Data processing method and communication device

Abstract

Provided are a data processing method and a communication device, the method comprising: a second device sending a first message to a first device, the first message comprising one or more of communication condition information of the second device, computing resource information of the second device, or model training information of the second device; the second device receives a second message from the first device, the second message including indication information indicating a quantized resolution, the quantized resolution relating to one or more of communication condition information, computing resource information, or model training information; further, the second device quantizes the first artificial intelligence AI cooperation data based on the quantization resolution to obtain second AI cooperation data; and sending the second AI cooperation data to the first device. Through the method, the second device can quantize the AI cooperation data through the dynamically changing quantization resolution, and the training efficiency of federal learning can be improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a data processing method and a communication device. Background Art

[0002] Federated learning is a distributed machine learning paradigm that enables multiple devices to conduct joint training and build shared machine learning models without sharing data resources. Specifically, in the process of federated learning, the distributed nodes participating in federated learning can conduct joint training by sharing artificial intelligence (AI) assisted data.

[0003] In order to reduce the communication overhead generated by each distributed node uploading and downloading AI-assisted data during the federated learning process, the AI-assisted data can usually be compressed in a quantized manner. In other words, the AI-assisted data is represented by multiple bits, and the number of these bits can determine the quantization level (or quantization resolution), thereby affecting the performance of the gradient quantization algorithm. Among them, the fewer bits used by the quantization algorithm, the lower the quantization resolution, the greater the quantization error introduced during the AI-assisted data upload process, and the longer the model convergence time; the more bits used by the quantization algorithm, the higher the quantization resolution, and the longer the communication time for transmitting AI-assisted data.

[0004] In the process of federated learning, what kind of quantization resolution AI distributed nodes use to assist in data quantization needs further study. Summary of the invention

[0005] The embodiments of the present application provide a data processing method and a communication device, and the second device can quantize the AI ​​collaborative data through dynamically changing quantization resolution, which is conducive to improving the training efficiency of collaborative learning.

[0006] In a first aspect, the present application provides a data processing method. Taking the execution of the method by a second device as an example, the method includes: the second device sends a first message to the first device, and the first message includes one or more of communication condition information of the second device, computing resource information of the second device, or model training information of the second device; the second device receives a second message from the first device, and the second message includes indication information, and the indication information indicates a quantization resolution, and the quantization resolution is related to one or more of the communication condition information, computing resource information, or model training information; further, the second device quantizes the first artificial intelligence AI collaboration data based on the quantization resolution to obtain second AI collaboration data; and sends the second AI collaboration data to the first device.

[0007] Based on the method described in the first aspect, the quantization resolution of the second device is dynamically adjusted in combination with the communication conditions, computing resources or model training information of the current second device, which is beneficial to improving the compatibility of the quantization resolution with the current second device. By quantizing and transmitting AI collaborative data at a quantization resolution adapted to the current second device, the training efficiency of collaborative learning is improved.

[0008] In one possible implementation, the communication condition information includes one or more of communication bandwidth information, transmission delay, reference signal received power RSRP, reference signal received quality RSRQ, signal to interference plus noise ratio SINR, bit error rate or throughput; the computing resource information includes one or more of the size of computing resources, the size of computing power or computing delay; the model training information includes one or more of the test set loss value, the training set loss value, the gradient norm size, the gradient size, and the model size.

[0009] In a possible implementation, the second device receives a third message from the first device, the third message being used to request assistance in updating the quantization resolution; the aforementioned first message is a response message corresponding to the third message. By implementing this possible implementation, it is helpful to improve the flexibility of triggering the second device to send the first message.

[0010] In one possible implementation, the third message includes one or more of the following information: a measurement identifier, an identifier of the second device, an identifier of the first device, an identifier of the model, indication information indicating the reason for updating the quantization resolution, an identifier of a training round corresponding to the model, an identifier of information requested for feedback, or configuration information for transmitting the first message; wherein the information requested for feedback includes one or more of communication condition information, computing resource information, or model training information; and the first message is generated based on the identifier of the information requested for feedback.

[0011] In a possible implementation, the indication information indicating the reason for updating the quantization resolution includes the time difference between the first delay and the average delay, or the ratio between the first delay and the average delay; wherein the first delay is the delay of the first device waiting for the second device to feedback the AI ​​collaboration data, and the average delay is the average value of the delay of the first device waiting for multiple distributed nodes to feedback the AI ​​collaboration data, and the multiple distributed nodes include the second device. By implementing this possible implementation, it is beneficial to shorten the delay of the second device to feedback the AT assistance data, thereby facilitating the improvement of the training efficiency of collaborative learning.

[0012] In a possible implementation, the configuration information for transmitting the first message includes one or more of a transmission period for transmitting the first message, a trigger condition for transmitting the first message, or a number of times of transmitting the first message. By implementing this possible implementation, the second device can automatically send the first message to the first device according to the configuration information to assist in updating the quantization resolution, thereby helping to save communication transmission resources.

[0013] In a possible implementation manner, the first message further includes one or more of the following information: a measurement identifier, a second device identifier, a first device identifier, an identifier of a model, and an identifier of a training wheel corresponding to the model.

[0014] In a second aspect, the present application provides a data processing method. Taking the execution of the method by a first device as an example, the method includes: the first device receives a first message from a second device, the first message including one or more of communication condition information of the second device, computing resource information of the second device, or model training information of the second device; further, based on one or more of the communication condition information, the computing resource information, or the model training information, a quantization resolution is determined; and a second message is sent to the second device, the second message including indication information, the indication information indicating the quantization resolution, and the quantization resolution is related to one or more of the communication condition information, the computing resource information, or the model training information; the first device receives second artificial intelligence AI collaboration data from the second device, and the second AI collaboration data is obtained by quantizing the first AI collaboration data based on the quantization resolution.

[0015] For the beneficial effects obtained by the method described in the second aspect, reference may be made to the method and beneficial effects described in the first aspect.

[0016] In one possible implementation, the communication condition information includes one or more of communication bandwidth information, transmission delay, reference signal received power RSRP, reference signal received quality RSRQ, signal to interference plus noise ratio SINR, bit error rate or throughput; the computing resource information includes one or more of the size of computing resources, the size of computing power or computing delay; the model training information includes one or more of the test set loss value, the training set loss value, the gradient norm size, the gradient size, and the model size.

[0017] In a possible implementation, the first device sends a third message to the second device, where the third message is used to request assistance in updating the quantization resolution; and the aforementioned first message is a response message corresponding to the third message.

[0018] In a possible implementation, the third message includes one or more of the following information: a measurement identifier, an identifier of the second device, an identifier of the first device, a model identifier, indication information indicating the reason for updating the quantization resolution, an identifier of a training round corresponding to the model, an identifier of information requested for feedback, and / or configuration information for transmitting the first message; wherein the information requested for feedback includes one or more of communication condition information, computing resource information, or model training information; and the first message is generated based on the identifier of the information requested for feedback.

[0019] In one possible implementation, the indication information indicating the reason for updating the quantization resolution includes the time difference between the first delay and the average delay, or the ratio between the first delay and the average delay; wherein the first delay is the delay of the first device waiting for the second device to feedback the AI ​​collaboration data, and the average delay is the average value of the delay of the first device waiting for multiple distributed nodes to feedback the AI ​​collaboration data, and the multiple distributed nodes include the second device.

[0020] In a possible implementation manner, the configuration information for transmitting the first message includes one or more of a transmission period for transmitting the first message, a triggering condition for transmitting the first message, or a number of times of transmitting the first message.

[0021] In a possible implementation manner, the first message further includes one or more of the following information: a measurement identifier, a second device identifier, a first device identifier, an identifier of a model, and an identifier of a training wheel corresponding to the model.

[0022] In a third aspect, the present application provides a communication device, which may be a second device, or a device in the second device, or a device that can be used in combination with the second device. Among them, 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 unit or module 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.

[0023] In a fourth aspect, the present application provides a communication device, which may be a first device, or a device in the first device, or a device that can be used in combination with the first 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. 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 method and beneficial effects described in the second aspect above.

[0024] In a fifth aspect, the present application provides a communication device, comprising 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 described in the first aspect through a logic circuit or by executing code instructions, or the processor is used to implement the method described in the second aspect through a logic circuit or by executing code instructions.

[0025] 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 is implemented, or the method described in the second aspect is implemented.

[0026] 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.

[0027] In an eighth aspect, the present application provides a communication system, comprising a communication device for executing the method described in the first aspect above, and a communication device for executing the method executed by the network device described in the second aspect above. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic diagram of a communication system provided in an embodiment of the present application;

[0029] Figure 2 This is a schematic diagram of an application framework of AI in an NR system provided in an embodiment of the present application;

[0030] Figure 3 It is a flowchart of a federated learning provided in an embodiment of the present application;

[0031] Figure 4aIt is a flowchart of a data processing method provided in an embodiment of the present application;

[0032] Figure 4b It is a flowchart of a collaborative learning provided by an embodiment of the present application;

[0033] Figure 4c is another flowchart of collaborative learning provided by an embodiment of the present application;

[0034] Figure 5 is a structural diagram of a communication device provided in an embodiment of the present application;

[0035] Figure 6 It is a structural diagram of another communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0036] In order to facilitate a specific understanding of the embodiments of the present application, the system architecture involved in the embodiments of the present application is first introduced below.

[0037] Figure 1 FIG. 1 is a schematic diagram of the architecture of a communication system 1000 used in an embodiment of the present application. Figure 1 As shown, 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 Figure 1 110a and 110b in the figure, collectively referred to as 110), and may also include at least one terminal (such as Figure 1 RAN 100 may also include other RAN nodes, such as wireless relay equipment and / or wireless backhaul equipment ( Figure 1 (not shown in the figure). The terminal 120 is connected to the RAN node 110 by wireless means, and the RAN node 110 is connected to the core network 200 by wireless or wired means. The core network device in the core network 200 and the RAN node 110 in the RAN 100 can be independent and different physical devices, or the same physical device that integrates the logical functions of the core network device and the logical functions of the RAN node. Terminals and terminals and RAN nodes and RAN nodes can be connected to each other by wire or wireless means. It should be noted that the RAN node 110 may also be referred to as the network device 110 in the following text.

[0038] RAN100 may be an evolved universal terrestrial radio access (E-UTRA) system, a new radio (NR) system, and a future radio access system defined in the 3rd generation partnership project (3GPP). RAN100 may also include two or more of the above-mentioned different radio access systems. RAN100 may also be an open RAN (O-RAN).

[0039] RAN nodes, also known as radio access network equipment, RAN entities or access nodes, are used to help terminals access the communication system wirelessly. 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 NodeB 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 a Figure 1 110a), or a micro base station or an indoor station (such as Figure 1 110b) in the figure, it can also be a relay node or a donor node.

[0040] In another application scenario, the cooperation of multiple RAN nodes can be used to help the terminal achieve wireless access, and different RAN nodes respectively implement part of the functions of the base station. For example, the RAN node can be a centralized unit (CU), a distributed unit (DU) or a radio unit (RU). The CU here completes the functions of the radio resource control protocol and the packet data convergence protocol (PDCP) of the base station, and can also complete the function of the service data adaptation protocol (SDAP); the DU completes the functions of the radio link control layer and the medium access control (MAC) layer of the base station, and can also complete the functions of part of the physical layer or all of the physical layer. For the specific description of the above-mentioned various protocol layers, please refer to the relevant technical specifications of 3GPP. RU can be used to implement the transceiver function of radio frequency signals. CU and DU can be two independent RAN nodes, or they can be integrated in the same RAN node, such as integrated in a baseband unit (BBU). RU can be included in a radio frequency device, 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.

[0041] In different systems, RAN nodes may have different names. For example, in an O-RAN system, CU may be called an open CU (open CU, O-CU), DU may be called an open DU (open DU, O-DU), and RU may be called an open RU (open RU, O-RU). The RAN node in the embodiment 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 a 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 the convenience of description, the following description takes a base station as an example of a RAN node.

[0042] A terminal is a device with wireless transceiver function, which can send signals to a base station or receive signals from a base station. A terminal can also be called a terminal device, user equipment (UE), a mobile station, a mobile terminal, etc. The terminal can be widely used in various scenarios, for example, device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IOT), virtual reality, augmented reality, industrial control, automatic driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, smart city, etc. The terminal can be a mobile phone, a tablet computer, a computer with wireless transceiver function, a wearable device, a vehicle, an airplane, a ship, a robot, a mechanical arm, a smart home device, etc. The embodiments of the present application do not limit the specific technology and specific device form adopted by the terminal.

[0043] Base stations and terminals can be fixed or movable. Base stations and terminals can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on the water surface; they can also be deployed on airplanes, balloons, and artificial satellites. The embodiments of this application do not limit the application scenarios of base stations and terminals.

[0044] The roles of the base station and the terminal can be relative, for example, Figure 1 The helicopter or drone 120i in the figure can be configured as a mobile base station. For the terminal 120j that accesses the wireless access network 100 through 120i, the terminal 120i is a base station; but for the base station 110a, 120i is a terminal, that is, 110a and 120i communicate through the wireless air interface protocol. Of course, 110a and 120i can also communicate through the interface protocol between base stations. In this case, relative to 110a, 120i is also a base station. Therefore, base stations and terminals can be collectively referred to as communication devices. Figure 1 110a and 110b in the figure may be referred to as communication devices having base station functions. Figure 1 120a-120j in the figure can be called communication devices with terminal functions.

[0045] Base stations and terminals, base stations and base stations, and terminals and terminals can communicate through authorized spectrum, unauthorized spectrum, or both; they can communicate 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.

[0046] 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 including the base station function. The control subsystem including the base station function 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 a modem) in the terminal, or by a device including the terminal function.

[0047] In this application, the base station sends a downlink signal or downlink information to the terminal, and the downlink information is carried on the downlink channel; the terminal sends an uplink signal or uplink information to the base station, and the uplink information is carried on the uplink channel. In order to communicate with the base station, the terminal needs to establish a wireless connection with the cell controlled by the base station. The cell with which the terminal has established a wireless connection is called the service cell of the terminal. When the terminal communicates with the service cell, it will also be interfered by signals from neighboring cells.

[0048] 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.

[0049] 1. AI

[0050] AI is a technology that simulates the human brain to perform complex calculations. AI can be applied to NR systems to improve network performance and user experience by intelligently collecting and analyzing data. Figure 2 As shown, Figure 2 This is the application framework diagram of AI in the NR system.

[0051] exist Figure 2In the data collection module, the data collection module is used as a database for AI model training and data analysis and reasoning, and is used to collect and store data inputs from, for example, gNB, gNB-CU, gNB-DU, UE or other network entities. The model training module is used to analyze the training data provided by the data collection module to obtain the optimal AI model. The model prediction module, based on the AI ​​model obtained by the model training module, analyzes the prediction data provided by the data collection module, and outputs the prediction results to the execution entity (including reasonable predictions of network operation, or guiding the network to execute corresponding adjustment strategies). Furthermore, the execution entity schedules and manages the network entity based on the prediction results output by the model prediction module, that is, the execution entity is used to schedule multiple network entities including itself to perform operations corresponding to the prediction results. After each network entity in the communication network performs the corresponding operation according to the prediction result, the specific performance of the communication network (which can be understood as the data generated during the operation of each network entity) will be collected and stored by the data collection module.

[0052] 2. Federated Learning

[0053] Federated learning is the joint training of AI models by multiple devices (referred to as distributed nodes participating in federated learning) without sharing data resources (or understood as model training data) to establish a shared AI model. According to the distribution of the data feature space and sample identity (ID) space of the training data between different distributed nodes, federated learning can be divided into the following three types: horizontal federated learning (HFL), vertical federated learning (VFL) and federated transfer learning (FTL).

[0054] Regardless of the type of federated learning, during the federated learning process, the distributed nodes participating in the federated learning can perform joint training of AI models by sharing AI collaborative data instead of sharing data resources. Among them, AI collaborative data includes one or more of the following data: gradient data, AI model, AI sub-model, AI model output, AI model intermediate value, etc.

[0055] For example, Figure 3As shown, device 1 and device 2 are distributed nodes participating in federated learning. During the process of federated learning, device 1 performs model training to obtain AI collaboration data #1, and device 2 performs model training to obtain AI collaboration data #2. Furthermore, device 1 sends AI collaboration data #1 to the centralized node (server) of the federated learning, and device 2 sends AI collaboration data #2 to the centralized node. The centralized node aggregates the AI ​​collaboration data #1 and AI collaboration data #2 to obtain AI collaboration data #12. Afterwards, the centralized node sends the AI ​​collaboration data #12 to device 1 and device 2, so that device 1 and device 2 perform model training according to the AI ​​collaboration data #12.

[0056] Usually, the amount of AI collaborative data is large. In order to reduce the communication overhead generated by each distributed node uploading and downloading AI collaborative data during federated learning, the AI ​​collaborative data can be quantized and compressed before transmission. Among them, quantization and compression of AI collaborative data can be understood as representing the AI ​​collaborative data to be transmitted (including upload or download) by multiple bits; the number of bits used to represent the AI ​​collaborative data can be understood as the quantization level corresponding to the AI ​​collaborative data, or called quantization resolution.

[0057] It is understandable that the fewer bits used to represent AI collaborative data, the lower the quantization resolution of the AI ​​collaborative data, and the greater the error introduced during the transmission process (which can be understood as the quantization error), which may increase the time it takes for the model to iterate and converge during the federated learning process; while, the more bits used to represent AI collaborative data, the higher the quantization resolution of the AI ​​collaborative data, the larger the amount of data transmitted during the transmission process, and the longer the transmission time, which may increase the time it takes for the model to train during the federated learning process. It can be seen that the quantization resolution will affect the training time of the federated learning model to a certain extent.

[0058] Typically, the distributed nodes of federated learning will quantize AI collaborative data based on a fixed (or understood as predetermined) quantization resolution. However, during the process of federated learning, the resources that the distributed nodes can use for federated learning (including but not limited to communication resources or computing resources, etc.) may change. In this case, if the AI ​​collaborative data is always quantized at a fixed quantization resolution, the quantization resolution will not match the current distributed nodes, which may increase the model training time of federated learning and reduce the training efficiency of federated learning.

[0059] In order to improve the training efficiency of collaborative learning, the present application provides a data processing method and a communication device. The data processing method and the communication device provided by the embodiment of the present application are described in detail below in conjunction with the accompanying drawings. Among them, the collaborative learning mentioned in the present application can be understood as an AI model training method in which multiple devices collaborate (or are understood as collaborative or joint) to perform training. For example, the collaborative learning can be distributed learning, federated learning, edge learning, or segmented learning.

[0060] Figure 4a is a flow chart of a data processing method provided in an embodiment of the present application. Figure 4a As shown, the data processing method includes the following steps S401 to S405: Figure 4a The method execution subject shown is described by taking the first device and the second device as an example. It can be understood that Figure 4a The method execution subject shown may also be a module (for example, a chip) in the first device and a module (for example, a chip) in the second device.

[0061] The first device and the second device are devices participating in collaborative learning. For example, the first device may be a centralized node corresponding to collaborative learning, and the second device may be a distributed node in collaborative learning (or a device for performing model training). Figure 1 Any one of the devices in the RAN 100 (such as a RAN node or terminal, etc.), the functional network element in the CN 200, or the server corresponding to the Internet 300 in the communication system. The second device may also be Figure 1 Any of the equipment in the RAN 100, the functional network element in the CN 200, or the server corresponding to the Internet 300 in the communication system; this application does not specifically limit this. Among them:

[0062] S401: The second device sends a first message to the first device, wherein the first message includes one or more of communication condition information of the second device, computing resource information of the second device, or model training information of the second device.

[0063] It can be understood that the second device is a device that uploads AI collaborative data during the collaborative learning process, and the first device is a device that determines the quantization resolution of the second device during the collaborative learning process. Taking the collaborative learning as federated learning as an example, the first device can be a service node (server) of the federated learning, and the second device can be one of the multiple client nodes (client) of the federated learning.

[0064] That is to say, before the first device updates (or is understood to determine) the quantization resolution of the second device, the first device receives a first message from the second device, and the first message is used to assist the first device in updating the quantization resolution of the second device. Specifically, the second device can indicate the communication environment in which the second device is located (or called corresponding) through the communication condition information in the first message, indicate the computing resources that the second device can use for collaborative learning through the computing information in the first message, and indicate the relevant information of the second device during the model training process (such as the size of AI collaborative data or quantization resolution related data, etc.) through the model training information in the first message. It should be noted that this application does not specifically limit the way in which the second device obtains communication condition information, computing resource information and model training information.

[0065] In a possible implementation, the communication condition information includes one or more of communication bandwidth information, transmission delay, reference signal receiving power (RSRP), reference signal receiving quality (RSRQ), signal to interference plus noise ratio (SINR), bit error rate or throughput. The computing resource information includes one or more of the size of computing resources, the size of computing power or computing delay. The model training information includes one or more of the test set loss value, the training set loss value, the gradient norm size, the gradient size, and the model size.

[0066] In a possible implementation, the first message sent by the second device may include one or more of the following information in addition to the communication condition information of the second device, the computing resource information of the second device, or the model training information of the second device: a measurement identifier, used to identify the reporting of the first message and to associate with the subsequent second message; a second device identifier, used to identify the device that sends the first message, or understood as a device whose quantization resolution is to be updated; a first device identifier, used to identify the device that receives the first message, or understood as a device for determining the quantization resolution; a model identifier, used to identify the AI ​​model corresponding to the first message; an identifier of the training wheel corresponding to the model, used to indicate that the first message is used to update the quantization resolution used when the second device feeds back which training wheel's AI collaboration data.

[0067] S402. The first device determines a quantization resolution according to one or more of the communication condition information, the computing resource information, or the model training information.

[0068] That is, after receiving the first message, the first device determines the quantization resolution according to the information included in the first message. In one possible implementation, the first device determines the quantization resolution according to one or more of the communication condition information, the computing resource information, or the model training information with the goal of minimizing the training time of the model.

[0069] It can be understood that in each round of training of the collaborative learning model, the first device will wait for multiple distributed nodes (including the second device) of collaborative learning to feedback AI collaborative data before executing subsequent steps (such as aggregating AI collaborative data, etc.). It can be seen that the training duration of the model is affected by the delay in the distributed nodes of collaborative learning to feedback AI collaborative data. In this case, in order to minimize the training duration of the model, the first device can determine a larger quantization resolution (i.e., a larger number of bits corresponding to the quantization resolution) for distributed nodes with large computing power, small delay, or large gradient norm to improve the accuracy of the AI ​​collaborative data fed back by the distributed nodes, and determine a smaller quantization resolution (i.e., a smaller number of bits corresponding to the quantization resolution) for distributed nodes with small computing power, large delay, or small gradient norm to reduce the communication delay between the first device and the distributed node, thereby reducing the training delay.

[0070] S403: The first device sends a second message to the second device, where the second message includes indication information indicating a quantization resolution.

[0071] That is, after the first device determines the quantization resolution of the second device, it indicates the quantization resolution to the second device through the second message. The indication information may indicate the quantization resolution by directly indicating the quantization resolution (i.e., the number of bits used for quantization), for example, the second message includes 4 bits or 16 bits, etc.; or it may indicate the quantization index corresponding to the quantization resolution, for example, if the quantization index included in the second message is 0, it means that the quantization is based on 4 bits, and if the quantization index included in the second message is 1, it means that the quantization is based on 16 bits.

[0072] In a possible implementation, in addition to the indication information indicating the quantization resolution, the second message may also include one or more of the following information: a measurement identifier, corresponding to the measurement identifier contained in the first message, used to identify the issuance of the second message; a second device identifier, used to identify the device receiving the second message, or understood as a device applying the quantization resolution; a first device identifier, used to identify the device sending the second message, or understood as a device determining the quantization resolution; a model identifier, used to identify the AI ​​model corresponding to the quantization resolution; an identifier of the training wheel corresponding to the model, used to indicate the quantization resolution used when the second device feeds back the AI ​​collaboration data of which training wheel is updated.

[0073] S404: The second device quantizes the first AI collaboration data based on the quantization resolution to obtain second AI collaboration data.

[0074] In the training round corresponding to the quantization resolution, the second device trains the model to obtain the first AI collaboration data (i.e., the AI ​​collaboration data before quantization). Further, the second device quantizes the first AI collaboration data based on the quantization resolution to obtain the second AI collaboration data (i.e., the AI ​​collaboration data after quantization).

[0075] S405. The second device sends second AI collaboration data to the first device.

[0076] It can be understood that after the second device obtains the second AI collaboration data, the second device sends a fourth message to the first device, and the fourth message includes the second AI collaboration data. In a possible implementation, in addition to the second AI collaboration data, the fourth message may also include one or more of the following information: a second device identifier, used to identify the device that sends the fourth message; a first device identifier, used to identify the device that receives the fourth message; a model identifier, used to identify the AI ​​model corresponding to the second AI collaboration data; an identifier of the training wheel corresponding to the model, used to indicate which training wheel the second AI collaboration data belongs to.

[0077] In summary, through Figure 4a The method shown can dynamically adjust the quantization resolution of the second device in combination with the communication conditions, computing resources or model training information of the current second device, which is beneficial to improving the compatibility of the quantization resolution with the current second device. By quantizing and transmitting AI collaborative data at a quantization resolution that is compatible with the current second device, it is beneficial to improve the training efficiency of collaborative learning.

[0078] Next, the collaborative learning process is described in detail in the following two cases according to the triggering method of triggering the second device to send the first message to the first device. It should be noted that in the following two cases, only the devices participating in the collaborative learning include the first device, the second device and the third device as an example. The devices participating in the collaborative learning may also include other devices (not shown in the figure), and this application does not make specific limitations on this.

[0079] Case 1: triggering the second device to send the first message to the first device through passive triggering.

[0080] That is to say, in situation one, the second device receives a third message from other devices (other devices except the second device, including the first device), and the third message is used to trigger the second device to send the aforementioned first message to the first device. In this case, the first message can be understood as a response message corresponding to the third message. For ease of understanding, the following text uses the example that the third message is sent by the first device to the second device for exemplary explanation, which should not be regarded as a specific limitation on the corresponding sending device of the third message.

[0081] See also Figure 4b As shown, the Figure 4b A flowchart of collaborative learning provided for this application. Figure 4b As shown, the collaborative learning includes the following steps S411 to S419. Among them:

[0082] S411. The first device sends a third message to the second device, where the third message is used to request the second device to assist in updating a quantization resolution.

[0083] It can be understood that, when the first device determines that there is a need to update the quantization resolution of the second device, the first device sends a third message to the second device, and the third message is used to request the second device to assist in updating the quantization resolution. It should be noted that the specific name of the third message is not specifically limited in this application. For example, the third message can be a request message for assisting in updating the quantization resolution.

[0084] Exemplarily, devices 1 to 5 are devices that perform federated learning tasks. Among them, device 1 is the server corresponding to the federated learning task, and devices 2 to 5 are clients that perform the federated learning task. In the process of executing the federated learning task, each device in devices 2 to 5 will feedback AI collaboration data to device 1. In a certain round (or training round (epoch)) of federated learning training, the delay for device 1 to wait for device 2 to feedback AI collaboration data #2 is delay 1, the delay for device 1 to wait for device 3 to feedback AI collaboration data #3 is delay 2, the delay for device 1 to wait for device 4 to feedback AI collaboration data #4 is delay 3, and the delay for device 1 to wait for device 5 to feedback AI collaboration data #5 is delay 4. Among them, delay 1 is the maximum value of delay 1 to delay 4, and the absolute value of the difference between the average delay of delay 1 to delay 4 and delay 1 is greater than the first threshold; the first threshold is a preset value greater than 0, and its specific value can be adjusted according to the specific application scenario. In this case, in order to reduce the time consumed by device 1 waiting for all clients to feedback AI collaboration data, device 1 can reduce the delay of device 1 waiting for device 2 to feedback AI collaboration data by adjusting the quantization resolution of device 2, which can be regarded as device 1 determining the need to update the quantization resolution of device 2 (that is, it can be understood as determining device 2 as the second device). Further, device 1 sends a third message to device 2, and the third message is used to request device 2 to assist device 1 in updating the quantization resolution, that is, requesting device 2 to send the first message to device 1.

[0085] In a possible implementation manner, the third message includes one or more of the following information:

[0086] ① Measurement ID, also known as measurement ID, is used to indicate the request for assistance in updating the quantization resolution.

[0087] ② The identifier of the second device, used to indicate the device whose quantization resolution is to be updated, or used to indicate the device that receives the third message, or used to indicate the device that sends the first message.

[0088] ③ The identifier of the first device, used to indicate a device for determining the quantization resolution, or used to indicate a device for sending the third message, or used to indicate a device for receiving the first message.

[0089] ④ Model identification, used to indicate that the second device needs to update the AI ​​model (or understand it as an AI collaboration model) of the quantization resolution. The model identification can be the identification of the AI ​​model, or associated information associated with the AI ​​model (such as the business identification corresponding to the AI ​​model), etc.

[0090] ⑤ Indication information indicating the reason for updating the quantization resolution, used to indicate the reason for this update of the quantization resolution. Exemplarily, the third message includes field A, and the reason for this update of the quantization resolution is indicated by the value of field A. When the value of field A is 1, the third message indicates that the reason for updating the quantization resolution is: the latency of the second device (i.e., the latency of the first device waiting for the second device to feedback the AI ​​collaborative data) is large, or it is understood that the efficiency of the second device to feedback the AI ​​collaborative data needs to be improved and the quantization resolution needs to be updated. When the value of field A is 0, the third message indicates that the reason for updating the quantization resolution is: the error of the AI ​​collaborative data fed back by the second device is large, or it is understood that the accuracy of the AI ​​collaborative data fed back by the second device needs to be improved and the quantization resolution needs to be updated.

[0091] In a possible implementation, the indication information indicating the reason for updating the quantization resolution includes the time difference between the first delay and the average delay, or the ratio between the first delay and the average delay. The first delay is the delay of the first device waiting for the second device to feedback the AI ​​collaboration data, and the average delay is the average value of the delay of the first device waiting for multiple distributed nodes to feedback the AI ​​collaboration data, and the multiple distributed nodes include the second device. It can be understood that when the reason for updating the quantization resolution is that the delay of the second device is large (or it is understood that the quantization resolution needs to be updated to improve the efficiency of the second device to feedback the AI ​​collaboration data), the indication information included in the third message can be the time difference between the first delay and the average delay, or the ratio between the first delay and the average delay.

[0092] ⑥ The identifier of the training wheel corresponding to the model is used to indicate the quantization resolution used when the second device feeds back the AI ​​collaboration data of which training wheel the first device updates based on the first message. For example, the identifier of the Nth training wheel of the model included in the third message indicates that the first device updates the quantization resolution used when the second device feeds back the AI ​​collaboration data of the Nth training wheel based on the first message, that is, when the second device subsequently feeds back the AI ​​collaboration data of the Nth training wheel, it can use the quantization resolution updated based on the first message for quantization. It should be noted that in this application, the training wheel may also be referred to as a wheel, and the whole text is the same.

[0093] ⑦ The identifier of the information requested for feedback is used to indicate that the first device expects (or understands as a request) information to be fed back by the second device. The information requested for feedback includes one or more of communication condition information, computing resource information, or model training information. The second device may subsequently generate a first message based on the identifier of the information requested for feedback, that is, the first message is generated based on the identifier of the information requested for feedback.

[0094] It should be noted that the information included in the first message may be part or all of the information requested for feedback by the third message. For example, the first device requests the second device to feedback communication condition information, computing resource information and model training information, then the identifier of the information requested for feedback in the third message sent by the first device includes: an identifier of the communication condition information (such as an identifier of the transmission delay or an identifier of the RSRP, etc.), an identifier of the computing resource information and an identifier of the model training information (such as an identifier of the test set loss value or an identifier of the gradient norm size). In this case, the second device may only feedback part (that is, feedback one or two of the communication condition information, computing resource information or model training information) or feedback all (that is, feedback of the communication condition information, computing resource information and model training information) according to the third message.

[0095] ⑧ Configuration information for transmitting the first message, used to instruct the second device how to transmit the first message. In a possible implementation, the configuration information for transmitting the first message includes one or more of a transmission period for transmitting the first message, a trigger condition for transmitting the first message, or a number of times of transmitting the first message. Wherein:

[0096] The transmission period of transmitting the first message may be determined based on a time interval value / index or based on a training round. Taking the transmission period determined based on a training round as an example, when the configuration information indicates that the transmission period of the first message is 3 training rounds, the configuration information indicates that the second device transmits the first message once every 3 training rounds.

[0097] The triggering condition for transmitting the first message may include one or more of the following conditions: 1. A triggering condition for triggering instant transmission of the first message (i.e., the second device transmits the first message once if the triggering condition is met), for example, the triggering condition is that the gradient norm of the second device is greater than or equal to the gradient norm threshold, the communication bandwidth is less than or equal to the bandwidth threshold, the channel quality (e.g., RSRP, RSRQ, SINR, bit error rate, throughput, etc.) is less than or equal to the channel quality threshold, the computing power is less than or equal to the computing power threshold, etc.; 2. A triggering condition for triggering periodic transmission of the first message (i.e., the second device transmits the first message once if the triggering condition is met, according to the transmission period of the first message, the period The trigger condition for periodically transmitting the first message is as follows: for example, the gradient norm of the second device is greater than or equal to the gradient norm threshold, the communication bandwidth is less than or equal to the bandwidth threshold, the channel quality is less than or equal to the channel quality threshold, the computing power is less than or equal to the computing power threshold, etc.; 3. The trigger condition for ending the periodic transmission of the first message (that is, when the second message periodically transmits the first message, if the second device meets the trigger condition, the second device stops periodically transmitting the first message), for example, the trigger condition is that the gradient norm of the second device is less than the gradient norm threshold, the communication bandwidth is greater than the bandwidth threshold, the channel quality is greater than the channel quality threshold, the computing power is greater than the computing power threshold, etc.

[0098] The number of times the first message is transmitted is used to indicate the total number of times the second device needs to transmit the first message after receiving the configuration information for transmitting the first message.

[0099] S412: The second device sends the first message to the first device according to the third message.

[0100] It can be understood that the first message is understood as a response message of the third message. For example, the measurement identifier in the first message corresponds to (or is understood to be the same as) the measurement identifier included in the third message, or the identifier of the model in the first message is the same as the identifier of the model included in the third message, or the identifier of the training wheel corresponding to the model in the first message is the same as the identifier of the training wheel corresponding to the model included in the third message.

[0101] It should be noted that the present application does not specifically limit the specific name of the first message. For example, the first message may be a response message for assisting in updating the quantization resolution. For a specific description of the first message, please refer to the specific description of the first message in S401 above, which will not be repeated here.

[0102] S413. The first device determines a quantization resolution according to one or more of the communication condition information, the computing resource information, or the model training information.

[0103] S414: The first device sends a second message to the second device, where the second message includes indication information indicating a quantization resolution.

[0104] S415. The second device quantizes the first AI collaboration data based on the quantization resolution to obtain second AI collaboration data.

[0105] S416. The second device sends second AI collaboration data to the first device.

[0106] The specific implementation process of S413 to S416 can refer to the description of the specific implementation process of S402 to S405 above, which will not be repeated here.

[0107] S417. The third device sends third AI collaboration data to the first device.

[0108] It can be understood that the third device trains the model to obtain AI collaboration data of the third device; further, the third device quantizes the AI ​​collaboration data of the third device based on its corresponding quantization resolution to obtain third AI collaboration data, and sends the third AI collaboration data to the first device.

[0109] It should be noted that the specific content of the AI ​​collaboration data of the third device can refer to the specific description of the aforementioned AI collaboration data. For example, the AI ​​collaboration data of the third device includes one or more of the following data: gradient data, AI model, AI sub-model, AI model output, AI model intermediate value, etc. The quantization resolution of the third device can also refer to the aforementioned Figure 4a The data processing method provided is used to update the quantitative resolution, and this application does not make any specific limitations on this.

[0110] S418. The first device aggregates the second AI collaboration data and the third AI collaboration data to obtain aggregated AI collaboration data.

[0111] That is to say, after the first device receives the AI ​​collaboration data fed back from all (or part) of the distributed nodes (such as the second device and the third device) participating in the collaborative learning, it aggregates the AI ​​collaboration data to obtain aggregated AI collaboration data.

[0112] S419. The first device sends the aggregated AI collaboration data to the second device and the third device respectively.

[0113] That is to say, after the first device obtains the aggregated AI collaborative data, the first device sends the aggregated AI collaborative data to the distributed nodes participating in the collaborative learning, so that the distributed nodes participating in the collaborative learning (including the second device and the third device) respectively perform model training based on the aggregated AI collaborative data.

[0114] It should be noted that when the first device sends aggregated AI collaborative data to the distributed nodes participating in collaborative learning, it can also indicate the identifier of the distributed node (for example, the identifier of the second device or the identifier of the third device) to indicate the device receiving the aggregated AI collaborative data; the first device identifier is used to identify the device sending the aggregated AI collaborative data; the identifier of the model is used to indicate the AI ​​model corresponding to the aggregated AI collaborative data; the identifier of the training wheel corresponding to the model is used to indicate which training wheel of the model the aggregated AI collaborative data corresponds to. It should also be noted that the aggregated AI collaborative data sent by the first device to the distributed nodes participating in collaborative learning can be unquantized data (that is, it can be understood as aggregated data obtained by aggregating the second AI collaborative data and the third AI collaborative data), or quantized data (that is, it can be understood as aggregated data obtained by aggregating the second AI collaborative data and the third AI collaborative data, and then subjected to an aggregated quantization resolution quantization operation). This application does not make specific restrictions on this. Among them, the aggregated quantization resolution can be understood as the quantization resolution of the quantized aggregated AI collaborative data, and the aggregated quantization resolution can also be determined based on one or more of the communication condition information of the first device, the communication condition information of the second device, or the computing resource information of the second device, and this application does not make specific restrictions.

[0115] Case 2: triggering the second device to send the first message to the first device by actively triggering the second device.

[0116] See also Figure 4c As shown, the Figure 4c Another collaborative learning process diagram provided for this application. Figure 4c As shown, the collaborative learning includes the following steps S421 to S428. Among them:

[0117] S421. The second device sends a first message to the first device, where the first message is used to request the first device to update a quantization resolution of the second device.

[0118] That is, when the second device determines that it has a need to update the quantization resolution, the second device actively sends a first message to the first device based on the need to update the quantization resolution. It should be noted that the present application does not specifically limit the specific name of the first message. For example, the first message can be a request message for updating the quantization resolution. For a specific description of the first message, please refer to the specific description of the first message in S401 above, which will not be repeated here.

[0119] For example, device 1 and device 2 are devices that perform federated learning tasks. Among them, device 1 is the server that performs the federated learning task, and device 2 is the client that performs the federated learning task. In the process of performing the federated learning task, device 2 will feedback AI collaboration data to device 1. During a round of training in federated learning, if device 2 detects that one or more of the following conditions are met: the gradient norm of the AI ​​model is greater than or equal to the gradient norm threshold, the communication bandwidth of the second device is less than or equal to the bandwidth threshold, the channel quality of the second device is less than or equal to the channel quality threshold, the computing power of the second device is less than or equal to the computing power threshold, etc., it can be regarded as having a need to update the quantization resolution of device 2, and device 2 actively sends a first message to the first device.

[0120] S422. The first device determines a quantization resolution according to one or more of the communication condition information, the computing resource information, or the model training information.

[0121] S423: The first device sends a second message to the second device, where the second message includes indication information indicating a quantization resolution.

[0122] S424. The second device quantizes the first AI collaboration data based on the quantization resolution to obtain second AI collaboration data.

[0123] S425. The second device sends second AI collaboration data to the first device.

[0124] S426. The third device sends third AI collaboration data to the first device.

[0125] S427. The first device aggregates the second AI collaboration data and the third AI collaboration data to obtain aggregated AI collaboration data.

[0126] S428. The first device sends the aggregated AI collaboration data to the second device and the third device respectively.

[0127] The specific implementation process of S422 to S428 can refer to the description of the specific implementation process of S413 to S419 mentioned above, which will not be repeated here.

[0128] It is understandable that in order to implement the functions in the above embodiments, the device (including the aforementioned first device or second device) includes a hardware structure and / or software module corresponding to the execution of each function. It should be easily appreciated by those skilled in the art that, in combination with the units and method steps of each example described in the embodiments disclosed in this application, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or a computer software transceiver unit driving the hardware depends on the specific application scenario and design constraints of the technical solution.

[0129] Figure 5 and Figure 6 The following is a schematic diagram of the structure of possible communication devices provided in the embodiments of the present application. These communication devices can be used to implement the functions of the devices in the above method embodiments, and thus can also achieve the beneficial effects of the above method embodiments. In the embodiments of the present application, the communication device can be Figure 4a to Figure 4c The first device in the embodiment may also be a module (such as a chip) applied to the first device, or the communication device may be Figure 4a to Figure 4c The second device in the embodiment may also be a module (such as a chip) applied to the second device.

[0130] like Figure 5 As shown, the communication device 500 includes a processing unit 510 and a transceiver unit 520. The communication device 500 is used to implement the above Figure 4a to Figure 4c The function of the first device or the function of the second device in the method embodiment shown in FIG.

[0131] When the communication device 500 is used to implement Figure 4a to Figure 4c The function of the first device in the method embodiment shown is: a transceiver unit 520 is used to send a first message to the first device, and the first message includes one or more of communication condition information of the second device, computing resource information of the second device, or model training information of the second device; the transceiver unit 520 is also used to receive a second message from the first device, and the second message includes indication information, and the indication information indicates a quantization resolution, and the quantization resolution is related to one or more of the communication condition information, computing resource information, or model training information; the processing unit 510 is used to quantize the first artificial intelligence AI collaboration data based on the quantization resolution to obtain the second AI collaboration data; the transceiver unit 520 is also used to send the second AI collaboration data to the first device.

[0132] In one possible implementation, the communication condition information includes one or more of communication bandwidth information, transmission delay, reference signal received power RSRP, reference signal received quality RSRQ, signal to interference plus noise ratio SINR, bit error rate or throughput; the computing resource information includes one or more of the size of computing resources, the size of computing power or computing delay; the model training information includes one or more of the test set loss value, the training set loss value, the gradient norm size, the gradient size, and the model size.

[0133] In a possible implementation, the transceiver unit 520 is further used to receive a third message from the first device, where the third message is used to request assistance in updating the quantization resolution; and the first message is a response message corresponding to the third message.

[0134] In a possible implementation, the third message includes one or more of the following information: a measurement identifier, an identifier of the second device, an identifier of the first device, an identifier of the model, indication information indicating the reason for updating the quantization resolution, an identifier of the training round corresponding to the model, an identifier of the information requested for feedback, or configuration information for transmitting the first message; wherein the information requested for feedback includes one or more of communication condition information, computing resource information, or model training information; and the first message is generated based on the identifier of the information requested for feedback.

[0135] In one possible implementation, the indication information indicating the reason for updating the quantization resolution includes the time difference between the first delay and the average delay, or the ratio between the first delay and the average delay; wherein the first delay is the delay of the first device waiting for the second device to feedback the AI ​​collaboration data, and the average delay is the average value of the delay of the first device waiting for multiple distributed nodes to feedback the AI ​​collaboration data, and the multiple distributed nodes include the second device.

[0136] In a possible implementation, the configuration information for transmitting the first message includes one or more of a transmission period for transmitting the first message, a triggering condition for transmitting the first message, or a number of times of transmitting the first message.

[0137] In a possible implementation, the first message further includes one or more of the following information: a measurement identifier, a second device identifier, a first device identifier, an identifier of the model, and an identifier of a training wheel corresponding to the model.

[0138] For more detailed description of the transceiver unit 520 and the processing unit 510, please refer to Figure 4a to Figure 4c Related description of the second device in the method embodiment shown.

[0139] like Figure 5 As shown, the communication device 500 includes a processing unit 510 and a transceiver unit 520. The communication device 500 is used to implement the above Figure 4a to Figure 4c The function of the first device in the method embodiment shown in FIG.

[0140] When the communication device 500 is used to implement Figure 4a to Figure 4cThe functions of the first device in the method embodiment shown are: a transceiver unit 520, used to receive a first message from a second device, the first message including one or more of communication condition information of the second device, computing resource information of the second device, or model training information of the second device; a processing unit 510, used to determine a quantization resolution based on one or more of the communication condition information, the computing resource information, or the model training information; the transceiver unit 520, is also used to send a second message to the second device, the second message including indication information, the indication information indicating a quantization resolution, and the quantization resolution is related to one or more of the communication condition information, the computing resource information, or the model training information; the transceiver unit 520 is also used to receive second artificial intelligence AI collaboration data from the second device, the second AI collaboration data being quantized based on the quantization resolution.

[0141] In one possible implementation, the communication condition information includes one or more of communication bandwidth information, transmission delay, reference signal received power RSRP, reference signal received quality RSRQ, signal to interference plus noise ratio SINR, bit error rate or throughput; the computing resource information includes one or more of the size of computing resources, the size of computing power or computing delay; the model training information includes one or more of the test set loss value, the training set loss value, the gradient norm size, the gradient size, and the model size.

[0142] In a possible implementation, the transceiver unit 520 is further used to send a third message to the second device, where the third message is used to request assistance in updating the quantization resolution; and the first message is a response message corresponding to the third message.

[0143] In one possible implementation, the third message includes one or more of the following information: a measurement identifier, an identifier of the second device, an identifier of the first device, a model identifier, indication information indicating the reason for updating the quantization resolution, an identifier of a training round corresponding to the model, an identifier of information requested for feedback, and / or configuration information for transmitting the first message; wherein the information requested for feedback includes one or more of communication condition information, computing resource information, or model training information; and the first message is generated based on the identifier of the information requested for feedback.

[0144] In one possible implementation, the indication information indicating the reason for updating the quantization resolution includes the time difference between the first delay and the average delay, or the ratio between the first delay and the average delay; wherein the first delay is the delay of the first device waiting for the second device to feedback the AI ​​collaboration data, and the average delay is the average value of the delay of the first device waiting for multiple distributed nodes to feedback the AI ​​collaboration data, and the multiple distributed nodes include the second device.

[0145] In a possible implementation, the configuration information for transmitting the first message includes one or more of a transmission period for transmitting the first message, a triggering condition for transmitting the first message, or a number of times of transmitting the first message.

[0146] In a possible implementation, the first message further includes one or more of the following information: a measurement identifier, a second device identifier, a first device identifier, an identifier of the model, and an identifier of a training wheel corresponding to the model.

[0147] For more detailed description of the transceiver unit 520 and the processing unit 510, please refer to Figure 4a to Figure 4c Related description of the first device in the method embodiment shown.

[0148] like Figure 6 As shown, the communication device 600 includes a processor 610 and an interface circuit 620. The processor 610 and the interface circuit 620 are coupled to each other. It is understood that the interface circuit 620 can be a transceiver or an input-output interface. Optionally, the communication device 600 may also include a memory 630 for storing instructions executed by the processor 610 or storing input data required by the processor 610 to execute instructions or storing data generated after the processor 610 executes instructions.

[0149] When the communication device 600 is used to implement Figure 4a to Figure 4c When the method is shown, the processor 610 is used to implement the function of the above-mentioned processing unit 510, and the interface circuit 620 is used to implement the function of the above-mentioned transceiver unit 520.

[0150] When the above-mentioned communication device is a chip applied to the first device, the chip implements the functions of the first device in the above-mentioned method embodiment. The chip receives information from the second device, which can be understood as the information is first received by other modules in the first device (such as a radio frequency module or an antenna), and then sent to the chip by these modules. The chip sends information to the second device, which can be understood as the information is first sent to other modules in the first device (such as a radio frequency module or an antenna), and then sent to the second device by these modules.

[0151] When the above-mentioned communication device is a chip applied to the second device, the chip implements the function of the second device in the above-mentioned method embodiment. The chip receives information from the first device, which can be understood as the information is first received by other modules in the second device (such as a radio frequency module or an antenna), and then sent to the chip by these modules. The chip sends information to the first device, which can be understood as the information is first sent to other modules in the second device (such as a radio frequency module or an antenna), and then sent to the first device by these modules.

[0152] In the present application, when entity A sends information to entity B, it can be that A sends it directly to B, or that A sends it to B indirectly through other entities. Similarly, when entity B receives information from entity A, it can be that entity B directly receives the information sent by entity A, or that entity B indirectly receives the information sent by entity A through other entities. Entities A and B here can be RAN nodes or terminals, or modules inside the RAN nodes or terminals. The sending and receiving of information can be information interaction between a RAN node and a terminal, for example, information interaction between a base station and a terminal; the sending and receiving of information can also be information interaction between two RAN nodes, for example, information interaction between a CU and a DU; the sending and receiving of information can also be information interaction between different modules inside a device, for example, information interaction between a terminal chip and other modules of the terminal, or information interaction between a base station chip and other modules in the base station.

[0153] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or 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.

[0154] 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, register, hard disk, mobile hard disk, CD-ROM or any other form of storage medium 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 a component of the processor. The processor and the 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 the storage medium can also be present in a base station or a terminal as discrete components.

[0155] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part 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 instruction is loaded and executed on a computer, the process or function described in the embodiment of the present application is executed 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 instruction may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instruction may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server, data center, etc. that integrates one or more available media. The available medium may be a magnetic medium, for example, a floppy disk, a hard disk, a tape; it may also be an optical medium, for example, a digital video disc; it may also be a semiconductor medium, for example, a solid-state hard disk. The computer-readable storage medium may be a volatile or nonvolatile storage medium, or may include both volatile and nonvolatile types of storage media.

[0156] In the various embodiments of the present application, unless otherwise specified or provided for in any logical conflict, the terms and / or descriptions between the different embodiments are consistent and may be referenced to each other, and the technical features in the different embodiments may be combined to form new embodiments according to their inherent logical relationships.

[0157] In the present application, "at least one" means one or more, and "more than one" 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 the present application, the character " / " generally indicates that the previous and next associated objects are in an "or" relationship; in the formula of the present 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.

[0158] It is understood that the various numbers involved in the embodiments of the present application are only for the convenience of description and are not used to limit the scope of the embodiments of the present application. The size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic.

Claims

1. A data processing method, characterized in that: The method comprises: Sending a first message to a first device, where the first message includes one or more of communication condition information of a second device, computing resource information of the second device, or model training information of the second device; receiving a second message from the first device, the second message including indication information, the indication information indicating a quantization resolution, the quantization resolution being related to one or more of the communication condition information, the computing resource information, or the model training information; quantizing the first artificial intelligence (AI) collaboration data based on the quantization resolution to obtain second AI collaboration data; Send the second AI collaboration data to the first device.

2. The method according to claim 1, characterized in that: The communication condition information includes one or more of communication bandwidth information, transmission delay, reference signal received power RSRP, reference signal received quality RSRQ, signal to interference plus noise ratio SINR, bit error rate or throughput; The computing resource information includes one or more of the size of the computing resource, the size of the computing capability, or the computing delay; The model training information includes one or more of a test set loss value, a training set loss value, a gradient norm size, a gradient size, and a model size.

3. The method according to claim 1 or 2, characterized in that: The method further comprises: receiving a third message from the first device, wherein the third message is used to request assistance in updating a quantization resolution; The first message is a response message corresponding to the third message.

4. The method according to claim 3, characterized in that: The third message includes one or more of the following information: a measurement identifier, an identifier of the second device, an identifier of the first device, an identifier of a model, indication information indicating a reason for updating a quantization resolution, an identifier of a training wheel corresponding to the model, an identifier of information requesting feedback, or configuration information for transmitting the first message; The information requested for feedback includes one or more of the communication condition information, the computing resource information or the model training information; and the first message is generated according to the identifier of the information requested for feedback.

5. The method according to claim 4, characterized in that: The indication information indicating the reason for updating the quantization resolution includes a time difference between the first delay and the average delay, or a ratio between the first delay and the average delay; The first delay is the delay of the first device waiting for the second device to feedback the AI ​​collaboration data, and the average delay is the average value of the delay of the first device waiting for multiple distributed nodes to feedback the AI ​​collaboration data, and the multiple distributed nodes include the second device.

6. The method according to claim 4 or 5, characterized in that: The configuration information for transmitting the first message includes one or more of a transmission period for transmitting the first message, a triggering condition for transmitting the first message, or a number of times of transmitting the first message.

7. The method according to any one of claims 1 to 6, characterized in that: The first message also includes one or more of the following information: A measurement identifier, the second device identifier, the first device identifier, an identifier of a model, and an identifier of a training wheel corresponding to the model.

8. A data processing method, characterized in that: The method comprises: receiving a first message from a second device, the first message including one or more of communication condition information of the second device, computing resource information of the second device, or model training information of the second device; Determine a quantization resolution based on one or more of the communication condition information, the computing resource information, or the model training information; Sending a second message to the second device, where the second message includes indication information, where the indication information indicates a quantization resolution, where the quantization resolution is related to one or more of the communication condition information, the computing resource information, or the model training information; Second artificial intelligence (AI) collaboration data is received from the second device, where the second AI collaboration data is obtained by quantizing the first AI collaboration data based on the quantization resolution.

9. The method according to claim 8, characterized in that: The communication condition information includes one or more of communication bandwidth information, transmission delay, reference signal received power RSRP, reference signal received quality RSRQ, signal to interference plus noise ratio SINR, bit error rate or throughput; The computing resource information includes one or more of the size of the computing resource, the size of the computing capability, or the computing delay; The model training information includes one or more of a test set loss value, a training set loss value, a gradient norm size, a gradient size, and a model size.

10. The method according to claim 8 or 9, characterized in that: The method further comprises: Sending a third message to the second device, where the third message is used to request assistance in updating the quantization resolution; The first message is a response message corresponding to the third message.

11. The method according to claim 10, characterized in that: The third message includes one or more of the following information: a measurement identifier, an identifier of the second device, an identifier of the first device, a model identifier, indication information indicating a reason for updating a quantization resolution, an identifier of a training round corresponding to the model, an identifier of information requesting feedback, and / or configuration information for transmitting the first message; The information requested for feedback includes one or more of the communication condition information, the computing resource information or the model training information; and the first message is generated according to the identifier of the information requested for feedback.

12. The method according to claim 11, characterized in that: The indication information indicating the reason for updating the quantization resolution includes a time difference between a first delay and an average delay, or a ratio between the first delay and the average delay; wherein the first delay is a delay for the first device to wait for the second device to feedback the AI ​​collaboration data, and the average delay is an average value of delays for the first device to wait for multiple distributed nodes to feedback the AI ​​collaboration data, and the multiple distributed nodes include the second device.

13. The method according to claim 11 or 12, characterized in that: The configuration information for transmitting the first message includes one or more of a transmission period for transmitting the first message, a triggering condition for transmitting the first message, or a number of times of transmitting the first message.

14. The method according to any one of claims 8 to 13, characterized in that: The first message also includes one or more of the following information: A measurement identifier, the second device identifier, the first device identifier, an identifier of a model, and an identifier of a training wheel corresponding to the model.

15. A communication device, characterized in that: The method comprises a module for executing the method according to any one of claims 1 to 7, or comprises a module for executing the method according to any one of claims 8 to 14.

16. A communication device, characterized in that: The invention comprises 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 7 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 8 to 14 through a logic circuit or by executing code instructions.

17. A computer-readable storage medium, characterized in that: The storage medium stores a computer program or instruction. When the computer program or instruction is executed by the communication device, the communication device implements the method as described in any one of claims 1 to 7, or implements the method as described in any one of claims 8 to 14.

18. A computer program product, characterized in that The computer program product comprises 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 7, or implements the method according to any one of claims 8 to 14.

Images