Distributed learning method and device

Abstract

The present application provides a distributed learning method and apparatus for combining wireless communication with distributed learning, which can save resources and improve the performance of distributed learning in a wireless environment. The method includes: a first node processes first data by using a first data model to obtain first intermediate data, and transmits the first intermediate data to a second node through a first channel. The first channel is updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data. The second intermediate data is a result of transmitting the first intermediate data to the second node through the first channel. The first channel is a channel between the first node and the second node.

Description

[Technical field]

[0001] This application claims priority to Chinese Patent Application No. 202110413617.6, entitled “DISTRIBUTED LEARNING METHOD AND APPARATUS,” filed with the State Intellectual Property Office of China on April 16, 2021, which is incorporated herein by reference in its entirety.

[0002] The present application relates to the field of communications, and in particular to a distributed learning method and apparatus. [Background technology]

[0003] In recent years, artificial intelligence (AI) technology has made remarkable developments in fields such as machine vision and natural language processing, and can be applied to various devices, but it also has high requirements for the capabilities of the devices. A neural network (NN) model is used as an example. In order to implement training and inference in the NN model, the device needs to have strong computing power, which is a big challenge for some devices.

[0004] In order to reduce the requirement of device capabilities, the NN model may be segmented at intermediate nodes in a specific layer or multiple specific layers of the NN model, and the multiple NN sub-models obtained are deployed on multiple devices for training and inference. Currently, it is necessary to perform operations such as channel coding on the intermediate layer information (tensor or vector) of forward propagation and the intermediate layer gradient information of backward propagation, and transmission between devices is performed through a channel. In this case, the encoding and decoding process occupies a large part of the resources, which causes resource waste. In addition, currently there is no technical solution that can improve the performance of distributed learning. Summary of the Invention

[0005] The present application provides a distributed learning method and apparatus for combining wireless communication with distributed learning, which can save resources and improve the performance of distributed learning in a wireless environment.

[0006] In order to achieve the above objectives, the following technical solutions are used in this application.

[0007] According to a first aspect, a distributed learning method is provided. The distributed learning method may be applied to a first node. The first node includes a first data model. The distributed learning method includes: processing first data by using the first data model to obtain first intermediate data, and transmitting the first intermediate data to a second node through a first channel. The first channel is updated based on error information of second intermediate data, information about the first channel, and the first intermediate data. The second intermediate data is a result of the transmission of the first intermediate data to the second node through the first channel. The first channel is a channel between the first node and the second node.

[0008] According to the distributed learning method of the first aspect, the channel is combined with the data model, and the first channel for transmitting data between the first node and the second node is used as an intermediate layer of the data model, and the first channel is optimized according to the error information of the second intermediate data, the information about the first channel, and the first intermediate data. In this way, the training of the first channel can improve the performance of distributed learning. In addition, in order to implement the integration of communication and computing, wireless transmission can be made to directly function for distributed learning, which can reduce the processing complexity and further save resources.

[0009] In a possible design manner, in a distributed learning method according to a first aspect, the first channel may include a second channel and a third channel, and the method may further include a step of updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data.

[0010] In this way, the first channel between the first node and the second node is divided into a second channel and a third channel. The second channel can be a controllable part or a controllable environment. By updating the controllable part of the first channel (i.e., the second channel), the performance of distributed learning can be improved.

[0011] In a possible design manner, the step of updating the second channel based on the error information of the second intermediate data, the information about the third channel, and the first intermediate data includes: receiving first information transmitted by the second node, and updating the second channel based on the first information and the first intermediate data, where the first information is determined based on the error information of the second intermediate data and the information about the third channel.

[0012] In this manner, to improve the performance of distributed learning, the controllable portion of the first channel (ie, the second channel) is updated based on the first information and the first intermediate data.

[0013] Optionally, the first node may receive the first information transmitted by the second node through the fourth channel. In response, the second node may transmit the first information to the first node through the fourth channel. That is, the channel between the first node and the second node in the reverse training process may be different from the channel between the first node and the second node in the forward training process, for example, in terms of frequency point or transmission mechanism.

[0014] For example, the first information may be transmitted on a conventional data / control channel or on a separate physical / logical channel dedicated to NN training. The transmission process may include operations to reduce the amount of data, such as sparsity, quantization, and / or entropy coding, or may include operations to reduce transmission errors, such as channel coding.

[0015] In a possible design manner, the distributed learning method according to the first aspect may further include transmitting a first signal to the second node through the first channel, the first signal being for determining the information about the first channel.

[0016] Optionally, the first signal may be a pilot signal, and in order to enable the second node to perform channel estimation and obtain channel information, the first node may transmit the pilot and set parameters of a controllable part of the channel, thereby updating the channel or the first data model to improve the performance of distributed learning.

[0017] In a possible design scheme, the step of updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data includes: receiving a third signal from the second node through the first channel; obtaining first information based on the third signal; and updating the first intermediate data based on the error information of the second intermediate data. Information and and updating the second channel based on the first intermediate data. The third signal is a signal obtained after a fourth signal is transmitted to the first node through the first channel, the fourth signal including a signal generated by mapping the error information of the second intermediate data to air interface resources. The first information is determined based on the error information of the second intermediate data and the information about the third channel.

[0018] In this way, the first information is obtained through over-the-air calculation by using the propagation characteristics of the signal on the air interface resource, whereby the second channel is updated based on the first channel and the first intermediate data, thereby improving the performance of distributed learning and reducing pilot overhead.

[0019] In a possible design manner, the distributed learning method according to the first aspect may further include transmitting a fifth signal to the second node through the first channel, the fifth signal including a signal generated by mapping third intermediate data to an air interface resource, the third intermediate data being for updating the first channel.

[0020] In this way, based on the channel reciprocity and by using the propagation characteristics of the signals on the air interface resources, the second node is able to obtain the third intermediate data and channel information through over-the-air calculation to avoid channel estimation, thereby reducing pilot overhead.

[0021] In a possible design manner, the distributed learning method according to the first aspect may further include updating the first data model based on error information of the first intermediate data to obtain a new first data model. In this way, the first node may update the first data model to improve the performance of distributed learning.

[0022] In a possible design manner, the distributed learning method according to the first aspect may further include receiving second information transmitted by the second node, the second information being for obtaining the error information of the first intermediate data. Thus, in order to improve the performance of distributed learning, the first node may update a first data model based on the error information of the first intermediate data.

[0023] In a possible design manner, the second information includes the error information of the first intermediate data. Alternatively, the second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data. In other words, the second information may be determined by the second node. Alternatively, the second information may be obtained through over-the-air calculation, which can save resources.

[0024] In a possible design manner, the distributed learning method according to the first aspect may further include: receiving an eighth signal from the second node; and obtaining the error information of the first intermediate data based on the eighth signal. Optionally, the eighth signal is a signal obtained after a seventh signal is transmitted to the first node through a channel, and the seventh signal includes a signal generated by mapping the error information of the second intermediate data to an air interface resource.

[0025] In this way, over-the-air computation of gradients of the intermediate layers in the first data model is implemented through processing of over-the-air signal propagation, which can reduce pilot overhead.

[0026] According to a second aspect, a method for distributed learning is provided. The method for distributed learning may be applied to a second node. The second node includes a second data model. The method for distributed learning may include: receiving second intermediate data through a first channel; and processing the second intermediate data by using the second data model to obtain output data. The second intermediate data is a result of a transmission of the first intermediate data sent by a first node through the first channel to a second node. The first channel is updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data. The first channel is a channel between the first node and the second node.

[0027] Optionally, the second node may determine a new channel based on the error information of the second intermediate data, the channel information of the channel, and the first intermediate data.

[0028] In a possible design manner, in the distributed learning method according to the second aspect, the first channel includes a second channel and a third channel, and the method may further include a step of transmitting the error information of the second intermediate data and information about the third channel to the first node.

[0029] In a possible design manner, the step of transmitting the error information of the second intermediate data and the information about the third channel to the first node includes transmitting first information to the first node, where the first information is determined based on the error information of the second intermediate data and the information about the third channel.

[0030] In a possible design manner, the distributed learning method according to the second aspect may further include: receiving a second signal from the first node through the first channel; and obtaining the information about the first channel based on the second signal, the second signal being a signal obtained after a first signal is transmitted to the second node through the first channel, the first signal being for determining the information about the first channel.

[0031] In a possible design manner, the step of transmitting the error information of the second intermediate data and information about the third channel to the first node includes transmitting a fourth signal to the first node through the first channel, the fourth signal including a signal generated by mapping the error information of the second intermediate data to air interface resources.

[0032] In a possible design manner, the distributed learning method according to the second aspect may further include: receiving a sixth signal from the first node through the first channel; acquiring third information based on the sixth signal; and updating the second channel based on the third information and the error information of the second intermediate data. The sixth signal is a signal acquired after a fifth signal is transmitted to the second node through the first channel. The fifth signal includes a signal generated by mapping third intermediate data to an air interface resource. The third intermediate data is for updating the first channel. The third information is determined based on the third intermediate data and the information about the third channel.

[0033] In a possible design manner, the distributed learning method according to the second aspect may further include transmitting second information to the first node, the second information being for obtaining the error information of the first intermediate data.

[0034] In a possible design manner, the second information includes the error information of the first intermediate data. Alternatively, the second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data.

[0035] In a possible design manner, the distributed learning method according to the second aspect may further include updating the second data model based on the output data to obtain a new second data model.

[0036] In a possible design manner, the distributed learning method according to the second aspect may further include transmitting a seventh signal to the first node. Optionally, the seventh signal includes a signal generated by mapping the error information of the second intermediate data to an air interface resource.

[0037] According to a third aspect, there is provided a distributed learning apparatus. The distributed learning apparatus includes a first data model. The apparatus includes a processing module and a transceiving module.

[0038] The processing module is configured to process the first data by using a first data model to obtain first intermediate data. The transceiver module is configured to transmit the first intermediate data to a second node through a first channel. The first channel is updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the second intermediate data being a result of the transmission of the first intermediate data to the second node through the first channel, the first channel being a channel between the distributed learning device and the second node.

[0039] In a possible design manner, the processing module is further configured to update the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data.

[0040] In a possible design manner, the transceiver module is further configured to receive first information transmitted by the second node, and the processing module is further configured to update the second channel based on the first information and the first intermediate data, the first information being determined based on the error information of the second intermediate data and the information about the third channel.

[0041] In a possible design manner, the transceiver module is further configured to receive a third signal from the second node through the first channel, the third signal being a signal obtained after a fourth signal is transmitted to the distributed learning device through the first channel, and the fourth signal includes a signal generated by mapping the error information of the second intermediate data to an air interface resource.

[0042] The processing module is further configured to obtain first information based on the third signal, the first information being determined based on the error information of the second intermediate data and the information about the third channel.

[0043] The processing module is further configured to update the second channel based on the first information and the first intermediate data.

[0044] In one possible design, the transceiver module is further configured to transmit a first signal to the second node over the first channel, the first signal being for determining the information about the first channel.

[0045] In a possible design manner, the transceiver module is further configured to transmit the fifth signal to the second node through the first channel, the fifth signal including a signal generated by mapping third intermediate data to air interface resources, the third intermediate data for updating the first channel.

[0046] In a possible design manner, the processing module is further configured to update the first data model based on error information of the first intermediate data to obtain a new first data model.

[0047] In a possible design manner, the transceiver module is further configured to receive second information sent by the second node, the second information being for obtaining the error information of the first intermediate data.

[0048] In a possible design manner, the second information includes the error information of the first intermediate data. Alternatively, the second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data.

[0049] It should be noted that the transceiver module in the third aspect may include a receiving module and a transmitting module. The receiving module is configured to receive data and / or signaling from the second node. The transmitting module is configured to transmit the data and / or signaling to the second node. The specific implementation form of the transceiver module is not specifically limited in this application.

[0050] Optionally, the distributed learning device according to the third aspect may further include a storage module, which stores a program or instructions, and when the processing module executes the program or instructions, the distributed learning device according to the third aspect is capable of executing the method according to the first aspect.

[0051] It should be noted that the distributed learning device according to the third aspect may be the first node, or may be, for example, a chip (system) or another component or assembly that may be located in the first node, which is not a limitation in this application.

[0052] In addition, for the technical effect of the distributed learning device according to the third aspect, please refer to the technical effect of the distributed learning method according to any possible implementation form in the first aspect, and the details will not be described again in this specification.

[0053] According to a fourth aspect, there is provided a distributed learning apparatus, the distributed learning apparatus including a second data model, the apparatus including a processing module and a transceiving module.

[0054] The transceiver module is configured to receive second intermediate data through a first channel, the second intermediate data being a result of a transmission of the first intermediate data sent by a first node through the first channel to the distributed learning device, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the first channel being a channel between the first node and the distributed learning device.

[0055] The processing module is configured to process the second intermediate data by using the second data model to obtain output data.

[0056] In a possible design manner, the first channel includes a second channel and a third channel, and the transceiver module is further configured to transmit the error information of the second intermediate data and information about the third channel to the first node.

[0057] In a possible design manner, the transceiver module is further configured to transmit first information to the first node, the first information being determined based on the error information of the second intermediate data and the information about the third channel.

[0058] In a possible design, the transceiver module is further configured to receive a second signal from the first node over the first channel, the second signal being a signal transmitted in response to a first signal transmitted over the first channel. Distributed learning device and a signal obtained after transmission to the first channel, the first signal being for determining the information about the first channel. The processing module is further configured to obtain the information about the first channel based on the second signal.

[0059] In a possible design manner, the transceiver module is further configured to transmit a fourth signal to the first node through the first channel, the fourth signal including a signal generated by mapping the error information of the second intermediate data to an air interface resource.

[0060] In a possible design manner, the transceiver module is further configured to receive a sixth signal from the first node through the first channel, the sixth signal being a signal obtained after a fifth signal is transmitted to the distributed learning device through the first channel, the fifth signal including a signal generated by mapping third intermediate data to air interface resources, the third intermediate data being for updating the first channel.

[0061] The processing module is further configured to obtain third information based on the sixth signal, the third information being determined based on the third intermediate data and the information about the third channel.

[0062] The processing module is further configured to update the second channel based on the third information and the error information of the second intermediate data.

[0063] In a possible design manner, the transceiver module is further configured to send second information to the first node, the second information being for obtaining the error information of the first intermediate data.

[0064] In a possible design manner, the second information includes the error information of the first intermediate data. Alternatively, the second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data.

[0065] In a possible design scenario, the processing module is further configured to update the second data model based on the output data to obtain a new second data model.

[0066] It should be noted that the transceiver module in the fourth aspect may include a receiving module and a transmitting module. The receiving module is configured to receive data and / or signaling from the first node. The transmitting module is configured to transmit the data and / or signaling to the first node. The specific implementation form of the transceiver module is not specifically limited in this application.

[0067] Optionally, the distributed learning device according to the fourth aspect may further include a storage module, which stores a program or instructions, and when the processing module executes the program or instructions, the distributed learning device according to the fourth aspect is capable of executing the method according to the second aspect.

[0068] It should be noted that the distributed learning device according to the fourth aspect may be the second node, or may be, for example, a chip (system) or another component or assembly that may be located in the second node, and this is not a limitation in the present application.

[0069] In addition, for the technical effect of the distributed learning device according to the fourth aspect, please refer to the technical effect of the distributed learning method according to any possible implementation form in the first aspect, and the details will not be described again in this specification.

[0070] According to a fifth aspect, there is provided an apparatus for distributed learning, the apparatus for distributed learning including a processor coupled to a memory, the memory configured to store a computer program.

[0071] The processor is configured to execute the computer program stored in the memory to enable the distributed learning device to perform the distributed learning method according to the executable implementation form of the first and second aspects.

[0072] In a possible design, the distributed learning apparatus according to the fifth aspect may further include a transceiver. The transceiver may be a transceiver circuit or an input / output port. The transceiver may be configured for the distributed learning apparatus to communicate with another device.

[0073] It should be noted that the input port may be configured to implement the receiving functionality included in the first and second aspects, and the output port may be configured to implement the transmitting functionality included in the first and second aspects.

[0074] In the present application, the distributed learning device according to the fifth aspect may be a first node or a second node, or a chip or chip system disposed inside the first node or the second node.

[0075] In addition, for the technical effects of the distributed learning device according to the fifth aspect, please refer to the technical effects of the distributed learning method according to any implementation form in the first and second aspects, and the details will not be described again in this specification.

[0076] According to a sixth aspect, there is provided a communication system, the communication system including a first node and a second node.

[0077] According to a seventh aspect, a chip system is provided. The chip system includes a processor and an input / output port. The processor is configured to implement the processing functions in the first and second aspects, and the input / output port is configured to implement the transmission and reception functions in the first and second aspects. Specifically, the input port may be configured to implement the reception functions included in the first and second aspects, and the output port may be configured to implement the transmission functions included in the first and second aspects.

[0078] In a possible design, the chip system further includes a memory, the memory configured to store program instructions and data for implementing the functions in the first and second aspects.

[0079] A chip system may include a chip, or may include a chip and other discrete components.

[0080] According to an eighth aspect, there is provided a computer readable storage medium comprising a computer program or instructions which, when executed on a computer, performs the distributed learning method according to the executable implementations of the first and second aspects.

[0081] According to a ninth aspect, there is provided a computer program product comprising a computer program or instructions which, when executed on a computer, perform the distributed learning method according to the executable implementations of the first and second aspects.

[0082] According to a tenth aspect, there is provided a computer program which, when run on a computer, performs the method according to any implementation of the first and second aspects. [Brief description of the drawings]

[0083] [Figure 1] FIG. 1 is a schematic diagram of the architecture of a communication system according to an embodiment of the present application;

[0084] [Diagram 2] FIG. 1 is a schematic diagram of a fully connected neural network according to an embodiment of the present application.

[0085] [Diagram 3] FIG. 1 is a schematic diagram of gradient descent according to an embodiment of the present application;

[0086] [Figure 4] FIG. 2 is a schematic diagram of neural network training according to an embodiment of the present application.

[0087] [Diagram 5] FIG. 2 is a schematic diagram of a segmentation of a type of neural network according to an embodiment of the present application.

[0088] [Figure 6] FIG. 2 is a schematic diagram of another type of segmentation of a neural network according to an embodiment of the present application.

[0089] [Figure 7] FIG. 13 is a schematic diagram of yet another type of segmentation of a neural network according to an embodiment of the present application.

[0090] [Figure 8] FIG. 2 is a schematic diagram of interactions of a distributed learning method according to an embodiment of the present application.

[0091] [Figure 9A] FIG. 2 is a schematic diagram of the application of a distributed learning method according to an embodiment of the present application; [Figure 9B] FIG. 2 is a schematic diagram of the application of a distributed learning method according to an embodiment of the present application;

[0092] [Figure 10A] FIG. 2 is a schematic diagram of the application of another distributed learning method according to an embodiment of the present application; [Figure 10B] FIG. 2 is a schematic diagram of the application of another distributed learning method according to an embodiment of the present application;

[0093] [Figure 11A] FIG. 13 is a schematic diagram of the application of yet another distributed learning method according to an embodiment of the present application; [Figure 11B] FIG. 13 is a schematic diagram of the application of yet another distributed learning method according to an embodiment of the present application;

[0094] [Figure 12(a)] FIG. 2 is a schematic diagram of interactions of another distributed learning method according to an embodiment of the present application; [Figure 12(b)] FIG. 2 is a schematic diagram of interactions of another distributed learning method according to an embodiment of the present application; [Figure 12(c)] FIG. 2 is a schematic diagram of interactions of another distributed learning method according to an embodiment of the present application;

[0095] [Figure 13] 1 is a schematic diagram of the structure of a distributed learning device according to an embodiment of the present invention;

[0096] [Figure 14] FIG. 2 is a schematic diagram of the structure of another distributed learning device according to an embodiment of the present application; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0097] The technical solutions of the present application are described below with reference to the accompanying drawings.

[0098] The technical solutions in the embodiments of the present application may be applied to various communication systems, such as wireless fidelity (Wi-Fi®) systems, vehicle to everything (V2X) communication systems, device-to-device (D2D) communication systems, Internet of vehicles communication systems, short-range wireless communication systems, satellite communication systems, NarrowBand-Internet of Things (NB-IoT) systems, long-term evolution (LTE) systems, 5th generation (5G) mobile communication systems, such as new radio (NR) systems, and future communication systems, such as 6th generation (6G) mobile communication systems. The technical solutions in the embodiments of the present application may be applied to the following application scenarios: enhanced mobile broadband (eMBB), URLLC (ultra-reliable low-latency communication, URLLC), and enhanced machine-type communication (eMTC).

[0099] All aspects, embodiments, or features are presented herein by describing a system that may include multiple devices, components, modules, etc. It is expressly understood that each system may include other devices, components, modules, etc. and / or may not include all devices, components, modules, etc. described with reference to the accompanying drawings. Additionally, combinations of these solutions may be used.

[0100] In addition, in the embodiments of the present application, terms such as "example" and "for example" denote serving as an example, illustration, or illustration. Any embodiment or design manner described in the present application as an "example" should not be described as preferred or having more advantages over another embodiment or design manner. Strictly speaking, use of the term "example" is intended to present concepts in a concrete manner.

[0101] The network architecture and service scenarios described in the embodiments of the present application are intended to more clearly explain the technical solutions in the embodiments of the present application, but do not constitute any limitations on the technical solutions provided in the embodiments of the present application. Those skilled in the art may learn that the technical solutions provided in the embodiments of the present application can also be applied to similar technical problems as the network architecture deepens and new service scenarios emerge.

[0102] In order to facilitate understanding of the embodiments of the present application, the following describes in detail a communication system applicable to the embodiments of the present application by using the communication system shown in Fig. 1 as an example. For example, Fig. 1 is a schematic diagram of the architecture of a communication system to which the distributed learning method according to an embodiment of the present application can be applied.

[0103] The communication system includes a first node and a second node. The first node may be a terminal device, and the second node may be a network device. Alternatively, the first node may be a network device, and the second node may be a terminal device. Alternatively, the first node may be a terminal device, and the second node may be a terminal device. As shown in FIG. 1, the communication system includes a terminal device, and there may be one or more terminal devices. The communication system may further include a network device.

[0104] A network device is a device located on the network side of a communication system and having radio transmission and reception capabilities, or a chip or chip system that may be located within a device. The network device may include, but is not limited to, an access point (AP) in a wireless fidelity (Wi-Fi) system, such as a home gateway, router, server, switch, or bridge, an evolved NodeB (eNB), a radio network controller (RNC), a home base station (e.g., a home evolved NodeB or home NodeB (HNB)), a wireless relay node, a wireless backhaul node, or a transmission point (transmission and reception point (TRP) or transmission point (TP)), or a gNB or transmission point (TRP or TP) in 5G, such as an NR system, or one antenna panel or a group of antenna panels (including multiple antenna panels) of a base station in a 5G system, or a network node forming a gNB or transmission point, such as a base band unit (BBU), a distributed unit (DU), or a road side unit (RSU) having a base station function or the like.The network device may alternatively be a radio controller in a cloud radio access network (CRAN) scenario, or a network device in a future evolved public land mobile network (PLMN), or a wearable or in-vehicle device, and further includes devices acting as a base station in device-to-device (D2D), vehicle-to-everything (V2X), machine-to-machine (M2M) communications, or Internet of Things communications.

[0105] A terminal device is a terminal that accesses a communication system and has radio transmission and reception capabilities, or a chip or chip system that may be located in a terminal. A terminal device may also be called User Equipment (UE), user equipment, access terminal, subscriber unit, subscriber station, mobile console, mobile station (MS), remote station, remote terminal, mobile device, user terminal, terminal, terminal unit, terminal station, terminal equipment, wireless communication device, user agent, or user device. For example, the terminal device in the embodiments of the present application may be a mobile phone, a wireless data card, a personal digital assistant (PDA) computer, a laptop computer, a tablet computer (Pad), a computer with wireless transmission and reception capabilities, a machine-type communication (MTC) terminal, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, an Internet of Things (IoT) terminal device, a wireless terminal in industrial control, a wireless terminal in unmanned driving (self-driving), a wireless terminal in remote medical, a wireless terminal in a smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, a wireless terminal in a smart home (e.g., a game console, a smart TV, a smart speaker, a smart refrigerator, and a fitness device), an in-vehicle terminal, or an RSU with a terminal function.An access terminal may be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a handheld device (handset) with wireless communication capabilities, a computing device, another processing device connected to a wireless modem, or a wearable device. As another example, the terminal device in the embodiment of the present application may be an express terminal in smart logistics (e.g., a device capable of monitoring the location of a vehicle carrying an item, or a device capable of monitoring the temperature and humidity of an item), a wireless terminal in smart agriculture (e.g., a wearable device capable of collecting relevant data of livestock and poultry), a wireless terminal in smart buildings (e.g., a smart elevator, a fire monitoring device, and a smart meter), a wireless terminal in smart medical care (e.g., a wearable device capable of monitoring the physiological status of a person or an animal), a wireless terminal in intelligent transportation (e.g., a smart bus, a smart vehicle, a shared bicycle, a charging pile monitoring device, a smart traffic light, a smart monitoring device, and a smart parking device), or a wireless terminal in smart retail (e.g., a vending machine, a self-checkout machine, and an unmanned convenience store). As another example, the terminal device in the present application may be an on-board module, an on-board module, an on-board component, an on-board chip, or an on-board unit built in a vehicle as one or more components or units. A vehicle is provided herein through an on-board module, an on-board component, an on-board chip, or an on-board unit. A distributed learning method It can be implemented.

[0106] It should be noted that the distributed learning method provided in the embodiment of the present application can be applied between any two nodes shown in Fig. 1, for example, between terminal devices, and between a terminal device and a network device. For specific implementation, please refer to the following method embodiment. Details will not be described again in this specification.

[0107] It should be noted that the solutions in the embodiments of the present application may be further applied to other communication systems, and the corresponding names may also be replaced with the names of corresponding functions in other communication systems.

[0108] It should be understood that Fig. 1 is only a simplified schematic diagram of an example for ease of understanding. The communication system may further include other devices not shown in Fig. 1.

[0109] In addition, those skilled in the art may learn that with the evolution of network architecture and the emergence of new service scenarios, the technical solutions provided in the embodiments of the present application can also be applied to similar technical problems.

[0110] In order to make the embodiments of the present application clearer, some contents and concepts related to the embodiments of the present application are described together below.

[0111] 1. Neural Networks

[0112] A neural network is an algorithmic network that can perform learning, summarization, and conclusions, and can be constructed in a computing node in the form of neural network software or hardware, for example, a neural network training program or executable script. In general, a deep neural network (DNN) model is used as an example. A DNN model includes multiple layers of neurons (operators). Each layer has multiple inputs and multiple outputs. The inputs or outputs are multidimensional arrays, also called tensors. Each layer has one or more weighting values, called weights. The output result of a particular layer, also called an eigenvalue, is equal to the result of a mathematical operation, such as multiplication of the input and the weight of the layer, and is usually related to a matrix multiplication operation.

[0113] A fully-connected neural network is a neural network, and a fully-connected neural network is also called a multi-layer perceptron (MLP). An MLP includes an input layer, an output layer, and multiple hidden layers (also called intermediate layers), each of which includes multiple neurons. The neurons in two adjacent layers of neurons are connected in pairs, as shown in FIG. 2.

[0114] For two adjacent layers of neurons, the output of a neuron in the lower layer, h, is the weighted sum of all neurons x in the upper layer that are connected to the neurons in the lower layer, passed through an activation function. The matrix can be expressed as the following equation: h=f(w×x+b), where w is the weight matrix, b is the offset vector, and f() is the activation function. The output of the neural network, y, is then expressed as y=f t (w t ×f t-1 (...)+b t ), where t is the number of layers of neurons included in the fully-connected neural network.

[0115] 2. Neural network training

[0116] A neural network may be understood as a mapping relationship between a set of input data and a set of output data. In general, a neural network may be randomly initialized, and the process of obtaining the mapping relationship based on the random number w and the random number b by using existing data may be referred to as neural network training.

[0117] For example, the training mode includes evaluating the output data of the neural network by using a loss function to obtain error information and back-propagating it. As shown in Figure 3, the weight matrix w and the deviation vector b can be iteratively optimized by a gradient descent method, and the optimal w and optimal b can be obtained when the value of the loss function reaches a minimum value.

[0118] For example, the gradient descent process can be expressed as equation (1):

number

[0119] The parameters (e.g., weights w and offsets b) can be optimized by using the above formula (1), where θ′ is the value of the parameter after optimization (which may also be referred to as the updated value), θ is the value of the parameter before optimization, L is a loss function, η is a learning rate, which may be for controlling the stride of the gradient descent, and the mathematical symbol ← denotes allocation from the right side to the left side.

[0120] Referring to FIG. 4, in the backpropagation process, the chain rule may be used to obtain partial derivatives; specifically, the gradient of the parameters of a neuron in the previous layer (layer j) may be calculated from the gradient of the parameters of a neuron in the next layer (layer i) by recursion, as shown in Equation (2) below.

number

[0121] In the above equation (2), L is the loss function, and w ij is the weight of the connection between neuron i and neuron j, and s i is the weighted sum of the inputs to neuron i, i.e.,

number

number

number

number

[0122] 3. Segmenting Neural Networks

[0123] Segmentation of Neural Network: The neural network is segmented into multiple sub-neural networks at a certain hidden layer based on the computation power and communication capability of the node, and the sub-neural networks are deployed on multiple nodes respectively.

[0124] Referring to Fig. 5, the first data model and the second data model are obtained after the data model is segmented in the intermediate layer. In the forward process of training or inference, the first node processes the input data by using the first data model to obtain intermediate data, and transmits the intermediate data to the second node through a channel. Then, the second node processes the intermediate data by using the second data model to obtain output data. In the backward process of training, the second node determines error information of the second data model based on the output data, updates parameters of the second data model, and sends the error information of the second data model to the first node. The first node determines error information of the first data model based on the error information of the second data model, and updates parameters of the first data model.

[0125] 4. Parameters of the first data model, parameters of the second data model, parameters of the first channel, and information about the first channel

[0126] Parameter of the first data model: Parameter θ of the first data model 1 may be weights and / or biases corresponding to the first data model.

[0127] Parameter of the second data model: Parameter θ of the second data model 2 may be weights and / or biases corresponding to the second data model.

[0128] In this embodiment of the present application, the first channel may include a controllable portion and may further include an uncontrollable portion. The controllable portion may include a controllable environment, and the uncontrollable portion may include an uncontrollable environment between two nodes. Alternatively, the first channel may include a trainable channel influenced by the controllable portion and may further include a direct channel formed by the uncontrollable portion.

[0129] Controllable Environment: To adjust the wireless channel environment, controllable units are deployed in the environment.

[0130] For example, the controllable units may include active devices such as relays, distributed antennas, or active intelligent reflective surfaces. Alternatively, the controllable units may include passive devices such as passive intelligent reflective surfaces.

[0131] 6 or 7, the first channel may include a second channel and may further include a third channel. In the present embodiment, the controllable portion or a trainable channel influenced by the controllable portion is referred to as the second channel, and the uncontrollable portion or a channel formed by the uncontrollable portion is referred to as the third channel.

[0132] Fig. 6 is a schematic diagram of the application of a distributed learning method according to an embodiment of the present application. Fig. 7 is a schematic diagram of the application of another distributed learning method according to an embodiment of the present application. As shown in Fig. 6 or Fig. 7, a first node includes a first data model, a second node includes a second data model, and the first node communicates with the second node through a first channel. Fig. 6 differs from Fig. 7 mainly in the display form of the first channel.

[0133] For example, the first channel includes a controllable portion and an uncontrollable portion, or the first channel includes a trainable channel influenced by the controllable portion and a channel formed by the uncontrollable portion. Referring to FIG. 6 and FIG. 7, the first channel between the first node and the second node is g 1 ×Φ×g 2 +h or

number

[0134] For example, the nth element of the controllable part's response Φ can be expressed as equation (3):

number

[0135] In the embodiment of the present application, the parameters of the first channel are as follows: g 1 , g 2 The parameters of the second channel may include one or more of g, h, and Φ. The parameters of the second channel may include Φ. The information about the first channel may include g 1 , g 2 , h, and Φ. The information about the third channel may include information obtained based on one or more of g 1 , g 2 , and h.

[0136] In the prior art, communication transmission is designed for lossless transmission. In order to ensure information error-free transmission, it is necessary to perform channel encoding on the information transmitted between the first node and the second node in the forward process and the reverse process, and it is necessary to decode the received information.

[0137] Through research, the applicant has found that the characteristics of the neural network and training method with excess parameters (i.e., the number of parameters is greater than the number of training samples), which is represented by the random gradient descent of the neural network, allow the neural network to have good error tolerance ability. Therefore, in the distributed learning process, the requirement for lossless transmission of information is low. However, the encoding and decoding process occupies a large part of the resources, which leads to resource waste. The design of separating the neural network learning from the communication brings about redundancy in wireless resource utilization. In addition, this solution cannot improve the performance of distributed learning.

[0138] In order to save resources and improve the performance of distributed learning, this application proposes to combine the channel with distributed learning and perform joint training on the channel and the neural network by treating the channel as a part of the neural network.

[0139] The distributed training method provided in the embodiments of the present application can be applied to multiple waveform systems, such as single carrier, discrete Fourier transform spread orthogonal frequency division multiplexing (DFT-s-OFDM) system, cyclic prefix orthogonal frequency division multiplexing (SC-OFDM) system, and orthogonal time-frequency space (OTFS) system. The wireless channel response of the systems may be modeled as a linear system. The main difference between the systems is the different resources. For example, single carrier defines symbols in the time domain, while OTFS etc. can be considered as performing further linear transformation on the symbols.

[0140] The distributed learning method provided in the embodiments of the present application can be applied to scenarios with one or multiple antennas, where the mapping (i.e., precoding) between neurons and antennas is also a linear transformation.

[0141] The distributed learning method provided in the embodiment of the present application can be applied to single AI task and multi-task multi-view scenario. In the multi-task scenario, there may be multiple second data models configured for multiple inference tasks. The multiple second data models are updated separately. The first data model and the controllable environment need to be updated by integrating (e.g., weighting) the middle layer error information of the multiple second data models. In the multi-view scenario, there may be multiple first data models configured to process data of different views for subsequent inference of the second data model. The update of the controllable environment is related to the transmission manner of the middle layer inference output of the multiple first data models. For example, when orthogonal / non-orthogonal multiple access is used, the multiple first data models may be updated separately. The distributed learning method provided in the embodiment of the present application is described by using an example in which the distributed learning method is applied to a single AI task.

[0142] The technical solution of this solution may be applied in a multi-hop network scenario, i.e. when the neural network is segmented into three or more.

[0143] The distributed learning method provided in the embodiment of the present application will be described in detail below with reference to Figures 8 to 12(a) and 12(b). Figures 9A to 11B are schematic diagrams of the application of the distributed learning method according to an embodiment of the present application. When the methods shown in Figures 8, 12(a), 12(b), and 12(c) are described, the contents of Figures 9A to 11B are described as examples.

[0144] For example, Figure 8 is a schematic diagram of the interaction of the distributed learning method according to one embodiment of the present application. OFDM system is used as an example for illustration. The distributed learning method may be applied to the communication between any two nodes shown in Figure 1.

[0145] As shown in FIG. 8, the distributed learning method includes the following steps.

[0146] S801: A first node processes first data by using a first data model to obtain first intermediate data.

[0147] For example, a first node includes a first data model, and the first data model is constructed within the first node.

[0148] For example, the first data model may be neural network software. The first data model may be constructed in the first node in the form of neural network software or hardware, for example, a neural network training program or executable script.

[0149] Optionally, the first data may be data in a training sample or data during inference. S801 may be applied to a training process or an inference process.

[0150] Referring to FIG. 6 or 7, a first node processes first data x based on a first data model to output first intermediate data z.

[0151] Referring to step (a) in FIG. 9A to FIG. 11B, the first data is x, and the first node is a node with a parameter θ 1 , processing x by using a first data model where

number

number

[0152] S802: A first node transmits first intermediate data to a second node through a first channel, and in response, the second node receives second intermediate data from the first node through the first channel.

[0153] For example, the second intermediate data is a result of transmission of the first intermediate data over the first channel to the second node.

[0154] Referring to Figure 6 or Figure 7, a first node transmits first intermediate data z through a first channel, and after the first intermediate data z passes through the first channel, second intermediate data c is output, and a second node receives the second intermediate data c.

[0155] In some embodiments, the second node may transmit a first channel (e.g., g 2 ×Φ×g 1 +h) from the first node to the first intermediate data (e.g.,

number

number

[0156] Referring to step (b) in FIG. 9A to FIG. 11B, the second node 100 transmits a first channel (e.g.,

number

number

number

[0157] For example, the first channel is updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data.

[0158] Optionally, the error information of the second intermediate data may include a gradient of the second intermediate data, or a normalized value of the gradient of the second intermediate data.

[0159] Optionally, the information about the first channel is one of the following: 1 , g 2 , h, and Φ.

[0160] For a specific implementation of updating the first channel, please refer to S1202a and S1202b shown in Figure 12(a), Figure 12(b), and Figure 12(c). The details will not be described again in this specification.

[0161] In this embodiment of the present application, the channel between the first node and the second node (i.e., the first channel) is trained, and the channel is used as an intermediate layer (e.g., residual layer) of the data model involved in training and inference, so that functions such as filtering or feature extraction required by DNN can be implemented in the wireless transmission process, thereby improving the performance of distributed learning. In addition, wireless transmission is enabled to function directly for distributed learning, thereby reducing the processing complexity and further saving resources.

[0162] Optionally, the values ​​of the parameters of the first data model are values ​​obtained after a previous training. The values ​​of the parameters of the first channel are values ​​obtained after a previous training.

[0163] That is, in the forward training process or inference process, the values ​​of the parameters of the first data model may be updated values ​​after previous training or initial values ​​(e.g., the first data model has not been trained), and the values ​​of the parameters of the first channel may be updated values ​​after previous training or initial values ​​(e.g., the first channel has not been trained).

[0164] In some embodiments, in S802, the first node transmitting the first intermediate data to the second node over the first channel may include: the first node modulating the first intermediate data into symbols and transmitting the symbols to the second node over the first channel.

[0165] For example, the first node may map the first intermediate data to an air interface resource, and send the first intermediate data to the second node, In this way, channel encoding and decoding is not required in the data transmission process, and resources can be saved.

[0166] For example, the number of modulated symbols is the same as the dimension of the intermediate layer of the data model.

[0167] For example, the symbol value and the hidden layer value may satisfy the following formula: s=z or s=W×z, that is, a certain linear transformation is satisfied between the symbol value and the hidden layer value, where s is the symbol value, z is the hidden layer value, and W is a linear transformation matrix. This may correspond to the case where the data model is a complex data model, that is, the neural network is a complex neural network. In the following, s=z is used as an example for explanation.

[0168] Referring to FIG. 6 or FIG. 7, when the dimension of the intermediate layer of the data model is four, the number of symbols is also four.

[0169] As another example, the number of modulated symbols is half the dimension of the hidden layer of the data model. In other words, the output of two neurons forms a complex-valued symbol, or a certain linear transformation can be satisfied between the value of the symbol and the value of the hidden layer.

[0170] For example, the following equation between the symbol value and the intermediate layer value:

number

[0171] In some embodiments, receiving the second intermediate data from the first node over the first channel by the second node at S802 may include: the second node performing dimensional matching on symbols of the received second intermediate data.

[0172] In this way, the second node does not need to perform decoding in the process of receiving the data, which can save resources.

[0173] Optionally, the second node may perform dimension matching depending on whether the data model is a complex data model.

[0174] For example, the number of demodulated symbols is equal to the dimension of the intermediate layer of the data model, which corresponds to the case where the data model is a complex data model.

[0175] As another example, one complex-valued symbol of the second intermediate data is expanded to two real numbers, which corresponds to the case where the data model is a real data model.

[0176] It should be noted that a specific implementation form in which the second node performs dimension matching depending on whether the data model is a complex data model is similar to a corresponding implementation form in which the first node modulates the first intermediate data into symbols and transmits the symbols to the second node through a first channel, and the details will not be described again in this specification.

[0177] It should be noted that the second node performing dimensional matching on the symbols of the received second intermediate data may be performed before the second data model of the second node processes the second intermediate data, e.g., may be performed before S803.

[0178] S803: A second node processes the second intermediate data by using the second data model to obtain output data.

[0179] For example, the second node includes a second data model. Similar to the first data model, the second data model may be constructed within the second node.

[0180] For example, the second data model may be neural network software and may be constructed in the second node in the form of neural network software or hardware, for example, a neural network training program or executable script.

[0181] In some embodiments, the second node has a parameter θ 2 The second intermediate data is generated by using a second data model

number

number

number

[0182] Referring to step (c) in FIG. 9A to FIG. 11B, using the antenna of the second node as an example, the second node 2 A second data model by using the second intermediate data

number

number

[0183] In some embodiments, the parameter θ 2 By using the second data model, the M antennas of the second node receive the second intermediate data

number

number

[0184] Optionally, a parameter θ of the second data model of the second node 2 The value of may be an updated value after a previous training.

[0185] That is, in the forward training process or inference process, the values ​​of the parameters of the second data model may be updated values ​​after previous training or initial values ​​(e.g., the second data model has not been trained).

[0186] By using the above-mentioned distributed learning method, the channel is combined with the data model, and the first channel for transmitting data between the first node and the second node is used as an intermediate layer of the data model, and the first channel is trained. The first channel is optimized based on the error information of the second intermediate data, the information about the first channel, and the first intermediate data, thereby improving the performance of distributed learning. In addition, in order to implement the integration of communication and computing, wireless transmission is enabled to directly function for distributed learning, thereby reducing the processing complexity and further saving resources.

[0187] It should be noted that S801-S803 may be a forward training process or an inference process. Optionally, in the inference process, the second node may feed back the acquired output data to the first node.

[0188] Optionally, the second node may determine a loss function based on the output results.

[0189] For example, the loss function is the following formula:

number

[0190] In a possible design manner, the distributed learning method provided in the embodiment of the present application may further include an inverse training process as shown in Figure 12(a), Figure 12(b), and Figure 12(c), including S1201, S1202a, S1202b, and S1203 to S1207. The distributed learning method shown in Figure 12(a), Figure 12(b), and Figure 12(c) may be used in combination with the distributed learning method shown in Figure 8.

[0191] S1201: The second node updates the second data model based on the output data to obtain a new second data model.

[0192] In some embodiments, S1201 may include the following steps 1 and 2.

[0193] Phase 1: A second node obtains an updated value of a parameter of a second data model.

[0194] Optionally, the second node may obtain updated values ​​of the parameters of the second data model by using error information of the parameters of the second data model, for example, the error information may include a gradient, or a normalized value of the gradient.

[0195] For example, error information for a parameter of the second data model may be determined by the second node based on the second intermediate data and the loss function.

[0196] Referring to step (d) in FIG. 9A to FIG. 11B, using an example in which the error information is a gradient, the second node receives second intermediate data (e.g.,

number

number

number

[0197] Phase 2: The second node updates the second data model based on the updated value of the parameter of the second data model.

[0198] For example, the second node is 2 For example, the value of a parameter of the second data model may be updated using

number

[0199] S1202a and S1202b are steps of updating the first channel. Optionally, the first channel may be updated by the first node (S1202a) or the second node (S1202b).

[0200] S1202a: The first node obtains a new first channel based on the error information of the second intermediate data, the information about the first channel, and the first intermediate data.

[0201] Optionally, the first channel includes a second channel and a third channel. S1202a may include: the first node may update the second channel based on the error information of the second intermediate data, the information about the third channel, and the first intermediate data.

[0202] S1202b: The second node obtains a new first channel based on the error information of the second intermediate data, the information about the first channel, and the first intermediate data.

[0203] Optionally, the first channel includes a second channel and a third channel. S1202b may include: the second node may update the second channel based on the error information of the second intermediate data, the information about the third channel, and the first intermediate data.

[0204] Optionally, the first channel may be updated by the first node or the second node.

[0205] For a specific implementation form of the information about the first channel, please refer to the above description of the information about the first channel with reference to Figures 6 and 7. The details will not be described again in this specification.

[0206] In a possible design manner, the distributed learning method provided in this embodiment of the present application may further include: the second node sends error information of the second intermediate data and information about the third channel to the first node. In response, the first node obtains the error information of the second intermediate data and the information about the third channel.

[0207] In some embodiments, the second node transmitting the error information of the second intermediate data and the information about the third channel to the first node may include S1204 shown in FIG. 12(a): the second node transmitting the first information to the first node. In response, the first node obtaining the error information of the second intermediate data and the information about the third channel may include: the first node receiving the first information transmitted by the second node.

[0208] Optionally, the first information may be determined by the second node based on error information of the second intermediate data and information about the third channel.

[0209] For example, referring to step (e) in FIG. 9A to FIG. 11B, the second node calculates, based on the output data y, the loss function, and the second intermediate data c, the error information of the second intermediate data as follows:

number

[0210] For example, referring to step (f) in FIG. 9A and FIG. 9B, the error information of the second intermediate data is

number

number

number

[0211] Optionally, the first node may receive the first information transmitted by the second node over the fourth channel, and in response, the second node may transmit the first information to the first node over the fourth channel.

[0212] For example, the fourth channel may be a channel between the first node and the second node in a reverse training process. For example, the fourth channel may be a control channel.

[0213] That is, the channel between the first node and the second node in the reverse training process may be different from the channel between the first node and the second node in the forward training process, for example, in terms of frequency point or transmission mechanism.

[0214] Optionally, the channel between the first node and the second node in the reverse training process may be the same as the channel between the first node and the second node in the forward training process.

[0215] For example, the first information may be transmitted on a conventional data / control channel, or may be transmitted on a separate physical / logical channel dedicated to DNN training. The transmission process may include operations to reduce the amount of data, such as sparsity, quantization, and / or entropy coding, or may include operations to reduce transmission errors, such as channel coding.

[0216] Optionally, the second node may transmit the error information of the second intermediate data and the information about the third channel to the first node separately. For example, the second node transmits the error information of the second intermediate data to the first node, and the second node transmits the information about the third channel to the first node. The order is not limited. In response, the first node receives the error information of the second intermediate data from the second node, and the first node receives the information about the third channel from the second node.

[0217] In this embodiment of the present application, if the channel between the first node and the second node in the reverse training process is different from the channel between the first node and the second node in the forward training process, the channel between the first node and the second node in the forward training process is referred to as the first channel, and the channel between the first node and the second node in the reverse training process is referred to as the fourth channel. If the channel between the first node and the second node in the reverse training process is the same as the channel between the first node and the second node in the forward training process, the channel between the first node and the second node in both the forward process and the reverse process is referred to as the first channel.

[0218] In some other embodiments, the second node transmitting the error information of the second intermediate data and information about the third channel to the first node may include S1205 shown in FIG. 12(b): the second node transmitting a fourth signal to the first node through the first channel.

[0219] In response, at S1205, the first node receives a third signal from the second node over the first channel.

[0220] Optionally, the fourth signal includes a signal generated by mapping error information of the second intermediate data to an air interface resource.

[0221] For example, the second node may receive error information of the second intermediate data.

number

[0222] For example, the second node may repeatedly transmit the fourth signal and adjust the parameter φ of the controllable portion each time it transmits the fourth signal.

[0223] For example, referring to step (x) in FIG. 10A and FIG. 10B ,

number

number

number

number

[0224] It should be noted that the order of adjusting the parameter φ of the controllable part is not limited in this embodiment of the present application, and the above is just an example devised in the present application for ease of explanation, on the premise that the first node can obtain error information of the first intermediate data based on the third signal.

[0225] Accordingly, in some other embodiments, the first node obtaining error information of the second intermediate data and information about the third channel may include: the first node receiving a third signal from the second node through the first channel, and the first node obtaining the first information based on the third signal.

[0226] Optionally, the third signal is a signal obtained after the fourth signal is transmitted over the first channel to the first node.

[0227] For example, referring to step (x) in FIG. 10A and FIG. 10B, the third signal received by the first node may be:

number

number

[0228] Optionally, the first information may be determined by the first node based on error information of the second intermediate data and information about the third channel.

[0229] For example, referring to step (x) in FIG. 10A and FIG. 10B, the first node:

number

number

number

[0230] In this way, by using the propagation characteristics of signals on the air interface resources based on the channel reciprocity, the first information (i.e., the error information of the second intermediate data and the information about the third channel) is obtained through over-the-air calculation to avoid channel estimation, thereby reducing pilot overhead.

[0231] In a possible design manner, the distributed learning method provided in this embodiment of the present application may further include: S1206, shown in Fig. 12(c): the first node transmits a fifth signal to the second node through the first channel, and in response, the second node receives a sixth signal from the first node through the first channel.

[0232] Optionally, the fifth signal includes a signal generated by mapping third intermediate data to an air interface resource, the third intermediate data being for updating the first channel.

[0233] For example, the third intermediate data may be a result of processing the first data by the first node. The first node may

number

[0234] For example, the first node may repeatedly transmit the fifth signal and adjust the parameter φ of the controllable portion each time it transmits the fifth signal.

[0235] For example, referring to step (y) in FIG. 11A and FIG. 11B ,

number

number

number

number

[0236] It should be noted that the order of adjusting the parameters φ of the controllable parts is not limited in this embodiment of the present application, and the above is just one example proposed in the present application for ease of explanation.

[0237] Optionally, the sixth signal is a signal obtained after the fifth signal is transmitted over the first channel to the second node.

[0238] Optionally, the first node may obtain the third information based on the sixth signal.

[0239] For example, the third information may be determined based on the third intermediate data and information about the third channel.

[0240] For example, referring to step (y) in FIG. 11A and FIG. 11B, the sixth signal received by the second node may be:

number

number

number

number

number

[0241] In this way, by using the propagation characteristics of signals on the air interface resources based on channel reciprocity, information about the third channel and the third intermediate data are obtained through over-the-air calculation to avoid channel estimation, thereby reducing pilot overhead.

[0242] In some embodiments, the first node or the second node may obtain the new first channel based on the first information and the first intermediate data.

[0243] In the following, updating the first channel (obtaining a new first channel) will be specifically described with reference to Examples 1 to 4.

[0244] Example 1: A method for updating the first channel may include the following steps 1.1 to 1.3.

[0245] Step 1.1: A first node determines parameter error information of a first channel based on first information and first intermediate data.

[0246] Optionally, the error information of the first channel parameter may include a gradient of the first channel parameter or a normalized value of the gradient of the first channel parameter.

[0247] For example, the parameter of the first channel may be the parameter φ of the controllable part of the first channel. For a specific implementation form of the parameter φ of the controllable part, please refer to the above description of the parameter of the first channel. The details will not be described again in this specification.

[0248] Optionally, the first node may determine error information of the parameter of the second channel based on the first information and the first intermediate data.

[0249] Referring to step (h) in FIG. 9A and FIG. 9B or FIG. 10A and FIG. 10B, the first node

number

number

number

[0250] Step 1.2: The first node determines an updated value of the parameter of the second channel based on the error information of the parameter of the second channel.

[0251] Referring to step (i) in FIG. 9A and FIG. 9B or FIG. 10A and FIG. 10B, the first node calculates a gradient of the parameter φ of the second channel.

number

number

number

number

number

number

[0252] Step 1.3: The first node updates the second channel based on the updated value of the parameter of the second channel.

[0253] Referring to step (j) in FIGS. 9A and 9B or 10A and 10B, the first node:

number

[0254] Optionally, the first node may be configured to adjust a parameter of the second channel.

[0255] For example, the first node has a function for adjusting a parameter of the second channel. For example, the first node includes a controller. The controller may be configured to adjust a parameter of the controllable part.

[0256] In some embodiments, after performing step 1.1 above, the first node may perform step 2.1 below to enable the second node to update the first channel.

[0257] Example 2: A scheme for updating the first channel may include the above-mentioned step 1.1 and the following steps 2.1 to 2.3: Step 1.1 may be performed first, and then steps 2.1 to 2.3 are performed.

[0258] Step 2.1: A first node transmits parameter error information of a second channel to a second node, and in response, the second node receives parameter error information of the second channel from the first node.

[0259] For a specific implementation of the parameter error information of the second channel, please refer to the above step 1.1. The details will not be described again in this specification.

[0260] Referring to step (k) in FIG. 9A and FIG. 9B , the first node:

number

[0261] Optionally, the first node transmits parameter error information of the second channel to the second node through the fourth channel, and in response, the second node receives parameter error information of the second channel from the first node through the fourth channel.

[0262] Step 2.2: The second node determines an updated value of the parameter of the second channel based on the error information of the parameter of the second channel.

[0263] It should be noted that the implementation of stage 2.2 is similar to stage 1.2 described above, with the main difference being that the first node is replaced by a second node, and the details will not be described again here.

[0264] Step 2.3: The second node updates the second channel based on the updated value of the parameter of the second channel.

[0265] Referring to step (l) in FIG. 9A and FIG. 9B , the second node:

number

[0266] For example, the second node has the capability to adjust a parameter of the second channel. For example, the second node includes a controller. The controller may be configured to adjust a parameter of the controllable portion.

[0267] In this way, when the first node does not have the capability to adjust the parameters of the second channel and the second node has the capability to adjust the parameters of the second channel, the first node may transmit error information of the parameters of the second channel to the second node to enable the second node to update the second channel.

[0268] In some embodiments, after performing step 1.2 above, the first node may perform step 3.1 below without performing step 1.3 to enable the second node to update the second channel.

[0269] Example 3: A scheme for updating the second channel may include steps 1.1 and 1.2 above, and steps 3.1 and 3.2 below.

[0270] Step 3.1: A first node transmits an updated value of a parameter of a second channel to a second node, and in response, the second node receives an updated value of the parameter of the second channel from the first node.

[0271] Referring to step (k) in FIG. 9A and FIG. 9B, the first node transmits an updated value of the parameter of the second channel.

number

[0272] Step 3.2: The second node updates the second channel based on the updated value of the parameter of the second channel.

[0273] Please note that for the implementation of step 3.2, reference may be made to step 2.3, and the details will not be described again in this specification.

[0274] For example, the second node has the capability to adjust a parameter of the second channel. For example, the second node includes a controller. The controller may be configured to adjust a parameter of the controllable portion.

[0275] In this manner, if the first node does not have the capability to adjust the parameter of the second channel and the second node has the capability to adjust the parameter of the second channel, the first node may transmit an updated value of the parameter of the second channel to the second node to enable the second node to update the second channel.

[0276] Example 4: A method for updating the first channel may include the following steps 4.1 to 4.3.

[0277] Step 4.1: The second node determines error information of a parameter of the second channel based on the error information of the second intermediate data, the information about the third channel, and the first intermediate data.

[0278] For example, the second node may update the second channel based on the third information and the error information of the second intermediate data.

[0279] Optionally, the error information of the second channel parameter may include a gradient of the second channel parameter or a normalized value of the gradient of the second channel parameter.

[0280] For example, the parameter of the second channel may be the parameter φ of the controllable part of the second channel. For the specific implementation form of the parameter φ of the controllable part, please refer to the above description of the parameter of the first channel. The details will not be described again in this specification.

[0281] Optionally, the second node may determine error information of a parameter of the second channel based on the error information of the second intermediate data, the information about the third channel, and the first intermediate data.

[0282] For example, the second node may determine error information of a parameter of the second channel based on the third information and the error information of the second intermediate data.

[0283] Referring to step (u) in FIG. 11A and FIG. 11B, the second node receives the third information

number

number

number

[0284] Step 4.2: The second node determines an updated value of the parameter of the second channel based on the error information of the parameter of the second channel.

[0285] Referring to step (v) in FIG. 11A and FIG. 11B, the second node calculates a gradient of the parameter φ of the second channel.

number

number

number

number

number

[0286] Step 4.3: The second node updates the second channel based on the updated value of the parameter of the second channel.

[0287] Referring to step (w) in FIG. 11A and FIG. 11B, the second node:

number

[0288] For example, the second node has the capability to adjust a parameter of the second channel. For example, the second node includes a controller. The controller may be configured to adjust a parameter of the controllable portion.

[0289] It should be noted that the manners of updating the first channel recorded in Examples 1 to 4 may further be applied when both the first node and the second node function to adjust the parameters of the channel.

[0290] In a possible design manner, the distributed learning method provided in this embodiment of the present application may include: a first node transmitting a first signal to a second node through a first channel, in response to which the second node receives a second signal from the first node through the first channel.

[0291] For example, the second signal is a signal obtained after the first signal is transmitted over the first channel to the second node.

[0292] Optionally, the first signal is for determining information about a third channel.

[0293] For example, the first signal may be a pilot signal, and the first node may transmit the pilot and set a parameter φ of the controllable part.

[0294] For example, referring to step (m) in FIG. 9A and FIG. 9B, a first node transmits a pilot to a second node, and the first node transmits a pilot signal φ 1 From φ N Set each to [0,0,...,0], and φ 1 From φ N respectively.

number

number

[0295] Optionally, the second node may obtain information about the first channel based on the second signal.

[0296] In this way, to improve the performance of distributed learning, the second node may perform channel estimation to obtain information about the channel and update the channel or first data model.

[0297] For example, referring to step (n) in FIG. 9A and FIG. 9B, the first node 1 From φ N It is assumed that the second node transmits a pilot signal to the second node, with the channel equation

number

[0298] The first node is φ 1 From φ N respectively.

number

number

number

[0299] The acquired information h for the first channel (e.g., the third channel) and the acquired information h+g for the first channel (e.g., the third channel) 2,n ×g 1,n Based on this, the second node derives information g 2,n ×g 1,n may be obtained.

[0300] For example, in step (j) in FIG. 9A and FIG. 9B and FIG. 7, the first node 1 From φ N The second node sets,k,to the value obtained after the previous training,k,, and the second node receives,k,information about the first channel.

number

[0301] It should be noted that the value of N may be equal to the actual number of controllable units, e.g., the number of distributed antennas or the number of elements of an intelligent reflecting surface. Alternatively, the value of N may be less than the actual number of controllable units. For example, in the case of a very large antenna array or an intelligent reflector, multiple controllable units may be controlled as a group to reduce overhead.

[0302] It should be noted that the order of steps (m) to (n) and the second node transmitting the first information to the first node is not limited in this embodiment of the present application. For example, steps (m) to (n) may be performed before the second node transmits the first information to the first node. That is, after obtaining information about the first channel (e.g., the third channel) by estimation, the second node may transmit the first information to the first node to update the first channel.

[0303] S1203: The first node updates the first data model based on the error information of the first intermediate data to obtain a new first data model.

[0304] In some embodiments, S1203 may include the following steps 3 to 5.

[0305] Step 3: The first node obtains error information of a parameter of the first data model based on the error information of the first intermediate data.

[0306] Referring to step (o) in FIG. 9A to FIG. 11B, the first node

number

number

[0307] Step 4: The first node obtains an updated value of the parameter of the first data model based on the error information of the parameter of the first data model.

[0308] Referring to step (o) in FIG. 9A to FIG. 11B, the first node receives error information of the parameters of the first data model.

number

number

[0309] Step 5: The first node updates the first data model based on the updated value of the parameter of the first data model.

[0310] Referring to step (o) in FIG. 9A to FIG. 11B, the first node is 1 For example, the value of a parameter of the first data model may be updated using

number

[0311] In a possible design manner, the distributed learning method provided in this embodiment of the present application may further include: S1207, shown in Fig. 12(a): the second node sends second information to the first node. In response, the first node receives the second information sent by the second node.

[0312] Optionally, the second information may be for obtaining error information of the first intermediate data.

[0313] In some embodiments, the second information may include error information of the first intermediate data. For example, the second information may include error information of the first intermediate data.

number

[0314] For example, referring to step (p) in FIG. 9A and FIG. 9B, the second node

number

[0315] Optionally, the second node may transmit the second information to the first node through the fourth channel. In response, the first node receives the second information transmitted by the second node through the fourth channel. In other words, the second node may transmit the second information to the first node through the control channel.

[0316] In some embodiments, before the second node transmits the second information to the first node, the distributed learning method provided in this embodiment of the present application may further include: the second node obtaining error information of the first intermediate data based on the error information of the second intermediate data and information about the first channel.

[0317] For example, referring to step (q) in FIG. 9A and FIG. 9B, the second node

number

number

number

number

number

[0318] That is, after obtaining the error information of the first intermediate data, the second node may send second information including the error information of the first intermediate data to the first node, so that the first node may update the first data model, thereby improving the performance of distributed learning.

[0319] In some other embodiments, the second information may include error information of the second intermediate data and information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data.

[0320] That is, the second node may transmit error information of the second intermediate data and information about the first channel to the first node, and in response, the first node receives the error information of the second intermediate data and information about the first channel transmitted by the second node.

[0321] In a possible design manner, the distributed learning method provided in this embodiment of the present application may further include: S1207 shown in Fig. 12(b) or Fig. 12(c): the second node sends a seventh signal to the first node. In response, the first node receives an eighth signal from the second node, and obtains error information of the first intermediate data according to the eighth signal.

[0322] Optionally, the eighth signal is a signal obtained after the seventh signal is transmitted to the first node through the first channel, and the seventh signal includes a signal generated by mapping error information of the second intermediate data to air interface resources.

[0323] For example, the second node may send error information of the second intermediate data and set parameters of the controllable part.

[0324] For example, the second node may generate error information of the second intermediate data based on the dimension matching.

number

[0325] For example, referring to step (p) in FIG. 10A to FIG. 11B, the second node

number

number

number

number

number

[0326] In this way, over-the-air calculation of the gradients of the hidden layers of the neural network of the first node is implemented through the process of over-the-air signal propagation, which can reduce pilot overhead.

[0327] Optionally, the distributed learning method provided in this embodiment of the present application may further include performing processing such as normalization, amplitude limiting, sparsity, and / or dynamic control on the parameter error information of the first data model, the parameter error information of the channel, and the parameter error information of the second data model to reduce the peak-to-average ratio of the transmission waveform, improve the signal-to-noise ratio, etc.

[0328] It should be noted that the order of S1201 to S1208 is not limited in this embodiment of the present application, provided that reverse training can be implemented.

[0329] By using the above-mentioned distributed learning method, the channel between the first node and the second node is trained, and the channel is used as an intermediate layer (e.g., residual layer) of the data model, and the first data model, the channel, and the second data model can participate in the trained inference, thereby improving the performance of distributed learning. In addition, wireless transmission can be used to directly function for distributed learning, thereby reducing the processing complexity and further saving resources.

[0330] The distributed learning method provided in the embodiment of the present application is described in detail above with reference to Figures 8 to 12(c). The distributed learning device provided in the embodiment of the present application is described in detail below with reference to Figures 13 to 14.

[0331] 13 is a schematic diagram of the structure of a distributed learning device configured to perform a distributed learning method according to an embodiment of the present application. The distributed learning device 1300 may be a first node or a second node, or may be a chip or another component with corresponding functions used in the first node or the second node. As shown in FIG. 13, the distributed learning device 1300 may include a processor 1301 and a transceiver 1303, and may further include a memory 1302. The processor 1301 is coupled to the memory 1302 and the transceiver 1303. For example, the processor 1301 may be connected to the memory 1302 and the transceiver 1303 by a communication bus. Alternatively, the processor 1301 may be used separately.

[0332] The components of distributed learning device 1300 are described in detail below with reference to FIG.

[0333] Processor 1301 is the control center of distributed learning device 1300 and may be a single processor or may be a general term for multiple processing elements.

[0334] The processor 1301 may execute software programs stored in the memory 1302 and access data stored in the memory 1302 to perform various functions of the distributed learning device 1300 .

[0335] In a specific implementation, in one embodiment, the processor 1301 may include one or more CPUs, for example, CPU0 and CPU1 shown in FIG.

[0336] In a specific implementation, in one embodiment, the distributed learning device 1300 may include multiple processors, such as the processor 1301 and the processor 1304 shown in FIG. 13. Each of the processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor in this specification may be one or more communication devices, circuits, and / or processing cores configured to process data (e.g., computer program instructions).

[0337] The memory 1302 may be integrated with the processor 1301 or may exist separately, and is coupled to the processor 1301 through an input / output port (not shown in FIG. 13) of the distributed learning device 1300. In the present embodiment of the present application, there is no particular limitation thereon.

[0338] For example, the input port may be configured to implement a receiving function performed by the first node or the second node in any one of the aforementioned method embodiments, and the output port may be configured to implement a transmitting function performed by the first node or the second node in any one of the aforementioned method embodiments.

[0339] The memory 1302 is configured to store a software program for implementing the solution of the present application, and the processor 1301 controls the execution. For specific implementation, please refer to the above-mentioned method embodiment. The details will not be described again in this specification.

[0340] The transceiver 1303 is configured to communicate with another device. For example, when the distributed learning device 1300 is a first node, the transceiver 1303 may be configured to communicate with a second node. As another example, when the distributed learning device 1300 is a second node, the transceiver 1303 may be configured to communicate with the first node. In addition, the transceiver 1303 may include a receiver and a transmitter (not shown separately in FIG. 13). The receiver is configured to implement a receiving function, and the transmitter is configured to implement a transmitting function. The transceiver 1303 may be integrated with the processor 1301 or may exist separately, and is coupled to the processor 1301 through an input / output port (not shown in FIG. 13) of the distributed learning device 1300. In this embodiment of the present application, there is no particular limitation thereon.

[0341] It should be noted that the structure of distributed learning device 1300 shown in Figure 13 does not constitute a limitation on the distributed learning device. An actual distributed learning device may include more or fewer components than shown in the figure, or some components may be combined, or a different arrangement of components may be used.

[0342] The operations of the first node in the above-mentioned steps S801, S802, S1202a, S1203, and S1206 may be performed by a processor 1301 in the distributed learning device 1300 shown in FIG. 13 by invoking application program code stored in memory 1302 that indicates to the first node to perform the operations.

[0343] The operations of the second node in the above steps S803, S1201, S1202b, S1204, S1205, S1207, and S1208 may be performed by the processor 1301 in the distributed learning device 1300 shown in Fig. 13 by calling application program code stored in the memory 1302 that instructs the second node to perform the operations. This embodiment is not limited thereto.

[0344] 14 is a schematic diagram of the structure of another distributed learning device according to an embodiment of the present application. For ease of explanation, FIG. 14 shows only the main components of the distributed learning device.

[0345] The distributed learning device 1400 includes a transceiver module 1401 and a processing module 1402. The distributed learning device 1400 may be a first node or a second node in the aforementioned method embodiments. The transceiver module 1401 may be referred to as a transceiver unit, and is configured to implement the transceiver function performed by the first node or the second node in any one of the aforementioned method embodiments.

[0346] It should be noted that the transceiver module 1401 may include a receiving module and a transmitting module (not shown in FIG. 14). The receiving module is configured to receive data and / or signaling from the first node. The transmitting module is configured to transmit data and / or signaling to the first node. The specific implementation of the transceiver module is not specifically limited in this application. The transceiver module may include a transceiver circuit, a transceiver machine, a transceiver, or a communication interface.

[0347] The processing module 1402 may be configured to implement the processing functions performed by the first node or the second node in any one of the embodiments of the method described above. The processing module 1402 may be a processor.

[0348] In this embodiment, distributed learning device 1400 is presented in the form of functional modules divided in an integrated manner. A "module" in this specification may be a specific ASIC, circuitry, processor executing one or more software or firmware programs, memory, integrated logic circuitry, and / or other components capable of providing the above-mentioned functionality. In one embodiment, one skilled in the art may recognize that distributed learning device 1400 is in the form of distributed learning device 1300 shown in FIG. 13.

[0349] For example, a processor 1301 in a distributed learning device 1300 shown in FIG. 13 may invoke computer-executable instructions stored in a memory 1302 to perform the distributed learning method in the method embodiment described above.

[0350] Specifically, the functions / implementation processes of the transceiver module 1401 and the processing module 1402 in FIG. 14 are implemented by the processor 1301 in the distributed learning device 1300 shown in FIG. 13 by calling computer-executable instructions stored in the memory 1302. Alternatively, the functions / implementation processes of the processing module 1402 in FIG. 14 may be implemented by the processor 1301 in the distributed learning device 1300 shown in FIG. 13 by calling computer-executable instructions stored in the memory 1302, The transmitting and receiving module 1401 The function / implementation process may be implemented by the transceiver 1303 in the distributed learning device 1300 shown in FIG.

[0351] The distributed learning device 1400 provided in this embodiment can execute the aforementioned distributed learning method, so please refer to the aforementioned method embodiments for the technical effects that can be obtained by the distributed learning device 1400. The details will not be described again in this specification.

[0352] In a possible design solution, the distributed learning device 1400 shown in FIG. 14 may be used in the communication system shown in FIG. 1 to perform the functions of the first node in the distributed learning method shown in FIG. 8 and / or FIGS. 12(a), 12(b), and 12(c).

[0353] The processing module 1402 is configured to process the first data by using the first data model to obtain first intermediate data.

[0354] The transceiver module 1401 is configured to transmit first intermediate data to the second node through a first channel, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the second intermediate data being a result of the transmission of the first intermediate data to the second node through the first channel, and the first channel being a channel between the distributed learning device and the second node.

[0355] Optionally, the distributed learning device 1400 may further include a storage module (not shown in FIG. 14), which stores a program or instruction. When the processing module 1402 executes the program or instruction, the distributed learning device 1400 can perform the function of the first node in the distributed learning method shown in FIG.

[0356] It should be noted that the distributed learning device 1400 may be a first node, or may be, for example, a chip (system) or another component or assembly that may be located within the first node, and this is not a limitation in this application.

[0357] In addition, for the technical effects of the distributed learning device 1400, please refer to the technical effects of the distributed learning method shown in Figure 8, Figure 12 (a), Figure 12 (b), and Figure 12 (c), and the details will not be described again in this specification.

[0358] In another possible design solution, the distributed learning device 1400 shown in FIG. 14 may be used in the communication system shown in FIG. 1 to perform the functions of a second node in the distributed learning method shown in FIG. 8 and / or FIGS. 12(a), 12(b), and 12(c).

[0359] The transceiver module 1401 is configured to receive second intermediate data through a first channel, the second intermediate data being a result of transmission of the first intermediate data sent by the first node to the distributed learning device through the first channel, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, and the first channel being a channel between the first node and the distributed learning device.

[0360] The processing module 1402 is configured to process the second intermediate data by using the second data model to obtain output data.

[0361] Optionally, the distributed learning device 1400 may further include a storage module (not shown in FIG. 14), which stores programs or instructions. When the processing module 1402 executes the programs or instructions, the distributed learning device 1400 can perform the functions of the second node in the distributed learning methods shown in FIG. 8, FIG. 12(a), FIG. 12(b), and FIG. 12(c).

[0362] It should be noted that the distributed learning device 1400 may be a second node or may be, for example, a chip (system) or another component or assembly that may be located in the second node, and this is not a limitation in this application.

[0363] In addition, for the technical effects of the distributed learning device 1400, please refer to the technical effects of the distributed learning method shown in Figure 8, Figure 12 (a), Figure 12 (b), and Figure 12 (c), and the details will not be described again in this specification.

[0364] An embodiment of the present application further provides a communication system, the communication system including a first node and a second node.

[0365] The first node is configured to perform the operations of the first node in the above-mentioned method embodiments. For specific implementation methods and processes, please refer to the above-mentioned method embodiments. The details will not be described again in this specification.

[0366] The second node is configured to perform the operations of the second node in the above-mentioned method embodiments. For specific implementation methods and processes, please refer to the above-mentioned method embodiments. The details will not be described again in this specification.

[0367] An embodiment of the present application provides a chip system, which includes a processor and an input / output port, the processor may be configured to implement a processing function included in the distributed learning method provided in the embodiment of the present application, and the input / output port may be configured to perform a transmission / reception function included in the distributed learning method provided in the embodiment of the present application.

[0368] For example, the input port may be configured to implement a receiving function included in the distributed learning method provided in an embodiment of the present application, and the output port may be configured to implement a transmitting function included in the distributed learning method provided in an embodiment of the present application.

[0369] In a possible design, the chip system may further include a memory configured to store program instructions and data for implementing functions involved in the distributed learning method provided in the embodiments of the present application.

[0370] A chip system may include a chip, or may include a chip and other discrete components.

[0371] An embodiment of the present application provides a computer-readable storage medium, which includes a computer program or instructions, which, when executed on a computer, executes the distributed learning method provided in the embodiment of the present application.

[0372] An embodiment of the present application provides a computer program product, which includes a computer program or instructions, which, when executed on a computer, performs the distributed learning method provided in the embodiment of the present application.

[0373] It should be understood that the processor in the embodiments of the present application may be a central processing unit (CPU). Alternatively, the processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or another programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or the like. The general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.

[0374] It may be further understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include a volatile memory and a non-volatile memory. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM) and is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) may be used, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchlink dynamic random access memory (synchlink DRAM, SLDRAM), and direct rambus random access memory (DR RAM).

[0375] The above-described embodiments may be implemented completely or partially by software, hardware (e.g., circuits), firmware, or any combination thereof. When software is used to implement the embodiments, all or some of the above-described embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the procedures or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored on a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired (e.g., infrared, wireless, or microwave) manner. The computer-readable storage medium may be any available medium accessible by a computer, or a data storage device such as a server or data center that integrates one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, hard disk, or magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium, etc. The semiconductor medium may be a solid state drive.

[0376] It should be understood that the term "and / or" in this specification describes only the correspondence between related objects and indicates that three relationships may exist. For example, A and / or B may represent the following three cases: only A is present, both A and B are present, and only B is present. A and B may be singular or plural. In addition, the character " / " in this specification usually indicates an "or" relationship between related objects, but may also indicate an "and / or" relationship. Please refer to the context for further understanding.

[0377] As used herein, at least one means one or more, and more means two or more. At least one of the following items, or similar expressions, refers to any combination of these items, including any combination of singular items or multiple items. For example, at least one of a, b, or c may refer to a, b, c, a and b, a and c, b and c, or a, b, and c, where a, b, and c may be singular or plural.

[0378] It should be understood that the sequence numbers of the above processes do not mean the execution sequence in various embodiments of the present application. The execution sequence of the processes should be determined based on the functions and internal logic of the processes, and should not be construed as any limitation on the implementation process of the embodiments of the present application.

[0379] Those skilled in the art may recognize that the units and algorithm steps may be implemented by electronic hardware or a combination of computer software and electronic hardware in combination with the examples described in the embodiments disclosed herein. Whether these functions are implemented by hardware or software is determined by the specific application and design constraints of the technical solution. Those skilled in the art may use different methods for each specific application to implement the described functions, but such implementation should not be considered as going beyond the scope of this application.

[0380] For the sake of convenience and concise description, the detailed operation processes of the above-mentioned systems, devices, and units shall refer to the corresponding processes in the above-mentioned method embodiments, and the details will not be described again here, as will be clearly understood by those skilled in the art.

[0381] In some embodiments provided in the present application, it should be understood that the disclosed system, device, and method may be implemented in other ways. For example, the embodiment of the device described is merely an example. For example, the division into multiple units is merely a logical division of functions, and may be divided in other ways in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not implemented. Furthermore, the shown or described interconnections or direct connections or communication connections may be implemented through some interfaces. Indirect connections or communication connections between devices or units may be implemented in electronic, mechanical, or other forms.

[0382] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one location or distributed across multiple network units. Some or all of these units may be selected based on actual requirements to achieve the objectives of the solutions of these embodiments.

[0383] In addition, the functional units in the embodiments of the present application may be integrated into one processing unit, or each of those units may exist physically alone, or two or more units may be integrated into one unit.

[0384] When these functions are implemented in the form of software functional units and sold or used as independent products, these functions may be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application may be essentially implemented in the form of a software product, or the part that contributes to the prior art, or some of these technical solutions may be implemented in the form of a software product. The computer software product is stored in a storage medium and includes some instructions for instructing a computer device (which may be a computer, a server, or a second node, etc.) to execute all or some of the steps in the method described in the embodiments of the present application. The aforementioned storage medium includes any medium that can store program code, such as a USB flash drive, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

[0385] The above description is merely a specific implementation of the present application, and is not intended to limit the scope of protection of the present application. Any variations or replacements that a person skilled in the art can easily think of within the technical scope disclosed in the present application shall be included in the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the scope of protection of the claims. 。 [Item 1] A method of distributed learning applied to a first node, the first node having a first data model, the method comprising: processing the first data by using the first data model to obtain first intermediate data; and transmitting the first intermediate data to a second node through a first channel, where the first channel is updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the second intermediate data being a result of transmission of the first intermediate data to the second node through the first channel, the first channel being a channel between the first node and the second node. The distributed learning method includes: [Item 2] the first channel having a second channel and a third channel, the method comprising: updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data. 2. The distributed learning method of item 1, further comprising: [Item 3] The step of updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data comprises: receiving first information transmitted by the second node, where the first information is determined based on the error information of the second intermediate data and the information about the third channel; and updating the second channel based on the first information and the first intermediate data. 3. The distributed learning method according to item 2, comprising: [Item 4] The step of receiving first information transmitted by the second node comprises: receiving the first information transmitted by the second node over a fourth channel. Item 3. The distributed learning method according to item 3. [Item 5] The step of updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data comprises: receiving a third signal from the second node through the first channel, where the third signal is a signal obtained after a fourth signal is transmitted to the first node through the first channel, and the fourth signal includes a signal generated by mapping the error information of the second intermediate data to an air interface resource; obtaining first information based on the third signal, where the first information is determined based on the error information of the second intermediate data and the information about the third channel; and updating the second channel based on the first information and the first intermediate data. 3. The distributed learning method according to item 2, comprising: [Item 6] The method comprising: transmitting a first signal over the first channel to the second node, the first signal being for determining the information about the first channel. 6. The distributed learning method according to any one of claims 1 to 5, further comprising: [Item 7] The method comprising: transmitting a fifth signal over the first channel to the second node, where the fifth signal comprises a signal generated by mapping third intermediate data to air interface resources, the third intermediate data being for updating the first channel. 7. The distributed learning method according to any one of claims 1 to 6, further comprising: [Item 8] The method comprising: updating the first data model based on error information of the first intermediate data to obtain a new first data model; 8. The distributed learning method according to any one of claims 1 to 7, further comprising: [Item 9] The method comprising: receiving second information transmitted by the second node, where the second information is for obtaining the error information of the first intermediate data; 9. The distributed learning method according to item 8, further comprising: [Item 10] the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data. 10. The distributed learning method according to item 9. [Item 11] The method comprising: receiving an eighth signal from the second node, where the eighth signal is obtained after a seventh signal is transmitted through a channel to the first node, the seventh signal having a signal generated by mapping the error information of the second intermediate data to an air interface resource; and obtaining the error information of the first intermediate data based on the eighth signal; 11. The distributed learning method according to any one of claims 1 to 10, further comprising: [Item 12] A method of distributed learning applied to a second node, the second node having a second data model, the method comprising: receiving second intermediate data over a first channel, where the second intermediate data is a result of a transmission of first intermediate data sent by a first node over the first channel to the second node, and the first channel is updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the first channel being a channel between the first node and the second node; and processing the second intermediate data by using the second data model to obtain output data; The distributed learning method includes: [Item 13] the first channel having a second channel and a third channel, the method comprising: transmitting the error information of the second intermediate data and information about the third channel to the first node. Item 13. The method of item 12, further comprising: [Item 14] The step of transmitting the error information of the second intermediate data and information about the third channel to the first node comprises: transmitting first information to the first node, where the first information is determined based on the error information of the second intermediate data and the information about the third channel. Item 14. The distributed learning method according to item 13, comprising: [Item 15] The step of transmitting the first information to the first node comprises: transmitting the first information to the first node over a fourth channel. Item 15. The distributed learning method according to item 14, comprising: [Item 16] The step of transmitting the error information of the second intermediate data and information about the third channel to the first node comprises: transmitting a fourth signal over the first channel to the first node, where the fourth signal includes a signal generated by mapping the error information of the second intermediate data to air interface resources. Item 14. The distributed learning method according to item 13, comprising: [Item 17] The method comprising: receiving a second signal from the first node over the first channel, the second signal being a signal obtained after a first signal is transmitted over the first channel to the second node, the first signal being for determining the information about the first channel; and obtaining said information about said first channel based on said second signal; 17. The distributed learning method according to any one of claims 12 to 16, further comprising: [Item 18] The method comprising: receiving a sixth signal from the first node over the first channel, where the sixth signal is obtained after a fifth signal is transmitted over the first channel to the second node, the fifth signal having a signal generated by mapping third intermediate data to air interface resources, the third intermediate data being for updating the first channel; obtaining third information based on the sixth signal, where the third information is determined based on the third intermediate data and the information about the third channel; and updating the second channel based on the third information and the error information of the second intermediate data. 18. The distributed learning method of any one of claims 12 to 17, further comprising: [Item 19] The method comprising: Sending second information to the first node, where the second information is for obtaining error information of the first intermediate data. 19. The distributed learning method of any one of claims 12 to 18, further comprising: [Item 20] the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data. 20. The distributed learning method according to item 19. [Item 21] The method comprising: updating the second data model based on the output data to obtain a new second data model. 21. The distributed learning method of any one of claims 12 to 20, further comprising: [Item 22] The method comprising: transmitting a seventh signal to the first node, where the seventh signal comprises a signal generated by mapping the error information of the second intermediate data to an air interface resource. 22. The distributed learning method of any one of claims 12 to 21, further comprising: [Item 23] A distributed learning device, the distributed learning device comprising a first data model, the device comprising a processing module and a transceiver module; the processing module is configured to process the first data by using the first data model to obtain first intermediate data; The transceiver module is configured to transmit the first intermediate data to a second node through a first channel, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the second intermediate data being a result of transmitting the first intermediate data to the second node through the first channel, and the first channel being a channel between the distributed learning device and the second node. Distributed learning device. [Item 24] the first channel having a second channel and a third channel; The processing module is further configured to update the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data. Item 24. The distributed learning device according to item 23. [Item 25] the transceiver module is further configured to receive first information transmitted by the second node, the first information being determined based on the error information of the second intermediate data and the information about the third channel; The processing module is further configured to update the second channel based on the first information and the first intermediate data. Item 25. The distributed learning device according to item 24. [Item 26] The transceiver module is further configured to receive the first information transmitted by the second node over a fourth channel. Item 26. The distributed learning device according to item 25. [Item 27] the transceiver module is further configured to receive a third signal from the second node through the first channel, the third signal being a signal obtained after a fourth signal is transmitted to the distributed learning device through the first channel, the fourth signal including a signal generated by mapping the error information of the second intermediate data to an air interface resource; the processing module is further configured to obtain first information based on the third signal, the first information being determined based on the error information of the second intermediate data and the information about the third channel; The processing module is further configured to update the second channel based on the first information and the first intermediate data. Item 25. The distributed learning device according to item 24. [Item 28] The transceiver module is further configured to transmit a first signal over the first channel to the second node, the first signal being for determining the information about the first channel. 28. A distributed learning device according to any one of items 23 to 27. [Item 29] The transceiver module is further configured to transmit a fifth signal to the second node over the first channel, the fifth signal including a signal generated by mapping third intermediate data to air interface resources, the third intermediate data being for updating the first channel. 29. A distributed learning device according to any one of items 23 to 28. [Item 30] The processing module is further configured to update the first data model based on error information of the first intermediate data to obtain a new first data model. 30. A distributed learning device according to any one of items 23 to 29. [Item 31] The transceiver module is further configured to receive second information sent by the second node, the second information being for obtaining the error information of the first intermediate data. Item 31. The distributed learning device according to item 30. [Item 32] the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data. Item 32. The distributed learning device according to item 31. [Item 33] the transceiver module is further configured to receive an eighth signal from the second node, the eighth signal being a signal obtained after a seventh signal is transmitted through a channel to the distributed learning device, the seventh signal including a signal generated by mapping the error information of the second intermediate data to an air interface resource; The processing module is further configured to obtain the error information of the first intermediate data based on the eighth signal. 33. A distributed learning device according to any one of items 23 to 32. [Item 34] A distributed learning device, the distributed learning device comprising a second data model, the device comprising a processing module and a transceiver module; the transceiver module is configured to receive second intermediate data through a first channel, the second intermediate data being a result of transmission of the first intermediate data sent by a first node through the first channel to the distributed learning device, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the first channel being a channel between the first node and the distributed learning device; The processing module is configured to process the second intermediate data by using the second data model to obtain output data. Distributed learning device. [Item 35] Item 35. The apparatus of item 34, wherein the first channel has a second channel and a third channel, and the transceiver module is further configured to transmit the error information of the second intermediate data and information about the third channel to the first node. [Item 36] The transceiver module is further configured to transmit first information to the first node, the first information being determined based on the error information of the second intermediate data and the information about the third channel. Item 36. The distributed learning device according to item 35. [Item 37] The transceiver module is further configured to transmit the first information to the first node over a fourth channel. Item 37. The distributed learning device according to item 36. [Item 38] The transceiver module is further configured to transmit a fourth signal to the first node through the first channel, the fourth signal including a signal generated by mapping the error information of the second intermediate data to an air interface resource. Item 36. The distributed learning device according to item 35. [Item 39] the transceiver module is further configured to receive a second signal from the first node through the first channel, the second signal being a signal obtained after a first signal is transmitted through the first channel to the distributed learning device, the first signal being for determining the information about the first channel; The processing module is further configured to obtain the information about the first channel based on the second signal. 39. A distributed learning device according to any one of items 34 to 38. [Item 40] the transceiver module is further configured to receive a sixth signal from the first node through the first channel, the sixth signal being a signal obtained after a fifth signal is transmitted to the distributed learning device through the first channel, the fifth signal including a signal generated by mapping third intermediate data to an air interface resource, the third intermediate data being for updating the first channel; the processing module is further configured to obtain third information based on the sixth signal, the third information being determined based on the third intermediate data and the information about the third channel; The processing module is further configured to update the second channel based on the third information and the error information of the second intermediate data. 40. A distributed learning device according to any one of items 34 to 39. [Item 41] The transceiver module is further configured to transmit second information to the first node, the second information being for obtaining error information of the first intermediate data. 41. A distributed learning device according to any one of items 34 to 40. [Item 42] the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data. Item 42. The distributed learning device according to item 41. [Item 43] The processing module is further configured to update the second data model based on the output data to obtain a new second data model. 43. A distributed learning device according to any one of items 34 to 42. [Item 44] the transceiver module is further configured to receive a second signal from the first node through the first channel, the second signal being a signal obtained after a first signal is transmitted through the first channel to the distributed learning device, the first signal being for determining the information about the first channel; The processing module is further configured to obtain the information about the first channel based on the second signal. 44. A distributed learning device according to any one of items 34 to 43. [Item 45] A distributed learning device, the distributed learning device comprising a processor, the processor coupled to a memory; the memory configured to store a computer program; The processor is configured to execute the computer program stored in the memory to enable the distributed learning device to perform the distributed learning method according to any one of items 1 to 22. Distributed learning device. [Item 46] 23. A computer-readable storage medium, the computer-readable storage medium comprising a computer program or instructions, the computer program or instructions being executed on a computer to perform the method according to any one of items 1 to 22. [Item 47] 23. A computer program product, the computer program product comprising a computer program or instructions, the computer program or instructions being executed on a computer to perform the method according to any one of items 1 to 22. [Item 48] 23. A communication device configured to perform the method according to any one of items 1 to 22. [Item 49] 23. A computer program which, when run on a computer, performs the method according to any one of items 1 to 22. [Item 50] A communication system comprising an apparatus according to any one of claims 23 to 33 and an apparatus according to any one of claims 34 to 44. [Item 51] 23. A chip system, the chip system comprising a processor, the processor configured to control a device in which the chip system is mounted to implement the method according to any one of items 1 to 22.

Claims

1. A distributed learning method applied to a first node, the first node having a first one of data models, the distributed learning method comprising: processing the first data by using the first data model to obtain first intermediate data; and transmitting the first intermediate data to a second node through a first channel, where the first channel is updated based on error information of a second intermediate data, information about the first channel, and the first intermediate data, the second intermediate data being a result of transmission of the first intermediate data to the second node through the first channel, the first channel being a channel between the first node and the second node. Equipped with The step of transmitting the first intermediate data over the first channel to the second node includes: modulating the first intermediate data into symbols and transmitting the symbols over the first channel to the second node; A method for distributed learning, wherein the number of modulated symbols is the same as a dimension of a hidden layer of the data model, or the number of modulated symbols is half the dimension of the hidden layer of the data model.

2. the first channel having a second channel and a third channel, and the distributed learning method comprises: updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data. The method of claim 1 further comprising:

3. The step of updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data comprises: receiving first information transmitted by the second node, where the first information is determined based on the error information of the second intermediate data and the information about the third channel; and updating the second channel based on the first information and the first intermediate data. The method of claim 2 , further comprising:

4. The step of receiving first information transmitted by the second node comprises: receiving the first information transmitted by the second node over a fourth channel. The distributed learning method according to claim 3.

5. The step of updating the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data comprises: receiving a third signal from the second node over the first channel, where the third signal is a signal obtained after a fourth signal is transmitted to the first node over the first channel, and the fourth signal includes a signal generated by mapping the error information of the second intermediate data to an air interface resource; obtaining first information based on the third signal, where the first information is determined based on the error information of the second intermediate data and the information about the third channel; and updating the second channel based on the first information and the first intermediate data. The method of claim 2 , further comprising:

6. The distributed learning method includes: transmitting a first signal over the first channel to the second node, the first signal being for determining the information about the first channel. The method of claim 1 further comprising:

7. The distributed learning method includes: Transmitting a fifth signal over the first channel to the second node, where the fifth signal comprises a signal generated by mapping third intermediate data to air interface resources, the third intermediate data being for updating the first channel. The method of claim 1 further comprising:

8. The distributed learning method includes: updating the first data model based on error information of the first intermediate data to obtain a new first data model; The method of claim 1 further comprising:

9. The distributed learning method includes: receiving second information transmitted by the second node, where the second information is for obtaining the error information of the first intermediate data; The method of claim 8 further comprising:

10. the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data. The distributed learning method of claim 9.

11. The distributed learning method comprises: receiving an eighth signal from the second node, where the eighth signal is obtained after a seventh signal is transmitted over a channel to the first node, the seventh signal comprising a signal generated by mapping the error information of the second intermediate data to an air interface resource; and obtaining the error information of the first intermediate data based on the eighth signal; The method of claim 1 further comprising:

12. A distributed learning method applied to a second node, the second node having a second one of the data models, the distributed learning method comprising: receiving second intermediate data over a first channel, where the second intermediate data is a result of a transmission of first intermediate data sent by a first node over the first channel to the second node, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the first channel being a channel between the first node and the second node; and processing the second intermediate data by using the second data model to obtain output data; Equipped with The step of receiving the second intermediate data from the first node over the first channel includes: performing dimensional matching on the received second intermediate data symbols; A distributed learning method, wherein the number of demodulated symbols is the same as the dimension of the intermediate layer of the data model, or one complex-valued symbol of the second intermediate data is expanded to two real numbers.

13. the first channel having a second channel and a third channel, and the distributed learning method comprises: transmitting the error information of the second intermediate data and information about the third channel to the first node. The method of claim 12 further comprising:

14. The step of transmitting the error information of the second intermediate data and information about the third channel to the first node comprises: transmitting first information to the first node, where the first information is determined based on the error information of the second intermediate data and the information about the third channel. The method of claim 13 , comprising:

15. The step of transmitting the first information to the first node comprises: transmitting the first information to the first node over a fourth channel. The method of claim 14 , comprising:

16. The step of transmitting the error information of the second intermediate data and information about the third channel to the first node comprises: transmitting a fourth signal over the first channel to the first node, where the fourth signal includes a signal generated by mapping the error information of the second intermediate data to air interface resources. The method of claim 13 , comprising:

17. The distributed learning method includes: receiving a second signal from the first node over the first channel, the second signal being a signal obtained after a first signal is transmitted over the first channel to the second node, the first signal being for determining the information about the first channel; and obtaining said information about said first channel based on said second signal; The method of claim 12 further comprising:

18. The distributed learning method includes: receiving a sixth signal from the first node over the first channel, where the sixth signal is obtained after a fifth signal is transmitted over the first channel to the second node, the fifth signal having a signal generated by mapping third intermediate data to air interface resources, the third intermediate data being for updating the first channel; obtaining third information based on the sixth signal, where the third information is determined based on the third intermediate data and the information about the third channel; and updating the second channel based on the third information and the error information of the second intermediate data. The method of claim 13 further comprising:

19. The distributed learning method includes: Sending second information to the first node, where the second information is for obtaining error information of the first intermediate data. The method of claim 12 further comprising:

20. the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data.

20. The distributed learning method of claim 19.

21. The distributed learning method includes: updating the second data model based on the output data to obtain a new second data model. The method of claim 12 further comprising:

22. The distributed learning method includes: transmitting a seventh signal to the first node, wherein the seventh signal comprises a signal generated by mapping the error information of the second intermediate data to air interface resources. The method of claim 12 further comprising:

23. A distributed learning device, the distributed learning device comprising a first data model of a data model, the distributed learning device comprising a processing module and a transceiver module; the processing module is configured to process the first data by using the first data model to obtain first intermediate data; the transceiver module is configured to transmit the first intermediate data to a second node through a first channel, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the second intermediate data being a result of the transmission of the first intermediate data to the second node through the first channel, the first channel being a channel between the distributed learning device and the second node; The transmitting of the first intermediate data to the second node over the first channel includes: modulating the first intermediate data into symbols and transmitting the symbols over the first channel to the second node; A distributed learning device, wherein the number of the modulated symbols is the same as a dimension of a hidden layer of the data model, or the number of the modulated symbols is half the dimension of the hidden layer of the data model.

24. the first channel having a second channel and a third channel; The processing module is further configured to update the second channel based on the error information of the second intermediate data, information about the third channel, and the first intermediate data.

24. The distributed learning device of claim 23.

25. the transceiver module is further configured to receive first information transmitted by the second node, the first information being determined based on the error information of the second intermediate data and the information about the third channel; The processing module is further configured to update the second channel based on the first information and the first intermediate data.

25. The distributed learning device of claim 24.

26. The transceiver module is further configured to receive the first information transmitted by the second node over a fourth channel.

26. The distributed learning device of claim 25.

27. the transceiver module is further configured to receive a third signal from the second node through the first channel, the third signal being a signal obtained after a fourth signal is transmitted to the distributed learning device through the first channel, the fourth signal including a signal generated by mapping the error information of the second intermediate data to an air interface resource; the processing module is further configured to obtain first information based on the third signal, the first information being determined based on the error information of the second intermediate data and the information about the third channel; The processing module is further configured to update the second channel based on the first information and the first intermediate data.

25. The distributed learning device of claim 24.

28. The transceiver module is further configured to transmit a first signal over the first channel to the second node, the first signal being for determining the information about the first channel.

24. The distributed learning device of claim 23.

29. The transceiver module is further configured to transmit a fifth signal to the second node over the first channel, the fifth signal including a signal generated by mapping third intermediate data to air interface resources, the third intermediate data being for updating the first channel.

24. The distributed learning device of claim 23.

30. The processing module is further configured to update the first data model based on error information of the first intermediate data to obtain a new first data model.

24. The distributed learning device of claim 23.

31. The transceiver module is further configured to receive second information transmitted by the second node, the second information being for obtaining the error information of the first intermediate data.

31. The distributed learning device of claim 30.

32. the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data.

32. The distributed learning device of claim 31.

33. the transceiver module is further configured to receive an eighth signal from the second node, the eighth signal being a signal obtained after a seventh signal is transmitted through a channel to the distributed learning device, the seventh signal including a signal generated by mapping the error information of the second intermediate data to an air interface resource; The processing module is further configured to obtain the error information of the first intermediate data based on the eighth signal.

24. The distributed learning device of claim 23.

34. A distributed learning device, the distributed learning device comprising a second data model of a data model, the distributed learning device comprising a processing module and a transceiver module; the transceiver module is configured to receive second intermediate data through a first channel, the second intermediate data being a result of transmission of the first intermediate data sent by a first node through the first channel to the distributed learning device, the first channel being updated based on error information of the second intermediate data, information about the first channel, and the first intermediate data, the first channel being a channel between the first node and the distributed learning device; the processing module is configured to process the second intermediate data by using the second data model to obtain output data; The receiving of the second intermediate data from the first node over the first channel includes: performing dimensional matching on the received second intermediate data symbols; A distributed learning device, wherein the number of demodulated symbols is the same as the dimension of an intermediate layer of the data model, or one complex-valued symbol of the second intermediate data is expanded to two real numbers.

35. 35. The distributed learning device of claim 34, wherein the first channel has a second channel and a third channel, and the transceiver module is further configured to transmit the error information of the second intermediate data and information about the third channel to the first node.

36. The transceiver module is further configured to transmit first information to the first node, the first information being determined based on the error information of the second intermediate data and the information about the third channel.

36. The distributed learning device of claim 35.

37. The transceiver module is further configured to transmit the first information to the first node over a fourth channel.

37. The distributed learning device of claim 36.

38. The transceiver module is further configured to transmit a fourth signal to the first node over the first channel, the fourth signal including a signal generated by mapping the error information of the second intermediate data to an air interface resource.

36. The distributed learning device of claim 35.

39. the transceiver module is further configured to receive a second signal from the first node through the first channel, the second signal being a signal obtained after a first signal is transmitted to the distributed learning device through the first channel, the first signal being for determining the information about the first channel; The processing module is further configured to obtain the information about the first channel based on the second signal.

35. The distributed learning device of claim 34.

40. the transceiver module is further configured to receive a sixth signal from the first node through the first channel, the sixth signal being a signal obtained after a fifth signal is transmitted to the distributed learning device through the first channel, the fifth signal including a signal generated by mapping third intermediate data to an air interface resource, the third intermediate data being for updating the first channel; the processing module is further configured to obtain third information based on the sixth signal, the third information being determined based on the third intermediate data and the information about the third channel; The processing module is further configured to update the second channel based on the third information and the error information of the second intermediate data.

36. The distributed learning device of claim 35.

41. The transceiver module is further configured to transmit second information to the first node, the second information being for obtaining error information of the first intermediate data.

35. The distributed learning device of claim 34.

42. the second information includes the error information of the first intermediate data; or The second information includes the error information of the second intermediate data and the information about the first channel, and the error information of the second intermediate data and the information about the first channel are for determining the error information of the first intermediate data.

42. The distributed learning device of claim 41.

43. The processing module is further configured to update the second data model based on the output data to obtain a new second data model.

35. The distributed learning device of claim 34.

44. the transceiver module is further configured to receive a second signal from the first node through the first channel, the second signal being a signal obtained after a first signal is transmitted to the distributed learning device through the first channel, the first signal being for determining the information about the first channel; The processing module is further configured to obtain the information about the first channel based on the second signal.

35. The distributed learning device of claim 34.

45. A distributed learning device, the distributed learning device comprising a processor, the processor coupled to a memory; the memory configured to store a computer program; The processor is configured to execute the computer program stored in the memory to enable the distributed learning device to perform the distributed learning method according to claim 1 or 12. Distributed learning device.

46. 13. A computer readable storage medium, the computer readable storage medium comprising a computer program or instructions, the computer program or instructions being executed on a computer to perform the distributed learning method of claim 1 or 12.

47. A communication device configured to perform the distributed learning method according to claim 1 or 12.

48. A computer program, which, when run on a computer, performs the distributed learning method according to claim 1 or 12.

49. A communication system comprising the distributed learning device according to claim 23 and the distributed learning device according to claim 34.

50. A chip system, the chip system comprising a processor, the processor configured to control a device in which the chip system is mounted to implement the distributed learning method of claim 1 or 12.

Images