Method for model data management, and communication device

Abstract

Provided are a method for model data management, and a communication device. The method comprises: a first device sending first information to a second device, wherein the first information is associated with a database, and model data, which is stored in the database, is used for model inference and / or model training. In the embodiments of the present application, a database is used to store model data used for model training and / or model inference. In addition, a first device may communicate with a second device by means of first information associated with the database, which is conducive to combining the database with a communication system, so as to use the model data, which is stored in the database, to perform model training and / or model inference, thereby improving the accuracy of a model training process and / or a model inference process of a model in the communication system.

Description

Method and communication device for model data management Technical Field

[0001] The present application relates to the field of communication technology, and more specifically, to a method and communication device for model data management. Background Art

[0002] During the model training process, the model's performance is typically fixed after training. However, the training data used during model training may become obsolete over time, making the model based on this training data incapable of adapting to current business needs, leading to performance degradation or even model failure.

[0003] During the model inference process, if the model performs inference based solely on a single input data set, the inference accuracy may decrease. For example, in scenarios where users interact with the model, they may ask the model a series of related questions. In this case, if the model performs inference based on only one question without considering the context of the question, the inference accuracy may decrease. Therefore, performing inference based on the question and its associated context helps improve the accuracy of model inference.

[0004] Based on the above introduction, it can be seen that both the model training process and the model inference process require the maintenance of a large amount of model data (for example, training data, input data, etc.). However, how to maintain this model data is an urgent problem to be solved.

[0005] Summary of the Invention

[0006] The present application provides a method and communication device for model data management. The following introduces various aspects of the present application.

[0007] In a first aspect, a method for model data management is provided, comprising: a first device sending first information to a second device, wherein the first information is associated with a database, and model data stored in the database is used for model reasoning and / or model training.

[0008] In a second aspect, a method for model data management is provided, including: a second device receives first information sent by a first device, the first information is associated with a database, and the model data stored in the database is used for model reasoning and / or model training.

[0009] In a third aspect, a communication device is provided, which is a first device and includes: a sending unit for sending first information to a second device, wherein the first information is associated with a database, and the model data stored in the database is used for model inference and / or model training.

[0010] In a fourth aspect, a communication device is provided, which is a second device and includes: a receiving unit for receiving first information sent by a first device, wherein the first information is associated with a database, and the model data stored in the database is used for model inference and / or model training.

[0011] In a fifth aspect, a communication device is provided, comprising a processor, a memory, and a communication interface, wherein the memory is used to store one or more computer programs, and the processor is used to call the computer program in the memory so that the communication device executes part or all of the steps in the method of the first aspect.

[0012] In a sixth aspect, an embodiment of the present application provides a communication system, which includes the first device and / or the second device described above. In another possible design, the system may also include other devices that interact with the first device and / or the second device in the solution provided in the embodiment of the present application.

[0013] In the seventh aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and the computer program enables a communication device (for example, a first device and / or a second device) to perform some or all of the steps in the methods of the above aspects.

[0014] In an eighth aspect, embodiments of the present application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a communication device (e.g., a first device and / or a second device) to perform some or all of the steps of the methods described in each of the above aspects. In some implementations, the computer program product can be a software installation package.

[0015] In a ninth aspect, an embodiment of the present application provides a chip comprising a memory and a processor, wherein the processor can call and run a computer program from the memory to implement some or all of the steps described in the methods of the above aspects.

[0016] In an embodiment of the present application, a database is used to store model data used for model training and / or model reasoning. In addition, the first device can communicate with the second device through the first information associated with the database, which helps to combine the database and the communication system to realize model training and / or model reasoning using the model data stored in the database, thereby improving the accuracy of the model training and / or model reasoning process of the model in the communication system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] FIG1 is a wireless communication system 100 used in an embodiment of the present application.

[0018] FIG2 is a schematic diagram of channel estimation and signal recovery applicable to an embodiment of the present application.

[0019] FIG3 is a schematic diagram of a channel-state information (CSI) feedback system based on an artificial intelligence (AI) model applicable to an embodiment of the present application.

[0020] FIG4 shows a schematic diagram of an AI model-based positioning solution applicable to an embodiment of the present application.

[0021] Figure 5 shows a schematic diagram of AI model-based beam management applicable to an embodiment of the present application.

[0022] FIG6 is a schematic diagram of a neural network applicable to an embodiment of the present application.

[0023] FIG7 is a schematic diagram of a convolutional neural network (CNN) applicable to an embodiment of the present application.

[0024] FIG8 is a schematic flowchart of a method for model data management in an embodiment of the present application.

[0025] FIG9 is a schematic flowchart of a method for model data management according to another embodiment of the present application.

[0026] FIG10 is a schematic diagram of a communication device according to an embodiment of the present application.

[0027] FIG11 is a schematic diagram of a communication device according to an embodiment of the present application.

[0028] FIG12 is a schematic structural diagram of a communication device according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] The technical solution in this application will be described below with reference to the accompanying drawings.

[0030] Figure 1 illustrates a wireless communication system 100 used in an embodiment of the present application. The wireless communication system 100 may include a network device 110 and a terminal device 120. The network device 110 may be a device that communicates with the terminal device 120. The network device 110 may provide communication coverage for a specific geographic area and may communicate with the terminal device 120 within the coverage area.

[0031] FIG1 exemplarily shows a network device and two terminals. Optionally, the wireless communication system 100 may include multiple network devices and each network device may include other numbers of terminal devices within its coverage area, which is not limited in the embodiments of the present application.

[0032] Optionally, the wireless communication system 100 may further include other network entities such as a network controller and a mobility management entity, which is not limited in the embodiment of the present application.

[0033] It should be understood that the technical solutions of the embodiments of the present application can be applied to various communication systems, such as: fifth generation (5G) system or new radio (NR), long term evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD), etc. The technical solutions provided in this application can also be applied to future communication systems, such as the sixth generation mobile communication system, satellite communication system, etc.

[0034] The terminal device in the embodiments of the present application may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station (MS), mobile terminal (MT), remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent or user device. The terminal device in the embodiments of the present application may refer to a device that provides voice and / or data connectivity to a user and can be used to connect people, objects and machines, such as a handheld device with wireless connection function, a vehicle-mounted device, etc. The terminal device in the embodiments of the present application can be a mobile phone, a tablet computer, a laptop computer, a PDA, a mobile internet device (MID), a wearable device, a virtual reality (VR) device, an augmented reality (AR) device, a wireless terminal in industrial control, a wireless terminal in self-driving, a wireless terminal in remote medical surgery, a wireless terminal in a smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, a wireless terminal in a smart home, etc. Optionally, the UE can be used to act as a base station. For example, the UE can act as a scheduling entity that provides sidelink signals between UEs in V2X or D2D, etc. For example, a cellular phone and a car communicate with each other using sidelink signals. The cellular phone and smart home devices communicate without relaying the communication signal through the base station.

[0035] The network device in the embodiments of the present application may be a device for communicating with a terminal device, and may also be referred to as an access network device or a radio access network device. For example, the network device may be a base station. The network device in the embodiments of the present application may refer to a radio access network (RAN) node (or device) that connects a terminal device to a wireless network. A base station can broadly cover various names as follows, or be replaced with the following names, such as: NodeB, evolved NodeB (eNB), next generation NodeB (gNB), relay station, access point, transmission point (TRP), transmission point (TP), master station MeNB, secondary station SeNB, multi-standard radio (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), positioning node, etc. A base station can be a macro base station, a micro base station, a relay node, a donor node or the like, or a combination thereof. A base station can also refer to a communication module, a modem or a chip used to be set in the aforementioned device or apparatus. The base station can also be a mobile switching center and a device that performs base station functions in device-to-device D2D, vehicle-to-everything (V2X), and machine-to-machine (M2M) communications, a network-side device in a 6G network, or a device that performs base station functions in future communication systems. The base station can support networks with the same or different access technologies. The embodiments of this application do not limit the specific technology and specific device form used by the network equipment.

[0036] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move based on the location of the mobile base station. In other examples, a helicopter or drone can be configured to act as a device that communicates with another base station.

[0037] In some deployments, the network device in the embodiments of the present application may refer to a CU or a DU, or the network device may include a CU and a DU. The gNB may also include an AAU.

[0038] The network equipment and terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; they can also be deployed in the air on aircraft, balloons, and satellites. The embodiments of this application do not limit the scenarios in which the network equipment and terminal devices are located.

[0039] It should be understood that all or part of the functions of the communication device in this application can also be implemented through software functions running on hardware, or through virtualization functions instantiated on a platform (such as a cloud platform).

[0040] Channel estimation and signal recovery based on AI model

[0041] Due to the complexity and time-varying nature of wireless channel environments, in wireless communication systems (e.g., the wireless communication systems described above), a receiver needs to recover received signals based on channel estimation results. Figure 2 is a schematic diagram of channel estimation and signal recovery applicable to embodiments of the present application.

[0042] As shown in FIG2 , in step S210 , the transmitter transmits, in addition to the data signal, a series of pilot signals known to the receiver on the time-frequency resources, such as the channel state information-reference signal (CSI-RS) and the demodulation reference signal (DMRS).

[0043] In step S211, the transmitter transmits the above data signal and pilot signal to the transmitter through the channel.

[0044] In step S212, after receiving the pilot signal, the receiver may perform channel estimation. In one possible implementation, the receiver may estimate channel information of the channel transmitting the pilot signal based on a pre-stored pilot sequence and the received pilot sequence using a channel estimation algorithm (e.g., a least squares (LS) channel estimation method).

[0045] In step S213, the receiver may recover the channel information on all time-frequency resources using an interpolation algorithm based on the channel information of the channel transmitting the pilot sequence, for subsequent channel state information (CSI) feedback or data recovery.

[0046] CSI feedback system based on AI model

[0047] In wireless communication systems, codebook-based solutions are primarily used to extract and provide feedback on channel characteristics. This means that after the receiver performs channel estimation, it selects the precoding matrix that best matches the current channel from a pre-set precoding codebook based on the estimation results and an optimization criterion. The receiver then feeds the precoding matrix index (PMI) information back to the transmitter via an air interface feedback link for precoding. In some implementations, the receiver can also provide the measured channel quality indicator (CQI) back to the transmitter for adaptive modulation and coding.

[0048] Figure 3 is a schematic diagram of a CSI feedback system based on an AI model applicable to an embodiment of the present application. As shown in Figure 3, the entire feedback system includes an AI encoder 311 and an AI decoder 321 part of the autoencoder, wherein the AI ​​encoder 311 is deployed at the transmitter 310 and the AI ​​decoder 321 is deployed at the receiver 320. The transmitter 310 compresses and encodes the CSI to be transmitted through the AI ​​encoder 311 to obtain compressed CSI. The compressed CSI is then fed back to the receiver 320 through the feedback link, and the receiver 320 decodes the compressed CSI through the AI ​​decoder 321 to obtain the recovered CSI. In this way, the communication overhead of feedback CSI can be saved without affecting the accuracy of CSI transmission.

[0049] Positioning based on AI models

[0050] In cellular wireless positioning, the straight-line propagation of electromagnetic waves between network devices and terminal devices is called line-of-sight (LOS) wireless propagation. In some cases, electromagnetic wave signals cannot propagate in a straight line due to obstruction by buildings or trees, which is usually called non-line-of-sight (NLOS) wireless propagation. Traditional positioning algorithms such as time difference of arrival (TDOA) and angle-of-arrival (AOA) are based on LOS channels and are no longer applicable in environments where NLOS is predominant. In most scenarios, the number of network devices with LOS channels to terminal devices is often small, resulting in the inability of traditional positioning algorithms to meet the requirements of high-precision positioning. In addition, there may be some non-ideal factors in actual systems, which can lead to reduced positioning accuracy.

[0051] Therefore, a high-precision positioning method based on AI models has been proposed for scenarios where LOS / NLOS channels coexist. Existing research results have shown that by using machine learning methods to train models based on large amounts of channel data and to explore the mapping relationship between channel responses and location coordinates, it is possible to address the limitations of traditional positioning algorithms in LOS / NLOS channel coexistence scenarios and improve positioning accuracy.

[0052] FIG4 shows a schematic diagram of an AI model-based positioning solution applicable to an embodiment of the present application. Referring to FIG4 , in a positioning solution based on an AI model 410 in a LOS / NLOS channel coexistence scenario, the channel response can be used as the input of the AI ​​model 410, and the location coordinates can be used as the output of the AI ​​model 410. The AI ​​model 410 learns the intrinsic relationship between the wireless channel and the location of the terminal device. In this way, even in a scenario where there are not enough LOS channels and / or in a scenario where there are non-ideal conditions, the positioning solution based on the AI ​​model 410 can also output the location coordinates of the terminal device with higher accuracy, which helps to meet the needs of high-precision positioning.

[0053] The above introduces several communication processes applicable to the AI ​​model. The following introduces the AI ​​model applicable to the embodiments of the present application. It should be noted that the AI ​​model applicable to the embodiments of the present application is not limited to the several AI models introduced below.

[0054] AI-based beam management

[0055] In the traditional beam selection process, it is usually necessary to traverse all combinations of receive beams and transmit beams to select the appropriate beam. However, traversing all combinations takes a long time, resulting in low beam selection efficiency.

[0056] For example, suppose the network equipment deploys 64 different downlink transmission directions in FR2 (carried by up to 64 synchronization signals and physical broadcast channel blocks (SSB)). Accordingly, the terminal device uses one or more antenna panels to simultaneously scan the receiving beams when receiving, and each antenna panel has 4 receiving beams. Then the terminal device needs to measure at least 256 beam pairs, which means that 256 resources of downlink resource overhead are required. From a time perspective, each SSB cycle is approximately 20ms, and 4 SSB cycles are required to complete the measurement of 4 receiving beams. Assuming that multiple receiving antenna panels can perform beam scanning simultaneously, it will take at least 80ms.

[0057] As the number of beams in future massive multiple-input, multiple-output (MIMO) systems increases, using beam scanning-based beam management solutions to match optimal beam pairs will only result in increased reference signal transmission overhead and beam scanning latency. Therefore, to avoid these issues, AI-based beam management was proposed in Release 18. The following describes this AI-based beam management solution, combining the training and prediction processes of the AI ​​model.

[0058] Assume that the AI ​​model is used to predict the available beams in beam set A. Accordingly, during the training phase, the beam measurement results of beam set B can be used as AI model training data. That is, the AI ​​model is trained based on the beam measurement results of beam set B so that the AI ​​model can predict the available beams from beam set A.

[0059] It should be noted that the beam measurement results of the above-mentioned beam set B may include the measurement results corresponding to the layer 1 (layer1, L1) measurement quantity, and / or the indication information of the selected beam in beam set B (for example, the transmitting beam identifier, the receiving beam identifier or the beam pair identifier, etc.).

[0060] In some implementations, the training data may also include label information of beam set A, and the label information is used to indicate one or more of the following beams in beam set A: optimal transmit beam, optimal receive beam, optimal beam pair, better multiple transmit beams, better multiple receive beams, better beam pair, etc.

[0061] As shown in Figure 5, in the prediction stage, the input of the AI ​​model 510 may include the link quality measurement results (for example, L1 measurement quantity) corresponding to the beams in the beam set A, and the prediction results output by the AI ​​model 510 may include the target beam selected from the beam set A, and the link quality corresponding to the target beam.

[0062] In some implementations, the target beam may be one or more beams. For example, if the target beam is a single beam, the target beam may be the optimal beam or a relatively optimal beam in beam set A. For example, if the target beam is multiple beams, the target beam may be multiple beams in beam set A that meet the requirements. "Meeting the requirements" may be understood as meaning that the link quality corresponding to the beam meets the requirements, for example, the link quality corresponding to the beam is greater than or equal to a threshold.

[0063] In other implementations, the target beam may refer to one or more beam pairs, each of which may include a receive beam or a transmit beam. For example, if the target beam is a single beam pair, the target beam may be the optimal beam pair or a relatively optimal beam pair in beam set A. For example, if the target beam is multiple beam pairs, the target beam may be multiple beam pairs in beam set A that meet the requirements. "Meeting the requirements" may be understood as meaning that the link quality corresponding to the beam pair meets the requirements, for example, the link quality corresponding to the beam pair is greater than or equal to a threshold.

[0064] It should be noted that the link quality in the embodiment of the present application can be determined by one or more measurement quantities described above. Of course, the link quality in the embodiment of the present application can also be determined based on other measurement quantities in future communication systems, and the embodiment of the present application is not limited to this.

[0065] In addition, the link quality is determined based on one or more measurement quantities, which can be understood as the link quality being obtained by processing one or more measurement quantities. Of course, the link quality can also be a measurement quantity, which is not limited in the present embodiment.

[0066] It should also be noted that if the prediction result only indicates one beam in the beam pair, the other beam in the beam pair can be determined by other means. For example, it can be determined by one or some of the processes P1 to P3 in the traditional beam selection process. Of course, it can also be determined by one or some of the processes U1 to U3 in the traditional beam selection process. The embodiments of the present application are not limited to this.

[0067] In some implementations, the beam set B may be a different beam set from the beam set A. In some implementations, the beam set B may be a subset of the beam set A. Accordingly, by measuring fewer beams (beams in the beam set B), predictions for more beams (beams in the beam set A) may be achieved. Compared with the above-mentioned scheme of selecting beams based on traversing all combinations, it helps to reduce the time of executing the beam selection process. Of course, in the embodiment of the present application, the beams in the beam set B and the beams in the beam set A may be completely different beams. For example, there is no intersection between the beam set B and the beam set A, but the beam direction corresponding to the beam set B may be similar to the beam direction corresponding to the beam set A.

[0068] In some other implementations, the beam set B may be exactly the same as the beam set A.

[0069] The above describes the communication process applicable to the AI ​​model in conjunction with Figures 2 to 5. The embodiments of the present application are not limited to this. The AI ​​model can also be applied to other communication processes specified in future communication protocols. For example, the AI ​​model can also be used to select the target cell in cell switching. For another example, the AI ​​model can also be used to select the target cell during cell reselection. For another example, the AI ​​model can also be used to predict communication link failure. For another example, the AI ​​model can also be used for communication link recovery. The following describes the AI ​​model applicable to the embodiments of the present application in conjunction with Figures 6 to 7.

[0070] AI models

[0071] In recent years, artificial intelligence research, exemplified by neural networks, has achieved remarkable success in many fields, and will continue to play a vital role in people's lives and production for a long time to come. A neural network can be understood as a computational model consisting of multiple interconnected neuron nodes. The connections between these nodes represent the weighted values ​​from input signals to output signals, often referred to as weights. Each node performs a weighted summation of different input signals and outputs the result through a specific activation function.

[0072] Common neural networks include CNN, recurrent neural network (RNN), deep neural network (DNN), etc.

[0073] The following describes a neural network applicable to embodiments of the present application in conjunction with FIG6 . The neural network shown in FIG6 can be divided into three categories based on the location of different layers: input layer 610, hidden layer 620, and output layer 630. Generally speaking, the first layer is the input layer 610, the last layer is the output layer 630, and the intermediate layers between the first and last layers are all hidden layers 620.

[0074] The input layer 610 is used to input data, where the input data can be, for example, a received signal received by a receiver. The hidden layer 620 is used to process the input data, for example, decompress the received signal. The output layer 630 is used to output processed output data, for example, a decompressed signal.

[0075] As shown in Figure 6, a neural network consists of multiple layers, each of which contains multiple neurons. The neurons between layers can be fully connected or partially connected. For connected neurons, the output of the neurons in the previous layer can serve as the input of the neurons in the next layer.

[0076] With the continuous advancement of neural network research, deep learning algorithms have been proposed in recent years. These algorithms introduce a large number of hidden layers into neural networks, forming DNNs. More hidden layers allow DNNs to better capture complex real-world situations. Theoretically, a model with more parameters has higher complexity and a greater "capacity," meaning it can handle more complex learning tasks. These neural network models are widely used in pattern recognition, signal processing, optimization and combination, anomaly detection, and other fields.

[0077] CNN is a deep neural network with a convolutional structure, and its structure is shown in FIG7 , which may include an input layer 710 , a convolutional layer 720 , a pooling layer 730 , a fully connected layer 740 , and an output layer 750 .

[0078] Each convolution layer 720 may include a plurality of convolution operators, which are also called kernels. The convolution operator can be regarded as a filter for extracting specific information from the input signal. The convolution operator can essentially be a weight matrix, which is usually predefined.

[0079] The weight values ​​in these weight matrices need to be obtained through a lot of training in practical applications. The weight matrices formed by the weight values ​​obtained through training can extract information from the input signal, thereby helping CNN to make correct predictions.

[0080] When CNN has multiple convolutional layers, the initial convolutional layer tends to extract more general features, which can also be called low-level features. As the depth of CNN increases, the features extracted by the subsequent convolutional layers become more and more complex.

[0081] Pooling layers 730 are often used periodically after convolutional layers to reduce the number of training parameters. For example, a single convolutional layer can be followed by a pooling layer, as shown in Figure 7, or multiple convolutional layers can be followed by one or more pooling layers. In signal processing, the sole purpose of a pooling layer is to reduce the spatial size of the extracted information.

[0082] The fully connected layer 740, after being processed by the convolution layer 720 and the pooling layer 730, is not sufficient for CNN to output the required output information. Because as before, the convolution layer 720 and the pooling layer 730 only extract features and reduce the parameters brought by the input data. However, in order to generate the final output information (for example, the bit stream of the original information transmitted by the transmitter), CNN also needs to use the fully connected layer 740. Generally, the fully connected layer 740 may include multiple hidden layers, and the parameters contained in the multiple hidden layers may be pre-trained based on relevant training data of a specific task type. For example, the task type may include decoding the data signal received by the receiver. For another example, the task type may also include channel estimation based on the pilot signal received by the receiver.

[0083] Following the multiple hidden layers in the fully connected layer 740, the final layer of the CNN is the output layer 750, which is used to output the results. Typically, this output layer 750 is configured with a loss function (e.g., a loss function similar to categorical cross entropy) to calculate the prediction error, or to evaluate the degree of difference between the output of the CNN model (also known as the predicted value) and the ideal result (also known as the true value).

[0084] To minimize the loss function, the CNN model needs to be trained. In some implementations, the CNN model can be trained using the backpropagation algorithm (BP). The BP training process consists of a forward propagation process and a backward propagation process. During the forward propagation process (e.g., the propagation from 710 to 750 in Figure 7 is forward propagation), the input data is fed into the aforementioned layers of the CNN model, processed layer by layer, and transmitted to the output layer. If the output result of the output layer differs significantly from the ideal result, the aforementioned loss function is minimized as the optimization goal, and the backward propagation process is switched to (e.g., the propagation from 750 to 710 in Figure 7 is backward propagation). The partial derivatives of the optimization goal with respect to each neuron weight are calculated layer by layer, forming the gradient of the optimization goal with respect to the weight vector, which serves as the basis for modifying the model weights. The CNN training process is completed during the weight modification process. When the aforementioned error reaches the desired value, the CNN training process ends.

[0085] It should be noted that the CNN shown in Figure 7 is only an example of a convolutional neural network. In specific applications, the convolutional neural network can also exist in the form of other network models, and the embodiments of the present application are not limited to this.

[0086] RNNs are designed to process sequential data. In traditional neural network models (for example, CNN models), the layers are fully connected, from the input layer to the hidden layer to the output layer, and the nodes within each layer are disconnected. However, these ordinary neural networks are inadequate for many problems. For example, if you want to predict the next word in a sentence, you generally need to use the previous word, because the previous and next words in a sentence are not independent. RNNs are called recurrent neural networks because the current output of a sequence is also related to the previous output. Specifically, the network remembers the previous information and applies it to the calculation of the current output. That is, the nodes between hidden layers are no longer disconnected but connected, and the input of the hidden layer includes not only the output of the input layer but also the output of the hidden layer at the previous moment. In theory, RNNs can process sequence data of any length.

[0087] Training an RNN is similar to training a traditional ANN (artificial neural network). The same backpropagation error algorithm is used, but there is a slight difference. If the RNN is expanded, the parameters W, U, and V are shared, while traditional neural networks are not. Furthermore, when using the gradient descent algorithm, the output of each step depends not only on the network state at the current step, but also on the state of the network at the previous steps. For example, at t = 4, three steps need to be propagated backward, and various gradients need to be added to the three subsequent steps. This learning algorithm is called backpropagation through time (BPTT).

[0088] Given the existence of artificial neural networks and convolutional neural networks, why do we still need recurrent neural networks? The reason is simple. Both convolutional and artificial neural networks assume that elements are independent of each other, and that inputs and outputs are also independent, like cats and dogs. However, in the real world, many elements are interconnected, such as the changes in stock prices over time. For example, someone said, "I love traveling, and my favorite place is Yunnan. I must visit __ someday." Everyone knows to fill in the blank with "Yunnan." This is because we infer this information based on the context, but achieving this is quite difficult. Therefore, recurrent neural networks were developed. Their essence is that they possess memory, just like humans. Therefore, their output depends on the current input and memory.

[0089] During the model training process, the model's performance is typically fixed after training. However, the training data used during model training may become obsolete over time, making the model trained based on this data unable to adapt to current business needs, resulting in degraded model performance or even failure. This problem is particularly severe in offline model training scenarios. Currently, one solution is to retrain (or update) the model using newly acquired data as training data to improve its fit with the current environment. Alternatively, the model is trained using data that matches the current business to improve its fit with the current environment. This process of training the model based on newly acquired data (or data that matches the current business) is called "online model training." However, this online model training approach requires frequent model training, increasing training costs. Especially for large models, the cost of a single model training session can reach tens of millions of dollars, making frequent online model training unacceptable.

[0090] For the model reasoning process of the model, if the model only performs model reasoning based on a single input data, the accuracy of the model reasoning may be reduced. For example, in a scenario where a user interacts with an AI model, the user may ask the model a series of questions, and these questions are related. At this time, if the model only performs model reasoning for one of the questions and does not care about the context of the question (for example, other questions related to the question), the accuracy of the model reasoning may be reduced. Therefore, performing model reasoning based on the question and the context associated with the question can help improve the accuracy of model reasoning.

[0091] Based on the above introduction, we can see that both the model training process and the model inference process require maintaining a large amount of model data (for example, training data, input data, context of input data, etc.). However, how to maintain this model data is an urgent problem to be solved.

[0092] To address the above issues, the applicant proposes maintaining the model data based on a database. In some implementations, the database can be independent of a particular model, or it can store model data for multiple models. This allows model data to be retrieved from the database during model training and / or model inference for multiple models. This approach, where multiple models can share a single database, helps reduce model data maintenance costs.

[0093] In other implementations, the model data may be model data used during model training and / or model data generated during model training. For example, the model data may include input data used during model training. For another example, the model data may include intermediate data generated during model training. For another example, the model data may include output data output after model training. For another example, the model data may include ideal output data used during model training.

[0094] In other implementations, the above-mentioned model data may be model data used for model reasoning, and / or model data generated by model reasoning. For example, the model data may include input data used for model reasoning. For another example, the model data may include intermediate data generated during the model reasoning process. For another example, the model data may include output data output after model reasoning. For another example, the model data may include context data (e.g., user context data) used for model reasoning. Taking the user's context data as an example, assuming that the model is used to locate user 1, accordingly, the usual travel route of user 1 can be queried in the database as the above-mentioned user's context data, which helps to improve the accuracy of the model in locating user 1.

[0095] Taking the model for reference signal received power (RSRP) prediction services as an example, the model data of the model may include input data and output data, where the input data may be [-121dBm, -130dBm, -97dBm], and the output data may include [-99dBm]. Accordingly, the model data may be expressed as {[-121dBm, -130dBm, -97dBm]; [-99dBm]}.

[0096] In an embodiment of the present application, a database can be used to maintain model data to improve the accuracy of model training and / or model reasoning. For example, during the model reasoning phase, the input data of model 1 can be used to query the database to obtain model data associated with similar input data recorded in the database. The model data associated with the input data can include the input data of the model and the output data of the model. The model data associated with the similar input data is input into model 1 together with the input data of model 1 for model reasoning, which helps to improve the accuracy of model reasoning.

[0097] In some scenarios, a database can store model data used for model training and / or model inference over a long period of time, helping to improve the accuracy of model inference and / or model training. Therefore, this database can also be called a "long-term memory database."

[0098] After introducing a database to record the above model data, model reasoning and / or model training can be performed based on the model data in the database, which helps improve the accuracy of model reasoning and / or model training. However, how to integrate the database with the communication system has become a new problem.

[0099] Therefore, to address the above issues, an embodiment of the present application proposes a method for model data management, in which a first device in a communication system can communicate with a second device via first information associated with a database, thereby facilitating integration of the database and the communication system to improve the accuracy of model training and / or model inference processes in the communication system. Figure 8 is a schematic flow chart of the method for model data management in an embodiment of the present application. The method shown in Figure 8 includes step S810.

[0100] In step S810, the first device sends first information to the second device. In some implementations, the first information is associated with a database, wherein the model data stored in the database is used for inference and / or model training.

[0101] In some implementations, the model data may include the first model data and / or the second model data described below. In the embodiment of the present application, the model data may be any data stored in the database described above, which will not be described here for brevity.

[0102] In the embodiments of the present application, the functions of the first information may be different for different scenarios. The first information in the embodiments of the present application will be described below in conjunction with embodiments 1 to 3.

[0103] Example 1

[0104] In some implementations, the first information is used to request to call first model data from a database, or in other words, the first information is used to call a database to obtain first model data, wherein the first model data is used for model reasoning and / or model training.

[0105] In some implementations, the first information includes one or more of the following: an index identifier; a model identifier of the model; part or all of the input data of the model; and a usage time of the first model data.

[0106] Taking the example that the first information includes an index identifier, the index identifier is used to query the first model data in the database.

[0107] In some implementations, the index identifier is used to query first model data that matches (or is similar to) the model input data. The first model data that matches the model input data can be understood as matching the model input data with the input data of the first model data. In other words, first model data can be retrieved from the database based on the index identifier, where the input data corresponding to the first model data is similar to the model input data.

[0108] For example, the first model data includes output data and context data, and the input data corresponding to the first model data is similar to the input data of the model. At this time, during the process of model training, the first model data can be obtained from the database, and the model can be combined with the first model data and the input data of the model to perform model inference on the model, which helps to improve the accuracy of model training.

[0109] In some implementations, the index identifier can be used to identify part of the information in the input data of the model. For example, the index identifier can be used to identify keywords in the input data of the model. In this case, the input data corresponding to the first model data matches the input data of the model, which can be understood as the keywords of the input data corresponding to the first model data match the keywords of the input data of the model. Of course, in an embodiment of the present application, the index identifier can be used to identify all the information in the input data of the model. In this case, the input data corresponding to the first model data matches the input data of the model, which can be understood as the entire information of the input data corresponding to the first model data matches the entire information of the input data of the model.

[0110] In some scenarios, the index identifier may be one of multiple index identifiers, and the multiple index identifiers are used to query the first model data from the database. In this case, the setting of each index identifier can refer to the above introduction, and for the sake of brevity, it will not be repeated here.

[0111] In some implementations, the above-mentioned multiple index identifiers can be presented in the form of a list (also called an "index identifier list"). Of course, multiple index identifiers can be presented in the form of an index identifier set, which is not limited in the embodiments of the present application.

[0112] For example, assuming that the index identifier ID is 1, it can identify keyword 1 of the model input data, and keyword 1 corresponds to the column vector 01001. If the index identifier ID is 3, it can identify keyword 2 of the model input data, and keyword 2 corresponds to the column vector 10101. In this case, the first information can carry the index identifier list {1, 3} to identify the first model data corresponding to query keyword 1 and keyword 2, avoiding the large amount of transmission resources occupied by carrying {01001, 10101} in the first information.

[0113] In the embodiment of the present application, the first message carries an index identifier to indicate the input data of the model, which helps avoid transmitting the input data of the model itself, which would result in a large transmission overhead of the first message. Of course, in the embodiment of the present application, if the above problem is not considered, the input data of the model can be directly carried in the first message.

[0114] Furthermore, in the embodiments of the present application, there are no limitations on the correspondence between the index identifier and the model input data. For example, the correspondence may be preconfigured by the first device, or predefined by a protocol, which helps reduce the transmission resources occupied by transmitting the correspondence. Of course, if the above issues are not considered, the correspondence may also be configured by the network device.

[0115] In the embodiments of the present application, the matching and / or similarity mentioned above are not limited. In some implementations, the input data corresponding to the first model data may be exactly the same as the input data of the model. In other implementations, the input data corresponding to the first model data may be slightly different from the input data of the model. For example, the keywords of the input data corresponding to the first model data are the same as the keywords of the input data of the model.

[0116] Taking the example that the first information includes the model identification of the model, in some implementations, the model identification is used to identify the model, and therefore, the model identification can also be called a "model identifier".

[0117] As mentioned above, databases typically store data for multiple models, and different models may have different corresponding data due to their varying functions. Therefore, the first information can directly carry the model identifier of a model, allowing queries for the first model data within the model data range corresponding to that model. This narrows the query scope for the first model data in the database, helping to improve the efficiency of querying the first model data.

[0118] For example, the database stores first model data for model 1 and first model data for model 2. For model 1 used for beam management, its corresponding model identifier is model identifier 1, and the first model data corresponding to model 1 stored in the database includes user location information. For model 2 used for mobility management, its corresponding model identifier is model identifier 2, and the first model data corresponding to model 2 stored in the database also includes user location information.

[0119] In this case, the first message can carry model identifier 1 to indicate that the user's location information to be obtained is for model 1. This helps avoid querying the database for the location information of the user corresponding to model 2, thereby improving query efficiency. On the other hand, it helps avoid the possibility of obtaining the location information of the user corresponding to model 2, thereby reducing unnecessary data transmission.

[0120] Taking the example that the first information includes part or all of the input data of the model, in some implementations, the first information can carry all of the input data of the model, which helps to improve the accuracy of querying the first model data.

[0121] For example, the first model data can be obtained based on all the input data of the model, wherein the input data corresponding to the first model data matches the entire input data of the model, or the input data corresponding to the first model data is similar to the entire input data of the model. For details on matching and / or similarity, please refer to the above description and will not be repeated here for the sake of brevity. In addition, the input data corresponding to the first model data can be all or part of the input data, which is not limited in this embodiment of the present application.

[0122] In other implementations, the first information may carry part of the input data of the model, which helps to reduce the overhead of transmitting the first information.

[0123] For example, the first model data can be obtained based on partial input data of the model, wherein the input data corresponding to the first model data matches the partial input data of the model, or the input data corresponding to the first model data is similar to the partial input data of the model. For details on the matching and / or similarity, please refer to the above description, which will not be repeated here for the sake of brevity. In addition, the input data corresponding to the first model data can be all or part of the input data, which is not limited in the embodiments of the present application.

[0124] Assuming that all input data of the model includes a complete precoding matrix, to reduce the transmission overhead of the input data, a sparse precoding matrix corresponding to the precoding matrix can be transmitted as part of the input data. The sparse precoding matrix can be obtained from the complete precoding matrix through matrix transformation. Carrying the sparse precoding matrix in the first information helps reduce the amount of data required to be transmitted on the air interface, thereby reducing the overhead of transmitting the first information.

[0125] For example, if the first information includes the usage time of the first model data, the usage time of the first model data may be the time during which the model uses the first model data. Assuming that the first model data is used for model training of the model, the usage time may be the time during which the model performs model training. Assuming that the first model data is used for model inference of the model, the usage time may be the time during which the model performs model inference.

[0126] In the embodiment of the present application, there is no limitation on the usage time. For example, the usage time can be used to indicate the latest time to use the first model data, or the usage time can be used to indicate the deadline for obtaining the first model data. Accordingly, the second device can refer to the usage time when sending the first model data. If the first model data is not transmitted after the latest time, then even if the subsequent first model data is successfully transmitted, the model can no longer perform model training and / or model reasoning based on the first model data. Therefore, in this case, the first model data may no longer be transmitted to avoid unnecessary model data transmission. For another example, the time to use the first model data can be a time before the actual time to use the first model data, so that the first model data can be transmitted in advance to reduce the failure of model training and / or model reasoning due to the first model data not being transmitted in time.

[0127] In the embodiments of the present application, the implementation of the above-mentioned usage time is not limited. In some implementations, the usage time of the first model data can be regarded as a time threshold. Accordingly, if the time threshold is exceeded, the first model data may not be transmitted to avoid unnecessary data transmission. If the time threshold is not exceeded, the first model data may be attempted to be transmitted.

[0128] In other implementations, the usage time can be maintained by a timer. Accordingly, during the timer's operation, an attempt can be made to transmit the first model data. If the timer expires, the transmission of the first model data can be stopped. The start of the timer can be determined based on the transmission of the first information. For example, the start time of the timer can be the time when the first information is sent. In another example, the start time of the timer can be separated from the time when the first information is sent by a time interval.

[0129] It should be noted that the first information in the embodiment of the present application is introduced above. In some implementations, the above-mentioned first information can be used separately to reduce the overhead of transmitting the first information. In other implementations, the above-mentioned first information can be used in combination with each other to improve the efficiency of querying the first model data. For example, the first information may include an index identifier and a model identifier of the model, so that the first model data can be queried using the index identifier in the data set corresponding to the model to improve the efficiency of querying the first model data. For another example, the first information may include part of the input data of the model and the model identifier of the model, so that the first model data can be queried using part of the input data in the data set corresponding to the model to improve the efficiency of querying the first model data. For example, the first information may include all the input data of the model and the model identifier of the model, so that the first model data can be queried using all the input data in the data set corresponding to the model to improve the efficiency of querying the first model data.

[0130] In addition, in the embodiments of the present application, the triggering method of the first information is not limited. The triggering method of the first information can be the same as the triggering method of the fourth information introduced below. For details, please refer to the triggering method of the fourth information in Example 3, that is, the fourth information below can be replaced with the first information. For the sake of brevity, it will not be repeated here.

[0131] In some implementations, if the second device receives the first message, the second device may indicate whether the first model data was found in the database through a response message to the first message. In other words, the method further includes: the second device sending a response message to the first device, the response message indicating whether the first model data was found in the database.

[0132] In some implementations, if the first model data is found in the database, the first model data may be carried in the response information. It should be noted that if the first model data is found in the database, the response information may indirectly indicate that the first model data was found in the database by carrying the first model data, thereby reducing the overhead of transmitting the response information. Of course, if the above issue is not considered, the first model data and the information indicating that the first model data was found in the database may be independent information and carried in the response information.

[0133] In other implementations, if the first model data cannot be transmitted, this can be indicated through response information. In some scenarios, if the first model data is not found in the database, the response information can be used to indicate that the first model data is not found in the database. For example, the response information indicating that the first model data is not found in the database can occupy 1 bit to reduce the overhead of transmitting the response information. For another example, the response information of the first information can be NACK (NACK usually occupies 1 bit) to reduce the overhead of transmitting the response information. For another example, the response information of the first information can be empty to indicate that the first model data is not found in the database.

[0134] In other scenarios, the first information may indicate the usage time of the first model data. In this case, if the transmission time of the first model data is later than the usage time indicated by the first information, the second device may indicate through the response information that the first model data cannot be transmitted, or in other words, indicate through the response information that the first model data will no longer be transmitted. For example, the response information may occupy 1 bit to indicate that the first model data cannot be transmitted, so as to reduce the overhead of transmitting the response information. For another example, the response information of the first information may be NACK, to indicate that the first model data cannot be transmitted, so as to reduce the overhead of transmitting the response information. For another example, the response information of the first information may be empty to indicate that the first model data cannot be transmitted, so as to reduce the overhead of transmitting the response information.

[0135] For example, if model 1 is used for model inference based on RSRP, and all input data 1 of model 1 is [-120dBm, -130dBm, -97dBm]. In this case, the first device can send a first message to the second device, where the first message carries input data 1. Accordingly, the second device can query matching input data 2 based on input data 1: [-121dBm, -130dBm, -97dBm]. Furthermore, the output data corresponding to input data 2 is [-99dBm]. In this case, the second device can use {input data 2, output data} as the first model data and send it to the first device in a response message to the first message.

[0136] Example 2

[0137] In some implementations, the first information is used to request that first model data in a database be updated to second model data.

[0138] In some implementations, both the first model data and the second model data are used for model reasoning and / or model training. It should be understood that the second model data has a similar meaning to the first model data. For details, see the above description of model data. For the sake of brevity, this description is omitted here.

[0139] For example, if new output data (i.e., second model data) is obtained after model inference is performed on the model based on the first model data, the first model data in the database can be updated to the second model data through the first information, where the second model data includes the new output data.

[0140] For another example, the first model data includes input data and output data. If a new model is obtained after model training based on the first model data, and new output data is obtained after the first model data is input into the new model, the new output data is the output data in the second model data, and the input data of the second model data is the input data in the first model data. At this time, the first model data in the database can be updated to the second model data through the first information, wherein the second model data includes the new output data.

[0141] In some implementations, after the first model data is updated to the second model data, the database may continue to store the first model data to increase the diversity of the data stored in the database, thereby helping to improve the accuracy of model training and / or model inference. Of course, in an embodiment of the present application, after the first model data is updated to the second model data, the database may no longer store the first model data, or in other words, the database may delete the first model data to save database storage space.

[0142] In some implementations, the second model data is one of multiple pieces of model data, and the multiple pieces of model data can be used to perform model inference and / or model training on the model.

[0143] In the embodiments of the present application, there is no limitation on the multiple pieces of model data. In some implementations, the multiple pieces of model data may be obtained by the first device performing multiple model inferences using a model (also referred to as the "first model"). Alternatively, the multiple pieces of model data may be obtained by the first device performing multiple model inferences using a single model. In other implementations, the multiple pieces of model data may be obtained by the first device performing model inferences based on multiple models, respectively, where the multiple models are different models in the first device.

[0144] In some implementations, the update of the first model data (or the update of the database) is based on one or more of the following triggers: periodic triggering; event-based triggering; model reasoning process-based triggering; model training process-based triggering; or third information-based triggering.

[0145] Taking the update of the first model data based on periodic triggering as an example, that is, within one cycle, if the update time of the first model data has not arrived, the first device can first cache the second model data, and when the update time of the first model data arrives, the first model data is updated to the second model data by sending the first information.

[0146] In the embodiment of the present application, there is no limitation on the period for updating the first model data. For example, the period can be determined based on one or more of the following methods: pre-configuration, pre-definition, and network device configuration.

[0147] Taking the example of an event-triggered update of the first model data, in some implementations, the event may be associated with the size of the available storage space on the first device, where the available storage space includes storage space available for storing the model data. For example, the event may include the remaining storage space available for storing the model data on the first device being less than a first threshold. For another example, the event may include the storage space occupied by the model data on the first device being greater than a third threshold.

[0148] Of course, in an embodiment of the present application, the above-mentioned available storage space may be the available storage space of the first device as a whole (for example, storage space including model data of the first model and storage space of model data of other models). At this time, the above-mentioned event may, for example, include the remaining storage space of the available storage space being less than threshold 1, or, for example, the event may include the used storage space in the available storage space being greater than threshold 2.

[0149] In an embodiment of the present application, the thresholds mentioned above (one or more of the first threshold, the third threshold, threshold 1 and threshold 2) can be determined based on one or more of the following methods: pre-configuration, pre-definition and network device configuration.

[0150] In addition, the model data mentioned above may include first model data and / or second model data. Of course, in the embodiment of the present application, the above model data may include model data of other models in addition to the model data of the aforementioned models, and the embodiment of the present application does not limit this.

[0151] In an embodiment of the present application, if the above-mentioned event occurs, it may indicate that the available storage space in the first device is insufficient or will soon be insufficient. At this time, the first model data can be updated and the second model data can be stored in the database. Accordingly, the second model data can no longer be stored in the first device, which helps to reduce the used storage space in the first device.

[0152] In other implementations, the above event may be associated with the amount of model data stored in the first device. For example, the event may include that the amount of model data stored in the first device is greater than a second threshold.

[0153] In the embodiment of the present application, the above-mentioned model data may include model data of the above-mentioned model, for example, the first model data and / or the second model data. Of course, in the embodiment of the present application, in addition to including the model data of the model mentioned above, the above-mentioned model data may also include model data of other models, and the embodiment of the present application is not limited to this.

[0154] In addition, in the embodiment of the present application, the second threshold may be determined based on one or more of the following methods: pre-configuration, pre-definition, and network device configuration.

[0155] In an embodiment of the present application, if the above-mentioned event occurs, it may indicate that the available storage space in the first device is insufficient or will soon be insufficient. At this time, the first model data can be updated and the second model data can be stored in the database. Accordingly, the second model data can no longer be stored in the first device, which helps to reduce the used storage space in the first device.

[0156] For example, the update of the first model data is triggered by a model inference process. The model inference process can be understood as a model inference process based on the first model data. In other words, each time the model inference generates the second model data, an update of the first model data is triggered. This means that the first model data is updated to the second model data. In this case, the update of the first model data is understood to be real-time.

[0157] For example, the update of the first model data is triggered by the model training process. The model training process can be understood as a model training process based on the first model data. In other words, each time the model training generates the second model data, an update of the first model data can be triggered. This means that the first model data is updated to the second model data. In this case, the update of the first model data is understood to be real-time.

[0158] Taking the example of an update of the first model data being triggered by third information, in some implementations, the third information is used to request the first device to perform model training based on the first model data to update the first model data. The third information may be sent by the second device, meaning that the update of the first model data may be based on a request from the second device.

[0159] In the embodiments of the present application, the triggering method of the third information is not limited. For example, if the second device detects that the first model data has not been updated for a long period of time, the second device may request an update through the third information. For another example, if the second device detects that the performance of the model trained based on the first model data has degraded, the second device may request an update through the third information.

[0160] In some implementations, the first model data may be stored by the first device in a database. For example, after model training and / or model inference, the first device may send the first model data to the second device to instruct the second device to store the first model data in the database. Of course, in the embodiments of the present application, the first model data may also be stored in other ways, which are not limited in the embodiments of the present application.

[0161] It should be noted that the information for storing the first model data can be used in combination with the first information, or the information for storing the first model data can be used alone, and this is not limited in this embodiment of the present application. In addition, in this embodiment of the present application, the first model data can include one or more groups of model data, each group of model data can include input data and / or output data.

[0162] Example 3

[0163] In some implementations, if the first model data stored in the database is used for model training, the above step S810 may include: in response to the completion of model training (or the completion of model training based on the first model data), the first device sends first information to the second device, and the first information is used to indicate the deletion of the first model data, or the first information is used to indicate the release of the first model data.

[0164] In some implementations, the deleting of the first model data may be replaced by deleting the first model data from a database, or the releasing of the first model data may be replaced by releasing the first model data in a database.

[0165] In some implementations, the above method further includes: the second device may send response information to the first device with respect to the first information, where the response information is used to indicate whether to release the first model data.

[0166] In some implementations, if the first model data is not released or is not successfully released, the response information may indicate that the first model data is not released or is not successfully released. For example, the information indicating that the first model data is not released or is not successfully released may occupy 1 bit to indicate that the first model data is not released or is not successfully released, thereby reducing the overhead of transmitting the response information. For another example, the response information of the first information may be NACK (NACK usually occupies 1 bit) to indicate that the first model data is not released or is not successfully released, thereby reducing the overhead of transmitting the response information. For another example, the response information of the first information may be empty to indicate that the first model data is not released or is not successfully released.

[0167] In other implementations, if the first model data is released, the release of the first model data can be indicated by response information. For example, the information indicating the release of the first model data can occupy 1 bit to reduce the overhead of transmitting the response information. For another example, the response information of the first information can be ACK (ACK typically occupies 1 bit) to reduce the overhead of transmitting the response information.

[0168] In some implementations, before step S810, the method further includes: the first device sending fourth information to the second device, where the fourth information is used to request the first model data in the database. In this case, the fourth information can be understood as the first information applicable to the model training process described in Example 1. Of course, in the embodiments of the present application, the fourth information can also be completely different from the first information in Example 1.

[0169] In some implementations, the transmission of the fourth information is triggered based on one of the following: periodic triggering; event triggering; or fifth information triggering.

[0170] Taking the transmission of the fourth information based on periodic triggering as an example, that is, the first device may periodically send the fourth information to the second device to request the first model data.

[0171] In some implementations, it can be understood that the process of model training the model based on the first model data can be periodic, which helps to improve the performance of the model.

[0172] In the embodiment of the present application, the above period is not limited. For example, the period can be determined based on one or more of the following methods: pre-configuration, pre-definition, and network device configuration.

[0173] Taking the transmission of the fourth information based on an event trigger as an example, in some implementations, the event may include a failure to train the model for a period of time. In other implementations, the event may include a degradation in the model's performance. Of course, the above-mentioned event may also be other events preconfigured by the manufacturer of the first device, and this embodiment of the application is not limited to this.

[0174] Taking the example of the transmission of the fourth information being triggered based on the fifth information, the fifth information is used to instruct the first device to obtain the first model data. In some implementations, the fifth information instructs the first device to obtain the first model data by instructing the first device to perform model training based on the first model data.

[0175] In some implementations, the fifth information is triggered by an event. In some implementations, the event may be associated with the amount of available storage space in the database, where available storage space includes storage space available for storing model data. For example, the event may include the remaining storage space available for storing model data in the database being less than a threshold. In another example, the event may include the storage space occupied by storing model data in the database being greater than a threshold.

[0176] In the embodiment of the present application, the thresholds mentioned above may be determined based on one or more of the following methods: pre-configuration, pre-definition, and network device configuration.

[0177] In an embodiment of the present application, if the above-mentioned event occurs, it may indicate that the available storage space in the database is insufficient or will soon be insufficient. At this time, the first device can be requested to obtain the first model data through the fifth information. Accordingly, the first device can request the first model data by sending the fourth information. After that, the first device can perform model training on the model based on the first model data. If the model training is completed, the database can no longer store the first model data, which helps to reduce the used storage space in the database.

[0178] In other implementations, the above event may be associated with the amount of model data stored in the database. For example, the event may include that the amount of model data stored in the database is greater than a threshold.

[0179] In the embodiment of the present application, the above-mentioned model data may include the model data of the above-mentioned model, for example, the first model data. Of course, in the embodiment of the present application, in addition to the model data of the model mentioned above, the above-mentioned model data may also include model data of other models, and the embodiment of the present application is not limited to this.

[0180] In addition, in the embodiment of the present application, the thresholds mentioned above can be determined based on one or more of the following methods: pre-configuration, pre-definition, and network device configuration.

[0181] In an embodiment of the present application, if the above-mentioned event occurs, it may indicate that the available storage space in the database is insufficient or will soon be insufficient. At this time, the first device can be requested to obtain the first model data through the fifth information. Accordingly, the first device can request the first model data by sending the fourth information. After that, the first device can perform model training on the model based on the first model data. If the model training is completed, the database can no longer store the first model data, which helps to reduce the used storage space in the database.

[0182] For ease of understanding, the following describes a method for model data management according to an embodiment of the present application in conjunction with FIG9 . The method shown in FIG9 includes steps S910 to S950 . Assume that a model 1 for performing model inference based on RSRP is deployed in a first device, and a database is deployed in a second device.

[0183] In step S910 , the second device sends fifth information to the first device to instruct the first device to obtain first model data.

[0184] In some implementations, the triggering method of the fifth information can refer to the above introduction.

[0185] In step S920 , in response to the fifth information, the first device sends fourth information to the second device to request the first model data in the database.

[0186] In step S930, in response to the fourth information, the second device sends response information of the fourth information to the first device, and the response information of the fourth information may include the first model data.

[0187] In step S940 , the first device performs model training on Model 1 based on the first model data.

[0188] In step S950, in response to the completion of the model training, the first device sends a first message to the second device to instruct the second device to delete the first model data from the database.

[0189] The above describes the method for deleting the first model data in an embodiment of the present application in conjunction with Example 3. In other scenarios, if the second device instructs the first device to obtain the first model data, the first device may temporarily be unable to use the first model data to train the model. In this case, the first device can instruct the second device to refuse to obtain the first model data from the database.

[0190] In an embodiment of the present application, the above-mentioned scheme of refusing to obtain the first model data can be used as another case of Example 3, that is, if the second device instructs the first device to obtain the first model data through the fifth information, the first device may temporarily be unable to use the first model data to train the model. At this time, the first device can send a response message of the fifth information to the second device to indicate the refusal to obtain the first model data. Of course, in an embodiment of the present application, the above-mentioned scheme of refusing to obtain the first model data can be used alone, that is, if the second device instructs the first device to obtain the first model data through the sixth information, the first device may temporarily be unable to use the first model data to train the model. At this time, the first device can send the first information to the second device to indicate the refusal to obtain the first model data.

[0191] In the embodiments of the present application, the information indicating the refusal to obtain the first model data is not limited. For example, the information may occupy one bit to indicate that the first device refuses to obtain the first model data, thereby reducing the overhead of transmitting the response information. For another example, the information may be a NACK (NACK typically occupies one bit) to indicate that the first device refuses to obtain the first model data, thereby reducing the overhead of transmitting the response information.

[0192] In some implementations, if the first device refuses to obtain the first model data, the second device may autonomously manage the database. For example, the second device may release some model data from the database using a first-in, first-out method. For another example, the second device may randomly release some model data from the database. For another example, the second device may no longer store new model data. This is not limited in the present embodiments.

[0193] In some implementations, the first device may be a device on which a model is deployed, or the first device may be a device on which a model is required to be inferred and / or trained. In other implementations, the second device may be a device on which a database is deployed, or the second device may be a device that accesses (e.g., updates) the database. In the embodiments of the present application, the device type of the first device and / or the second device is not specifically limited. This will be described in detail below and will not be repeated here for the sake of brevity.

[0194] In the embodiments of the present application, the solutions of the above embodiments 1 to 3 can be used alone, or they can be used in combination with each other, which is not limited in the embodiments of the present application.

[0195] In the embodiment of the present application, the first device and the second device may be different devices, or the first device and the second device may be the same device, which is not limited in the embodiment of the present application.

[0196] In the embodiments of the present application, the first device and / or the second device are not limited. For example, the first device is a terminal device, and the second device is a network device. In another example, the first device is a network device, and the second device is a terminal device. In another example, the first device is a first terminal device, and the second device is a second terminal device. In another example, the first device is a first network device, and the second device is a second network device.

[0197] In the embodiment of the present application, if the second device is a terminal device, the database can be deployed on the terminal device. Generally, most measurement information is directly obtained by the terminal device, so deploying the database on the terminal device helps avoid frequent transmission of large amounts of model data.

[0198] In an embodiment of the present application, if the second device is a network device, the database can be deployed on the network device. Typically, a network device can provide services for multiple terminal devices. Therefore, model data corresponding to multiple terminal devices can be stored in the database, which helps to increase the diversity of model data in the database and improve the accuracy of model training and / or training inference.

[0199] The network device is an access network device, a core network device, an AI / ML model-related information management device, or an operation administration and maintenance (OAM) device. Exemplarily, the access network device is any one of the following: a gNB, a centralized unit (CU), a distributed unit (DU), a centralized unit-control plane (CU-CP), or a centralized unit-user plane (CU-UP).

[0200] Exemplarily, the core network device is any one of the following: location management function (LMF) network element, network slice selection function (NSSF), authentication server function (AUSF), unified data management (UDM), access and mobility management function (AMF), session management function (SMF), policy control function (PCF), user plane function (UPF), sensing control function (SF), network data analysis (NWDAF) network element.

[0201] In the embodiments of the present application, the models mentioned above are not limited. For example, the model can be an AI model, for example, any of the AI ​​models mentioned above, or a new model proposed by future technology. For another example, the model can be an ML model.

[0202] The method embodiment of the present application is described in detail above in conjunction with Figures 1 to 9 . The device embodiment of the present application is described in detail below in conjunction with Figures 10 to 12 . It should be understood that the description of the method embodiment corresponds to the description of the device embodiment. Therefore, for parts not described in detail, reference can be made to the above method embodiment.

[0203] FIG10 is a schematic diagram of a communication device according to an embodiment of the present application. The communication device 1000 shown in FIG10 is a first device, and the communication device 1000 includes: a sending unit 1010.

[0204] The sending unit 1010 is used to send first information to the second device, where the first information is associated with a database, and the model data stored in the database is used for model reasoning and / or model training.

[0205] In some implementations, the first information is used to request to call first model data from the database, and the first model data is used for the model to perform model inference and / or model training.

[0206] In some implementations, the first information includes one or more of the following: a model identifier of the model; an index identifier for querying the first model data; part or all of the input data of the model; and the time when the model uses the first model data.

[0207] In some implementations, the first information includes the index identifier, and the input data corresponding to the first model data queried based on the index identifier matches the input data of the model.

[0208] In some implementations, the communication device further includes: a first receiving unit, configured to receive response information to the first information sent by the second device, wherein the response information is used to indicate whether the first model data is found in the database.

[0209] In some implementations, if the first model data is found in the database, the response information carries the first model data.

[0210] In some implementations, the first information is used to request that the first model data in the database be updated to the second model data.

[0211] In some implementations, the first model data and the second model data are stored in the database.

[0212] In some implementations, the second model data is one of multiple model data, and the multiple model data are obtained by performing multiple model inferences on the first model in the first device; and / or the multiple model data are obtained by performing model inference on the first device based on multiple models respectively.

[0213] In some implementations, the update of the first model data is based on one or more of the following triggers: periodic triggering; event triggering; model reasoning process triggering; model training process triggering; or third information triggering, wherein the third information is used to request the first device to perform model training based on the first model data to update the first model data.

[0214] In some implementations, the update of the first model data is based on the event trigger, and the event includes one or more of the following: the remaining storage space for storing model data in the first device is less than a first threshold; the amount of model data stored in the first device is greater than a second threshold.

[0215] In some implementations, the first model data stored in the database is used for model training, and the sending unit is used to send the first information to the second device in response to the completion of the model training, where the first information is used to indicate deletion of the first model data.

[0216] In some implementations, the communication device further includes: a second receiving unit, configured to receive response information sent by the second device to the first information, wherein the response information is used to indicate whether to release the first model data.

[0217] In some implementations, before the first device sends the first information to the second device, the sending unit is configured to send fourth information to the second device, where the fourth information is used to request the first model data in the database.

[0218] In some implementations, the transmission of the fourth information is triggered based on one of the following: periodic triggering; or triggered based on fifth information, where the fifth information is used to instruct the first device to obtain the first model data.

[0219] In some implementations, the fourth information is triggered based on the fifth information, and the fifth information instructs the first device to obtain the first model data by instructing the first device to perform model training based on the first model data.

[0220] In some implementations, the fifth information is triggered based on an event, and the event includes a remaining storage space of the database being less than a threshold.

[0221] In some implementations, the first information is used to instruct the first device to refuse to obtain the first model data from the database, and the communication device further includes: a third receiving unit, used to receive sixth information sent by the second device, and the sixth information is used to instruct the first device to obtain the first model data.

[0222] FIG11 is a schematic diagram of a communication device according to an embodiment of the present application. The communication device 1100 shown in FIG11 is a second device, and the communication device 1100 includes a receiving unit 1110 .

[0223] The receiving unit 1110 is used to receive first information sent by a first device, where the first information is associated with a database, and the model data stored in the database is used for model inference and / or model training.

[0224] In some implementations, the first information is used to request to call first model data from the database, and the first model data is used for the model to perform model inference and / or model training.

[0225] In some implementations, the first information includes one or more of the following: a model identifier of the model; an index identifier for querying the first model data; part or all of the input data of the model; and the time when the model uses the first model data.

[0226] In some implementations, the first information includes the index identifier, and the input data corresponding to the first model data queried based on the index identifier matches the input data of the model.

[0227] In some implementations, the communication device further includes: a first sending unit, configured to send response information of the first information to the first device, wherein the response information is used to indicate whether the first model data is found in the database.

[0228] In some implementations, if the first model data is found in the database, the response information carries the first model data.

[0229] In some implementations, the first information is used to request that the first model data in the database be updated to the second model data.

[0230] In some implementations, the first model data and the second model data are stored in the database.

[0231] In some implementations, the second model data is one of multiple model data, and the multiple model data are obtained by performing multiple model inferences on the first model in the first device; and / or the multiple model data are obtained by performing model inference on the first device based on multiple models respectively.

[0232] In some implementations, the update of the first model data is based on one or more of the following triggers: periodic triggering; event-based triggering; model reasoning process-based triggering; model training process-based triggering; or third information-based triggering, wherein the third information is used to request the first device to perform model training based on the first model data to update the first model data.

[0233] In some implementations, the update of the first model data is triggered based on the event, and the event includes one or more of the following: the remaining storage space for storing model data in the first device is less than a first threshold; the amount of model data stored in the first device is greater than a second threshold.

[0234] In some implementations, the first model data stored in the database is used for model training, and the receiving unit is used to receive the first information sent by the first device in response to the completion of the model training, where the first information is used to indicate deletion of the first model data.

[0235] In some implementations, the communication device further includes: a second sending unit, configured to send response information to the first information to the first device, where the response information is used to indicate whether to release the first model data.

[0236] In some implementations, before the first device sends the first information to the second device, the receiving unit is further configured to receive fourth information sent by the first device, where the fourth information is used to request the first model data in the database.

[0237] In some implementations, the transmission of the fourth information is triggered based on one of the following: periodic triggering; or triggered based on fifth information, where the fifth information is used to instruct the first device to obtain the first model data.

[0238] In some implementations, the fourth information is triggered based on the fifth information, and the fifth information instructs the first device to obtain the first model data by instructing the first device to perform model training based on the first model data.

[0239] In some implementations, the fifth information is triggered based on an event, and the event includes a remaining storage space of the database being less than a threshold.

[0240] In some implementations, the first information is used by the first device to refuse to obtain the first model data from the database, and the communication device further includes: a third sending unit, used to send sixth information to the first device, and the sixth information is used to instruct the first device to obtain the first model data.

[0241] In an optional embodiment, the sending unit 1010 may be a transceiver 1230. The communication device 1000 may further include a transceiver 1230 and a memory 1220, as specifically shown in FIG12 .

[0242] In an optional embodiment, the receiving unit 1110 may be a transceiver 1230. The communication device 1100 may further include a transceiver 1230 and a memory 1220, as specifically shown in FIG12 .

[0243] Figure 12 is a schematic block diagram of a communication device according to an embodiment of the present application. The dashed lines in Figure 12 indicate that the unit or module is optional. The device 1200 may be used to implement the method described in the above method embodiment. The device 1200 may be a chip, a terminal device, or a network device.

[0244] The device 1200 may include one or more processors 1210. The processor 1210 may support the device 1200 to implement the method described in the above method embodiment. The processor 1210 may be a general-purpose processor or a special-purpose processor. For example, the processor may be a central processing unit (CPU). Alternatively, the processor may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0245] The apparatus 1200 may further include one or more memories 1220. The memories 1220 store programs that can be executed by the processor 1210, causing the processor 1210 to perform the methods described in the above method embodiments. The memories 1220 may be independent of the processor 1210 or integrated into the processor 1210.

[0246] The apparatus 1200 may further include a transceiver 1230. The processor 1210 may communicate with other devices or chips via the transceiver 1230. For example, the processor 1210 may transmit and receive data with other devices or chips via the transceiver 1230.

[0247] The present application also provides a computer-readable storage medium for storing a program. The computer-readable storage medium can be applied to a terminal or network device provided in the present application, and the program enables a computer to execute the method performed by the terminal or network device in each embodiment of the present application.

[0248] The present application also provides a computer program product. The computer program product includes a program. The computer program product can be applied to a terminal or network device provided in the present application, and the program causes a computer to execute the method performed by the terminal or network device in each embodiment of the present application.

[0249] The embodiments of the present application also provide a computer program. The computer program can be applied to the terminal or network device provided in the embodiments of the present application, and the computer program enables a computer to execute the method performed by the terminal or network device in each embodiment of the present application.

[0250] It should be understood that the terms "system" and "network" in this application can be used interchangeably. In addition, the terms used in this application are only used to explain the specific embodiments of this application and are not intended to limit this application. The terms "first", "second", "third", and "fourth" in the specification, claims, and drawings of this application are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.

[0251] In the embodiments of this application, the term "indication" may refer to a direct indication, an indirect indication, or an indication of an association. For example, "A indicates B" may refer to a direct indication of B, e.g., B can obtain information through A; it may refer to an indirect indication of B, e.g., A indicates C, e.g., B can obtain information through C; or it may refer to an association between A and B.

[0252] In the embodiment of the present application, "B corresponding to A" means that B is associated with A and B can be determined based on A. However, it should be understood that determining B based on A does not mean determining B based solely on A, but B can also be determined based on A and / or other information.

[0253] In the embodiments of the present application, the term "corresponding" may indicate a direct or indirect correspondence between the two, or an association relationship between the two, or a relationship between indication and indication, configuration and configuration, etc.

[0254] In the embodiments of the present application, "pre-definition" or "pre-configuration" may be implemented by pre-storing corresponding codes, tables, or other methods that can be used to indicate relevant information in a device (e.g., a terminal device and a network device). The present application does not limit the specific implementation method. For example, pre-definition may refer to information defined in a protocol.

[0255] In the embodiments of the present application, "protocol" may refer to a standard protocol in the field of communications, for example, it may include an LTE protocol, a NR protocol, and related protocols used in future communication systems, and this application does not limit this.

[0256] In the embodiments of this application, the term "and / or" is simply a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0257] In various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0258] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0259] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0260] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0261] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be read by a computer or a data storage device such as a server or data center that includes one or more available media integrated therein. Available media may be magnetic media (eg, floppy disks, hard disks, tapes), optical media (eg, digital versatile discs (DVDs)), or semiconductor media (eg, solid state disks (SSDs)).

[0262] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for model data management, characterized in that, it includes: A first device sends first information to a second device, the first information is associated with a database, and the model data stored in the database is used for model inference and / or model training.

2. The method according to claim 1, characterized in that, the first information is used to request to call first model data from the database, and the first model data is used for the model to perform model inference and / or model training.

3. The method according to claim 2, characterized in that, the first information includes one or more of the following: The model identifier of the model; The index identifier used to query the first model data; Part or all of the input data of the model; The time when the model uses the first model data.

4. The method according to claim 3, characterized in that, the first information includes the index identifier, and the input data corresponding to the first model data queried based on the index identifier matches the input data of the model.

5. The method according to any one of claims 2-4, characterized in that, the method further includes: The first device receives the response information of the first information sent by the second device, and the response information is used to indicate whether the first model data is queried in the database.

6. The method according to claim 5, characterized in that, if the first model data is queried in the database, the response information carries the first model data.

7. The method according to claim 1, characterized in that, the first information is used to request to update the first model data in the database to second model data.

8. The method according to claim 7, characterized in that, the first model data and the second model data are stored in the database.

9. The method according to claim 7 or 8, characterized in that, the second model data belongs to one of multiple model data, and the multiple model data is obtained by the first model in the first device performing multiple model inferences; and / or the multiple model data is obtained by the first device performing model inferences based on multiple models respectively.

10. The method according to any one of claims 7-9, characterized in that, the update of the first model data is triggered based on one or more of the following: Periodic trigger; Trigger based on an event; Trigger based on the model inference process; Trigger based on the model training process; Trigger based on third information, and the third information is used to request the first device to perform model training based on the first model data to update the first model data.

11. The method according to claim 10, characterized in that, the update of the first model data is triggered based on the event, and the event includes one or more of the following: The remaining storage space in the first device for storing model data is less than a first threshold; The data volume of the model data stored in the first device is greater than a second threshold.

12. The method according to claim 1, characterized in that, The first model data stored in the database is used for the model training, and the first device sends first information to the second device, including: In response to the end of the model training, the first device sends the first information to the second device, and the first information is used to instruct to delete the first model data.

13. The method according to claim 12, wherein, the method further includes: The first device receives response information sent by the second device for the first information, and the response information is used to indicate whether to release the first model data.

14. The method according to claim 12 or 13, wherein, before the first device sends the first information to the second device, the method further includes: The first device sends fourth information to the second device, and the fourth information is used to request the first model data in the database.

15. The method according to claim 14, wherein, the transmission of the fourth information is triggered by one of the following: Periodic trigger; Triggered based on fifth information, and the fifth information is used to instruct the first device to obtain the first model data.

16. The method according to claim 15, wherein, the fourth information is triggered based on the fifth information, and the fifth information instructs the first device to perform model training based on the first model data to instruct the first device to obtain the first model data.

17. The method according to claim 15 or 16, wherein, the fifth information is triggered based on an event, and the event includes that the remaining storage space of the database is less than a threshold.

18. The method according to claim 1, wherein, the first information is used to instruct the first device to refuse to obtain the first model data from the database. Before the first device sends the first information to the second device, the method further includes: The first device receives sixth information sent by the second device, and the sixth information is used to instruct the first device to obtain the first model data.

19. A method for model data management, wherein, includes: The second device receives first information sent by the first device, the first information is associated with a database, and the model data stored in the database is used for model inference and / or model training.

20. The method according to claim 19, wherein, the first information is used to request to call the first model data from the database, and the first model data is used for the model to perform model inference and / or model training.

21. The method according to claim 20, wherein, the first information includes one or more of the following: The model identifier of the model; The index identifier for querying the first model data; Part or all of the input data of the model; The time when the model uses the first model data.

22. The method according to claim 21, wherein, the first information includes the index identifier, and the input data corresponding to the first model data queried based on the index identifier matches the input data of the model.

23. The method according to any one of claims 20 - 22, characterized in that, the method further comprises: the second device sends response information of the first information to the first device, and the response information is used to indicate whether the first model data is queried in the database.

24. The method according to claim 23, characterized in that, if the first model data is queried in the database, the response information carries the first model data.

25. The method according to claim 19, characterized in that, the first information is used to request to update the first model data in the database to the second model data.

26. The method according to claim 25, characterized in that, the first model data and the second model data are stored in the database.

27. The method according to claim 25 or 26, characterized in that, the second model data belongs to one of multiple model data, and the multiple model data are obtained by performing multiple model inferences through the first model in the first device; and / or the multiple model data are obtained by performing model inferences through the first device based on multiple models respectively.

28. The method according to any one of claims 25 - 27, characterized in that, the update of the first model data is triggered based on one or more of the following: periodic trigger; event - based trigger; trigger based on the model inference process; trigger based on the model training process; trigger based on the third information, and the third information is used to request the first device to perform model training based on the first model data to update the first model data.

29. The method according to claim 28, characterized in that, the update of the first model data is triggered based on the event, and the event includes one or more of the following: the remaining storage space in the first device for storing model data is less than a first threshold; the data volume of the model data stored in the first device is greater than a second threshold.

30. The method according to claim 19, characterized in that, the first model data stored in the database is used for the model training, and the first device sends the first information to the second device, including: in response to the end of the model training, the second device receives the first information sent by the first device, and the first information is used to indicate deleting the first model data.

31. The method according to claim 30, characterized in that, the method further comprises: the second device sends response information for the first information to the first device, and the response information is used to indicate whether to release the first model data.

32. The method according to claim 30 or 31, characterized in that, before the first device sends the first information to the second device, the method further comprises: the second device receives the fourth information sent by the first device, and the fourth information is used to request the first model data in the database.

33. The method according to claim 32, characterized in that, the transmission of the fourth information is triggered based on one of the following: periodic trigger; Triggered based on the fifth piece of information, where the fifth piece of information is used to instruct the first device to obtain the first model data.

34. The method according to claim 33, wherein, the fourth piece of information is triggered based on the fifth piece of information, and the fifth piece of information instructs the first device to perform model training based on the first model data, so as to instruct the first device to obtain the first model data.

35. The method according to claim 33 or 34, wherein, the fifth piece of information is triggered based on an event, and the event includes that the remaining storage space of the database is less than a threshold.

36. The method according to claim 19, wherein, the first piece of information is used to instruct the first device to refuse to obtain the first model data from the database. Before the first device sends the first piece of information to the second device, the method further includes: the second device sends the sixth piece of information to the first device, and the sixth piece of information is used to instruct the first device to obtain the first model data.

37. A communication device, wherein, the communication device is the first device, including: a sending unit, configured to send the first piece of information to the second device, where the first piece of information is associated with a database, and the model data stored in the database is used for model inference and / or model training.

38. The communication device according to claim 37, wherein, the first piece of information is used to request to call the first model data from the database, and the first model data is used for the model to perform model inference and / or model training.

39. The communication device according to claim 38, wherein, the first piece of information includes one or more of the following: the model identifier of the model; the index identifier used to query the first model data; part or all of the input data of the model; the time when the model uses the first model data.

40. The communication device according to claim 39, wherein, the first piece of information includes the index identifier, and the input data corresponding to the first model data queried based on the index identifier matches the input data of the model.

41. The communication device according to any one of claims 38-40, wherein, the communication device further includes: a first receiving unit, configured to receive the response information of the first piece of information sent by the second device, and the response information is used to indicate whether the first model data is queried in the database.

42. The communication device according to claim 41, wherein, if the first model data is queried in the database, the response information carries the first model data.

43. The communication device according to claim 37, wherein, the first piece of information is used to request to update the first model data in the database to the second model data.

44. The communication device according to claim 43, wherein, the first model data and the second model data are stored in the database.

45. The communication device according to claim 43 or 44, wherein, The second model data belongs to one of multiple model data, and the multiple model data is obtained by performing multiple model inferences through a first model in the first device; and / or The multiple model data is obtained by the first device performing model inferences based on multiple models respectively.

46. The communication device according to any one of claims 43-45, wherein The update of the first model data is triggered based on one or more of the following: Periodic trigger; Trigger based on an event; Trigger based on the model inference process; Trigger based on the model training process; Trigger based on third information, where the third information is used to request the first device to perform model training based on the first model data to update the first model data.

47. The communication device according to claim 46, wherein The update of the first model data is triggered based on the event, and the event includes one or more of the following: The remaining storage space in the first device for storing model data is less than a first threshold; The data volume of the model data stored in the first device is greater than a second threshold.

48. The communication device according to claim 37, wherein The first model data stored in the database is used for the model training, The sending unit is configured to send the first information to the second device in response to the end of the model training, and the first information is used to instruct to delete the first model data.

49. The communication device according to claim 48, wherein The communication device further includes: A second receiving unit, configured to receive response information sent by the second device for the first information, and the response information is used to indicate whether to release the first model data.

50. The communication device according to claim 48 or 49, wherein Before the first device sends the first information to the second device, the sending unit is configured to send fourth information to the second device, and the fourth information is used to request the first model data in the database.

51. The communication device according to claim 50, wherein The transmission of the fourth information is triggered based on one of the following: Periodic trigger; Trigger based on fifth information, where the fifth information is used to instruct the first device to obtain the first model data.

52. The communication device according to claim 51, wherein The fourth information is triggered based on the fifth information, and the fifth information instructs the first device to perform model training based on the first model data to instruct the first device to obtain the first model data.

53. The communication device according to claim 51 or 52, wherein The fifth information is triggered based on an event, and the event includes that the remaining storage space of the database is less than a threshold.

54. The communication device according to claim 37, wherein The first information is used to instruct the first device to refuse to obtain the first model data from the database, and the communication device further includes: A third receiving unit, configured to receive sixth information sent by the second device, where the sixth information is used to instruct the first device to obtain the first model data.

55. A communication device, characterized in that the communication device is a second device, including: a receiving unit, configured to receive first information sent by a first device, where the first information is associated with a database, and model data stored in the database is used for model inference and / or model training.

56. The communication device according to claim 55, characterized in that the first information is used to request to call the first model data from the database, and the first model data is used for the model to perform model inference and / or model training.

57. The communication device according to claim 56, characterized in that the first information includes one or more of the following: a model identifier of the model; an index identifier used to query the first model data; part or all of the input data of the model; the time when the model uses the first model data.

58. The communication device according to claim 57, characterized in that the first information includes the index identifier, and the input data corresponding to the first model data queried by the index identifier matches the input data of the model.

59. The communication device according to any one of claims 56-58, characterized in that the communication device further includes: a first sending unit, configured to send a response message of the first information to the first device, where the response message is used to indicate whether the first model data is found in the database.

60. The communication device according to claim 59, characterized in that if the first model data is found in the database, the response message carries the first model data.

61. The communication device according to claim 55, characterized in that the first information is used to request to update the first model data in the database to second model data, and the second model data is used for the model to perform model inference and / or model training.

62. The communication device according to claim 61, characterized in that the first model data and the second model data are stored in the database.

63. The communication device according to claim 61 or 62, characterized in that the second model data is one of multiple model data, and the multiple model data is obtained by the first model in the first device performing multiple model inferences; and / or the multiple model data is obtained by the first device performing model inferences based on multiple models respectively.

64. The communication device according to any one of claims 61-63, characterized in that the update of the first model data is triggered based on one or more of the following: periodic triggering; event-based triggering; triggering based on the model inference process; triggering based on the model training process; triggering based on third information, where the third information is used to request the first device to perform model training based on the first model data to update the first model data.

65. The communication device according to claim 64, characterized in that The update of the first model data is triggered based on the event, and the event includes one or more of the following: The remaining storage space in the first device for storing model data is less than a first threshold; The data volume of the model data stored in the first device is greater than a second threshold.

66. The communication device according to claim 55, wherein, The first model data stored in the database is used for model training. The receiving unit is configured to receive the first information sent by the first device in response to the end of the model training, and the first information is used to indicate deleting the first model data.

67. The communication device according to claim 66, wherein, The communication device further includes: A second sending unit, configured to send response information for the first information to the first device, and the response information is used to indicate whether to release the first model data.

68. The communication device according to claim 66 or 67, wherein, The receiving unit is further configured to receive the fourth information sent by the first device, and the fourth information is used to request the first model data in the database.

69. The communication device according to claim 68, wherein, The transmission of the fourth information is triggered based on one of the following: Periodic trigger; Triggered based on the fifth information, and the fifth information is used to indicate that the first device acquires the first model data.

70. The communication device according to claim 69, wherein, The fourth information is triggered based on the fifth information, and the fifth information indicates that the first device performs model training based on the first model data to indicate that the first device acquires the first model data.

71. The communication device according to claim 69 or 70, wherein, The fifth information is triggered based on an event, and the event includes that the remaining storage space of the database is less than a threshold.

72. The communication device according to claim 55, wherein, The first information is used to instruct the first device to refuse to acquire the first model data from the database. The communication device further includes: A third sending unit, configured to send sixth information to the first device, and the sixth information is used to instruct the first device to acquire the first model data.

73. A communication device, wherein, It includes a transceiver, a memory, and a processor. The memory is used to store programs, and the processor is used to call the programs in the memory and control the transceiver to receive or send signals, so that the communication device executes the method according to any one of claims 1-36.

74. A device, wherein, It includes a processor, configured to call a program from a memory, so that the device executes the method according to any one of claims 1-36.

75. A chip, wherein, It includes a processor, configured to call a program from a memory, so that the device installed with the chip executes the method according to any one of claims 1-36.

76. A computer-readable storage medium, wherein, A program is stored thereon, and the program causes a computer to execute the method according to any one of claims 1-36.

77. A computer program product, characterized in that, it includes a program which causes a computer to execute the method according to any one of claims 1-36.

78. A computer program, characterized in that, the computer program causes a computer to execute the method according to any one of claims 1-36.

Images