Model management method, communication node and storage medium

Abstract

Provided in the present application are a model management method, a communication node and a storage medium. The method comprises: receiving a model management parameter, or obtaining the model management parameter in a predefined manner (110); on the basis of the model management parameter, measuring the power overhead of a model (120); and reporting control information, wherein the control information comprises at least one of the following: the power overhead, a power overhead level, and a requested model operation instruction (130).

Description

Model management methods, communication nodes, and storage media Technical Field

[0001] This application relates to the field of mobile communication technology, such as a model management method, a communication node, and a storage medium. Background Technology

[0002] In recent years, the introduction of Artificial Intelligence (AI) / Machine Learning (ML) models into mobile communication technologies has inevitably led to new energy consumption during model training, inference, and related functions. The hardware upon which AI / ML models are based, such as Central Processing Units (CPUs), Graphics Processing Units (GPUs), and Neural Processing Units (NPUs), requires a staggering amount of energy to handle massive computations. For example, the Thermal Design Power (TDP) of the NVIDIA H100 GPU is approximately 700 watts. Processing large models requires a large number of GPUs, resulting in enormous energy consumption. When an AI model is running on a battery-powered device, due to the limited battery capacity, model management and control are essential to maintain the longest possible battery life.

[0003] In general, AI models can improve the performance and efficiency of mobile communication systems, but this comes at the cost of the AI ​​models' own power consumption. Currently, there is no unified architecture that can manage all models and functions, making it impossible to reduce the energy consumption of models running in communication systems. Furthermore, it is difficult to strike a balance between the conflicting goals of increasing power consumption and improving mobile communication performance. Summary of the Invention

[0004] This application provides a model management method, a communication node, and a storage medium.

[0005] This application provides a model management method applied to a first communication node, including:

[0006] Receive model management parameters or obtain model management parameters through predefined methods;

[0007] The power overhead of the model is measured based on the model management parameters.

[0008] Report control information, which includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

[0009] This application provides a model management method applied to a second communication node, including:

[0010] Send model management parameters, which are used to instruct the first communication node to measure the power overhead of the model;

[0011] Receive control information reported by the first communication node;

[0012] The control information includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

[0013] This application also provides a communication node, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the model management method in any of the above embodiments.

[0014] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the model management method in any of the above embodiments. Attached Figure Description

[0015] Figure 1 is a schematic diagram of an AI / ML framework provided in one embodiment;

[0016] Figure 2 is a flowchart of a model management method provided in one embodiment;

[0017] Figure 3 is a flowchart of another model management method provided in one embodiment;

[0018] Figure 4 is a schematic diagram of the power overhead of a management model provided in one embodiment;

[0019] Figure 5 is a schematic diagram of the power overhead of another management model provided in one embodiment;

[0020] Figure 6 is a schematic diagram of a non-continuous operation cycle of a management model provided in one embodiment;

[0021] Figure 7 is a schematic diagram illustrating the relationship between configuration power overhead levels and model influencing factors in one embodiment;

[0022] Figure 8 is a schematic diagram of a model determined by a terminal according to an embodiment;

[0023] Figure 9 is a schematic diagram of a model determined by a network device according to an embodiment;

[0024] Figure 10 is a schematic diagram of an embodiment of a network device determining an activation or deactivation model;

[0025] Figure 11 is a schematic diagram of a model management device according to an embodiment;

[0026] Figure 12 is a schematic diagram of another model management device provided in one embodiment;

[0027] Figure 13 is a schematic diagram of the hardware structure of a communication node according to an embodiment. Detailed Implementation

[0028] Artificial Intelligence (AI) encompasses devices, components, software, algorithms, or modules with self-learning capabilities, including Machine Learning (ML), Deep Learning, Reinforcement Learning, Transfer Learning, Deep Reinforcement Learning, and Meta-learning. In some cases, AI is implemented through AI networks (or neural networks). A neural network consists of multiple layers, each containing at least one node. Typically, a neural network includes an input layer, an output layer, and at least one hidden layer. Each layer of the neural network uses at least one of the following: fully connected layers, dense layers, convolutional layers, transposed convolutional layers, directly connected layers, activation functions, normalization layers, and pooling layers. In other cases, each layer of the neural network may include a sub-neural network, such as a residual network block (or ResNet block), a dense network (DenseNet block), or a recurrent neural network (RNN). AI networks include a neural network model and / or the corresponding neural network parameters. The neural network model can be simply referred to as the network model, and the neural network parameters can be simply referred to as network parameters. A network model defines the architecture of a neural network, including the number of layers, the size of each layer, the activation function, the connections, the convolutional kernels and strides, and the convolution type (e.g., 1D convolution, 2D convolution, 3D convolution, hollow convolution, transposed convolution, separable convolution, grouped convolution, or expanded convolution). Network parameters are the weights and / or biases of each layer in the network model and their values. A network model can correspond to multiple sets of different neural network parameter values ​​to adapt to different scenarios. Network parameter values ​​can be obtained through offline and / or online training, such as training the neural network model by inputting at least one sample and label. A neural network model can correspond to multiple different neural network parameter values.

[0029] AI / ML (Artificial Intelligence / Machine Learning) is a promising area for enhancing mobile communication systems. Introducing AI / ML technologies into mobile communication systems, such as 5G, 5G-Advanced, and 6G, can improve system efficiency. For example, AI / ML inference and prediction can reduce the overhead of reference signals, reduce the overhead of channel state information feedback, or improve the accuracy of terminal positioning.

[0030] In connection with communication systems employing AI / ML technologies, throughout this application, "model" is a general term used to describe the processing methods, functions, features, or groups of features that a device in a mobile system can perform. "Model" can be understood as a function, functional module, processing method, information processing method, implementation method, functional group, configuration, configuration set, dataset (e.g., for model training), or data-driven algorithm, etc.

[0031] Figure 1 is a schematic diagram of an AI / ML framework provided in one embodiment. As shown in Figure 1, the data collection function module can provide input data for model training, management, and inference functions. The input data required for the model training function is training data (also known as training data). The input data required for managing the model or function is monitoring data. The input data required for the AI / ML inference function is inference data. The collected monitoring data may be labeled data, providing a calibration-like function, and is sent to the model management function module for comparison with the inference output to output corresponding decisions or management commands.

[0032] The model training module has the function of performing model training, validation and testing.

[0033] The model management module can perform the following functions: (1) instructing relevant model operations, such as model selection / activation / deactivation / switching / rollback, etc. (2) monitoring model performance. (3) making decisions or instructions to ensure correct inference operations based on data received from the data collection and inference functions. (4) this module is also responsible for requesting models from the model storage function for model transfer / delivery requests. (5) inputting performance feedback / retraining requests to the model training module for model (re)training or updating.

[0034] The model inference module takes data from the data collection function as input and provides output from the application model or function. The inference output is the final output of the entire AI / ML model or function, used by other units in the mobile system. Output data can also be input into the management module internally for monitoring the performance of the AI / ML model or function.

[0035] Model storage is a function responsible for storing trained / updated models that can be used to perform inference functions.

[0036] All the functional modules involved in the model mentioned in the above framework belong to the processing process. The processing processes mentioned in the embodiments of this application can all be implemented using the model and the above framework.

[0037] Models improve the performance and efficiency of mobile communication systems, but this comes at the cost of the model's own power consumption. Therefore, a balance must be struck between these two conflicting goals: balancing the power consumption cost with the improved mobile communication performance. This application's embodiments define a unified architecture to manage all model and function implementations, thereby better controlling their power consumption. This architecture can be implemented through a model management function module to provide unified power consumption control for both models and functions.

[0038] Figure 2 is a flowchart of a model management method provided in one embodiment. This method can be applied to a first communication node, which can be the node where the model resides, the managed node, or the node whose power overhead is controlled. As shown in Figure 2, the method provided in this embodiment includes the following steps:

[0039] Step 110: Receive model management parameters or obtain model management parameters through a predefined method.

[0040] Step 120: Measure the power overhead of the model according to the model management parameters.

[0041] Step 130: Report control information, which includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

[0042] In this embodiment, "model" primarily refers to an AI model. Model management parameters can be understood as parameters used to manage the model, such as parameters instructing model-related operations, including model selection, activation, deactivation, switching, and / or rollback; parameters for monitoring model performance, including power overhead, model status, and / or model parameters; and decisions or instructions made regarding the model, such as whether to measure or report control information, whether to train and / or update the model. Model management parameters can be obtained through definition (e.g., system-defined or default parameters) or indicated or configured by the second communication node. The first communication node reports control information to the second communication node based on the measured power overhead of the model. The first communication node then performs unified power overhead control on the model according to the control information. Power overhead can also be referred to as power consumption or power usage.

[0043] In one embodiment, the model management parameters include at least one of the following: measurement cycle, measurement start point, measurement duration, and reporting format.

[0044] In one embodiment, the reporting format includes at least one of the following:

[0045] The reported measurement value must be at least power or work.

[0046] The reported measurements are quantified between upper and lower bounds of the measurements at equal or unequal intervals, which are obtained by predefinition.

[0047] In one embodiment, reporting control information includes one of the following:

[0048] Periodically report control information;

[0049] Report control information upon receiving a reporting instruction.

[0050] In one embodiment, the method further includes:

[0051] Report auxiliary information for model operation, including at least one of the following: model operation time information and level information;

[0052] The time information includes at least one of the following: period, duration, and activation start time.

[0053] In one embodiment, the method further includes:

[0054] Receive activation or deactivation instructions;

[0055] Upon receiving an activation instruction, the model is run;

[0056] Upon receiving a deactivation instruction, the model is stopped from running.

[0057] In one embodiment, the request model operation instruction includes at least one of the following instructions: request model activation, request model deactivation, request model configuration instruction, request model transmission, and request configuration power overhead level.

[0058] In one embodiment, the request model operation instruction includes at least one of the following: the index of the model, and related parameters associated with the model index.

[0059] In one embodiment, the method further includes:

[0060] Based on the indication information of the non-continuous operation cycle, the operation of the model is controlled to alternate between an active state and a dormant state.

[0061] In one embodiment, the method further includes:

[0062] The indication information for the discontinuous operation cycle is received through at least one of radio resource control, medium access control (MAC) layer signaling, and downlink control information (DCI).

[0063] The indication information for the discontinuous operation cycle is provided by the second communication node based on the reported power overhead semi-static or dynamic configuration or indication.

[0064] In one embodiment, the indication information for the discontinuous operation cycle includes at least one of the following:

[0065] The length of the discontinuous operation cycle, the starting point of the discontinuous operation cycle, the length of the active time, the starting point of the active time, the length of the dormant time, and the starting point of the dormant time.

[0066] In one embodiment, the indication information for discontinuous operation cycles includes indication information corresponding to one or more discontinuous operation cycles.

[0067] In one embodiment, the indication information of the discontinuous operation cycle is bound to the operating status of the first communication node.

[0068] In one embodiment, the indication information of the discontinuous operation cycle is bound to the energy-saving mode of the first communication node.

[0069] In one embodiment, the discontinuous operation cycle is consistent with or matches the activation period of the energy-saving mode; or,

[0070] The duration of the activation period within the discontinuous operation cycle is consistent with or matches the activation period of the energy-saving mode.

[0071] The activation period matching the energy-saving mode includes:

[0072] The activation period is less than or equal to that of the energy-saving mode.

[0073] In one embodiment, the method further includes:

[0074] To obtain the relationship between control information and the influencing factors of the model;

[0075] The influencing factors include at least one of the following: model type, model accuracy, model depth, model width, number of model parameters, and split points of a bilateral model.

[0076] In one embodiment, the relationship between the control information and the influencing factors of the model is determined by a second communication node or obtained based on predefined information.

[0077] In one embodiment, the method further includes:

[0078] The updated model is received, and the updated model is determined by the second communication node based on the control information.

[0079] In one embodiment, the method further includes:

[0080] The model is determined based on the power overhead level configured in the second communication node.

[0081] In one embodiment, the relationship between control information and the influencing factors of the model is adaptively adjusted by the second communication node based on the reported control information and / or the model's adjustment information.

[0082] In one embodiment, the method further includes:

[0083] Receive random access resource configuration information, wherein the partitions of the random access resources are associated with the control information of the model;

[0084] Based on the random access resource configuration information, select random access resources and transmit random access signals.

[0085] Figure 3 is a flowchart of a model management method provided in one embodiment. This method can be applied to a second communication node, which can be a node for managing the model or a node for controlling power overhead. As shown in Figure 3, the method provided in this embodiment includes steps 210 and 220.

[0086] In step 210, model management parameters are sent, which are used to instruct the first communication node to measure the power overhead of the model.

[0087] In step 220, control information reported by the first communication node is received. The control information includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

[0088] In one embodiment, the model management parameters include at least one of the following: measurement cycle, measurement start point, measurement duration, and reporting format.

[0089] In one embodiment, the reporting format includes at least one of the following:

[0090] The reported measurement value must be at least power or work.

[0091] The reported measurements are quantified between upper and lower bounds of the measurements at equal or unequal intervals, which are obtained by predefinition.

[0092] In one embodiment, receiving the control information includes one of the following:

[0093] The control information is received periodically;

[0094] The control information is received when a reporting instruction is sent.

[0095] In one embodiment, the method further includes:

[0096] Receive auxiliary information for model operation, the auxiliary information including at least one of the following: model operation time information, level information;

[0097] The time information includes at least one of the following: period, duration, and activation start time.

[0098] In one embodiment, the method further includes:

[0099] Send activation instruction or deactivation instruction;

[0100] The activation instruction information is used to instruct the first communication node to run the artificial intelligence model;

[0101] The deactivation instruction is used to instruct the first communication node to stop running the artificial intelligence model.

[0102] In one embodiment, the request model operation instruction includes at least one of the following instructions: request model activation, request model deactivation, request model configuration instruction, request model transmission, and request configuration power overhead level.

[0103] In one embodiment, the method further includes:

[0104] The system determines indication information for discontinuous operation cycles, which is used to indicate the time during which the artificial intelligence model is in an active or dormant state.

[0105] In one embodiment, the method further includes:

[0106] Based on the semi-static or dynamic configuration of the power overhead or the indication information indicating the discontinuous operation cycle;

[0107] The indication information for the discontinuous operation cycle is transmitted via at least one of radio resource control, MAC layer signaling, and DCI signaling.

[0108] In one embodiment, the method further includes:

[0109] Determine the relationship between control information and the influencing factors of the model;

[0110] The influencing factors include at least one of the following: model type, model accuracy, model depth, model width, number of model parameters, and split points of a bilateral model.

[0111] In one embodiment, the relationship between control information and the influencing factors of the model is determined by a second communication node or obtained based on predefined information.

[0112] In one embodiment, the method further includes:

[0113] The model is updated based on the control information;

[0114] The updated model is sent to the first communication node.

[0115] In one embodiment, the method further includes:

[0116] Configure and send power overhead levels;

[0117] The power overhead level is used to assist the first communication node in determining the model based on the power overhead level.

[0118] In one embodiment, the method further includes:

[0119] Based on the reported control information and / or model adjustment information, adaptively adjust the relationship between the control information and the influencing factors of the model.

[0120] In one embodiment, the method further includes:

[0121] Determine the random access resource configuration information, wherein the partitioning of the random access resource is associated with the control information of the model.

[0122] The model management method of this application is illustrated by some embodiments below.

[0123] Example 1

[0124] This embodiment mainly describes the power consumption measurement and reporting process of the model. If power consumption control of the model is required, it is first necessary to understand the real-time power consumption of the model during operation in the device, and the device needs to perform power consumption measurement and reporting.

[0125] Assuming that the main control body for power overhead is located in the model management module, the model management module is located in the first communication node (such as a network device), and the model is located in the second communication node (such as a terminal), this embodiment defines the signaling for power overhead control and management between the network device and the terminal.

[0126] For example, the model management module is located in a network device (such as a core network, wireless access network, base station, or network management device), while the model is located in the terminal. In this embodiment, the relevant signaling between the network device and the terminal is defined.

[0127] For example, the model management module is located in a network device (such as a core network, wireless access network, or network management device), while the model is located in a base station. This embodiment can be extended to support this scenario.

[0128] Figure 4 is a schematic diagram of the power overhead of a management model provided in one embodiment. As shown in Figure 4, the terminal can periodically measure the power overhead during model operation and report it periodically. Before measurement, the configured power overhead measurement period, measurement start point, and / or reporting format, etc., can be obtained from the network device. The periodicity can be configured according to different states of the terminal, such as Radio Resource Control (RRC) linked state, inactive state, or idle state, or it can be set according to other measurement periods of the terminal. This can concentrate multiple measurements in the same time period as much as possible to avoid unnecessary power overhead caused by scattered measurements.

[0129] Figure 5 is a schematic diagram of the power overhead of another management model provided in one embodiment. As shown in Figure 5, the network device can also request power overhead measurement results from the terminal device non-periodically. Before or accompanied by request signaling, model management parameters such as the measurement duration, measurement start point, or reporting format of the power overhead can be configured. The above-mentioned model management parameters can also be auxiliary information for model management. After the terminal obtains these model management parameters, the network device can request or instruct the terminal to report power overhead. After measuring the power overhead, the terminal reports the power overhead non-periodically.

[0130] In some embodiments, periodic measurement reporting can be used in combination with aperiodic measurement reporting. For example, when the period is set to be long, aperiodic measurement reporting can be initiated within a certain period to compensate for the problem of excessive delay in periodic measurement reporting.

[0131] After acquiring power overhead, network devices can decide on their next course of action based on the power overhead measurement results.

[0132] In this embodiment, the model management parameters include information about auxiliary power overhead measurement, such as measurement cycle, measurement duration, measurement start point, or reporting format, which can be defined as follows and indicated by RRC signaling or MAC layer signaling to the device where the power overhead measurement is located.

[0133] Measurement cycle: Periodic measurement reporting can be performed according to the configuration of the measurement cycle, which can be set in units of superframe, frame, subframe, time slot and / or symbol.

[0134] Measurement Duration: The measurement duration is configured to specify the actual measurement time for power consumption. In the case of periodic measurement reporting, the measurement duration should not exceed the measurement cycle. For non-periodic measurement reporting, the measurement duration should not exceed the next measurement start point. Measurement duration can be set in units of superframes, frames, subframes, time slots, and / or symbols. A longer measurement duration results in smoother and more accurate long-term measurements, but it also consumes more power. Therefore, a balance must be struck between measurement duration and measurement accuracy.

[0135] Measurement Start Point: This configures the specific starting location of the measurement, and is required for both periodic and aperiodic measurement reporting. When measurement reporting is periodic, the measurement start point can be defined as the relative time offset relative to the superframe header, frame header, subframe header, slot header, and / or the first symbol. When measurement reporting is aperiodic, the measurement start point is configured with reference to the signaling indicating power overhead reporting. For example, if the signaling indicating power overhead reporting is DCI, the measurement start point can be delayed by a time length ΔT relative to this DCI signaling. ΔT can be the length of one or more slots, the length of one or more symbols, or the time deviation of the sum of the lengths of one or more slots and one or more symbols.

[0136] Reporting Format: The basic measurement value for power overhead reporting can be energy consumption per unit time, i.e., power value, in watts (W). Alternatively, if the measurement duration is fixed, energy consumption, i.e., work, in joules, kilowatt-hours, or other equivalent units can also be reported. Due to limitations in measurement reporting overhead, power overhead reporting should be quantized before submission. The reported measurement value is quantized between the upper and lower bounds of the measurement at equal or unequal intervals, which are predefined. For example, power reporting can be quantized into 256 levels, represented using 8 bits in binary. 00000000 indicates power below X watts, 11111111 indicates power above Y watts, and the intermediate binary indices quantify the power difference between X and Y watts at equal or unequal intervals. This quantization method can be expressed in a table and stored in the network model and terminals using predefined methods. Once network devices and terminals reach a consensus on the power overhead level through signaling interaction, measurement reporting can also directly report the power overhead level.

[0137] The aforementioned model management parameters, including measurement cycle, measurement duration, measurement start point, and / or reporting format, can also be determined or obtained through predefined methods.

[0138] When the power overhead control unit is located in the core network and the model is located in the base station, a similar signaling procedure can be used between the core network and the base station to report power overhead.

[0139] Whether power overhead measurement reporting is periodic or aperiodic, the effective measurement time may not perfectly match the model's activation or running time. Therefore, to better adapt the power overhead measurement period and the model activation / running period, the device hosting the model can report auxiliary information for model activation or running. This reporting includes model running time information and / or level information. Time information includes, but is not limited to, the model activation or running cycle, duration, activation start time, and / or activation / running level. Level information, such as the activation / running level, defines the possible power overhead levels during model operation. For example, model training has the highest power overhead, model inference has the second highest, and the model only handles basic management functions with the lowest power overhead. The device can report model training time information and activation / running level separately, or it can report model inference time information and activation / running level separately. The auxiliary information for model activation or running reported by the device hosting the model can be managed by the model management device, such as the network device configuring periodic reporting, or the network device instructing the device hosting the model (such as a terminal) to report aperiodically.

[0140] Example 2

[0141] This embodiment mainly describes the activation or deactivation process of the model. It is assumed that the control subject for model activation or deactivation is located in the model management module, which is located in the first communication node (such as a network device), while the model is located in the second communication node (such as a terminal).

[0142] Based on the power overhead measurement report from the device where the model is located, the network device can comprehensively judge the balance between the performance gain obtained by the model and its own power consumption, and decide whether to activate the model, deactivate the model, or put the model into an inactive state.

[0143] For models whose activation or deactivation times exhibit a certain periodicity, network devices can also set discontinuous processing (DP) cycles for the devices hosting the models. During this cycle, the model alternates between active and dormant states. During active periods (On-duration), the model can be started; during off-duration, it can be shut down to conserve energy until the next active period.

[0144] Figure 6 is a schematic diagram of a non-continuous operation cycle of a management model provided in one embodiment. As shown in Figure 6, the specific duration or timing of the DP cycle can be semi-statically or dynamically configured by the network device based on the power overhead reporting results of the device where the model is located, and notified to the terminal through RRC signaling, MAC layer signaling, or DCI signaling. These specific parameters may involve the length of the DP cycle, the definition of the cycle start point, the length of the active time, the sleep time, and / or the definition of the start point, etc.

[0145] The DP period can be set in units of superframes, frames, subframes, time slots, and / or symbols.

[0146] Measurement Duration: Measurement duration indicates the actual length of time for power consumption measurement. When periodic measurements are reported, the measurement duration should not exceed the measurement cycle; when non-periodic measurements are reported, the measurement duration should not exceed the next measurement start point. Measurement duration can be set in units of superframes, frames, subframes, time slots, and / or symbols. A longer measurement duration results in smoother and more accurate long-term measurements, but it also consumes more power. Therefore, a balance needs to be struck between measurement duration and measurement accuracy.

[0147] The DP cycle start point can be defined as the relative time offset relative to the superframe header, frame header, subframe header, slot header, and / or the first symbol.

[0148] Active time or sleep time can be defined simultaneously, or only one of them can be defined, with the other derived by reasoning from the period definition. When defining active time, its length unit and starting point can be set in units of superframe, frame, subframe, time slot, and / or symbol.

[0149] Network devices can be configured with multiple DP cycles, including at least one long DP cycle and one short DP cycle, to adapt to service requirements.

[0150] The DP cycle setting and configuration can also be aligned with or bound to the device's operating state. For example, different DP cycle parameters can be configured for the terminal device's RRC linked state, inactive state, and / or idle state. For instance, the DP cycle is shorter when the device is linked, while the DP cycle progressively becomes longer when the device is inactive or idle.

[0151] The DP cycle setting and configuration can also be aligned or bound to other energy-saving modes of the device. For example, when there is a strong binding relationship between the activation period configuration of the model operation and the terminal's energy-saving mode, the DP cycle can be kept consistent with or matched with the activation period of the energy-saving mode, and the duration of the activation period within the DP cycle can also be kept consistent with or matched with the activation period of the energy-saving mode. The matching method can be that the DP cycle is less than or equal to the transmission cycle, or the duration of the activation period is less than or equal to the transmission time.

[0152] Network devices decide whether to activate or deactivate a model based not only on the power consumption measurement results reported by the device hosting the model, but also by the device hosting the model directly sending a request to the network device to activate or deactivate the model or function. For example, if a terminal, acting as the device hosting the model, performs power consumption measurements and finds that its model's power consumption exceeds a threshold or expectation, it can directly request the network device to configure a deactivation instruction; conversely, if it finds that the model needs performance improvements, it can also directly request the network device to configure an activation instruction.

[0153] Table 1 is an organizational structure table of operation instructions for requesting activation or deactivation of models. As shown in Table 1, the terminal, as the device where the model resides, may simultaneously maintain multiple types of models, such as models performing beam management, models performing channel prediction and compression, and models performing positioning functions. The activation or deactivation request sent by the terminal can be for a specific model, a group of models, or all models; even models with the same function, such as beam management models, may be in different states, such as training state, inference state, or low-power maintenance state. The activation or deactivation request sent by the terminal can be targeted at different states of the model, for example, deactivating the training state but activating the inference state.

[0154] Table 1. Organizational structure of a request model for activation or deactivation.

[0155] Example 3

[0156] This embodiment mainly describes the power overhead levels of the management model. It is assumed that the control entity for model activation or deactivation is located in the model management module, which is located in the first communication node (e.g., a network device), while the model is located in the second communication node (e.g., a terminal).

[0157] Referring to Example 2, in order to reduce the energy or power overhead of the model, deactivation can be adopted. However, after deactivation, it is difficult to obtain the performance gain that the model can achieve. Therefore, between the two extreme states of activation and deactivation, some appropriate compromises and balances can be made. This can be achieved by setting certain levels of power overhead and selecting a suitable model for the device based on the different configurations of the models corresponding to different power overhead levels. In this example, power overhead and power consumption are synonymous.

[0158] In this embodiment, the energy consumption influencing factors of the model may depend on one or more of the following aspects:

[0159] 1. Model size and complexity, including:

[0160] Number of parameters: The more parameters a model has, the higher its complexity and the more computational resources it requires to process. Large models such as Transformer and BERT consume a lot of power due to their huge number of parameters and deep network structures.

[0161] Depth: The depth of a network directly affects the model's learning ability and computational requirements. The more layers there are, the more forward and backward propagation computations are required, and the higher the energy consumption.

[0162] 2. Training algorithm, including:

[0163] Optimization Algorithm: The optimization algorithm used (such as Adam or stochastic gradient descent (SGD)) affects the convergence speed and stability, indirectly affecting the energy consumption of the training process. Some algorithms may require more iterations to converge, increasing the overall energy consumption.

[0164] Batch size: The batch size during training affects GPU utilization and memory requirements. Larger batches can improve the efficiency of parallel processing, but may also increase memory pressure and power consumption.

[0165] 3. Number of training iterations and durations, including:

[0166] Number of epochs: The more epochs the model trains, the greater the total computational cost required to complete the training, and the higher the energy consumption.

[0167] Early stopping technique: Using the early stopping technique, training can be stopped when it is verified that the loss is no longer improving, thereby reducing unnecessary computation and energy consumption.

[0168] 4. Hardware efficiency, including:

[0169] Processor type: The type of hardware used (such as CPU or GPU) greatly affects energy efficiency. GPUs are generally more energy efficient than CPUs when performing large-scale matrix calculations in parallel.

[0170] Hardware utilization: Insufficient utilization of hardware resources leads to energy waste. Optimizing hardware utilization can reduce energy consumption during idle periods.

[0171] 5. Model deployment and inference, including:

[0172] Quantization: By quantizing the model, floating-point parameters are converted into low-precision (such as int8) format, which can reduce model size and runtime energy consumption.

[0173] Pruning: Removing redundant connections in the network (weight pruning) can reduce model complexity and energy consumption during the inference phase.

[0174] Distillation: By training a smaller model (student model) to mimic the behavior of a larger model (teacher model), energy consumption during inference can be reduced without significantly decreasing accuracy.

[0175] 6. Software and tool optimization, including:

[0176] Software framework: The software framework used to implement the model (such as TensorFlow, PyTorch) also affects runtime efficiency and energy consumption. Different frameworks may differ in memory management and computational optimization.

[0177] Compiler optimization: Through advanced compiler optimization techniques, such as the integration of optimization and operation, runtime efficiency can be improved and energy consumption reduced.

[0178] Specifically, for the models used in mobile communication systems or terminals, under the same conditions of hardware, software algorithms, and implementation tools, the magnitude of model power consumption and its controllable influencing factors mainly come from the model type, accuracy (quantization bits), model depth and width, number of model parameters, and split points of the two-sided model. If the hardware conditions of the implemented models differ, the power consumption of the models is related to the hardware structure, type, and data transmission channel size. The model's power management mechanism needs to flexibly handle the above-mentioned factors related to the model's own parameters, hardware and software, and / or tools, so as to change, adjust, and optimize the model configuration within the constraints of power consumption.

[0179] Figure 7 is a schematic diagram illustrating the relationship between power overhead levels and model influencing factors according to an embodiment. As shown in Figure 7, network devices can establish associations or mappings between power overhead levels and various influencing factors of the model. These levels and their relationships with the model's influencing factors can be predefined and stored in the network devices and terminals. Alternatively, when the network devices have configured the levels and relationships, they can be configured to the terminals via signaling. The corresponding levels and relationships can be terminal-specific or shared by multiple terminals.

[0180] It should be noted that other control information, such as power overhead and / or request model operation instructions, may also be related to model influencing factors.

[0181] Table 2 shows the relationship between power overhead levels and model influencing factors in one embodiment. It takes the case of increasing power overhead from level 1 to level 3 as an example.

[0182] Table 2 Relationship between power overhead level and model influencing factors

[0183] MLP stands for Multi-Layer Perceptron; CNN stands for Convolutional Neural Network.

[0184] Figure 8 is a schematic diagram illustrating a model determined by a terminal according to an embodiment. As shown in Figure 8, when the power overhead control and management center determines in real time the applicable power overhead level for the model, the network device can configure or instruct the power overhead level to the device where the model resides, such as the terminal. The terminal determines the various settings or parameters for executing the model function based on the pre-configured mapping relationship between each level and various influencing factors of the model. Furthermore, the terminal can first request to configure the power overhead level, and then the network device configures or instructs the power overhead level to the terminal. Subsequent operations by the terminal are similar to those described above.

[0185] Figure 9 is a schematic diagram illustrating a model determination process by a network device according to an embodiment. As shown in Figure 9, when the power overhead control and management center determines in real time the applicable power overhead level for the model, the network device can also select a suitable model based on the determined power overhead level and transmit the model to the device where the model resides. The specific transmission methods for the model are diverse, such as transmitting the model's index, transmitting the dataset or dataset index used for model training, or transmitting the model's structure and parameters.

[0186] The device hosting the model can determine its power overhead level based on its own power overhead measurement results and the mapping relationship between power overhead and power overhead level, and report this to the device managing the model. The mapping relationship between power overhead and power overhead level is optionally configured by the device managing the model for the device hosting the model. The device managing the model further adjusts the desired power overhead level based on the reported power overhead level and determines the model accordingly. Once the model is determined, it is transmitted to the device hosting the model.

[0187] Figure 10 is a schematic diagram illustrating how a network device determines whether to activate or deactivate a model, according to an embodiment. As shown in Figure 10, the device managing the model can also directly decide whether to activate or deactivate the model.

[0188] The device hosting the model can also directly request network devices to configure a specific type or level of model based on its own power consumption measurements. The requested model can reduce or enhance power consumption to achieve energy saving or performance improvement. The request may include the index of the configured model or related parameters or descriptions associated with the model index.

[0189] The establishment of power overhead levels and their mapping to the model can be initially configured by network administrators. However, this initial configuration only considers the current state of the model at the time of configuration, without taking into account further model evolution and long-term optimization of power overhead in system terminals. Therefore, during long-term model operation, network devices still need to adaptively adjust the power overhead levels or the mapping between power overhead levels and influencing factors of the model based on reported power overhead measurements or model structure / parameter adjustment information. This adaptive adjustment process can be completed through manual intervention or artificial intelligence management.

[0190] The power overhead level of the model is essentially determined once the model is defined. The device hosting the model, such as a terminal, can map the power overhead level to random access resources, that is, associate the random access resources of a partition with the power overhead level. During the random access process, the terminal selects the corresponding random access resources based on the power overhead level of the model. It can implicitly notify the network device hosting the model management function of the power overhead level of the model it has configured, thereby determining the bandwidth, modulation scheme, waveform, number of Multiple-In Multiple-Out (MIMO) layers, bandwidth part (BWP), carrier, reference signal configuration, and other parameters for subsequent access to adapt to the model.

[0191] This application also provides a model management device. Figure 11 is a schematic diagram of the structure of a model management device provided in one embodiment. As shown in Figure 11, the model management device includes:

[0192] The parameter acquisition module 310 is configured to receive model management parameters or obtain model management parameters by a predefined method.

[0193] Measurement module 320 is configured to measure the power overhead of the model according to the model management parameters;

[0194] The information reporting module 330 is configured to report control information, which includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

[0195] In one embodiment, the model management parameters include at least one of the following:

[0196] Measurement cycle, measurement starting point, measurement duration, and reporting format.

[0197] In one embodiment, the reporting format includes at least one of the following:

[0198] The reported measurement value must be at least power or work.

[0199] The reported measurements are quantified between upper and lower bounds of the measurements at equal or unequal intervals, which are obtained by predefinition.

[0200] In one embodiment, the reported control information includes one of the following:

[0201] Periodically report control information;

[0202] Report control information upon receiving a reporting instruction.

[0203] In one embodiment, the reporting module is further configured to: report auxiliary information for model operation, the auxiliary information including at least one of the following: model operation time information, level information;

[0204] The time information includes at least one of the following: period, duration, and activation start time.

[0205] In one embodiment, the device further includes:

[0206] The receiving module is configured to receive activation or deactivation instructions.

[0207] The running module is configured to run the model upon receiving an activation instruction.

[0208] The stop module is configured to stop running the model upon receiving a deactivation instruction.

[0209] In one embodiment, the request model operation instruction includes at least one of the following instructions: request model activation, request model deactivation, request model configuration instruction, request model transmission, and request power overhead level configuration.

[0210] In one embodiment, the request model operation instruction includes at least one of the following: the index of the model, and related parameters associated with the model index.

[0211] In one embodiment, the device further includes:

[0212] The control module is configured to control the model to alternate between an active state and a dormant state based on indication information from discontinuous operating cycles.

[0213] In one embodiment, the indication information of the discontinuous operation cycle is bound to the energy-saving mode of the first communication node.

[0214] In one embodiment, the discontinuous operation cycle is consistent with or matches the activation period of the energy-saving mode; or,

[0215] The duration of the activation period within the discontinuous operation cycle is consistent with or matches the activation period of the energy-saving mode.

[0216] The activation period matching the energy-saving mode includes:

[0217] The activation period is less than or equal to that of the energy-saving mode.

[0218] In one embodiment, the device further includes:

[0219] The relationship acquisition module is set up to acquire the relationship between control information and the influencing factors of the model.

[0220] The influencing factors include at least one of the following: model type, model accuracy, model depth, model width, number of model parameters, and split points of a bilateral model.

[0221] In one embodiment, the relationship between control information and the influencing factors of the model is determined by a second communication node or obtained based on predefined information.

[0222] In one embodiment, the device further includes:

[0223] The model receiving module is configured to receive the updated model, which is determined by the second communication node based on the control information.

[0224] In one embodiment, the device further includes:

[0225] The model determination module is configured to determine the model based on the power overhead level configured for the second communication node.

[0226] In one embodiment, the relationship between control information and the influencing factors of the model is adaptively adjusted by the second communication node based on the reported control information and / or the model's adjustment information.

[0227] In one embodiment, the device further includes:

[0228] The random access module is configured to receive random access resource configuration information, wherein the partitions of the random access resources are associated with the control information of the model; and select random access resources and transmit random access signals according to the random access resource configuration information.

[0229] The model management device proposed in this embodiment belongs to the same inventive concept as the model management method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in any of the above embodiments. Furthermore, this embodiment has the same beneficial effects as the model management method.

[0230] This application also provides a model management device. Figure 12 is a schematic diagram of the structure of a model management device provided in one embodiment. As shown in Figure 12, the model management device includes:

[0231] The parameter sending module 410 is configured to send model management parameters, which are used to instruct the first communication node on the power overhead of the measurement model;

[0232] The information receiving module 420 is configured to receive control information reported by the first communication node;

[0233] The control information includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

[0234] In one embodiment, the model management parameters include at least one of the following:

[0235] Measurement cycle, measurement starting point, measurement duration, and reporting format.

[0236] In one embodiment, the reporting format includes at least one of the following:

[0237] The reported measurement value must be at least power or work.

[0238] The reported measurements are quantified between upper and lower bounds of the measurements at equal or unequal intervals, which are obtained by predefinition.

[0239] In one embodiment, receiving the control information includes one of the following:

[0240] The control information is received periodically;

[0241] The control information is received when a reporting instruction is sent.

[0242] In one embodiment, the information receiving module 420 is further configured to:

[0243] Receive auxiliary information for model operation, the auxiliary information including at least one of the following: model operation time information, level information; wherein, the time information includes at least one of the following: period, duration, activation start time.

[0244] In one embodiment, the device further includes:

[0245] The activation module is configured to send activation or deactivation instructions.

[0246] The activation instruction information is used to instruct the first communication node to run the artificial intelligence model;

[0247] The deactivation instruction is used to instruct the first communication node to stop running the artificial intelligence model.

[0248] In one embodiment, the request model operation instruction includes at least one of the following instructions: request model activation, request model deactivation, request model configuration instruction, request model transmission, and request power overhead level configuration.

[0249] In one embodiment, the device further includes:

[0250] The cycle determination module is configured to determine indication information for discontinuous operation cycles, which is used to indicate the time when the artificial intelligence model is in an active state and a dormant state.

[0251] In one embodiment, the device further includes:

[0252] The cycle indication module is configured to semi-statically or dynamically configure or indicate indication information for the discontinuous operation cycle based on the power overhead;

[0253] The indication information for the discontinuous operation cycle is transmitted via at least one of radio resource control, MAC layer signaling, and DCI signaling.

[0254] In one embodiment, the device further includes:

[0255] The relationship determination module is set up to determine the relationship between control information and the influencing factors of the model;

[0256] The influencing factors include at least one of the following: model type, model accuracy, model depth, model width, number of model parameters, and split points of a bilateral model.

[0257] In one embodiment, the relationship between control information and the influencing factors of the model is determined by a second communication node or obtained based on predefined information.

[0258] In one embodiment, the device further includes:

[0259] The update module is configured to update the model based on the control information and send the updated model to the first communication node.

[0260] In one embodiment, the device further includes:

[0261] The power overhead level configuration module is configured to configure and send power overhead levels; the power overhead level is used to assist the first communication node in determining the model based on the power overhead level.

[0262] In one embodiment, the device further includes:

[0263] The adjustment module is configured to adaptively adjust the relationship between the control information and the influencing factors of the model based on the reported control information and / or the model's adjustment information.

[0264] In one embodiment, the device further includes:

[0265] The determination module is configured to determine random access resource configuration information, wherein the partitions of the random access resources are associated with the control information of the model.

[0266] The model management device proposed in this embodiment belongs to the same inventive concept as the model management method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in any of the above embodiments. Furthermore, this embodiment has the same beneficial effects as the model management method.

[0267] This application also provides a communication node. Figure 13 is a schematic diagram of the hardware structure of a communication node provided in an embodiment. As shown in Figure 13, the communication node provided in this application includes a processor 510 and a memory 520. The processor 510 in the communication node can be one or more, and Figure 13 shows one processor 510 as an example. The memory 520 is configured to store one or more programs. The one or more programs are executed by the one or more processors 510, so that the one or more processors 510 implement the model management method as described in the embodiment of this application.

[0268] The communication node also includes: a communication device 530, an input device 540, and an output device 550.

[0269] The processor 510, memory 520, communication device 530, input device 540 and output device 550 in the communication node can be connected by a bus or other means. Figure 13 shows an example of connection by bus.

[0270] Input device 540 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the communication node. Output device 550 may include display devices such as a display screen.

[0271] The communication device 530 may include a receiver and a transmitter. The communication device 530 is configured to perform information transmission and reception communication under the control of the processor 510.

[0272] The memory 520, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the model management method described in the embodiments of this application (e.g., parameter acquisition module 310, measurement module 320, and information reporting module 330 in the model management device). The memory 520 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the communication node, etc. Furthermore, the memory 520 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 520 may further include memory remotely located relative to the processor 510, and these remote memories can be connected to the communication node via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0273] This application embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements any of the model management methods described in this application embodiment. The method includes: receiving model management parameters or obtaining model management parameters by a predefined method; measuring the power overhead of a model based on the model management parameters; and reporting control information, the control information including at least one of the following: the power overhead, a power overhead level, and a request for model operation instruction. Alternatively, the method includes: sending model management parameters, the model management parameters being used to instruct a first communication node to measure the power overhead of a model; and receiving control information reported by the first communication node. The control information includes at least one of the following: the power overhead, a power overhead level, and a request for model operation instruction.

[0274] This application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements any of the model management methods described in this application.

[0275] The computer storage medium in this application embodiment can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable CD-ROM, optical storage device, magnetic storage device, or any suitable combination thereof. The computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0276] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device.

[0277] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, radio frequency (RF), etc., or any suitable combination thereof.

[0278] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0279] This application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the model management method as described in any of the above embodiments.

[0280] The above description is merely an exemplary embodiment of this application and is not intended to limit the scope of protection of this application.

[0281] Those skilled in the art will understand that the term user terminal encompasses any suitable type of wireless user equipment, such as mobile phones, portable data processing portable web browsers, or vehicle-mounted mobile stations.

[0282] Generally, the various embodiments of this application can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. For example, some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, although this application is not limited thereto.

[0283] Embodiments of this application can be implemented by executing computer program instructions through the data processor of a mobile device, for example, in a processor entity, or through hardware, or through a combination of software and hardware. The computer program instructions can be assembly instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages.

[0284] Any block diagram of logical flow in the accompanying drawings of this application may represent program steps, or may represent interconnected logic circuits, modules, and functions, or may represent a combination of program steps and logic circuits, modules, and functions. The computer program may be stored in memory. The memory may be of any type suitable to the local technical environment and may be implemented using any suitable data storage technology, such as, but not limited to, read-only memory (ROM), random access memory (RAM), optical storage devices and systems (Digital Video Disc (DVD) or Compact Disk (CD), etc.). Computer-readable media may include non-transitory storage media. The data processor may be of any type suitable to the local technical environment, such as, but not limited to, general-purpose computers, special-purpose computers, microprocessors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), and processors based on multi-core processor architectures.

Claims

1. A model management method, applied to a first communication node, comprising: Receive model management parameters or obtain model management parameters through predefined methods; The power overhead of the model is measured based on the model management parameters. Report control information, which includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

2. The method according to claim 1, wherein, The model management parameters include at least one of the following: Measurement cycle, measurement starting point, measurement duration, and reporting format.

3. The method according to claim 2, wherein, The reporting format includes at least one of the following: The reported measurement value must be at least power or work. The reported measurements are quantified between upper and lower bounds of the measurements at equal or unequal intervals, which are obtained by predefinition.

4. The method according to claim 1, wherein, The reported control information includes one of the following: Periodically report control information; In response to receiving a reporting instruction, control information is reported.

5. The method according to claim 1, further comprising: Report auxiliary information for the operation of the model, wherein the auxiliary information includes at least one of the following: The model's runtime information and level information; The time information includes at least one of the following: period, duration, and activation start time.

6. The method according to claim 1, further comprising: Receive activation or deactivation instructions; In response to determining that an activation instruction has been received, the model is run; Upon confirming that a deactivation instruction has been received, the model is stopped from running.

7. The method according to claim 1, wherein, The request model operation instruction includes at least one of the following instructions: request model activation, request model deactivation, request model configuration instruction, request model transmission, and request configuration power overhead level.

8. The method according to claim 7, wherein, The requested model operation instruction includes at least one of the following: the model index, and related parameters associated with the model index.

9. The method according to claim 1, further comprising: Based on the indication information of the non-continuous operation cycle, the operation of the model is controlled to alternate between an active state and a dormant state.

10. The method according to claim 9, wherein, The indication information for the discontinuous operation cycle is bound to the energy-saving mode of the first communication node.

11. The method according to claim 10, wherein, The discontinuous operation cycle is consistent with or matches the activation period of the energy-saving mode; or, The duration of the activation period within the discontinuous operation cycle is consistent with or matches the activation period of the energy-saving mode. The matching of the activation period with the energy-saving mode includes: being less than or equal to the activation period of the energy-saving mode.

12. The method according to claim 1, further comprising: Obtain the relationship between the control information and the influencing factors of the model; The influencing factors include at least one of the following: model type, model accuracy, model depth, model width, number of model parameters, and split points of a bilateral model.

13. The method according to claim 12, wherein, The relationship between the control information and the influencing factors of the model is determined by the second communication node, or obtained based on predefined information.

14. The method according to claim 1, further comprising: The updated model is received, and the updated model is determined by the second communication node based on the control information.

15. The method according to claim 1, further comprising: The model is determined based on the power overhead level configured in the second communication node.

16. The method according to claim 12, wherein, The relationship between the control information and the influencing factors of the model is adaptively adjusted by the second communication node based on the reported control information and / or the adjustment information of the model.

17. The method according to claim 1, further comprising: Receive random access resource configuration information, wherein the partitions of the random access resources are associated with the control information of the model; Based on the random access resource configuration information, select random access resources and transmit random access signals.

18. A model management method, applied to a second communication node, comprising: Send model management parameters, which are used to instruct the first communication node to measure the power overhead of the model; Receive control information reported by the first communication node; The control information includes at least one of the following: power overhead, power overhead level, and request model operation instruction.

19. The method according to claim 18, wherein, The model management parameters include at least one of the following: Measurement cycle, measurement starting point, measurement duration, and reporting format.

20. The method according to claim 19, wherein, The reporting format includes at least one of the following: The reported measurement value must be at least power or work. The reported measurements are quantified between upper and lower bounds of the measurements at equal or unequal intervals, which are obtained by predefinition.

21. The method according to claim 18, wherein, The receiving of control information reported by the first communication node includes one of the following: Periodically receive control information reported by the first communication node; In response to determining the sending and reporting instruction, the system receives control information reported by the first communication node.

22. The method of claim 18, further comprising: Receive auxiliary information for the model operation, the auxiliary information including at least one of the following: time information and level information of the model operation; The time information includes at least one of the following: period, duration, and activation start time.

23. The method of claim 18, further comprising: Send activation instruction or deactivation instruction; The activation instruction information is used to instruct the first communication node to run the model; The deactivation instruction is used to instruct the first communication node to stop running the model.

24. The method according to claim 18, wherein, The request model operation instruction includes at least one of the following instructions: request model activation, request model deactivation, request model configuration instruction, request model transmission, and request configuration power overhead level.

25. The method of claim 18, further comprising: The indication information for discontinuous operation cycles is determined, which is used to indicate the time when the model is in an active state and a dormant state.

26. The method of claim 18, further comprising: Determine the relationship between the control information and the influencing factors of the model; The influencing factors include at least one of the following: model type, model accuracy, model depth, model width, number of model parameters, and split points of a bilateral model.

27. The method according to claim 26, wherein, The relationship between the control information and the influencing factors of the model is determined by the second communication node, or obtained based on predefined information.

28. The method of claim 18, further comprising: The model is updated based on the control information; The updated model is sent to the first communication node.

29. The method of claim 18, further comprising: Configure and send the power overhead level; The power overhead level is used to assist the first communication node in determining the model based on the power overhead level.

30. The method of claim 26, further comprising: Based on the reported control information and / or the adjustment information of the model, the relationship between the control information and the influencing factors of the model is adaptively adjusted.

31. The method of claim 18, further comprising: Determine the random access resource configuration information, wherein the partition of the random access resource is associated with the control information of the model.

32. A communication node, comprising: Memory, and one or more processors; The memory is configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the model management method as described in any one of claims 1-17, or implement the model management method as described in any one of claims 18-31.

33. A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the model management method as described in any one of claims 1-17, or to implement the model management method as described in any one of claims 18-31.

Images