Neural network model training method and communication apparatus

Abstract

The present application provides a neural network model training method and a communication apparatus. The method comprises: a sending end device performing forward reasoning on training data comprised in a first training process by means of a first neural network model, to obtain a first forward reasoning result; sending the first forward reasoning result to a receiving end device; further, the sending end device receiving a first feedback result from the receiving end device for the first forward reasoning result; obtaining a first intermediate result generated when performing the forward reasoning on the training data comprised in the first training process; the sending end device calculating a gradient of the first training process on the basis of the first feedback result, the first forward reasoning result, and the first intermediate result; and further, the sending end device updating a first model parameter to a second model parameter on the basis of the gradient corresponding to the first training process. According to the neural network model training method, the performance of a wireless communication system using the neural network model is optimized.

Description

A neural network model training method and communication device Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a neural network model training method and a communication device. Background Art

[0002] Combining wireless communication systems with neural networks—that is, training the neural network models within wireless communication systems through data-driven training—helps improve the performance of wireless communication systems. Neural network models can be applied to various devices in wireless communication systems (including receivers and transmitters). For example, applying a neural network model to the channel decoding module in a receiver device through data-driven training can help improve the channel decoding performance of that receiver device.

[0003] Neural network models typically used in wireless communication systems are trained in a training system before being deployed in wireless communication system devices. The resulting neural network model, trained using this method, exhibits poor compatibility with the communication environment (or, more specifically, the communication channels within that environment), which can impact the performance of the wireless communication system in which it is applied. Improving the compatibility between neural network models and communication channels is a pressing issue.

[0004] Summary of the Invention

[0005] The present application provides a neural network model training method and a communication device. For a neural network model deployed on a device, the training method updates the neural network model through training data of a communication network, which is beneficial to improving the adaptability of the neural network model to the communication channel, thereby facilitating optimizing the performance of a wireless communication system that applies the neural network model.

[0006] In a first aspect, the present application provides a neural network model training method, which is applied to a sending device, or to a module in the sending device (such as a chip or a chip system, etc.). Taking the application to the sending device as an example, the method includes: the sending device sends a first forward reasoning result to the receiving device, and the first forward reasoning result is obtained by the first neural network model based on the first model parameters to perform forward reasoning on the training data included in the first training process. The first training process is any one of the at least one training processes corresponding to the first neural network model, and the first neural network model is a model deployed on the sending device; the sending device receives a first feedback result for the first forward reasoning result from the receiving device; the sending device obtains a first intermediate result, which is generated when forward reasoning is performed on the training data included in the first training process; the sending device calculates the gradient of the first training process based on the first feedback result, the first forward reasoning result and the first intermediate result; further, the sending device updates the first model parameter to the second model parameter based on the gradient corresponding to the first training process.

[0007] Based on the method described in the first aspect, after the first neural network model is deployed on the transmitting device, the transmitting device and the receiving device exchange information through the communication system (or understood as the real channel of the actual communication environment). That is, after the transmitting device performs forward reasoning on the training data through the first neural network model, the forward reasoning result is sent to the receiving device through the channel of the actual communication network; when the receiving device receives the forward reasoning result through the channel, the forward reasoning result will be affected by the transmission environment of the channel (that is, it can be understood as carrying information about the real channel), so the feedback result obtained based on such forward reasoning result can also be understood as carrying information about the real channel; the transmitting device subsequently receives the feedback result and updates the model parameters of the first neural network model deployed with the transmitting device based on the feedback result, which is conducive to improving the adaptability of the neural network model to the actual communication scenario, thereby facilitating the optimization of the performance of the wireless communication system to which the neural network model is applied.

[0008] In one possible implementation, the sending device obtains the first intermediate result from the parameter cache space corresponding to the first training process. By implementing this possible implementation, the sending device directly obtains the first intermediate result from the parameter cache space, which helps save computing resources of the sending device.

[0009] In one possible implementation, after the transmitting device calculates the gradient of the first training process based on the first feedback result, the first forward reasoning result, and the first intermediate result, the transmitting device clears the first intermediate result from the parameter cache space corresponding to the first training process. By implementing this possible implementation, the transmitting device clears the first intermediate result from the parameter cache space after using the first intermediate result, which helps conserve cache resources of the transmitting device.

[0010] In a possible implementation, the sending end device may perform forward reasoning again based on the training data included in the first training process to obtain a first intermediate result.

[0011] In one possible embodiment, the first training process includes at least one sub-process, each sub-process includes at least one training data, the first forward reasoning result includes the forward reasoning result of each training data in the training data included in the first training process, the first intermediate result includes the intermediate result corresponding to the forward reasoning result of each training data, the first feedback result includes the feedback result corresponding to the forward reasoning result of each training data, and the forward reasoning result of each training data is obtained based on the first model parameter.

[0012] In one possible implementation, at least one of the following is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or channel conditions: the amount of training data included in the first training process; or the number of subprocesses included in the first training process; or the amount of training data included in the subprocesses. Implementing this possible implementation helps improve the compatibility between the neural network model training process and the communication device, thereby improving training effectiveness.

[0013] In one possible implementation, the transmitting device updates the first model parameter to the third model parameter based on the gradient corresponding to the second training process; wherein the second training process is any one of the at least one training process except the first training process, the gradient of the second training process is obtained by reverse gradient calculation based on the second feedback result, the second forward reasoning result, and the second intermediate result, the second forward reasoning result is obtained by forward reasoning on the training data included in the second training process based on the first model parameter, the second intermediate result is generated when forward reasoning on the training data included in the first training process, and the second feedback result is the feedback result for the second forward reasoning result; further, the transmitting device updates the model parameter of the first neural network model to the fourth model parameter based on the second model parameter and the third model parameter. By implementing this possible implementation, the transmitting device can update the model parameters of the first neural network model in parallel through multiple training processes, and finally update the model parameters of the first neural network model based on the model parameters obtained from multiple training processes, which is conducive to improving the training speed of the first neural network model.

[0014] In one possible implementation, the number of training processes of at least one training process corresponding to the first neural network model is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or channel conditions.

[0015] In one possible embodiment, the capabilities of the transmitting device include one or more of the computing speed of the transmitting device, the size of the cache space of the transmitting device, the forward reasoning time of the transmitting device for a reference model, and / or the reverse gradient calculation time; the capabilities of the receiving device include one or more of the computing speed of the receiving device, the size of the cache space of the receiving device, the forward reasoning time of the receiving device for a reference model, and / or the reverse gradient calculation time; the channel conditions include one or more of the wireless resources, channel quality, communication rate, and / or communication delay for communication between the transmitting device and the receiving device.

[0016] In a possible implementation, the transmitting device receives capability information from the receiving device, where the capability information is used to indicate the capability of the receiving device.

[0017] In one possible implementation, different training processes of the first neural network model correspond to different parameter cache spaces. By implementing this possible implementation, the parameter cache spaces can be used to distinguish between the various training processes, which helps improve the efficiency of the sending end device in obtaining the intermediate results of each training process.

[0018] In one possible implementation, different sub-processes in the first training process correspond to different parameter cache spaces. By implementing this possible implementation, each sub-process can be distinguished by the parameter cache space, which is conducive to improving the efficiency of the sending end device in obtaining the intermediate results of each sub-process.

[0019] In the second aspect, the present application provides a neural network model training method, which is applied to a receiving device, or to a module in the receiving device (such as a chip or a chip system, etc.). Taking the application to the receiving device as an example, the method includes: the receiving device receives a first forward reasoning result from the sending device, and the first forward reasoning result is obtained by the first neural network model based on the first model parameters to perform forward reasoning on the training data included in the first training process. The first training process is any one of the at least one training processes corresponding to the first neural network model, and the first neural network model is a model deployed on the sending device; the receiving device sends the first feedback result to the sending device based on the first forward reasoning result.

[0020] Based on the method described in the second aspect, after the first neural network model is deployed on the transmitting device, the receiving device receives the forward reasoning result through the channel of the communication system. It can be understood that the forward reasoning result will be affected by the transmission environment of the channel (that is, it carries the information of the real channel). Therefore, the feedback result obtained based on such a forward reasoning result can also be understood as carrying the information of the real channel. In this case, the receiving device sends the feedback result to the transmitting device, which is conducive to improving the adaptability of the feedback result to the actual communication scenario, thereby facilitating the optimization of the performance of the wireless communication system using the first neural network model. Based on the beneficial effects of other embodiments described in the second aspect, reference can be made to the beneficial effects of the embodiments described in the first aspect, and no further details will be given later.

[0021] In one possible embodiment, the first training process includes at least one sub-process, each sub-process includes at least one training data, the first forward reasoning result includes the forward reasoning result of each training data in the training data included in the first training process, the first intermediate result includes the intermediate result corresponding to the forward reasoning result of each training data, the first feedback result includes the feedback result corresponding to the forward reasoning result of each training data, and the forward reasoning result of each training data is obtained based on the first model parameter.

[0022] In one possible implementation, at least one of the following is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or the channel conditions: the amount of training data included in the first training process; or the number of sub-processes included in the first training process; or the amount of training data included in the sub-processes.

[0023] In one possible implementation, a receiving device receives a second forward reasoning result from a sending device, where the second forward reasoning result is obtained by performing forward reasoning on training data included in a second training process based on the first model parameters, where the second training process is any one of at least one training process other than the first training process; the receiving device performs a reverse gradient calculation based on the second forward reasoning result to obtain a second feedback result; and sends the second feedback result to the sending device, where the second feedback result is used to update the model parameters of the first neural network model in combination with the first feedback result.

[0024] In one possible implementation, the number of training processes of at least one training process corresponding to the first neural network model is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or channel conditions.

[0025] In one possible embodiment, the capabilities of the transmitting device include one or more of the computing speed of the transmitting device, the size of the cache space of the transmitting device, the forward reasoning time of the transmitting device for a reference model, and / or the reverse gradient calculation time; the capabilities of the receiving device include one or more of the computing speed of the receiving device, the size of the cache space of the receiving device, the forward reasoning time of the receiving device for a reference model, and / or the reverse gradient calculation time; the channel conditions include one or more of the wireless resources, channel quality, communication rate, and / or communication delay for communication between the transmitting device and the receiving device.

[0026] In one possible implementation, a second neural network model is deployed in a receiving device. The receiving device updates the fifth model parameter of the second neural network model to the sixth model parameter based on the first forward reasoning result; and updates the fifth model parameter of the second neural network model to the seventh model parameter based on the second forward reasoning result; and further, obtains the eighth model parameter based on the sixth model parameter and the seventh model parameter. By implementing this possible implementation, the receiving device can concurrently update the model parameters of the second neural network model using the forward reasoning results of multiple training processes, and finally update the model parameters of the second neural network model based on the model parameters obtained from the multiple training processes, which is conducive to improving the training speed of the second neural network model.

[0027] In a possible implementation, the receiving device is deployed with a second neural network model, and the receiving device updates the fifth model parameter of the second neural network model to a sixth model parameter based on the first forward reasoning result.

[0028] In a third aspect, the present application provides a communication device, which may be a transmitting end device, a device in a transmitting end device, or a device that can be used in conjunction with a transmitting end device. The communication device may also be a chip system. The communication device may execute the method described in the first aspect. The functions of the communication device may be implemented by hardware, or by hardware executing corresponding software implementations. The hardware or software includes one or more units or modules corresponding to the above functions. The units or modules may be software and / or hardware. The operations and beneficial effects performed by the communication device may refer to the method and beneficial effects described in the first aspect above.

[0029] In a fourth aspect, the present application provides a communication device, which may be a receiving device, a device in a receiving device, or a device that can be used in conjunction with a receiving device. The communication device may also be a chip system. The communication device may execute the method described in the second aspect. The functions of the communication device may be implemented by hardware, or by hardware executing corresponding software implementations. The hardware or software includes one or more units or modules corresponding to the above functions. The unit or module may be software and / or hardware. The operations and beneficial effects performed by the communication device may refer to the methods and beneficial effects described in the second aspect above.

[0030] In a fifth aspect, the present application provides a communication device, which includes a processor and an interface circuit, wherein the interface circuit is used to receive signals from other communication devices outside the communication device and transmit them to the processor or send signals from the processor to other communication devices outside the communication device, and the processor is used to implement the method as described in the first aspect through a logic circuit or executing code instructions, or the processor is used to implement the method as described in the second aspect through a logic circuit or executing code instructions.

[0031] In a sixth aspect, the present application provides a computer-readable storage medium, in which a computer program or instruction is stored. When the computer program or instruction is executed by a communication device, the method described in the first aspect or the method described in the second aspect is implemented.

[0032] In a seventh aspect, the present application provides a computer program product comprising instructions, which, when a communication device reads and executes the instructions, causes the communication device to execute the method as described in the first aspect, or causes the communication device to execute the method as described in the second aspect.

[0033] In an eighth aspect, the present application provides a communication system comprising a communication device for executing the method described in the first aspect and a communication device for executing the method described in the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] FIG1a is a schematic diagram of the architecture of a communication system provided in an embodiment of the present application;

[0035] FIG1b is another schematic diagram of a wireless communication system applicable to an embodiment of the present application;

[0036] FIG2 is a schematic diagram of a wireless communication system based on a neural network transmitter and receiver provided in an embodiment of the present application;

[0037] FIG3 is a schematic diagram of different modules in a transceiver optimized by several neural network models provided in an embodiment of the present application;

[0038] FIG4 is a schematic diagram of a training process of an intelligent air interface provided in an embodiment of the present application;

[0039] FIG5 is a flow chart of a neural network model training method provided in an embodiment of the present application;

[0040] FIG6 is a flow chart of another neural network model training method provided in an embodiment of the present application;

[0041] FIG7 is a flow chart of another neural network model training method provided in an embodiment of the present application;

[0042] FIG8 is a schematic structural diagram of a communication device provided in an embodiment of the present application;

[0043] FIG9 is a schematic structural diagram of another communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of this application clearer, this application will be further described in detail below with reference to the accompanying drawings.

[0045] The terms "first" and "second" and the like in the specification, claims, and drawings of this application are used to distinguish between different objects, not to describe a particular order. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of operations or units is not limited to the listed operations or units, but may optionally include operations or units not listed, or may optionally include other operations or units inherent to the process, method, product, or apparatus.

[0046] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0047] In order to facilitate a detailed understanding of the embodiments of the present application, the system architecture involved in the embodiments of the present application is first introduced below.

[0048] Figure 1a is a schematic diagram of the architecture of a communication system 1000 used in an embodiment of the present application. As shown in Figure 1a, the communication system includes a radio access network (RAN) 100 and a core network 200. Optionally, the communication system 1000 may also include the Internet 300. The RAN 100 includes at least one RAN node (such as 110a and 110b in Figure 1a, collectively referred to as 110) and may also include at least one terminal (such as 120a-120j in Figure 1a, collectively referred to as 120). The RAN 100 may also include other RAN nodes, such as wireless relay devices and / or wireless backhaul devices (not shown in Figure 1a). The terminal 120 is wirelessly connected to the RAN node 110, and the RAN node 110 is wirelessly or wiredly connected to the core network 200. The core network devices in the core network 200 and the RAN node 110 in the RAN 100 may be independent and different physical devices, or they may be the same physical device that integrates the logical functions of the core network devices and the logical functions of the RAN nodes. Terminals and RAN nodes may be connected to each other via wired or wireless means.

[0049] RAN100 may be an evolved universal terrestrial radio access (E-UTRA) system, a new radio (NR) system, or a future radio access system defined in the 3rd Generation Partnership Project (3GPP). RAN100 may also include two or more of the aforementioned different radio access systems. RAN100 may also be an open RAN (O-RAN).

[0050] A RAN node, also known as a radio access network device, RAN entity, or access node, and hereinafter referred to as a network device, facilitates wireless access to a communication system by a terminal. In one application scenario, a RAN node can be a base station, an evolved NodeB (eNodeB), a transmission reception point (TRP), a next-generation NodeB (gNB) in a fifth-generation (5G) mobile communication system, a next-generation base station in a sixth-generation (6G) mobile communication system, or a base station in a future mobile communication system. A RAN node can be a macro base station (such as 110a in Figure 1a), a micro base station, an indoor station (such as 110b in Figure 1a), a relay node, or a donor node.

[0051] In another application scenario, multiple RAN nodes can collaborate to help terminals achieve wireless access, with different RAN nodes implementing portions of the base station's functions. For example, a RAN node can be a centralized unit (CU), a distributed unit (DU), or a radio unit (RU). The CU implements the base station's radio resource control protocol and packet data convergence protocol (PDCP) functions, as well as the service data adaptation protocol (SDAP) functions. The DU implements the base station's radio link control layer and medium access control (MAC) layer functions, as well as some or all of the physical layer functions. For detailed descriptions of each of the above protocol layers, please refer to the relevant 3GPP technical specifications. The RU can be used to implement the transmission and reception of radio frequency signals. The CU and DU can be two independent RAN nodes, or they can be integrated into the same RAN node, such as in a baseband unit (BBU). The RU can be included in radio frequency equipment, such as a remote radio unit (RRU) or an active antenna unit (AAU). The CU can be further divided into two types of RAN nodes: CU-control plane and CU-user plane.

[0052] The RAN node can support one or more types of fronthaul interfaces. Different fronthaul interfaces correspond to DUs and RUs with different functions. If the fronthaul interface between the DU and the RU is the common public radio interface (CPRI), the DU is configured to implement one or more baseband functions, and the RU is configured to implement one or more radio frequency functions. If the fronthaul interface between the DU and the RU is the enhanced common public radio interface (eCPRI), relative to CPRI, part of the downlink and / or uplink baseband functions are moved from the DU to the RU for implementation. The division between the DU and the RU is different, corresponding to different types (category, Cat) of eCPRI, such as eCPRI Cat A, B, C, D, E, and F.

[0053] Taking eCPRI Cat A as an example, for downlink transmission, based on layer mapping, the DU is configured to implement layer mapping and one or more functions preceding it (i.e., one or more of coding, rate matching, scrambling, modulation, and layer mapping). Other functions after layer mapping (e.g., RE mapping, digital beamforming (BF), or one or more of inverse fast Fourier transform (IFFT) / cyclic prefix (CP) addition) are moved to the RU for implementation. For uplink transmission, based on RE demapping, the DU is configured to implement demapping and one or more functions preceding it (i.e., one or more of decoding, rate matching, descrambling, demodulation, inverse discrete Fourier transform (IDFT), channel equalization, and RE demapping). Other functions after demapping (e.g., one or more of digital BF or fast Fourier transform (FFT) / CP removal) are moved to the RU for implementation. It is understandable that for the functional description of DU and RU corresponding to various types of eCPRI, reference can be made to the eCPRI protocol, which will not be described in detail here.

[0054] In one possible design, the processing unit for implementing baseband functions in the BBU is called a baseband high layer (BBH) unit, and the processing unit for implementing baseband functions in the RRU / AAU / RRH is called a baseband low layer (BBL) unit.

[0055] In different systems, RAN nodes may have different names. For example, in an O-RAN system, a CU may be called an open CU (O-CU), a DU may be called an open DU (O-DU), and a RU may be called an open RU (O-RU). The RAN node in the embodiments of the present application may be implemented by a software module, a hardware module, or a combination of a software module and a hardware module. For example, the RAN node may be a server loaded with the corresponding software module. The embodiments of the present application do not limit the specific technology and specific device form adopted by the RAN node. For ease of description, the following description takes a base station as an example of a RAN node.

[0056] A terminal is a device with wireless transceiver capabilities that can send signals to a base station or receive signals from a base station. A terminal may also be referred to as a terminal device, user equipment (UE), mobile station, mobile terminal, etc. Terminals can be widely used in various scenarios, such as device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IoT), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wearable, smart transportation, smart city, etc. A terminal can be a mobile phone, tablet computer, computer with wireless transceiver capabilities, wearable device, vehicle, airplane, ship, robot, robotic arm, smart home device, etc. The embodiments of this application do not limit the specific technology and specific device form adopted by the terminal.

[0057] Base stations and terminals can be fixed or mobile. They can be deployed on land, indoors or outdoors, handheld or vehicle-mounted; on water; or on aircraft, balloons, and satellites. The embodiments of this application do not limit the application scenarios of base stations and terminals.

[0058] The roles of base stations and terminals can be relative. For example, the helicopter or drone 120i in Figure 1a can be configured as a mobile base station. To terminals 120j accessing the wireless access network 100 via 120i, terminal 120i is a base station. However, to base station 110a, 120i is a terminal, meaning that communication between 110a and 120i occurs via a wireless air interface protocol. Of course, communication between 110a and 120i can also occur via a base station-to-base station interface protocol. In this case, 120i is also a base station relative to 110a. Therefore, base stations and terminals can be collectively referred to as communication devices. 110a and 110b in Figure 1a can be referred to as communication devices with base station functionality, while 120a-120j in Figure 1a can be referred to as communication devices with terminal functionality.

[0059] Communication between base stations and terminals, between base stations, and between terminals can be carried out through authorized spectrum, unauthorized spectrum, or both; communication can be carried out through spectrum below 6 gigahertz (GHz), spectrum above 6 GHz, or spectrum below 6 GHz and spectrum above 6 GHz. The embodiments of the present application do not limit the spectrum resources used for wireless communication.

[0060] In the embodiments of the present application, the functions of the base station may also be performed by a module (such as a chip) in the base station, or by a control subsystem that includes the base station functions. The control subsystem that includes the base station functions here may be a control center in the above-mentioned application scenarios such as smart grid, industrial control, smart transportation, and smart city. The functions of the terminal may also be performed by a module (such as a chip or modem) in the terminal, or by a device that includes the terminal functions.

[0061] Refer to FIG. 1 b , which is another schematic diagram of a wireless communication system applicable to an embodiment of the present application.

[0062] As shown in Figure 1b, the wireless communication system includes a RAN intelligent controller (RIC). As an example, RIC can be used to implement AI-related functions. As an example, the RIC includes a near-real time RIC (near-real time RIC, near-RT RIC) and a non-real time RIC (non-real time RIC, Non-RT RIC). Among them, the non-real-time RIC mainly processes non-real-time information, such as data that is not sensitive to delay, and the delay of the data can be in the order of seconds. Real-time RIC mainly processes near real-time information, such as data that is relatively sensitive to delay, and the delay of the data is in the order of tens of milliseconds.

[0063] The near real-time RIC is used for model training and reasoning. For example, it is used to train an AI model and use the AI ​​model for reasoning. The near real-time RIC can obtain network-side and / or terminal-side information from a RAN node (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or a terminal. This information can be used as training data or reasoning data. Optionally, the near real-time RIC can deliver the reasoning result to the RAN node and / or the terminal. Optionally, the reasoning result can be exchanged between the CU and the DU, and / or between the DU and the RU. For example, the near real-time RIC delivers the reasoning result to the DU, and the DU sends it to the RU.

[0064] The non-real-time RIC is also used for model training and reasoning. For example, it is used to train an AI model and use the model for reasoning. The non-real-time RIC can obtain network-side and / or terminal-side information from RAN nodes (such as CU, CU-CP, CU-UP, DU and / or RU) and / or terminals. This information can be used as training data or reasoning data, and the reasoning results can be submitted to the RAN node and / or terminal. Optionally, the reasoning results can be exchanged between the CU and the DU, and / or between the DU and the RU. For example, the non-real-time RIC submits the reasoning results to the DU, and the DU sends it to the RU.

[0065] The near real-time RIC and non-real-time RIC may also be separately configured as a network element. Optionally, the near real-time RIC and non-real-time RIC may also be part of other devices. For example, the near real-time RIC is configured in a RAN node (e.g., a CU or DU), while the non-real-time RIC is configured in an OAM, a cloud server, a core network device, or other network device.

[0066] In practical applications, the wireless communication system may include multiple network devices (also called access network devices) and multiple terminal devices at the same time, without limitation. A network device may serve one or more terminal devices at the same time. A terminal device may also access one or more network devices at the same time. The embodiments of the present application do not limit the number of terminal devices and network devices included in the wireless communication system.

[0067] To facilitate understanding of the embodiments of the present application, the following describes several aspects of the embodiments of the present application:

[0068] (1) “Sending” and “receiving” in the embodiments of the present application indicate the direction of signal transmission. For example, “sending information to XX” can be understood as the destination of the information being XX, which can include direct sending through the air interface, as well as indirect sending through the air interface by other units or modules. “Receiving information from YY” can be understood as the source of the information being YY, which can include direct receiving from YY through the air interface, as well as indirect receiving from YY through the air interface from other units or modules. “Sending” can also be understood as the “output” of the chip interface, and “receiving” can also be understood as the “input” of the chip interface.

[0069] In other words, sending and receiving can be performed between devices, for example, between a network device and a terminal device, or can be performed within a device, for example, sending or receiving between components, modules, chips, software modules or hardware modules within the device through a bus, wiring or interface.

[0070] It is understandable that information may be processed between the source and destination of information transmission, such as coding, modulation, etc., but the destination can understand the valid information from the source. Similar expressions in this application can be understood similarly and will not be repeated.

[0071] (2) In the embodiments of the present application, "indication" may include direct indication and indirect indication, and may also include explicit indication and implicit indication. The information indicated by a certain information (such as the indication information described below) is called information to be indicated. In the specific implementation process, there are many ways to indicate the information to be indicated, such as but not limited to, directly indicating the information to be indicated, such as the information to be indicated itself or the index of the information to be indicated. The information to be indicated may also be indirectly indicated by indicating other information, wherein the other information is associated with the information to be indicated; or only a part of the information to be indicated may be indicated, while the other part of the information to be indicated is known or agreed in advance. For example, the indication of specific information may be achieved by means of the arrangement order of each information agreed in advance (such as predefined by the protocol), thereby reducing the indication overhead to a certain extent. The present application does not limit the specific method of indication. It is understandable that for the sender of the indication information, the indication information can be used to indicate the information to be indicated, and for the receiver of the indication information, the indication information can be used to determine the information to be indicated.

[0072] In order to facilitate understanding of the relevant contents of the embodiments of the present application, some of the terms involved in the embodiments of the present application are explained below. This part is only for ease of understanding and cannot be regarded as a disclosure or specific limitation of the technical solution of the present application.

[0073] 1. Neural Networks

[0074] A neural network can be composed of neural units, which can be represented by x s The output of the operation unit can be shown in formula (1).

[0075] Where, s = 1, 2, ... n, n is a natural number greater than 1, W s is x s The weight of the neural unit, b is the bias of the neural unit. f is the activation function of the neural unit, which is used to introduce nonlinear characteristics into the neural network to convert the input signal in the neural unit into the output signal. The output signal of the activation function can be used as the input of the next convolutional layer. The activation function can be a sigmoid function. A neural network is a network formed by connecting many of the above-mentioned single neural units together, that is, the output of one neural unit can be the input of another neural unit. The input of each neural unit can be connected to the local receptive field of the previous layer to extract the features of the local receptive field. The local receptive field can be an area composed of several neural units.

[0076] It should be noted that the neural network model mentioned in this application may be one or more of a neural network model, a deep neural network (DNN) network model, a convolutional neural network (CNN) network model, a recurrent neural network (RNN) network model, a generative adversarial network network model, or a variation (or combination) of a combination thereof, and this application does not specifically limit this.

[0077] 2. Intelligent air interface technology

[0078] Intelligent air interface technology can be understood as a technology that applies a neural network model in a wireless communication system to optimize the performance of the wireless communication system. Please refer to Figure 2, which shows a wireless communication system based on a neural network transmitter and receiver (hereinafter collectively referred to as a transceiver) provided by this application. In the system shown in Figure 2, the transceiver that applies the neural network model can optimize the performance of transmitting and receiving signals through data-driven training.

[0079] It can be understood that optimizing the performance of a transceiver by applying a neural network model is to optimize the channel coding, modulation, waveform, pilot and other modules used for signal processing in the transceiver by using a neural network model. For example, please refer to FIG3 , which is a schematic diagram of optimizing different modules in a transceiver using several neural network models provided in an embodiment of the present application. Among them, FIG3a is a schematic diagram of optimizing a modulation module and a waveform by using a neural network model; FIG3b is a schematic diagram of optimizing a coding module and a modulation module by using a neural network model; and FIG3c is a schematic diagram of optimizing a reference signal by using a neural network model.

[0080] In 3a of Figure 3, after the transmitted bits are encoded by forward error correction (FEC) to obtain coded bits, the coded bits are modulated by the modulation neural network model (i.e., the NN-Mod module in 3a of Figure 3), mapped to the physical resources, transformed into time domain signals by inverse fast Fourier transform (IFFT), and then processed by the time signal neural network model (i.e., the T-NN module in 3a of Figure 3) before being sent; the received signal is first processed by the time signal neural network model, transformed into frequency domain signals by fast Fourier transform (FFT), and then demodulated by the demodulation neural network model (i.e., the NN-DeMod module in 3a of Figure 3) before being sent to FEC decoding.

[0081] In 3b of Figure 3, the bits to be transmitted are processed by the coding modulation neural network model (i.e., the NN-CoMo module in 3b of Figure 3) to obtain modulated symbols, which are then processed by modules such as IFFT and sent; the receiving end sends the signals processed by modules such as FFT directly to the demodulation and decoding neural network model (i.e., the NN-DeCoMo module in 3b of Figure 3) for processing to obtain estimated transmitted bits.

[0082] In 3c of Figure 3, the transmitting end generates a reference signal based on the reference signal neural network model (i.e., the NN-RS module in 3c of Figure 3), and the receiving end processes the received signal based on the reference signal and the corresponding channel estimation neural network model (i.e., the NN-CE module in 3c of Figure 3) to obtain an estimated channel.

[0083] 3. Intelligent air interface training

[0084] The training of the intelligent air interface can be understood as the training of the neural network model used in the wireless communication system. Please refer to Figure 4, which is a schematic diagram of the training process of an intelligent air interface provided by this application. Among them, the sending neural network model performs forward reasoning on the training data, and sends the forward reasoning result to the receiving neural network model; the receiving neural network model performs forward reasoning and reverse gradient calculation based on the forward reasoning result to obtain a feedback result; the receiving neural network model sends the feedback result to the sending neural network model; the sending neural network model performs reverse gradient calculation based on the feedback result, and updates the model parameters of the sending neural network model based on the gradient obtained by the reverse gradient calculation. Exemplarily, the update of the tth round of the sending neural network model is as shown in formula (2).

[0085] Among them, w t+1 is the parameter of the t+1th round, that is, the parameter after the tth round update; w t is the parameter of the tth round; z k is the output of the sending neural network model corresponding to the kth training sample, l k is the loss function corresponding to the tth round, K is the number of samples in this round of training; η t is the learning rate; is the gradient of the loss function with respect to the output of the neural network model.

[0086] That is, during any round of training, the neural network model is first sent for forward inference, obtaining the forward inference results and intermediate results. The intermediate inference results are cached. Then, the loss function is calculated from the last layer, and the gradient is calculated layer by layer according to the chain rule. The gradient calculation process requires the forward inference results of each layer. The gradients calculated for multiple training data are added together and used to send the model parameter updates to the neural network model.

[0087] Typically, the training process shown in Figure 4 is performed offline. This means that the neural network model is trained in the training system using a large amount of simulated data or offline data collected over the air interface. Once the training is complete, the neural network model is deployed on the device. However, the differences between the way information is exchanged between the sending and receiving neural network models during training and in actual applications can affect the performance of wireless communication systems that utilize neural network models.

[0088] This application provides a neural network model training method that is beneficial for improving the adaptability of the neural network model to actual application scenarios, thereby improving the performance of wireless communication systems that use the neural network model. The following is a further introduction to the neural network model training method and communication device provided by this application in conjunction with the accompanying drawings:

[0089] Please refer to Figure 5, which is a flow chart of a neural network model training method provided by an embodiment of the present application. As shown in Figure 5, the neural network model training method includes the following S501 to S505. The execution subject of the method shown in Figure 5 can be a sending end device and a receiving end device, or the execution subject of the method shown in Figure 5 can be a module in the sending end device and a module in the receiving end device, or the execution subject of the method shown in Figure 5 can be a chip of the sending end device and a chip of the receiving end device. Figure 5 is illustrated by taking the sending end device and the receiving end device as the execution subject of the method as an example. It should be noted that the sending end device (or receiving end device) mentioned in this application can be the network device shown in Figure 1a, or it can be the terminal device shown in Figure 1a, and this application does not make specific restrictions. Among them:

[0090] S501. The transmitting device sends a first forward reasoning result to the receiving device. The first forward reasoning result is obtained by the first neural network model performing forward reasoning on the training data included in the first training process based on the first model parameters. The first training process is any one of the at least one training process corresponding to the first neural network model. The first neural network model is a model deployed on the transmitting device.

[0091] It should be understood that in this application, the neural network model deployed in the transmitting device is referred to as the first neural network model. The first neural network model corresponds to at least one training process, and the transmitting device can update the model parameters of the first neural network model based on the training data included in each training process (or understood as corresponding). In the specific implementation of S501 to S505, only the updating of the model parameters of the first neural network model based on the training data included in a training process in the at least one training process (referred to as the first training process in this application) is used as an example for explanation. The first model parameters are the initial model parameters in the process of updating the model parameters based on the training data of the first training process. Among them, the training data can be understood as data used for training the neural network model; for example, the training data can be training data with a label (or called an annotation), which is used to update the model parameters of the neural network model (including the first neural network model or the second neural network model mentioned in this application). When the training data has a label, the label is related to the task of the neural network model, and the label can be understood as the result obtained after the neural network model processes the training data. For example, if the neural network model is used for data recovery (i.e., the task of the neural network model is data recovery), the label of the training data is the true value of the training data (or the data to be recovered); if the neural network model is used for data classification (i.e., the task of the neural network model is data classification), the label of the training data is the true category of the training data (or the data to be classified). This application does not specifically limit the specific form of the training data.

[0092] That is, the transmitting device obtains the training data included in the first training process and performs forward reasoning on the training data using the first neural network model to obtain a first forward reasoning result and a first intermediate result. The first intermediate result is generated when forward reasoning is performed on the training data included in the first training process. That is, the first intermediate result can be understood as a general term for the information generated by the first neural network model during the process of generating the first forward reasoning result. Furthermore, the transmitting device sends the first forward reasoning result to the receiving device.

[0093] It should be noted that the forward reasoning result mentioned in this application is the result to be sent to the receiving device after calculation (or understood as reasoning) by the first neural network model; or it can be understood as the calculation result obtained by the last layer of neural network in the first neural network model in the process of calculating the forward reasoning result. The forward reasoning result mentioned in this application can also be called a calculation result, or an inference result, etc. in other schemes, and this application does not specifically limit its name. The intermediate result mentioned in this application is the information generated in the process of the first neural network model calculating the forward reasoning result based on the training data, or it can be called an intermediate state; or it can also be understood as the calculation result obtained by the other layers of neural networks in the first neural network model except the last layer of neural network in the process of calculating the forward reasoning result.

[0094] In a possible implementation, after the sending end device generates the first intermediate result, it caches the first intermediate result; that is, it stores the first intermediate result in the parameter cache space corresponding to the first training process.

[0095] S502: The transmitting device receives a first feedback result for the first forward reasoning result from the receiving device.

[0096] That is, after the receiving device receives the first forward reasoning result from the sending device, it calculates the first forward reasoning result through the second neural network model deployed on the receiving device (for example, including at least one of forward reasoning, loss function calculation, or reverse gradient calculation), obtains a first feedback result, and feeds back the first feedback result to the sending device. The first feedback result can be the loss result of the reasoning calculation of the neural network model of the receiving device (i.e., the second neural network model mentioned in this application), or the gradient obtained by the reverse gradient calculation of the second neural network model, which is not specifically limited in this application.

[0097] In one possible embodiment, the first training process includes at least one sub-process, each sub-process includes at least one training data. In this case, the first forward reasoning result includes the forward reasoning result of each training data in the training data included in the first training process, the first intermediate result includes the intermediate result corresponding to the forward reasoning result of each training data, and the first feedback result includes the feedback result corresponding to the forward reasoning result of each training data. The forward reasoning result of each training data is obtained based on the first model parameter.

[0098] Exemplarily, the first training process includes subprocess 1 and subprocess 2. Subprocess 1 includes training data 1 and training data 2, and subprocess 2 includes training data 3 and training data 4. The intermediate results, forward reasoning results, and feedback results corresponding to each training data are shown in Table 1.

[0099] Table 1

[0100] In this case, the first forward reasoning result includes forward reasoning result 1 to forward reasoning result 4, the first intermediate result includes intermediate result 1 to intermediate result 4, and the first feedback result includes feedback result 1 to feedback result 4.

[0101] In one possible implementation, at least one of the following is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or channel conditions: the amount of training data included in the first training process; or the number of sub-processes included in the first training process; or the amount of training data included in the sub-processes. The capabilities of the transmitting device include one or more of the following: the computing speed of the transmitting device, the size of the transmitting device's cache space, the forward reasoning time for the reference model of the transmitting device, and / or the reverse gradient calculation time. The capabilities of the receiving device include one or more of the following: the computing speed of the receiving device, the size of the receiving device's cache space, the forward reasoning time for the reference model of the receiving device, and / or the reverse gradient calculation time. Channel conditions include one or more of the following: wireless resources, channel quality, communication rate, and / or communication delay for communication between the transmitting and receiving devices. It should be noted that the reference model mentioned in this application is a model used to measure the computing time (e.g., forward reasoning time and / or reverse gradient calculation time) of the transmitting end. This application does not specifically limit the name of the reference model, nor does it specifically limit the specific model structure of the reference model.

[0102] That is, after a certain device (e.g., a transmitting device, a receiving device, or a network device) obtains at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or the channel conditions, it determines the amount of training data included in the first training process based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or the channel conditions; or determines the number of sub-processes included in the first training process; or determines the amount of training data included in each sub-process in the first training process. For ease of understanding, the following description uses the example of a network device determining the amount of training data included in the first training process based on the capabilities of the transmitting device and the capabilities of the receiving device as an example, which should not be regarded as a specific limitation of the present application.

[0103] In Example 1, the transmitting device receives capabilities from the receiving device and determines the amount of training data included in the first training process based on the capabilities of the transmitting device and the receiving device. Further, the transmitting device sends indication information indicating the amount of training data included in the first training process to the receiving device.

[0104] For example, the transmitting device is network device N and the receiving device is terminal device B. In this case, network device N obtains the capabilities of terminal device B and determines the amount of training data included in the first training process based on the capabilities of network device N and terminal device B. Furthermore, network device N sends indication information to terminal device B, where the indication information is used to indicate the amount of training data included in the first training process.

[0105] In Example 2, the receiving device receives the capabilities of the sending device and determines the amount of training data included in the first training process based on the capabilities of the sending device and the receiving device. Further, the receiving device sends indication information indicating the amount of training data included in the first training process to the sending device.

[0106] For example, the transmitting device is terminal device A and the receiving device is network device N. In this case, network device N obtains the capabilities of terminal device A and determines the amount of training data included in the first training process based on the capabilities of terminal device A and network device N. Furthermore, network device N sends indication information to terminal device A, where the indication information is used to indicate the amount of training data included in the first training process.

[0107] In Example 3, a device having a communication connection with a transmitting device and a receiving device is configured to obtain capabilities of the transmitting device and the receiving device, and determine the amount of training data included in the first training process based on the capabilities of the transmitting device and the receiving device. Furthermore, indication information indicating the amount of training data included in the first training process is sent to the transmitting device and the receiving device.

[0108] For example, the transmitting end device is terminal device A, and the receiving end device is terminal device B. In this case, network device N providing services for terminal device A and terminal device B obtains the capabilities of terminal device A and terminal device B, and determines the amount of training data included in the first training process based on the capabilities of terminal device A and terminal device B. Furthermore, base station network device N sends indication information to terminal device A and terminal device B, respectively, to indicate the amount of training data included in the first training process.

[0109] S503: The sending end device obtains a first intermediate result, where the first intermediate result is generated when forward reasoning is performed on the training data included in the first training process.

[0110] It is understood that if the sending end device caches the first intermediate result after generating it in S501, the sending end device can obtain the first intermediate result from the parameter cache space corresponding to the first training process. If the sending end device does not cache the first intermediate result after generating it in S501, the sending end device can perform forward reasoning again based on the training data included in the first training process to obtain the first intermediate result.

[0111] In a possible implementation, in order to improve the efficiency of the sending end device in obtaining the intermediate results corresponding to each forward reasoning result. The intermediate results of different training processes can be distinguished according to the parameter cache space; that is, different training processes corresponding to the first neural network model correspond to different parameter cache spaces, that is, the intermediate results of the training data belonging to the same training process are stored in the same parameter cache space, and the intermediate results of the training data belonging to different training processes are stored in different parameter cache spaces. Alternatively, the intermediate results of different sub-processes of the same training process can also be distinguished according to the parameter cache space; that is, different sub-processes in the first training process correspond to different parameter cache spaces, that is, the intermediate results of the training data belonging to the same sub-process are stored in the same parameter cache space, and the intermediate results of the training data belonging to different sub-processes are stored in different parameter cache spaces.

[0112] S504. The sending device calculates the gradient of the first training process according to the first feedback result, the first forward reasoning result, and the first intermediate result.

[0113] It is understandable that the first forward reasoning result generated by the transmitting device does not carry information about the real channel; however, the first forward reasoning result received by the receiving device is transmitted through the real channel, that is, the first forward reasoning result received by the receiving device will be affected by the transmission environment of the real channel during the transmission process (that is, it carries information about the real channel). Therefore, the first feedback result obtained by the receiving device based on such a first forward reasoning result can also be understood as carrying information about the real channel. Furthermore, after the receiving device sends the first feedback result to the transmitting device, the transmitting device calculates the gradient of the first training process based on the first feedback result (containing information about the real channel), the first forward reasoning result (not containing information about the real channel), and the first intermediate result.

[0114] In one possible implementation, if the sending device caches the first intermediate result after generating the first intermediate result in S501, then after the sending device executes S504, in order to save the cache resources of the sending device, the sending device may clear the first intermediate result from the parameter cache space corresponding to the first training process.

[0115] S505: The transmitting device updates the first model parameters to second model parameters according to the gradient corresponding to the first training process.

[0116] After obtaining the gradient corresponding to the first training process, the transmitting device updates the model parameters of the first neural network model to second model parameters. It is understood that the first model parameters are the parameters of the first neural network model before the transmitting device performs the first training process update, and the second model parameters are the parameters of the first neural network model after the transmitting device performs the first training process update.

[0117] It should be understood that the first neural network model corresponds to at least one training process. The sending end device obtains the gradients corresponding to each training process according to the method of obtaining the gradients corresponding to the first training process (i.e., the aforementioned S501 to S504), and updates the model parameters of the first neural network model according to the gradients of all training processes corresponding to the first neural network model.

[0118] The following describes a specific method in which the transmitting device updates the model parameters of the first neural network model according to the gradients of all training processes corresponding to the first neural network model.

[0119] Method 1: Iterative update of each training process.

[0120] As shown in FIG6 , the first neural network model corresponds to N training processes, where N is a positive integer greater than 1. The first training process described in FIG5 is one of the N training processes, and the third training process is any one of the N training processes other than the first training process. FIG6 illustrates the third training process as the next training process after the first training process, which should not be construed as a specific limitation of the present application.

[0121] That is to say, the sending device is based on W t The first neural network model corresponding to the first model parameter performs forward inference on the training data of the first training process to obtain a first forward inference result, and sends the first forward inference result to the receiving device. Further, the sending device receives a first feedback result from the receiving device for the first forward inference result, and calculates the gradient of the first training process based on the first feedback result. Further, the sending device updates W according to the gradient corresponding to the first training process. t W t+1 (i.e., the second model parameter). Then, the transmitting device is based on W t+1The first neural network model corresponding to the second model parameter performs forward inference on the training data of the third training process to obtain a third forward inference result, and sends the third forward inference result to the receiving device. Further, the sending device receives a third feedback result from the receiving device for the third forward inference result, and calculates the gradient of the third training process based on the third feedback result. Further, the sending device updates W according to the gradient corresponding to the third training process. t+1 (i.e. the second model parameter) is W t+2 .

[0122] Method 2: Mixed updates of various training processes (including parallel updates and iterative updates).

[0123] As shown in FIG7 , the first neural network model corresponds to N training processes, where N is a positive integer greater than 1. The first training process described in FIG5 is one of the N training processes, and the second training process is any one of the N training processes other than the first training process. FIG7 exemplifies the second approach by taking the first and second training processes as training processes of a neural network model based on the same initial model parameters as an example.

[0124] That is to say, the sending device is based on W t (i.e., the first model parameter) performs forward reasoning on the training data of the first training process to obtain a first forward reasoning result, and sends the first forward reasoning result to the receiving device. Then, the sending device performs forward reasoning on the training data of the first training process to obtain a first forward reasoning result. t (i.e., the first model parameters) corresponding to the first neural network model, forward inference is performed on the training data of the second training process to obtain a second forward inference result, and the second forward inference result is sent to the receiving device. Further, the sending device receives a first feedback result from the receiving device for the first forward inference result, and calculates the gradient of the first training process based on the first feedback result, the first forward inference result, and the first intermediate result (the intermediate result generated when forward inference is performed on the training data included in the first training process). Further, the sending device updates W according to the gradient corresponding to the first training process. t W t+1 (i.e., the second model parameters). The transmitting device receives a second feedback result from the receiving device for the second forward reasoning result, and calculates the gradient of the second training process based on the second feedback result, the second forward reasoning result, and the second intermediate result (the intermediate result generated when forward reasoning is performed on the training data included in the second training process). Further, the transmitting device updates W according to the gradient corresponding to the first training process. t W t+1' (ie, the third model parameter). Further, the transmitting end device is based on the second model parameter (ie, Wt+1 ) and the third model parameter (i.e. W t+1' ), the model parameters of the first neural network model are obtained (or understood as updated) as the fourth model parameters.

[0125] It should be noted that the second model parameter (ie W t+1 ) and the third model parameter (i.e. W t+1' ), the fourth model parameter is obtained, including either of the following two interpretations:

[0126] ①, the sending end device integrates the second model parameters (i.e. W t+1 ) and the third model parameter (i.e. W t+1' ), and obtain the fourth model parameter. For example, the fourth model parameter is a weighted average of the second model parameter and the third model parameter.

[0127] ②. The transmitting end device iterates the model parameters of the first neural network model multiple times based on the iteration of the second model parameters and the training data of multiple training processes to obtain the model parameters W after multiple iterations. And the transmitting end device iterates the model parameters of the first neural network model multiple times based on the training data of multiple training processes to obtain the model parameters W' after multiple iterations. Finally, the model parameters W and the model parameters W' are fused to obtain the fourth model parameters. For example, after obtaining the second model parameters (i.e., W t+1 ) and the third model parameter (i.e. W t+1' ) then updates the second model parameter (i.e. W) of the first neural network model based on the training data of the training process P1 in the N training processes (i.e. at least one training process corresponding to the first neural network model). t+1 ) is the model parameter W t+2 Based on the training data of the training process P2 in the N training processes, the third model parameter of the first neural network model (ie, W t+1' ) is the model parameter W t+2' Furthermore, the transmitting device can fuse the model parameters W t+2 and model parameters W t+2' , and obtain the fourth model parameter.

[0128] It can be seen that by implementing the neural network model training method described in Figure 5 of the present application, after the first neural network model is deployed on the sending device, the sending device performs forward reasoning on the training data through the first neural network model, and sends the forward reasoning result to the receiving device through the channel of the communication system. The forward reasoning result will be affected by the transmission environment of the channel (that is, it carries information about the real channel). Therefore, the feedback result obtained based on such forward reasoning result can also be understood as carrying information about the real channel. When the sending device subsequently updates the model parameters of the first neural network model based on the feedback result, it is beneficial to improve the adaptability of the neural network model to the communication system, thereby helping to optimize the performance of the wireless communication system that applies the neural network model.

[0129] It should be noted that in the neural network model training method shown in Figure 5, only the update process of the neural network model deployed on the sending end device (i.e., the aforementioned first neural network model) is described, and the update process of the neural network model deployed on the receiving end device (i.e., the second neural network model mentioned in this application) is not limited. That is to say, in the update process of the first neural network model, the receiving end device can update the model parameters of the second neural network model based on the forward reasoning results from the sending end device (including the aforementioned first forward reasoning results to third forward reasoning results); the receiving end device may also not update the model parameters of the second neural network model; this application does not make specific restrictions on this. Moreover, in the case where the receiving end device updates the model parameters of the second neural network model based on the forward reasoning results from the sending end device, the application does not make specific restrictions on what update method the receiving end device uses to update the model parameters of the second neural network model. For ease of understanding, this application provides two examples of receiving end devices updating the model parameters of the second neural network model.

[0130] In one possible implementation, each time the receiving device receives a forward reasoning result from the transmitting device, the model parameters of the second neural network model are updated according to the forward reasoning result, and a hybrid update (including parallel update and iterative update) is performed based on the forward reasoning results of multiple training processes. In the following examples, the model parameters of the second neural network model before the update are used as the fifth model parameters to illustrate the update process of the second neural network model.

[0131] For example, let's continue with the example corresponding to Figure 7 above. After the receiving end device receives the first forward reasoning result, it calculates the first forward reasoning result based on the second neural network model corresponding to the fifth model parameter to obtain a first feedback result, and updates the fifth model parameter of the second neural network model to the sixth model parameter based on the first forward reasoning result. After the receiving end device receives the second forward reasoning result, it calculates the second forward reasoning result based on the second neural network model corresponding to the fifth model parameter to obtain a second feedback result, and updates the fifth model parameter of the second neural network model to the seventh model parameter based on the second forward reasoning result. Furthermore, the receiving end device obtains (or understands as updates) the model parameter of the second neural network model as the eighth model parameter based on the sixth model parameter and the seventh model parameter. Specifically, regarding the understanding of the receiving end device obtaining the eighth model parameter based on the sixth model parameter and the seventh model parameter, reference can be made to the description in Figure 7 about the sending end device obtaining the eighth model parameter based on the second model parameter (i.e., W t+1 ) and the third model parameter (i.e. W t+1' ) to obtain the understanding of the fourth model parameter, which will not be repeated here.

[0132] In one possible implementation, each time the receiving device receives a forward reasoning result from the sending device, the model parameters of the second neural network model are updated according to the forward reasoning result, and the forward reasoning results of multiple training processes are iteratively updated in sequence.

[0133] For example, continuing with the example corresponding to Figure 6 above. In the first training process, after the receiving end device receives the first forward reasoning result, it calculates the first forward reasoning result based on the second neural network model corresponding to the fifth model parameter to obtain a first feedback result, and updates the fifth model parameter of the second neural network model to the sixth model parameter based on the first forward reasoning result. In the third training process, after the receiving end device receives the third forward reasoning result, it calculates the third forward reasoning result based on the second neural network model corresponding to the sixth model parameter to obtain a third feedback result, and updates the sixth model parameter of the second neural network model to the ninth model parameter based on the third forward reasoning result.

[0134] It should also be noted that a training process in an embodiment of the present application includes: a process in which the first neural network model performs forward reasoning on a set of training data (i.e., the training data of the training process), a process in which feedback results corresponding to the set of training data are received, a process in which gradients are calculated based on the forward reasoning results, intermediate results, and feedback results of the set of training data, and a process in which the model parameters of the first neural network model are updated based on the gradients. A sub-process in an embodiment of the present application includes: a process in which the first neural network model performs forward reasoning on a set of training data (i.e., the training data of the sub-process), a process in which feedback results corresponding to the set of training data are received, and a process in which gradients are calculated based on the forward reasoning results, intermediate results, and feedback results of the set of training data. This application does not specifically limit the names of training processes and sub-processes.

[0135] For example, in one application scenario, different services of the sending device correspond to different neural network models. In this case, one service can be called a training process, and the process of updating the model parameters of the neural network model in a certain service is called a sub-process (that is, the training process mentioned in this application).

[0136] For example, as shown in Figure 7, the sending device will obtain the fourth model parameter as the end of a training process, and will regard each update process of the first neural network model based on the feedback result as the end of a sub-process (i.e., the training process mentioned in this application).

[0137] Please refer to Figure 8, which shows a structural diagram of a communication device 800 of an embodiment of the present application. The communication device shown in Figure 8 can be a sending end device, or a device in a sending end device, or a device that can be used in combination with a sending end device. The communication device shown in Figure 8 may include a communication unit 801 and a processing unit 802; the communication device shown in Figure 8 can be a receiving end device, or a device in a receiving end device, or a device that can be used in combination with a receiving end device. The communication device shown in Figure 8 may include a communication unit 801 and a processing unit 802. Specifically, the processing unit 802 is used to process data, and the data may be data received by the communication unit 801, and the processed data may also be sent by the communication unit 801;

[0138] In one embodiment, the communication device 800 is a transmitting end device, a device in a transmitting end device, or a device that can be used in conjunction with the transmitting end device, wherein:

[0139] The communication unit 801 is used to send a first forward reasoning result to a receiving device, where the first forward reasoning result is obtained by the first neural network model performing forward reasoning on the training data included in the first training process based on the first model parameters. The first training process is any one of the at least one training processes corresponding to the first neural network model, and the first neural network model is a model deployed on the sending device; the communication unit 801 is also used to receive a first feedback result from the receiving device for the first forward reasoning result; the processing unit 802 is used to obtain a first intermediate result, where the first intermediate result is generated when forward reasoning is performed on the training data included in the first training process; the processing unit 802 is used to calculate the gradient of the first training process based on the first feedback result, the first forward reasoning result and the first intermediate result; the processing unit 802 is also used to update the first model parameter to the second model parameter based on the gradient corresponding to the first training process.

[0140] In a possible implementation, the processing unit 802 is further configured to obtain a first intermediate result from a parameter cache space corresponding to the first training process.

[0141] In one possible implementation, after calculating the gradient of the first training process based on the first feedback result, the first forward reasoning result, and the first intermediate result, the processing unit 802 is further used to clear the first intermediate result from the parameter cache space corresponding to the first training process.

[0142] In a possible implementation, the processing unit 802 is further configured to perform forward reasoning again based on the training data included in the first training process to obtain a first intermediate result.

[0143] In one possible implementation, the first training process includes at least one sub-process, each sub-process includes at least one training data, the first forward reasoning result includes the forward reasoning result of each training data in the training data included in the first training process, the first intermediate result includes the intermediate result corresponding to the forward reasoning result of each training data, the first feedback result includes the feedback result corresponding to the forward reasoning result of each training data, and the forward reasoning result of each training data is obtained based on the first model parameter.

[0144] In one possible implementation, at least one of the following is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or the channel conditions: the amount of training data included in the first training process; or the number of sub-processes included in the first training process; or the amount of training data included in the sub-processes.

[0145] In one possible implementation, the processing unit 802 is further used to update the first model parameters to third model parameters based on the gradient corresponding to the second training process; wherein the second training process is any one of the at least one training process other than the first training process, the gradient of the second training process is obtained by reverse gradient calculation based on the second feedback result, the second forward reasoning result and the second intermediate result, the second forward reasoning result is obtained by forward reasoning on the training data included in the second training process based on the first model parameters, the second intermediate result is generated when forward reasoning is performed on the training data included in the first training process, and the second feedback result is a feedback result for the second forward reasoning result; the processing unit 802 is also used to update the model parameters of the first neural network model to fourth model parameters based on the second model parameters and the third model parameters.

[0146] In one possible implementation, the number of training processes of at least one training process corresponding to the first neural network model is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device and / or the channel conditions.

[0147] In one possible implementation, the capabilities of the transmitting device include one or more of the computing speed of the transmitting device, the size of the cache space of the transmitting device, the forward reasoning time of the transmitting device for a reference model, and / or the reverse gradient calculation time; the capabilities of the receiving device include one or more of the computing speed of the receiving device, the size of the cache space of the receiving device, the forward reasoning time of the receiving device for a reference model, and / or the reverse gradient calculation time; the channel conditions include one or more of the wireless resources, channel quality, communication rate, and communication delay for communication between the transmitting device and the receiving device.

[0148] In a possible implementation, the communication unit 801 is configured to receive capability information from a receiving device, where the capability information indicates the capability of the receiving device.

[0149] In a possible implementation, different training processes of the first neural network model correspond to different parameter cache spaces.

[0150] In a possible implementation, different sub-processes in the first training process correspond to different parameter cache spaces.

[0151] In one embodiment, the communication device 800 is a receiving device, a device in a receiving device, or a device that can be used in conjunction with a receiving device, wherein:

[0152] The communication unit 801 is used to receive a first forward reasoning result from a sending end device, where the first forward reasoning result is obtained by the first neural network model performing forward reasoning on the training data included in the first training process based on the first model parameters, where the first training process is any one of at least one training process corresponding to the first neural network model, and the first neural network model is a model deployed on the sending end device; the communication unit 801 is also used to send a first feedback result to the sending end device based on the first forward reasoning result.

[0153] In one possible implementation, the first training process includes at least one sub-process, each sub-process includes at least one training data, the first forward reasoning result includes the forward reasoning result of each training data in the training data included in the first training process, the first intermediate result includes the intermediate result corresponding to the forward reasoning result of each training data, the first feedback result includes the feedback result corresponding to the forward reasoning result of each training data, and the forward reasoning result of each training data is obtained based on the first model parameter.

[0154] In one possible implementation, at least one of the following is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device, and / or the channel conditions: the amount of training data included in the first training process; or the number of sub-processes included in the first training process; or the amount of training data included in the sub-processes.

[0155] In one possible implementation, the communication unit 801 is further used to receive a second forward reasoning result from the sending end device, where the second forward reasoning result is obtained by forward reasoning the training data included in the second training process based on the first model parameters, and the second training process is any one of the at least one training process except the first training process; the communication unit 801 is also used to send a second feedback result to the sending end device based on the second forward reasoning result, where the second feedback result is used to update the model parameters of the first neural network model in combination with the first feedback result.

[0156] In one possible implementation, the number of training processes of at least one training process corresponding to the first neural network model is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device and / or the channel conditions.

[0157] In one possible implementation, the capabilities of the transmitting device include one or more of the computing speed of the transmitting device, the size of the cache space of the transmitting device, the forward reasoning time of the transmitting device for a reference model, and / or the reverse gradient calculation time; the capabilities of the receiving device include one or more of the computing speed of the receiving device, the size of the cache space of the receiving device, the forward reasoning time of the receiving device for a reference model, and / or the reverse gradient calculation time; the channel conditions include one or more of the wireless resources, channel quality, communication rate, and / or communication delay for communication between the transmitting device and the receiving device.

[0158] In one possible implementation, the receiving device is deployed with a second neural network model, and the processing unit 802 is further used to update the fifth model parameter of the second neural network model to a sixth model parameter based on the first forward reasoning result; update the fifth model parameter of the second neural network model to a seventh model parameter based on the second forward reasoning result; and obtain the eighth model parameter based on the sixth model parameter and the seventh model parameter.

[0159] In one possible implementation, the receiving device is deployed with a second neural network model, and the processing unit 802 is further used to update the fifth model parameter of the second neural network model to a sixth model parameter based on the first forward reasoning result.

[0160] For a more detailed description of the communication unit 801 and the processing unit 802 , reference may be made to the relevant description of the transmitting end device or the receiving end device in the method embodiment shown in FIG5 .

[0161] As shown in Figure 9, communication device 900 includes a processor 910 and an interface circuit 920. Processor 910 and interface circuit 920 are coupled to each other. It is understood that interface circuit 920 can be a transceiver or an input / output interface. Optionally, communication device 900 may also include a memory 930 for storing instructions executed by processor 910, input data required by processor 910 to execute instructions, or data generated after processor 910 executes instructions.

[0162] When the communication device 900 is used to implement the method shown in FIG. 5 , the processor 910 is used to implement the functions of the processing unit 802 , and the interface circuit 920 is used to implement the functions of the communication unit 801 .

[0163] When the above-mentioned communication device is a chip applied to a transmitting device, the terminal chip implements the functions of the transmitting device in the above-mentioned method embodiment. The transmitting device chip receives information from the receiving device, which can be understood as the information being first received by other modules in the transmitting device (such as a radio frequency module or antenna) and then sent to the transmitting device chip by these modules. The transmitting device chip sends information to the receiving device, which can be understood as the information being first sent to other modules in the transmitting device (such as a radio frequency module or antenna) and then sent to the receiving device by these modules.

[0164] When the communication device is a chip used in a receiving device, the receiving device chip implements the functions of the receiving device in the above-mentioned method embodiment. The receiving device chip receives information from the transmitting device, which can be understood as the information being first received by other modules in the receiving device (such as a radio frequency module or antenna) and then transmitted to the receiving device chip by these modules. The receiving device chip sends information to the transmitting device, which can be understood as the information being transmitted to other modules in the receiving device (such as a radio frequency module or antenna) and then transmitted to the transmitting device by these modules.

[0165] In this application, when entity A sends information to entity B, it can be done directly from A to B or indirectly through another entity. Similarly, when entity B receives information from entity A, it can be done directly from entity B or indirectly through another entity. Entities A and B herein can be RAN nodes or terminals, or modules within a RAN node or terminal. The sending and receiving of information can be information exchange between a RAN node and a terminal, for example, between a base station and a terminal; the sending and receiving of information can also be information exchange between two RAN nodes, for example, between a CU and a DU; the sending and receiving of information can also be information exchange between different modules within a device, for example, between a terminal chip and other modules in the terminal, or between a base station chip and other modules within the base station.

[0166] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0167] The method steps in the embodiments of the present application can be implemented in hardware or in software instructions that can be executed by a processor. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disk, mobile hard disk, CD-ROM or any other form of storage medium well known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. The storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC. In addition, the ASIC can be located in a base station or a terminal. The processor and storage medium can also exist in a base station or a terminal as discrete components.

[0168] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are performed in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device, or other programmable device. The computer program or instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions may be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video disk; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or nonvolatile storage medium, or may include both volatile and nonvolatile types of storage media.

[0169] In the various embodiments of the present application, unless otherwise specified or there is a logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced by each other. The technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0170] In this application, "at least one" means one or more, and "more" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In the text description of this application, the character " / " generally indicates that the previous and next associated objects are in an "or" relationship; in the formula of this application, the character " / " indicates that the previous and next associated objects are in a "division" relationship. "Including at least one of A, B and C" can mean: including A; including B; including C; including A and B; including A and C; including B and C; including A, B and C.

[0171] It is understood that the various numbers used in the embodiments of this application are merely for ease of description and are not intended to limit the scope of the embodiments of this application. The order of the sequence numbers of the above-mentioned processes does not necessarily imply a specific order of execution; the order of execution of the processes should be determined by their functions and inherent logic.

Claims

1. A neural network model training method, characterized in that: The method comprises: Sending a first forward reasoning result to a receiving device, where the first forward reasoning result is obtained by forward reasoning the first neural network model on the training data included in the first training process based on the first model parameter, where the first training process is any one of at least one training process corresponding to the first neural network model, and the first neural network model is a model deployed on the sending device; Receiving a first feedback result from the receiving end device for the first forward reasoning result; Obtaining a first intermediate result, where the first intermediate result is generated when forward reasoning is performed on the training data included in the first training process; Calculating a gradient of the first training process according to the first feedback result, the first forward reasoning result, and the first intermediate result; According to the gradient corresponding to the first training process, the first model parameter is updated to the second model parameter.

2. The method according to claim 1, characterized in that: The obtaining of the first intermediate result comprises: Obtain a first intermediate result from the parameter cache space corresponding to the first training process.

3. The method according to claim 2, characterized in that: After calculating the gradient of the first training process according to the first feedback result, the first forward reasoning result and the first intermediate result, the method further includes: The first intermediate result is cleared from the parameter cache space corresponding to the first training process.

4. The method according to claim 1, characterized in that: The obtaining of the first intermediate result comprises: Forward reasoning is performed again based on the training data included in the first training process to obtain a first intermediate result.

5. The method according to any one of claims 1 to 4, characterized in that: The first training process includes at least one sub-process, each sub-process includes at least one training data, the first forward reasoning result includes the forward reasoning result of each training data in the training data included in the first training process, the first intermediate result includes the intermediate result corresponding to the forward reasoning result of each training data, the first feedback result includes the feedback result corresponding to the forward reasoning result of each training data, and the forward reasoning result of each training data is obtained based on the first model parameter.

6. The method according to claim 5, characterized in that: At least one of the following is determined based on at least one of the capability of the transmitting end device, the capability of the receiving end device and / or a channel condition: the amount of training data included in the first training process; or, the number of sub-processes included in the first training process; Or, the number of training data included in the sub-process.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: According to the gradient corresponding to the second training process, the first model parameter is updated to a third model parameter; wherein the second training process is any one of the at least one training process except the first training process, the gradient of the second training process is obtained by performing reverse gradient calculation based on the second feedback result, the second forward reasoning result, and the second intermediate result, the second forward reasoning result is obtained by performing forward reasoning on the training data included in the second training process based on the first model parameter, the second intermediate result is generated when performing forward reasoning on the training data included in the first training process, and the second feedback result is a feedback result for the second forward reasoning result; According to the second model parameters and the third model parameters, the model parameters of the first neural network model are updated to fourth model parameters.

8. The method according to claim 7, characterized in that: The number of training processes of the at least one training process corresponding to the first neural network model is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device and / or channel conditions.

9. The method according to claim 6 or 8, characterized in that: The capabilities of the transmitting device include one or more of the computing speed of the transmitting device, the size of the cache space of the transmitting device, the forward reasoning time of the transmitting device for a reference model, and / or the reverse gradient calculation time; the capabilities of the receiving device include one or more of the computing speed of the receiving device, the size of the cache space of the receiving device, the forward reasoning time of the receiving device for a reference model, and / or the reverse gradient calculation time; the channel conditions include one or more of the wireless resources, channel quality, communication rate, and / or communication delay for communication between the transmitting device and the receiving device.

10. The method according to any one of claims 6 to 9, characterized in that: The method further comprises: Capability information is received from the receiving end device, where the capability information is used to indicate the capability of the receiving end device.

11. The method according to any one of claims 1 to 10, characterized in that: Different training processes of the first neural network model correspond to different parameter cache spaces.

12. The method according to any one of claims 5 to 10, characterized in that: Different sub-processes in the first training process correspond to different parameter cache spaces.

13. A neural network model training method, characterized in that: The method comprises: Receiving a first forward reasoning result from a transmitting end device, where the first forward reasoning result is obtained by forward reasoning a first neural network model on training data included in a first training process based on a first model parameter, where the first training process is any one of at least one training process corresponding to the first neural network model, and the first neural network model is a model deployed on the transmitting end device; Based on the first forward reasoning result, the first feedback result is sent to the sending end device.

14. The method according to claim 13, characterized in that: The first training process includes at least one sub-process, each sub-process includes at least one training data, the first forward reasoning result includes the forward reasoning result of each training data in the training data included in the first training process, the first intermediate result includes the intermediate result corresponding to the forward reasoning result of each training data, the first feedback result includes the feedback result corresponding to the forward reasoning result of each training data, and the forward reasoning result of each training data is obtained based on the first model parameter.

15. The method according to claim 13 or 14, characterized in that: At least one of the following is determined based on at least one of the capability of the transmitting end device, the capability of the receiving end device and / or a channel condition: the amount of training data included in the first training process; or, the number of sub-processes included in the first training process; Or, the number of training data included in the sub-process.

16. The method according to any one of claims 13 to 15, characterized in that: The method further comprises: receiving a second forward reasoning result from the transmitting end device, where the second forward reasoning result is obtained by forward reasoning the training data included in a second training process based on the first model parameter, where the second training process is any one of the at least one training process except the first training process; Based on the second forward reasoning result, the second feedback result is sent to the sending end device, and the second feedback result is used to update the model parameters of the first neural network model in combination with the first feedback result.

17. The method according to claim 16, characterized in that: The number of training processes of the at least one training process corresponding to the first neural network model is determined based on at least one of the capabilities of the transmitting device, the capabilities of the receiving device and / or channel conditions.

18. The method according to claim 15 or 17, characterized in that: The capabilities of the transmitting device include one or more of the computing speed of the transmitting device, the size of the cache space of the transmitting device, the forward reasoning time of the transmitting device for a reference model, and / or the reverse gradient calculation time; the capabilities of the receiving device include one or more of the computing speed of the receiving device, the size of the cache space of the receiving device, the forward reasoning time of the receiving device for a reference model, and / or the reverse gradient calculation time; the channel conditions include one or more of the wireless resources, channel quality, communication rate, and / or communication delay for communication between the transmitting device and the receiving device.

19. The method according to claim 16, characterized in that: The receiving end device is deployed with a second neural network model, and the method further includes: Based on the first forward reasoning result, a fifth model parameter of the second neural network model is updated to a sixth model parameter; Update the fifth model parameter of the second neural network model to a seventh model parameter based on the second forward reasoning result; Based on the sixth model parameter and the seventh model parameter, an eighth model parameter is obtained.

20. The method according to claim 13, characterized in that: The receiving end device is deployed with a second neural network model, and the method further includes: Based on the first forward reasoning result, the fifth model parameter of the second neural network model is updated to the sixth model parameter.

21. A communication device, characterized in that: The method comprises a module for executing the method according to any one of claims 1 to 12, or comprises a module for executing the method according to any one of claims 13 to 20.

22. A communication device, characterized in that: It includes a processor and an interface circuit, wherein the interface circuit is used to receive signals from other communication devices outside the communication device and transmit them to the processor or send signals from the processor to other communication devices outside the communication device, and the processor is used to implement the method as described in any one of claims 1 to 12 through a logic circuit or by executing code instructions, or the processor is used to implement the method as described in any one of claims 13 to 20 through a logic circuit or by executing code instructions.

23. The method according to claim 22, characterized in that: The communication device is a chip or a chip system.

24. A computer-readable storage medium, characterized in that: The storage medium stores a computer program or an instruction. When the computer program or the instruction is executed by the communication device, the method according to any one of claims 1 to 12 is implemented, or the method according to any one of claims 13 to 20 is implemented.

25. A computer program product, characterized in that The computer program product includes a computer program or instructions, and when the computer program or instructions are executed by a communication device, the communication device implements the method according to any one of claims 1 to 12, or implements the method according to any one of claims 13 to 20.

Images