Communication method, apparatus, device, and storage medium

Abstract

The application provides a communication method, device and equipment and a storage medium. The method comprises the following steps: a network device sends a training time delay constraint instruction to a terminal device participating in federated learning; the terminal device receives the training time delay constraint instruction from the network device; and the training time delay constraint instruction is used to indicate the reporting deadline of a model training result. Since the network device can send the training time delay constraint instruction to the terminal device participating in the federated learning, the training time delay constraint instruction is used to indicate the reporting deadline of the model training result, so that in the case that there are two or more terminal devices participating in the federated learning, different terminal devices can upload the model training result according to the training time delay constraint instruction, so as to reduce the time difference of the model training result uploaded by different terminal devices participating in the federated learning, thereby ensuring the overall training time delay of the federated learning, reducing the possibility of late reporting of the terminal device, and helping to reduce the waste of network resources.

Description

Technical Field

[0001] This application relates to the field of network communication technology, and in particular to a communication method, apparatus, device and storage medium. Background Technology

[0002] Federated learning, through distributed training and encryption technology, can protect user privacy and data security while obtaining a model that meets performance requirements.

[0003] For example, multiple terminal devices can download model data from a network device, then use their respective data to train the model, and upload the trained model to the network device. The network device then aggregates the models uploaded by the terminal devices to obtain an updated model.

[0004] In the existing federated learning process, the time it takes for each terminal device to upload the model varies, which may increase the overall training latency of federated learning; or the terminal device may report the model training results too late, resulting in the model training results contributing too little to the current iteration and wasting transmission resources. Summary of the Invention

[0005] This application provides a communication method, apparatus, device, and storage medium to solve the problems existing in the prior art.

[0006] In a first aspect, this application provides a communication method, the method comprising:

[0007] Send a training delay constraint instruction to the terminal devices participating in the i-th round of federated learning. The training delay constraint instruction is used to indicate the deadline for reporting the training results of the i-th round of model, where i is a positive integer greater than or equal to 0.

[0008] In some embodiments, the training latency constraint indicator is used to indicate the deadline for reporting the training results of the i-th round of the model, including:

[0009] The training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

[0010] In some embodiments, the method further includes:

[0011] Receive rejection response messages from N terminal devices participating in the i-th round of federated learning. The rejection response messages are used to indicate that the training latency constraint indication is not supported, where N is a positive integer greater than or equal to 1.

[0012] In some embodiments, the method further includes:

[0013] Send notification messages to N terminal devices to cancel their participation in the i-th round of federated learning.

[0014] In some embodiments, the method further includes:

[0015] Receive relaxation instruction information from the terminal device participating in the i-th round of federated learning. The relaxation instruction information is used to indicate the conditions for dropping data packets of model training results and / or that all data packets of model training results are useful.

[0016] In some embodiments, sending a training latency constraint instruction to the terminal device participating in the i-th round of federated learning includes:

[0017] Send configuration information or session establishment request messages to the terminal devices participating in the i-th round of federated learning. The configuration information includes training latency constraint indications, and the session establishment request messages include training latency constraint indications.

[0018] In some embodiments, sending a training latency constraint instruction to the terminal device participating in the i-th round of federated learning includes:

[0019] Send control protocol data units to the terminal devices participating in the i-th round of federated learning. The control protocol data units include training latency constraint indications.

[0020] In some embodiments, the method further includes:

[0021] Send data packet acceleration instructions to K terminal devices among the terminal devices participating in the i-th round of federated learning. The data packet acceleration instructions are also used to indicate the acceleration of the data packet to be sent for the model training results of the i-th round, where K is a positive integer greater than or equal to 1.

[0022] In some embodiments, sending a training latency constraint instruction to the terminal device participating in the i-th round of federated learning includes:

[0023] The confirmation information is sent to the terminal devices participating in the i-th round of federated learning, which includes a training latency constraint indication. The terminal devices participating in the i-th round of federated learning are the terminal devices among the terminal devices participating in the i-1 round of federated learning, and i is a positive integer greater than or equal to 1.

[0024] In some embodiments, the method further includes:

[0025] Receive data packets containing the model training results from the terminal devices participating in the i-th round of federated learning;

[0026] If the difference between the arrival time of the model training result of the first terminal device among the terminal devices participating in the i-th round of federated learning and the reporting deadline is greater than or equal to the first threshold, the timeout reporting count of the first terminal device will be incremented by 1 in the data packet timeout reporting record table.

[0027] In some embodiments, the method further includes:

[0028] If the number of timeout reports by the first terminal device exceeds the second threshold, the first terminal device will be restricted or canceled from continuing to participate in the i-th round of federated learning.

[0029] In some embodiments, the method further includes:

[0030] Send a device capability condition indication to the terminal devices participating in the i-th round of federated learning. The device capability condition indication is used to indicate the device's computing power processing capability threshold.

[0031] In some embodiments, the method further includes:

[0032] Receive device capability information from M terminal devices participating in the i-th round of federated learning, where M is a positive integer greater than or equal to 1.

[0033] In some embodiments, the method further includes:

[0034] Based on the device capability information of the terminal devices, update the candidate list of terminal devices participating in federated learning. The candidate list includes the identifier of at least one terminal device.

[0035] In some embodiments, the method further includes:

[0036] If a cell handover is detected in the second terminal device among M terminal devices, the device capability information of the second terminal device is sent to the network device that will serve the second terminal device after the cell handover.

[0037] Secondly, this application provides a communication method, the method comprising:

[0038] Receive a training delay constraint indication from the network device. The training delay constraint indication is used to indicate the deadline for reporting the training results of the i-th round of the model, where i is a positive integer greater than or equal to 0.

[0039] In some embodiments, the training latency constraint indicator is used to indicate the deadline for reporting the training results of the i-th round of the model, including:

[0040] The training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

[0041] In some embodiments, the method further includes:

[0042] If it is determined that the training latency constraint indication is not supported, a rejection response message is sent to the network device.

[0043] In some embodiments, the method further includes:

[0044] Receive a notification message from a network device that it has cancelled its participation in round i of federated learning.

[0045] In some embodiments, the method further includes:

[0046] Send a relaxation instruction to the network device. The relaxation instruction is used to indicate the conditions for dropping data packets of the model training results and / or that all data packets of the model training results are useful.

[0047] In some embodiments, receiving a training latency constraint indication from a network device includes:

[0048] Receive configuration information or session establishment request messages from network devices. The configuration information includes training latency constraint indications, and the session establishment request messages include training latency constraint indications.

[0049] In some embodiments, receiving a training latency constraint indication from a network device includes:

[0050] Receive control protocol data units from network devices, the control protocol data units including training delay constraint indications.

[0051] In some embodiments, the method further includes:

[0052] Receive a packet acceleration instruction from the network device. The packet acceleration instruction is used to indicate the accelerated transmission of the data packets to be sent for the model training results of the i-th round.

[0053] In some embodiments, receiving a training latency constraint indication from a network device includes:

[0054] The system receives an acknowledgment message for the data packet containing the model training results of the (i-1)th round from the network device. The acknowledgment message includes a training latency constraint indication, where i is a positive integer greater than or equal to 1.

[0055] In some embodiments, the method further includes:

[0056] Receive device capability condition indication from network devices, which indicates the device's computing power processing capacity threshold.

[0057] In some embodiments, the method further includes:

[0058] If it is determined that the device capability condition indication is not met, then device capability information is sent to the network device.

[0059] Thirdly, this application provides a communication device, the device comprising:

[0060] The sending module is used to send a training delay constraint indication to the terminal device participating in the i-th round of federated learning. The training delay constraint indication is used to indicate the deadline for reporting the training results of the i-th round of model, where i is a positive integer greater than or equal to 0.

[0061] Fourthly, this application provides a communication device, the device comprising:

[0062] The receiving module is used to receive a training delay constraint indication from the network device, wherein the training delay constraint indication is used to indicate the reporting deadline of the i-th round of model training results, and i is a positive integer greater than or equal to 0.

[0063] Fifthly, this application provides a communication device, including a memory, a transceiver, and a processor:

[0064] A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations:

[0065] A training delay constraint indication is sent to the terminal device participating in the i-th round of federated learning. The training delay constraint indication is used to indicate the deadline for reporting the training results of the i-th round of model, where i is a positive integer greater than or equal to 0.

[0066] Sixthly, this application provides a communication device, including a memory, a transceiver, and a processor:

[0067] A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations:

[0068] Receive a training delay constraint indication from a network device, the training delay constraint indication being used to indicate the deadline for reporting the training results of the i-th round of the model, where i is a positive integer greater than or equal to 0.

[0069] In a seventh aspect, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the above-described communication method.

[0070] The communication method, apparatus, device, and storage medium provided in this application enable the network to send training delay constraint instructions to the terminal devices participating in federated learning, indicating the reporting deadline for model training results. Therefore, when there are two or more terminal devices participating in federated learning, each terminal device can upload model training results according to the training delay constraint instructions. This reduces the time difference in uploading model training results between different terminal devices participating in federated learning, thereby ensuring the overall training delay of federated learning. It also reduces the possibility of terminal devices reporting too late, helping to reduce the waste of network resources. Attached Figure Description

[0071] Figure 1 This is a schematic diagram illustrating the application scenario of this application;

[0072] Figure 2 A diagram illustrating federated learning;

[0073] Figure 3 A schematic diagram of the communication method provided in the embodiments of this application;

[0074] Figure 4 Another schematic diagram of the communication method provided in the embodiments of this application;

[0075] Figure 5 Another schematic diagram of the communication method provided in the embodiments of this application;

[0076] Figure 6 Another schematic diagram of the communication method provided in the embodiments of this application;

[0077] Figure 7 Another schematic diagram of the communication method provided in the embodiments of this application;

[0078] Figure 8 Another schematic diagram of the communication method provided in the embodiments of this application;

[0079] Figure 9 Another schematic diagram of the communication method provided in the embodiments of this application;

[0080] Figure 10 Another schematic diagram of the communication method provided in the embodiments of this application;

[0081] Figure 11 Another schematic diagram of the communication method provided in the embodiments of this application;

[0082] Figure 12 Another schematic diagram of the communication method provided in the embodiments of this application;

[0083] Figure 13 A schematic diagram of a communication device provided in an embodiment of this application;

[0084] Figure 14 A schematic diagram of a communication device provided in an embodiment of this application;

[0085] Figure 15 This is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0086] Figure 16 This is a schematic diagram of the structure of a communication device provided in an embodiment of this application. Detailed Implementation

[0087] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms "a" and "the" as used in the embodiments of this application are also intended to include the plural forms unless the context clearly indicates otherwise.

[0088] It should be understood that the term "and / or" used in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Furthermore, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0089] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0090] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0091] Figure 1 This is a schematic diagram illustrating the application scenario of this application. Figure 2 A diagram illustrating federated learning, such as Figure 1 as well as Figure 2As shown, terminal devices can communicate with network devices. During federated learning, after initialization, the network device sends resource requests to multiple terminal devices. The terminal devices report resource information to the network device based on the resource requests. After selecting clients based on the resource information, the network device distributes the global model to each terminal device. The terminal devices can update the model distributed by the network device based on their own data, and after completing the update, they upload the model update results to the network device, including model weights, gradient values, and other information. After receiving the model update results from the terminal devices, the network device can send an acknowledgment signal (ACK signal) back to the terminal devices. After the last terminal device uploads its model update results, the network device aggregates the model based on the model update results uploaded by all terminal devices to obtain the updated model.

[0092] The above process can be called the first iteration of federated learning. After receiving the updated model, the network device can redistribute the updated model to each terminal device. The terminal devices repeat the above iterative process, eventually allowing the network device to obtain the second updated model, thus completing the second iteration of federated learning. It can be understood that the number of iterations in federated learning is at least one.

[0093] Federated learning can be divided into synchronous federated learning and asynchronous federated learning.

[0094] Synchronous federated learning refers to a learning process where the training latency in each iteration is determined by the terminal device that last reports the model update results. The network devices can only complete federated aggregation after all necessary training updates have been correctly collected. Therefore, synchronous federated learning requires high latency; significant differences in the upload times of various terminal devices will increase the overall training latency.

[0095] Asynchronous federated learning refers to network devices adaptively setting hybrid weight values ​​to alleviate the synchronization overhead of synchronous federated learning. In other words, network devices do not need to simultaneously receive model update results from all terminal devices before aggregating the models. Although asynchronous federated learning has lower latency requirements, latency is still significant. For example, a terminal device might report gradient information only after several iterations, but the corresponding hybrid weights might be zero. This means the model training result contributes too little to the current iteration, rendering the terminal device's reporting of that gradient information meaningless and wasting transmission resources.

[0096] The communication method provided in this application is intended to solve the above-mentioned technical problems of the prior art.

[0097] The main concept of this application is as follows: During the federated learning process, the network device sends a training delay constraint instruction to the terminal devices participating in the federated learning. Correspondingly, the terminal devices receive the training delay constraint instruction. In the above communication process, the training delay constraint instruction is used to indicate the deadline for reporting the model training results. Thus, the terminal devices participating in the federated learning can upload the model training results according to the reporting deadline indicated by the network device. Therefore, when there are two or more terminal devices participating in the federated learning, it helps to reduce the time difference in uploading the model training results by different terminal devices, thereby ensuring the overall training delay of the federated learning. In addition, it also helps to reduce the possibility of the terminal devices reporting too late and reduce the waste of transmission resources.

[0098] It is understood that the processing steps of the communication method in this application can be implemented by a terminal device or a network device.

[0099] Among them, network equipment can be access network equipment, specifically, it can be a base station (BTS) and / or base station controller in Global System for Mobile communication (GSM) or Code Division Multiple Access (CDMA), or a base station (NodeB, NB) and / or Radio Network Controller (RNC) in Wideband Code Division Multiple Access (WCDMA), or an evolved Node B (eNB or eNodeB) in Long Term Evolution (LTE), or a relay station or access point, or a base station (gNB) in a 5G network, or a base station in a future 6G network, etc.

[0100] In addition, network equipment can also be core network equipment, such as 4G core network equipment (Evolved Packet Core, EPC), 5G core network equipment (5G Core, 5GC), and core network equipment in future 6G networks. This application does not limit the specific type of network equipment.

[0101] Terminal devices can be either wireless or wired. A wireless terminal can be a device that provides voice and / or other data connectivity to a user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem. A wireless terminal can communicate with one or more core network devices via a Radio Access Network (RAN). A wireless terminal can be a mobile terminal, such as a mobile phone (or "cellular" phone) or a computer with a mobile terminal, for example, a portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile device that exchanges voice and / or data with the RAN. Furthermore, a wireless terminal can also be a Personal Communication Service (PCS) phone, a cordless phone, a Session Initiation Protocol (SIP) phone, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA), or other similar devices. A wireless terminal can also be referred to as a system, subscriber unit, subscriber station, mobile station, mobile station, remote station, remote terminal, access terminal, user terminal, user agent, user device, or user equipment; no specific terminology is used here. Optionally, the aforementioned terminal devices can also be smartwatches, tablets, or other similar devices.

[0102] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0103] In some embodiments, a communication method is provided, applied to Figure 1 The application scenarios shown and Figure 2 The federated learning process is shown.

[0104] Take the i-th round of federated learning as an example. Figure 3 The diagram shown is a schematic representation of a communication method provided in an embodiment of this application, which specifically includes the following steps:

[0105] S310, The network device sends a training delay constraint instruction to the terminal device participating in the i-th round of federated learning; correspondingly, the terminal device receives the training delay constraint instruction from the network device.

[0106] In this embodiment, the training latency constraint indicator is used to indicate the reporting deadline for the i-th round of model training results, where i is a positive integer greater than or equal to 0. The reporting deadline indicated by the training latency constraint indicator can be a preset value or determined by the network device based on actual conditions.

[0107] For example, the deadline for reporting the model training results in the i-th round can refer to the time T1 when the terminal device sends the last data packet of the model training results to the network device. That is, the terminal device must send all data packets of the model training results to the network device before time T1.

[0108] For example, the deadline for reporting the model training results in the i-th round can also be the time T2 when the network device receives the last data packet of the model training results. That is, when the terminal device reports the model training results to the network device, it needs to take into account information such as the number of data packets and the data packet transmission time to ensure that the network device can receive all the data packets of the model training results before time T2.

[0109] For example, the deadline for reporting the model training results in the i-th round can also be the time T3 when the terminal device sends the first data packet of the model training results to the network device. That is, the terminal device must start sending the first data packet of the model training results to the network device before time T3.

[0110] For example, the deadline for reporting the model training results in the i-th round can also be the time T4 when the network device receives the first data packet of the model training results. That is, when the terminal device reports the model training results to the network device, it needs to ensure that the network device can receive the first data packet of the model training results before time T4.

[0111] The above is an example of the reporting deadline for the i-th round of model training results. It can be understood that the reporting deadline can also be other types of time nodes, such as: the time T5 when the terminal device starts sending (or finishes sending) a preset number of data packets in the model training results, or the time T6 when the network device starts receiving (or finishes receiving) a preset number of data packets in the model training results, etc. The preset number can be a specific value, such as 10 data packets, or a certain proportion of the total number of data packets, such as 1 / 2, 1 / 3, etc. of the total number of data packets.

[0112] In some embodiments, when there are two or more terminal devices participating in the i-th round of federated learning, the reporting deadlines indicated by the training delay constraint instructions sent by the network device to different terminal devices may be the same or different.

[0113] For example, network devices can determine the reporting deadline for each terminal device based on the device capability information of the terminal devices (such as model training efficiency), and send corresponding training delay constraint instructions to each terminal device according to the determined reporting deadline.

[0114] In this embodiment, the network device can send training latency constraint indications to the terminal device using various different message formats, including but not limited to: sending training latency constraint indications in configuration information or session establishment request messages, sending training latency constraint indications in control protocol data units, and sending them in confirmation information (e.g., ... Figure 2 The ACK signal in the training delay constraint indication is sent, etc.

[0115] Upon receiving a training latency constraint instruction from the network device, each terminal device participating in federated learning can report its model training results to the network device according to the reporting deadline indicated by the instruction. For example, upon receiving the training latency constraint instruction, different terminal devices can adjust their own configuration parameters, such as increasing power or prioritizing training tasks, to meet the time constraint on reporting model training results. This reduces the time difference in model uploads between different terminal devices and lowers the possibility of late reporting.

[0116] In some embodiments, the training latency constraint indicator is used to indicate the deadline for reporting the training results of the i-th round of the model, and can be implemented in the following ways:

[0117] Method 1: The training latency constraint indicator is used to indicate the duration between the receiving time of the i-th round of model training and the reporting deadline. The receiving time of the i-th round of model training refers to the time it takes for the terminal device to receive the model data before model training begins.

[0118] Specifically, the model data received by the terminal device refers to... Figure 2 In the federated learning process shown, the network device distributes model data to the terminal device during the "model distribution" phase. The terminal device can then use this model data to train the model and obtain the model update results.

[0119] For example, the receiving time of the i-th training model can be either the time T7 when the terminal device receives the first data packet of model data for this iteration, or the time T8 when the terminal device receives the last data packet of model data for this iteration. This embodiment does not limit this.

[0120] The training latency constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline. Specifically, the reporting deadline can be a time node after the receiving time with a certain interval, that is: reporting deadline = receiving time of model data + a certain duration, that is, the terminal device takes a time node after a certain duration after the receiving time of model data as the reporting deadline.

[0121] For example, the above duration can be a preset fixed duration value, such as 10 minutes or one hour, or it can be set by the network device according to the actual situation.

[0122] Therefore, by using the training delay constraint indicator to indicate the duration between the receiving time and the reporting deadline of the i-th round of training model, the terminal device can accurately determine the reporting deadline based on the training delay constraint indicator.

[0123] Method 2: The training delay constraint indication is used to indicate the duration between the reception time of the training delay constraint indication and the reporting deadline. The reception time of the training delay constraint indication is the time it takes for the terminal device to receive the training delay constraint indication sent by the network device.

[0124] For example, the reception time of the training delay constraint indication can be the time T9 when the terminal device receives the first data packet for the training delay constraint indication, or the time T10 when the terminal device receives the last data packet for the training delay constraint indication. This embodiment does not limit this.

[0125] In this embodiment, the training delay constraint indication is used to indicate the duration between the receiving time of the training delay constraint indication and the reporting deadline. Specifically, the reporting deadline can be a time node that is a certain duration after the receiving time, that is: the reporting deadline = the receiving time of the training delay constraint indication + a certain duration, that is, the terminal device takes a certain duration after the receiving time of the training delay constraint indication as the reporting deadline.

[0126] Optionally, the above duration can be a preset fixed duration value, such as 10 minutes or one hour, or it can be set in real time by the network device according to the actual situation.

[0127] Therefore, by using the training delay constraint indication to indicate the duration between the receiving time of the training delay constraint indication and the reporting deadline, the terminal device can accurately determine the reporting deadline based on the training delay constraint indication.

[0128] As an optional embodiment, the training latency constraint indication sent by the network device to the terminal device can also be a specific time node, such as 10:00 AM or 1:00 PM.

[0129] In some embodiments, the training latency constraint indication sent by the network device to the terminal device may be applied only to one iteration process. That is, the network device sends a training latency constraint indication to the terminal device in each iteration, and the terminal device uses the training latency constraint indication issued by the network device in that iteration as the standard in each iteration.

[0130] Alternatively, in some embodiments, the training latency constraint indication sent by the network device to the terminal device can be applied to all subsequent iterations. That is, the network device may send a training latency constraint indication to the terminal device once in the i-th iteration (where i is a positive integer greater than or equal to 0), and the terminal device will use this training latency constraint indication in all subsequent iterations. For example, in the first round of federated learning, the network device sends a training constraint indication to the terminal device. In the next round of federated learning, the network device will not send a training constraint indication again. The terminal device can calculate the reporting deadline for the model training results of the next round of federated learning based on the training constraint indication sent in the first round. Then, the terminal device reports the model training results for that round based on the calculated reporting deadline.

[0131] Alternatively, the training latency constraint indication sent by the network device to the terminal device can also be applied to subsequent iterations. That is, the network device can first send a training latency constraint indication to the terminal device in the i-th iteration (i is a positive integer greater than or equal to 0), and the terminal device can use this training latency constraint indication in subsequent iterations. After completing multiple iterations, if the network device sends a new training latency constraint indication to the terminal device, the terminal device can use the newly issued training latency constraint indication from the network device in subsequent iterations, and so on.

[0132] Furthermore, in some embodiments, when a terminal device receives a training latency constraint instruction from a network device, it can determine whether it supports the training latency constraint instruction based on the reporting deadline indicated by the training latency constraint instruction and factors such as the efficiency of the terminal device in training its own model. That is, whether the time for the terminal device to report the model training results can meet the reporting deadline indicated by the training latency constraint instruction.

[0133] Specifically, Figure 4Another schematic diagram of the communication method provided in the embodiments of this application is shown below. Figure 4 As shown, it specifically includes:

[0134] S410. If the terminal device determines that it does not support the training delay constraint indication, it sends a rejection response message to the network device. Correspondingly, the network device receives rejection response messages from N terminal devices participating in the i-th round of federated learning. The rejection response message is used to indicate that the training delay constraint indication is not supported, and N is a positive integer greater than or equal to 1.

[0135] In some other embodiments of this application, if the terminal device determines that it supports the training delay constraint instruction, it may send a receive response message to the network device, or it may not send a response message to the network device, that is, it can report the model training results to the network device normally according to the reporting deadline indicated by the training delay constraint instruction.

[0136] Therefore, in scenarios where terminal devices do not support training latency constraint instructions, by sending a rejection response message to the network device, the network device can be informed of the specific situation of the terminal device in a timely manner. This allows the network device to execute corresponding processing strategies to deal with the situation where the terminal device does not support training latency constraint instructions, thus preventing the terminal device from affecting the overall training process of federated learning.

[0137] In some embodiments, when a network device receives a rejection response message from a terminal device, it may adopt certain processing strategies to prevent terminal devices that do not support training latency constraint instructions from affecting the overall training process of federated learning.

[0138] Specifically, let's take an example where N terminal devices participating in the i-th round of federated learning send rejection response messages to the network device. For instance... Figure 5 The diagram shown is another schematic representation of the communication method provided in this application embodiment, specifically including:

[0139] S510. The network device sends a notification message to the N terminal devices to cancel their participation in the i-th round of federated learning. Correspondingly, the terminal devices receive the notification message from the network device to cancel their participation in the i-th round of federated learning.

[0140] In this embodiment, if the network device receives rejection response messages from N terminal devices participating in the i-th round of federated learning, the network device determines that the N terminal devices do not support the training delay constraint indication, that is, the N terminal devices cannot report the model training results to the network device before the reporting deadline indicated by the training delay constraint indication. Therefore, the network device can send a notification message to the N terminal devices to notify them to cancel their participation in the i-th round of federated learning.

[0141] After receiving a notification message from the network device that it has cancelled its participation in the i-th round of federated learning, the terminal device may choose not to update the model in this iteration, that is, it may choose not to participate in the model update of the i-th round of federated learning.

[0142] Therefore, after receiving a rejection response message from a terminal device, the network device can notify the terminal device to cancel its participation in this round of federated learning, thus avoiding the impact of terminal devices that do not support training latency constraint instructions on the overall training process of federated learning.

[0143] In some embodiments, when a terminal device receives a training latency constraint instruction from a network device, if it determines that it does not support the training latency constraint instruction, it can further estimate the time for uploading the model training results and send the estimated upload time to the network device.

[0144] The terminal device may send the estimated upload time and the rejection response message to the network device at the same time, or it may send the estimated upload time and the rejection response message to the network device separately. This embodiment does not limit this.

[0145] After receiving the estimated upload time from the terminal device, the network device can compare the estimated upload time with the corresponding upload deadline and determine further processing strategies based on the time difference between the two.

[0146] For example, if the time difference between the two is within the preset time range, it indicates that the estimated upload time exceeds the upload deadline by a small margin, and the impact of the terminal device on the overall training latency is also small. Therefore, the network device can send a notification message to the terminal device to continue participating in the i-th round of federated learning; after receiving the notification message to continue participating in the i-th round of federated learning, the terminal device performs model training normally and reports the model training results to the network device.

[0147] For example, if the time difference between the two exceeds the preset time range, it indicates that the estimated upload time exceeds the upload deadline by a significant margin, and the terminal device has a greater impact on the overall training latency. Therefore, the network device can send a notification message to the terminal device to cancel its participation in the i-th round of federated learning. After receiving the notification message from the network device, the terminal device can choose not to update the model for this iteration.

[0148] In the above process, the preset time range can be a preset value or it can be determined by the network device according to the actual situation.

[0149] In some embodiments, since the model training results reported by the terminal device are model update information, the data volume may be large, for example, the model update information may have 100 data packets. However, the model training results can only be used by the network device when all 100 data packets are received. If at least one data packet is lost, the model training results will be unusable, causing the terminal device's reporting behavior to fail and wasting network resources. Based on this, the terminal device can send relevant indication information to the network device to avoid the situation where not all data packets are received completely.

[0150] For example, such as Figure 6 The diagram shown is another schematic representation of the communication method provided in this application embodiment, specifically including:

[0151] S610, The terminal device sends a relaxation instruction message to the network device; correspondingly, the network device receives the relaxation instruction message sent by the terminal device participating in the i-th round of federated learning.

[0152] In this embodiment, the relaxation instruction information is used to indicate that the packet dropping conditions for the model training results are relaxed, and / or that all packets of the model training results are useful. Upon receiving the relaxation instruction information, the network device can relax the packet dropping conditions for the model training results, for example, by adjusting the discard timer, and / or, based on the relaxation instruction information, know that all packets of the model training results are useful, meaning that all packets of the model training results uploaded by the terminal device are considered as a whole, and the network device does not perform packet dropping processing.

[0153] For example, the terminal device may send a relaxation instruction to the network device before reporting the model training results; or it may send the relaxation instruction to the network device when reporting the model training results. This embodiment does not limit this.

[0154] Thus, by sending a relaxation instruction to the network device, the terminal device enables the network device to fully receive all data packets of the model training results, ensuring the effectiveness of the terminal device's reporting behavior and avoiding the waste of network resources.

[0155] In some embodiments, the terminal device may not send relaxation instruction information, but instead directly send network parameter configuration information to the network device, such as the parameter configuration of the discard timer. In this way, the network device can make settings according to the configuration information sent by the terminal device, thereby relaxing the data packet discarding conditions of the model training results.

[0156] In some embodiments, when the i-th round of federated learning is the first iteration (i.e., i=1), the network device may carry a training delay constraint indication during the process of sending a Radio Resource Control (RRC) message to the terminal device.

[0157] Specifically, taking sending a training latency constraint indication in the configuration information or session establishment request message as an example, such as... Figure 7 The diagram shown is another schematic representation of the communication method provided in this application embodiment, specifically including:

[0158] S710: The network device sends configuration information or a session establishment request message to the terminal devices participating in the i-th round of federated learning. The configuration information includes a training latency constraint indication, and the session establishment request message includes a training latency constraint indication. Correspondingly, the terminal devices receive the configuration information or session establishment request message from the network device.

[0159] In the first iteration (i.e. the first round of federated learning), the network device sends an RRC message to the terminal device, which may be configuration information or a session establishment request message. The session establishment request message may be a Data Radio Bearer (DRB) establishment request message or a Multicast Broadcast Service Radio Bearer (MRB) establishment request message.

[0160] Therefore, after receiving configuration information, DRB establishment request, or MRB establishment request, the terminal device receives the overall training model sent by the network device in the local iteration, and then performs subsequent model training processing. After obtaining the model training result, it reports the model training result to the network device according to the training latency constraint indication in the configuration information or session establishment request message, thereby ensuring the overall training latency.

[0161] In some embodiments, the network device may also send a training delay constraint instruction to the terminal device via a Control Protocol Data Unit (Control PDU).

[0162] Specifically, taking the transmission of training delay constraint indications in the control protocol data unit as an example, such as... Figure 8 The diagram shown is another schematic representation of the communication method provided in this application embodiment, specifically including:

[0163] S810, the network device sends a control protocol data unit to the terminal device participating in the i-th round of federated learning. The control protocol data unit includes a training latency constraint indication. Correspondingly, the terminal device receives the control protocol data unit from the network device.

[0164] In this scenario, network devices can send training delay constraint indications to terminal devices without using existing control protocol data units (CRTs). Instead, they can use a newly added type of CRT to do so. In this case, the training delay constraint indication can contain complete time indication information. Upon receiving the CRT from the network device, the terminal device can directly extract the training delay constraint indication from the CRT to determine the reporting deadline.

[0165] Alternatively, network devices can use existing control protocol data units (CRTs), adding training delay constraint indications to the reserved bits of existing CRTs. In this case, the training delay constraint indication can be referential information containing time indication information. After receiving the CRT sent by the network device, the terminal device can extract the referential information of the time indication from the CRT and then determine the reporting deadline based on the referential information.

[0166] For example, by adding a training delay constraint indication to the reserved bits of the existing control protocol data unit, the network device can send a configuration table to the terminal device. This configuration table contains the correspondence between different time indication information and different reference information. Thus, the terminal device can determine the corresponding time indication information based on the configuration table and the specified information sent by the network device, thereby obtaining the corresponding reporting deadline.

[0167] For example, if 2 bits are reserved, the values ​​00, 01, 10, and 11 can represent four different time indication information.

[0168] Optionally, the network device may send the control protocol data unit to the terminal device before sending the model data, or it may send the control protocol data unit to the terminal device when sending the model data. This embodiment does not limit this.

[0169] In some embodiments, the network device may instruct the terminal device to speed up data packet transmission to increase the speed at which the network device receives model training results, thereby helping to improve the overall efficiency of federated learning.

[0170] Specifically, such as Figure 9 The diagram shown is another schematic representation of the communication method provided in this application embodiment, specifically including:

[0171] S910, the network device sends a data packet acceleration instruction to K terminal devices participating in the i-th round of federated learning. The data packet acceleration instruction is also used to indicate the accelerated transmission of the data packets to be sent for the model training results of the i-th round, where K is a positive integer greater than or equal to 1. Correspondingly, the terminal devices receive the data packet acceleration instruction from the network device.

[0172] In this embodiment, the network device can send both the data packet acceleration instruction and the training latency constraint instruction to the terminal device at the same time, or it can send both the data packet acceleration instruction and the training latency constraint instruction to the terminal device separately. This embodiment does not limit the specific method of sending both methods.

[0173] Furthermore, network devices can use the same message to send data packets to speed up transmission indications and training latency constraint indications, or they can use different types of message to send data packets to speed up transmission indications and training latency constraint indications. This embodiment does not limit this.

[0174] After receiving the instruction to speed up data packet transmission, the terminal device accelerates the transmission speed of the remaining data packets to ensure the overall training latency of federated learning.

[0175] In some embodiments, except for the first iteration, the network device may indicate the reporting deadline to the terminal device during the process of sending confirmation information to the terminal device.

[0176] Specifically, taking sending a training delay constraint instruction in the confirmation message as an example, such as... Figure 10 The diagram shown is another schematic representation of the communication method provided in this application embodiment, specifically including:

[0177] S1010: The network device sends an acknowledgment message for the data packet containing the model training results of the (i-1)th round to the terminal devices participating in the i-th round of federated learning. The acknowledgment message includes a training latency constraint indication. The terminal devices participating in the i-th round of federated learning are the terminal devices among those participating in the (i-1)th round of federated learning, where i is a positive integer greater than or equal to 1. Correspondingly, the terminal devices receive the acknowledgment message for the data packet containing the model training results of the (i-1)th round from the network device.

[0178] The acknowledgment information can be an ACK signal. For example, refer to... Figure 2 as well as Figure 10 After receiving the model training results of the (i-1)th round reported by the terminal device, the network device will send an ACK signal back to the terminal device. Therefore, the network device can send an ACK signal carrying a training delay constraint to the terminal device. This training delay constraint is used to indicate the reporting deadline of the terminal device in the next round (i.e., the i-th round) iteration.

[0179] In some embodiments, network devices may introduce a recording mechanism to record the number of times the actual reporting time of a terminal device exceeds the reporting deadline.

[0180] Specifically, such as Figure 11 The diagram shown is another schematic representation of the communication method provided in this application embodiment, specifically including:

[0181] S1110. The network device receives data packets of model training results from terminal devices participating in the i-th round of federated learning;

[0182] S1120. If the difference between the arrival time of the model training result of the first terminal device among the terminal devices participating in the i-th round of federated learning and the reporting deadline is greater than or equal to the first threshold, then the network device will increment the timeout reporting count of the first terminal device by 1 in the data packet timeout reporting record table.

[0183] Wherein, if the reporting deadline is used to indicate the deadline for the terminal device to send the first data packet of the model training result, or to indicate the deadline for the network device to receive the first data packet of the model training result, then the "arrival time" in step S1120 can refer to the time when the network device receives the first data packet in the model training result. That is, the network device compares the time when it receives the first data packet in the model training result with the reporting deadline to obtain the difference between the two, and compares it with the first threshold.

[0184] Alternatively, if the reporting deadline is used to indicate the deadline for the terminal device to send the last data packet of the model training results, or to indicate the deadline for the network device to receive the last data packet of the model training results, then the "arrival time" in step S1120 can refer to the time when the network device receives the last data packet of the model training results. That is, the network device compares the time when it receives the last data packet of the model training results with the reporting deadline to obtain the difference between the two, and compares it with the first threshold.

[0185] Optionally, embodiments of this application may also introduce a penalty mechanism, such as... Figure 11 The method shown may also include:

[0186] S1130. If the number of timeout reports by the first terminal device exceeds the second threshold, the network device will restrict or cancel the first terminal device from continuing to participate in the i-th round of federated learning.

[0187] Specifically, if the number of timeout reports exceeds the second threshold, the network device can restrict or cancel the first terminal device from continuing to participate in the i-th round of federated learning. That is, the first terminal device cannot participate in this round of iteration, or cannot participate in all subsequent iterations.

[0188] Optionally, if the network device only restricts the first terminal device from participating in the current iteration, then the first terminal device can also participate in the subsequent (i+1)th iteration. In this case, the network device continues to count the number of timeout reports from the first terminal device. If the number of timeout reports exceeds the third threshold, the network device restricts the first terminal device from participating in all subsequent iterations, that is, permanently removes the first terminal device from the list of terminal devices participating in federated learning.

[0189] Optionally, the network device may send a notification message to the first terminal device to notify the first terminal device to cancel its participation in the current iteration or to cancel its participation in all subsequent iterations.

[0190] Therefore, by recording the number of timeouts reported by terminal devices, network devices can help optimize the list of terminal devices participating in federated learning and ensure the overall training latency of federated learning.

[0191] In some embodiments, in order to facilitate the maintenance of the list of terminal devices participating in federated learning by the network device, the terminal device may report its own device capabilities to the network device.

[0192] Specifically, Figure 12 Another schematic diagram of the communication method provided in the embodiments of this application is shown below. Figure 12 As shown, the method includes:

[0193] S1210, the network device sends a device capability condition indication to the terminal devices participating in the i-th round of federated learning. The device capability condition indication is used to indicate the device's computing power processing capability threshold. Correspondingly, the terminal devices receive the device capability condition indication from the network device.

[0194] Among them, the device computing power processing capability threshold, such as the computing power threshold of the terminal device, and / or the computing power change rate threshold, etc., so that after receiving the device capability condition indication sent by the network device, the terminal device can determine whether the device itself meets the corresponding device computing power processing capability threshold according to the device capability condition indication.

[0195] Optionally, after determining whether it meets the device capability condition indication, the terminal device can provide feedback to the network device based on the determination result.

[0196] For details, please refer to Figure 12 The methods also include:

[0197] S1230. If a terminal device determines that it does not meet the device capability condition indication, it sends device capability information to the network device. Correspondingly, the network device receives device capability information from M terminal devices participating in the i-th round of federated learning, where M is a positive integer greater than or equal to 1.

[0198] If a terminal device determines that it does not meet the device capability condition indication, such as its own computing power not meeting the computing power threshold in the device capability condition indication, and / or its own computing power change rate not meeting the computing power change rate threshold in the device capability condition indication, then the terminal device will report its own device capability information to the network device, such as real-time computing power load and / or real-time computing power change rate.

[0199] Optional, see reference Figure 12 The methods also include:

[0200] S1250. The network device updates the candidate list of terminal devices participating in federated learning based on the device capability information of the terminal devices. The candidate list includes the identifier of at least one terminal device.

[0201] After receiving the device capability information reported by the terminal device, the network device maintains the list of terminal devices participating in federated learning based on the device capability information. For example, it determines the list of terminal devices participating in the (i+1)th round of federated learning, or removes terminal devices with device capabilities lower than the preset value from the list of terminal devices participating in federated learning, so as to ensure the accuracy of federated learning and the overall training latency.

[0202] In some embodiments, considering that the location of the terminal device may change, network devices can transmit device capability information of the terminal device to each other.

[0203] Specifically, the method also includes: when a network device detects that a second terminal device among M terminal devices has undergone cell handover, it sends the device capability information of the second terminal device to the network device that will serve the second terminal device after the cell handover.

[0204] When the network device is a base station, after the source base station detects that the second terminal device has undergone cell handover, it can send the device capability information of the second terminal device to the target base station that will serve the second terminal device after the cell handover through the Xn interface; or send it to the AMF (Access and Mobility Management Function) through the NG interface, and then the AMF will send it to the target base station through the NG interface.

[0205] In addition, when the network device is a core network device, if the core network device connected to the terminal device also changes during cell handover, the source core network device can send the device capability information of the second terminal device to the target core network device used to serve the second terminal device after cell handover through the core network internal interface (such as N14).

[0206] Optionally, when the network device is a core network device, if the core network device connected to the terminal device also changes during cell handover, the source core network device can send the device capability information of the second terminal device to the source base station through the NG interface. The source base station then sends the device capability information of the second terminal device to the target base station that will serve the second terminal device after cell handover through the Xn interface. Finally, the target base station sends the information to the target core network device through the NG interface.

[0207] Optionally, when the network device is a core network device, if the core network device connected to the terminal device also changes during cell handover, the source core network device can send the device capability information of the second terminal device to OAM (Operation Administration and Maintenance), and OAM will then send the device capability information of the second terminal device to the target core network device.

[0208] Therefore, when the location of a terminal device changes, the source network device can send the stored terminal device capability information to the target network device, thereby facilitating the target network device to adjust its network configuration and appropriately call or page the terminal device.

[0209] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0210] The above embodiments can be used individually or in combination to achieve different technical effects.

[0211] In some embodiments, a communication device is provided, which is applied to a network device.

[0212] Figure 13 A schematic diagram of the communication device provided in the embodiments of this application, such as... Figure 13 As shown, the device includes:

[0213] The sending module 131 is used to send a training delay constraint indication to the terminal device participating in the i-th round of federated learning. The training delay constraint indication is used to indicate the deadline for reporting the training results of the i-th round of model, where i is a positive integer greater than or equal to 0.

[0214] In some embodiments, the training delay constraint indicator is used to indicate the reporting deadline for the i-th round of model training results, including: the training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

[0215] In some embodiments, the system further includes a receiving module, configured to receive rejection response messages from N terminal devices participating in the i-th round of federated learning, wherein the rejection response messages indicate that the training latency constraint indication is not supported, and N is a positive integer greater than or equal to 1.

[0216] In some embodiments, the sending module is further configured to: send a notification message to N terminal devices to cancel their participation in the i-th round of federated learning.

[0217] In some embodiments, the receiving module is further configured to: receive relaxation instruction information sent from the terminal device participating in the i-th round of federated learning, wherein the relaxation instruction information is used to indicate the relaxation of the data packet dropping conditions of the model training results and / or the fact that all data packets of the model training results are useful.

[0218] In some embodiments, sending a training latency constraint indication to a terminal device participating in the i-th round of federated learning includes: sending configuration information or a session establishment request message to the terminal device participating in the i-th round of federated learning, wherein the configuration information includes a training latency constraint indication, and the session establishment request message includes a training latency constraint indication.

[0219] In some embodiments, sending a training latency constraint indication to a terminal device participating in the i-th round of federated learning includes: sending a control protocol data unit to the terminal device participating in the i-th round of federated learning, wherein the control protocol data unit includes the training latency constraint indication.

[0220] In some embodiments, the sending module is further configured to: send a data packet accelerated transmission instruction to K terminal devices among the terminal devices participating in the i-th round of federated learning, the data packet accelerated transmission instruction being further configured to indicate accelerated transmission of the data packet to be sent for the model training results of the i-th round, where K is a positive integer greater than or equal to 1.

[0221] In some embodiments, sending a training latency constraint instruction to a terminal device participating in the i-th round of federated learning includes: sending confirmation information of a data packet containing the model training results of the (i-1)-th round to the terminal device participating in the i-th round of federated learning. The confirmation information includes the training latency constraint instruction. The terminal device participating in the i-th round of federated learning is a terminal device among the terminal devices participating in the (i-1)-th round of federated learning, and i is a positive integer greater than or equal to 1.

[0222] In some embodiments, the receiving module is further configured to: receive data packets of model training results from terminal devices participating in the i-th round of federated learning; if the difference between the arrival time of the model training result of the first terminal device among the terminal devices participating in the i-th round of federated learning and the reporting deadline is greater than or equal to a first threshold, increment the timeout reporting count of the first terminal device by 1 in the data packet timeout reporting record table.

[0223] In some embodiments, the system further includes a processing module, configured to restrict or cancel the first terminal device from continuing to participate in the i-th round of federated learning if the number of timeout reports by the first terminal device exceeds a second threshold.

[0224] In some embodiments, the sending module is further configured to: send a device capability condition indication to the terminal device participating in the i-th round of federated learning, wherein the device capability condition indication is used to indicate the device's computing power processing capability threshold.

[0225] In some embodiments, the receiving module is further configured to: receive device capability information of M terminal devices among the terminal devices participating in the i-th round of federated learning, where M is a positive integer greater than or equal to 1.

[0226] In some embodiments, the processing module is further configured to: update the candidate list of terminal devices participating in federated learning based on the device capability information of the terminal devices, wherein the candidate list includes the identifier of at least one terminal device.

[0227] In some embodiments, the sending module is further configured to: detect that a second terminal device among M terminal devices has undergone cell handover, and send the device capability information of the second terminal device to the network device used to serve the second terminal device after the cell handover.

[0228] In some embodiments, a communication device is provided for use in a terminal device.

[0229] Figure 14 A schematic diagram of the communication device provided in the embodiments of this application, such as... Figure 14 As shown, the device includes:

[0230] The receiving module 141 is used to receive a training delay constraint indication from the network device, wherein the training delay constraint indication is used to indicate the reporting deadline of the i-th round of model training results, and i is a positive integer greater than or equal to 0.

[0231] In some embodiments, the training delay constraint indicator is used to indicate the reporting deadline for the i-th round of model training results, including: the training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

[0232] In some embodiments, the device further includes a sending module, configured to send a rejection response message to the network device if it is determined that the training delay constraint indication is not supported.

[0233] In some embodiments, the receiving module is further configured to: receive a notification message from the network device indicating that it has cancelled its participation in the i-th round of federated learning.

[0234] In some embodiments, the sending module is further configured to: send relaxation indication information to the network device, the relaxation indication information being used to indicate the relaxation of the packet dropping conditions for the model training results, and / or that all packets of the model training results are useful.

[0235] In some embodiments, receiving a training latency constraint indication from a network device includes: receiving configuration information or a session establishment request message from the network device, wherein the configuration information includes a training latency constraint indication, and the session establishment request message includes a training latency constraint indication.

[0236] In some embodiments, receiving a training latency constraint indication from a network device includes: receiving a control protocol data unit from the network device, the control protocol data unit including the training latency constraint indication.

[0237] In some embodiments, the receiving module is further configured to: receive a data packet accelerated transmission indication from a network device, the data packet accelerated transmission indication being used to indicate accelerated transmission of the data packets to be sent for the model training results of the i-th round.

[0238] In some embodiments, receiving a training latency constraint indication from a network device includes: receiving confirmation information of a data packet containing the model training results of the (i-1)th round from the network device, wherein the confirmation information includes a training latency constraint indication, and i is a positive integer greater than or equal to 1.

[0239] In some embodiments, the receiving module is further configured to: receive a device capability condition indication from a network device, the device capability condition indication being used to indicate a device computing power processing capability threshold.

[0240] In some embodiments, the sending module is further configured to: if it is determined that the device capability condition indication is not met, send device capability information to the network device.

[0241] Specific limitations regarding the communication device can be found in the limitations regarding the communication method above, and will not be repeated here. Each module in the aforementioned communication device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the operations corresponding to each module.

[0242] In some embodiments, a communication device is provided.

[0243] Figure 15 This is a schematic diagram of the structure of the communication device provided in the embodiments of this application, such as... Figure 15 As shown, the communication device includes a memory 151, a transceiver 152, and a processor 153.

[0244] The memory is used to store computer programs; the transceiver is used to send and receive data under the control of the processor; the processor is used to read the computer programs in the memory and perform the following operations: send a training delay constraint indication to the terminal devices participating in the i-th round of federated learning, wherein the training delay constraint indication is used to indicate the reporting deadline of the model training results in the i-th round, and i is a positive integer greater than or equal to 0.

[0245] In some embodiments, the training latency constraint indicator is used to indicate the deadline for reporting the training results of the i-th round of the model, including:

[0246] The training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

[0247] In some embodiments, the operation further includes:

[0248] Receive rejection response messages from N terminal devices participating in the i-th round of federated learning. The rejection response messages are used to indicate that the training latency constraint indication is not supported, where N is a positive integer greater than or equal to 1.

[0249] In some embodiments, the operation further includes:

[0250] Send notification messages to N terminal devices to cancel their participation in the i-th round of federated learning.

[0251] In some embodiments, the operation further includes:

[0252] Receive relaxation instruction information from the terminal device participating in the i-th round of federated learning. The relaxation instruction information is used to indicate the conditions for dropping data packets of model training results and / or that all data packets of model training results are useful.

[0253] In some embodiments, sending a training latency constraint instruction to the terminal device participating in the i-th round of federated learning includes:

[0254] Send configuration information or session establishment request messages to the terminal devices participating in the i-th round of federated learning. The configuration information includes training latency constraint indications, and the session establishment request messages include training latency constraint indications.

[0255] In some embodiments, sending a training latency constraint instruction to the terminal device participating in the i-th round of federated learning includes:

[0256] Send control protocol data units to the terminal devices participating in the i-th round of federated learning. The control protocol data units include training latency constraint indications.

[0257] In some embodiments, the operation further includes:

[0258] Send data packet acceleration instructions to K terminal devices among the terminal devices participating in the i-th round of federated learning. The data packet acceleration instructions are also used to indicate the acceleration of the data packet to be sent for the model training results of the i-th round, where K is a positive integer greater than or equal to 1.

[0259] In some embodiments, sending a training latency constraint instruction to the terminal device participating in the i-th round of federated learning includes:

[0260] The confirmation information is sent to the terminal devices participating in the i-th round of federated learning, which includes a training latency constraint indication. The terminal devices participating in the i-th round of federated learning are the terminal devices among the terminal devices participating in the i-1 round of federated learning, and i is a positive integer greater than or equal to 1.

[0261] In some embodiments, the operation further includes:

[0262] Receive data packets containing the model training results from the terminal devices participating in the i-th round of federated learning;

[0263] If the difference between the arrival time of the model training result of the first terminal device among the terminal devices participating in the i-th round of federated learning and the reporting deadline is greater than or equal to the first threshold, the timeout reporting count of the first terminal device will be incremented by 1 in the data packet timeout reporting record table.

[0264] In some embodiments, the operation further includes:

[0265] If the number of timeout reports by the first terminal device exceeds the second threshold, the first terminal device will be restricted or canceled from continuing to participate in the i-th round of federated learning.

[0266] In some embodiments, the operation further includes:

[0267] Send a device capability condition indication to the terminal devices participating in the i-th round of federated learning. The device capability condition indication is used to indicate the device's computing power processing capability threshold.

[0268] In some embodiments, the operation further includes:

[0269] Receive device capability information from M terminal devices participating in the i-th round of federated learning, where M is a positive integer greater than or equal to 1.

[0270] In some embodiments, the operation further includes:

[0271] Based on the device capability information of the terminal devices, update the candidate list of terminal devices participating in federated learning. The candidate list includes the identifier of at least one terminal device.

[0272] In some embodiments, the operation further includes:

[0273] If a cell handover is detected in the second terminal device among M terminal devices, the device capability information of the second terminal device is sent to the network device that will serve the second terminal device after the cell handover.

[0274] In some embodiments, a communication device is provided.

[0275] Figure 16 This is a schematic diagram of the structure of the communication device provided in the embodiments of this application, such as... Figure 16 As shown, the communication device includes a memory 161, a transceiver 162, and a processor 163.

[0276] The memory is used to store computer programs; the transceiver is used to send and receive data under the control of the processor; the processor is used to read the computer programs in the memory and perform the following operations: receiving a training delay constraint indication from a network device, the training delay constraint indication indicating the reporting deadline for the i-th round of model training results, where i is a positive integer greater than or equal to 0.

[0277] In some embodiments, the training latency constraint indicator is used to indicate the deadline for reporting the training results of the i-th round of the model, including:

[0278] The training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

[0279] In some embodiments, the operation further includes:

[0280] If it is determined that the training latency constraint indication is not supported, a rejection response message is sent to the network device.

[0281] In some embodiments, the operation further includes:

[0282] Receive a notification message from a network device that it has cancelled its participation in round i of federated learning.

[0283] In some embodiments, the operation further includes:

[0284] Send a relaxation instruction to the network device. The relaxation instruction is used to indicate the conditions for dropping data packets of the model training results and / or that all data packets of the model training results are useful.

[0285] In some embodiments, receiving a training latency constraint indication from a network device includes:

[0286] Receive configuration information or session establishment request messages from network devices. The configuration information includes training latency constraint indications, and the session establishment request messages include training latency constraint indications.

[0287] In some embodiments, receiving a training latency constraint indication from a network device includes:

[0288] Receive control protocol data units from network devices, the control protocol data units including training delay constraint indications.

[0289] In some embodiments, the operation further includes:

[0290] Receive a packet acceleration instruction from the network device. The packet acceleration instruction is used to indicate that the packet to be sent for the model training results of the i-th round should be accelerated.

[0291] In some embodiments, receiving a training latency constraint indication from a network device includes:

[0292] The system receives an acknowledgment message for the data packet containing the model training results of the (i-1)th round from the network device. The acknowledgment message includes a training latency constraint indication, where i is a positive integer greater than or equal to 1.

[0293] In some embodiments, the operation further includes:

[0294] Receive device capability condition indication from network devices, which indicates the device's computing power processing capacity threshold.

[0295] In some embodiments, the operation further includes:

[0296] If it is determined that the device capability condition indication is not met, then device capability information is sent to the network device.

[0297] In the aforementioned communication devices, the memory and processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines, such as a bus connection. The memory stores computer-executable instructions that implement data access control methods, including at least one software functional module that can be stored in the memory in the form of software or firmware. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory.

[0298] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), and Electrically Erasable Programmable Read-Only Memory (EEPROM). The memory stores programs, which the processor executes upon receiving execution instructions. Furthermore, the software programs and modules within the memory may include an operating system, which can include various software components and / or drivers for managing system tasks (e.g., memory management, storage device control, power management), and can communicate with various hardware or software components to provide an operating environment for other software components.

[0299] The processor can be an integrated circuit chip with signal processing capabilities. The aforementioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0300] In some embodiments, a computer-readable storage medium is provided, which stores computer-executable instructions that, when executed by a processor, are used to implement the steps of various method embodiments of the present application.

[0301] In some embodiments, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of various method embodiments of the present application.

[0302] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0303] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0304] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A communication method, characterized in that, The method includes: Send a training delay constraint indication to the terminal device participating in the i-th round of federated learning. The training delay constraint indication is used to indicate the deadline for reporting the training results of the i-th round of model, where i is a positive integer greater than or equal to 0. If the terminal device does not support the training latency constraint indication, it also receives rejection response messages from N terminal devices participating in the i-th round of federated learning. The rejection response messages are used to indicate that the training latency constraint indication is not supported; N is a positive integer greater than or equal to 1. Send a notification message to the N terminal devices to cancel their participation in the i-th round of federated learning.

2. The method according to claim 1, characterized in that, The training delay constraint indicator is used to indicate the deadline for reporting the training results of the i-th round of the model, including: The training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

3. The method according to claim 1, characterized in that, The method further includes: Receive relaxation instruction information from the terminal device participating in the i-th round of federated learning. The relaxation instruction information is used to indicate the relaxation of the data packet dropping conditions of the model training results and / or the fact that all data packets of the model training results are useful.

4. The method according to claim 1, characterized in that, Sending training latency constraint instructions to the terminal devices participating in the i-th round of federated learning includes: Send configuration information or session establishment request messages to the terminal devices participating in the i-th round of federated learning. The configuration information includes the training latency constraint indication, and the session establishment request message includes the training latency constraint indication.

5. The method according to claim 1, characterized in that, Sending training latency constraint instructions to the terminal devices participating in the i-th round of federated learning includes: A control protocol data unit is sent to the terminal device participating in the i-th round of federated learning. The control protocol data unit includes the training delay constraint indication.

6. The method according to claim 1, characterized in that, The method further includes: Send a data packet acceleration instruction to K terminal devices among the terminal devices participating in the i-th round of federated learning. The data packet acceleration instruction is also used to indicate that the data packets to be sent for the model training results of the i-th round should be accelerated. K is a positive integer greater than or equal to 1.

7. The method according to claim 1 or 6, characterized in that, Sending training latency constraint instructions to the terminal devices participating in the i-th round of federated learning includes: A confirmation message is sent to the terminal device participating in the i-th round of federated learning to send the data packet containing the model training results of the (i-1)th round. The confirmation message includes the training delay constraint indication. The terminal device participating in the i-th round of federated learning is one of the terminal devices participating in the (i-1)th round of federated learning, and i is a positive integer greater than or equal to 1.

8. The method according to claim 1, characterized in that, The method further includes: Receive data packets containing the model training results from the terminal devices participating in the i-th round of federated learning; If the difference between the arrival time of the model training result of the first terminal device among the terminal devices participating in the i-th round of federated learning and the reporting deadline is greater than or equal to the first threshold, the timeout reporting count of the first terminal device is incremented by 1 in the data packet timeout reporting record table.

9. The method according to claim 8, characterized in that, The method further includes: If the number of timeout reports by the first terminal device exceeds the second threshold, the first terminal device is restricted or canceled from continuing to participate in the i-th round of federated learning.

10. The method according to claim 1, characterized in that, The method further includes: Send a device capability condition indication to the terminal device participating in the i-th round of federated learning. The device capability condition indication is used to indicate the device's computing power processing capability threshold.

11. The method according to claim 10, characterized in that, The method further includes: Receive device capability information of M terminal devices among the terminal devices participating in the i-th round of federated learning, where M is a positive integer greater than or equal to 1.

12. The method according to claim 11, characterized in that, The method further includes: Based on the device capability information of the terminal devices, update the candidate list of terminal devices participating in federated learning, wherein the candidate list includes the identifier of at least one terminal device.

13. The method according to claim 11 or 12, characterized in that, The method further includes: If a cell handover is detected in the second terminal device among the M terminal devices, the device capability information of the second terminal device is sent to the network device that will serve the second terminal device after the cell handover.

14. A communication method, characterized in that, The method includes: N terminal devices participating in the i-th round of federated learning receive a training delay constraint instruction from the network device. The training delay constraint instruction is used to indicate the deadline for reporting the model training results of the i-th round, where N is a positive integer greater than or equal to 1 and i is a positive integer greater than or equal to 0. If it is determined that the training latency constraint indication is not supported, a rejection response message is sent to the network device, the rejection response message being used to indicate that the training latency constraint indication is not supported; Receive a notification message from the network device that it has cancelled its participation in the i-th round of federated learning.

15. The method according to claim 14, characterized in that, The training delay constraint indicator is used to indicate the deadline for reporting the training results of the i-th round of the model, including: The training delay constraint indicator is used to indicate the duration between the receiving time of the i-th round of training model and the reporting deadline.

16. The method according to claim 14, characterized in that, The method further includes: Send a relaxation instruction to the network device, the relaxation instruction being used to indicate the relaxation of the packet dropping conditions for the model training results and / or that all packets of the model training results are useful.

17. The method according to claim 14, characterized in that, The receiving of training latency constraint indication from the network device includes: Receive configuration information or session establishment request message from the network device, wherein the configuration information includes the training latency constraint indication, and the session establishment request message includes the training latency constraint indication.

18. The method according to claim 14, characterized in that, The receiving of training latency constraint indication from the network device includes: The system receives a control protocol data unit from the network device, the control protocol data unit including the training delay constraint indication.

19. The method according to claim 14, characterized in that, The method further includes: Receive a data packet acceleration instruction from the network device, the data packet acceleration instruction being used to indicate the accelerated transmission of the data packets to be sent for the model training results of the i-th round.

20. The method according to claim 14 or 19, characterized in that, The receiving of training latency constraint indication from the network device includes: The acknowledgment information received from the network device is a data packet containing the model training results of the (i-1)th round, the acknowledgment information including the training delay constraint indication, where i is a positive integer greater than or equal to 1.

21. The method according to claim 14, characterized in that, The method further includes: Receive a device capability condition indication from the network device, the device capability condition indication being used to indicate a threshold for the device's computing power processing capability.

22. The method according to claim 21, characterized in that, The method further includes: If it is determined that the device capability condition indication is not met, then device capability information is sent to the network device.

23. A communication device, characterized in that, The device includes: The sending module is used to send a training delay constraint indication to the terminal device participating in the i-th round of federated learning. The training delay constraint indication is used to indicate the deadline for reporting the training results of the i-th round of model, where i is a positive integer greater than or equal to 0. The receiving module is configured to, when the terminal device does not support the training delay constraint indication, also receive rejection response messages sent by N terminal devices participating in the i-th round of federated learning, wherein the rejection response messages are used to indicate that the training delay constraint indication is not supported, and N is a positive integer greater than or equal to 1; The sending module is also used to send a notification message to the N terminal devices to cancel their participation in the i-th round of federated learning.

24. A communication device, characterized in that, The device includes: a receiving module for N terminal devices participating in the i-th round of federated learning to receive a training delay constraint indication from a network device, wherein the training delay constraint indication is used to indicate the reporting deadline for the model training results of the i-th round, wherein N is a positive integer greater than or equal to 1, and i is a positive integer greater than or equal to 0; The sending module is configured to send a rejection response message to the network device if it is determined that the training delay constraint indication is not supported; the rejection response message is used to indicate that the training delay constraint indication is not supported. The receiving module is also configured to receive a notification message from the network device indicating that the user has cancelled participation in the i-th round of federated learning.

25. A communication device, characterized in that, Includes memory, transceiver, and processor: A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: A training delay constraint indication is sent to the terminal devices participating in the i-th round of federated learning. The training delay constraint indication is used to indicate the deadline for reporting the model training results in the i-th round, where i is a positive integer greater than or equal to 0. If the terminal device does not support the training delay constraint indication, a rejection response message is also received from N terminal devices participating in the i-th round of federated learning. The rejection response message is used to indicate that the training delay constraint indication is not supported, where N is a positive integer greater than or equal to 1. Send a notification message to the N terminal devices to cancel their participation in the i-th round of federated learning.

26. A communication device, characterized in that, Includes memory, transceiver, and processor: A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and performing the following operations: N terminal devices participate in the i-th round of federated learning; receive a training delay constraint indication from a network device, the training delay constraint indication being used to indicate the deadline for reporting the model training results of the i-th round, where N is a positive integer greater than or equal to 1 and i is a positive integer greater than or equal to 0; if it is determined that the training delay constraint indication is not supported, a rejection response message is sent to the network device, the rejection response message being used to indicate that the training delay constraint indication is not supported; receive a notification message from the network device canceling participation in the i-th round of federated learning.

27. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the communication method as described in any one of claims 1-22.

Images