WO2024017176A1 - 模型训练方法、装置、网络侧设备及终端设备 - Google Patents

模型训练方法、装置、网络侧设备及终端设备 Download PDF

Info

Publication number
WO2024017176A1
WO2024017176A1 PCT/CN2023/107627 CN2023107627W WO2024017176A1 WO 2024017176 A1 WO2024017176 A1 WO 2024017176A1 CN 2023107627 W CN2023107627 W CN 2023107627W WO 2024017176 A1 WO2024017176 A1 WO 2024017176A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
information
terminal device
pseudo
network side
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2023/107627
Other languages
English (en)
French (fr)
Inventor
贾承璐
邬华明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Vivo Mobile Communication Co Ltd
Original Assignee
Vivo Mobile Communication Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Vivo Mobile Communication Co Ltd filed Critical Vivo Mobile Communication Co Ltd
Publication of WO2024017176A1 publication Critical patent/WO2024017176A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Definitions

  • This application belongs to the field of communication technology, and specifically relates to a model training method, device, network side equipment and terminal equipment.
  • AI Artificial Intelligence
  • AI technology is a data-driven technology that can use labeled training data to train artificial intelligence models, and then apply the trained artificial intelligence models to actual scenarios to process corresponding businesses. It can be seen that in AI-based applications, the accuracy of the AI model depends largely on the size and quality of the data set.
  • Embodiments of the present application provide a model training method, device, network side equipment and terminal equipment, which can solve the problem that existing model training methods limit the application in many directions in the field of wireless communications and the accuracy of the model.
  • the first aspect provides a model training method, including:
  • the network side device sends the information of the first model to the terminal device, where the first model is obtained by training the initial model based on the labeled data set;
  • the network side device receives the relevant information of the pseudo label sent by the terminal device, wherein the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model, and the second model
  • the model is determined based on information from the first model
  • the network side device trains the first model based on the labeled data set and the relevant information of the pseudo label to obtain a third model.
  • a model training method including:
  • the terminal device receives the information of the first model sent by the network side device, wherein the first model is obtained by training the initial model based on the labeled data set;
  • the terminal device sends the relevant information of the pseudo label to the network side device, wherein the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model, and the second model The model is based on the The information of the first model is determined.
  • a model training device including:
  • the model information sending module is used to send the information of the first model to the terminal device, wherein the first model is obtained by training the initial model based on the labeled data set;
  • a pseudo label information receiving module configured to receive pseudo label related information sent by the terminal device, where the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model, so The second model is determined based on the information of the first model;
  • a training module configured to train the first model based on the labeled data set and the relevant information of the pseudo-label to obtain a third model.
  • a model training device including:
  • a model information receiving module configured to receive information about the first model sent by the network side device, where the first model is obtained by training an initial model based on a labeled data set;
  • a pseudo label information sending module configured to send pseudo label related information to the network side device, where the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model,
  • the second model is determined based on the information of the first model.
  • a terminal device in a fifth aspect, includes a processor and a memory.
  • the memory stores programs or instructions that can be run on the processor.
  • the program or instructions are executed by the processor, the following is implemented: The steps of the method described in the first aspect.
  • a network-side device including a processor and a memory.
  • the memory stores programs or instructions that can be run on the processor.
  • the program or instructions are executed by the processor, the following implementations are implemented: The steps of the method described in the second aspect.
  • a seventh aspect provides a model training system, including: a network side device and a terminal device.
  • the network side device can be used to perform the steps of the model training method described in the first aspect.
  • the terminal device can be used to perform The steps of the model training method described in the second aspect above.
  • a readable storage medium is provided. Programs or instructions are stored on the readable storage medium. When the programs or instructions are executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method are implemented as described in the first aspect. The steps of the method described in the second aspect.
  • a chip in a ninth aspect, includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the method described in the first aspect. , or implement the method described in the second aspect.
  • a computer program/program product is provided, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement the first aspect or the second aspect. The steps of the method described in this aspect.
  • an embodiment of the present application provides a model training device, which is configured to perform the steps of the model training method described in the first or second aspect.
  • the network side device can send the information of the first model to the terminal device, thereby receiving the relevant information of the pseudo label sent by the terminal device, wherein the first model is an initial model based on the labeled data set.
  • the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model.
  • the second model is determined based on the information of the first model.
  • Figure 1 is a block diagram of a wireless communication system applicable to the embodiment of the present application.
  • Figure 2 is a schematic diagram of a neural network in an embodiment of the present application.
  • Figure 3 is a schematic diagram of the neurons of the neural network in the embodiment of the present application.
  • Figure 4 is a flow chart of a model training method in an embodiment of the present application.
  • Figure 5 is a flow chart of another model training method in the embodiment of the present application.
  • Figure 6 is a schematic flowchart of a specific implementation of the model training method in the embodiment of the present application.
  • Figure 7 is a structural block diagram of a model training device in an embodiment of the present application.
  • Figure 8 is a structural block diagram of another model training device in an embodiment of the present application.
  • Figure 9 is a structural block diagram of a communication device in an embodiment of the present application.
  • Figure 10 is a structural block diagram of a terminal device in an embodiment of the present application.
  • Figure 11 is a structural block diagram of a network side device in an embodiment of the present application.
  • Figure 12 is a structural block diagram of another network side device in an embodiment of the present application.
  • first, second, etc. in the description and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in sequences other than those illustrated or described herein, and that "first" and “second” are distinguished objects It is usually one type, and the number of objects is not limited.
  • the first object can be one or multiple.
  • “and/or” in the description and claims indicates at least one of the connected objects, and the character “/" generally indicates that the related objects are in an "or” relationship.
  • LTE Long Term Evolution
  • LTE-Advanced, LTE-A Long Term Evolution
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • FDMA Frequency Division Multiple Access
  • OFDMA Orthogonal Frequency Division Multiple Access
  • SC-FDMA Single-carrier Frequency Division Multiple Access
  • system and “network” in the embodiments of this application are often used interchangeably, and the described technology can be used not only for the above-mentioned systems and radio technologies, but also for other systems and radio technologies.
  • NR New Radio
  • the following description describes a New Radio (NR) system for example purposes, and NR terminology is used in much of the following description, but these techniques can also be applied to applications other than NR system applications, such as 6th generation Generation, 6G) communication system.
  • 6G 6th generation Generation
  • FIG. 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable.
  • the wireless communication system includes a terminal device 11 and a network side device 12.
  • the terminal device 11 may be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a palmtop computer, a netbook, or a super mobile personal computer.
  • Tablet Personal Computer Tablet Personal Computer
  • laptop computer laptop computer
  • PDA Personal Digital Assistant
  • PDA Personal Digital Assistant
  • UMPC Ultra-mobile personal computer
  • MID Mobile Internet Device
  • AR augmented reality
  • VR virtual reality
  • VUE vehicle-mounted equipment
  • PUE pedestrian terminal
  • smart home home equipment with wireless communication functions, such as refrigerators, TVs, washing machines or furniture, etc.
  • PC personal computers
  • teller machines or Terminal-side devices such as self-service machines
  • 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.
  • the network side device 12 may include an access network device or a core network device, where the access network device 12 may also be called a radio access network device, a radio access network (Radio Access Network, RAN), a radio access network function or Wireless access network unit.
  • the access network device 12 may include a base station, a WLAN access point or a WiFi node, etc.
  • the base station may be called a Node B, an evolved Node B (eNB), an access point, a Base Transceiver Station (BTS), a radio Base station, radio transceiver, Basic Service Set (BSS), Extended Service Set (ESS), Home Node B, Home Evolved Node B, Transmitting Receiving Point (TRP) or all
  • eNB evolved Node B
  • BTS Base Transceiver Station
  • BSS Basic Service Set
  • ESS Extended Service Set
  • Home Node B Home Evolved Node B
  • TRP Transmitting Receiving Point
  • Core network equipment may include but is not limited to at least one of the following: core network nodes, core network functions, mobility management entities (Mobility Management Entity, MME), access mobility management functions (Access and Mobility Management Function, AMF), session management functions (Session Management Function, SMF), User Plane Function (UPF), Policy Control Function (PCF), Policy and Charging Rules Function (PCRF), Edge Application Service Discovery function (Edge Application Server Discovery Function, EASDF), Unified Data Management (UDM), Unified Data Repository (UDR), Home Subscriber Server (HSS), centralized network configuration ( Centralized network configuration (CNC), Network Repository Function (NRF), Network Exposure Function (NEF), local NEF (Local NEF, or L-NEF), binding support function (Binding Support Function, BSF), application function (Application Function, AF), etc.
  • MME mobility management entities
  • AMF Access and Mobility Management Function
  • SMF Session Management Function
  • UPF User Plane Function
  • PCF Policy Control Function
  • PCF Policy
  • Figure 4 is an implementation flow chart of a model training method provided by an embodiment of the present application.
  • the method may include the following steps:
  • Step 401 The network side device sends the information of the first model to the terminal device.
  • the first model is obtained by training the initial model based on the labeled data set.
  • the network side device may be the access network device in Figure 1, such as a base station or a newly defined artificial intelligence processing node on the access network side, or it may be the core network device in Figure 1, such as network data Analysis function (Network Data Analytics Function, NWDAF), location management function (Location Management Function, LMF), or a newly defined processing node on the core network side, or a combination of multiple nodes mentioned above.
  • NWDAF Network Data Analytics Function
  • LMF Location Management Function
  • a newly defined processing node on the core network side or a combination of multiple nodes mentioned above.
  • the network side device can first determine the target service to be processed (such as positioning service, information push service), and then obtain the labeled data of the target service. That is to say, in the embodiment of this application, for a certain business, the labeled data of the business can be obtained, thereby training a model capable of processing the business.
  • the target service to be processed such as positioning service, information push service
  • the labeled data set includes at least one labeled training data.
  • the network side device can obtain the historical business data of the target business through various methods such as data collection and collection. Some historical business data contains real labels, and some historical business data does not contain real labels, but it can be obtained through manual annotation or automatic annotation. Label it. Historical business data that contains real labels or can be labeled can be used as labeled training data to form a labeled data set.
  • a terminal device with positioning function can continuously report measurement data to the network side device during movement.
  • the measurement data can include wireless measurement information and real location information.
  • the real position information can be used as the real position label of the measurement data to obtain labeled training data, and then obtain a labeled data set.
  • the tester can move with the terminal device in hand, and obtain the measurement data through the terminal device.
  • the measurement data contains wireless measurement information, and after labeling the measurement data based on the current real location information, report it to the network side device.
  • the network-side device can obtain measurement data containing wireless measurement information and real location labels, which can be used as labeled training data to obtain a labeled data set.
  • the network side device obtains the measurement data containing wireless measurement information, it automatically labels the location based on the preset labeling rules to obtain labeled training data, and then obtains a labeled data set.
  • the initial model can be randomly generated based on a certain distribution, for example, an initial model can be generated based on a Gaussian distribution with mean a and variance b, or a pre-trained model can also be used as the initial model.
  • the initial model may be an artificial intelligence model, such as any type of fully connected neural network, convolutional neural network, decision tree, support vector machine, or Bayesian classifier.
  • the neural network model As an example, its schematic diagram can be shown in Figure 2.
  • the neural network is composed of neurons, and the schematic diagram of the neurons is shown in Figure 3.
  • a1, a2,...aK represents the input
  • w represents the weight (i.e., multiplicative coefficient)
  • b represents the bias (i.e., additive coefficient)
  • ⁇ (.) represents the activation function.
  • Common activation functions include Sigmoid (mapping variables between 0 and 1), tanh (translation and contraction of Sigmoid), linear rectification function/rectified linear unit (Rectified Linear Unit, ReLU), etc.
  • the model training process is introduced as follows:
  • the parameters of the neural network can be optimized through the gradient optimization algorithm.
  • Gradient optimization algorithms are a type of algorithm that minimize or maximize an objective function (sometimes also called a loss function), and the objective function is often a mathematical combination of model parameters and data.
  • an objective function sometimes also called a loss function
  • the objective function is often a mathematical combination of model parameters and data.
  • a neural network model f(.) can be constructed, and the predicted output f(x) can be obtained based on the input (f(x)-Y), this is the loss function.
  • the optimization goal of the gradient optimization algorithm is to find appropriate w (ie, weight) and b (ie, bias) to minimize the value of the above loss function. The smaller the loss value, the closer the model is to the real situation.
  • BP error Back Propagation
  • the basic idea of BP algorithm is that the learning process consists of two processes: forward propagation of signals and back propagation of errors.
  • the input sample is passed in from the input layer, processed layer by layer by each hidden layer, and then transmitted to the output layer. If the actual output of the output layer does not match the expected output, it will enter the error backpropagation stage.
  • Error backpropagation is to propagate the output error back to the input layer layer by layer through the hidden layer in some form, and allocate the error to all units in each layer, thereby obtaining the error signal of each layer unit. This error signal is used as a correction for each layer.
  • This process of adjusting the weights of each layer in forward signal propagation and error back propagation is carried out over and over again.
  • the process of continuous adjustment of weights is the learning and training process of the network. This process continues until the error of the network output is reduced to an acceptable level, or until a preset number of learning times.
  • common optimization algorithms include gradient descent (Gradient Descent), stochastic gradient descent (Stochastic Gradient Descent, SGD), mini-batch gradient descent (mini-batch gradient descent), momentum method (Momentum), Nesterov (the name of the inventor, Specifically, they are stochastic gradient descent with momentum) adaptive gradient descent (ADAptive GRADient descent, Adagrad), Adagrad's extended algorithm (Adadelta), root mean square error reduction (root mean square prop, RMSprop), adaptive momentum estimation (Adaptive Moment Estimation, Adam), etc.
  • gradient descent Gradient Descent
  • stochastic gradient descent stochastic Gradient Descent, SGD
  • mini-batch gradient descent mini-batch gradient descent
  • momentum method Momentum
  • Nesterov the name of the inventor, Specifically, they are stochastic gradient descent with momentum) adaptive gradient descent (ADAptive GRADient descent, Adagrad), Adagrad's extended algorithm (Adadelta), root mean square error reduction (root mean square prop, RMSprop), adaptive momentum estimation
  • the network side device trains the initial model based on labeled data to obtain the first model, it can send the information of the first model to the terminal device, so that the terminal device can obtain the first model according to the information of the first model.
  • Information deploy the second model on the terminal device side. Therefore, in this embodiment of the present application, the second model may be the same as the first model.
  • Step 402 The network side device receives the relevant information of the pseudo label sent by the terminal device.
  • the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model, and the second model is determined based on the information of the first model.
  • the terminal device deploys the second model on the side of the terminal device based on the information of the first model, it can input the unlabeled data set to the second model, thereby outputting the pseudo-label of the unlabeled data set, and then converting the pseudo-label Relevant information is sent to the network side device.
  • the terminal device can collect unlabeled data in advance, so that after receiving the information of the first model sent by the network side device, it can directly call the collected unlabeled data for use, thus saving the time of obtaining pseudo labels and thus saving the entire model training. time.
  • historical business data with real labels may be difficult to obtain, or it may be difficult to annotate real labels.
  • Such historical business data can be used as unlabeled training data to form an unlabeled data set.
  • the measurement data only contains wireless measurement information and does not contain the corresponding real location information, and it is impossible to label the real location label.
  • Such measurement data can be used as unlabeled training data to form an unlabeled data set.
  • the above-mentioned unlabeled data set and the labeled data set belong to the same service.
  • the above-mentioned labeled data set is a data set for the positioning service
  • the unlabeled data set is also a data set for the positioning service. set.
  • Step 403 The network side device trains the first model based on the labeled data set and the relevant information of the pseudo label to obtain a third model.
  • the above-mentioned initial model, the first model, the second model and the third model are the same type of models, such as all artificial intelligence models of the fully connected neural network type, or all convolutional neural network types.
  • supervised learning is a method of training a model only through labeled data
  • unsupervised learning is a method of training a model only through unlabeled data.
  • the network side device trains the model based on the labeled data set. Initial model to obtain the first model, and then use the first model (i.e., the second model) to determine the pseudo labels of the unlabeled data set, and then train the first model based on the pseudo labels and labeled data sets to obtain the third model, namely Embodiments of the present application can simultaneously utilize partially labeled and partially unlabeled data for model training, which is a type of semi-supervised learning.
  • the model training method of semi-supervised learning is applied to the field of wireless communication, and is implemented by the interaction between the network side device and the terminal device.
  • the network side device can send the information of the first model to the terminal device, thereby receiving the relevant information of the pseudo label sent by the terminal device, wherein the first model is The pseudo-label is obtained by training the initial model based on the labeled data set.
  • the pseudo-label is the pseudo-label output by the second model after the unlabeled data set is input to the second model.
  • the second model is based on the first The model information is determined.
  • the information of the first model includes at least one of the following:
  • Item A-1 Structural information of the first model
  • Item A-2 Parameter information of the first model
  • Item A-3 Configuration information of the first model
  • Item A-4 Information on computing power requirements of the first model
  • Item A-5 Storage capacity requirement information of the first model
  • Item A-6 The task identifier associated with the first model
  • Item A-7 The role of the first model
  • Item A-8 Life cycle information of the first model.
  • the structural information of the first model may include the type of model (such as fully connected neural network, convolutional neural network, hands-on deep learning (Transformer) or others), the input structure and content of the model, and the output of the model. At least one of the structure and content, the number of layers of the network, the number of neurons in each layer, the activation function type of each layer, and the normalization method of each layer (such as batch normalization, layer normalization) .
  • model such as fully connected neural network, convolutional neural network, hands-on deep learning (Transformer) or others
  • Transformer hands-on deep learning
  • the parameter information of the first model may include: at least one of the weight and bias of the model;
  • the configuration information of the first model may include: at least one of the loss function, the optimizer and the status of the optimizer (such as learning rate and learning rate reduction strategy, etc.), and recommended batch size;
  • the computing power requirements information of the first model is used to indicate the computing power required for inference of the first model
  • the storage capacity requirement information of the first model is used to indicate the storage capacity required by the first model.
  • it can include: such as the number of digits to store floating point numbers (for example, floating point (float) 32/float 64/ 7), at least one of the storage space required to store the first model, and the memory space required for inference of the first model.
  • the task identifier associated with the first model is used to indicate the task that the first model is used to process.
  • the role of the first model is used to indicate whether the first model is used for specific services or for generating pseudo labels.
  • the first model generated by the network side device is used to generate pseudo labels. .
  • the life cycle information of the first model may include at least one of the effective time, the expiration time, and the duration of the model's validity.
  • the method before the network side device receives the pseudo label related information sent by the terminal device, the method further includes:
  • the network side device sends target information to the terminal device, where the target information is used to instruct the terminal device to report relevant information of data.
  • the network side device can also send target information to the terminal device to instruct the terminal device how to report pseudo label related information.
  • the target information includes at least one of the following:
  • Item B-1 Format of reported data
  • Item B-2 Instruction information used to indicate the confidence or error or accuracy of the reported pseudo-label
  • Item B-3 Confidence threshold or error threshold or accuracy threshold for filtering pseudo labels
  • Item B-4 Data reporting grouping method
  • Item B-5 Instruction information used to indicate the timestamp information of the reported data.
  • the terminal device For item B-1, it is used to instruct the terminal device to report the format of the pseudo label.
  • the format is ⁇ channel state information, location label>, where "channel state information" is an example of unlabeled data, and "location label” is Pseudo label for "Channel State Information”.
  • the terminal device reports the relevant information of the pseudo label in the format indicated by item B-1.
  • the terminal device can filter the pseudo-labels of the pseudo-label data set obtained above based on the pseudo-label's confidence threshold, error threshold, or accuracy threshold, thereby eliminating unreasonable pseudo-labels;
  • item B-4 it is used to indicate the grouping method of the relevant information of the pseudo label reported by the terminal device. For example, every N pieces of data are one group, and N is an integer greater than 0; when the target information includes item B-4, the terminal device Report pseudo-label related information according to the grouping method indicated in item B-4.
  • the timestamp information used to indicate that the terminal device can report data to the network side device may include at least one of the timestamp information of data collection and the timestamp information of pseudo label generation.
  • the method before the network side device receives the pseudo label related information sent by the terminal device, the method further includes:
  • the network side device sends downlink reference signal information used to collect the unlabeled data set to the terminal device.
  • the network side device can also instruct the terminal device to collect downlink reference signal information of unlabeled data, so that the terminal device can collect the downlink reference signal information according to the downlink reference signal information.
  • Reference signal information is collected to collect unlabeled data sets.
  • the downlink reference signal information may include at least one of frequency resources, time domain resources, air domain resources, and port resources.
  • the relevant information of the pseudo label includes at least one of the following:
  • Item C-1 the pseudo label and the unlabeled data corresponding to the pseudo label
  • Item C-2 Confidence or error or accuracy of the pseudo-label
  • Item C-3 Collection timestamp information of the unlabeled data set
  • Item C-4 Generation timestamp information of the pseudo label
  • Item C-5 The identification information of the terminal device that collects the unlabeled data set
  • Item C-6 Collect the identification information of the area where the terminal equipment of the unlabeled data set is located;
  • Item C-7 Identification information of the sending and receiving points associated with the unlabeled data set.
  • the relevant information of the pseudo label may include the above item C-2 (i.e., pseudo label confidence or error or accuracy);
  • the relevant information of the pseudo label may include the above item C-3 (that is, the collection timestamp information of the unlabeled data set) ) and at least one of the C-4 items (pseudo label generation timestamp information).
  • the identification information of the area where the terminal device of the unlabeled data set is located is collected, which may be, for example, the area ID;
  • the identification information of the transmitting and receiving point (TPR) associated with the unlabeled data set can be the TPR ID.
  • the method before the network side device receives the pseudo label related information sent by the terminal device, the method further includes:
  • the network side device allocates reporting resources for the pseudo label related information to the terminal device.
  • the network side device can allocate reporting resources to the terminal device, so that the terminal device reports the relevant information of the pseudo label to the network side device through the reporting resources.
  • the reported resources include at least one of frequency resources, time domain resources, air domain resources, and port resources.
  • Figure 5 is an implementation flow chart of a model training method provided by an embodiment of the present application.
  • the method may include the following steps:
  • Step 501 The terminal device receives the information of the first model sent by the network side device.
  • the network side device may be the terminal device 11 in Figure 1.
  • the terminal device 11 for examples of the terminal device 11, please refer to the foregoing text and will not be described again here.
  • the first model is obtained by training an initial model based on a labeled data set.
  • the network side device obtains the labeled data set, and trains the initial model based on the labeled data set, thereby obtaining the first model, and then sends the information of the first model to the terminal device.
  • the labeled data set includes at least one labeled training data.
  • the network side device can obtain the historical business data of the target business through various methods such as data collection and collection. Some historical business data contains real labels, and some historical business data does not contain real labels, but it can be obtained through manual annotation or automatic annotation. Label it. Historical business data that contains real labels or can be labeled can be used as labeled training data to form a labeled data set.
  • a terminal device with positioning function can continuously report measurement data to the network side device during movement.
  • the measurement data can include wireless measurement information and real location information.
  • the real position information can be used as the real position label of the measurement data to obtain labeled training data, and then obtain a labeled data set.
  • the tester can move with the terminal device in hand, and obtain the measurement data through the terminal device.
  • the measurement data contains wireless measurement information, and after labeling the measurement data based on the current real location information, report it to the network side device.
  • the network-side device can obtain measurement data containing wireless measurement information and real location labels, which can be used as labeled training data to obtain a labeled data set.
  • the network-side device after the network-side device obtains the measurement data containing wireless measurement information, it automatically labels the location based on the preset labeling rules to obtain labeled training data, and then obtains a labeled data set.
  • the initial model can be randomly generated according to a certain distribution, for example, an initial model can be generated based on a Gaussian distribution with mean a and variance b, or a pre-trained model can also be used as the initial model; the initial model can be artificial intelligence Models, such as any type of fully connected neural network, convolutional neural network, decision tree, support vector machine, and Bayesian classifier.
  • Step 502 The terminal device sends the relevant information of the pseudo label to the network side device.
  • the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model, and the second model is determined based on the information of the first model.
  • the network side device trains the initial model based on labeled data to obtain the first model, it can send the information of the first model to the terminal device, so that the terminal device can, based on the information of the first model, A second model is deployed on the device side. Therefore, in this embodiment of the present application, the second model may be the same as the first model.
  • the terminal device can collect unlabeled data in advance, so that after receiving the information of the first model sent by the network side device, it can directly call the collected unlabeled data for use, thus saving the time of obtaining pseudo labels and thus saving the entire model training. time.
  • historical business data with real labels may be difficult to obtain, or it may be difficult to annotate real labels.
  • Such historical business data can be used as unlabeled training data to form an unlabeled data set.
  • the measurement data only contains wireless measurement information and does not contain the corresponding real location information, and it is impossible to label the real location label.
  • Such measurement data can be used as unlabeled training data to form an unlabeled data set.
  • the above-mentioned unlabeled data set and the labeled data set belong to the same service.
  • the above-mentioned labeled data set is a data set for the positioning service
  • the unlabeled data set is also a data set for the positioning service. set.
  • the terminal device deploys the second model on the side of the terminal device based on the information of the first model, it can input the unlabeled data set to the second model, thereby outputting the pseudo labels of the unlabeled data set, and then converting the pseudo labels to the second model.
  • the relevant information of the tag is sent to the network side device.
  • the network side device After receiving the relevant information of the pseudo-label, the network side device trains the above-mentioned first model based on the labeled data and the relevant information of the pseudo-label, thereby obtaining the third model.
  • This third model is used to handle business.
  • the above-mentioned initial model, the first model, the second model and the third model are the same type of models, such as artificial intelligence models of the fully connected neural network type, or artificial intelligence models of the convolutional neural network type. Intelligent models, or artificial intelligence models that are both decision tree types.
  • supervised learning is a method of training a model only through labeled data
  • unsupervised learning is a method of training a model only through unlabeled data.
  • the network side device is trained based on the labeled data set.
  • Initial model to obtain the first model, and then use the first model (i.e., the second model) to determine the pseudo labels of the unlabeled data set, and then train the first model based on the pseudo labels and labeled data sets to obtain the third model, namely
  • the embodiments of the present application can simultaneously use part of the labeled and part of the unlabeled data for model training, which is a kind of semi-supervised learning.
  • the model training method of semi-supervised learning is applied to the wireless communication field. Domain is implemented by the interaction between network side equipment and terminal equipment.
  • the network side device can send the information of the first model to the terminal device, thereby receiving the relevant information of the pseudo label sent by the terminal device, wherein the first model is The pseudo-label is obtained by training the initial model based on the labeled data set.
  • the pseudo-label is the pseudo-label output by the second model after the unlabeled data set is input to the second model.
  • the second model is based on the first The model information is determined.
  • the method further includes:
  • the terminal device determines the confidence of each pseudo label
  • the terminal device eliminates the relevant information of pseudo labels whose confidence is less than the confidence threshold
  • the terminal device determines the error of each of the pseudo tags
  • the terminal device removes the relevant information of the pseudo tags whose errors are greater than the error threshold
  • the terminal device determines the accuracy of each of the pseudo tags
  • the terminal device eliminates the relevant information of pseudo labels whose accuracy is less than the accuracy threshold.
  • the confidence of the pseudo label is less than the confidence threshold, or the error is greater than the error threshold, or the accuracy is less than the accuracy threshold, it means that the error of the pseudo label is large.
  • unreasonable pseudo labels can be eliminated by eliminating pseudo labels whose confidence is less than the confidence threshold, or pseudo labels whose error is greater than the error threshold, or pseudo labels whose accuracy is less than the accuracy threshold, thereby further Improve the accuracy of model training.
  • the confidence threshold, error threshold, and accuracy threshold can be specified by the network side device or independently determined by the terminal device. For the determination criteria of these thresholds, for example, if the accuracy requirements of the model are higher, a larger confidence threshold can be selected. Or a larger accuracy threshold or a smaller error threshold. On the contrary, you can choose a smaller confidence threshold or a smaller accuracy threshold or a larger error threshold; for example, if the sampling time is long, you can choose a larger The confidence threshold or the larger accuracy threshold or the smaller error threshold. On the contrary, you can choose a smaller confidence threshold or a smaller accuracy threshold or a larger error threshold.
  • the information of the first model includes at least one of the following:
  • Item A-1 Structural information of the first model
  • Item A-2 Parameter information of the first model
  • Item A-3 Configuration information of the first model
  • Item A-4 Information on computing power requirements of the first model
  • Item A-5 Storage capacity requirement information of the first model
  • Item A-6 The task identifier associated with the first model
  • Item A-7 The role of the first model
  • Item A-8 Life cycle information of the first model.
  • the structural information of the first model may include the type of model (such as fully connected neural network, convolutional neural network, hands-on deep learning (Transformer) or others), the input structure and content of the model, and the output of the model. At least one of the structure and content, the number of layers of the network, the number of neurons in each layer, the activation function type of each layer, and the normalization method of each layer (such as batch normalization, layer normalization) .
  • model such as fully connected neural network, convolutional neural network, hands-on deep learning (Transformer) or others
  • Transformer hands-on deep learning
  • the parameter information of the first model may include: at least one of the weight and bias of the model;
  • the configuration information of the first model may include: at least one of the loss function, the optimizer and the status of the optimizer (such as learning rate and learning rate reduction strategy, etc.), and recommended batch size;
  • the computing power requirements information of the first model is used to indicate the computing power required for inference of the first model
  • the storage capacity requirement information of the first model is used to indicate the storage capacity required by the first model.
  • it can include: such as the number of digits to store floating point numbers (for example, floating point (float) 32/float 64/ 7), at least one of the storage space required to store the first model, and the memory space required for inference of the first model.
  • the task identifier associated with the first model is used to indicate the task that the first model is used to process.
  • the role of the first model is used to indicate whether the first model is used for specific services or for generating pseudo labels.
  • the first model generated by the network side device is used to generate pseudo labels. .
  • the life cycle information of the first model may include at least one of the effective time, the expiration time, and the duration of the model's validity.
  • the method further includes:
  • the terminal device receives target information sent by the network side device, wherein the target information is used to indicate relevant information of data reported by the terminal device;
  • the terminal device sends the relevant information of the pseudo label to the network side device, including:
  • the terminal device sends the relevant information of the pseudo label to the network side device according to the target information.
  • the network side device can also send target information to the terminal device to instruct the terminal device how to report pseudo label related information.
  • the target information includes at least one of the following:
  • Item B-1 Format of reported data
  • Item B-2 Instruction information used to indicate the confidence or error or accuracy of the reported pseudo-label
  • Item B-3 Confidence threshold or error threshold or accuracy threshold for screening pseudo labels
  • Item B-4 Data reporting grouping method
  • Item B-5 Instruction information used to indicate the timestamp information of the reported data.
  • the terminal device For item B-1, it is used to instruct the terminal device to report the format of the pseudo label.
  • the format is ⁇ channel state information, location label>, where "channel state information" is an example of unlabeled data, and "location label” is "Channel status Information" pseudo label.
  • the terminal device reports the relevant information of the pseudo label in the format indicated by item B-1.
  • the terminal device can filter the pseudo-labels of the pseudo-label data set obtained above based on the pseudo-label's confidence threshold, error threshold, or accuracy threshold, thereby deleting unreasonable pseudo-labels;
  • item B-4 it is used to indicate the grouping method of the relevant information of the pseudo label reported by the terminal device. For example, every N pieces of data are one group, and N is an integer greater than 0; when the target information includes item B-4, the terminal device Report pseudo-label related information according to the grouping method indicated in item B-4.
  • the timestamp information used to indicate that the terminal device can report data to the network side device may include at least one of the timestamp information of data collection and the timestamp information of pseudo label generation.
  • the method further includes:
  • the terminal device receives the downlink reference signal information sent by the network side device;
  • the terminal device collects the label-free data set according to the downlink reference signal information.
  • the network side device can also instruct the terminal device to collect downlink reference signal information of unlabeled data, so that the terminal device can collect the downlink reference signal information according to the downlink reference signal information.
  • Reference signal information is collected to collect unlabeled data sets.
  • the downlink reference signal information may include at least one of frequency resources, time domain resources, air domain resources, and port resources.
  • the relevant information of the pseudo label includes at least one of the following:
  • Item C-1 the pseudo label and the unlabeled data corresponding to the pseudo label
  • Item C-2 Confidence or error or accuracy of the pseudo-label
  • Item C-3 Collection timestamp information of the unlabeled data set
  • Item C-4 Generation timestamp information of the pseudo label
  • Item C-5 The identification information of the terminal device that collects the unlabeled data set
  • Item C-6 Collect the identification information of the area where the terminal equipment of the unlabeled data set is located;
  • Item C-7 Identification information of the sending and receiving points associated with the unlabeled data set.
  • the relevant information of the pseudo label may include the above item C-2 (i.e., pseudo label confidence or error or accuracy);
  • the relevant information of the pseudo label may include the above item C-3 (that is, the collection timestamp information of the unlabeled data set) ) and at least one of the C-4 items (pseudo label generation timestamp information).
  • the identification information of the transmitting and receiving point (TPR) associated with the unlabeled data set can be the TPR ID.
  • the method further includes:
  • the terminal device obtains the reporting resources allocated by the network side device to the terminal device;
  • the terminal device sends the relevant information of the pseudo label to the network side device, including:
  • the terminal device sends the relevant information of the pseudo label to the network side device through the reporting resource.
  • the network side device can allocate reporting resources to the terminal device, so that the terminal device reports the relevant information of the pseudo label to the network side device through the reporting resources.
  • the reported resources include at least one of frequency resources, time domain resources, air domain resources, and port resources.
  • Step H1 The network side device collects a labeled data set, where the labeled data set can include the carrier to interference ratio (Carrier to Interference Ratio, CIR) and the real label of the location;
  • CIR Carrier to Interference Ratio
  • Step H2 The network side device trains the initial neural network (such as a fully connected neural network (Deep-Learning Neural Network, DNN)) based on the labeled data set to obtain the first model;
  • the initial neural network such as a fully connected neural network (Deep-Learning Neural Network, DNN)
  • Step H3 The network side device transmits the information of the first model to the terminal device; for the specific content of the information of the first model, please refer to the above description and will not be repeated here;
  • Step H4 The terminal device obtains an unlabeled data set, where the labeled data set may include a carrier-to-interference ratio;
  • Step H5 The terminal device deploys the second model based on the information of the first model
  • Step H6 The terminal device inputs the unlabeled data set into the second model to obtain the pseudo-positioning (i.e., pseudo-label) of the unlabeled data set;
  • Step H7 The terminal device eliminates the pseudo positioning and its data whose confidence is less than the confidence threshold, or the pseudo positioning and its data whose error is greater than the error threshold, or the pseudo positioning and its data whose accuracy is less than the accuracy threshold in the aforementioned pseudo positioning;
  • Step H8 The terminal device sends the remaining pseudo-positioning related information to the network side device;
  • Step H9 The network side device trains the first model based on the aforementioned labeled data set and the received data to be pseudo-labeled, to obtain a third model.
  • model training requires four processes.
  • the first process is the model training process.
  • the network side device trains the AI model based on real labeled data, that is, performs supervised learning;
  • the second process is In the pseudo label generation process, the terminal device uses the AI model trained in the first process to generate pseudo labels for unlabeled data.
  • the third process is the data screening process.
  • the terminal device filters the pseudo label data generated in the second process.
  • the error in deleting labels is large. samples;
  • the fourth process is the model training process.
  • the network-side device trains the supervised learning of the AI model based on real labeled data and pseudo-labeled data.
  • the accuracy of the AI model depends largely on the size and quality of the data set.
  • the amount of data collected with accurate labels requires a lot of manpower and time, which limits the application of AI technology in many directions in the field of wireless communications.
  • the embodiments of this application provide the necessary information interaction process and possible training methods introduced by the pseudo-label-based semi-supervised learning method in wireless communications. Compared with traditional supervised learning methods, embodiments of the present application can make full use of unlabeled data to significantly improve the accuracy of AI models.
  • the execution subject may be a model training device.
  • a model training device executing a model training method is used as an example to illustrate the model training device provided by the embodiment of the present application.
  • model training device 70 which can be applied to network-side equipment.
  • model training device 70 includes:
  • the model information sending module 701 is used to send the information of the first model to the terminal device, where the first model is obtained by training the initial model based on the labeled data set;
  • the pseudo-label information receiving module 702 is configured to receive pseudo-label related information sent by the terminal device, where the pseudo-label is a pseudo-label output by the second model after the unlabeled data set is input to the second model, The second model is determined based on the information of the first model;
  • the training module 703 is used to train the first model based on the relevant information of the labeled data set and the pseudo label to obtain a third model.
  • the network side device can send the information of the first model to the terminal device, thereby receiving the relevant information of the pseudo label sent by the terminal device, wherein the first model is based on the labeled data.
  • the pseudo-label is obtained by training the initial model on the set.
  • the pseudo-label is the pseudo-label output by the second model after the unlabeled data set is input to the second model.
  • the second model is determined based on the information of the first model. of.
  • the information of the first model includes at least one of the following:
  • the life cycle information of the first model is the life cycle information of the first model.
  • the device also includes:
  • the target information sending module is configured to send target information to the terminal device before the pseudo label information receiving module receives the relevant information of the pseudo label sent by the terminal device, wherein the target information is used to instruct the terminal device to report Information about the data.
  • the target information includes at least one of the following:
  • Instruction information used to indicate the timestamp information of the reported data is
  • the device also includes:
  • the reference signal information sending module is configured to send downlink reference signal information for collecting the label-free data set to the terminal device before the pseudo label information receiving module receives the pseudo label related information sent by the terminal device.
  • the relevant information of the pseudo label includes at least one of the following:
  • the device also includes:
  • a reporting resource allocation module is configured to allocate reporting resources for the pseudo label related information to the terminal device before the pseudo label information receiving module receives the pseudo label related information sent by the terminal device.
  • the model training device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip.
  • the electronic device may be a network side device.
  • the network side device may include, but is not limited to, the types of network side device 12 listed above.
  • model training device provided by the embodiment of the present application can implement each process implemented by the method embodiment in Figure 4 and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • the model training device 80 includes:
  • the model information receiving module 801 is used to receive the information of the first model sent by the network side device, wherein the first model is obtained by training the initial model based on the labeled data set;
  • the pseudo label information sending module 802 is used to send pseudo label related information to the network side device, where the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model. , the second model is determined based on the information of the first model.
  • the network side device can send the information of the first model to the terminal device, thereby receiving the relevant information of the pseudo label sent by the terminal device, wherein the first model is based on the labeled data.
  • set The pseudo-label is obtained by training the initial model.
  • the pseudo-label is the pseudo-label output by the second model after the unlabeled data set is input to the second model.
  • the second model is determined based on the information of the first model.
  • the embodiments of this application provide the necessary information interaction process and possible training methods introduced in wireless communications by the pseudo-label-based semi-supervised learning method. Compared with traditional supervised learning methods, embodiments of the present application can make full use of unlabeled data to significantly improve the accuracy of AI models and expand the application of AI technology in many directions in the field of wireless communications.
  • the device also includes:
  • the first screening module is configured to determine the confidence of each pseudo label before the pseudo label information sending module sends the relevant information of the pseudo label to the network side device, and set the confidence to be less than the confidence threshold. Removing information related to pseudo-labels;
  • the second screening module is configured to determine the error of each pseudo label before the pseudo label information sending module sends the relevant information of the pseudo label to the network side device, and select the pseudo label whose error is greater than the error threshold. Relevant information is eliminated;
  • the third screening module is configured to determine the accuracy of each pseudo label before the pseudo label information sending module sends the relevant information of the pseudo label to the network side device, and set the accuracy to be less than the accuracy threshold. Relevant information of pseudo labels is eliminated.
  • the information of the first model includes at least one of the following:
  • the life cycle information of the first model is the life cycle information of the first model.
  • the device also includes:
  • the target information receiving module is configured to receive the target information sent by the network side device before the pseudo label information sending module sends the relevant information of the pseudo label to the network side device, wherein the target information is expressed in Relevant information used to instruct the terminal device to report data;
  • the relevant information of the pseudo label is sent to the network side device.
  • the target information includes at least one of the following:
  • Instruction information used to indicate the timestamp information of the reported data is
  • the device also includes:
  • a reference signal information receiving module configured to receive downlink reference signal information sent by the network side device before the second data acquisition module obtains the unlabeled data set;
  • the data acquisition module is configured to collect the unlabeled data set according to the downlink reference signal information.
  • the relevant information of the pseudo label includes at least one of the following:
  • the device also includes:
  • a reporting resource acquisition module configured to acquire the reporting resources allocated by the network side device to the terminal device before the pseudo label information sending module sends the relevant information of the pseudo label to the network side device;
  • the relevant information of the pseudo label is sent to the network side device through the reporting resource.
  • the model training device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip.
  • the electronic device may be a terminal device.
  • the terminal device may include, but is not limited to, the types of terminal device 11 listed above.
  • model training device provided by the embodiment of the present application can implement each process implemented by the method embodiment in Figure 5 and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • this embodiment of the present application also provides a communication device 900, which includes a processor 901 and a memory 902.
  • the memory 902 stores programs or instructions that can be run on the processor 901, such as , when the communication device 900 is a network-side device, when the program or instruction is executed by the processor 901, each step of the model training method embodiment described in the first aspect is implemented, and the same technical effect can be achieved.
  • the communication device 900 is a terminal device, when the program or instruction is executed by the processor 901, each step of the model training method embodiment described in the second aspect is implemented, and the same technical effect can be achieved. To avoid duplication, the steps are not included here. Again.
  • FIG. 10 it is a schematic diagram of the hardware structure of a terminal device that implements an embodiment of the present application.
  • the terminal device 1000 includes but is not limited to: radio frequency unit 1001, network module 1002, audio output unit 1003, input unit 1004, sensor 1005, display unit 1006, user input unit 1007, interface unit 1008, memory 1009, processor 1010, etc. at least some parts of it.
  • the terminal device 1000 may also include a power supply (such as a battery) that supplies power to various components.
  • the power supply may be logically connected to the processor 1010 through a power management system, thereby managing charging, discharging, and function through the power management system. Consumption management and other functions.
  • the structure of the terminal device shown in Figure 10 does not constitute a limitation on the terminal device.
  • the terminal device may include more or less components than shown in the figure, or combine certain components, or arrange different components, which will not be described again here. .
  • the input unit 1004 may include a graphics processing unit (Graphics Processing Unit, GPU) 10041 and a microphone 10042.
  • the graphics processor 10041 is responsible for the image capture device (GPU) in the video capture mode or the image capture mode. Process the image data of still pictures or videos obtained by cameras (such as cameras).
  • the display unit 1006 may include a display panel 10061, which may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like.
  • the user input unit 1007 includes at least one of a touch panel 10071 and other input devices 10072 .
  • Touch panel 10071 also known as 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 physical keyboards, function keys (such as volume control keys, switch keys, etc.), trackballs, mice, and joysticks, which will not be described again here.
  • the radio frequency unit 1001 after receiving downlink data from the network side device, can transmit it to the processor 1010 for processing; in addition, the radio frequency unit 1001 can send uplink data to the network side device.
  • the radio frequency unit 1001 includes, but is not limited to, an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, etc.
  • Memory 1009 may be used to store software programs or instructions as well as 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, an application program or instructions required for at least one function (such as a sound playback function, Image playback function, etc.) etc.
  • memory 1009 may include volatile memory or nonvolatile memory, or memory 1009 may include both volatile and nonvolatile memory.
  • non-volatile memory can be read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM, PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electrically removable memory.
  • Volatile memory can be random access memory (Random Access Memory, RAM), static random access memory (Static RAM, SRAM), dynamic random access memory (Dynamic RAM, DRAM), synchronous dynamic random access memory (Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double Data Rate SDRAM, DDRSDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synch link DRAM) , SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DRRAM).
  • RAM Random Access Memory
  • SRAM static random access memory
  • DRAM dynamic random access memory
  • synchronous dynamic random access memory Synchronous DRAM, SDRAM
  • Double data rate synchronous dynamic random access memory Double Data Rate SDRAM, DDRSDRAM
  • enhanced SDRAM synchronous dynamic random access memory
  • Synch link DRAM synchronous link dynamic random access memory
  • SLDRAM direct memory bus random access memory
  • Direct Rambus RAM Direct Rambus RAM
  • the processor 1010 may include one or more processing units; optionally, the processor 1010 integrates an application processor and a modem processor, where the application processor mainly handles operations related to the operating system, user interface, application programs, etc., Modem processors mainly process wireless communication signals, such as baseband processors. It can be understood that the above modem processor may not be integrated into the processor 1010.
  • the radio frequency unit 1001 is used to receive the information of the first model sent by the network side device, wherein the first The model is obtained by training the initial model based on the labeled data set;
  • the radio frequency unit 1001 is configured to send the relevant information of the pseudo label to the network side device, wherein the pseudo label is a pseudo label output by the second model after the unlabeled data set is input to the second model, and the third model The second model is determined based on the information of the first model.
  • the network side device can send the information of the first model to the terminal device, thereby receiving the relevant information of the pseudo label sent by the terminal device, wherein the first model is based on the labeled data.
  • the pseudo-label is obtained by training the initial model on the set.
  • the pseudo-label is the pseudo-label output by the second model after the unlabeled data set is input to the second model.
  • the second model is determined based on the information of the first model. of.
  • the processor 1010 is used to:
  • the information of the first model includes at least one of the following:
  • the life cycle information of the first model is the life cycle information of the first model.
  • the radio frequency unit 1001 before sending the pseudo tag related information to the network side device, the radio frequency unit 1001 is also used to:
  • Target information sent by the network side device, wherein the target information is used to indicate relevant information of data reported by the terminal device;
  • the radio frequency unit 1001 sends the relevant information of the pseudo tag to the network side device, specifically for:
  • the relevant information of the pseudo label is sent to the network side device.
  • the target information includes at least one of the following:
  • Instruction information used to indicate the timestamp information of the reported data is
  • the radio frequency unit 1001 is also used for:
  • the processor 1010 obtains the unlabeled data set, specifically for:
  • the unlabeled data set is collected according to the downlink reference signal information.
  • the relevant information of the pseudo label includes at least one of the following:
  • the radio frequency unit 1001 before sending the pseudo tag related information to the network side device, the radio frequency unit 1001 is also used to:
  • the radio frequency unit 1001 sends the relevant information of the pseudo tag to the network side device, specifically for:
  • the relevant information of the pseudo label is sent to the network side device through the reporting resource.
  • the network side device 1100 includes: an antenna 111, a radio frequency device 112, a baseband device 113, a processor 114 and a memory 115.
  • the antenna 111 is connected to the radio frequency device 112 .
  • the radio frequency device 112 receives information through the antenna 111 and sends the received information to the baseband device 113 for processing.
  • the baseband device 113 processes the information to be sent and sends it to the radio frequency device 112.
  • the radio frequency device 112 processes the received information and then sends it out through the antenna 111.
  • the method performed by the network side device in the above embodiment can be implemented in the baseband device 113, which includes a baseband processor.
  • the baseband device 113 may include, for example, at least one baseband board on which multiple chips are disposed, as shown in FIG. Program to perform the network device operations shown in the above method embodiments.
  • the network side device may also include a network interface 116, which is, for example, a common public radio interface (CPRI).
  • a network interface 116 which is, for example, a common public radio interface (CPRI).
  • CPRI common public radio interface
  • the network side device 1100 in this embodiment of the present invention also includes: stored in the memory 115 and capable of processing
  • the processor 114 calls the instructions or programs in the memory 115 to execute the method of executing each module shown in Figure 6, and achieves the same technical effect. To avoid duplication, it will not be described again here.
  • the network side device 1200 includes: a processor 1201, a network interface 1202, and a memory 1203.
  • the network interface 1202 is, for example, a common public radio interface (CPRI).
  • CPRI common public radio interface
  • the network side device 1200 in this embodiment of the present invention also includes: instructions or programs stored in the memory 1203 and executable on the processor 1201.
  • the processor 1201 calls the instructions or programs in the memory 1203 to execute each of the steps shown in Figure 4.
  • the method of module execution and achieving the same technical effect will not be described in detail here to avoid duplication.
  • Embodiments of the present application also provide a readable storage medium.
  • Programs or instructions are stored on the readable storage medium.
  • the program or instructions are executed by a processor, each process of the above model training method embodiment is implemented, and the same can be achieved. The technical effects will not be repeated here to avoid repetition.
  • the processor is the processor in the terminal device described in the above embodiment.
  • the readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc.
  • An embodiment of the present application further provides a chip.
  • the chip includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the above model training method embodiment. Each process can achieve the same technical effect. To avoid duplication, it will not be described again here.
  • chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-chip or system-on-chip, etc.
  • Embodiments of the present application further provide a computer program/program product.
  • the computer program/program product is stored in a storage medium.
  • the computer program/program product is executed by at least one processor to implement the above model training method embodiment.
  • Each process can achieve the same technical effect. To avoid repetition, we will not go into details here.
  • Embodiments of the present application also provide a model training system, including: a terminal device and a network side device.
  • the terminal can be used to perform the steps of the model training method described in the second aspect above.
  • the network side device can be used to perform the above steps. The steps of the model training method described in the first aspect.
  • the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology.
  • the computer software product is stored in a storage medium (such as ROM/RAM, disk , CD), including several instructions to cause a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of this application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种模型训练方法、装置、网络侧设备及终端设备，属于通信技术领域，本申请实施例的模型训练方法包括：网络侧设备将第一模型的信息发送给终端设备，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；所述网络侧设备接收所述终端设备发送的伪标签的相关信息，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的；所述网络侧设备基于所述有标签数据集和所述伪标签的相关信息，对所述第一模型进行训练，得到第三模型。

Description

模型训练方法、装置、网络侧设备及终端设备
相关申请的交叉引用
本申请要求在2022年7月21日提交中国专利局、申请号为202210873126.4、名称为“模型训练方法、装置、网络侧设备及终端设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请属于通信技术领域，具体涉及一种模型训练方法、装置、网络侧设备及终端设备。
背景技术
目前，人工智能(Artificial Intelligence，AI)技术在各领域得到了广泛应用，并起到了较好作用。由此可见，将AI技术融入到无线通信网络中，显著提升吞吐量、时延以及用户容量等技术指标，是未来的无线通信网络的重要任务。
其中，AI技术是一种数据驱动技术，可以利用有标签的训练数据对人工智能模型进行训练，然后将训练得到的人工智能模型应用到实际场景中，进行相应业务的处理。由此可见，在基于AI的应用中，AI模型的准确性很大程度上依赖于数据集的规模和质量。
然而，很多情况下，采集有准确标签的数据需要消耗大量的人力和时间，而为了降低训练成本，只能采用较小的训练数据，从而降低了AI模型的精度。或者某些场景下，能采集到的训练数据更小，从而无法应用AI技术。
发明内容
本申请实施例提供一种模型训练方法、装置、网络侧设备及终端设备，能够解决现有的模型训练方法限制了在无线通信领域诸多方向的应用以及模型精度的问题。
第一方面，提供了一种模型训练方法，包括：
网络侧设备将第一模型的信息发送给终端设备，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
所述网络侧设备接收所述终端设备发送的伪标签的相关信息，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的；
所述网络侧设备基于所述有标签数据集和所述伪标签的相关信息，对所述第一模型进行训练，得到第三模型。
第二方面，提供了一种模型训练方法，包括：
终端设备接收网络侧设备发送的第一模型的信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
所述终端设备将伪标签的相关信息发送给所述网络侧设备，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述 第一模型的信息确定的。
第三方面，提供了一种模型训练装置，包括：
模型信息发送模块，用于将第一模型的信息发送给终端设备，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
伪标签信息接收模块，用于接收所述终端设备发送的伪标签的相关信息，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的；
训练模块，用于基于所述有标签数据集和所述伪标签的相关信息，对所述第一模型进行训练，得到第三模型。
第四方面，提供了一种模型训练装置，包括：
模型信息接收模块，用于接收网络侧设备发送的第一模型的信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
伪标签信息发送模块，用于将伪标签的相关信息发送给所述网络侧设备，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。
第五方面，提供了一种终端设备，该终端包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，所述程序或指令被所述处理器执行时实现如第一方面所述的方法的步骤。
第六方面，提供了一种网络侧设备，包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，所述程序或指令被所述处理器执行时实现如第二方面所述的方法的步骤。
第七方面，提供了一种模型训练系统，包括：网络侧设备和终端设备，所述网络侧设备可用于执行如上述第一方面所述的模型训练方法的步骤，所述终端设备可用于执行如上述第二方面所述的模型训练方法的步骤。
第八方面，提供了一种可读存储介质，所述可读存储介质上存储程序或指令，所述程序或指令被处理器执行时实现如第一方面所述的方法的步骤，或者实现如第二方面所述的方法的步骤。
第九方面，提供了一种芯片，所述芯片包括处理器和通信接口，所述通信接口和所述处理器耦合，所述处理器用于运行程序或指令，实现如第一方面所述的方法，或实现如第二方面所述的方法。
第十方面，提供了一种计算机程序/程序产品，所述计算机程序/程序产品被存储在存储介质中，所述计算机程序/程序产品被至少一个处理器执行以实现如第一方面或者第二方面所述的方法的步骤。
第十一方面，本申请实施例提供了一种模型训练装置，所述装置用于执行如第一方面或第二方面所述的模型训练方法的步骤。
在本申请实施例中，网络侧设备能够将第一模型的信息发送给终端设备，从而接收终端设备发送的伪标签的相关信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。本申请实施例通过将大量无标签数据转化为伪标签数据，将伪标签数据作为有标签数据样本进行模型训练，能够充分利用无标签数据的数据量，提升AI模型的训练精度，另外，由于可利用无标签数据转化为伪标签数据来扩大数据样本，可扩展AI技术在无线通信领域诸多方向的应用。附 图说明
图1是本申请实施例可应用的一种无线通信系统的框图；
图2是本申请实施例中神经网络的示意图；
图3是本申请实施例中神经网络的神经元的示意图；
图4是本申请实施例中的一种模型训练方法的流程图；
图5是本申请实施例中的另一种模型训练方法的流程图；
图6是本申请实施例中模型训练方法的具体实施方式的流程示意图；
图7是本申请实施例中的一种模型训练装置的结构框图；
图8是本申请实施例中的另一种模型训练装置的结构框图；
图9是本申请实施例中的一种通信设备的结构框图；
图10是本申请实施例中的一种终端设备的结构框图；
图11是本申请实施例中的一种网络侧设备的结构框图；
图12是本申请实施例中另一种网络侧设备的结构框图。
具体实施例
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员所获得的所有其他实施例，都属于本申请保护的范围。
本申请的说明书和权利要求书中的术语“第一”、“第二”等是用于区别类似的对象，而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换，以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施，且“第一”、“第二”所区别的对象通常为一类，并不限定对象的个数，例如第一对象可以是一个，也可以是多个。此外，说明书以及权利要求中“和/或”表示所连接对象的至少其中之一，字符“/”一般表示前后关联对象是一种“或”的关系。
值得指出的是，本申请实施例所描述的技术不限于长期演进型(Long Term Evolution，LTE)/LTE的演进(LTE-Advanced，LTE-A)系统，还可用于其他无线通信系统，诸如码分多址(Code Division Multiple Access，CDMA)、时分多址(Time Division Multiple Access，TDMA)、频分多址(Frequency Division Multiple Access，FDMA)、正交频分 多址(Orthogonal Frequency Division Multiple Access，OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access，SC-FDMA)和其他系统。本申请实施例中的术语“系统”和“网络”常被可互换地使用，所描述的技术既可用于以上提及的系统和无线电技术，也可用于其他系统和无线电技术。以下描述出于示例目的描述了新空口(New Radio，NR)系统，并且在以下大部分描述中使用NR术语，但是这些技术也可应用于NR系统应用以外的应用，如第6代(6th Generation，6G)通信系统。
图1示出本申请实施例可应用的一种无线通信系统的框图。无线通信系统包括终端设备11和网络侧设备12。其中，终端设备11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)或称为笔记本电脑、个人数字助理(Personal Digital Assistant，PDA)、掌上电脑、上网本、超级移动个人计算机(ultra-mobile personal computer，UMPC)、移动上网装置(Mobile Internet Device，MID)、增强现实(augmented reality，AR)/虚拟现实(virtual reality，VR)设备、机器人、可穿戴式设备(Wearable Device)、车载设备(VUE)、行人终端(PUE)、智能家居(具有无线通信功能的家居设备，如冰箱、电视、洗衣机或者家具等)、游戏机、个人计算机(personal computer，PC)、柜员机或者自助机等终端侧设备，可穿戴式设备包括：智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链等)、智能腕带、智能服装等。需要说明的是，在本申请实施例并不限定终端设备11的具体类型。网络侧设备12可以包括接入网设备或核心网设备，其中，接入网设备12也可以称为无线接入网设备、无线接入网(Radio Access Network,RAN)、无线接入网功能或无线接入网单元。接入网设备12可以包括基站、WLAN接入点或WiFi节点等，基站可被称为节点B、演进节点B(eNB)、接入点、基收发机站(Base Transceiver Station，BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set，BSS)、扩展服务集(Extended Service Set，ESS)、家用B节点、家用演进型B节点、发送接收点(Transmitting Receiving Point，TRP)或所述领域中其他某个合适的术语，只要达到相同的技术效果，所述基站不限于特定技术词汇，需要说明的是，在本申请实施例中仅以NR系统中的基站为例进行介绍，并不限定基站的具体类型。核心网设备可以包含但不限于如下至少一项：核心网节点、核心网功能、移动管理实体(Mobility Management Entity，MME)、接入移动管理功能(Access and Mobility Management Function，AMF)、会话管理功能(Session Management Function，SMF)、用户平面功能(User Plane Function，UPF)、策略控制功能(Policy Control Function，PCF)、策略与计费规则功能单元(Policy and Charging Rules Function，PCRF)、边缘应用服务发现功能(Edge Application Server Discovery Function，EASDF)、统一数据管理(Unified Data Management，UDM)，统一数据仓储(Unified Data Repository，UDR)、归属用户服务器(Home Subscriber Server，HSS)、集中式网络配置(Centralized network configuration，CNC)、网络存储功能(Network Repository Function，NRF)，网络开放功能(Network Exposure Function，NEF)、本地 NEF(Local NEF，或L-NEF)、绑定支持功能(Binding Support Function，BSF)、应用功能(Application Function，AF)等。需要说明的是，在本申请实施例中仅以NR系统中的核心网设备为例进行介绍，并不限定核心网设备的具体类型。
下面结合附图，通过一些实施例及其应用场景对本申请实施例提供的模型训练方法进行详细地说明。
第一方面，参见图4所示，为本申请实施例所提供的一种模型训练方法的实施流程图，该方法可以包括以下步骤：
步骤401：网络侧设备将第一模型的信息发送给终端设备。
其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的.
在本申请实施例中，网络侧设备可以是图1中的接入网设备，如基站或接入网侧新定义的人工智能处理节点，还可以是图1中的核心网设备，如网络数据分析功能(Network Data Analytics Function，NWDAF)、定位管理功能(Location Management Function，LMF)、或者核心网侧新定义的处理节点，还可以是上述多个节点的组合。
这里，网络侧设备可以先确定待处理的目标业务(例如定位业务、信息推送业务)，然后获取目标业务的有标签数据。即本申请实施例中，可以针对某一项业务，获取该项业务的有标签数据，从而训练能够处理该项业务的模型。
其中，有标签数据集中包括至少一个有标签训练数据。网络侧设备可以通过数据采集、收集等多种方式获得目标业务的历史业务数据，有的历史业务数据包含真实标签，有的历史业务数据不包含真实标签，但通过人工标注或自动标注等方式可以对其进行标签标注。包含真实标签或者可进行标签标注的历史业务数据可以作为有标签训练数据，构成有标签数据集。
例如，在定位业务中，具有定位功能的终端设备在移动过程中，可以不断将测量数据上报给网络侧设备，测量数据可以包括无线测量信息和真实位置信息，网络侧设备接收到测量数据后，可以将其中的真实位置信息作为该测量数据的真实位置标签，得到有标签训练数据，进而得到有标签数据集。或者，可以通过测试人员手持终端设备进行移动，通过终端设备获取测量数据，该测量数据中包含无线测量信息，并在基于当前真实位置信息对该测量数据进行标签标注后，上报给网络侧设备，网络侧设备即可获得包含无线测量信息和真实位置标签的测量数据，可以将其作为有标签训练数据，进而得到有标签数据集。或者，网络侧设备获得包含无线测量信息的测量数据后，基于预设的标签标注规则自动对其进行位置标签标注，得到有标签训练数据，进而得到有标签数据集。
另外，初始模型可以根据某一分布随机生成，例如根据均值为a，方差为b的高斯分布生成一初始模型，或者，还可以采用预训练的模型作为初始模型。
例如，所述初始模型可以为人工智能模型，例如全连接神经网络、卷积神经网络、决策树、支持向量机、贝叶斯分类器中的任意一种类型。以神经网络模型为例，其示意图可如图2所示。另外，神经网络由神经元组成，神经元的示意图如图3所示。其中在 图3中，a1,a2,…aK表示输入，w表示权值(即乘性系数)，b表示偏置(即加性系数)，σ(.)表示激活函数。常见的激活函数包括Sigmoid(将变量映射到0、1之间)、tanh(对Sigmoid的平移和收缩)、线性整流函数/修正线性单元(Rectified Linear Unit，ReLU)等。
此外，以神经网络模型为例，对模型训练的过程进行如下介绍：
其中，神经网络的参数可以通过梯度优化算法进行优化。梯度优化算法是一类最小化或者最大化目标函数(有时候也称为损失函数)的算法，而目标函数往往是模型参数和数据的数学组合。例如给定数据X和其对应的标签Y，可以构建一个神经网络模型f(.)，则根据输入x就可以得到预测输出f(x)，并且可以计算出预测值和真实值之间的差距(f(x)-Y)，这个就是损失函数。其中，梯度优化算法的优化目标是找到合适的w(即权值)和b(即偏置)使上述的损失函数的值达到最小，而损失值越小，则说明模型越接近于真实情况。
目前常见的优化算法，基本都是基于误差反向传播(error Back Propagation，BP)算法。BP算法的基本思想是，学习过程由信号的正向传播与误差的反向传播两个过程组成。正向传播时，输入样本从输入层传入，经各隐层逐层处理后，传向输出层。若输出层的实际输出与期望的输出不符，则转入误差的反向传播阶段。误差反传则是将输出误差以某种形式通过隐层向输入层逐层反传，并将误差分摊给各层的所有单元，从而获得各层单元的误差信号，此误差信号即作为修正各单元权值的依据。这种信号正向传播与误差反向传播的各层权值调整过程，是周而复始地进行的。其中，权值不断调整的过程，也就是网络的学习训练过程。此过程一直进行到网络输出的误差减少到可接受的程度，或进行到预先设定的学习次数为止。
另外，常见的优化算法有梯度下降(Gradient Descent)、随机梯度下降(Stochastic Gradient Descent，SGD)、小批量梯度下降(mini-batch gradient descent)、动量法(Momentum)、Nesterov(发明者的名字，具体为带动量的随机梯度下降)自适应梯度下降(ADAptive GRADient descent，Adagrad)、Adagrad的扩展算法(Adadelta)、均方根误差降速(root mean square prop，RMSprop)、自适应动量估计(Adaptive Moment Estimation，Adam)等。
这些优化算法在误差反向传播时，都是根据损失函数得到的误差/损失，对当前神经元求导数/偏导，加上学习速率、之前的梯度/导数/偏导等影响，得到梯度，将梯度传给上一层。
由上述可知，在本申请实施例中，网络侧设备基于有标签数据对初始模型训练得到第一模型之后，可以将第一模型的信息发送给终端设备，以使得终端设备可以根据第一模型的信息，在终端设备一侧部署第二模型。因此，本申请实施例中，第二模型可以与第一模型相同。
步骤402：所述网络侧设备接收所述终端设备发送的伪标签的相关信息。
其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。
其中，终端设备根据第一模型的信息，在终端设备一侧部署第二模型之后，则可以将无标签数据集输入至第二模型，从而输出无标签数据集的伪标签，进而将伪标签的相关信息发送给网络侧设备。
另外，终端设备可以预先采集无标签数据，从而在接收到网络侧设备发送的第一模型的信息之后，直接调用已采集的无标签数据使用，从而节省伪标签的获取时间，进而节省整个模型训练的时间。
此外，针对某项业务而言，带有真实标签的历史业务数据可能难以获取，或者难以进行真实标签的标注，则可以将这样的历史业务数据作为无标签训练数据，构成无标签数据集。比如，定位业务中，测量数据中只包含无线测量信息，不包含相应的真实位置信息，也无法进行真实位置标签的标注，这样的测量数据可以作为无标签训练数据，构成无标签数据集。
需要说明的是，在本申请实施例中，上述无标签数据集和有标签数据集属于同一业务，例如上述有标签数据集为定位业务的数据集，则无标签数据集也为定位业务的数据集。
步骤403：所述网络侧设备基于所述有标签数据集和所述伪标签的相关信息，对所述第一模型进行训练，得到第三模型。
其中，需要说明的是，上述初始模型，第一模型，第二模型和第三模型为同种类型的模型，如均为全连接神经网络类型的人工智能模型，或者均为卷积神经网络类型的人工智能模型，或者均为决策树类型的人工智能模型。
另外，监督学习是一种仅通过有标签数据训练模型的方法；无监督学习是一种仅通过无标签数据训练模型的方法，而本申请实施例中，由网络侧设备基于有标签数据集训练初始模型，得到第一模型，然后利用第一模型(即第二模型)确定无标签数据集的伪标签，进而基于伪标签和有标签数据集，对第一模型训练，得到第三模型，即本申请的实施例，可以同时利用部分有标签的和部分无标签的数据进行模型训练，属于一种半监督学习。并且，在本申请实施例中，是将半监督学习的模型训练方法应用于无线通信领域，由网络侧设备和终端设备进行交互来实现。
由上述步骤401至403可知，在本申请实施例中，网络侧设备能够将第一模型的信息发送给终端设备，从而接收终端设备发送的伪标签的相关信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。由此可见，本申请实施例给出了基于伪标签的半监督学习方法在无线通信中引入的必要信息交互流程以及可能的训练方式。相比于传统的监督学习方法，本申请的实施例，能够充分利用无标签的数据显著提升AI模型的精度，并扩展AI技术在无线通信领域诸 多方向的应用。
可选地，所述第一模型的信息包括如下中的至少一项：
A-1项：所述第一模型的结构信息；
A-2项：所述第一模型的参数信息；
A-3项：所述第一模型的配置信息；
A-4项：所述第一模型的计算能力要求信息；
A-5项：所述第一模型的存储能力要求信息；
A-6项：所述第一模型关联的任务标识；
A-7项：所述第一模型的作用；
A-8项：所述第一模型的生命周期信息。
对于A-1项，第一模型的结构信息可以包括模型的类型(例如全连接神经网络、卷积神经网络、动手学深度学习(Transformer)或其他)、模型的输入结构和内容、模型的输出结构和内容、网络的层数、每一层的神经元数量、每一层的激活函数类型、每一层的归一化方式(如批归一化、层归一化)中的至少一项。
对于A-2项，第一模型的参数信息可以包括：模型的权值、偏置中的至少一项；
对于A-3项，第一模型的配置信息可以包括：损失函数、优化器及优化器的状态(例如学习率及学习率下降的策略等)、推荐的批处理大小中的至少一项；
对于A-4项，第一模型的计算能力要求信息，用于指示第一模型推理所需的算力；
对于A-5项，第一模型的存储能力要求信息，用于指示第一模型所需的存储能力，例如可以包括：如浮点数存储的位数(例如浮点(float)32/float 64/…)、存储第一模型所需存储空间、第一模型推理所需内存空间中的至少一项。
对于A-6项，第一模型关联的任务标识，用于指示第一模型用于处理的任务。
对于A-7项，第一模型的作用，用于指示第一模型用于具体业务还是用于生成伪标签，而在本申请实施例中，网络侧设备生成的第一模型用于生成伪标签。
对于A-8项，第一模型的生命周期信息可以包括生效时间、失效时间、模型有效的持续时间中的至少一项。
可选地，所述网络侧设备接收所述终端设备发送的伪标签的相关信息之前，所述方法还包括：
所述网络侧设备向所述终端设备发送目标信息，其中，所述目标信息用于指示所述终端设备上报数据的相关信息。
由此可知，网络侧设备还可以向终端设备发送目标信息，以指示终端设备如何上报伪标签的相关信息。
可选地，所述目标信息包括如下中至少一项：
B-1项：上报数据的格式；
B-2项：用于指示上报伪标签的置信度或误差或准确性的指示信息；
B-3项：筛选伪标签的置信度阈值或误差阈值或准确性阈值；
B-4项：数据的上报分组方式；
B-5项：用于指示上报数据的时间戳信息的指示信息。
对于B-1项，用于指示终端设备上报伪标签的格式，例如格式为<信道状态信息，位置标签>，其中，“信道状态信息”在这里为无标签数据的举例，“位置标签”为“信道状态信息”的伪标签。当该目标信息包括B-1项时，终端设备按照B-1项指示的格式上报伪标签的相关信息。
对于B-2项，用于指示终端设备可以上报伪标签的置信度或误差或准确性；
对于B-3项，终端设备可以根据伪标签的置信度阈值或误差阈值或准确性阈值，对前述得到的伪标签数据集的伪标签进行筛选，从而剔除不合理的伪标签；
对于B-4项，用于指示终端设备上报伪标签的相关信息的分组方式，例如每N条数据为一组，N为大于0的整数；当该目标信息包括B-4项时，终端设备按照B-4项指示的分组方式上报伪标签的相关信息。
对于B-5项，用于指示终端设备可以向网络侧设备上报数据的时间戳信息，该时间戳信息可以包括数据采集的时间戳信息、伪标签生成的时间戳信息中的至少一项。
可选地，所述网络侧设备接收所述终端设备发送的伪标签的相关信息之前，所述方法还包括：
所述网络侧设备向所述终端设备发送用于采集所述无标签数据集的下行参考信号信息。
由此可知，在本申请实施例中，在终端设备基于第二模型生成伪标签之前，网络侧设备还可以指示终端设备用于采集无标签数据的下行参考信号信息，从而使得终端设备根据该下行参考信号信息，采集无标签数据集。
其中，该下行参考信号信息可以包括频率资源、时域资源、空域资源、端口资源中的至少一者。
可选地，所述伪标签的相关信息包括如下中至少一项：
C-1项：所述伪标签和与所述伪标签对应的无标签数据；
C-2项：所述伪标签的置信度或误差或准确性；
C-3项：所述无标签数据集的采集时间戳信息；
C-4项：所述伪标签的生成时间戳信息；
C-5项：采集所述无标签数据集的终端设备的标识信息；
C-6项：采集所述无标签数据集的终端设备所处的区域的标识信息；
C-7项：所述无标签数据集关联的发送接收点的标识信息。
其中，当前述目标信息包括B-2项(即用于指示上报伪标签的置信度或误差或准确性的指示信息)时，该伪标签的相关信息可以包括上述C-2项(即伪标签的置信度或误差或准确性)；
当前述目标信息包括B-5项(即用于指示上报数据的时间戳信息的指示信息)时，该伪标签的相关信息可以包括上述C-3项(即无标签数据集的采集时间戳信息)和C-4项(伪标签的生成时间戳信息)中的至少一项。
对于C-5项，采集无标签数据集的终端设备的标识信息，例如可以为终端设备的ID；
对于C-6项，采集所述无标签数据集的终端设备所处的区域的标识信息，例如可以为区域ID；
对于C-7项，无标签数据集关联的发送接收点(TPR)的标识信息，可以为TPR ID。
可选地，所述网络侧设备接收所述终端设备发送的伪标签的相关信息之前，所述方法还包括：
所述网络侧设备为所述终端设备分配所述伪标签的相关信息的上报资源。
由此可知，在终端设备上报伪标签的相关信息之前，网络侧设备可以为终端设备分配上报资源，以使得终端设备通过该上报资源，向网络侧设备上报伪标签的相关信息。
其中，该上报资源包括频率资源、时域资源、空域资源、端口资源中的至少一者。
第二方面，参见图5所示，为本申请实施例所提供的一种模型训练方法的实施流程图，该方法可以包括以下步骤：
步骤501：终端设备接收网络侧设备发送的第一模型的信息。
在本申请实施例中，网络侧设备可以是图1中的终端设备11，其中，对于终端设备11的举例可参见前文，此处不再赘述。
其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的。
即在本申请实施例中，网络侧设备获取有标签数据集，并基于有标签数据集对初始模型进行训练，从而获得第一模型，进而将第一模型的信息发送给终端设备。
其中，有标签数据集中包括至少一个有标签训练数据。网络侧设备可以通过数据采集、收集等多种方式获得目标业务的历史业务数据，有的历史业务数据包含真实标签，有的历史业务数据不包含真实标签，但通过人工标注或自动标注等方式可以对其进行标签标注。包含真实标签或者可进行标签标注的历史业务数据可以作为有标签训练数据，构成有标签数据集。
例如，在定位业务中，具有定位功能的终端设备在移动过程中，可以不断将测量数据上报给网络侧设备，测量数据可以包括无线测量信息和真实位置信息，网络侧设备接收到测量数据后，可以将其中的真实位置信息作为该测量数据的真实位置标签，得到有标签训练数据，进而得到有标签数据集。或者，可以通过测试人员手持终端设备进行移动，通过终端设备获取测量数据，该测量数据中包含无线测量信息，并在基于当前真实位置信息对该测量数据进行标签标注后，上报给网络侧设备，网络侧设备即可获得包含无线测量信息和真实位置标签的测量数据，可以将其作为有标签训练数据，进而得到有标签数据集。或者，网络侧设备获得包含无线测量信息的测量数据后，基于预设的标签标注规则自动对其进行位置标签标注，得到有标签训练数据，进而得到有标签数据集。
另外，初始模型可以根据某一分布随机生成，例如根据均值为a，方差为b的高斯分布生成一初始模型，或者，还可以采用预训练的模型作为初始模型；所述初始模型可以为人工智能模型，例如全连接神经网络、卷积神经网络、决策树、支持向量机、贝叶斯分类器中的任意一种类型。
步骤502：所述终端设备将伪标签的相关信息发送给所述网络侧设备。
其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。
在本申请实施例中，网络侧设备基于有标签数据对初始模型训练得到第一模型之后，可以将第一模型的信息发送给终端设备，以使得终端设备可以根据第一模型的信息，在终端设备一侧部署第二模型。因此，本申请实施例中，第二模型可以与第一模型相同。
其中，终端设备可以预先采集无标签数据，从而在接收到网络侧设备发送的第一模型的信息之后，直接调用已采集的无标签数据使用，从而节省伪标签的获取时间，进而节省整个模型训练的时间。
另外，针对某项业务而言，带有真实标签的历史业务数据可能难以获取，或者难以进行真实标签的标注，则可以将这样的历史业务数据作为无标签训练数据，构成无标签数据集。比如，定位业务中，测量数据中只包含无线测量信息，不包含相应的真实位置信息，也无法进行真实位置标签的标注，这样的测量数据可以作为无标签训练数据，构成无标签数据集。
需要说明的是，在本申请实施例中，上述无标签数据集和有标签数据集属于同一业务，例如上述有标签数据集为定位业务的数据集，则无标签数据集也为定位业务的数据集。
由上述可知，终端设备根据第一模型的信息，在终端设备一侧部署第二模型之后，则可以将无标签数据集输入至第二模型，从而输出无标签数据集的伪标签，进而将伪标签的相关信息发送给网络侧设备。
其中，网络侧设备接收到伪标签的相关信息之后，则基于有标签数据和伪标签的相关信息，对上述第一模型进行训练，从而得到第三模型。该第三模型用于处理业务。
需要说明的是，上述初始模型，第一模型，第二模型和第三模型为同种类型的模型，如均为全连接神经网络类型的人工智能模型，或者均为卷积神经网络类型的人工智能模型，或者均为决策树类型的人工智能模型。
另外，监督学习是一种仅通过有标签数据训练模型的方法；无监督学习是一种仅通过无标签数据训练模型的方法，而本申请实施例中，由网络侧设备基于有标签数据集训练初始模型，得到第一模型，然后利用第一模型(即第二模型)确定无标签数据集的伪标签，进而基于伪标签和有标签数据集，对第一模型训练，得到第三模型，即本申请的实施例，可以同时利用部分有标签的和部分无标签的数据进行模型训练，属于一种半监督学习。并且，在本申请实施例中，是将半监督学习的模型训练方法应用于无线通信领 域，由网络侧设备和终端设备进行交互来实现。
由上述步骤501至502可知，在本申请实施例中，网络侧设备能够将第一模型的信息发送给终端设备，从而接收终端设备发送的伪标签的相关信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。由此可见，本申请实施例给出了基于伪标签的半监督学习方法在无线通信中引入的必要信息交互流程以及可能的训练方式。相比于传统的监督学习方法，本申请的实施例，能够充分利用无标签的数据显著提升AI模型的精度，并扩展AI技术在无线通信领域诸多方向的应用。
可选地，所述终端设备将伪标签的相关信息发送给所述网络侧设备之前，所述方法还包括：
所述终端设备确定各个所述伪标签的置信度；
所述终端设备将置信度小于置信度阈值的伪标签的相关信息剔除；
或者，
所述终端设备确定各个所述伪标签的误差；
所述终端设备将误差大于误差阈值的伪标签的相关信息剔除；
或者
所述终端设备确定各个所述伪标签的准确性；
所述终端设备将准确性小于准确性阈值的伪标签的相关信息剔除。
其中，伪标签的置信度小于置信度阈值，或者误差大于误差阈值，或者准确性小于准确性阈值，则表示该伪标签的误差较大。在本申请实施例中，通过剔除置信度小于置信度阈值的伪标签，或者误差大于误差阈值的伪标签，或者准确性小于准确性阈值的伪标签，可以将不合理的伪标签剔除，从而进一步提升模型训练的准确度。
另外，置信度阈值、误差阈值、准确性阈值可以由网络侧设备指定或者终端设备自主确定，对于这些阈值的确定标准，比如，对于模型的精度要求较高，则可以选择较大的置信度阈值或者较大的准确度阈值或者较小的误差阈值，反之，可以选择较小的置信度阈值或者较小的准确度阈值或者较大的误差阈值；比如，采样时间较长，则可以选择较大的置信度阈值或者较大的准确度阈值或者较小的误差阈值，反之，可以选择较小的置信度阈值或者较小的准确度阈值或者较大的误差阈值。
可选地，所述第一模型的信息包括如下中的至少一项：
A-1项：所述第一模型的结构信息；
A-2项：所述第一模型的参数信息；
A-3项：所述第一模型的配置信息；
A-4项：所述第一模型的计算能力要求信息；
A-5项：所述第一模型的存储能力要求信息；
A-6项：所述第一模型关联的任务标识；
A-7项：所述第一模型的作用；
A-8项：所述第一模型的生命周期信息。
对于A-1项，第一模型的结构信息可以包括模型的类型(例如全连接神经网络、卷积神经网络、动手学深度学习(Transformer)或其他)、模型的输入结构和内容、模型的输出结构和内容、网络的层数、每一层的神经元数量、每一层的激活函数类型、每一层的归一化方式(如批归一化、层归一化)中的至少一项。
对于A-2项，第一模型的参数信息可以包括：模型的权值、偏置中的至少一项；
对于A-3项，第一模型的配置信息可以包括：损失函数、优化器及优化器的状态(例如学习率及学习率下降的策略等)、推荐的批处理大小中的至少一项；
对于A-4项，第一模型的计算能力要求信息，用于指示第一模型推理所需的算力；
对于A-5项，第一模型的存储能力要求信息，用于指示第一模型所需的存储能力，例如可以包括：如浮点数存储的位数(例如浮点(float)32/float 64/…)、存储第一模型所需存储空间、第一模型推理所需内存空间中的至少一项。
对于A-6项，第一模型关联的任务标识，用于指示第一模型用于处理的任务。
对于A-7项，第一模型的作用，用于指示第一模型用于具体业务还是用于生成伪标签，而在本申请实施例中，网络侧设备生成的第一模型用于生成伪标签。
对于A-8项，第一模型的生命周期信息可以包括生效时间、失效时间、模型有效的持续时间中的至少一项。
可选地，所述终端设备将伪标签的相关信息发送给所述网络侧设备之前，所述方法还包括：
所述终端设备接收所述网络侧设备发送的目标信息，其中，所述目标信息用于指示所述终端设备上报数据的相关信息；
所述终端设备将伪标签的相关信息发送给所述网络侧设备，包括：
所述终端设备根据所述目标信息，将所述伪标签的相关信息发送给所述网络侧设备。
由此可知，网络侧设备还可以向终端设备发送目标信息，以指示终端设备如何上报伪标签的相关信息。
可选地，所述目标信息包括如下中至少一项：
B-1项：上报数据的格式；
B-2项：用于指示上报伪标签的置信度或误差或准确性的指示信息；
B-3项：筛选伪标签的置信度阈值或误差阈值或准确性阈值；
B-4项：数据的上报分组方式；
B-5项：用于指示上报数据的时间戳信息的指示信息。
对于B-1项，用于指示终端设备上报伪标签的格式，例如格式为<信道状态信息，位置标签>，其中，“信道状态信息”在这里为无标签数据的举例，“位置标签”为“信道状态 信息”的伪标签。当该目标信息包括B-1项时，终端设备按照B-1项指示的格式上报伪标签的相关信息。
对于B-2项，用于指示终端设备可以上报伪标签的置信度或误差或准确性；
对于B-3项，终端设备可以根据伪标签的置信度阈值或误差阈值或准确性阈值，对前述得到的伪标签数据集的伪标签进行筛选，从而删除不合理的伪标签；
对于B-4项，用于指示终端设备上报伪标签的相关信息的分组方式，例如每N条数据为一组，N为大于0的整数；当该目标信息包括B-4项时，终端设备按照B-4项指示的分组方式上报伪标签的相关信息。
对于B-5项，用于指示终端设备可以向网络侧设备上报数据的时间戳信息，该时间戳信息可以包括数据采集的时间戳信息、伪标签生成的时间戳信息中的至少一项。
可选地，所述终端设备获取无标签数据集之前，所述方法还包括：
所述终端设备接收所述网络侧设备发送的下行参考信号信息；
所述终端设备获取无标签数据集的过程，包括：
所述终端设备根据所述下行参考信号信息，采集所述无标签数据集。
由此可知，在本申请实施例中，在终端设备基于第二模型生成伪标签之前，网络侧设备还可以指示终端设备用于采集无标签数据的下行参考信号信息，从而使得终端设备根据该下行参考信号信息，采集无标签数据集。
其中，该下行参考信号信息可以包括频率资源、时域资源、空域资源、端口资源中的至少一者。
可选地，所述伪标签的相关信息包括如下中至少一项：
C-1项：所述伪标签和与所述伪标签对应的无标签数据；
C-2项：所述伪标签的置信度或误差或准确性；
C-3项：所述无标签数据集的采集时间戳信息；
C-4项：所述伪标签的生成时间戳信息；
C-5项：采集所述无标签数据集的终端设备的标识信息；
C-6项：采集所述无标签数据集的终端设备所处的区域的标识信息；
C-7项：所述无标签数据集关联的发送接收点的标识信息。
其中，当前述目标信息包括B-2项(即用于指示上报伪标签的置信度或误差或准确性的指示信息)时，该伪标签的相关信息可以包括上述C-2项(即伪标签的置信度或误差或准确性)；
当前述目标信息包括B-5项(即用于指示上报数据的时间戳信息的指示信息)时，该伪标签的相关信息可以包括上述C-3项(即无标签数据集的采集时间戳信息)和C-4项(伪标签的生成时间戳信息)中的至少一项。
对于C-5项，采集无标签数据集的终端设备的标识信息，例如可以为终端设备的ID；
对于C-6项，采集所述无标签数据集的终端设备所处的区域的标识信息，例如可以 为区域ID；
对于C-7项，无标签数据集关联的发送接收点(TPR)的标识信息，可以为TPR ID。
可选地，所述终端设备将伪标签的相关信息发送给所述网络侧设备之前，所述方法还包括：
所述终端设备获取所述网络侧设备为所述终端设备分配的上报资源；
所述终端设备将伪标签的相关信息发送给所述网络侧设备，包括：
所述终端设备通过所述上报资源，将所述伪标签的相关信息发送给所述网络侧设备。
由此可知，在终端设备上报伪标签的相关信息之前，网络侧设备可以为终端设备分配上报资源，以使得终端设备通过该上报资源，向网络侧设备上报伪标签的相关信息。
其中，该上报资源包括频率资源、时域资源、空域资源、端口资源中的至少一者。
综上所述，本申请实施例的模型训练方法应用于用于定位业务的AI模型的训练时，具体实施方式，可如图6所示，具体如下所述：
步骤H1：网络侧设备采集有标签数据集，其中，该有标签数据集可以包括载波干扰比(Carrier to Interference Ratio，CIR)和位置的真实标签；
步骤H2：网络侧设备基于有标签数据集，对初始神经网络(例如全连接神经网络((Deep-Learning Neural Network，DNN))进行训练，得到第一模型；
步骤H3：网络侧设备将第一模型的信息传输给终端设备；其中，第一模型的信息的具体内容可参见前文所述，此处不再赘述；
步骤H4：终端设备获取无标签数据集，其中，该有标签数据集可以包括载波干扰比；
步骤H5：终端设备根据第一模型的信息，部署第二模型；
步骤H6：终端设备将无标签数据集输入至第二模型，得到无标签数据集的伪定位(即伪标签)；
步骤H7：终端设备将前述伪定位中置信度小于置信度阈值的伪定位及其数据，或者误差大于误差阈值的伪定位及其数据，或者准确性小于准确性阈值的伪定位及其数据剔除；
步骤H8：终端设备将剩余的伪定位的相关信息发送给网络侧设备；
步骤H9：网络侧设备根据前述有标签数据集和接收到的待伪标签的数据，对前述第一模型进行训练，得到第三模型。
由此可见，在本申请实施例中，模型训练需要有四个过程，第一过程为模型训练过程，网络侧设备基于有真实标签数据的训练AI模型，即进行有监督学习；第二过程为伪标签生成过程，终端设备对无标签数据利用第一过程训练的AI模型生成伪标签；第三过程为数据筛选过程，终端设备对第二过程生成的伪标签数据进行筛选，删除标签误差较大的样本；第四过程为模型训练过程，网络侧设备基于有真实标签的数据和伪标签数据训练AI模型的有监督学习。
其中，在基于AI的应用中，AI模型的准确性很大程度上依赖于数据集的规模和质 量，而有准确标签的数据采集需要消耗大量的人力和时间，这限制了AI技术在无线通信领域诸多方向的应用。而本申请的实施例，给出了基于伪标签的半监督学习方法在无线通信引入的必要信息交互流程以及可能的训练方式。相比于传统的监督学习方法，本申请的实施例能够充分利用无标签的数据显著提升AI模型的精度。
本申请实施例提供的模型训练方法，执行主体可以为模型训练装置。本申请实施例中以模型训练装置执行模型训练方法为例，说明本申请实施例提供的模型训练装置。
第三方面，本申请实施例提供了一种模型训练装置，该装置可以应用于网络侧设备，如图7所示，该模型训练装置70包括：
模型信息发送模块701，用于将第一模型的信息发送给终端设备，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
伪标签信息接收模块702，用于接收所述终端设备发送的伪标签的相关信息，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的；
训练模块703，用于基于所述有标签数据集和所述伪标签的相关信息，对所述第一模型进行训练，得到第三模型。
由此可见，在本申请实施例中，网络侧设备能够将第一模型的信息发送给终端设备，从而接收终端设备发送的伪标签的相关信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。由此可见，本申请实施例给出了基于伪标签的半监督学习方法在无线通信中引入的必要信息交互流程以及可能的训练方式。相比于传统的监督学习方法，本申请的实施例，能够充分利用无标签的数据显著提升AI模型的精度，并扩展AI技术在无线通信领域诸多方向的应用。
可选地，所述第一模型的信息包括如下中的至少一项：
所述第一模型的结构信息；
所述第一模型的参数信息；
所述第一模型的配置信息；
所述第一模型的计算能力要求信息；
所述第一模型的存储能力要求信息；
所述第一模型关联的任务标识；
所述第一模型的作用；
所述第一模型的生命周期信息。
可选地，所述装置还包括：
目标信息发送模块，用于在伪标签信息接收模块接收所述终端设备发送的伪标签的相关信息之前，向所述终端设备发送目标信息，其中，所述目标信息用于指示所述终端设备上报数据的相关信息。
可选地，所述目标信息包括如下中至少一项：
上报数据的格式；
用于指示上报伪标签的置信度或误差或准确性的指示信息；
筛选伪标签的置信度阈值或误差阈值或准确性阈值；
数据的上报分组方式；
用于指示上报数据的时间戳信息的指示信息。
可选地，所述装置还包括：
参考信号信息发送模块，用于在伪标签信息接收模块接收所述终端设备发送的伪标签的相关信息之前，向所述终端设备发送用于采集所述无标签数据集的下行参考信号信息。
可选地，所述伪标签的相关信息包括如下中至少一项：
所述伪标签和与所述伪标签对应的无标签数据；
所述伪标签的置信度或误差或准确性；
所述无标签数据集的采集时间戳信息；
所述伪标签的生成时间戳信息；
采集所述无标签数据集的终端设备的标识信息；
采集所述无标签数据集的终端设备所处的区域的标识信息；
所述无标签数据集关联的发送接收点的标识信息。
可选地，所述装置还包括：
上报资源分配模块，用于在伪标签信息接收模块接收所述终端设备发送的伪标签的相关信息之前，为所述终端设备分配所述伪标签的相关信息的上报资源。
本申请实施例中的模型训练装置可以是电子设备，例如具有操作系统的电子设备，也可以是电子设备中的部件，例如集成电路或芯片。该电子设备可以是网络侧设备。示例性的，网络侧设备可以包括但不限于上述所列举的网络侧设备12的类型。
本申请实施例提供的模型训练装置能够实现图4的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
第四方面，本申请实施例提供了一种模型训练装置，该装置可以应用于终端设备，如图8所示，该模型训练装置80包括：
模型信息接收模块801，用于接收网络侧设备发送的第一模型的信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
伪标签信息发送模块802，用于将伪标签的相关信息发送给所述网络侧设备，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。
由此可见，在本申请实施例中，网络侧设备能够将第一模型的信息发送给终端设备，从而接收终端设备发送的伪标签的相关信息，其中，所述第一模型是基于有标签数据集 对初始模型进行训练获得的，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。由此可见，本申请实施例给出了基于伪标签的半监督学习方法在无线通信中引入的必要信息交互流程以及可能的训练方式。相比于传统的监督学习方法，本申请的实施例，能够充分利用无标签的数据显著提升AI模型的精度，并扩展AI技术在无线通信领域诸多方向的应用。
可选地，所述装置还包括：
第一筛选模块，用于在所述伪标签信息发送模块将所述伪标签的相关信息发送给所述网络侧设备之前，确定各个所述伪标签的置信度，并将置信度小于置信度阈值的伪标签的相关信息剔除；
或者，
第二筛选模块，用于在所述伪标签信息发送模块将所述伪标签的相关信息发送给所述网络侧设备之前，确定各个所述伪标签的误差，并将误差大于误差阈值的伪标签的相关信息剔除；
或者，
第三筛选模块，用于在所述伪标签信息发送模块将所述伪标签的相关信息发送给所述网络侧设备之前，确定各个所述伪标签的准确性，并将准确性小于准确性阈值的伪标签的相关信息剔除。
可选地，所述第一模型的信息包括如下中的至少一项：
所述第一模型的结构信息；
所述第一模型的参数信息；
所述第一模型的配置信息；
所述第一模型的计算能力要求信息；
所述第一模型的存储能力要求信息；
所述第一模型关联的任务标识；
所述第一模型的作用；
所述第一模型的生命周期信息。
可选地，所述装置还包括：
目标信息接收模块，用于在所述伪标签信息发送模块将所述伪标签的相关信息发送给所述网络侧设备之前，接收所述网络侧设备发送的目标信息，其中，所述目标信息用于指示所述终端设备上报数据的相关信息；
所述伪标签信息发送模块，具体用于：
根据所述目标信息，将所述伪标签的相关信息发送给所述网络侧设备。
可选地，所述目标信息包括如下中至少一项：
上报数据的格式；
用于指示上报伪标签的置信度或误差或准确性的指示信息；
筛选伪标签的置信度阈值或误差阈值或准确性阈值；
数据的上报分组方式；
用于指示上报数据的时间戳信息的指示信息。
可选地，所述装置还包括：
参考信号信息接收模块，用于在所述第二数据获取模块获取无标签数据集之前，接收所述网络侧设备发送的下行参考信号信息；
数据获取模块用于：根据所述下行参考信号信息，采集所述无标签数据集。
可选地，所述伪标签的相关信息包括如下中至少一项：
所述伪标签和与所述伪标签对应的无标签数据；
所述伪标签的置信度或误差或准确性；
所述无标签数据集的采集时间戳信息；
所述伪标签的生成时间戳信息；
采集所述无标签数据集的终端设备的标识信息；
采集所述无标签数据集的终端设备所处的区域的标识信息；
所述无标签数据集关联的发送接收点的标识信息。
可选地，所述装置还包括：
上报资源获取模块，用于在所述伪标签信息发送模块将所述伪标签的相关信息发送给所述网络侧设备之前，获取所述网络侧设备为所述终端设备分配的上报资源；
所述伪标签信息发送模块具体用于：
通过所述上报资源，将所述伪标签的相关信息发送给所述网络侧设备。
本申请实施例中的模型训练装置可以是电子设备，例如具有操作系统的电子设备，也可以是电子设备中的部件，例如集成电路或芯片。该电子设备可以是终端设备。示例性的，终端设备可以包括但不限于上述所列举的终端设备11的类型。
本申请实施例提供的模型训练装置能够实现图5的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
可选地，如图9所示，本申请实施例还提供一种通信设备900，包括处理器901和存储器902，存储器902上存储有可在所述处理器901上运行的程序或指令，例如，该通信设备900为网络侧设备时，该程序或指令被处理器901执行时实现上述第一方面所述的模型训练方法实施例的各个步骤，且能达到相同的技术效果。该通信设备900为终端设备时，该程序或指令被处理器901执行时实现上述第二方面所述的模型训练方法实施例的各个步骤，且能达到相同的技术效果，为避免重复，这里不再赘述。
如图10所示，为实现本申请实施例的一种终端设备的硬件结构示意图。
该终端设备1000包括但不限于：射频单元1001、网络模块1002、音频输出单元1003、输入单元1004、传感器1005、显示单元1006、用户输入单元1007、接口单元1008、存储器1009以及处理器1010等中的至少部分部件。
本领域技术人员可以理解，终端设备1000还可以包括给各个部件供电的电源(比如电池)，电源可以通过电源管理系统与处理器1010逻辑相连，从而通过电源管理系统实现管理充电、放电、以及功耗管理等功能。图10中示出的终端设备结构并不构成对终端设备的限定，终端设备可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置，在此不再赘述。
应理解的是，本申请实施例中，输入单元1004可以包括图形处理单元(Graphics Processing Unit，GPU)10041和麦克风10042，图形处理器10041对在视频捕获模式或图像捕获模式中由图像捕获装置(如摄像头)获得的静态图片或视频的图像数据进行处理。显示单元1006可包括显示面板10061，可以采用液晶显示器、有机发光二极管等形式来配置显示面板10061。用户输入单元1007包括触控面板10071以及其他输入设备10072中的至少一种。触控面板10071，也称为触摸屏。触控面板10071可包括触摸检测装置和触摸控制器两个部分。其他输入设备10072可以包括但不限于物理键盘、功能键(比如音量控制按键、开关按键等)、轨迹球、鼠标、操作杆，在此不再赘述。
本申请实施例中，射频单元1001接收来自网络侧设备的下行数据后，可以传输给处理器1010进行处理；另外，射频单元1001可以向网络侧设备发送上行数据。通常，射频单元1001包括但不限于天线、放大器、收发信机、耦合器、低噪声放大器、双工器等。
存储器1009可用于存储软件程序或指令以及各种数据。存储器1009可主要包括存储程序或指令的第一存储区和存储数据的第二存储区，其中，第一存储区可存储操作系统、至少一个功能所需的应用程序或指令(比如声音播放功能、图像播放功能等)等。此外，存储器1009可以包括易失性存储器或非易失性存储器，或者，存储器1009可以包括易失性和非易失性存储器两者。其中，非易失性存储器可以是只读存储器(Read-Only Memory，ROM)、可编程只读存储器(Programmable ROM，PROM)、可擦除可编程只读存储器(Erasable PROM，EPROM)、电可擦除可编程只读存储器(Electrically EPROM，EEPROM)或闪存。易失性存储器可以是随机存取存储器(Random Access Memory，RAM)，静态随机存取存储器(Static RAM，SRAM)、动态随机存取存储器(Dynamic RAM，DRAM)、同步动态随机存取存储器(Synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(Double Data Rate SDRAM，DDRSDRAM)、增强型同步动态随机存取存储器(Enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(Synch link DRAM，SLDRAM)和直接内存总线随机存取存储器(Direct Rambus RAM，DRRAM)。本申请实施例中的存储器1009包括但不限于这些和任意其它适合类型的存储器。
处理器1010可包括一个或多个处理单元；可选地，处理器1010集成应用处理器和调制解调处理器，其中，应用处理器主要处理涉及操作系统、用户界面和应用程序等的操作，调制解调处理器主要处理无线通信信号，如基带处理器。可以理解的是，上述调制解调处理器也可以不集成到处理器1010中。
其中，射频单元1001用于接收网络侧设备发送的第一模型的信息，其中，所述第一 模型是基于有标签数据集对初始模型进行训练获得的；
射频单元1001用于将伪标签的相关信息发送给所述网络侧设备，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。
由此可知，在本申请实施例中，网络侧设备能够将第一模型的信息发送给终端设备，从而接收终端设备发送的伪标签的相关信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。由此可见，本申请实施例给出了基于伪标签的半监督学习方法在无线通信中引入的必要信息交互流程以及可能的训练方式。相比于传统的监督学习方法，本申请的实施例，能够充分利用无标签的数据显著提升AI模型的精度，并扩展AI技术在无线通信领域诸多方向的应用。
可选地，射频单元1001将伪标签的相关信息发送给所述网络侧设备之前，处理器1010用于：
确定各个所述伪标签的置信度；
将置信度小于置信度阈值的伪标签的相关信息剔除；
或者，
确定各个所述伪标签的误差；
将误差大于误差阈值的伪标签的相关信息剔除；
或者
确定各个所述伪标签的准确性；
将准确性小于准确性阈值的伪标签的相关信息剔除。
可选地，所述第一模型的信息包括如下中的至少一项：
所述第一模型的结构信息；
所述第一模型的参数信息；
所述第一模型的配置信息；
所述第一模型的计算能力要求信息；
所述第一模型的存储能力要求信息；
所述第一模型关联的任务标识；
所述第一模型的作用；
所述第一模型的生命周期信息。
可选地，射频单元1001将伪标签的相关信息发送给所述网络侧设备之前，还用于：
接收所述网络侧设备发送的目标信息，其中，所述目标信息用于指示所述终端设备上报数据的相关信息；
射频单元1001将伪标签的相关信息发送给所述网络侧设备，具体用于：
根据所述目标信息，将所述伪标签的相关信息发送给所述网络侧设备。
可选地，所述目标信息包括如下中至少一项：
上报数据的格式；
用于指示上报伪标签的置信度或误差或准确性的指示信息；
筛选伪标签的置信度阈值或误差阈值或准确性阈值；
数据的上报分组方式；
用于指示上报数据的时间戳信息的指示信息。
可选地，射频单元1001还用于：
接收所述网络侧设备发送的下行参考信号信息；
处理器1010获取无标签数据集，具体用于：
根据所述下行参考信号信息，采集所述无标签数据集。
可选地，所述伪标签的相关信息包括如下中至少一项：
所述伪标签和与所述伪标签对应的无标签数据；
所述伪标签的置信度或误差或准确性；
所述无标签数据集的采集时间戳信息；
所述伪标签的生成时间戳信息；
采集所述无标签数据集的终端设备的标识信息；
采集所述无标签数据集的终端设备所处的区域的标识信息；
所述无标签数据集关联的发送接收点的标识信息。
可选地，射频单元1001将伪标签的相关信息发送给所述网络侧设备之前，还用于：
获取所述网络侧设备为所述终端设备分配的上报资源；
射频单元1001将伪标签的相关信息发送给所述网络侧设备，具体用于：
通过所述上报资源，将所述伪标签的相关信息发送给所述网络侧设备。
本申请实施例还提供一种网络侧设备，如图11所示，该网络侧设备1100包括：天线111、射频装置112、基带装置113、处理器114和存储器115。天线111与射频装置112连接。在上行方向上，射频装置112通过天线111接收信息，将接收的信息发送给基带装置113进行处理。在下行方向上，基带装置113对要发送的信息进行处理，并发送给射频装置112，射频装置112对收到的信息进行处理后经过天线111发送出去。
以上实施例中网络侧设备执行的方法可以在基带装置113中实现，该基带装置113包括基带处理器。
基带装置113例如可以包括至少一个基带板，该基带板上设置有多个芯片，如图11所示，其中一个芯片例如为基带处理器，通过总线接口与存储器115连接，以调用存储器115中的程序，执行以上方法实施例中所示的网络设备操作。
该网络侧设备还可以包括网络接口116，该接口例如为通用公共无线接口(common public radio interface，CPRI)。
具体地，本发明实施例的网络侧设备1100还包括：存储在存储器115上并可在处理 器114上运行的指令或程序，处理器114调用存储器115中的指令或程序执行图6所示各模块执行的方法，并达到相同的技术效果，为避免重复，故不在此赘述。
本申请实施例还提供了一种网络侧设备。如图12所示，该网络侧设备1200包括：处理器1201、网络接口1202和存储器1203。其中，网络接口1202例如为通用公共无线接口(common public radio interface，CPRI)。
具体地，本发明实施例的网络侧设备1200还包括：存储在存储器1203上并可在处理器1201上运行的指令或程序，处理器1201调用存储器1203中的指令或程序执行图4所示各模块执行的方法，并达到相同的技术效果，为避免重复，故不在此赘述。
本申请实施例还提供一种可读存储介质，所述可读存储介质上存储有程序或指令，该程序或指令被处理器执行时实现上述模型训练方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
其中，所述处理器为上述实施例中所述的终端设备中的处理器。所述可读存储介质，包括计算机可读存储介质，如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。
本申请实施例另提供了一种芯片，所述芯片包括处理器和通信接口，所述通信接口和所述处理器耦合，所述处理器用于运行程序或指令，实现上述模型训练方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
应理解，本申请实施例提到的芯片还可以称为系统级芯片，系统芯片，芯片系统或片上系统芯片等。
本申请实施例另提供了一种计算机程序/程序产品，所述计算机程序/程序产品被存储在存储介质中，所述计算机程序/程序产品被至少一个处理器执行以实现上述模型训练方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
本申请实施例还提供了一种模型训练系统，包括：终端设备及网络侧设备，所述终端可用于执行如上第二方面所述的模型训练方法的步骤，所述网络侧设备可用于执行如上第一方面所述的模型训练方法的步骤。
需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外，需要指出的是，本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能，还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能，例如，可以按不同于所描述的次序来执行所描述的方法，并且还可以添加、省去、或组合各种步骤。另外，参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法 可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以计算机软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述，但是本申请并不局限于上述的具体实施方式，上述的具体实施方式仅仅是示意性的，而不是限制性的，本领域的普通技术人员在本申请的启示下，在不脱离本申请宗旨和权利要求所保护的范围情况下，还可做出很多形式，均属于本申请的保护之内。

Claims (20)

  1. 一种模型训练方法，其中，所述方法包括：
    网络侧设备将第一模型的信息发送给终端设备，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
    所述网络侧设备接收所述终端设备发送的伪标签的相关信息，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的；
    所述网络侧设备基于所述有标签数据集和所述伪标签的相关信息，对所述第一模型进行训练，得到第三模型。
  2. 根据权利要求1所述的方法，其中，所述第一模型的信息包括如下中的至少一项：
    所述第一模型的结构信息；
    所述第一模型的参数信息；
    所述第一模型的配置信息；
    所述第一模型的计算能力要求信息；
    所述第一模型的存储能力要求信息；
    所述第一模型关联的任务标识；
    所述第一模型的作用；
    所述第一模型的生命周期信息。
  3. 根据权利要求1所述的方法，其中，所述网络侧设备接收所述终端设备发送的伪标签的相关信息之前，所述方法还包括：
    所述网络侧设备向所述终端设备发送目标信息，其中，所述目标信息用于指示所述终端设备上报数据的相关信息。
  4. 根据权利要求3所述的方法，其中，所述目标信息包括如下中至少一项：
    上报数据的格式；
    用于指示上报伪标签的置信度或误差或准确性的指示信息；
    筛选伪标签的置信度阈值或误差阈值或准确性阈值；
    数据的上报分组方式；
    用于指示上报数据的时间戳信息的指示信息。
  5. 根据权利要求1所述的方法，其中，所述网络侧设备接收所述终端设备发送的伪标签的相关信息之前，所述方法还包括：
    所述网络侧设备向所述终端设备发送用于采集所述无标签数据集的下行参考信号信息。
  6. 根据权利要求1所述的方法，其中，所述伪标签的相关信息包括如下中至少一项：
    所述伪标签和与所述伪标签对应的无标签数据；
    所述伪标签的置信度或误差或准确性；
    所述无标签数据集的采集时间戳信息；
    所述伪标签的生成时间戳信息；
    采集所述无标签数据集的终端设备的标识信息；
    采集所述无标签数据集的终端设备所处的区域的标识信息；
    所述无标签数据集关联的发送接收点的标识信息。
  7. 根据权利要求1所述的方法，其中，所述网络侧设备接收所述终端设备发送的伪标签的相关信息之前，所述方法还包括：
    所述网络侧设备为所述终端设备分配所述伪标签的相关信息的上报资源。
  8. 一种模型训练方法，其中，所述方法包括：
    终端设备接收网络侧设备发送的第一模型的信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
    所述终端设备将伪标签的相关信息发送给所述网络侧设备，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。
  9. 根据权利要求8所述的方法，其中，所述终端设备将伪标签的相关信息发送给所述网络侧设备之前，所述方法还包括：
    所述终端设备确定各个所述伪标签的置信度；
    所述终端设备将置信度小于置信度阈值的伪标签的相关信息剔除；
    或者，
    所述终端设备确定各个所述伪标签的误差；
    所述终端设备将误差大于误差阈值的伪标签的相关信息剔除；
    或者
    所述终端设备确定各个所述伪标签的准确性；
    所述终端设备将准确性小于准确性阈值的伪标签的相关信息剔除。
  10. 根据权利要求8所述的方法，其中，所述第一模型的信息包括如下中的至少一项：
    所述第一模型的结构信息；
    所述第一模型的参数信息；
    所述第一模型的配置信息；
    所述第一模型的计算能力要求信息；
    所述第一模型的存储能力要求信息；
    所述第一模型关联的任务标识；
    所述第一模型的作用；
    所述第一模型的生命周期信息。
  11. 根据权利要求8所述的方法，其中，所述终端设备将伪标签的相关信息发送给所述网络侧设备之前，所述方法还包括：
    所述终端设备接收所述网络侧设备发送的目标信息，其中，所述目标信息用于指示所述终端设备上报数据的相关信息；
    所述终端设备将伪标签的相关信息发送给所述网络侧设备，包括：
    所述终端设备根据所述目标信息，将所述伪标签的相关信息发送给所述网络侧设备。
  12. 根据权利要求11所述的方法，其中，所述目标信息包括如下中至少一项：
    上报数据的格式；
    用于指示上报伪标签的置信度或误差或准确性的指示信息；
    筛选伪标签的置信度阈值或误差阈值或准确性阈值；
    数据的上报分组方式；
    用于指示上报数据的时间戳信息的指示信息。
  13. 根据权利要求8所述的方法，其中，所述方法还包括：
    所述终端设备接收所述网络侧设备发送的下行参考信号信息；
    所述终端设备获取无标签数据集的过程，包括：
    所述终端设备根据所述下行参考信号信息，采集所述无标签数据集。
  14. 根据权利要求8所述的方法，其中，所述伪标签的相关信息包括如下中至少一项：
    所述伪标签和与所述伪标签对应的无标签数据；
    所述伪标签的置信度或误差或准确性；
    所述无标签数据集的采集时间戳信息；
    所述伪标签的生成时间戳信息；
    采集所述无标签数据集的终端设备的标识信息；
    采集所述无标签数据集的终端设备所处的区域的标识信息；
    所述无标签数据集关联的发送接收点的标识信息。
  15. 根据权利要求8所述的方法，其中，所述终端设备将伪标签的相关信息发送给所述网络侧设备之前，所述方法还包括：
    所述终端设备获取所述网络侧设备为所述终端设备分配的上报资源；
    所述终端设备将伪标签的相关信息发送给所述网络侧设备，包括：
    所述终端设备通过所述上报资源，将所述伪标签的相关信息发送给所述网络侧设备。
  16. 一种模型训练装置，其中，所述装置包括：
    模型信息发送模块，用于将第一模型的信息发送给终端设备，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
    伪标签信息接收模块，用于接收所述终端设备发送的伪标签的相关信息，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的；
    训练模块，用于基于所述有标签数据集和所述伪标签的相关信息，对所述第一模型进行训练，得到第三模型。
  17. 一种模型训练装置，其中，所述装置包括：
    模型信息接收模块，用于接收网络侧设备发送的第一模型的信息，其中，所述第一模型是基于有标签数据集对初始模型进行训练获得的；
    伪标签信息发送模块，用于将伪标签的相关信息发送给所述网络侧设备，其中，所述伪标签为无标签数据集输入至第二模型后，所述第二模型输出的伪标签，所述第二模型是根据所述第一模型的信息确定的。
  18. 一种网络侧设备，其中，包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，所述程序或指令被所述处理器执行时实现如权利要求1至7任一项所述的模型训练方法的步骤。
  19. 一种终端设备，其中，包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，所述程序或指令被所述处理器执行时实现如权利要求8至15任一项所述的模型训练方法的步骤。
  20. 一种可读存储介质，其中，所述可读存储介质上存储程序或指令，所述程序或指令被处理器执行时实现如权利要求1-7任一项所述的模型训练方法，或者实现如权利要求8至15任一项所述的模型训练方法的步骤。
PCT/CN2023/107627 2022-07-21 2023-07-17 模型训练方法、装置、网络侧设备及终端设备 Ceased WO2024017176A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210873126.4A CN117493864A (zh) 2022-07-21 2022-07-21 模型训练方法、装置、网络侧设备及终端设备
CN202210873126.4 2022-07-21

Publications (1)

Publication Number Publication Date
WO2024017176A1 true WO2024017176A1 (zh) 2024-01-25

Family

ID=89617140

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/107627 Ceased WO2024017176A1 (zh) 2022-07-21 2023-07-17 模型训练方法、装置、网络侧设备及终端设备

Country Status (2)

Country Link
CN (1) CN117493864A (zh)
WO (1) WO2024017176A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2025237165A1 (zh) * 2024-05-16 2025-11-20 维沃移动通信有限公司 数据收集方法、装置、终端及网络侧设备

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118414869A (zh) * 2024-02-22 2024-07-30 上海移远通信技术股份有限公司 用于定位的方法及通信设备
WO2025199902A1 (zh) * 2024-03-28 2025-10-02 北京小米移动软件有限公司 模型训练方法、通信设备和存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113705769A (zh) * 2021-05-17 2021-11-26 华为技术有限公司 一种神经网络训练方法以及装置
CN114154530A (zh) * 2021-10-26 2022-03-08 北京航空航天大学 一种心电时序信号房颤检测模型训练方法及装置
CN114169389A (zh) * 2021-10-22 2022-03-11 福建亿榕信息技术有限公司 一种类别扩增的目标检测模型训练方法和存储设备

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113095346A (zh) * 2020-01-08 2021-07-09 华为技术有限公司 数据标注的方法以及数据标注的装置
CN114422380B (zh) * 2020-10-09 2023-06-09 维沃移动通信有限公司 神经网络信息传输方法、装置、通信设备和存储介质
CN113326826A (zh) * 2021-08-03 2021-08-31 新石器慧通（北京）科技有限公司 网络模型的训练方法、装置、电子设备及存储介质
CN114399995B (zh) * 2022-01-20 2025-07-15 腾讯科技（深圳）有限公司 语音模型的训练方法、装置、设备及计算机可读存储介质

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113705769A (zh) * 2021-05-17 2021-11-26 华为技术有限公司 一种神经网络训练方法以及装置
CN114169389A (zh) * 2021-10-22 2022-03-11 福建亿榕信息技术有限公司 一种类别扩增的目标检测模型训练方法和存储设备
CN114154530A (zh) * 2021-10-26 2022-03-08 北京航空航天大学 一种心电时序信号房颤检测模型训练方法及装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2025237165A1 (zh) * 2024-05-16 2025-11-20 维沃移动通信有限公司 数据收集方法、装置、终端及网络侧设备

Also Published As

Publication number Publication date
CN117493864A (zh) 2024-02-02

Similar Documents

Publication Publication Date Title
CN115913486B (zh) 信息上报方法、装置、终端及可读存储介质
WO2024017176A1 (zh) 模型训练方法、装置、网络侧设备及终端设备
WO2023160534A1 (zh) 数据采集方法及设备
CN113284142A (zh) 图像检测方法、装置、计算机可读存储介质及计算机设备
CN116233906B (zh) 测量方法、装置、设备及存储介质
CN116939713A (zh) 信息反馈方法、装置及设备
WO2024008111A1 (zh) 数据采集方法及装置
WO2023040886A1 (zh) 数据采集方法及装置
WO2024067281A1 (zh) Ai模型的处理方法、装置及通信设备
WO2023125951A1 (zh) 通信模型配置方法、装置和通信设备
WO2024120445A1 (zh) 模型输入信息的确定方法、装置、设备、系统及存储介质
WO2023186091A1 (zh) 样本确定方法、装置及设备
WO2024208260A1 (zh) 基于ai的csi预测的数据采集方法及装置
WO2024083004A1 (zh) Ai模型配置方法、终端及网络侧设备
WO2023040888A1 (zh) 数据传输方法及装置
CN117896838A (zh) Ai模型推理方法、设备及可读存储介质
CN118214750A (zh) Ai算力上报方法、终端及网络侧设备
JP2025507145A (ja) 通信ネットワークにおけるデータ処理方法、ネットワーク側機器及び可読記憶媒体
WO2023155839A1 (zh) Ai模型的在线学习方法、装置、通信设备及可读存储介质
CN117390440A (zh) 人工智能模型训练方法、装置、网络侧设备、终端及介质
CN116776988A (zh) 通信网络中的数据处理方法及网络侧设备
CN117910589A (zh) 模型请求方法、装置、通信设备及可读存储介质
WO2024093799A1 (zh) 数据标注方法、装置、终端设备及网络侧设备
WO2025092999A1 (zh) 模型性能监督方法、装置及设备
WO2025108195A1 (zh) 模型确定方法、装置及通信设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23842239

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 23842239

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 23842239

Country of ref document: EP

Kind code of ref document: A1