Method and apparatus for updating model, and terminal, network-side device and medium

Abstract

The present application belongs to the technical field of communications. Disclosed are a method and apparatus for updating a model, and a terminal, a network-side device and a medium. The method for updating a model in the embodiments of the present application comprises: a terminal acquiring first information, which first information comprises: information related to a first target layer of a first AI unit after model input dimensions are updated, wherein the first target layer is used for determining or updating application configurations of the first AI unit; and the terminal updating the first AI unit in the terminal on the basis of the first information.

Description

Model updating method, device, terminal, network side equipment and medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Chinese patent application number 202311870273.7 filed in China on December 29, 2023, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present application belongs to the field of communication technology, and specifically relates to a model updating method, apparatus, terminal, network-side equipment, and medium. Background Art

[0004] Currently, updating the model input dimensions supported by the Channel State Information (CSI) compression model can be achieved by updating the CSI compression model. The specific methods for updating the CSI compression model generally include any of the following: 1. The network device retransmits the updated CSI compression model encoder to the terminal, or the terminal retransmits the updated CSI compression model decoder to the network device; 2. The network device and the terminal retrain the CSI compression model; 3. The network device and the terminal retransmit relevant model information.

[0005] However, according to the above method, since the encoder, decoder or related model information of the CSI compression model is re-transmitted, a large air interface overhead is required, and the overhead of the above joint training is much greater than the overhead of model transmission, the overhead of updating the model input dimensions supported by the model is relatively large. Summary of the Invention

[0006] The embodiments of the present application provide a model updating method, apparatus, terminal, network-side equipment, and medium, which can solve the problem of high overhead in updating the input dimensions supported by the model.

[0007] In a first aspect, a model updating method is provided, which is executed by a terminal, and the method includes: the terminal obtains first information, the first information including: information related to a first target layer of a first artificial intelligence (AI) unit after updating a model input dimension, the first target layer being used to determine or update an applicable configuration of the first AI unit; the terminal updates the first AI unit in the terminal according to the first information.

[0008] In a second aspect, a model updating method is provided, which is executed by a network side device, and the method includes: the network side device sends target information to the terminal, the target information is used for the terminal to obtain first information, the first information is used to update the first AI unit in the terminal, and the first information includes: information related to the first target layer of the first AI unit after the model input dimension is updated, and the first target layer is used to determine or update the applicable configuration of the first AI unit.

[0009] In a third aspect, a model updating device is provided, which includes an acquisition module and an update module; the acquisition module is used to obtain first information, the first information including: information related to the first target layer of the first AI unit after updating the model input dimension, the first target layer is used to determine or update the applicable configuration of the first AI unit; the update module is used to update the first AI unit in the terminal according to the first information obtained by the acquisition module.

[0010] In a fourth aspect, a model updating device is provided, which includes a sending module; the sending module is used to send target information to the terminal, the target information is used for the terminal to obtain first information, the first information is used to update the first AI unit in the terminal, and the first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit.

[0011] In a fifth aspect, a terminal is provided, comprising a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method described in the first aspect are implemented.

[0012] In a sixth aspect, a terminal is provided, comprising a processor and a communication interface, wherein the processor is used to obtain first information, the first information comprising: information related to a first target layer of the first AI unit after updating the model input dimension, the first target layer being used to determine or update the applicable configuration of the first AI unit; and updating the first AI unit in the terminal based on the obtained first information.

[0013] In the seventh aspect, a network side device is provided, which includes a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the program or instructions are executed by the processor, the steps of the method described in the second aspect are implemented.

[0014] In an eighth aspect, a network side device is provided, comprising a processor and a communication interface, wherein the communication interface is used to send target information to a terminal, the target information is used for the terminal to obtain first information, the first information is used to update a first AI unit in the terminal, and the first information includes: information related to a first target layer of the first AI unit after updating the model input dimension, the first target layer is used to determine or update an applicable configuration of the first AI unit.

[0015] In the ninth aspect, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method described in the second aspect are implemented.

[0016] In the tenth aspect, a wireless communication system is provided, comprising: a terminal and a network side device, wherein the terminal can be used to execute the steps of the method described in the first aspect, and the network side device can be used to execute the steps of the method described in the second aspect.

[0017] In the eleventh aspect, a chip is provided, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the method as described in the first aspect, or to implement the method as described in the second aspect.

[0018] In the twelfth aspect, a computer program / program product is provided, which is stored in a storage medium and is executed by at least one processor to implement the steps of the method described in the first aspect, or to implement the steps of the method described in the second aspect.

[0019] In an embodiment of the present application, a terminal can obtain first information, the first information including: information related to a first target layer of a first AI unit after updating a model input dimension, the first target layer being used to determine or update an applicable configuration of the first AI unit; and based on the first information, the first AI unit in the terminal is updated. Through this solution, since the terminal can update the first AI unit in the terminal based on the information related to the first target layer of the first AI unit after updating the model input dimension to update the model input dimension supported by the first AI unit, and the first target layer used to determine or update the applicable configuration of the first AI unit is generally simpler in structure and has fewer related parameters, the overhead required to update the model input dimensions supported by the first AI unit based on the information related to the first target layer is much less than the overhead required to update the model in related technologies, thereby reducing the overhead of updating the model input dimensions supported by the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] FIG1 is a block diagram of a wireless communication system applicable to embodiments of the present application;

[0021] FIG2 is a schematic diagram of the structure of a neural network in the related art;

[0022] FIG3 is a schematic diagram of neurons constituting a neural network in the related art;

[0023] FIG4 is a flow chart of a model updating method provided in an embodiment of the present application;

[0024] FIG5 is a schematic diagram of an AI CSI compression model with an adaptation layer in a model updating method provided in an embodiment of the present application;

[0025] FIG6 is a schematic diagram showing a method for updating a model provided by an embodiment of the present application, in which an adaptation layer includes an encoder input;

[0026] FIG7 is a second schematic diagram of an adaptation layer including encoder input in a model updating method provided by an embodiment of the present application;

[0027] FIG8 is a schematic diagram showing a model updating method provided by an embodiment of the present application in which the adaptation layer does not include an encoder input;

[0028] FIG9 is a second schematic diagram of a model updating method provided in an embodiment of the present application in which the adaptation layer does not include an encoder input;

[0029] FIG10 is a third schematic diagram of a model updating method provided in an embodiment of the present application in which the adaptation layer does not include an encoder input;

[0030] FIG11 is a fourth schematic diagram of a model updating method provided in an embodiment of the present application in which the adaptation layer does not include an encoder input;

[0031] FIG12 is a flowchart of another model updating method provided in an embodiment of the present application;

[0032] FIG13 is a schematic structural diagram of a model updating device provided in an embodiment of the present application;

[0033] FIG14 is a schematic structural diagram of another model updating device provided in an embodiment of the present application;

[0034] FIG15 is a schematic diagram of a communication device provided in an embodiment of the present application;

[0035] FIG16 is a schematic diagram of the hardware structure of a terminal provided in an embodiment of the present application;

[0036] FIG17 is a schematic diagram of the hardware structure of the network side device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will be combined with the accompanying drawings in the embodiments of this application to clearly describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0038] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same type, and do not limit the number of objects, for example, the first object can be one or more. In addition, "or" in this application represents at least one of the connected objects. For example, "A or B" covers three options, namely, Option 1: including A but not including B; Option 2: including B but not including A; Option 3: including both A and B. The character " / " generally indicates that the objects associated before and after are in an "or" relationship.

[0039] The term "indication" in this application can be either a direct indication (or explicit indication) or an indirect indication (or implicit indication). A direct indication can be understood as the sender explicitly informing the receiver of specific information, the operation to be performed, or the requested result, etc. in the instruction sent; an indirect indication can be understood as the receiver determining the corresponding information based on the instruction sent by the sender, or making a judgment and determining the operation to be performed or the requested result, etc. based on the judgment result.

[0040] It is worth noting that the technology described in the embodiments of the present application is not limited to the Long Term Evolution (LTE) / LTE-Advanced (LTE-A) system, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA) or other systems. The terms "system" and "network" in the embodiments of the present application are often used interchangeably, and the technology described can be used for the systems and radio technologies mentioned above, as well as for other systems and radio technologies. The following description describes a New Radio (NR) system for illustrative purposes, and NR terminology is used in most of the following description, but these technologies can also be applied to systems other than NR systems, such as 6th generation (6G) systems. th Generation, 6G) communication system.

[0041] FIG1 is a block diagram of a wireless communication system applicable to an embodiment of the present application. The wireless communication system includes a terminal 11 and a network-side device 12. The terminal 11 may be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), a notebook computer, a personal digital assistant (PDA), a handheld computer, a netbook, an ultra-mobile personal computer (UMPC), a mobile internet device (MID), an augmented reality (AR), a virtual reality (VR) device, a robot, a wearable device (Wearable Device), an aircraft (Flight Vehicle), a vehicle-mounted device (VUE), a ship-mounted device, a pedestrian user equipment (PUE), a smart home (home appliances with wireless communication capabilities, such as refrigerators, televisions, washing machines, or furniture), a game console, a personal computer (PC), an ATM, or a self-service machine, or other terminal-side devices. Wearable devices include: smart watches, smart bracelets, smart headphones, smart glasses, smart jewelry (smart bracelets, smart bracelets, smart rings, smart necklaces, smart anklets, smart anklets, etc.), smart wristbands, smart clothing, etc. Among them, the vehicle-mounted device can also be called a vehicle-mounted terminal, a vehicle-mounted controller, a vehicle-mounted module, a vehicle-mounted component, a vehicle-mounted chip or a vehicle-mounted unit, etc. It should be noted that the specific type of the terminal 11 is not limited in the embodiment of the present application. The network side device 12 may include an access network device or a core network device, wherein the access network device may also be called a radio access network (Radio Access Network, RAN) device, a radio access network function or a radio access network unit. The access network device may include a base station, a wireless local area network (WLAN) access point (AS) or a wireless fidelity (WiFi) node, etc.Among them, the base station can be referred to as Node B (NB), Evolved Node B (eNB), the next generation Node B (gNB), New Radio Node B (NR Node B), access point, Relay Base Station (RBS), Serving Base Station (SBS), Base Transceiver Station (BTS), radio base station, radio transceiver, Basic Service Set (BSS), Extended Service Set (ESS), Home Node B (HNB), Home evolved Node B (home evolved Node B), Transmission Reception Point (TRP) or other appropriate terms in the relevant field. As long as the same technical effect is achieved, the base station is not limited to specific technical vocabulary. It should be noted that in the embodiment of the present application, only the base station in the NR system is used as an example for introduction, and the specific type of the base station is not limited.

[0042] The following, in combination with the accompanying drawings, describes in detail the model updating method, apparatus, terminal, network-side equipment and medium provided in the embodiments of the present application through some embodiments and their application scenarios.

[0043] According to information theory, accurate CSI is crucial to channel capacity. Especially for multi-antenna systems, the transmitter can optimize signal transmission based on CSI to better match the channel state. For example, CSI can include channel quality indicator (CQI) and precoding matrix indicator (PMI). CQI can be used to select an appropriate modulation and coding scheme (MCS) for link adaptation, while PMI can be used to implement eigenbeamforming to maximize the strength of the received signal or to suppress interference such as inter-cell interference or interference between multiple users. Therefore, since the introduction of multi-input multi-output (MIMO) technology, the acquisition of accurate CSI has been a research hotspot.

[0044] Typically, a base station sends a Channel State Information Reference Signal (CSI-RS) on certain time-frequency resources in a certain time slot. The terminal then performs channel estimation based on the received CSI-RS, calculates the channel information for this slot, and feeds the PMI back to the base station using a codebook. The base station can then combine the channel information from the codebook feedback from the terminal and use it for data precoding and multi-user scheduling before the next CSI report.

[0045] To reduce CSI feedback overhead, the terminal can change the PMI reported for each subband to reporting it based on delay (i.e., delay). Since the channels in the delay domain are more concentrated, only PMIs with fewer delays can approximately represent the PMIs of all subbands. This means that the delay domain information is compressed before reporting. Similarly, to reduce overhead, the base station can precode the CSI-RS and send the encoded CSI-RS to the terminal. The terminal receives the channel corresponding to the encoded CSI-RS and only needs to select several ports with higher strength from the ports indicated by the network and report the coefficients corresponding to these ports.

[0046] Furthermore, in order to better compress channel information, neural network or machine learning methods can be used.

[0047] AI is currently widely used in various fields. Among them, AI modules can be implemented in various ways, such as neural networks, decision trees, support vector machines, Bayesian classifiers, etc. Taking the AI ​​module as a neural network as an example, Figure 2 shows a schematic diagram of the structure of a neural network. As shown in Figure 2, the neural network is composed of neurons; Figure 3 shows a schematic diagram of neurons in the neural network. As shown in Figure 3, a is the input, w is the weight (multiplicative coefficient), b is the bias (additive coefficient), and σ(.) is the activation function. Among them, common activation functions include Sigmoid (i.e., S-shaped growth curve), tanh (i.e., hyperbolic tangent function), linear rectifier function (Rectified Linear Unit, ReLU), etc.

[0048] Neural network parameters can be optimized using gradient optimization algorithms. Gradient optimization algorithms are a class of algorithms that minimize or maximize an objective function (also known as a loss function), which is often a mathematical combination of model parameters and data. For example, given data x and its corresponding label Y, a neural network model f(.) is constructed. Through this neural network model f(.), x can be input to obtain a predicted output f(x). The difference between the predicted value and the true value (f(x) - Y) can be calculated. This is the loss function. Appropriate values ​​for w and b can minimize the value of this loss function. The smaller the loss value, the closer the neural network model is to the true situation.

[0049] Currently, the most common optimization algorithms for neural network parameters are based on the back propagation (BP) algorithm. The basic idea of ​​the BP algorithm is that the learning process consists of two steps: forward propagation of signals and back propagation of errors. During forward propagation, input samples are passed from the input layer, processed layer by layer through each hidden layer, and then transmitted to the output layer. If the actual output of the output layer does not match the expected output, the error begins back propagation. Back propagation of errors involves propagating the output error back through the hidden layers to the input layer in some form, distributing the error to all units in each layer, thereby obtaining an error signal for each unit in each layer. This error signal serves as the basis for adjusting the weights of each unit. This process of adjusting the weights of each layer, including forward propagation of signals and back propagation of errors, is repeated repeatedly. This process of continuous weight adjustment is the network's learning and training process, which continues until the error in the network output is reduced to an acceptable level or until a pre-set number of learning cycles is reached.

[0050] Common optimization algorithms include Gradient Descent (GD), Stochastic Gradient Descent (SGD), mini-batch gradient descent, Momentum, Nesterov (specifically Stochastic Gradient Descent with momentum), Adaptive Gradient Descent (Adagrad), Adadelta, Root Mean Square Error (RMSprop), and Adaptive Moment Estimation (Adam). During error backpropagation, these optimization algorithms calculate the derivative / partial derivative of the current neuron based on the error / loss obtained from the loss function, add the learning rate, previous gradients / derivatives, and partial derivatives, and finally obtain the gradient, which is then passed to the previous layer.

[0051] Specifically, the channel information can be compressed and encoded on the terminal side, and the compressed content can be decoded on the base station side to recover the channel information. In this case, the base station decoder and the terminal encoder need to be jointly trained to achieve a reasonable match. The terminal encoder and the base station decoder form a joint neural network, which is jointly trained by the network. After training is complete, the base station sends the encoder network to the terminal. After estimating the CSI-RS and calculating the channel information, the terminal can pass the calculated channel information or the original estimated channel information through the encoder to obtain the encoding result and send the encoding result to the base station. The base station receives the encoded result and inputs it into the decoder to recover the channel information.

[0052] The following describes in detail the training method of the CSI compression model.

[0053] The CSI compression model is a typical two-end model use case. This means that the complete CSI compression model needs to be deployed on different network nodes. Currently, the encoder is typically deployed on the terminal, and the decoder is deployed on the network end. Models or sub-models deployed on multiple nodes must be paired to function properly. Considering the characteristics of two-end models, the following basic types of training collaboration have been identified:

[0054] 1. Joint training at single entity (also known as type 1)

[0055] The joint training framework on a single node refers to training a complete encoder and decoder model on a certain network node (for example, a terminal, a network, or a third-party server node, etc.), and then deploying the corresponding model module to the target node through methods such as model transfer, for example, transferring the encoder part to the terminal and the decoder part to the network.

[0056] 2. Joint training at multiple entities (also known as type 2)

[0057] A multi-node joint training framework involves multiple nodes participating in the training process, with each node independently calculating the forward and backpropagation information required for local model training and updating its own model parameters. Because the training process requires forward and backpropagation of the entire model, including the encoder and decoder, the participating nodes must transfer the corresponding forward and backpropagation information. After training is complete, the model no longer needs to be transferred between nodes.

[0058] 3. Separate (or step-by-step) training on multiple nodes (separate training, also known as type 3)

[0059] A separate (or step-by-step) training framework on multiple nodes involves first training a reference model on a specific node, then sending information about the reference model to the target node. The target node then uses this information to train the model it needs, ensuring that the models or sub-models on each node can be used in conjunction with each other. For example, the network first trains a complete encoder-decoder model, determines that the resulting decoder is the one that will actually be used, and then sends information about the encoder corresponding to that decoder (typically, the encoder's input and output data) to the terminal. The terminal then uses this information to train its own encoder. This training framework can be further divided into two scenarios: terminal-first training (i.e., UE-first training) and network-first training (i.e., NW-first training). Terminal-first training involves first training a complete model on the terminal side, then sending the network the information needed to train a matching model (typically, the input and output data of the model to be trained). In contrast, network-first training involves first training a complete model on the network side, then sending the terminal the information needed to train a matching model (typically, the input and output data of the model to be trained).

[0060] Typically, a model is scalable. Model scalability refers to a single model's ability to adapt to multiple input / output configurations simultaneously. In CSI compression, model scalability primarily considers the following metrics: the number of encoder input subbands, the number of encoder input ports, and the encoder output payload length.

[0061] For example, taking the payload length output by the encoder as an example, a CSI compression model needs to be able to simultaneously support as many of the above configuration combinations as possible. For example, a CSI compression model can support both 64-bit and 116-bit payloads. When the network needs to switch the payload from 64-bit to 116-bit, it can rely on the existing model implementation without retraining the model, thereby reducing the cost of at least one of the following: replacing the model and maintaining multiple models simultaneously. Generally speaking, to achieve a certain amount of scalability of the CSI compression model, it is necessary to consider multiple situations during the model training phase and combine it with a special model structure (for example, the model's adaptation layer).

[0062] As another example, taking the number of subbands and ports of the encoder input as an example, a CSI compression model is required to support as many of the above configuration combinations as possible at the same time. For example, a CSI compression model can support 13 subbands and 26 subbands, or 32 ports and 16 ports, etc. When the network configuration is changed, the model does not need to be retrained / deployed, thereby saving the overhead and delay of model updates. Methods for achieving scalability of input dimensions include but are not limited to data zero padding and dimensionality splitting; among them, data zero padding generally refers to when the dimension supported by the model is greater than the current input data dimension, the input can be reduced to the dimension supported by the model by filling in zeros; splitting generally refers to when the dimension supported by the model is less than the current input data dimension, the data can be split into several lower-dimensional data, and then processed separately by the model. It should be noted that in order to use these two methods, the possible data dimensions must be known during the training phase. It is difficult for the model to directly process input data that was not considered when training the model.

[0063] Theoretically, it's possible to consider all possible scalability requirements for CSI compression models during model training, enabling a single model to handle all configurations. However, given the difficulty of model training and the fact that certain network parameters are sometimes not known in advance during training, it's difficult to fully consider all possible scalability requirements for each model in practice. In such cases, retraining the model is still necessary to address this issue.

[0064] However, when CSI compression requires an input dimension configuration that was not considered in the previous model training, it is often necessary to update the model to meet this requirement. In the existing AI CSI compression training types, the methods for updating the model include: 1. In type1, the network generally needs to re-transmit the complete encoder to the terminal side (or re-transmit the decoder to the network side); 2. In type2, the network and the terminal need to re-train the model; 3. In type3, the relevant model information (for example, the input and output data sets of the model, etc.) needs to be re-transmitted to update the model. Generally, the delay and air interface overhead of re-transmitting the model are lower than the related overhead of retraining the model. Therefore, type1 has an advantage in flexible model updates compared to type2 and type3. However, when the structure of the encoder is more complex or the parameter scale is large, the overhead of model transmission and deployment is still high. As a result, the overhead of updating the model input dimensions supported by the model is large.

[0065] In order to solve the above problems, in the model updating method provided in the embodiment of the present application, since the terminal can update the first AI unit in the terminal based on the information related to the first target layer of the first AI unit after the model input dimension is updated, so as to update the model input dimension supported by the first AI unit, and the first target layer used to determine or update the applicable configuration of the first AI unit is usually simpler in structure and has fewer related parameters, the overhead required to update the model input dimension supported by the first AI unit based on the information related to the first target layer is much less than the overhead required to update the model in the related technology, thereby reducing the overhead of updating the model input dimension supported by the model.

[0066] The present invention provides a model updating method, and Figure 4 shows a flow chart of the model updating method provided by the present invention. As shown in Figure 4, the model updating method provided by the present invention may include the following steps 401 and 402.

[0067] Step 401: The terminal obtains first information.

[0068] Among them, the above-mentioned first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit.

[0069] Optionally, in the embodiment of the present application, the first target layer may also be referred to as an adaptation layer (ie, adaptation layer).

[0070] For example, taking the encoder with the AI ​​CSI compression model as the first AI unit, as shown in Figure 5, the decoder is consistent with the AI ​​CSI compression model in the related art, and the encoder is further divided into two parts: a fixed layer and an adaptation layer. The scope of the fixed layer and the adaptation layer in the encoder includes but is not limited to the following:

[0071] 1. After the freezing layer, and then through the adaptation layer, the freezing layer refers to the part from the input of the encoder to a certain intermediate node, and the adaptation layer refers to the part from the end of the freezing layer to the encoding output;

[0072] 2. First pass through the adaptation layer, then pass through the freezing layer. The adaptation layer refers to the part from the encoder input to a certain intermediate node, while the freezing layer refers to the part from the end of the adaptation layer to the encoder output;

[0073] 3. First pass through adaptation layer 1, then pass through freezing layer, and then pass through adaptation layer 2;

[0074] 4. First pass through freezing layer 1, then through adaptation layer, and then through freezing layer 2.

[0075] It should be noted that the AI ​​unit in the embodiments of the present application may also be referred to as an AI model, a machine learning (ML) model, an ML unit, an AI structure, an AI function, an AI characteristic, a neural network, a neural network function or a neural network function, etc.; or the AI ​​unit may also refer to a processing unit that can implement specific algorithms, formulas, processing procedures or capabilities related to AI, or a processing method, algorithm, function, module or unit for a specific data set, or a processing method, algorithm, function, module or unit running on AI / ML related hardware; the specific requirements can be determined according to actual usage requirements, and the embodiments of the present application are not limited thereto.

[0076] Optionally, in an embodiment of the present application, the above-mentioned specific data set may include the input and / or output of the above-mentioned AI unit.

[0077] Optionally, the identifier of the above-mentioned AI unit can be an AI model identifier, an AI structure identifier, an AI algorithm identifier, or an identifier of a specific data set associated with the AI ​​unit, or an identifier of a specific scenario, environment, channel feature or device related to AI / ML, or an identifier of an AI / ML-related function, feature, capability or module; the specific identifier can be determined according to actual usage requirements and is not limited in the embodiments of this application.

[0078] For a detailed description of the model input dimensions, please refer to the relevant description in the related technology. To avoid repetition, it will not be repeated here.

[0079] Optionally, in the embodiment of the present application, the first information may include any one of the following:

[0080] Information of the first target layer mentioned above;

[0081] The information of the first target layer and the first mapping relationship;

[0082] The information of the first target layer, the updated model input dimension of the first AI unit, and the first mapping relationship.

[0083] The first mapping relationship includes any of the following:

[0084] A mapping relationship between the input of the first AI unit after the model input dimension is updated and the input of the second target layer of the first AI unit after the model input dimension is updated;

[0085] A mapping relationship between the output of the second target layer of the first AI unit after the model input dimension is updated and the input of the first target layer of the first AI unit after the model input dimension is updated;

[0086] The second target layer is a layer in the first AI unit other than the first target layer.

[0087] Optionally, in the embodiment of the present application, the second target layer may also be referred to as a frozen layer (ie, a fixed layer).

[0088] Optionally, in the embodiment of the present application, the information of the first target layer may include but is not limited to at least one of the following:

[0089] the identifier of the first target layer;

[0090] Information about at least a portion of the model structure of the first target layer;

[0091] An identifier of at least a portion of the model structure of the first target layer.

[0092] Optionally, in an embodiment of the present application, the applicable configuration of the first AI unit may include: configuration related to the input dimension of the first AI unit and partial parameter configuration of the first AI unit, or any other configuration related to the first AI unit.

[0093] In an embodiment of the present application, since the above-mentioned first information can include different information related to the information of the above-mentioned first target layer, the content of the first information is enriched, so that when the terminal updates the first AI unit according to the first information, the flexibility of updating the first AI unit can be improved.

[0094] Optionally, in an embodiment of the present application, the above-mentioned first information may be received by the terminal from the network side device, and in this case, the first information may be obtained by the network side device according to the model training reported by the terminal; wherein, the model may include at least one of the following: the above-mentioned first AI unit (for example, encoder), the above-mentioned second target layer (for example, the part of the encoder other than the adaptation layer).

[0095] Optionally, in the embodiment of the present application, the first target layer may satisfy at least one of the following 1.1 to 1.11:

[0096] 1.1. The first target layer is at least one layer before the first AI unit;

[0097] 1.2. The input of the first target layer is the input of the first AI unit;

[0098] 1.3. The output of the first target layer is the input of the portion of the first AI unit outside the first target layer;

[0099] 1.4. The first target layer is at least one layer after the first AI unit;

[0100] 1.5. The output of the first target layer is the output of the first AI unit;

[0101] 1.6. The input of the first target layer is the output of the first AI unit outside the first target layer;

[0102] 1.7. The first target layer includes the first N layers and the last M layers of the first AI unit. The number of layers included in the first AI unit is greater than N + M, where N and M are both positive integers.

[0103] 1.8. The input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit;

[0104] 1.9. The first target layer includes the layers between the first P layers and the last Q layers of the first AI unit. The number of layers included in the first AI unit is greater than P + Q, where P and Q are both positive integers.

[0105] 1.10. The input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit. The number of layers in the first AI unit is greater than P + Q, where P and Q are both positive integers.

[0106] 1.11. The model structure of the first target layer is defined by the protocol or agreed upon by the terminal and the network side device.

[0107] Optionally, in an embodiment of the present application, when the first target layer satisfies at least one of 1.1 to 1.3 above, the first target layer is located at the head of the first AI unit.

[0108] For example, taking the encoder of the AI ​​CSI compression model as an example, as shown in Figure 6, the input of adaptation layer 0 includes the input of encoder 0; as shown in Figure 7, the input of adaptation layer 1 includes the input of encoder 0; it can be seen that the adaptation layer is located at the head of the encoder at this time.

[0109] When the adaptation layer includes the encoder input, the network device can directly adjust the model input by issuing a new adaptation layer. After receiving the updated adaptation layer, the terminal can combine the adaptation layer with the frozen layer to form a complete encoder model. The structure of the adaptation layer can be determined by protocol specifications or through negotiation between the terminal and the network device. Because changes to the adaptation layer involve simple adjustments to the dimensions, the structure of the adaptation layer should be designed with some room for change.

[0110] Optionally, in an embodiment of the present application, when the first target layer satisfies at least one of 1.4 to 1.6 above, the first target layer is located at the tail of the first AI unit.

[0111] For example, taking the encoder of the AI ​​CSI compression model as the first AI unit, as shown in FIG8 , the input of the adaptation layer 0 does not include the input of the encoder 0; as shown in FIG9 , the input of the adaptation layer 1 does not include the input of the encoder 0; it can be seen that the adaptation layer is located at the tail of the encoder at this time.

[0112] Among them, when the configured new model input dimension is smaller than the previously used model input dimension, the network-side device can adjust the adaptation layer and decoder according to the new input data based on the known frozen layer information, and then inform the terminal of the updated adaptation layer and input mapping method. The input mapping method refers to how to process the new input data into a legal frozen layer input. For example, zero padding can be used to fill the low-dimensional input with a high-dimensional input. When the configured new model input dimension is larger than the previously used model input dimension, it is difficult to process a high-dimensional input into a legal low-dimensional input without losing information, so the situation will become more complicated.

[0113] In order to keep the frozen layer weights unchanged as much as possible, as shown in Figures 10 and 11, the entire encoder model can be updated; in order to reuse the frozen layer, the high-dimensional input is split into multiple legal low-dimensional inputs, and then processed separately through the frozen layer. When a high-dimensional input cannot be split into an integer number of low-dimensional input combinations, it can be processed into legal frozen layer inputs by padding some inputs with zeros (for example, if the frozen layer supports 13 subbands and 32-port inputs and the new input configuration is 20 subbands and 32-ports, the new input can be split into sub-input 0 consisting of the first 13 subbands and sub-input 1 consisting of the last 7 subbands, and then sub-input 1 is padded with 6 32-port subbands containing all 0s to make it a legal input) or overlapping (for example, if the frozen layer supports 13 subbands and 32-port inputs and the new input configuration is 20 subbands and 32-ports, the new input can be split into sub-input 0 consisting of the first 13 subbands and sub-input 1 consisting of the last 13 subbands, where the last 6 subbands of sub-input 0 overlap with the first 6 subbands of sub-input 1, that is, the subbands of sub-input 0 are 1 to 13, and the subbands of sub-input 1 are 8 to 20).

[0114] It should be noted that while Figure 11 shows multiple frozen layer modules, it is not necessary to use multiple frozen layer modules to process data separately. The same frozen layer module can also be used serially to process the input. After the frozen layer is processed, the data is combined and passed through the adaptation layer to obtain the encoder output. In this case, the network device needs to inform the terminal how to use the frozen layer to process the new input while transmitting the adaptation layer to the terminal.

[0115] In order for the network-side equipment to effectively adjust the adaptation layer and decoder, the terminal needs to inform the network-side equipment of the encoder's frozen layer information. Information about the frozen layer can generally be divided into several categories: 1. Relatively high-level model structure information, such as the model's backbone network structure and related hyperparameters; 2. Data set information used to limit model input and output; 3. Specific model structure and weight information. Since it involves changes in the model input dimension, the specific model structure and weight information are mainly considered. Using 1 or 2 will cause the network-side equipment to be unable to obtain sufficient frozen layer information, resulting in a significant decrease in the performance of the adjusted adaptation layer and decoder.

[0116] It should be noted that the model weights shared by the terminal to the network-side device are not strictly required to be the frozen layer model actually deployed by the terminal. The terminal can also share a reference model that has the same functionality as the actual model, but the model design and specific weights can be different. Whether the terminal shares the reference model is determined by the terminal implementation (this information is transparent to the network-side device), but the terminal must ensure that the adaptation layer and decoder trained based on the reference model can be used in conjunction with the model actually used by the terminal.

[0117] Optionally, in an embodiment of the present application, when the model structure of the above-mentioned first target layer is defined by a protocol or agreed upon by the terminal and the network side device, the number of layers of the first target layer can be 1, 2 or several, and the structure of the first target layer can be a multilayer perceptron (MLP) or a convolutional neural network (CNN), etc.

[0118] In the embodiment of the present application, since the above-mentioned first target layer can meet at least one of the above 1.1 to 1.11, it is possible to update different first target layer architectures, thereby improving the flexibility of the update.

[0119] Optionally, in the embodiment of the present application, the first information includes information of the first target layer. Exemplarily, the step 401 can be implemented by the following step 401a.

[0120] Step 401a: The terminal obtains information of the first target layer based on the model information transmitted by the network-side device.

[0121] The above model information includes any of the following:

[0122] at least some model parameters of the first target layer;

[0123] At least part of the model parameters of the first target layer, and at least part of the model structure of the first target layer.

[0124] Optionally, in an embodiment of the present application, the above-mentioned model information may be issued by a network-side device or a third-party server node.

[0125] In an embodiment of the present application, since the terminal can obtain the information of the above-mentioned first target layer based on the model information transmitted by the network side device, the terminal does not need to be trained again, thereby simplifying the process of the terminal obtaining the information of the first target layer.

[0126] Optionally, in the embodiment of the present application, the first information includes information of the first target layer. Exemplarily, the step 401 can be implemented by the following step 401b.

[0127] Step 401b: The terminal obtains information of the first target layer through training based on the first data received from the network-side device.

[0128] Among them, the above-mentioned first data is used to train at least one of the following: the above-mentioned first AI unit, the above-mentioned first target layer.

[0129] In an embodiment of the present application, since the information of the above-mentioned first target layer can be obtained by the terminal based on the above-mentioned first data training received from the network side device, it is not necessary for the network side device to transmit the information of the first target layer to the terminal, thereby reducing the overhead of transmitting the information of the first target layer.

[0130] Optionally, in an embodiment of the present application, the updated model input dimension of the first AI unit may be obtained by the terminal according to any one of the following items: second information, second mapping relationship.

[0131] The second information includes at least one of the following:

[0132] The complete model input dimensions supported by the first AI unit as indicated by the network device;

[0133] The first change value indicated by the network-side device is the change value of the model input dimension after the update of the above-mentioned first AI unit compared with the model input dimension before the update.

[0134] The second mapping relationship includes: a mapping relationship between the output of the first AI unit after the model input dimension is updated, and the input of the first AI unit after the model input dimension is updated.

[0135] In the embodiment of the present application, since the updated model input dimension of the first AI unit can be obtained by the terminal based on the second information or the second mapping relationship, the flexibility of obtaining the updated model input dimension of the first AI unit can be improved.

[0136] Optionally, in an embodiment of the present application, the first mapping relationship may be acquired by the terminal according to the third information.

[0137] The third information includes at least one of the following:

[0138] A complete mapping method between the input of the first AI unit after the model input dimension is updated, as indicated by the network-side device, and the input of the second target layer of the first AI unit after the model input dimension is updated;

[0139] The mapping method serial number indicated by the network side device is determined by the mapping method between the input of the above-mentioned AI unit after the model input dimension is updated according to the protocol agreement and the input of the second target layer of the AI ​​unit after the model input dimension is updated.

[0140] In an embodiment of the present application, since the above-mentioned first mapping relationship can be obtained by the terminal according to at least one of the above-mentioned complete mapping method and the above-mentioned mapping method serial number indicated by the network side, the flexibility of obtaining the first mapping relationship can be improved.

[0141] Step 402: The terminal updates the first AI unit in the terminal according to the first information.

[0142] Optionally, in an embodiment of the present application, after the terminal updates the first AI unit, it can use the updated first AI unit to perform related actions.

[0143] In the model updating method provided in the embodiment of the present application, since the terminal can update the first AI unit in the terminal based on the information related to the first target layer of the first AI unit after the model input dimension is updated, so as to update the model input dimension supported by the first AI unit, and the first target layer used to determine or update the applicable configuration of the first AI unit is usually simpler in structure and has fewer related parameters, the overhead required to update the model input dimension supported by the first AI unit based on the information related to the first target layer is much less than the overhead required to update the model in the related art, thereby reducing the overhead of updating the model input dimension supported by the model.

[0144] Optionally, in an embodiment of the present application, before the above step 401, the model updating method provided in the embodiment of the present application may further include the following step 403.

[0145] Step 403: The terminal reports or registers information of the first target layer supported by the terminal to the network side device.

[0146] The information of the first target layer supported by the terminal includes at least one of the following:

[0147] The identifier of the first target layer supported by the terminal;

[0148] Information about at least a portion of the model structure of the first target layer supported by the terminal;

[0149] An identifier of at least a portion of the model structure of the first target layer supported by the terminal.

[0150] In an embodiment of the present application, since the terminal can first report or register the information of the first target layer supported by the terminal before obtaining the above-mentioned first information, the understanding of the terminal and the network side device can be consistent, thereby improving the accuracy of the models deployed on the terminal and the network side device respectively.

[0151] Optionally, in an embodiment of the present application, the above step 403 can be specifically implemented through the following step 403a or step 403b.

[0152] Step 403a: The terminal reports or registers information of the first target layer supported by the terminal to the network device according to the instruction of the network device.

[0153] Optionally, in an embodiment of the present application, the above-mentioned indication may be an indication sent by the network side device to the terminal during the terminal capability reporting phase.

[0154] Step 403b: The terminal autonomously reports or registers information of the first target layer supported by the terminal to the network side device.

[0155] Optionally, in an embodiment of the present application, the terminal may autonomously report or register information of the first target layer supported by the terminal to the network-side device during the terminal capability reporting phase.

[0156] In an embodiment of the present application, since the terminal can report or register the information of the first target layer supported by the terminal to the network side device according to the instructions of the network side device or autonomously, the flexibility of reporting or registering the information of the first target layer supported by the terminal to the network side device can be improved.

[0157] The embodiment of the present application provides another model updating method, and Figure 12 shows a flow chart of the model updating method provided by the embodiment of the present application. As shown in Figure 12, the model updating method provided by the embodiment of the present application may include the following step 121.

[0158] Step 121: The network-side device sends target information to the terminal.

[0159] Among them, the above-mentioned target information is used by the terminal to obtain first information, and the first information is used to update the first AI unit in the terminal. The first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit.

[0160] Optionally, in the embodiment of the present application, the first information may include any one of the following:

[0161] Information of the first target layer mentioned above;

[0162] The information of the first target layer and the first mapping relationship;

[0163] The information of the first target layer, the updated model input dimension of the first AI unit, and the first mapping relationship.

[0164] The first mapping relationship includes any of the following:

[0165] A mapping relationship between the input of the first AI unit after the model input dimension is updated and the input of the second target layer of the first AI unit after the model input dimension is updated;

[0166] A mapping relationship between the output of the second target layer of the first AI unit after the model input dimension is updated and the input of the first target layer of the first AI unit after the model input dimension is updated;

[0167] The second target layer is a layer in the first AI unit other than the first target layer.

[0168] Optionally, in the embodiment of the present application, the first target layer may satisfy at least one of the following conditions:

[0169] The first target layer is at least one layer before the first AI unit;

[0170] The input of the first target layer is the input of the first AI unit;

[0171] The output of the first target layer is the input of the portion of the first AI unit outside the first target layer;

[0172] The first target layer is at least one layer after the first AI unit;

[0173] The output of the first target layer is the output of the first AI unit.

[0174] The input of the first target layer is the output of the first AI unit outside the first target layer;

[0175] The first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N+M, where N and M are both positive integers;

[0176] The input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit;

[0177] The first target layer includes layers between the first P layers and the last Q layers of the first AI unit, and the number of layers included in the first AI unit is greater than P+Q, where P and Q are both positive integers;

[0178] The input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit. The number of layers included in the first AI unit is greater than P + Q, and P and Q are both positive integers.

[0179] The model structure of the first target layer is defined by a protocol or agreed upon by the terminal and the network side device.

[0180] Optionally, in an embodiment of the present application, the target information may include model information, and the model information is used by the terminal to obtain the information of the first target layer.

[0181] The above model information includes any of the following:

[0182] at least some model parameters of the first target layer;

[0183] At least part of the model parameters of the first target layer, and at least part of the model structure of the first target layer.

[0184] Optionally, in an embodiment of the present application, the target information may include first data, and the first data is used for: the terminal to train at least one of the first AI unit and the first target layer to obtain information of the first target layer.

[0185] Optionally, in an embodiment of the present application, the target information may include second information, which is used by the terminal to obtain the updated model input dimension of the first AI unit.

[0186] The second information includes at least one of the following:

[0187] The complete model input dimensions supported by the first AI unit as indicated by the network device;

[0188] The first change value indicated by the network-side device is the change value of the model input dimension after the update of the above-mentioned first AI unit compared with the model input dimension before the update.

[0189] Optionally, in an embodiment of the present application, the target information may include third information, and the third information is used by the terminal to obtain the first mapping relationship.

[0190] The third information includes at least one of the following:

[0191] A complete mapping method between the input of the first AI unit after the model input dimension is updated, as indicated by the network-side device, and the input of the second target layer of the first AI unit after the model input dimension is updated;

[0192] The mapping method serial number indicated by the network-side device is determined by the mapping method between the input of the above-mentioned first AI unit after the model input dimension is updated according to the protocol agreement and the input of the second target layer of the first AI unit after the model input dimension is updated.

[0193] In the model updating method provided in the embodiment of the present application, since the network side device can send target information for obtaining the first information to the terminal, the terminal can update the first AI unit in the terminal according to the information related to the first target layer of the first AI unit after the model input dimension is updated, and then update the model input dimension supported by the first AI unit. The first target layer used to determine or update the applicable configuration of the first AI unit is usually simpler in structure and has fewer related parameters. Therefore, the overhead required to update the model input dimension supported by the first AI unit according to the information related to the first target layer is much less than the overhead required to update the model in the related technology, thereby reducing the overhead of updating the model input dimension supported by the model.

[0194] Optionally, in an embodiment of the present application, before the above step 121, the model updating method provided in the embodiment of the present application may further include the following step 122.

[0195] Step 122: The network-side device receives information of the first target layer supported by the terminal reported or registered by the terminal.

[0196] The information of the first target layer supported by the terminal includes at least one of the following:

[0197] The identifier of the first target layer supported by the terminal;

[0198] Information about at least a portion of the model structure of the first target layer supported by the terminal;

[0199] An identifier of at least a portion of the model structure of the first target layer supported by the terminal.

[0200] Optionally, in an embodiment of the present application, before the above step 122, the model updating method provided in the embodiment of the present application may further include the following step 123.

[0201] Step 123: The network side device sends instruction information to the terminal.

[0202] The above indication information is used to instruct the terminal to report or register information of the first target layer supported by the terminal.

[0203] Optionally, in an embodiment of the present application, the network side device can monitor whether the first AI unit in the terminal changes, or whether the degree of change exceeds the transformation threshold value. When the first AI unit in the terminal changes, or the degree of change exceeds the transformation threshold value, the terminal is required to report the model and related information used for the network side device to train the above-mentioned first target layer.

[0204] Exemplarily, the network-side device can periodically / event-trigger the terminal to report the currently used encoder information (for example, the model ID corresponding to the first AI unit); or, the terminal can actively execute an encoder update event; or, the terminal reports the encoder input (original channel / channel to be compressed) and encoder output, and the network-side device uses the decoder for inference to determine whether the terminal-side encoder has changed.

[0205] For other descriptions of the embodiments of the present application and the technical effects that can be achieved by each technical feature, please refer to the relevant descriptions in the above-mentioned terminal-side method embodiment. In order to avoid repetition, they will not be repeated here.

[0206] The model updating method provided in the embodiment of the present application can be executed by a model updating device. In the embodiment of the present application, the model updating device provided in the embodiment of the present application is described by taking the execution of the model updating method by the model updating device as an example.

[0207] 13 , an embodiment of the present application provides a model updating device 130 , which may include: an acquisition module 131 and an updating module 132 .

[0208] Acquisition module 131 may be configured to acquire first information, including information related to a first target layer of the first AI unit after updating the model input dimension. The first target layer is used to determine or update the applicable configuration of the first AI unit. Update module 132 may be configured to update the first AI unit in the terminal based on the first information acquired by acquisition module 131.

[0209] In one possible implementation, the first information may include any one of the following: information about the first target layer; information about the first target layer, and a first mapping relationship; information about the first target layer, the updated model input dimension of the first AI unit, and the first mapping relationship. The first mapping relationship includes any one of the following: a mapping relationship between the input of the first AI unit after the model input dimension is updated and the input of the second target layer of the first AI unit after the model input dimension is updated; a mapping relationship between the output of the second target layer of the first AI unit after the model input dimension is updated and the input of the first target layer of the first AI unit after the model input dimension is updated. The second target layer is a layer in the first AI unit other than the first target layer.

[0210] In one possible implementation, the first target layer may satisfy at least one of the following: the first target layer is at least one layer before the first AI unit; the input of the first target layer is the input of the first AI unit; the output of the first target layer is the input of the part of the first AI unit outside the first target layer; the first target layer is at least one layer after the first AI unit; the output of the first target layer is the output of the first AI unit; the input of the first target layer is the output of the part of the first AI unit outside the first target layer; the first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N+M. N and M are both positive integers; the input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit; the first target layer includes layers between the first P layers and the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the model structure of the first target layer is defined by the protocol or agreed upon by the terminal and the network side device.

[0211] In one possible implementation, the first information includes information about the first target layer. The acquisition module 131 may be specifically configured to acquire the information about the first target layer based on model information transmitted by the network-side device. The model information may include any of the following: at least some model parameters of the first target layer; at least some model parameters of the first target layer; and at least some model structure of the first target layer.

[0212] In one possible implementation, the first information includes information about the first target layer. Acquisition module 131 may be configured to train and obtain the information about the first target layer based on first data received from a network-side device. The first data is used to train at least one of: the first AI unit; and the first target layer.

[0213] In one possible implementation, the updated model input dimension of the first AI unit is obtained based on any of the following: second information, a second mapping relationship. The second information includes at least one of the following: a complete model input dimension supported by the first AI unit, as indicated by a network device; or a first change value indicated by the network device, the first change value being the change in the updated model input dimension of the first AI unit compared to the pre-update model input dimension. The second mapping relationship includes a mapping relationship between the output of the first AI unit after the updated model input dimension and the input of the first AI unit after the updated model input dimension.

[0214] In one possible implementation, the first mapping relationship is obtained based on third information. The third information includes at least one of the following: a complete mapping method between the input of the first AI unit after the model input dimension is updated, as indicated by the network-side device, and the input of the second target layer of the first AI unit after the model input dimension is updated; and a mapping method sequence number indicated by the network-side device, where the mapping method sequence number is determined based on the mapping method between the input of the first AI unit after the model input dimension is updated, as agreed upon in the protocol, and the input of the second target layer of the first AI unit after the model input dimension is updated.

[0215] In one possible implementation, the model updating device 130 may further include a processing module. The processing module may be configured to report or register information about the first target layer supported by the terminal to the network-side device before the acquisition module 131 acquires the first information. The information about the first target layer supported by the terminal includes at least one of the following: an identifier of the first target layer supported by the terminal; information about at least a portion of the model structure of the first target layer supported by the terminal; or an identifier of at least a portion of the model structure of the first target layer supported by the terminal.

[0216] In one possible implementation, the above-mentioned processing module can be specifically used to: report or register the information of the first target layer supported by the terminal to the network side device according to the instructions of the network side device; or autonomously report or register the information of the first target layer supported by the terminal to the network side device.

[0217] In the model updating device provided in the embodiment of the present application, since the model updating device can update the first AI unit in the terminal based on the information related to the first target layer of the first AI unit after the model input dimension is updated, so as to update the model input dimension supported by the first AI unit, and the first target layer used to determine or update the applicable configuration of the first AI unit is usually simpler in structure and has fewer related parameters, the overhead required to update the model input dimension supported by the first AI unit based on the information related to the first target layer is much less than the overhead required to update the model in the related art, thereby reducing the overhead of updating the model input dimension supported by the model.

[0218] The model updating device in the embodiments of the present application can be an electronic device, such as an electronic device with an operating system, or a component of an electronic device, such as an integrated circuit or chip. The electronic device can be a terminal. For example, the terminal can include, but is not limited to, the types of terminal 11 listed above, and is not specifically limited in the embodiments of the present application.

[0219] The model updating device provided in the embodiment of the present application can implement the various processes implemented in the above-mentioned terminal-side method embodiment and achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0220] In conjunction with FIG. 14 , an embodiment of the present application provides another model updating device 140 , which may include a sending module 141 .

[0221] Among them, the sending module 141 can be used to send target information to the terminal, and the target information is used for the terminal to obtain first information, and the first information is used to update the first AI unit in the terminal. The first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit.

[0222] In one possible implementation, the first information may include any one of the following: information about the first target layer; information about the first target layer, and a first mapping relationship; information about the first target layer, the updated model input dimension of the first AI unit, and the first mapping relationship. The first mapping relationship includes any one of the following: a mapping relationship between the input of the first AI unit after the model input dimension is updated and the input of the second target layer of the first AI unit after the model input dimension is updated; a mapping relationship between the output of the second target layer of the first AI unit after the model input dimension is updated and the input of the first target layer of the first AI unit after the model input dimension is updated. The second target layer is a layer in the first AI unit other than the first target layer.

[0223] In one possible implementation, the first target layer may satisfy at least one of the following: the first target layer is at least one layer before the first AI unit; the input of the first target layer is the input of the first AI unit; the output of the first target layer is the input of the part of the first AI unit outside the first target layer; the first target layer is at least one layer after the first AI unit; the output of the first target layer is the output of the first AI unit; the input of the first target layer is the output of the part of the first AI unit outside the first target layer; the first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N+M. N and M are both positive integers; the input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit; the first target layer includes layers between the first P layers and the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the model structure of the first target layer is defined by the protocol or agreed upon by the terminal and the network side device.

[0224] In one possible implementation, the target information may include model information, which is used by the terminal to obtain the information of the first target layer. The model information includes any of the following: at least some model parameters of the first target layer; at least some model parameters of the first target layer; and at least some model structure of the first target layer.

[0225] In one possible implementation, the target information may include first data, and the first data is used for: the terminal to train at least one of the first AI unit and the first target layer to obtain information of the first target layer.

[0226] In one possible implementation, the target information may include second information used by the terminal to obtain the updated model input dimensions of the first AI unit. The second information includes at least one of the following: a complete model input dimension supported by the first AI unit, as indicated by a network device; and a first change value indicated by the network device, where the first change value is the change in the updated model input dimension of the first AI unit compared to the model input dimension before the update.

[0227] In one possible implementation, the target information may include third information, which is used by the terminal to obtain the first mapping relationship. The third information includes at least one of the following: a complete mapping method between the input of the first AI unit after the model input dimension is updated, as indicated by the network-side device, and the input of the second target layer of the first AI unit after the model input dimension is updated; and a mapping method serial number indicated by the network-side device, which is determined according to the mapping method between the input of the first AI unit after the model input dimension is updated, as agreed upon in the protocol, and the input of the second target layer of the first AI unit after the model input dimension is updated.

[0228] In one possible implementation, the model updating device 140 may further include a receiving module. The receiving module may be configured to receive information about the first target layer supported by the terminal, as reported or registered by the terminal, before the sending module 141 sends the target information to the terminal. The information about the first target layer supported by the terminal includes at least one of the following: an identifier of the first target layer supported by the terminal; information about at least a portion of the model structure of the first target layer supported by the terminal; or an identifier of at least a portion of the model structure of the first target layer supported by the terminal.

[0229] In one possible implementation, the sending module 141 can also be used to send indication information to the terminal before the above-mentioned receiving module receives the information of the first target layer supported by the terminal reported or registered by the terminal, and the indication information is used to indicate the information of the first target layer supported by the terminal reported or registered by the terminal.

[0230] In the model updating device provided in the embodiment of the present application, since the model updating device can send target information for obtaining the first information to the terminal, the terminal can update the first AI unit in the terminal according to the information related to the first target layer of the first AI unit after the model input dimension is updated, and then update the model input dimension supported by the first AI unit. The first target layer used to determine or update the applicable configuration of the first AI unit is usually simpler in structure and has fewer related parameters. Therefore, the overhead required to update the model input dimension supported by the first AI unit according to the information related to the first target layer is much less than the overhead required to update the model in the related technology, thereby reducing the overhead of updating the model input dimension supported by the model.

[0231] The model updating device in the embodiments of the present application can be an electronic device, such as an electronic device with an operating system, or a component in the electronic device, such as an integrated circuit or chip. The electronic device can be a device other than a terminal. For example, the other device can be a server, a network attached storage (NAS), etc., which is not specifically limited in the embodiments of the present application.

[0232] The model updating device provided in the embodiment of the present application can implement the various processes implemented in the above-mentioned network side device method embodiment and achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0233] As shown in Figure 15, an embodiment of the present application further provides a communication device 100, including a processor 101 and a memory 102. The memory 102 stores a program or instruction that can be run on the processor 101. For example, when the communication device 100 is a terminal, the program or instruction, when executed by the processor 101, implements the various steps of the above-mentioned terminal-side method embodiment and can achieve the same technical effect. When the communication device 100 is a network-side device, the program or instruction, when executed by the processor 101, implements the various steps of the above-mentioned network-side device method embodiment and can achieve the same technical effect. To avoid repetition, they are not described here.

[0234] An embodiment of the present application also provides a terminal, including a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the steps in the above-mentioned terminal-side method embodiment. The processor is used to obtain first information, and the first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit; and based on the first information obtained, the first AI unit in the terminal is updated. This terminal embodiment corresponds to the above-mentioned terminal-side method embodiment, and each implementation process and implementation method of the above-mentioned method embodiment can be applied to this terminal embodiment, and can achieve the same technical effect. Specifically, Figure 16 is a schematic diagram of the hardware structure of a terminal that implements an embodiment of the present application.

[0235] The terminal 1000 includes but is not limited to: a radio frequency unit 1001, a network module 1002, an audio output unit 1003, an input unit 1004, a sensor 1005, a display unit 1006, a user input unit 1007, an interface unit 1008, a memory 1009 and at least some of the components of the processor 1010.

[0236] Those skilled in the art will appreciate that the terminal 1000 may also include a power supply (such as a battery) to power various components. The power supply may be logically connected to the processor 1010 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The terminal structure shown in FIG16 does not limit the terminal. The terminal may include more or fewer components than shown, or combine certain components, or arrange the components differently, which will not be described in detail here.

[0237] It should be understood that in an embodiment of the present application, the input unit 1004 may include a graphics processing unit (GPU) 10041 and a microphone 10042, and the graphics processor 10041 processes the image data of a static picture or video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 1006 may include a display panel 10061, and the display panel 10061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 1007 includes a touch panel 10071 and at least one of other input devices 10072. The touch panel 10071 is also called a touch screen. The touch panel 10071 may include two parts: a touch detection device and a touch controller. Other input devices 10072 may include, but are not limited to, a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, and an operating stick, which will not be repeated here.

[0238] In the embodiment of the present application, after receiving downlink data from a network-side device, the RF unit 1001 may transmit the data to the processor 1010 for processing. Furthermore, the RF unit 1001 may send uplink data to the network-side device. Typically, the RF unit 1001 includes, but is not limited to, an antenna, an amplifier, a transceiver, a coupler, a low-noise amplifier, a duplexer, and the like.

[0239] The memory 1009 can be used to store software programs or instructions and various data. The memory 1009 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, applications or instructions required for at least one function (such as a sound playback function, an image playback function, etc.). In addition, the memory 1009 may include a volatile memory or a non-volatile memory. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM), and a direct memory bus random access memory (DRRAM). The memory 1009 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0240] Processor 1010 may include one or more processing units. Optionally, processor 1010 integrates an application processor and a modem processor. The application processor primarily handles operations related to the operating system, user interface, and application programs, while the modem processor primarily processes wireless communication signals, such as a baseband processor. It is understood that the modem processor may not be integrated into processor 1010.

[0241] Among them, processor 1010 can be used to obtain first information, which includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit; and update the first AI unit in terminal 1000 based on the obtained first information.

[0242] In one possible implementation, the first information may include any one of the following: information about the first target layer; information about the first target layer, and a first mapping relationship; information about the first target layer, the updated model input dimension of the first AI unit, and the first mapping relationship. The first mapping relationship includes any one of the following: a mapping relationship between the input of the first AI unit after the model input dimension is updated and the input of the second target layer of the first AI unit after the model input dimension is updated; a mapping relationship between the output of the second target layer of the first AI unit after the model input dimension is updated and the input of the first target layer of the first AI unit after the model input dimension is updated. The second target layer is a layer in the first AI unit other than the first target layer.

[0243] In a possible implementation, the first target layer may satisfy at least one of the following: the first target layer is at least one layer before the first AI unit; the input of the first target layer is the input of the first AI unit; the output of the first target layer is the input of the part of the first AI unit other than the first target layer; the first target layer is at least one layer after the first AI unit; the output of the first target layer is the output of the first AI unit; the input of the first target layer is the output of the part of the first AI unit other than the first target layer; the first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N+M, N and M is a positive integer; the input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit; the first target layer includes the layers between the first P layers and the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the model structure of the first target layer is defined by the protocol or agreed upon by the terminal 1000 and the network side device.

[0244] In one possible implementation, the first information includes information about the first target layer. Processor 1010 may be specifically configured to obtain the information about the first target layer based on model information transmitted by the network-side device. The model information includes any of the following: at least some model parameters of the first target layer; at least some model parameters of the first target layer; and at least some model structure of the first target layer.

[0245] In one possible implementation, the first information includes information about the first target layer. Processor 1010 may be specifically configured to train the first target layer information based on first data received from a network-side device. The first data is used to train at least one of: the first AI unit; and the first target layer.

[0246] In one possible implementation, the updated model input dimension of the first AI unit is obtained based on any of the following: second information, a second mapping relationship. The second information includes at least one of the following: a complete model input dimension supported by the first AI unit, as indicated by a network device; or a first change value indicated by the network device, the first change value being the change in the updated model input dimension of the first AI unit compared to the pre-update model input dimension. The second mapping relationship includes a mapping relationship between the output of the first AI unit after the updated model input dimension and the input of the first AI unit after the updated model input dimension.

[0247] In one possible implementation, the first mapping relationship is obtained based on third information. The third information includes at least one of the following: a complete mapping method between the input of the first AI unit after the model input dimension is updated, as indicated by the network-side device, and the input of the second target layer of the first AI unit after the model input dimension is updated; and a mapping method sequence number indicated by the network-side device, where the mapping method sequence number is determined based on the mapping method between the input of the first AI unit after the model input dimension is updated, as agreed upon in the protocol, and the input of the second target layer of the first AI unit after the model input dimension is updated.

[0248] In one possible implementation, the processor 1010 may be further configured to, before obtaining the first information, report or register information of the first target layer supported by the terminal 1000 to the network-side device. The information of the first target layer supported by the terminal 1000 includes at least one of the following: an identifier of the first target layer supported by the terminal 1000; information of at least a portion of the model structure of the first target layer supported by the terminal 1000; or an identifier of at least a portion of the model structure of the first target layer supported by the terminal 1000.

[0249] In one possible implementation, the processor 1010 can be specifically used to: report or register information of the first target layer supported by the terminal 1000 to the network side device according to the instructions of the network side device; or autonomously report or register information of the first target layer supported by the terminal 1000 to the network side device.

[0250] In the terminal provided in the embodiment of the present application, since the terminal can update the first AI unit in the terminal based on the information related to the first target layer of the first AI unit after the model input dimension is updated, so as to update the model input dimension supported by the first AI unit, and the first target layer used to determine or update the applicable configuration of the first AI unit is usually simpler in structure and has fewer related parameters, the overhead required to update the model input dimension supported by the first AI unit based on the information related to the first target layer is much less than the overhead required to update the model in the related art, thereby reducing the overhead of updating the model input dimension supported by the model.

[0251] It can be understood that the implementation process of each implementation method mentioned in this embodiment can refer to the relevant description of the above-mentioned terminal side method embodiment and achieve the same or corresponding technical effects. To avoid repetition, it will not be repeated here.

[0252] The embodiment of the present application also provides a network side device, including a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the steps of the above-mentioned network side device method embodiment. Among them, the communication interface is used to send target information to the terminal, and the target information is used for the terminal to obtain first information, and the first information is used to update the first AI unit in the terminal, and the first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit. This network side device embodiment corresponds to the above-mentioned network side device method embodiment, and each implementation process and implementation method of the above-mentioned method embodiment can be applied to this network side device embodiment, and can achieve the same technical effect.

[0253] Specifically, embodiments of the present application also provide a network-side device. As shown in Figure 17, network-side device 1700 includes an antenna 171, a radio frequency device 172, a baseband device 173, a processor 174, and a memory 175. Antenna 171 is connected to radio frequency device 172. In the uplink direction, radio frequency device 172 receives information via antenna 171 and sends the received information to baseband device 173 for processing. In the downlink direction, baseband device 173 processes the information to be transmitted and sends it to radio frequency device 172. Radio frequency device 172 processes the received information and then sends it through antenna 171.

[0254] The method executed by the network-side device in the above embodiment may be implemented in the baseband device 173 , which includes a baseband processor.

[0255] The baseband device 173 may include, for example, at least one baseband board, on which multiple chips are arranged, as shown in Figure 17, one of which is, for example, a baseband processor, which is connected to the memory 175 through a bus interface to call the program in the memory 175 and execute the network device operations shown in the above method embodiment.

[0256] The network side device may further include a network interface 176, which is, for example, a Common Public Radio Interface (CPR17).

[0257] Specifically, the network side device 1700 of the embodiment of the present application also includes: instructions or programs stored in the memory 175 and can be run on the processor 174. The processor 174 calls the instructions or programs in the memory 175 to execute the methods executed by each module shown in Figure 14 and achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0258] Among them, the radio frequency device 172 can be used to send target information to the terminal, and the target information is used by the terminal to obtain first information. The first information is used to update the first AI unit in the terminal. The first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit.

[0259] In one possible implementation, the first information may include any one of the following: information about the first target layer; information about the first target layer, and a first mapping relationship; information about the first target layer, the updated model input dimension of the first AI unit, and the first mapping relationship. The first mapping relationship includes any one of the following: a mapping relationship between the input of the first AI unit after the model input dimension is updated and the input of the second target layer of the first AI unit after the model input dimension is updated; a mapping relationship between the output of the second target layer of the first AI unit after the model input dimension is updated and the input of the first target layer of the first AI unit after the model input dimension is updated. The second target layer is a layer in the first AI unit other than the first target layer.

[0260] In a possible implementation, the first target layer may satisfy at least one of the following: the first target layer is at least one layer before the first AI unit; the input of the first target layer is the input of the first AI unit; the output of the first target layer is the input of the part of the first AI unit other than the first target layer; the first target layer is at least one layer after the first AI unit; the output of the first target layer is the output of the first AI unit; the input of the first target layer is the output of the part of the first AI unit other than the first target layer; the first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N+M, N and M is a positive integer; the input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit; the first target layer includes the layers between the first P layers and the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P+Q, and P and Q are both positive integers; the model structure of the first target layer is defined by the protocol or agreed upon by the terminal and the network side device 1700.

[0261] In one possible implementation, the target information may include model information, which is used by the terminal to obtain the information of the first target layer. The model information includes any of the following: at least some model parameters of the first target layer; at least some model parameters of the first target layer; and at least some model structure of the first target layer.

[0262] In one possible implementation, the target information may include first data, and the first data is used for: the terminal to train at least one of the first AI unit and the first target layer to obtain information of the first target layer.

[0263] In one possible implementation, the target information may include second information used by the terminal to obtain the updated model input dimensions of the first AI unit. The second information includes at least one of the following: a complete model input dimension supported by the first AI unit, as indicated by the network device 1700; or a first change value indicated by the network device 1700, where the first change value is the change in the updated model input dimension of the first AI unit compared to the model input dimension before the update.

[0264] In one possible implementation, the target information may include third information, which is used by the terminal to obtain the first mapping relationship. The third information includes at least one of the following: a complete mapping method between the input of the first AI unit after the model input dimension is updated, as indicated by the network-side device 1700, and the input of the second target layer of the first AI unit after the model input dimension is updated; a mapping method serial number indicated by the network-side device 1700, which is determined according to the mapping method between the input of the first AI unit after the model input dimension is updated, and the input of the second target layer of the first AI unit after the model input dimension is updated, as agreed in the protocol.

[0265] In one possible implementation, the radio frequency device 172 may be further configured to receive information about a first target layer supported by the terminal, reported or registered by the terminal, before sending the target information to the terminal. The information about the first target layer supported by the terminal includes at least one of the following: an identifier of the first target layer supported by the terminal; information about at least a portion of a model structure of the first target layer supported by the terminal; or an identifier of at least a portion of a model structure of the first target layer supported by the terminal.

[0266] In one possible implementation, the radio frequency device 172 may also be used to send indication information to the terminal before receiving the information of the first target layer supported by the terminal reported or registered by the terminal, where the indication information is used to instruct the terminal to report or register the information of the first target layer supported by the terminal.

[0267] In the network-side device provided in the embodiment of the present application, since the network-side device can send target information for obtaining the first information to the terminal, the terminal can update the first AI unit in the terminal according to the information related to the first target layer of the first AI unit after the model input dimension is updated, and then update the model input dimension supported by the first AI unit. The first target layer used to determine or update the applicable configuration of the first AI unit is usually simpler in structure and has fewer related parameters. Therefore, the overhead required to update the model input dimension supported by the first AI unit according to the information related to the first target layer is much less than the overhead required to update the model in the related technology, thereby reducing the overhead of updating the model input dimension supported by the model.

[0268] It can be understood that the implementation process of each implementation method mentioned in this embodiment can refer to the relevant description of the above-mentioned network side device method embodiment and achieve the same or corresponding technical effects. To avoid repetition, it will not be repeated here.

[0269] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned model updating method embodiment are implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0270] The processor is the processor in the terminal described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. In some examples, the readable storage medium may be a non-transitory readable storage medium.

[0271] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned model updating method embodiment and achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0272] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0273] An embodiment of the present application further provides a computer program / program product, which is stored in a storage medium. The computer program / program product is executed by at least one processor to implement the various processes of the above-mentioned model updating method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0274] An embodiment of the present application also provides a communication system, including: a terminal and a network side device, wherein the terminal can be used to execute the steps of the terminal side method described above, and the network side device can be used to execute the steps of the network side device method described above.

[0275] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0276] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of a computer software product plus a necessary general-purpose hardware platform, or of course, by hardware. The computer software product is stored in a storage medium (such as ROM, RAM, magnetic disk, optical disk, etc.) and includes a number of instructions for enabling a terminal or network-side device to execute the methods described in each embodiment of the present application.

[0277] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms of implementation methods without departing from the purpose of this application and the scope of protection of the claims. These implementation methods are all within the protection of this application.

Claims

1. A model update method, the method comprising: The terminal obtains first information, the first information including: information related to a first target layer of a first artificial intelligence (AI) unit after updating the input dimension of the model, where the first target layer is used to determine or update an applicable configuration of the first AI unit; The terminal updates the first AI unit in the terminal according to the first information.

2. The method according to claim 1, wherein, The first information includes any one of the following: Information of the first target layer; Information of the first target layer, and a first mapping relationship; Information of the first target layer, the input dimension of the model after updating the first AI unit, and a first mapping relationship; Wherein, the first mapping relationship includes any one of the following: A mapping relationship between an input of the first AI unit after updating the input dimension of the model and an input of a second target layer of the first AI unit after updating the input dimension of the model; A mapping relationship between an output of the second target layer of the first AI unit after updating the input dimension of the model and an input of the first target layer of the first AI unit after updating the input dimension of the model; Wherein, the second target layer is a layer other than the first target layer in the first AI unit.

3. The method according to claim 1 or 2, wherein, The first target layer satisfies at least one of the following: The first target layer is at least one of the previous layers of the first AI unit; An input of the first target layer is an input of the first AI unit; An output of the first target layer is an input of a part other than the first target layer in the first AI unit; The first target layer is at least one of the subsequent layers of the first AI unit; An output of the first target layer is an output of the first AI unit; An input of the first target layer is an output of a part other than the first target layer in the first AI unit; The first target layer includes the first N layers and the subsequent M layers of the first AI unit, the number of layers included in the first AI unit is greater than N + M, and both N and M are positive integers; An input of the first target layer is an input of the first AI unit, and an output of the first target layer is an output of the first AI unit; The first target layer includes layers between the first P layers and the subsequent Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P + Q, and both P and Q are positive integers; An input of the first target layer is an output of the first P layers of the first AI unit, and an output of the first target layer is an output of the subsequent Q layers of the first AI unit, the number of layers included in the first AI unit is greater than P + Q, and both P and Q are positive integers; The model structure of the first target layer is defined by a protocol or agreed upon by the terminal and a network-side device.

4. The method according to claim 2, wherein, The first information includes information of the first target layer; The terminal obtaining the first information includes: The terminal obtains the information of the first target layer based on model information transmitted by a network-side device; Wherein, the model information includes any one of the following: At least some model parameters of the first target layer; At least some model parameters of the first target layer, and at least some of the model structure of the first target layer.

5. The method according to claim 2, wherein The first information includes information about the first target layer; The terminal obtains the first information, including: The terminal trains to obtain the information about the first target layer based on the first data received from the network-side device; Wherein, the first data is used to train at least one of the following: the first AI unit, the first target layer.

6. The method according to claim 2, wherein, The updated model input dimension of the first AI unit is obtained by the terminal according to any one of the following: second information, second mapping relationship; Wherein, the second information includes at least one of the following: The complete model input dimension supported by the first AI unit indicated by the network-side device; The first change value indicated by the network-side device, where the first change value is the change value of the model input dimension of the first AI unit after update compared with the model input dimension before update; The second mapping relationship includes: the mapping relationship between the output of the first AI unit after updating the model input dimension and the input of the first AI unit after updating the model input dimension.

7. The method according to claim 2, wherein, The first mapping relationship is obtained by the terminal according to the third information; Wherein, the third information includes at least one of the following: The complete mapping method between the input of the first AI unit after updating the model input dimension and the input of the second target layer of the first AI unit after updating the model input dimension indicated by the network-side device; The mapping method serial number indicated by the network-side device, where the mapping method serial number is determined according to the mapping method between the input of the first AI unit after updating the model input dimension and the input of the second target layer of the first AI unit after updating the model input dimension agreed by the protocol.

8. The method according to any one of claims 1 to 7, wherein Before the terminal obtains the first information, the method further includes: The terminal reports or registers to the network-side device the information about the first target layer supported by the terminal; Wherein, the information about the first target layer supported by the terminal includes at least one of the following: The identifier of the first target layer supported by the terminal; The information about at least part of the model structure of the first target layer supported by the terminal; The identifier of at least part of the model structure of the first target layer supported by the terminal.

9. The method according to claim 8, wherein, The terminal reports or registers to the network-side device the information about the first target layer supported by the terminal, including: The terminal reports or registers to the network-side device the information about the first target layer supported by the terminal according to the indication of the network-side device; Or, The terminal independently reports or registers to the network-side device the information about the first target layer supported by the terminal.

10. A model update method, the method includes: The network-side device sends target information to the terminal, the target information is used for the terminal to obtain the first information, the first information is used to update the first AI unit in the terminal, and the first information includes: information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit.

11. The method according to claim 10, wherein, The first information includes any one of the following: The information about the first target layer; The information about the first target layer, and the first mapping relationship; The information of the first target layer, the updated model input dimension of the first AI unit, and the first mapping relationship; Wherein, the first mapping relationship includes any one of the following: The mapping relationship between the input of the first AI unit after updating the model input dimension and the input of the second target layer of the first AI unit after updating the model input dimension; The mapping relationship between the output of the second target layer of the first AI unit after updating the model input dimension and the input of the first target layer of the first AI unit after updating the model input dimension; Wherein, the second target layer is a layer in the first AI unit other than the first target layer.

12. The method according to claim 10 or 11, wherein The first target layer satisfies at least one of the following: The first target layer is at least the previous layer of the first AI unit; The input of the first target layer is the input of the first AI unit; The output of the first target layer is the input of the part other than the first target layer in the first AI unit; The first target layer is at least the subsequent layer of the first AI unit; The output of the first target layer is the output of the first AI unit; The input of the first target layer is the output of the part other than the first target layer in the first AI unit; The first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N + M, and both N and M are positive integers; The input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit; The first target layer includes the layers between the first P layers and the last Q layers of the first AI unit, and the number of layers included in the first AI unit is greater than P + Q, and both P and Q are positive integers; The input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit, and the number of layers included in the first AI unit is greater than P + Q, and both P and Q are positive integers; The model structure of the first target layer is defined by a protocol or agreed upon by the terminal and the network-side device.

13. The method according to claim 11, wherein, The target information includes model information, and the model information is used for the terminal to obtain the information of the first target layer; Wherein, the model information includes any one of the following: At least some model parameters of the first target layer; At least some model parameters of the first target layer and at least some model structures of the first target layer.

14. The method according to claim 11, wherein, The target information includes first data, and the first data is used for: the terminal to train at least one of the first AI unit and the first target layer to obtain the information of the first target layer.

15. The method according to claim 11, wherein, The target information includes second information, and the second information is used for the terminal to obtain the updated model input dimension of the first AI unit; Wherein, the second information includes at least one of the following: The complete model input dimension supported by the first AI unit indicated by the network-side device; The first change value indicated by the network-side device, where the first change value is the change value of the model input dimension after the update of the first AI unit compared to the model input dimension before the update.

16. The method according to claim 11, wherein The target information includes third information, and the third information is used for the terminal to obtain the first mapping relationship. Wherein, the third information includes at least one of the following: The complete mapping method between the input of the first AI unit after updating the model input dimension indicated by the network-side device and the input of the second target layer of the first AI unit after updating the model input dimension. The mapping method serial number indicated by the network-side device, and the mapping method serial number is determined according to the mapping method between the input of the first AI unit after updating the model input dimension agreed upon by the protocol and the input of the second target layer of the first AI unit after updating the model input dimension.

17. The method according to any one of claims 10 to 16, wherein, Before the network-side device sends the target information to the terminal, the method further includes: The network-side device receives the information of the first target layer supported by the terminal reported or registered by the terminal. Wherein, the information of the first target layer supported by the terminal includes at least one of the following: The identifier of the first target layer supported by the terminal. The information of at least part of the model structure of the first target layer supported by the terminal. The identifier of at least part of the model structure of the first target layer supported by the terminal.

18. The method according to claim 17, wherein Before the network-side device receives the information of the first target layer supported by the terminal reported or registered by the terminal, the method further includes: The network-side device sends indication information to the terminal, and the indication information is used to indicate the terminal to report or register the information of the first target layer supported by the terminal.

19. A model update device, the device includes an acquisition module and an update module; The obtaining module is configured to obtain first information, where the first information includes: Information related to the first target layer of the first AI unit after updating the model input dimension, and the first target layer is used to determine or update the applicable configuration of the first AI unit. The update module is used to update the first AI unit in the terminal according to the first information acquired by the acquisition module.

20. The apparatus according to claim 19, wherein The first information includes any one of the following: The information of the first target layer. The information of the first target layer and the first mapping relationship. The information of the first target layer, the model input dimension after the update of the first AI unit, and the first mapping relationship. Wherein, the first mapping relationship includes any one of the following: The mapping relationship between the input of the first AI unit after updating the model input dimension and the input of the second target layer of the first AI unit after updating the model input dimension. The mapping relationship between the output of the second target layer of the first AI unit after updating the model input dimension and the input of the first target layer of the first AI unit after updating the model input dimension. Wherein, the second target layer is the layer other than the first target layer in the first AI unit.

21. The device according to claim 19 or 20, wherein The first target layer satisfies at least one of the following: The first target layer is at least the previous layer of the first AI unit. The input of the first target layer is the input of the first AI unit. The output of the first target layer is the input to the part of the first AI unit other than the first target layer; The first target layer is at least the subsequent layer of the first AI unit; The output of the first target layer is the output of the first AI unit; The input of the first target layer is the output of the part of the first AI unit other than the first target layer; The first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N + M, where both N and M are positive integers; The input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit; The first target layer includes the layers between the first P layers and the last Q layers of the first AI unit, and the number of layers included in the first AI unit is greater than P + Q, where both P and Q are positive integers; The input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit, and the number of layers included in the first AI unit is greater than P + Q, where both P and Q are positive integers; The model structure of the first target layer is defined by a protocol or is agreed upon between the terminal and the network-side device.

22. The apparatus according to any one of claims 19 to 21, wherein, The device further includes a processing module; The processing module is configured to report or register information about the first target layer supported by the terminal to the network-side device before the acquisition module acquires the first information; Wherein, the information about the first target layer supported by the terminal includes at least one of the following: The identifier of the first target layer supported by the terminal; Information about at least part of the model structure of the first target layer supported by the terminal; The identifier of at least part of the model structure of the first target layer supported by the terminal.

23. A model update device, the device includes a sending module; The sending module is configured to send target information to a terminal, where the target information is used for the terminal to obtain first information, and the first information is used to update a first AI unit in the terminal. The first information includes: Information related to the first target layer of the first AI unit after updating the input dimension of the model, where the first target layer is used to determine or update the applicable configuration of the first AI unit.

24. The device according to claim 23, wherein, The first information includes any one of the following: The information of the first target layer; The information of the first target layer, and a first mapping relationship; The information of the first target layer, the updated model input dimension of the first AI unit, and a first mapping relationship; Wherein, the first mapping relationship includes any one of the following: The mapping relationship between the input of the first AI unit after updating the model input dimension and the input of the second target layer of the first AI unit after updating the model input dimension; The mapping relationship between the output of the second target layer of the first AI unit after updating the model input dimension and the input of the first target layer of the first AI unit after updating the model input dimension; Wherein, the second target layer is the layer of the first AI unit other than the first target layer.

25. The device according to claim 23 or 24, wherein The first target layer satisfies at least one of the following: The first target layer is at least the first layer of the first AI unit; The input of the first target layer is the input of the first AI unit; The output of the first target layer is the input to the part of the first AI unit other than the first target layer; The first target layer is at least one layer behind the first AI unit; The output of the first target layer is the output of the first AI unit; The input of the first target layer is the output of the part of the first AI unit outside the first target layer; The first target layer includes the first N layers and the last M layers of the first AI unit, and the number of layers included in the first AI unit is greater than N + M, where N and M are both positive integers; The input of the first target layer is the input of the first AI unit, and the output of the first target layer is the output of the first AI unit; The first target layer includes the layers between the first P layers and the last Q layers of the first AI unit, and the number of layers included in the first AI unit is greater than P + Q, where P and Q are both positive integers; The input of the first target layer is the output of the first P layers of the first AI unit, and the output of the first target layer is the output of the last Q layers of the first AI unit, and the number of layers included in the first AI unit is greater than P + Q, where P and Q are both positive integers; The model structure of the first target layer is defined by a protocol or agreed upon between the terminal and the network-side device.

26. The apparatus according to any one of claims 23 to 25, wherein, The device further includes a receiving module; The receiving module is configured to receive the information of the first target layer supported by the terminal reported or registered by the terminal before the sending module sends the target information to the terminal; Wherein, the information of the first target layer supported by the terminal includes at least one of the following: The identifier of the first target layer supported by the terminal; The information of at least part of the model structure of the first target layer supported by the terminal; The identifier of at least part of the model structure of the first target layer supported by the terminal.

27. A terminal, comprising a processor and a memory, where the memory stores a program or instruction that can run on the processor, and when the program or instruction is executed by the processor, it implements the steps of the model update method according to any one of claims 1 to 9.

28. A network-side device, comprising a processor and a memory, where the memory stores a program or instruction that can run on the processor, and when the program or instruction is executed by the processor, it implements the steps of the model update method according to any one of claims 10 to 18.

29. A readable storage medium, where a program or instruction is stored on the readable storage medium, and when the program or instruction is executed by a processor, it implements the model update method according to any one of claims 1 to 9, or implements the steps of the model update method according to any one of claims 10 to 18.

Images