenario, or indoor scenario.

It should be noted that the splitting position in this embodiment is used to indicate which intermediate layer in the first model outputs data as the first data, and the data output at that output position has been processed by that intermediate layer and other layers (e.g., other intermediate layers) before that intermediate layer.

For example, the number of segmentation positions of the first model in the segmentation information is 2. The position information of the segmentation positions in the first model in the segmentation information are: ReLU activation layer #1 and ReLU activation layer #4. Then the segmentation positions indicated by the segmentation information are: segmentation position #1 and segmentation position #2, as shown in Figures 10 and 11. Figure 10 is a schematic diagram of the segmentation position of the first model in an embodiment of this application, and Figure 11 is a schematic diagram of the segmentation position of the first model in another embodiment of this application. Taking the first model as a convolutional neural network (CNN) as an example, the first model includes multiple convolutional layers and multiple activation function layers. Taking convolutional layer #1 as an example, the weights (W) of the convolutional layer #1 are a matrix of size 192x192x1x7, and the bias (B) of the convolutional layer #1 is 192. Figures 10 and 11 show a partial structure of the first model, where the split position #1 is specifically the output of the activation function layer (ReLU) #1 in the first model, and the split position #2 is specifically the output of the activation function layer (ReLU) #4 in the first model.

It should be noted that the partial structure of the model shown in Figure 10 and the partial structure of the model shown in Figure 11 belong to the same first model. That is, one branch of the first model is shown in Figure 10, and another branch of the first model is shown in Figure 11.

Optionally, if the first and/or second (or third and/or fourth) information is compressed data, the configuration information may further include compression-related configuration information. Specifically, the configuration information may also include: first compression configuration information, and/or second compression configuration information, wherein: the first compression configuration information indicates the compression method and compression parameters of the first and/or third information, and the second compression configuration information indicates the compression method and compression parameters of the second and/or fourth information. The sending end sends the first and/or third information to the receiving end, i.e., the forward interaction data sent by the sending end to the receiving end includes: the first and/or third information; the receiving end sends the second and/or fourth information to the sending end, i.e., the reverse interaction data sent by the receiving end to the sending end includes: the second and/or fourth information. In other words, the first compression configuration information indicates the compression method and compression parameters of the forward interaction data, and the second compression configuration information indicates the compression method and compression parameters of the reverse interaction data.

In one possible implementation, the forward interaction data and the reverse interaction data use the same compression method and compression parameters. In this case, the first compression configuration information and the second compression configuration information are the same compression configuration information, which can also be called common compression configuration information.

In another possible implementation, the forward and reverse interaction data use different compression methods and parameters. In this case, the configuration information sent from the sending end to the receiving end may include first compressed configuration information and second compressed configuration information; or, the configuration information sent from the receiving end to the sending end may include first compressed configuration information and second compressed configuration information; or, the configuration information sent from the sending end to the receiving end includes first compressed configuration information, and the configuration information sent from the receiving end to the sending end includes second compressed configuration information; or, the configuration information sent from the sending end to the receiving end includes second compressed configuration information, and the configuration information sent from the receiving end to the sending end includes first compressed configuration information.

Optionally, the aforementioned first compression configuration information and/or second compression configuration information may be carried in configuration information; or the first compression configuration information may be carried in first information and/or third information, and the second compression configuration information may be carried in second information and/or fourth information; or it may be carried in other independent information, and the embodiments of this application do not limit this.

Regarding compression methods, the embodiments of this application include, but are not limited to: no compression, fixed quantization, entropy coding, fixed quantization and entropy coding, dictionary compression, or dictionary compression and entropy coding, etc.

Regarding compression parameters, embodiments of this application include, but are not limited to: quantization parameters, such as quantization boundaries or quantization bits, entropy coding parameters, or codebook dimensions, etc.

For example, the first compression configuration information and/or the second compression configuration information can be indicated using index information. Based on this index information, the corresponding compression configuration information can be determined from the compression configuration information table. The sending end and the receiving end configure this compression information table respectively. The compression configuration information table is shown in Table 2, for example.

Table 2

Optionally, the codebook dimension in the compression parameters can also be indicated by index information. Based on this index information, the corresponding dictionary compression parameters (i.e., codebook dimensions) can be determined from the dictionary compression parameter table. The dictionary compression parameter table is shown in Table 3.

Table 3

It should be noted that the aforementioned compressed configuration information can be carried within the configuration information, or it can be sent together with the first, second, third, and/or fourth information. This application embodiment does not impose any limitations on this. For example, the forward interaction data sent from the sending end to the receiving end includes the first and/or third information, and this forward interaction data also includes the first compressed configuration information. As another example, the reverse interaction data sent from the receiving end to the sending end includes the second and/or fourth information, and this reverse interaction data also includes the second compressed configuration information.

It should be noted that the various information items included in the above configuration information can be sent separately or together, and this application embodiment does not impose any restrictions on this.

S2. The sending end sends the first information and/or the third information to the receiving end.

In step S2, after the sending end and receiving end configure the first model according to the configuration information, the sending end trains the first model and then sends the data output by the intermediate layer of the first model during the training process to the receiving end. In this embodiment, the data output by the intermediate layer of the first model is referred to as the first data. In addition to the first data, the first information also includes first training label data, which is used to update the first model. Specifically, the first training label data includes: the index information of the first training label in the training label set, or the first training label, wherein the training label set includes one or more training labels. This training label set is the training label set indicated by the training label set information in the configuration information.

In one possible implementation, the sending end can determine the training label set based on the training label set information in the configuration information. Then, the sending end trains the first model based on the first training label in the training label set. The sending end uses the data output from the intermediate layer of the first model as the first data. In this possible implementation, the first training label data includes: the index information of the first training label in the training label set.

In another possible implementation, when the configuration information does not include training label set information, the sending and receiving ends train by collecting training samples in real time. These training samples are called the first training samples. The training label corresponding to the first training sample is called the first training label. In this possible implementation, the first training label data includes: the first training label.

During the training of the first model, to improve training efficiency, the sending end can send one or more first data points to the receiving end. In one possible implementation, when the segmentation information included in the configuration information indicates that the first model includes multiple segmentation positions, the multiple feature data (or other data) output by the multiple segmentation positions are considered as one set of first data. In the first information, the multiple data points included in a single set of first data can be ordered according to the hierarchical size of the segmentation position that outputs the first data.

In another possible implementation, the sending end acquires multiple distinct first training samples. Then, the sending end trains a first model using these multiple first training samples, with each training iteration using a different first training sample. During the training process of the first model using multiple first training samples, multiple sets of first data are output.

Optionally, the sending end can acquire multiple first training samples by directly acquiring multiple different first training samples. For example, the sending end can acquire multiple different first training samples through multiple acquisitions by a sensor. Alternatively, the sending end can acquire a common training sample and then perform different processing on that common training sample to obtain multiple different first training samples. Specifically, different processing on the common training sample includes, but is not limited to: data cleaning, data standardization, data normalization, feature scaling, encoding, feature selection, feature extraction, data augmentation, or resampling.

Optionally, after obtaining a common training sample, the sending end performs different processing on the common training sample to obtain multiple first data.

In another possible implementation, after acquiring a first training sample, the sending end uses this first training sample to process the first model one or more times. This processing can be training processing. During the multiple processing steps using the same first training sample, the sending end outputs one or more first data points. In these multiple processing steps, the processing method and/or processing parameters are inconsistent each time. For example, the initialization parameters of the first model are different in each processing step, other input information of the first model is different in each processing step, or the sampling operation of the first model is different in each processing step.

After acquiring one or more first data points and one or more first training label data points at the sending end, the sending end sends first information to the receiving end. The first information includes the one or more first data points and one or more first training label data points. The one or more first data points and one or more first training label data points in the first information can be grouped in various different ways to improve generalization. The different grouping methods of the first information are described below.

In grouping method A, the first information includes one or more first data and one first training label data, wherein the first training label data includes a first training label (or index information indicating the first training label) corresponding to the one or more first data. For example, as shown in Figure 6, Figure 6 is a schematic diagram of the structure of the first information in an embodiment of this application. In Figure 6, the first information includes n first data and one first training label data, where n is a positive integer greater than or equal to 1.

In grouping method B, the first information includes one or more first data points and multiple first training label data points, wherein each first data point corresponds to one first training label data point. For example, as shown in Figure 7, which is another structural diagram of the first information in an embodiment of this application, the first information includes n first data points and n first training label data points, where n is a positive integer greater than or equal to 1. Each first data point and its corresponding first training label data point are divided into a group of data, meaning the first information includes n groups of data, each group including one first data point and its corresponding first training label data point. For example, the nth group of data includes first data point n and first training label data point n.

Optionally, the sending end may also inform the receiving end of the grouping method of the first information and the number of one or more first data points included in the first information. Specifically, the first information may further include: grouping indication information, which indicates the grouping method of the one or more first data points included in the first information; and/or, group number information, which indicates the number of one or more first data points included in the first information. For example, the grouping indication information indicates whether the first information is grouped using grouping method A or grouping using grouping method B. As another example, the group number information indicates that the first information includes n groups of data (n first data points and n first training label data points, each group including one first data point and one first training label data point).

For example, the first information can be supplemented with indication information to indicate whether the first information adopts grouping method A or grouping method B. For example, when the indication information is 0, it indicates that the first information adopts grouping method A; when the indication information is 1, it indicates that the first information adopts grouping method B.

In step S2, in addition to sending the first information to the receiving end, the sending end may also send third information to the receiving end. Specifically, the third information includes, but is not limited to: third data, input location information of the third data in the first model, dimension information of the third data, first type indication information, or identification information of the first model. The first type indication information is used to indicate the data type of the third data. The third data includes any one or more of the following information: first channel information, first environment information, or multimodal information. The first type indication information is related to the channel between the sending end and the receiving end, and the first environment information is related to the environment in which the sending end and the receiving end are located.

For example, the channel information in the embodiments of this application includes, but is not limited to: channel capacity information, channel bandwidth information, channel noise information, channel modulation method, channel coding method, channel state information, or channel bit error rate, etc.

For example, the environmental information in the embodiments of this application includes, but is not limited to: terrain information, climate condition information, electromagnetic environment information, network topology information, network load information, signal quality information, or wireless spectrum information.

For example, the multimodal information in the embodiments of this application includes, but is not limited to, image information, point cloud information, text information, or audio information.

For example, the first type of indication information is shown in Table 4.

Table 4

For example, the dimension information of the third data refers to the data dimension of the third data. For instance, when the third data includes the first channel information, and the first channel information is a J*K dimension matrix, then the dimension information of the third data is J*K, where J is a positive integer and K is a positive integer.

For example, the identification information of the first model is used to indicate which models and/or the types of models included in the first model. The identification information of the first model is shown in Table 5, for example.

Table 5

For example, the input position information for inputting third data in the first model is used to instruct the receiving end to input the third data at that input position in the first model after receiving the third information. For ease of understanding, please refer to Figure 12, which is another structural schematic diagram of the first model in an embodiment of this application. The input position information for inputting third data in the first model instructs the input of third data at the output position of the activation function layer (ReLU) #6 of the first model. When the third data also includes first channel information and first environment information, after receiving the third information (including the third data), the receiving end inputs the first channel information and first environment information included in the third data at the output position of the activation function layer (ReLU) #6 of the first model.

It should be noted that the first and third information mentioned above can be carried in the same message, that is, the sending end sends the first and third information at the same time; or the first and third information mentioned above can be carried in different messages, that is, the sending end sends the first and third information separately. This application embodiment does not limit this.

The following describes how the first and/or third information is carried:

Implementation Method A: The first and/or third information is carried in a radio resource control (RRC) message. Specifically, the DataInfo field, in RRC sequence format, indicates the specific type of the first and/or third information.

In one possible implementation, the data type is implicitly indicated by the presence of an element representing the data type in the DataInfo field.

In one example, the data type of the data included in the forward interaction data is indicated as follows: "DataInfo::SEQUENCE{

forward-inter-Data ForwardInterData OPTIONAL,

model-Data ModelData OPTIONAL,

}”, where “ForwardInterData” indicates that the RRC message carries forward interaction data, and “ModelData” indicates that the RRC message carries model-related data.

In another example, the data type of the data included in the forward interaction data is indicated as follows: "DataInfo::SEQUENCE{

ai-DataInfo AIDataInfo OPTIONAL,

sensing-DataInfo SensingDataInfo OPTIONAL,

}

AIDataInfo::SEQUENCE{

Forward-inter-Data ForwardInterData OPTIONAL,

Model-Data ModelData OPTIONAL,

The specific meaning of "}" is: This RRC message carries AI data, which specifically includes forward interaction data and model-related data.

In another possible implementation, a DataType field indicates whether a certain data type exists. If it exists, the data type is "true"; otherwise, if it does not exist, the data type is "false".

In one example, the data type of the data included in the forward interaction data is indicated in the following way:

“DataType::＝SEQUENCE{

includeForwardInterData ENUMERATED{true}OPTIONAL

}

DataInfo::＝SEQUENCE{

Forward-inter-Data ForwardInterData

The `includeForwardInterData ENUMERATED{true}` indicates that the RRC message carries forward interactive data.

In another example, the "forwardInterData" field indicates one or more of the following information: the identifier of the first model (modelIndicate), data configuration information (dataConfig), grouping indication information (groupMode), number of groups information (groupNum), feature data (featureData), or other data (otherData). The "dataConfig" field indicates one or more of the following information: the data type of other data (otherDataType), the location information of other data (otherDataPosition), the dimension information of other data (otherDataDim), or compression configuration information (configCompress). Specifically: "ForwardInterData::＝SEQUENCE{

modelIndicate OCTET STRING,OPTIONAL,

dataConfig ConfigModel,OPTIONAL,

groupMode OCTET STRING,OPTIONAL,

groupNum OCTET STRING,OPTIONAL,

featureData OCTET STRING,OPTIONAL,

trainingLabel OCTET STRING,OPTIONAL,

otherData OCTET STRING,OPTIONAL

}

ConfigModel::＝SEQUENCE{

otherDataType OCTET STRING OPTIONAL,

otherDataPosition OCTET STRING OPTIONAL,

otherDataDim OCTET STRING OPTIONAL,

configCompress OCTET STRING OPTIONAL

}".

Optionally, the configuration information in step S1 can also be carried in an RRC message. The specific carrying method is similar to that of the first information and/or the third information carried in an RRC message, and will not be elaborated here.

Implementation method B: The first and/or third information is carried in a media access control (MAC) message.

In one possible implementation, the data type carried by the MAC message is indicated in the MAC subheader using a Logical Channel ID (LCID) or an Extended Logical Channel ID (eLCID). Therefore, when first and/or third information is carried in a MAC message, the data type of the first and/or third information can be indicated by the LCID or eLCID in the MAC subheader.

For example, since the values of LCID from 0 to 34 are currently in use, LCID = 35 can be used as the data type for the first and/or third information (i.e., forward interaction data). When the receiving end receives a MAC message and finds that the value of the LCID field of the MAC message is 35, it determines that the MAC message carries the first and/or third information. Furthermore, LCID = 36 can also be used as the data type for the second and/or fourth information (i.e., reverse interaction data).

In another possible implementation, LCID or eLCID is used to indicate that the MAC message carries AI data. For example, LCID=35 indicates that the data type carried by the MAC message is AI data. AI data includes forward interaction data (first information and/or third information) and reverse interaction data (second information and/or fourth information). Therefore, in order to distinguish whether the AI data specifically carries forward interaction data or reverse interaction data, the type field of the MAC message is used to indicate this.

It should be noted that both the type field and the payload mentioned above can be carried on the MAC Control Element (MAC CE). The payload can be used to carry AI data.

As an example, please refer to Figure 16, which is a schematic diagram of the structure of a MAC message carrying forward interaction data in an embodiment of this application. In Figure 16, the LCID of the MAC message is 35, indicating that the MAC message carries AI data; the type of the MAC message is 0, indicating that the AI data carried by the MAC message specifically includes forward interaction data.

Another example is shown in Figure 17, which is a schematic diagram of the structure of a MAC message carrying reverse interaction data in an embodiment of this application. In Figure 17, the LCID of the MAC message is 35, indicating that the MAC message carries AI data; the type of the MAC message is 1, indicating that the AI data carried by the MAC message specifically includes reverse interaction data.

Optionally, the configuration information in step S1 can also be carried in a MAC message. The specific carrying method is similar to that of the first information and/or the third information carried in a MAC message, and will not be elaborated here.

It should be noted that the information included in the first or third information mentioned above can be sent separately or together, and this application embodiment does not impose any restrictions on this.

For example, as shown in Figure 13, which is a schematic diagram of a forward interaction data structure in an embodiment of this application, the forward interaction data sent from the sending end to the receiving end includes: first information and third information. The first information includes n first data points (first data 1 to first data n) and first training label data. Optionally, the forward interaction data shown in Figure 13 may also include configuration information (see step S1) and/or compressed configuration information (e.g., first compressed configuration information).

For example, as shown in Figure 14, which is another structural diagram of forward interaction data in an embodiment of this application, the forward interaction data sent from the sending end to the receiving end includes: first information and third information. The first information includes n first data items (first data 1 to first data n), n first training label data items (first training label data 1 to first training label data n), and n third information items (third information 1 to third information n). The first data items correspond to the first training label data and the third information items. Optionally, the forward interaction data shown in Figure 14 may also include configuration information (see step S1) and/or compressed configuration information (e.g., first compressed configuration information).

It is understandable that when one or more pieces of first data included in the first information share a single configuration information (e.g., one or more pieces of first data include data output from the same model), then the first information may include only one such configuration information. When one or more pieces of first data included in the first information correspond to different configuration information, then the first information includes multiple configuration information.

S3. The receiving end sends the second and/or fourth information to the sending end.

After step S2, the receiving end receives forward interaction data (including first information and/or third information) from the sending end. The receiving end then obtains the training loss function (loss) of the first model based on the forward interaction data and updates the decoding model (i.e., the receiving end model) in the first model. Then, during the reverse update process, the receiving end obtains second data based on the first model. The second data includes model update data output from the reverse of the intermediate layers of the first model.

For example, the second data includes, but is not limited to, gradients. Specifically, the gradient can be the gradient of the bias or the gradient of the weights, etc.

In one possible implementation, the second information includes multiple second data, where each second data corresponds to a first data. For example, as shown in Figure 8, which is a schematic diagram of the structure of the second information in an embodiment of this application. The sending end sends the second information to the receiving end, which includes m second data, where m is a positive integer.

In another possible implementation, the second information includes weighted values of multiple second-number data points (N), wherein the first information also includes the weights of the first data points. The receiving end determines multiple second-number data points based on one or more first data points. Then, the receiving end performs weighted processing on the multiple second-number data points according to the weights of the first data points corresponding to each second-number data point, to obtain a weighted value for the multiple second-number data points. For example, if the weights of each first data point are equal, then the second information includes the average value of the multiple second-number data points. For example, as shown in Figure 9, Figure 9 is another structural schematic diagram of the second information in an embodiment of this application. The sending end sends the second information to the receiving end, which includes weighted values of m second-number data points (i.e., a total of m second-number data points from second-number data 1 to second-number data m), where m is a positive integer.

In step S3, in addition to sending the second information to the sending end, the receiving end can also send a fourth information to the sending end. Specifically, the fourth information includes, but is not limited to:

The fourth data includes: second type indication information, termination interaction instruction of the first model, dimension information of the fourth data, or identification information of the first model. The second type indication information is used to indicate the data type of the fourth data. The termination interaction instruction of the first model is used to instruct the sending end to stop sending information related to the first model. The second channel information is related to the channel between the sending end and the receiving end. The second environment information is related to the environment in which the sending end and the receiving end are located. The fourth data includes any one or more of the following information: second channel information, second environment information, multimodal information, training residual information generated by the receiving end during the training of the first model, or performance information of the updated first model.

For example, the second channel information is similar to the first channel information in step S2 above, and will not be described in detail here.

For example, the second environmental information is similar to the first environmental information in step S2 above, and will not be described in detail here.

For example, the multimodal information is similar to the multimodal information in step S2 above, and will not be described in detail here.

For example, the second type of indication information, the dimension information of the fourth data, or the identification information of the first model are similar to the first type of indication information, the dimension information of the third data, or the identification information of the first model in the aforementioned step S2, and will not be elaborated here.

Once the sending end receives the training residual information generated by the receiving end during the training of the first model, the sending end can use this training residual information to train the first model.

Upon receiving a termination interaction command from the first model, the sending end ceases exchanging data related to the first model with the receiving end, such as first information and/or third information. Optionally, the sending end may also stop training the first model.

After receiving the performance information of the first model, the sending end can determine whether to continue exchanging data related to the first model, such as first information and/or third information, based on this performance information. If the sending end determines that the performance of the first model meets the requirements based on the performance information, it can stop exchanging data related to the first model with the receiving end. Optionally, the sending end can also stop training the first model. For example, the performance information of the first model includes: test loss function, test accuracy, precision, and/or recall.

It should be noted that the various pieces of information included in the second or fourth information mentioned above can be sent separately or together, and this application embodiment does not impose any restrictions on this.

For example, as shown in Figure 15, which is a structural schematic diagram of the second information in an embodiment of this application, the sending end sends the second information and the fourth information to the receiving end. The second information includes the weighted values of m second data (i.e., the weighted values of a total of m second data from second data 1 to second data m), where m is a positive integer. Since each of the m second data corresponds to one fourth information, the reverse interaction data also includes m fourth information (i.e., the m fourth information from fourth information 1 to fourth information m), where fourth information 1 corresponds to second data 1, and so on, with fourth information m corresponding to second data m.

Optionally, the weighted value of the m second data points can be the average of the m second data points.

Optionally, the reverse interaction data shown in Figure 15 may also include configuration information (see step S1) and/or compressed configuration information (e.g., second compressed configuration information).

In another possible implementation, the second information includes the weighted values of the second data 1 to the second data m and a fourth information corresponding to the m second data from the second data 1 to the second data m.

It is understandable that when multiple pieces of second data included in the second information share a single configuration information (e.g., multiple pieces of second data include model update data output from the same model), then the second information may include only one such configuration information. When multiple pieces of second data included in the second information correspond to different configuration information, then the second information includes multiple configuration information.

Optionally, after step S3, if the sending end or the receiving end determines that it is necessary to continue updating the first model, steps S2 to S3 can be repeated until the performance of the first model meets the service requirements.

Optionally, during the execution of steps S1 to S3, if the sending end determines that other models need to be trained, the sending end can send the configuration information of the model to the receiving end (similar to step S1). Then, the sending end and the receiving end exchange the relevant data of the model, similar to steps S2 to S3 above, which will not be elaborated here.

Optionally, during the execution of steps S1 to S3, if the sending end or the receiving end needs to modify the configuration information of the first model, the sending end can send the updated configuration information to the receiving end (similar to step S1), or the receiving end can send the updated configuration information to the sending end (similar to step S1). Then, the sending end and the receiving end exchange the updated relevant data of the first model, similar to steps S2 to S3 above, which will not be elaborated here.

In this embodiment, the sending end sends first information to the receiving end. The first information includes one or more first data points and one or more first training label data points. The first data points include data output from the intermediate layers of the first model, and the first training label data is used to update the first model. Then, the sending end receives second information sent by the receiving end. The second information instructs the intermediate layers of the first model to output model update data in reverse. Through this method, the sending end and receiving end can collaboratively train a neural network model, which can be flexibly applied to various different scenarios and tasks, improving compatibility. Furthermore, the first information sent by the sending end to the receiving end includes one or more first data points and one or more first training label data points. By sending rich training-related data to the receiving end, the generalization ability of the model is improved. In addition, by defining a unified interaction format for the data exchanged between the sending end and the receiving end, the sending end and the receiving end can flexibly train for different training scenarios and tasks, improving compatibility. Furthermore, the sending end and the receiving end can support training multiple models within a single training cycle, improving training efficiency. The sending end and the receiving end can also adjust the trained model within a single training cycle, further improving training efficiency. Because the data exchanged between the sending and receiving ends supports multiple compression methods, the amount of data transmitted can be effectively reduced, thus improving communication efficiency.

Please refer to Figure 18, which is a schematic diagram of the structure of a communication device 1800 in an embodiment of this application. This communication device 1800 can realize the functions of a transmitting end or a receiving end in the above method embodiments, and therefore can also achieve the beneficial effects of the above method embodiments. In this embodiment, the communication device 1800 can be a transmitting end or a receiving end, or it can be an integrated circuit or component inside the transmitting end or receiving end, such as a chip, baseband chip, modem chip, SoC chip containing a modem core, system-in-package (SIP) chip, communication module, chip system, processor, etc.

It should be noted that the transceiver unit 1802 may include a transmitting unit and a receiving unit, which are used to perform transmitting and receiving respectively.

In one possible implementation, when the device 1800 is used to execute the method performed by the transmitting end or the receiving end in FIG5 and related embodiments, the device 1800 includes a processing unit 1801 and a transceiver unit 1802.

In one example, the communication device 1800 is used at a transmitting end, and the communication device 1800 includes:

The transceiver unit 1802 is used to send first information to the receiving end. The first information includes one or more first data and one or more first training label data. The first data includes data output from the intermediate layer of the first model. The first training label data is used to update the first model.

The transceiver unit 1802 is also used to receive second information sent by the receiving end, the second information indicating the model update data output in reverse by the intermediate layer of the first model.

In one possible implementation, the first training label data includes:

The first training label, or the index information of the first training label in the training label set, wherein the training label set includes one or more training labels.

In one possible implementation, the first information further includes:

Grouping indication information, wherein the grouping indication information is used to indicate the grouping method of the plurality of first data included in the first information;

And/or, group number information, which indicates the number of the plurality of first data included in the first information.

In one possible implementation, there is a weighted sum of multiple second data, each second data corresponding to one first data.

In one possible implementation, the first information may also include the weight of the first data.

In one possible implementation, the transceiver unit 1802 is further configured to acquire a first training sample, wherein the training label corresponding to the first training sample is the first training label.

The processing unit 1801 is used to process the first model multiple times using the first training samples and output the plurality of first data, wherein the processing method and/or processing parameters are different each time.

In one possible implementation, the transceiver unit 1802 is further configured to acquire multiple first training samples, wherein the training label corresponding to the first training sample is the first training label.

The processing unit 1801 is further configured to process the first model once or multiple times using the plurality of first training samples, and output the plurality of first data, wherein the first training samples used in each processing are different.

In one possible implementation, the transceiver unit 1802 is also used to acquire public training samples;

The processing unit 1801 is further configured to process the common training samples to generate the plurality of first training samples, wherein the plurality of first training samples are different from each other.

In one possible implementation, the transceiver unit 1802 is further configured to send third information to the receiving end, the third information including any one or more of the following:

The third data, the input location information of the third data in the first model, the dimension information of the third data, the identification information of the first model, or the first type indication information, wherein the first type indication information is used to indicate the data type of the third data, wherein the third data includes any one or more of the following information: first channel information, first environment information, or multimodal information.

In one possible implementation, the transceiver unit 1802 is further configured to receive fourth information sent by the receiving end, the fourth information including any one or more of the following:

The fourth data, the second type indication information, the termination interaction instruction of the first model, the dimension information of the fourth data, or the identification information of the first model, the second type indication information is used to indicate the data type of the fourth data, and the termination interaction instruction of the first model is used to instruct the sending end to stop sending information related to the first model, wherein the fourth data includes any one or more of the following information: second channel information, second environment information, multimodal information, training residual information generated by the receiving end during the training of the first model, or the performance information of the updated first model.

In one possible implementation, the first model includes one or more segmentation locations;

The first data includes: data output from intermediate layers corresponding to one or more segmentation positions in the first model.

In one possible implementation, the transceiver unit 1802 is further configured to send configuration information to the receiving end, the configuration information being used to configure the training task for training the first model, the configuration information including any one or more of the following:

Model information, wherein the model information includes any one or more of the following: the identification information of the first model, the structural information of the first model, the training task information of the first model, the initialization parameters of the first model, or the parameters of the first model to be updated.

Segmentation information, wherein the segmentation information includes: the number of segmentation positions of the first model, and/or, the position information of the segmentation positions in the first model;

Alternatively, training information, wherein the training information includes any one or more of the following: training label set information, validation set information, number of training iterations, learning rate, or loss function, wherein the training label set includes one or more training labels;

And/or, the transceiver unit 1802 is also used to receive the configuration information sent by the receiving end.

In one possible implementation, the configuration information further includes:

First compression configuration information, and/or, second compression configuration information, wherein:

The first compression configuration information is used to indicate the compression method and compression parameters of the first information and/or the third information.

The second compression configuration information is used to indicate the compression method and compression parameters of the second information and/or the fourth information.

In one possible implementation, the first information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message.

And/or, the second information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

In one possible implementation, the third information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message.

And/or, the fourth information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

In another example, the communication device 1800 is applied at the receiving end, and the communication device 1800 includes:

The transceiver unit 1802 is used to receive first information sent by the sender. The first information includes one or more first data and one or more first training label data. The first data includes data output from the intermediate layer of the first model. The first training label data is used to update the first model.

The transceiver unit 1802 is also used to send second information to the sending end, the second information indicating the model update data output in reverse by the intermediate layer of the first model.

In one possible implementation, the first training label data includes:

The first training label, or the index information of the first training label in the training label set, wherein the training label set includes one or more training labels.

In one possible implementation, the first information further includes:

Grouping indication information, wherein the grouping indication information is used to indicate the grouping method of the plurality of first data included in the first information;

And/or, group number information, which indicates the number of the plurality of first data included in the first information.

In one possible implementation, the second information includes: the weighted value of the plurality of second data.

In one possible implementation, the first information may also include the weight of the first data.

In one possible implementation, the first information may also include the weight of the first data.

In one possible implementation, the transceiver unit 1802 is further configured to receive third information transmitted by the transmitting end, the third information including any one or more of the following:

The third data, the input location information of the third data in the first model, the dimension information of the third data, the identification information of the first model, or the first type indication information, wherein the first type indication information is used to indicate the data type of the third data, wherein the third data includes any one or more of the following information: first channel information, first environment information, or multimodal information.

In one possible implementation, the transceiver unit 1802 is further configured to send fourth information to the sending end, the fourth information including any one or more of the following:

The fourth data, the second type indication information, the termination interaction instruction of the first model, the dimension information of the fourth data, or the identification information of the first model, the second type indication information is used to indicate the data type of the fourth data, and the termination interaction instruction of the first model is used to instruct the sending end to stop sending information related to the first model, wherein the fourth data includes any one or more of the following information: second channel information, second environment information, multimodal information, training residual information generated by the receiving end during the training of the first model, or the performance information of the updated first model.

In one possible implementation, the first model includes one or more segmentation locations;

The first data includes: data output from intermediate layers corresponding to one or more segmentation positions in the first model;

And/or, the second data includes: model update data output in reverse from one or more of the segmentation positions in the first model.

In one possible implementation, the transceiver unit 1802 is further configured to receive configuration information from the receiving end, the configuration information being used to configure the training task for training the first model, the configuration information including any one or more of the following:

Model information, wherein the model information includes any one or more of the following: the identification information of the first model, the structural information of the first model, the training task information of the first model, the initialization parameters of the first model, or the parameters of the first model to be updated.

Segmentation information, wherein the segmentation information includes: the number of segmentation positions of the first model, and/or, the position information of the segmentation positions in the first model;

Alternatively, training information, wherein the training information includes any one or more of the following: training label set information, validation set information, number of training iterations, learning rate, or loss function, wherein the training label set includes one or more training labels;

And/or, the transceiver unit 1802 is also used to send the configuration information to the sending end.

In one possible implementation, the configuration information further includes:

First compression configuration information, and/or, second compression configuration information, wherein:

The first compression configuration information is used to indicate the compression method and compression parameters of the first information and/or the third information.

The second compression configuration information is used to indicate the compression method and compression parameters of the second information and/or the fourth information.

In one possible implementation, the first information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message.

And/or, the second information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

In one possible implementation, the third information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message.

And/or, the fourth information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message.

In one possible design, when the communication device 1800 is a terminal device or a communication module within a terminal (or network device), the function of the processing unit 1801 can be implemented by one or more processors. Specifically, the processor may include a modem chip, or a SoC chip or SIP chip containing a modem core. The function of the transceiver unit 1802 can be implemented by transceiver circuitry.

In one possible design, when the communication device 1800 is a circuit or chip responsible for communication functions in a terminal (or network device), such as a modem chip or a SoC chip or SIP chip containing a modem core, the function of the processing unit 1801 can be implemented by a circuit system in the aforementioned chip that includes one or more processors or processor cores. The function of the transceiver unit 1802 can be implemented by the interface circuit or data transceiver circuit on the aforementioned chip.

It should be noted that the information execution process of the unit of the above-mentioned communication device 1800 can be specifically described in the method embodiment shown above in this application, and will not be repeated here.

Please refer to Figure 19, which is another schematic structural diagram of the communication device 1900 provided in this application. The communication device 1900 includes a logic circuit 1901 and an input/output interface 1902. The communication device 1900 can be a chip or an integrated circuit.

In Figure 18, the transceiver unit 1802 can be a communication interface, which can be the input/output interface 1902 in Figure 19. The input/output interface 1902 can include an input interface and an output interface. Alternatively, the communication interface can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit. The logic circuit 1901 and the input/output interface 1902 can also perform other steps executed by the transmitting or receiving end in any embodiment and achieve corresponding beneficial effects, which will not be elaborated further here.

In one possible implementation, the processing unit 1801 shown in FIG18 can be the logic circuit 1901 in FIG19.

Optionally, the logic circuit 1901 can be a processing device, the functions of which can be partially or entirely implemented in software.

Optionally, the processing apparatus may include a memory and a processor, wherein the memory is used to store a computer program, and the processor reads and executes the computer program stored in the memory to perform the corresponding processing and/or steps in any of the method embodiments.

Optionally, the processing device may consist of only a processor. A memory for storing computer programs is located outside the processing device, and the processor is connected to the memory via circuitry/wires to read and execute the computer programs stored in the memory. The memory and processor may be integrated together or physically independent of each other.

Optionally, the processing device may be one or more chips, or one or more integrated circuits. For example, the processing device may be one or more field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), system-on-chips (SoCs), central processing units (CPUs), network processors (NPs), digital signal processors (DSPs), microcontroller units (MCUs), programmable logic controllers (PLDs), or other integrated chips, or any combination of the above chips or processors.

Please refer to Figure 20. Figure 20 is a structural schematic diagram of the communication device 2000 involved in the above embodiments provided by the embodiments of this application. Specifically, the communication device 2000 can be the communication device as the sending end (or receiving end) in the above embodiments. The example shown in Figure 20 is that the sending end is implemented by the sending end (or the component in the sending end), or the example shown in Figure 20 is that the receiving end is implemented by the receiving end (or the component in the receiving end).

The present invention is a possible logical structure diagram of the communication device 2000, which may include, but is not limited to, at least one processor 2001 and a communication port 2002.

In Figure 18, the transceiver unit 1802 can be a communication interface, which can be the communication port 2002 in Figure 20. The communication port 2002 can include an input interface and an output interface. Alternatively, the communication port 2002 can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

Further optionally, the device may also include at least one of a memory 2003 and a bus 2004. In embodiments of this application, the at least one processor 2001 is used to control the operation of the communication device 2000.

Furthermore, the processor 2001 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, etc. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

It should be noted that the communication device 2000 shown in Figure 20 can be used to implement the steps implemented by the sending end or the receiving end in the aforementioned method embodiments, and to achieve the corresponding technical effects of the sending end or the receiving end. The specific implementation of the communication device shown in Figure 20 can be referred to the description in the aforementioned method embodiments, and will not be repeated here.

Another example is shown in Figure 21, which is a schematic diagram of the communication device 2100 involved in the above embodiments provided by the present application. Specifically, the communication device 2100 can be a communication device serving as a receiving end (or a transmitting end) as described in the above embodiments. The example shown in Figure 21 illustrates a receiving end implemented through a receiving end (or a component within a receiving end), or a transmitting end implemented through a transmitting end (or a component within a transmitting end). The structure of this communication device can be referenced from the structure shown in Figure 21.

The communication device 2100 includes at least one processor 2111 and at least one network interface 2114. Further optionally, the communication device also includes at least one memory 2112, at least one transceiver 2113, and one or more antennas 2115. The processor 2111, memory 2112, transceiver 2113, and network interface 2114 are connected, for example, via a bus. In this embodiment, the connection may include various interfaces, transmission lines, or buses, etc., and this embodiment is not limited thereto. The antenna 2115 is connected to the transceiver 2113. The network interface 2114 enables the communication device to communicate with other communication devices through a communication link. For example, the network interface 2114 may include a network interface between the communication device and core network equipment, such as an S1 interface; the network interface may also include a network interface between the communication device and other communication devices (e.g., other receivers, other transmitters, or core network equipment), such as an X2 or Xn interface.

In Figure 18, the transceiver unit 1802 can be a communication interface, which can be the network interface 2114 in Figure 21. The network interface 2114 can include an input interface and an output interface. Alternatively, the network interface 2114 can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

The processor 2111 is primarily used to process communication protocols and communication data, control the entire communication device, execute software programs, and process data from the software programs, for example, to support the actions described in the embodiments of the communication device. The communication device may include a baseband processor and a central processing unit (CPU). The baseband processor is primarily used to process communication protocols and communication data, while the CPU is primarily used to control the entire communication device, execute software programs, and process data from the software programs. The processor 2111 in Figure 21 can integrate the functions of both a baseband processor and a CPU. Those skilled in the art will understand that the baseband processor and CPU can also be independent processors interconnected via technologies such as buses. Those skilled in the art will understand that the communication device may include multiple baseband processors to adapt to different network standards, and multiple CPUs to enhance its processing capabilities. The various components of the communication device can be connected via various buses. The baseband processor can also be described as a baseband processing circuit or a baseband processing chip. The CPU can also be described as a central processing circuit or a central processing chip. The function of processing communication protocols and communication data can be built into the processor or stored in memory as a software program, which is then executed by the processor to implement the baseband processing function.

The memory is primarily used to store software programs and data. The memory 2112 can exist independently or be connected to the processor 2111. Optionally, the memory 2112 can be integrated with the processor 2111, for example, integrated within a single chip. The memory 2112 can store program code that executes the technical solutions of the embodiments of this application, and its execution is controlled by the processor 2111. The various types of computer program code being executed can also be considered as drivers for the processor 2111.

Figure 21 shows only one memory and one processor. In actual communication devices, there can be multiple processors and multiple memories. Memory can also be called storage medium or storage device, etc. Memory can be a storage element on the same chip as the processor, i.e., an on-chip storage element, or it can be a separate storage element; the embodiments of this application do not limit this.

Transceiver 2113 can be used to support the reception or transmission of radio frequency (RF) signals between a communication device and a terminal. Transceiver 2113 can be connected to antenna 2115. Transceiver 2113 includes a transmitter Tx and a receiver Rx. Specifically, one or more antennas 2115 can receive RF signals. The receiver Rx of transceiver 2113 receives the RF signals from the antennas, converts the RF signals into digital baseband signals or digital intermediate frequency (IF) signals, and provides the digital baseband signals or IF signals to processor 2111 so that processor 2111 can perform further processing on the digital baseband signals or IF signals, such as demodulation and decoding. Furthermore, the transmitter Tx in transceiver 2113 is also used to receive modulated digital baseband signals or IF signals from processor 2111, convert the modulated digital baseband signals or IF signals into RF signals, and transmit the RF signals through one or more antennas 2115. Specifically, the receiver Rx can selectively perform one or more stages of downmixing and analog-to-digital conversion on the radio frequency signal to obtain a digital baseband signal or a digital intermediate frequency (IF) signal. The order of these downmixing and IF conversion processes is adjustable. The transmitter Tx can selectively perform one or more stages of upmixing and digital-to-analog conversion on the modulated digital baseband signal or digital IF signal to obtain a radio frequency signal. The order of these upmixing and IF conversion processes is also adjustable. The digital baseband signal and the digital IF signal can be collectively referred to as digital signals.

The transceiver 2113 can also be called a transceiver unit, transceiver, transceiver device, etc. Optionally, the device in the transceiver unit that performs the receiving function can be regarded as the receiving unit, and the device in the transceiver unit that performs the transmitting function can be regarded as the transmitting unit. That is, the transceiver unit includes a receiving unit and a transmitting unit. The receiving unit can also be called a receiver, input port, receiving circuit, etc., and the transmitting unit can be called a transmitter, transmitter, or transmitting circuit, etc.

It should be noted that the communication device 2100 shown in Figure 21 can be used to implement the steps implemented by the receiving end or the sending end in the aforementioned method embodiments, and to achieve the corresponding technical effects of the receiving end or the sending end. The specific implementation of the communication device 2100 shown in Figure 21 can be referred to the description in the aforementioned method embodiments, and will not be repeated here.

This application also provides a computer-readable storage medium for storing one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor performs the method described in the possible implementations of the sending end, receiving end, or second communication device in the foregoing embodiments.

This application also provides a computer program product (or computer program) that, when executed by the processor, executes the method described above for the possible implementation of the sending or receiving end.

This application also provides a chip system including at least one processor for supporting a communication device in implementing the functions involved in the possible implementations of the communication device described above. Optionally, the chip system further includes an interface circuit that provides program instructions and/or data to the at least one processor. In one possible design, the chip system may further include a memory for storing the program instructions and data necessary for the communication device. The chip system may be composed of chips or may include chips and other discrete devices, wherein the communication device may specifically be a transmitting end or a receiving end in the aforementioned method embodiments.

This application also provides a communication system, the network system architecture of which includes the sending end or receiving end in any of the above embodiments.

Optionally, the communication system may also include other communication devices that communicate with the transmitting end or the receiving end.

In the 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 through some interfaces, indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms. Whether a function is implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

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 technical solution of this application, in essence, or the part that contributes, 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, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims (27)

A communication method, characterized in that the method is applied at a sending end, the method comprising: The sending end sends first information to the receiving end. The first information includes one or more first data and one or more first training label data. The first data includes data output from the intermediate layer of the first model. The first training label data is used to update the first model. The sending end receives the second information sent by the receiving end, the second information indicating the model update data output in reverse by the intermediate layer of the first model. The method according to claim 1, wherein the first training label data comprises: The first training label, or the index information of the first training label in the training label set, wherein the training label set includes one or more training labels. The method according to claim 1 or 2, wherein the first information further includes: Grouping indication information, wherein the grouping indication information is used to indicate the grouping method of the plurality of first data included in the first information; And/or, group number information, which indicates the number of the plurality of first data included in the first information. The method according to any one of claims 1-3 is characterized in that, The second information includes: a weighted value of multiple second data, each of the second data corresponding to one of the first data. The method according to claim 4 is characterized in that the first information further includes the weight of the first data. The method according to any one of claims 1-5, characterized in that the method further comprises: The sending end sends third information to the receiving end, the third information including any one or more of the following: The third data, the input location information of the third data in the first model, the dimension information of the third data, the identification information of the first model, or the first type indication information, wherein the first type indication information is used to indicate the data type of the third data, wherein the third data includes any one or more of the following information: first channel information, first environment information, or multimodal information. The method according to any one of claims 1-6, characterized in that the method further comprises: The sending end receives the fourth information sent by the receiving end, the fourth information including any one or more of the following: The fourth data, the second type indication information, the termination interaction instruction of the first model, the dimension information of the fourth data, or the identification information of the first model, the second type indication information is used to indicate the data type of the fourth data, and the termination interaction instruction of the first model is used to instruct the sending end to stop sending information related to the first model, wherein the fourth data includes any one or more of the following information: second channel information, second environment information, multimodal information, training residual information generated by the receiving end during the training of the first model, or the performance information of the updated first model. The method according to any one of claims 1-7, wherein the first model includes one or more segmentation locations; The first data includes: data output from intermediate layers corresponding to one or more segmentation positions in the first model. The method according to any one of claims 1-8, characterized in that the method further comprises: The sending end sends configuration information to the receiving end. The configuration information is used to configure the training task for training the first model. The configuration information includes any one or more of the following: Model information, wherein the model information includes any one or more of the following: the identification information of the first model, the structural information of the first model, the training task information of the first model, the initialization parameters of the first model, or the parameters of the first model to be updated. Segmentation information, wherein the segmentation information includes: the number of segmentation positions of the first model, and/or, the position information of the segmentation positions in the first model; Alternatively, training information, wherein the training information includes any one or more of the following: training label set information, validation set information, number of training iterations, learning rate, or loss function, wherein the training label set includes one or more training labels; And/or, the sending end receives the configuration information sent by the receiving end. The method according to claim 9, wherein the configuration information further includes: First compression configuration information, and/or, second compression configuration information, wherein: The first compression configuration information is used to indicate the compression method and compression parameters of the first information and/or the third information. The second compression configuration information is used to indicate the compression method and compression parameters of the second information and/or the fourth information. The method according to any one of claims 1-10, characterized in that, The first information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message; And/or, the second information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message. The method according to any one of claims 6-11, characterized in that, The third information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message; And/or, the fourth information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message. A communication method, characterized in that the method is applied at a receiving end, the method comprising: The receiving end receives first information sent by the sending end. The first information includes one or more first data and one or more first training label data. The first data includes data output from the intermediate layer of the first model. The first training label data is used to update the first model. The receiving end sends a second message to the sending end, the second message indicating the model update data output in reverse by the intermediate layer of the first model. The method according to claim 13, wherein the first training label data comprises: The first training label, or the index information of the first training label in the training label set, wherein the training label set includes one or more training labels. The method according to claim 13 or 14, wherein the first information further includes: Grouping indication information, wherein the grouping indication information is used to indicate the grouping method of the plurality of first data included in the first information; And/or, group number information, which indicates the number of the plurality of first data included in the first information. The method according to any one of claims 13-15 is characterized in that, The second information includes: a weighted value of multiple second data, each of the second data corresponding to one of the first data. The method according to claim 16 is characterized in that the first information further includes the weight of the first data. The method according to any one of claims 13-17, characterized in that the method further comprises: The receiving end receives third information sent by the sending end, the third information including any one or more of the following: The third data, the input location information of the third data in the first model, the dimension information of the third data, the identification information of the first model, or the first type indication information, wherein the first type indication information is used to indicate the data type of the third data, wherein the third data includes any one or more of the following information: first channel information, first environment information, or multimodal information. The method according to any one of claims 13-18, characterized in that the method further comprises: The receiving end sends fourth information to the sending end, the fourth information including any one or more of the following: The fourth data, the second type indication information, the termination interaction instruction of the first model, the dimension information of the fourth data, or the identification information of the first model, the second type indication information is used to indicate the data type of the fourth data, and the termination interaction instruction of the first model is used to instruct the sending end to stop sending information related to the first model, wherein the fourth data includes any one or more of the following information: second channel information, second environment information, multimodal information, training residual information generated by the receiving end during the training of the first model, or the performance information of the updated first model. The method according to any one of claims 13-19, wherein the first model includes one or more segmentation locations; The first data includes: data output from intermediate layers corresponding to one or more segmentation positions in the first model. The method according to any one of claims 13-20, characterized in that the method further comprises: The receiving end receives configuration information from the receiving end, the configuration information being used to configure the training task for training the first model, the configuration information including any one or more of the following: Model information, wherein the model information includes any one or more of the following: the identification information of the first model, the structural information of the first model, the training task information of the first model, the initialization parameters of the first model, or the parameters of the first model to be updated. Segmentation information, wherein the segmentation information includes: the number of segmentation positions of the first model, and/or, the position information of the segmentation positions in the first model; Alternatively, training information, wherein the training information includes any one or more of the following: training label set information, validation set information, number of training iterations, learning rate, or loss function, wherein the training label set includes one or more training labels; And/or, the receiving end sends the configuration information to the sending end. According to the method of claim 21, the configuration information further includes: First compression configuration information, and/or, second compression configuration information, wherein: The first compression configuration information is used to indicate the compression method and compression parameters of the first information and/or the third information. The second compression configuration information is used to indicate the compression method and compression parameters of the second information and/or the fourth information. The method according to any one of claims 13-22 is characterized in that, The first information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message; And/or, the second information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message. The method according to any one of claims 19-23 is characterized in that, The third information is carried in a Radio Resource Control (RRC) message or a Media Access Control (MAC) message; And/or, the fourth information is carried in a Radio Resource Control (RRC) message, or a Media Access Control (MAC) message. A communication device, characterized in that it includes a module for performing the method as described in any one of claims 1 to 24. A communication device, characterized in that it includes at least one processor, said at least one processor being configured to perform the method as described in any one of claims 1 to 24. A readable storage medium, characterized in that the storage medium stores a computer program or instructions that, when executed by a communication device, implement the method as described in any one of claims 1 to 24.