WO2023155839A1

WO2023155839A1 - Online learning method and apparatus for ai model, and communication device and readable storage medium

Info

Publication number: WO2023155839A1
Application number: PCT/CN2023/076492
Authority: WO
Inventors: 贾承璐; 孙布勒; 王园园
Original assignee: 维沃移动通信有限公司
Priority date: 2022-02-21
Filing date: 2023-02-16
Publication date: 2023-08-24
Also published as: CN116668309A

Abstract

The present application discloses an online learning method and apparatus for an artificial intelligence (AI) model, and a communication device and a readable storage medium. The online learning method for the AI model of embodiments of the present application comprises: a second device configures a first AI model for a first device; and the second device configures online learning information of the first AI model for the first device.

Description

AI model online learning method, device, communication device and readable storage medium

Cross References to Related Applications

This application claims priority to Chinese Patent Application No. 202210157466.7 filed in China on February 21, 2022, the entire contents of which are hereby incorporated by reference.

technical field

The application belongs to the field of electronic information technology, and specifically relates to an AI model online learning method, device, communication device and readable storage medium.

Background technique

With the widespread application of artificial intelligence (AI) in various fields, it has become an important task for wireless communication networks to integrate AI into wireless communication networks to improve technical indicators such as network throughput, delay, and user capacity. At present, there are many ways to realize the AI module in the wireless communication network. For example, neural network (neural network, NN) decision tree (decision tree, DT), support vector machine (support vector machine, SVM), genetic algorithm (genetic algorithm, GA), etc.

In related technologies, the AI model is usually trained offline, and then the trained AI model is deployed in a wireless communication system. However, when the wireless communication environment changes, the accuracy of the output result of the AI model is low. In this way, the calculation accuracy of the AI model is poor.

Contents of the invention

Embodiments of the present application provide an online AI model learning method, device, communication device, and readable storage medium, which can solve the problem of invalidation of the AI model caused by dynamic changes in the wireless communication environment in actual scenarios.

In the first aspect, an online learning method of an AI model is provided, the method includes: a second device configures a first AI model for a first device; the second device configures online learning information of the first AI model for the first device .

In a second aspect, an online AI model learning device is provided, which includes: a configuration module, wherein: the configuration module is used for the second device to configure the first AI model for the first device; the configuration module also The online learning information is used for the second device to configure the first AI model for the first device.

In a third aspect, an online learning method of an AI model is provided, the method includes: an acquisition module and an execution module, wherein: the acquisition module is used for the first device to acquire the first AI model; the execution module, The first device performs online learning on the first AI model based on the online learning information of the first AI model.

In a fourth aspect, an online learning device for an AI model is provided, the device includes: a first device acquires a first AI model; the first device, based on the online learning information of the first AI model, AI models for online learning.

In a fifth aspect, a communication device is provided, the communication device includes a processor and a memory, the memory stores programs or instructions that can run on the processor, and the programs or instructions are implemented when executed by the processor The steps of the method as described in the first aspect.

In a sixth aspect, a communication device is provided, including a processor and a communication interface, wherein the processor is configured to configure a first AI model for a first device; and configure online learning of the first AI model for the first device information.

In a seventh aspect, a communication device is provided, the communication device includes a processor and a memory, the memory stores programs or instructions that can run on the processor, and the programs or instructions are implemented when executed by the processor The steps of the method as described in the first aspect.

In an eighth aspect, a network side device is provided, including a processor and a communication interface, wherein the above-mentioned processor is used to obtain a first AI model, and based on the online learning information of the first AI model, the first AI model Take online learning.

In the ninth aspect, a readable storage medium is provided, and programs or instructions are stored on the readable storage medium, and when the programs or instructions are executed by a processor, the steps of the method described in the first aspect are realized, or the steps of the method described in the first aspect are realized, or The steps of the method described in the third aspect.

In a tenth aspect, a chip is provided, the chip includes a processor and a communication interface, the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the method as described in the first aspect , or implement the method described in the third aspect.

In an eleventh aspect, a computer program/program product is provided, and the computer program/program product is stored in a storage medium In the above, the computer program/program product is executed by at least one processor to implement the first aspect, or to implement the steps of the online learning method of the AI model as described in the third aspect.

In this embodiment of the present application, the first device acquires a first AI model, and performs online learning on the first AI model based on online learning information of the first AI model. With this method, by deploying the first AI model on the first device side and configuring the first model with parameters required for online learning, the first AI model can be continuously adjusted online on the first device side, thereby maintaining The predictive performance of the first AI model, thereby ensuring the service quality of the first device.

Description of drawings

FIG. 1 is a block diagram of a wireless communication system provided by an embodiment of the present application;

Fig. 2 is one of the schematic flow charts of the online learning method of the AI model provided by the embodiment of the present application;

Fig. 3 is the second schematic flow diagram of the online learning method of the AI model provided by the embodiment of the present application;

Fig. 4 is one of the structural schematic diagrams of the online learning device of the AI model provided by the embodiment of the present application;

Fig. 5 is the second structural schematic diagram of the online learning device of the AI model provided by the embodiment of the present application;

FIG. 6 is a schematic structural diagram of a communication device provided by an embodiment of the present application;

FIG. 7 is a schematic diagram of a hardware structure of a terminal provided by an embodiment of the present application;

FIG. 8 is one of the schematic diagrams of the hardware structure of the network side device provided by the embodiment of the present application;

FIG. 9 is the second schematic diagram of the hardware structure of the network side device provided by the embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all of them. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in this application belong to the protection scope of this application.

The terms "first", "second" and the like in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific sequence or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the application are capable of operation in sequences other than those illustrated or described herein and that "first" and "second" distinguish objects. It is usually one category, and the number of objects is not limited. For example, there may be one or more first objects. In addition, "and/or" in the description and claims means at least one of the connected objects, and the character "/" generally means that the related objects are an "or" relationship.

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

Fig. 1 shows a block diagram of a wireless communication system to which the embodiment of the present application is applicable. The wireless communication system includes a terminal 11 and a network side device 12 . Wherein, the terminal 11 can 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, a super mobile personal computer (ultra-mobile personal computer, UMPC), mobile Internet device (Mobile Internet Device, MID), augmented reality (augmented reality, AR) / virtual reality (virtual reality, VR) equipment, robot, wearable device (Wearable Device) , vehicle equipment (VUE), pedestrian terminal (PUE), smart home (home equipment with wireless communication functions, such as refrigerators, TVs, washing machines or furniture, etc.), game consoles, personal computers (personal computers, PCs), teller machines or self-service 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. It should be noted that, the embodiment of the present application does not limit the specific type of the terminal 11 . 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., and the base station may be called a node B, an evolved node B (eNB), an access point, a base transceiver station (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 other in the field An appropriate term, as long as the same technical effect is achieved, the base station is not limited to a specific technical vocabulary. It should be noted that in the embodiment of this application, only the base station in the NR system is used as an example to introduce, and the specific details of the base station are not limited. type. The core network equipment may include but not limited to at least one of the following: core network node, core network function, mobility management entity (Mobility Management Entity, MME), access mobility management function (Access and Mobility Management Function, AMF), session management function (Session Management Function, SMF), user plane function (User Plane Function, UPF), policy control function (Policy Control Function, PCF), policy and charging rules function unit (Policy and Charging Rules Function, PCRF), edge application service Discovery function (Edge Application Server Discovery Function, EASDF), unified data management (Unified Data Management, UDM), unified data storage (Unified Data Repository, UDR), home subscriber server (Home Subscriber Server, HSS), centralized network configuration ( Centralized network configuration, CNC), network storage function (Network Repository Function, NRF), network exposure function (Network Exposure Function, NEF), local NEF (Local NEF, or L-NEF), binding support function (Binding Support Function, BSF), Application Function (Application Function, AF), etc. It should be noted that, in the embodiment of the present application, only the core network equipment in the NR system is used as an example for introduction, and the specific type of the core network equipment is not limited.

Some terms involved in the embodiments of the present invention are explained below:

(1) Artificial Intelligence (AI): Artificial intelligence is a very broad science, which consists of different fields, such as machine learning, computer vision and so on.

(2) Machine Learning (Machine Learning, ML): Machine learning is an important branch of artificial intelligence, which mainly studies how to make computers have the ability to learn by themselves. Machine learning algorithms include neural network (neural network, NN) decision tree (decision tree, DT), support vector machine (support vector machine, SVM), genetic algorithm (genetic algorithm, GA) and so on.

(3) Neural network: A neural network consists of a large number of nodes, which are called neurons. Among them, the composition information of neurons includes: input (a1, a2,...aK) weight/multiplicative coefficient (w), bias/additive coefficient (b), activation function (σ(.)). Common activation functions include Sigmoid, tanh, ReLU (Rectified Linear Unit), linear rectification function, corrected linear unit) and so on.

Further, the parameters of the neural network can be optimized by gradient optimization algorithm. The gradient optimization algorithm is a class of algorithms that minimize or maximize an objective function (sometimes called a loss function), and the objective function is often a mathematical combination of model parameters and data. For example, given data X and its corresponding label Y, after constructing a neural network model f(.), with the model, the predicted output f(x) can be obtained according to the input x, and the predicted value and the real value can be calculated The gap between (f(x)-Y), which is the loss function. If a suitable W,b is found to minimize the value of the above loss function, the smaller the loss value, the closer the model is to the real situation.

For example, currently common optimization algorithms are usually based on BP (error Back Propagation, error back propagation) algorithm. The basic idea of the BP algorithm is that the learning process consists of two processes: the forward propagation of the signal and the back propagation of the error. During forward propagation, the input samples are passed in from the input layer, processed layer by layer by each hidden layer, and passed 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 transmit the output error layer by layer through the hidden layer to the input layer in some form, and distribute the error to all the units of each layer, so as to obtain the error signal of each layer unit, and this error signal is used as the correction unit Basis for weight. This weight adjustment process of each layer of signal forward propagation and error back propagation is carried out repeatedly. The process of continuously adjusting the weights is also the learning and training process of the network. This process has been carried out until the error of the network output is reduced to an acceptable level, or until the preset number of learning times.

For example, 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), stochastic gradient descent with momentum (Nesterov), adaptive gradient descent (ADAptive GRADient descent, Adagrad), Adadelta, root mean square error deceleration (root mean square prop, RMSprop), adaptive momentum estimation (Adaptive Moment Estimation, Adam), etc.

As an example, the above optimization algorithm is based on the error/loss obtained by the loss function when the error is backpropagated, and calculates the derivative/partial derivative of the current neuron, plus the learning rate, the previous gradient/derivative/partial derivative, etc., Get the gradient and pass the gradient to the previous layer.

The online learning method provided by the embodiment of the present application will be described in detail below through some embodiments and application scenarios with reference to the accompanying drawings.

The current AI research in the field of wireless communication mainly focuses on offline learning and deployment. Since the wireless environment is constantly changing, the fixed AI model obtained through offline training will gradually fail in the dynamic environment. How to improve the model in the new changing environment The adaptability among them has become an urgent problem to be solved.

This application is an example to solve the above problems, and proposes an online learning method for AI models. Further, the realization of online learning has the following difficulties: 1) limited by the storage capacity and data collection capacity of the device (for example, the time cost and hardware cost of collecting data are relatively high), it is usually difficult to obtain a large enough data set for online training; 2) Limited by the computing power of the device and the limited data set, it may not be possible to perform multiple rounds of model fine-tuning or over-fitting will result after multiple rounds of model fine-tuning; 3) For wireless communication, there are also communication delay limitations and communication The problem of continuity, which puts forward requirements on the time of the first equipment data collection and the time of online learning.

FIG. 2 shows a flow chart of an online learning method for an AI model provided by an embodiment of the present application. As shown in Figure 2, the implementation of this application The online learning method of the AI model provided by the example may include the following steps 201 and 202:

Step 201: the second device configures the first AI model for the first device.

In the embodiment of the present application, the above-mentioned second device may include at least one of the following: core network equipment, access network equipment, and terminal; the above-mentioned first device may include at least one of the following: core network equipment, access network equipment, and terminal .

Exemplarily, the second device is a core network device, and correspondingly, the first device may be an access network device or a terminal.

Exemplarily, the second device is an access network device, and correspondingly, the first device may be a core network device or a terminal.

Exemplarily, the second device is a terminal, and correspondingly, the first device may be a core network device or an access network device.

In the embodiment of the present application, the above-mentioned first AI model is an AI model obtained by offline training on the second device side.

Optionally, in the embodiment of the present application, the algorithm of the first AI model may include at least one of the following: a neural network, a decision tree, a support vector machine, and a Bayesian classifier.

Optionally, in this embodiment of the present application, the above-mentioned first AI model may be an AI model used for terminal positioning, network optimization, processing of large input data sets, and network recommendation for users.

Optionally, in this embodiment of the present application, the second device may train the AI model based on a preset learning framework, so as to obtain the above-mentioned first AI model. Exemplarily, the first AI model is a neural network model as an example. The second device can train the neural network model based on a preset learning framework to obtain the first neural network model.

Optionally, in this embodiment of the present application, the second device may send the trained first AI model to the first device, and deploy the first AI model on the side of the first device.

Step 202: the second device configures online learning information of the first AI model for the first device.

In this embodiment of the application, the second device can send the information required for the online learning of the first AI model to the first device, and the first device can, based on the online learning information sent by the second device, The learning information performs online learning on the first AI model.

In this embodiment of the present application, the online learning information is configured by the network device, or determined independently by the second device.

Optionally, in this embodiment of the present application, the foregoing first device may be a terminal, and the second device may be a core network device.

Exemplarily, the core network device sends the first AI model to the terminal, and sends the online learning information of the first AI model to the terminal, and the terminal receives the first AI model sent by the core network device, and the online learning information of the first AI model information.

In the online learning method provided in the embodiment of the present application, the second device configures the first AI model for the first device, and configures the online learning information of the first AI model for the first device. With this method, by deploying the first AI model on the first device side and configuring the first model with parameters required for online learning, the first AI model can be continuously adjusted online on the first device side, thereby maintaining The predictive performance of the first AI model, thereby ensuring the service quality of the first device.

Optionally, in this embodiment of the application, the online learning information includes at least one of the following:

How eLearning is triggered;

Conditions for discontinuation of online learning;

Parameter configuration information for online learning;

Datasets for online learning.

Exemplarily, the above-mentioned triggering manner is related to the state information of the first device and the channel information of a channel related to the first device. For example, when the moving speed of the first device is fast, or when the channel environment of the working channel of the first device changes, online learning of the first AI model may be triggered, so that the first AI model continuously adapts to changing environment.

Exemplarily, the first device may suspend the online learning of the first AI model when the number of online learning of the first AI model is greater than the preset number of iterations; or, when the first AI model reaches the preset accuracy , suspending the online learning of the first AI model; or, in the case that the error information of the output result of the first AI model is small, suspending the online learning of the first AI model. Since the number of online learning of the first AI model is greater than the preset number of iterations, or the accuracy of the first AI model reaches the preset accuracy, indicating that the current first AI model is valid, the online learning of the first AI model can be terminated, to save power consumption.

Optionally, in this embodiment of the application, the trigger condition corresponding to the above trigger mode includes at least one of the following:

The state information of the above-mentioned first device satisfies a first preset condition;

The amount of data collected by the first device is greater than a first threshold;

The measurement information of the above-mentioned first device satisfies a second preset condition;

The error information of the output result of the first AI model is greater than the second threshold;

The statistical information of the first information corresponding to the above-mentioned first AI model satisfies a third preset condition;

The above-mentioned statistical information of the measurement information of the first device satisfies a fourth preset condition;

Optionally, the above status information includes at least one of the following: moving speed, beam switching information, and cell switching information.

Optionally, the amount of data collected by the above-mentioned first device may be the amount of online data collected by the first device, for example, the first device collects in real time received channel information.

For example, when the first device is moving fast, the first device is triggered to learn the first AI model online; or, when the amount of data collected by the first device is large, the first device is triggered to learn the first AI model. The AI model performs online learning; or, when the measurement information of the first device indicates that the channel environment of the current channel changes, the first device is triggered to perform online learning on the first AI model; or, when the output result of the first AI model When the error of the first AI model is relatively large or the accuracy of the first AI model is low, the first device is triggered to perform online learning on the first AI model. When the terminal moves rapidly or the channel environment changes, the first AI model may fail. In this way, the first device can learn the first AI model online based on information such as its own moving speed, the channel environment of the relevant channel, and the accuracy value of the AI model when the first AI model fails, thereby improving the first AI model. The prediction accuracy of the AI model.

Exemplarily, the channel information may include at least one of the following: signal launch angle information, signal arrival angle angle information, signal delay information in the channel, signal quality in the channel, and so on.

Optionally, the above-mentioned first threshold may be 3000, 5000 or 7000 and so on.

Optionally, the measurement information includes at least one of the following: first measurement information of a reference signal received by the first device, and second measurement information collected by a sensor of the first device. Exemplarily, the first measurement information includes at least one of the following: instantaneous measurement information of the reference signal, and statistical measurement information of the reference signal. Exemplarily, the instantaneous measurement information of the reference signal may be: measurement information of the reference signal at a specific moment; the statistical measurement information of the reference signal may be: measurement information of the reference signal within a period of time.

Exemplarily, the aforementioned reference signal includes at least one of the following: a synchronization signal block SSB, a CSI reference signal CSI-RS, a sounding reference signal SRS, and a positioning reference signal PRS. Optionally, the aforementioned sensors may include at least one of the following: vision sensors, radar sensors, position sensors and the like.

Optionally, the above-mentioned first information includes at least one of the following: input information of the above-mentioned first AI model, and output information of the above-mentioned first AI model.

Exemplarily, taking the first device as a terminal as an example, in the case where the first information includes the input information of the first AI model, the first information may be the working channel or the channel information of the surrounding channel of the terminal. In the first information In the case where the output information of the first AI model is included, the first information may be location information of the terminal.

Optionally, in this embodiment of the present application, the statistical information of the above-mentioned first information includes at least one of the following:

The first statistic of the above-mentioned first information in the first time window, the second statistic corresponding to the above-mentioned first information in at least two consecutive second time windows, and the at least two terminals under the first cell at the first moment Statistical information of the first information, and correlation information of the above-mentioned first information.

Wherein, the above-mentioned second statistic is calculated based on the statistic in each second time window.

Optionally, the above statistical information may include at least one of the following: mean value, variance and so on.

Optionally, the foregoing statistical information may include temporal statistical information and spatial statistical information. For example, the statistical information on time may be: statistical information on channels of the same terminal within a continuous period of time, and the statistical information on space may be statistical information on channels of multiple different terminals under one cell.

Exemplarily, the statistical information of the above-mentioned first information is used to represent whether the wireless network environment within the action area of the first AI model changes.

Exemplarily, the first information is channel information as an example. The average value of channel information in a certain continuous time window is less than a certain threshold, or the correlation index of channel information in two time windows is lower than a certain threshold, or, the correlation between the front and rear data in the current time window is lower than In the case of a certain threshold, if the channel environment representing the relevant channel of the first device changes, the first device can be triggered to perform online learning of the first AI model to adapt to the current channel environment, thereby improving the prediction accuracy of the first AI model .

The statistical information of the above-mentioned first information is explained below by taking the first information as channel information as an example.

Exemplarily, when the statistical information of the first information includes the first statistical quantity of the first information in the first time window, the statistical information of the first information may be: The mean or variance of the channel information.

Exemplarily, when the statistical information of the first information includes the second statistical quantity corresponding to the above-mentioned first information in at least two consecutive second time windows, the above-mentioned statistical information of the first information may be: The mean value determined by the mean value of the channel information in each consecutive time window in consecutive time windows, for example, the mean value of the channel information in time window 1 is a, the mean value of channel information in time window 2 is b, and the mean value of channel information in time window 3 is The mean value of is c, then the statistical information of the channel information is the mean value of a, b and c.

Exemplarily, when the statistical information of the first information includes the statistical information of the first information of at least two terminals under the first cell at the first moment, the statistical information of the above-mentioned first information may be: The mean value of the channel information of different terminals at a certain moment, for example, the mean values of the channel information of terminal A, terminal B and terminal C in the same cell at time 1 are d, e and f respectively, then the statistical information of the channel information is Means of d, e and f.

Exemplarily, when the statistical information of the first information includes the correlation information of the first information, the statistical information of the above-mentioned first information It may be: the correlation index of the channel information of the two time windows is lower than a certain threshold, for example, the distance between the previous and subsequent data in the current time window is smaller than a certain threshold.

Optionally, in this embodiment of the present application, the statistical information of the measurement information of the first device includes at least one of the following:

The third statistic of the measurement information in the third time window, the fourth statistic of the measurement information corresponding to at least two consecutive fourth time windows, and the correlation information of the measurement information.

Wherein, the above fourth statistic is calculated based on the statistic in each fourth time window.

Optionally, the above correlation information includes at least one of the following: distance between data, covariance, and correlation coefficient.

Exemplarily, it is taken that the measurement information is channel measurement information of a reference signal received by the first device as an example. The statistical information of the above measurement information may be: the variance of the channel measurement information in a certain continuous time window, or the average of the mean values of the channel measurement information in two consecutive time windows, or the channel measurement information in the current time window distance between the data.

Exemplarily, the above statistical information of the measurement information is used to represent whether the wireless network environment within the active area of the first AI model changes.

Exemplarily, the measurement information is channel information as an example. The mean value of the measurement information in a continuous time window is less than a certain threshold, or the correlation index of the measurement information in two time windows is lower than a certain threshold, or, the correlation between the front and rear data in the current time window is lower than In the case of a certain threshold, if the wireless environment representing the role of the first AI model changes, the first device can be triggered to perform online learning of the first AI model to adapt to the current wireless environment, thereby improving the prediction accuracy of the first AI model .

Further optionally, in the embodiment of the present application, the trigger condition includes: the status information of the first device meets a first preset condition; the satisfaction of the first preset condition includes at least one of the following:

The moving speed of the first device is greater than a third threshold;

The beam switching information indicates that beam switching occurs to the first device, and the beam switching frequency is greater than a fourth threshold;

The above cell switching information indicates that the cell switching occurs to the first device.

Exemplarily, the above-mentioned third threshold may be 60km/h, 80km/h, 100km/h and so on.

Further optionally, in the embodiment of the present application, the above-mentioned trigger condition includes: the measurement information of the above-mentioned first device meets the second preset condition; optionally, the above-mentioned meeting the second preset condition includes: the measurement information of the first device It indicates that the channel environment of the relevant channel of the first device changes.

Exemplarily, it is taken that the measurement information is a reference signal received by the first device as an example. The above-mentioned second preset condition may be: the first device estimates the downlink channel according to the measurement of CSI-RS, and detects that the channel environment changes, such as changing from a line of sight (LOS) environment to a non-line of sight (not line of sight) environment. sight, NLOS) environment, such as the signal-to-noise ratio SINR is lower than a certain threshold and so on.

Exemplarily, it is taken that the measurement information is the measurement information collected by the sensor of the first device as an example. The above-mentioned second preset condition may be: the measurement information obtained by the visual sensor indicates that the first device is in an LOS environment.

Further optionally, in the embodiment of the present application, the above-mentioned trigger conditions include: the statistical information of the first information corresponding to the above-mentioned first AI model satisfies a third preset condition; the above-mentioned meeting the third preset condition includes at least one of the following:

The above-mentioned first statistic is greater than the maximum value of the first threshold interval;

The above-mentioned second statistic is greater than the maximum value of the second threshold interval;

The correlation information of the first information collected in at least two time windows satisfies the first condition;

Correlation information between different first pieces of information collected within the current time window satisfies the second condition.

Exemplarily, it is assumed that the first statistic is an average value of channel information of the terminal within a certain continuous time window. The foregoing third preset condition may be: the first device detects that the statistics of channel information in a certain continuous time window exceed the maximum value of a certain threshold interval.

Exemplarily, the second statistic is an average value of average values of channel information of the terminal in multiple consecutive time windows as an example. The foregoing third preset condition may be: the first device detects that the average value of the average values of the channel information in multiple consecutive time windows exceeds the maximum value of a certain threshold interval.

Exemplarily, the first information is channel information as an example. The above-mentioned third preset condition may be: the correlation index of the channel information in the two time windows is lower than a certain threshold, or the correlation of the previous and subsequent data in the current time window is lower than a certain threshold.

Further optionally, in the embodiment of the present application, the trigger condition includes: the statistical information of the measurement information of the first device satisfies a fourth preset condition; the satisfaction of the fourth preset condition includes at least one of the following:

The above third statistic is greater than the maximum value of the third threshold interval;

The above fourth statistic is greater than the maximum value of the fourth threshold interval;

The correlation information of the measurement information collected in at least two time windows satisfies the third condition;

Correlation information between different measurement information collected in the current time window satisfies the fourth condition;

The difference between the distribution of the measurement information and the reference distribution is greater than a fifth threshold, and the reference distribution is information configured by the second device for the first device.

Exemplarily, in the case where the third statistic is the mean value of the channel measurement information within a certain continuous time window, the above-mentioned fourth preset condition It may be: the mean value of the channel measurement information in a certain continuous time window exceeds the maximum value of a certain threshold interval.

Exemplarily, in the case where the fourth statistic is an average value calculated based on the average value of the channel measurement information in each fourth time window, the above fourth preset condition may be: based on the average value of the channel measurement information in each fourth time window The mean calculated by the mean exceeds the maximum value of a certain threshold interval.

Exemplarily, the above third condition may be: the covariance of the data of the measurement information respectively collected in at least two time windows is smaller than a certain threshold.

Exemplarily, the above fourth condition may be: the distance between different data of the measurement information collected within the current time window is smaller than a certain threshold.

Exemplarily, the above-mentioned distribution of the measurement information is a statistical distribution of the measurement information. Exemplarily, the above reference distribution is the statistical distribution of the first AI model. It can be understood that the training set for offline training of the first AI model obeys the benchmark distribution, and the performance of the first AI model is best when the measurement information obeys the benchmark distribution.

Exemplarily, the index describing the difference between the distribution of the measurement information and the reference distribution may include at least one of the following: Wasserstein distance; Kullback-Leibler divergence; Hellinger distance and the like.

Optionally, in this embodiment of the present application, the trigger condition of the trigger mode includes at least one of the following:

The second device instructs the first device to perform online learning;

The output accuracy of the first AI model is less than or equal to the sixth threshold.

Further optionally, in this embodiment of the present application, the above-mentioned second device instructs the first device to perform online learning, including at least one of the following:

The second device instructs the first device to perform online learning periodically;

The second device instructs the first device to conduct online learning semi-periodically;

The second device instructs the first device to perform online learning aperiodically.

Wherein, the cycle adopted by the above-mentioned first device to perform online learning periodically or semi-periodically is: a cycle pre-configured by the above-mentioned second device, or a cycle independently configured by the above-mentioned first device.

Further optionally, the above-mentioned second device instructs the first device to perform online learning half-periodically, including:

The second device instructs the first device to perform online learning semi-periodically through the first signaling, and the first signaling includes at least one of the following: medium access control-control element MAC-CE, downlink control information DCI.

Optionally, in this embodiment of the application, the above suspension conditions include at least one of the following:

The number of online learning of the first AI model is greater than the preset number of iterations;

The first AI model reaches the preset accuracy;

The error information of the output result of the first AI model is smaller than the seventh threshold;

The second device abruptly instructs the first device to end the current online learning process;

the target task associated with the first AI model is aborted;

The difference information between the measurement information distribution of the first device and the reference distribution is smaller than the eighth threshold.

Exemplarily, the target task associated with the first AI model may be a task currently performed by the first AI model, such as locating a terminal, making network recommendations for a user, and so on.

Optionally, in this embodiment of the application, the above parameter configuration information includes at least one of the following:

The online learning mode of the first AI model;

the size of the sample batch of the first AI model;

the state of the optimizer of the first AI model;

A division method of the first data set of the first AI model;

Composition information of the first data set of the first AI model;

The contribution weight of the first data set of the first AI model to the update of the first AI model;

AI model identification associated with the first information;

a baseline distribution of the first AI model;

Wherein, the above-mentioned first data set is at least one of the following: the original data set used by the above-mentioned first AI model, the data set newly collected by the above-mentioned first AI model; the parameter information of the above-mentioned reference distribution includes at least one of the following: variance, mean , standard deviation; the size of the above sample batch refers to the size of the number of samples included in a sample batch (Batch).

Optionally, the above-mentioned original data set is: a data set used for offline training of the first AI model (that is, an old data set), and the above-mentioned newly collected data set is: deploying the first AI model online on the first device side, That is, after the first AI model is configured for the first device, the data set (ie, the new data set) collected by the first device in a new environment.

Optionally, the above-mentioned online learning mode includes any one of the following: an instantaneous training mode (ie, One-shot mode), and a continuous learning mode. Exemplarily, in the One-shot mode, the first device performs online learning when the collected data reaches a specified amount; In the continuous learning mode, the first device continuously performs online learning as the amount of collected data increases.

Exemplarily, the size of the above batch (Batch) is N, and N is a positive integer.

Exemplarily, the state of the optimizer of the above-mentioned first AI model may include a loss function, a learning rate, and the like.

Exemplarily, the first data set division method may include division ratios of the training set, verification set, and test set, and the like.

Exemplarily, the composition information of the above data sets includes the ratio between the number of original data sets and the number of newly collected data sets. It should be noted that the use of the original data set during training in the embodiment of the present application can effectively prevent over-fitting of newly collected data during the online learning process, thereby effectively improving the performance of the AI model.

Exemplarily, the contribution weight of the first data set to the update of the first AI model may be: the contribution weights of the old data set and the new data set to the update of the first AI model. For example, when the first AI model performs online learning, it may be Assign smaller weights to the original dataset and larger weights to the new dataset.

Further optionally, in the embodiment of the present application, the parameter configuration information includes an online learning mode of the first AI model, and the online learning mode is an instantaneous training mode, and the parameter configuration information further includes at least one of the following: the first device The amount of data collected, and the length of time for collecting the above data amount.

Further optionally, in the embodiment of the present application, the parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is a continuous learning mode, and the parameter configuration information further includes at least one of the following: The time interval between two online learning, the data volume interval between two adjacent online learning.

Exemplarily, the data volume interval between two adjacent online learning sessions may be 100, for example, online training is performed every time 100 sets of data are collected.

Fig. 3 shows a flowchart of an online learning method for an AI model provided by an embodiment of the present application. As shown in Figure 3, the online learning method of the AI model provided by the embodiment of the present application may include the following steps 301 and 302:

Step 301: the first device acquires a first AI model.

Step 302: The first device performs online learning on the first AI model based on the online learning information of the first AI model.

In this embodiment of the present application, the first device may perform online learning on the first AI model obtained through offline training based on the online learning information of the first AI model, and obtain the first AI model after parameter adjustment.

In this embodiment of the present application, the online learning information is information determined by the first device.

In the online learning method of the AI model provided in the embodiment of the present application, the first device acquires the first AI model, and performs online learning on the first AI model based on the online learning information of the first AI model. With this method, by deploying the first AI model on the first device side and configuring the first model with parameters required for online learning, the first AI model can be continuously adjusted online on the first device side, thereby maintaining The predictive performance of the first AI model, thereby ensuring the service quality of the first device.

How eLearning is triggered;

Conditions for discontinuation of online learning;

Parameter configuration information for online learning;

Datasets for online learning.

Optionally, in this embodiment of the application, the above step 301 may include the following step 301a:

Step 301a: the first device receives the first AI model configured by the second device.

Optionally, in the embodiment of the present application, the above-mentioned second device includes at least one of the following: core network equipment, access network equipment, and terminal; the above-mentioned first device includes at least one of the following: core network equipment, access network equipment equipment, and terminals.

Optionally, the second device may send the first AI model obtained through offline training to the first device, and the first device may receive the first AI model sent by the second device.

Optionally, in the embodiment of the present application, before the above step 302, the online learning method provided in the embodiment of the present application further includes the following step A1:

Step A1: the first device obtains the online learning information of the first AI model from the second device.

Optionally, in the case where the second device can send the information required for the online learning of the first AI model to the first device, the first device can receive the online learning information sent by the second device, and based on the online learning information, can The first AI model for online learning.

Optionally, the above-mentioned first device may be a terminal, and the second device may be a core network device.

In this way, the second device configures the first AI model for the first device, and configures the parameters required for online learning of the first AI model, so that the first AI model can be continuously adjusted online on the first device side, thereby maintaining The predictive performance of the first AI model, thereby ensuring the service quality of the first device.

Optionally, in the embodiment of the present application, the online learning method provided in the embodiment of the present application further includes the following step 303:

Step 303: The first device configures online learning information of the first AI model for the third device.

Optionally, the third device includes at least one of the following: a core network device, an access network device, and a terminal.

Optionally, the above-mentioned first device may be a core network device, and the third device may be a terminal.

Optionally, the above-mentioned second device can autonomously execute the online learning method of the AI model, or, the second device can deploy the AI model to the first device, and configure the information required for the online learning of the AI model for the first device, by The first device executes the online learning of the AI model, or the first device can deploy the AI model to the third device, and configure the information required for the online learning of the AI model for the first device, and the third device executes the online learning of the AI model. study.

The state information of the first device satisfies a first preset condition;

The measurement information of the first device satisfies a second preset condition;

The statistical information of the first information corresponding to the first AI model satisfies a third preset condition;

Statistical information of the measurement information of the first device satisfies a fourth preset condition;

Wherein, the above state information includes at least one of the following: moving speed, beam switching information, and cell switching information.

Optionally, the amount of data collected by the first device may be the amount of online data collected by the first device, for example, channel information collected by the first device in real time.

Exemplarily, the above channel information may include at least one of the following: signal emission angle information, time delay information of signals in the channel, signal quality in the channel, and so on.

Exemplarily, the first device may obtain status information in real time or periodically, and perform online learning on the first AI model when the status information satisfies a first preset condition.

For example, during the calculation process of the first AI model, the first device may obtain the error information of the output result of the first AI model in real time or periodically, or the prediction accuracy of the first AI model, and when the error information is greater than the second In the case of the threshold value, online learning is performed on the first AI model.

Exemplarily, the first device may collect statistics on input information and output information of the first AI model, and perform online learning on the first AI model when the statistical information of the input information or output information satisfies a third preset condition.

Exemplarily, the first device may detect the reference signal or the measurement information collected by the sensor, and when the measurement information satisfies the second preset condition, and/or the statistical information of the measurement information satisfies the fourth preset condition, the first device An AI model for online learning.

Optionally, in this embodiment of the present application, the statistical information of the above-mentioned first information includes at least one of the following: the first statistical quantity of the first information in the first time window, the first statistical quantity of the first information in at least two consecutive second time windows The second statistical quantity corresponding to the information is the statistical information of the first information of at least two terminals under the first cell at the first moment, and the correlation information of the first information.

The statistical information of the above-mentioned first information is explained below by using the first information as channel information.

Exemplarily, when the statistical information of the first information includes the second statistical quantity corresponding to the above-mentioned first information in at least two consecutive second time windows, the above-mentioned statistical information of the first information may be: The mean value of the mean value of channel information in continuous time windows, for example, the mean value of channel information in time window 1 is a, the mean value of channel information in time window 2 is b, and the mean value of channel information in time window 3 is c, then the channel information The statistic for is the mean of a, b, and c.

Exemplarily, when the statistical information of the first information includes the correlation information of the first information, the statistical information of the first information may be: the correlation index of the channel information of the two time windows is lower than a certain threshold, For example, the distance between the previous and subsequent data in the current time window is smaller than a certain threshold.

Optionally, in this embodiment of the present application, the statistical information of the measurement information of the first device includes at least one of the following: a third statistic for the measurement information within a third time window, and the measurement information is in at least two consecutive The fourth statistic corresponding to the fourth time window is the correlation information of the above measurement information.

Exemplarily, the measurement information is channel measurement information of a reference signal received by the first device as an example. The statistical information of the above measurement information may be: the variance of the channel measurement information in a certain continuous time window, or the average of the mean values of the channel measurement information in two consecutive time windows, or the channel measurement information in the current time window distance between the data.

Further optionally, in the embodiment of the present application, the trigger condition includes: the state information of the first device meets a first preset condition; the satisfaction of the first preset condition includes at least one of the following:

The moving speed of the first device is greater than a third threshold;

Further optionally, in the embodiment of the present application, the trigger condition includes: the measurement information of the first device meets a second preset condition; the meeting the second preset condition includes: the measurement information of the first device indicates that the first device The channel environment of the relevant channel changes.

Wherein, the above-mentioned measurement information includes at least one of the following: first measurement information of the reference signal received by the first device, and second measurement information collected by the sensor of the first device; optionally, the above-mentioned first measurement information includes at least one of the following : Instantaneous measurement information of the reference signal, statistical measurement information of the reference signal;

Wherein, the above-mentioned reference signal includes at least one of the following: a synchronization signal block SSB, a CSI reference signal CSI-RS, a sounding reference signal SRS, and a positioning reference signal PRS.

Further optionally, in this embodiment of the present application, the above-mentioned trigger conditions include: statistics of the second information corresponding to the above-mentioned first AI model The information satisfies the third preset condition; the above-mentioned meeting of the third preset condition includes at least one of the following:

Correlation information among different measurement information collected in the current time window satisfies the fourth condition;

The difference between the distribution of the measurement information and the reference distribution is greater than the fifth threshold, and the above reference distribution is information configured by the second device for the first device.

Exemplarily, in the case where the third statistic is the mean value of the channel measurement information in a certain continuous time window, the above fourth preset condition may be: the mean value of the channel measurement information in a certain continuous time window exceeds a certain The maximum value of the threshold interval.

Optionally, in this embodiment of the present application, the trigger condition of the above trigger method includes at least one of the following:

The second device instructs the first device to perform online learning;

Wherein, the cycle adopted by the above-mentioned first device to perform online learning periodically or semi-periodically is: a cycle preconfigured by the second device, or a cycle independently configured by the first device.

Further optionally, the second device instructs the first device to perform online learning semi-periodically, including:

The second device instructs the first device to perform online learning semi-periodically through the first signaling, and the first signaling includes at least one of the following: medium access control-control element MAC-CE, and downlink control information DCI.

The first AI model reaches the preset accuracy;

the target task associated with the first AI model is aborted;

Exemplarily, during the process of online learning of the first AI model, the first device may detect in real time or periodically whether the suspension condition is met, and if the suspension condition is satisfied, suspend the online learning of the first AI model.

Exemplarily, the aforementioned target task associated with the first AI model may be a task currently performed by the first AI model, such as locating a terminal, making network recommendations for a user, and so on.

In an example, the first device may stop online learning of the first AI model when it detects that the number of iterations of the first AI model is greater than a preset number of iterations (eg, 10,000). In another example, the second device may stop online learning of the first AI model when detecting that the accuracy value of the first AI model reaches a preset accuracy. In yet another example, the second device may stop the online learning of the first AI model when it detects that the error of the output result of the first AI model is smaller than a preset error value. In this way, the first device can stop the online learning of the AI model based on the number of times of online learning of the above-mentioned AI model, the achieved accuracy, the error of the output result, and the difference from the reference distribution, so as to improve the prediction accuracy of the model. case, saving power consumption.

It should be noted that suspending the online learning of the first AI model may be temporarily stopping the online learning of the first AI model, or ending the online learning of the first AI model.

The online learning mode of the first AI model;

the size of the sample batch of the first AI model;

the state of the optimizer of the first AI model;

A division method of the first data set of the first AI model;

Composition information of the first data set of the first AI model;

The contribution weight of the first data set of the first AI model to the update of the AI model;

AI model identification associated with online learning information;

The baseline distribution of the first AI model;

Wherein, the above-mentioned first data set is at least one of the following: the original data set used by the first AI model, and the newly collected data set by the first AI model. The above parameter information of the benchmark distribution includes at least one of the following: variance, mean, and standard deviation.

Optionally, the above-mentioned original data set is: a data set used for offline training of the first AI model (that is, an old data set), and the above-mentioned newly collected data set is: deployed online on the first device side, that is, the first AI model. After the device configures the first AI model, the data set (ie, the new data set) collected by the first device in a new environment.

Optionally, the above-mentioned online learning mode includes any one of the following: an instantaneous training mode (ie, One-shot mode), and a continuous learning mode. Exemplarily, in the One-shot mode, the first device performs online learning when the collected data reaches a specified amount; in the continuous learning mode, the first device continuously performs online learning as the amount of collected data increases .

Exemplarily, the first device may obtain parameter configuration information of the first AI model, and perform online learning on the first AI model based on the parameter configuration information, so as to ensure the prediction accuracy of the AI model in a changing environment.

Further optionally, in the embodiment of the present application, the parameter configuration information includes an online learning mode of the first AI model, and the online learning mode is an instantaneous training mode, and the parameter configuration information further includes at least one of the following: The amount of data collected and the length of time for collecting data.

Further optionally, in the embodiment of the present application, the above-mentioned parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is a continuous learning mode, and the above-mentioned parameter configuration information also includes at least one of the following: The time interval of online learning, the data volume interval between two adjacent online learning.

The online learning method of the AI model provided in the embodiment of the present application may be executed by an online learning device of the AI model. In the embodiment of this application, the online method of executing the AI model by the online device of the AI model is taken as an example to illustrate the online method of the AI model provided by the embodiment of the application. learning device.

An embodiment of the present application provides an online AI model learning device 400. As shown in FIG. A device deploys a first AI model; the configuration 401 is further used for the second device to configure online learning information of the first AI model for the first device.

How eLearning is triggered;

Conditions for discontinuation of online learning;

Parameter configuration information for online learning;

Datasets for online learning.

Optionally, in the embodiment of this application,

The trigger condition corresponding to the trigger mode includes at least one of the following:

The state information of the first device satisfies a first preset condition;

The error information of the output result of the first AI model is greater than a second threshold;

Wherein, the state information includes at least one of the following: moving speed, beam switching information, cell switching information;

The first information includes at least one of the following: input information of the first AI model, and output information of the first AI model.

Optionally, in this embodiment of the present application, the statistical information of the first information includes at least one of the following: a first statistical quantity of the first information within a first time window, at least two consecutive second time windows A second statistic corresponding to the first information, statistical information of the first information of at least two terminals under the first cell at the first moment, and correlation information of the first information;

The second statistics are calculated based on statistics in each of the second time windows;

The statistical information of the measurement information of the first device includes at least one of the following: a third statistic for the above measurement information within a third time window, where the measurement information corresponds to at least two consecutive fourth time windows. Four statistics, correlation information of the measurement information;

The fourth statistic is calculated based on the statistic in each of the fourth time windows;

The correlation information includes at least one of the following: distance between data, covariance, and correlation coefficient.

Optionally, in this embodiment of the present application, the trigger condition includes: the status information of the first device meets a first preset condition;

Said meeting the first preset condition includes at least one of the following:

The moving speed of the first device is greater than a third threshold;

The beam switching information indicates that beam switching occurs for the first device, and the beam switching frequency is greater than a fourth threshold;

The cell switching information indicates that a cell switching occurs to the first device.

Optionally, in this embodiment of the present application, the trigger condition includes: the measurement information of the reference signal received by the first device satisfies a second preset condition;

The meeting the second preset condition includes: the measurement information of the reference signal indicates that the channel environment of the relevant channel of the first device changes;

Wherein, the measurement information includes at least one of the following: first measurement information of a reference signal received by the first device, and second measurement information collected by a sensor of the first device;

The first measurement information includes at least one of the following: instantaneous measurement information of the reference signal, statistical measurement information of the reference signal;

The reference signal includes at least one of the following: a synchronization signal block SSB, a CSI reference signal CSI-RS, a sounding reference signal SRS, and a positioning reference signal PRS.

Optionally, in the embodiment of the present application, the trigger condition includes: the statistical information of the first information corresponding to the first AI model satisfies a third preset condition;

Said meeting the third preset condition includes at least one of the following:

The first statistic is greater than the maximum value of the first threshold interval;

The second statistic is greater than the maximum value of the second threshold interval;

The correlation information of the first information collected in at least two time windows satisfies a first condition;

Optionally, in this embodiment of the present application, the trigger condition includes: statistical information of the measurement information of the first device meets a fourth preset condition;

Said meeting the fourth preset condition includes at least one of the following:

The third statistic is greater than the maximum value of the third threshold interval;

The fourth statistic is greater than the maximum value of the fourth threshold interval;

A difference between the distribution of the measurement information and a reference distribution is greater than a fifth threshold, where the reference distribution is information configured by the second device for the first device.

Optionally, in the embodiment of this application,

The trigger condition of the trigger mode includes at least one of the following:

The second device instructs the first device to perform online learning;

Optionally, in this embodiment of the present application, the second device instructs the first device to perform online learning, including at least one of the following:

The second device instructs the first device to periodically perform online learning;

The second device instructs the first device to perform online learning semi-periodically;

Wherein, the period adopted by the first device to perform online learning periodically or semi-periodically is: a period preconfigured by the second device, or a period independently configured by the first device.

Optionally, in this embodiment of the present application, the second device instructs the first device to perform online learning semi-periodically, including:

The second device instructs the first device to perform online learning semi-periodically through a first signaling, and the first signaling includes at least one of the following: medium access control-control element MAC-CE, downlink control information DCI .

Optionally, in this embodiment of the application, the termination condition includes at least one of the following:

The first AI model reaches a preset accuracy;

The error information of the output result of the first AI model is less than the seventh threshold;

a target task associated with the first AI model is aborted;

The difference information between the measurement information distribution of the first device and the reference distribution is smaller than an eighth threshold.

Optionally, in this embodiment of the present application, the parameter configuration information includes at least one of the following:

The online learning mode of the AI model;

the size of the sample batch of the AI model;

the state of the optimizer of the AI model;

The division method of the first data set of the AI model;

Composition information of the first data set of the AI model;

The contribution weight of the first data set of the AI model to the update of the AI model;

An AI model identifier associated with the first information;

a baseline distribution of the first AI model;

Wherein, the first data set is at least one of the following: the original data set used by the first AI model, the data set newly collected by the first AI model;

The parameter information of the reference distribution includes at least one of the following: variance, mean, and standard deviation.

Optionally, in this embodiment of the present application, the parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is an instantaneous training mode, and the parameter configuration information further includes at least one of the following : the amount of data collected by the first device, and the length of time for collecting the data.

Optionally, in this embodiment of the present application, the parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is a continuous learning mode, and the parameter configuration information further includes at least one of the following : the time interval between two adjacent online learning, and the data volume interval between two adjacent online learning.

Optionally, in this embodiment of the present application, the second device includes at least one of the following: core network equipment, access network equipment, and terminal; the first device includes at least one of the following: core network equipment, access Network equipment, terminals.

In the online AI model learning apparatus provided in the embodiment of the present application, the second device configures the first AI model for the first device, and configures the online learning information of the first AI model for the first device. With this method, by deploying the first AI model on the first device side and configuring the first model with parameters required for online learning, the first AI model can be continuously adjusted online on the first device side, thereby maintaining The predictive performance of the first AI model, thereby ensuring the service quality of the first device.

An embodiment of the present application provides an online learning device 500 for an AI model. As shown in FIG. The first device acquires a first AI model; the execution module 502 is configured to enable the first device to learn online the first AI model based on the online learning information of the first AI model.

Optionally, in this embodiment of the application, the first device acquires the first AI model, including:

The first device receives the first AI model configured by the second device.

Optionally, in the embodiment of this application, the acquisition module is specifically used to

Obtain online learning information of the first AI model from the second device.

Optionally, in the embodiment of the present application, the apparatus further includes: a configuration module, configured to configure the online learning information of the first AI model for the third device.

How eLearning is triggered;

Conditions for discontinuation of online learning;

Parameter configuration information for online learning;

Datasets for online learning.

Optionally, in this embodiment of the present application, the trigger condition corresponding to the trigger mode includes at least one of the following:

The state information of the first device satisfies a first preset condition;

Said meeting the first preset condition includes at least one of the following:

The moving speed of the first device is greater than a third threshold;

Optionally, in this embodiment of the present application, the trigger condition includes: the statistical information of the second information corresponding to the first AI model satisfies a third preset condition;

Said meeting the third preset condition includes at least one of the following:

The second device instructs the first device to perform online learning;

The first AI model reaches a preset accuracy;

a target task associated with the first AI model is aborted;

The online learning mode of the AI model;

The size of the sample batch of the AI model;

the state of the optimizer of the AI model;

The division method of the first data set of the AI model;

Composition information of the first data set of the AI model;

An AI model identifier associated with the first information;

a baseline distribution of the first AI model;

Optionally, in this embodiment of the present application, the second device, the first device, and the third device include at least one of the following: a core network device, an access network device, and a terminal.

In the apparatus for online learning of an AI model provided in the embodiment of the present application, the first device acquires a first AI model, and performs online learning on the first AI model based on online learning information of the first AI model. With this method, by deploying the first AI model on the first device side and configuring the first model with parameters required for online learning, the first AI model can be continuously adjusted online on the first device side, thereby maintaining The predictive performance of the first AI model, thereby ensuring the service quality of the first device.

The online learning apparatus for the AI model in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or a component in the electronic device, such as an integrated circuit or a chip. The electronic device may be a terminal, or other devices other than the terminal. Exemplarily, the terminal may include, but not limited to, the types of terminal 11 listed above, and other devices may be servers, Network Attached Storage (NAS), etc., which are not specifically limited in this embodiment of the present application.

The AI model online learning device provided by the embodiment of the present application can realize the various processes realized by the method embodiments in Fig. 1 to Fig. 3 and achieve the same technical effect. To avoid repetition, details are not repeated here.

Optionally, as shown in FIG. 6 , this embodiment of the present application also provides a communication device 600, including a processor 601 and a memory 602, and the memory 602 stores programs or instructions that can run on the processor 601, such as When the communication device 600 is a terminal, when the program or instruction is executed by the processor 601, each step of the above embodiment of the online learning method of the AI model can be realized, and the same technical effect can be achieved. When the communication device 600 is a network-side device, when the program or instruction is executed by the processor 601, the steps of the above-mentioned online learning method for the AI model are implemented, and the same technical effect can be achieved. To avoid repetition, details are not repeated here. .

Take the first device as a terminal as an example.

The embodiment of the present application further provides a terminal, including a processor and a communication interface, and the processor is configured to acquire a first AI model, and perform online learning on the first AI model based on online learning information of the first AI model. This terminal embodiment corresponds to the above-mentioned terminal-side method embodiment, and each implementation process and implementation mode of the above-mentioned method embodiment can be applied to this terminal embodiment, and can achieve the same technical effect. Specifically, FIG. 7 is a schematic diagram of a hardware structure of a terminal implementing an embodiment of the present application.

The terminal 700 includes, but is not limited to: a radio frequency unit 701, a network module 702, an audio output unit 703, an input unit 704, a sensor 705, a display unit 706, a user input unit 707, an interface unit 708, a memory 709, and a processor 710. At least some parts.

Those skilled in the art can understand that the terminal 700 may also include a power supply (such as a battery) for supplying power to various components, and the power supply may be logically connected to the processor 710 through the power management system, so as to manage charging, discharging, and power consumption through the power management system. Management and other functions. The terminal structure shown in FIG. 7 does not constitute a limitation on the terminal, and the terminal may include more or fewer components than shown in the figure, or combine some components, or arrange different components, which will not be repeated here.

It should be understood that, in this embodiment of the present application, the input unit 704 may include a graphics processing unit (Graphics Processing Unit, GPU) 7041 and a microphone 7042, and the graphics processor 7041 is used by the image capture device ( Such as the image data of the still picture or video obtained by the camera) for processing. The display unit 706 may include a display panel 7061, and the display panel 7061 may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 707 includes at least one of a touch panel 7071 and other input devices 7072 . The touch panel 7071 is also called a touch screen. The touch panel 7071 may include two parts, a touch detection device and a touch controller. Other input devices 7072 may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, switch buttons, etc.), trackballs, mice, and joysticks, which will not be described in detail here.

In the embodiment of the present application, the radio frequency unit 701 may transmit the downlink data from the network side device to the processor 710 for processing after receiving the downlink data; in addition, the radio frequency unit 701 may send uplink data to the network side device. Generally, the radio frequency unit 701 includes, but is not limited to, an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, and the like.

The memory 709 can be used to store software programs or instructions as well as various data. The memory 709 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 by at least one function (such as a sound playing function, image playback function, etc.), etc. Furthermore, memory 709 may include volatile memory or nonvolatile memory, or, memory 709 may include both volatile and nonvolatile memory. Among them, the 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), electronically programmable Erase Programmable Read-Only Memory (Electrically EPROM, EEPROM) or Flash. 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 connection dynamic random access memory (Synch link DRAM , SLDRAM) and Direct Memory Bus Random Access Memory (Direct Rambus RAM, DRRAM). The memory 709 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

The processor 710 may include one or more processing units; optionally, the processor 710 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to the operating system, user interface, and application programs, etc., The modem processor mainly handles the line communication signals, such as baseband processors. It can be understood that the foregoing modem processor may not be integrated into the processor 710 .

Wherein, the processor 710 is used for the first device to acquire the first AI model, and the processor 710 is also used for the first device to perform the first AI model based on the online learning information of the first AI model. An AI model for online learning.

Optionally, in this embodiment of the present application, the radio frequency unit 701 is configured to receive the first AI model configured by the second device.

Optionally, in this embodiment of the present application, the processor 710 is specifically configured to acquire online learning information of the first AI model from the second device.

Optionally, in this embodiment of the present application, the processor 710 is further configured to configure online learning information of the first AI model for the third device.

How eLearning is triggered;

Conditions for discontinuation of online learning;

Parameter configuration information for online learning;

Datasets for online learning.

The state information of the first device satisfies a first preset condition;

Said meeting the first preset condition includes at least one of the following:

The moving speed of the first device is greater than a third threshold;

Said meeting the third preset condition includes at least one of the following:

The second device instructs the first device to perform online learning;

The first AI model reaches a preset accuracy;

a target task associated with the first AI model is aborted;

The online learning mode of the AI model;

The size of the sample batch of the AI model;

the state of the optimizer of the AI model;

The division method of the first data set of the AI model;

Composition information of the first data set of the AI model;

An AI model identifier associated with the first information;

a baseline distribution of the first AI model;

In the terminal provided in the embodiment of the present application, the terminal acquires a first AI model, and performs online learning on the first AI model based on online learning information of the first AI model. With this method, by deploying the first AI model on the terminal side, and configuring the first model on The parameters required for online learning enable continuous online adjustment of the first AI model on the terminal side, thereby maintaining the predictive performance of the first AI model and ensuring the service quality of the terminal.

Take the second device being the network side device as an example.

The embodiment of the present application also provides a network side device, including a processor and a communication interface, the processor is configured to configure a first AI model for a first device; and configure online learning information of the first AI model for the first device. The network-side device embodiment corresponds to the above-mentioned network-side device method embodiment, and each implementation process and implementation mode of the above-mentioned method embodiment can be applied to this network-side device embodiment, and can achieve the same technical effect.

Specifically, the embodiment of the present application also provides a network side device. As shown in FIG. 8 , the network side device 800 includes: an antenna 81 , a radio frequency device 82 , a baseband device 83 , a processor 84 and a memory 85 . The antenna 81 is connected to a radio frequency device 82 . In the uplink direction, the radio frequency device 82 receives information through the antenna 81, and sends the received information to the baseband device 83 for processing. In the downlink direction, the baseband device 83 processes the information to be sent and sends it to the radio frequency device 82 , and the radio frequency device 82 processes the received information and sends it out through the antenna 81 .

The method performed by the network side device in the above embodiments may be implemented in the baseband device 83, where the baseband device 83 includes a baseband processor.

The baseband device 83 can include at least one baseband board, for example, a plurality of chips are arranged on the baseband board, as shown in FIG. The program executes the network device operations shown in the above method embodiments.

The network side device may also include a network interface 86, such as a common public radio interface (common public radio interface, CPRI).

Specifically, the network side device 800 in this embodiment of the present invention further includes: instructions or programs stored in the memory 85 and operable on the processor 84, and the processor 84 invokes the instructions or programs in the memory 85 to execute the various programs shown in FIG. The method of module execution achieves the same technical effect, so in order to avoid repetition, it is not repeated here.

Specifically, the embodiment of the present application also provides a network side device. As shown in FIG. 9 , the network side device 900 includes: a processor 901 , a network interface 902 and a memory 903 . Wherein, the network interface 902 is, for example, a common public radio interface (common public radio interface, CPRI).

Specifically, the network-side device 900 in this embodiment of the present invention also includes: instructions or programs stored in the memory 903 and executable on the processor 901, and the processor 901 invokes the instructions or programs in the memory 903 to execute the various programs shown in FIG. The method of module execution achieves the same technical effect, so in order to avoid repetition, it is not repeated here.

The embodiment of the present application also provides a readable storage medium, the readable storage medium stores a program or an instruction, and when the program or instruction is executed by the processor, each process of the above-mentioned online learning method embodiment of the AI model is realized, and The same technical effect can be achieved, so in order to avoid repetition, details will not be repeated here.

Wherein, the processor is the processor in the terminal described in the foregoing embodiments. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk, and the like.

The 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, and the processor is used to run programs or instructions to realize the online learning method of the AI model The various processes of the embodiment can achieve the same technical effect, so in order to avoid repetition, details are not repeated here.

It should be understood that the chip mentioned in the embodiment of the present application may also be called a system-on-chip, a system-on-chip, a system-on-a-chip, or a system-on-a-chip.

The embodiment of the present application further provides a computer program/program product, 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 realize the above-mentioned online learning of the AI model Each process of the method embodiment can achieve the same technical effect, and will not be repeated here to avoid repetition.

The embodiment of the present application also provides a communication system, including: a terminal and a network-side device, the terminal can be used to execute the steps of the online learning method of the AI model as described above, and the network-side device can be used to execute the above-mentioned The steps of the online learning method of the AI model.

It should be noted that, in this document, the term "comprising", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions are performed, for example, the described methods may be performed in an order different from that described, and various steps may also be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on this In this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of computer software products, and the computer software products are stored in a storage medium (such as ROM/RAM, disk, CD) contains several instructions to enable a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in various embodiments of the present application.

The embodiments of the present application have been described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementations. The above-mentioned specific implementations are only illustrative and not restrictive. Those of ordinary skill in the art will Under the inspiration of this application, without departing from the purpose of this application and the scope of protection of the claims, many forms can also be made, all of which belong to the protection of this application. .

Claims

An online learning method of an artificial intelligence AI model, said method comprising:

The first device obtains the first AI model;

The first device performs online learning on the first AI model based on the online learning information of the first AI model.
The method according to claim 1, wherein the acquiring the first AI model by the first device comprises:

The first device receives the first AI model configured by the second device.
The method according to claim 1 or 2, wherein, before the first device performs online learning on the first AI model based on the online learning information of the first AI model, the method further includes:

The first device acquires online learning information of the first AI model from the second device.
The method according to claim 1, wherein the method further comprises:

The first device configures the online learning information of the first AI model for the third device.
The method according to claim 1, wherein the online learning information includes at least one of the following:

How eLearning is triggered;

Conditions for discontinuation of online learning;

Parameter configuration information for online learning;

Datasets for online learning.
The method according to claim 5, wherein,

The trigger condition corresponding to the trigger mode includes at least one of the following:

The state information of the first device satisfies a first preset condition;

The amount of data collected by the first device is greater than a first threshold;

The measurement information of the first device satisfies a second preset condition;

The error information of the output result of the first AI model is greater than a second threshold;

The statistical information of the first information corresponding to the first AI model satisfies a third preset condition;

Statistical information of the measurement information of the first device satisfies a fourth preset condition;

Wherein, the state information includes at least one of the following: moving speed, beam switching information, cell switching information;

The first information includes at least one of the following: input information of the first AI model, and output information of the first AI model.
The method of claim 6, wherein,

The statistical information of the first information includes at least one of the following: a first statistic of the first information in a first time window, a second statistic corresponding to the first information in at least two consecutive second time windows Quantity, statistical information of the first information of at least two terminals under the first cell at the first moment, and correlation information of the first information;

The second statistics are calculated based on statistics in each of the second time windows.
The method according to claim 7, wherein the third preset condition includes at least one of the following:

The first statistic is greater than the maximum value of the first threshold interval;

The second statistic is greater than the maximum value of the second threshold interval;

The correlation information of the first information collected in at least two time windows satisfies a first condition;

Correlation information between different first pieces of information collected within the current time window satisfies the second condition.
The method of claim 6, wherein,

The statistical information of the measurement information of the first device includes at least one of the following: a third statistic for the measurement information within a third time window, where the measurement information corresponds to at least two consecutive fourth time windows The fourth statistic, the correlation information of the measurement information;

The fourth statistic is calculated based on the statistic in each of the fourth time windows;

The correlation information includes at least one of the following: distance between data, covariance, and correlation coefficient.
The method according to claim 9, wherein said meeting the fourth preset condition includes at least one of the following:

The third statistic is greater than the maximum value of the third threshold interval;

The fourth statistic is greater than the maximum value of the fourth threshold interval;

The correlation information of the measurement information collected in at least two time windows satisfies the third condition;

Correlation information among different measurement information collected in the current time window satisfies the fourth condition;

A difference between the distribution of the measurement information and a reference distribution is greater than a fifth threshold, where the reference distribution is information configured by the second device for the first device.
The method according to claim 6, wherein the trigger condition comprises: the status information of the first device meets a first preset condition;

Said meeting the first preset condition includes at least one of the following:

The moving speed of the first device is greater than a third threshold;

The beam switching information indicates that beam switching occurs for the first device, and the beam switching frequency is greater than a fourth threshold;

The cell switching information indicates that a cell switching occurs to the first device.
The method according to claim 6, wherein the trigger condition comprises: the measurement information of the first device meets a second preset condition;

The meeting the second preset condition includes: the measurement information indicates that the channel environment of the relevant channel of the first device changes;

Wherein, the measurement information includes at least one of the following: first measurement information of a reference signal received by the first device, and second measurement information collected by a sensor of the first device;

The first measurement information includes at least one of the following: instantaneous measurement information of the reference signal, statistical measurement information of the reference signal;

The reference signal includes at least one of the following: a synchronization signal block SSB, a CSI reference signal CSI-RS, a sounding reference signal SRS, and a positioning reference signal PRS.
The method according to claim 5, wherein,

The trigger condition of the trigger mode includes at least one of the following:

The second device instructs the first device to perform online learning;

The output accuracy of the first AI model is less than or equal to the sixth threshold.
The method of claim 13, wherein,

The second device instructs the first device to perform online learning, including at least one of the following:

The second device instructs the first device to periodically perform online learning;

The second device instructs the first device to perform online learning semi-periodically;

The second device instructs the first device to perform online learning aperiodically;

Wherein, the period adopted by the first device to perform online learning periodically or semi-periodically is: a period preconfigured by the second device, or a period independently configured by the first device.
The method according to claim 5, wherein the termination condition includes at least one of the following:

The number of online learning of the first AI model is greater than the preset number of iterations;

The first AI model reaches a preset accuracy;

The error information of the output result of the first AI model is less than the seventh threshold;

The second device abruptly instructs the first device to end the current online learning process;

a target task associated with the first AI model is aborted;

The difference information between the measurement information distribution of the first device and the reference distribution is smaller than an eighth threshold.
The method according to claim 5, wherein the parameter configuration information includes at least one of the following:

The online learning mode of the first AI model;

the size of the sample batch of the first AI model;

the state of the optimizer of the first AI model;

A division method of the first data set of the first AI model;

Composition information of the first data set of the first AI model;

The contribution weight of the first data set of the first AI model to the update of the first AI model;

AI model identification associated with the online learning information;

a baseline distribution of the first AI model;

Wherein, the first data set is at least one of the following: the original data set used by the first AI model, the data set newly collected by the first AI model;

The parameter information of the reference distribution includes at least one of the following: variance, mean, and standard deviation.
The method according to claim 16, wherein the parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is an instantaneous training mode, and the parameter configuration information further includes at least one of the following : the amount of data collected by the first device, and the length of time for collecting the data.
The method according to claim 16, wherein the parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is a continuous learning mode, and the parameter configuration information further includes at least one of the following : adjacent two The time interval between online learning and the data volume interval between two adjacent online learning.
The method according to claim 1, wherein the second device, the first device and the third device comprise at least one of the following: a core network device, an access network device, and a terminal.
An online learning method of an artificial intelligence AI model, said method comprising:

The second device configures the first AI model for the first device;

The second device configures online learning information of the first AI model for the first device.
The method according to claim 20, wherein the online learning information includes at least one of the following:

How eLearning is triggered;

Conditions for discontinuation of online learning;

Parameter configuration information for online learning;

Datasets for online learning.
The method of claim 21, wherein,

The trigger condition corresponding to the trigger mode includes at least one of the following:

The state information of the first device satisfies a first preset condition;

The amount of data collected by the first device is greater than a first threshold;

The measurement information of the first device satisfies a second preset condition;

The error information of the output result of the first AI model is greater than a second threshold;

The statistical information of the first information corresponding to the first AI model satisfies a third preset condition;

Statistical information of the measurement information of the first device satisfies a fourth preset condition;

Wherein, the state information includes at least one of the following: moving speed, beam switching information, cell switching information;

The first information includes at least one of the following: input information of the first AI model, and output information of the first AI model.
The method of claim 22, wherein,

The statistical information of the first information includes at least one of the following: a first statistic of the first information in a first time window, a second statistic corresponding to the first information in at least two consecutive second time windows Quantity, statistical information of the first information of at least two terminals under the first cell at the first moment, and correlation information of the first information;

The second statistics are calculated based on statistics in each of the second time windows.
The method according to claim 23, wherein the third preset condition includes at least one of the following:

The first statistic is greater than the maximum value of the first threshold interval;

The second statistic is greater than the maximum value of the second threshold interval;

The correlation information of the first information collected in at least two time windows satisfies a first condition;

Correlation information between different first pieces of information collected within the current time window satisfies the second condition.
The method of claim 22, wherein,

The statistical information of the measurement information of the first device includes at least one of the following: a third statistic for the measurement information within a third time window, where the measurement information corresponds to at least two consecutive fourth time windows The fourth statistic, the correlation information of the measurement information;

The fourth statistic is calculated based on the statistic in each of the fourth time windows;

The correlation information includes at least one of the following: distance between data, covariance, and correlation coefficient.
The method according to claim 25, wherein said meeting the fourth preset condition includes at least one of the following:

The third statistic is greater than the maximum value of the third threshold interval;

The fourth statistic is greater than the maximum value of the fourth threshold interval;

The correlation information of the measurement information collected in at least two time windows satisfies the third condition;

Correlation information among different measurement information collected in the current time window satisfies the fourth condition;

A difference between the distribution of the measurement information and a reference distribution is greater than a fifth threshold, where the reference distribution is information configured by the second device for the first device.
The method according to claim 22, wherein the trigger condition comprises: the status information of the first device meets a first preset condition;

Said meeting the first preset condition includes at least one of the following:

The moving speed of the first device is greater than a third threshold;

The beam switching information indicates that beam switching occurs for the first device, and the beam switching frequency is greater than a fourth threshold;

The cell switching information indicates that a cell switching occurs to the first device.
The method according to claim 22, wherein the trigger condition comprises: the measurement information of the first device meets a second preset condition;

The meeting the second preset condition includes: the measurement information indicates that the channel environment of the relevant channel of the first device changes;

Wherein, the measurement information includes at least one of the following: first measurement information of a reference signal received by the first device, and second measurement information collected by a sensor of the first device;

The first measurement information includes at least one of the following: instantaneous measurement information of the reference signal, statistical measurement information of the reference signal;

The reference signal includes at least one of the following: a synchronization signal block SSB, a CSI reference signal CSI-RS, a sounding reference signal SRS, and a positioning reference signal PRS.
The method of claim 21, wherein,

The trigger condition of the trigger mode includes at least one of the following:

The second device instructs the first device to perform online learning;

The output accuracy of the first AI model is less than or equal to the sixth threshold.
The method of claim 29, wherein,

The second device instructs the first device to perform online learning, including at least one of the following:

The second device instructs the first device to periodically perform online learning;

The second device instructs the first device to perform online learning semi-periodically;

The second device instructs the first device to perform online learning aperiodically;

Wherein, the period adopted by the first device to perform online learning periodically or semi-periodically is: a period preconfigured by the second device, or a period independently configured by the first device.
The method according to claim 21, wherein the termination condition comprises at least one of the following:

The number of online learning of the first AI model is greater than the preset number of iterations;

The first AI model reaches a preset accuracy;

The error information of the output result of the first AI model is less than the seventh threshold;

The second device abruptly instructs the first device to end the current online learning process;

a target task associated with the first AI model is aborted;

The difference information between the measurement information distribution of the first device and the reference distribution is smaller than an eighth threshold.
The method according to claim 21, wherein the parameter configuration information includes at least one of the following:

The online learning mode of the first AI model;

the size of the sample batch of the first AI model;

the state of the optimizer of the first AI model;

A division method of the first data set of the first AI model;

Composition information of the first data set of the first AI model;

The contribution weight of the first data set of the first AI model to the update of the AI model;

AI model identification associated with the online learning information;

a baseline distribution of the first AI model;

Wherein, the first data set is at least one of the following: the original data set used by the first AI model, the data set newly collected by the first AI model;

The parameter information of the reference distribution includes at least one of the following: variance, mean, and standard deviation.
The method according to claim 32, wherein the parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is an instantaneous training mode, and the parameter configuration information further includes at least one of the following : the amount of data collected by the first device, and the length of time for collecting the data.
The method according to claim 33, wherein the parameter configuration information includes the online learning mode of the first AI model, and the online learning mode is a continuous learning mode, and the parameter configuration information further includes at least one of the following : The time interval between two adjacent online learning, and the data volume interval between two adjacent online learning.
The method according to claim 20, wherein the second device includes at least one of the following: core network equipment, access network equipment, and terminal; the first device includes at least one of the following: core network equipment, access Network equipment, terminals.
An online learning device of an AI model, the device comprising: a configuration module, wherein:

The configuration module is used for the second device to configure the first AI model for the first device;

The configuration module is further used for the second device to configure online learning information of the first AI model for the first device.
An online learning device of an artificial intelligence AI model, said device comprising: an acquisition module and an execution module, wherein:

The obtaining module is used for the first device to obtain a first AI model;

The execution module is used for the first device to perform online learning on the first AI model based on the online learning information of the first AI model.
A communication device, comprising a processor and a memory, the memory stores programs or instructions that can run on the processor, and when the programs or instructions are executed by the processor, any one of claims 1 to 19 is implemented The steps of the online learning method of the AI model.
A communication device, comprising a processor and a memory, the memory stores programs or instructions that can run on the processor, and when the programs or instructions are executed by the processor, any one of claims 20 to 35 is implemented The steps of the online learning method of the AI model.
A readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the online learning method of the AI model according to any one of claims 1-19 is implemented, or The steps of realizing the online learning method of the AI model as described in any one of claims 20 to 35.
A computer program product, the program product is executed by at least one processor to realize the online learning method of the AI model according to any one of claims 1 to 19, or to realize the method according to any one of claims 20 to 35 The steps of the online learning method of the AI model.
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, and realize the AI described in any one of claims 1 to 19 The online learning method of the AI model, or the steps of realizing the online learning method of the AI model according to any one of claims 20 to 35.